Vous êtes sur la page 1sur 7

Composite Structures 94 (2011) 239–245

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Experimental and numerical study of fastener pull-through failure in

GFRP laminates
G. Catalanotti a,⇑, P.P. Camanho a, P. Ghys b, A.T. Marques a
DEMec, Faculdade de Engenharia, Universidade do Porto, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal
ALSTOM Transport, Rue Albert Dhalenne, 48, 93482 Saint-Ouen, France

a r t i c l e i n f o a b s t r a c t

Article history: An experimental and numerical study of the fastener pull-through failure mode in glass–fiber reinforced
Available online 2 July 2011 plastic (GFRP) laminates using both phenolic and vinylester resins is presented. It is shown that the type
of resin does not affect the mechanical response of the joint when a pull-through test is performed
Keywords: because similar values of the sub-critical initial and final failure loads are obtained. Moreover, consider-
A. Hybrid structures ing that the joint is considering to fail when the sub-critical failure load is reached, a methodology to pre-
B. Bolted joints dict the pull-through failure mode is proposed. It is observed that the main failure mechanism is the
C. Pull-through
delamination of the plies; therefore, the prediction of the sub-critical initial failure load is performed
using a three-dimensional finite element model where cohesive elements are used to simulate delamina-
tion. The predictions agree remarkably well with the experimental results.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction study the influence of the different parameters on the pull-through

failure, namely: the geometry of the fastener head, the thickness of
GFRP laminates are used in marine, railway and automotive the laminate, and the stacking sequence. It was observed that the
industries in non-structural parts and, more recently, in the main failed specimens exhibit intralaminar damage similar to that
load-carrying structures. The use of composites leads to a reduc- shown in composite panels when a low-velocity impact is applied
tion of the weight (and consequently of the cost of the transporta- in the transversal direction. In fact, the damaged zones (intralam-
tion), a reduction of manufacturing costs (simplification of the inar and interlaminar damage) are distributed conically with re-
design and reduction of the costs required for the assembly), and spect to the axis of the fastener.
to a reduction of the recurring cost (composites require less main- Banbury et al. [9] performed numerical analysis to study the
tenance than metals). pull-through damage mechanism. The finite element (FE) analysis
Due to their high specific stiffness and strength and to the flex- indicated that:
ibility in their use, GFRP are nowadays used together with metals
in the design of hybrid low-cost train structures [1]. Hybrid struc-  shear stresses in the vicinity of the bolt head are responsible for
tures are interesting for the railway industry because they may re- the intralaminar matrix cracking in through-the-thickness
sult in a mass reduction of about 12–24% and in a cost reduction of direction;
about 20% [2]. One of the main design requirements of the railway  tensile in-plane stresses are responsible for the flexural defor-
industry is the calculation of the strength of hybrid bolted joints, mation of the material in particular in low-modulus laminate;
which correspond to the critical regions of the structures.  matrix cracking was observed to be the primary failure mecha-
While the prediction of the strength of composite bolted joints nism while delamination (caused by the high interlaminar
under in-plane failure mechanism has been throughly investigated shear and peel stress) was the secondary mechanism.
in the literature [3–7], few attempts have been made to predict
out-of-plane failure mechanism, such as fastener pull-through Moreover, a numerical procedure to simulate the progressive
[8–11]. damage in the material was proposed. A progressive damage mod-
Banbury and Kelly [8] investigated the pull-through failure of el together with the maximum principal strain criterion were used
carbon laminates manufactured using both plane weave and unidi- and a good agreement between the experiments and the numerical
rectional prepregs. An experimental campaign was performed to predictions, both in terms of failure load, damaged zones and dam-
age mechanisms, was obtained.
⇑ Corresponding author. Kelly and Hallström [10] performed an experimental and
E-mail address: giuseppe.catalanotti@fe.up.pt (G. Catalanotti). numerical investigation of a laminate subjected to transversal

0263-8223/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
240 G. Catalanotti et al. / Composite Structures 94 (2011) 239–245

Fig. 1. Hybrid connection.

loads. Different geometries, materials and lay-ups were investi- The laminates were manufactured using the technique of resin
gated. The damage onset was observed to occur at a load of infusion and they have the quasi-isotropic lay-up reported in
approximately 20–30% of the failure load, and the different failure Table 1. The mechanical properties of these laminates are reported
mechanism were identified. As mentioned in previous investiga- in Table 2 where: Ei is the Young’s modulus in i direction, mij is the
tions [8], the damaged zones show both interlaminar and intralam- Poisson’s ration in i–j directions, Gij is the shear modulus in i–j
inar matrix cracking. A three-dimensional finite element method directions, XT is the longitudinal tensile strength, XC is the longitu-
was proposed to predict the first-ply failure. dinal compressive strength, YT is the transverse tensile strength, YC
Elder et al. [11] proposed a simplified three-dimensional finite
elements model to model pull-through failure of composite lami-
nates. It was concluded that simplified models allow to obtain a
Table 1
good prediction of the pull-through failure for quasi-isotropic lam-
Orientation pattern for GF-vinylester/phenolic composite.
inates even if additional efforts are required to properly define the
cohesive parameters used in the numerical model. Ply Type of product Name Supplier
Fastener pull-through is particularly important for train struc- 14 ±45° – 610 g/m2 EBX 600 SELCOM
tures where several hybrid connections are present: 13 90° – 600 g/m2 Roving UD 600 CHOMARAT GAZECHIM
12 ±45° – 610 g/m2 EBX 600 SELCOM
11 90° – 600 g/m2 Roving UD 600 CHOMARAT GAZECHIM
 the connection between the main frame and the floor; 10 ±45° – 610 g/m2 EBX 600 SELCOM
 the connection between the carbody shell and the top floor (in 9 0° – 1246 g/m2 UNIE 1200 SELCOM
the case of a double deck carbody shell); 8 ±45° – 610 g/m2 EBX 600 SELCOM
 the connection between the main frame and the carbody shell; 7 ±45° – 610 g/m2 EBX 600 SELCOM
6 0° – 1246 g/m2 UNIE 1200 SELCOM
 the connection between the carbody shell and the roof.
5 ±45° – 610 g/m2 EBX 600 SELCOM
4 90° – 600 g/m2 Roving UD 600 CHOMARAT GAZECHIM
Fig. 1 represents the connection between the main frame and 3 ±45° – 610 g/m2 EBX 600 SELCOM
the side of the carbody shell. This connection uses two different 2 ±90° – 600 g/m2 Roving UD 600 CHOMARAT GAZECHIM
materials because the internal panel cannot be toxic, while the 1 ±45° – 610 g/m2 EBX 600 SELCOM

external panel must have a good fire resistance [12]. For this rea- Isoftalic poliester gelcoat GCI S90000 VM10 SAF 1 POLYPROCESS
son, the internal panel is manufactured using phenolic resin, while
the external panel is manufactured using vinylester resin.
The aim of this paper is to experimentally study the fastener
pull-through failure mode in GFRP laminates and to propose a
Table 2
numerical technique to predict the response of a bolted joints un- Mechanical properties of GF-V and GF-P UD laminate.
der out-of-planes loads. Taking into account that in industrial
Materials GF-vinylester GF-phenolic
applications several resins are used to satisfy the current legisla-
tion (in particular about the fire behavior [12]) the study presented E1 (MPa) 42,830 35,200
E2 = E3 (MPa) 1530 3000
here concerns two different resins: phenolic and vinylester.
m12 0.35 0.35
m13 = m23 0.3 0.3
2. Experiments G12 (MPa) 2800 3400
G13 = G23 (MPa) 2800 3400
XT (MPa) 350 355
2.1. Materials
YT = ZT (MPa) 35 35
XC (MPa) 300 300
The composites investigated in this study are: YC = ZC (MPa) 30 30
ST (MPa) 10 19
SL (MPa) 22 14
 Fiber glass-vinylester composite (GF-V).
q (kg/m3) 1863 1900
 Fiber glass-phenolic composite (GF-P).
G. Catalanotti et al. / Composite Structures 94 (2011) 239–245 241

is the transverse compressive strength, ST is the transverse shear

strength, SL is the longitudinal shear strength and q is the density.
The material was tested after a heat aging treatment according
to the AFNOR norm [13].

2.2. Pull-through tests

The pull-through test is performed following the norm ASTM

D7332 – Standard Test Method for Measuring the Fastener Pull-
Through Resistance of a Fiber-Reinforced Polymer Matrix Composite
This test method proposes two procedures, A and B. Procedure
A is used to enhance the fastener design while procedure B is used
to study other design variables. Both procedures use flat test spec-
imens with a circular hole in the center where the fastener is in-
stalled. In procedure A two specimens are joined using a fastener Fig. 3. Equipment set-up for PT test.

and one plate is rotated of 45° with respect to the other. Each test
specimen contains four additional holes at the corners where the
the other with a 10 mm diameter hole. In the following, each
fasteners are installed to connect the specimens to the test fixtures.
specimen was indicated with the denomination PT-M-D where M
In procedure B, the load is applied to the test specimen using a
indicates the material (P for phenolic and V for vinylester) and D
yoke as shown in Fig. 2. Since procedure A is more complex and
indicates the diameter in millimeters (6 or 10).
has inherent problems associated with the flexural stiffness of
the specimen to test, procedure B was used.
2.3. Derived properties
The test was conducted using an INSTRON-4208 test machine.
Fig. 3 shows the experimental set-up used. The test machine was
The load–displacement for a pull-through test is used to iden-
equipped with a 100 kN load cell. The speed of the machine (dis-
tify three important characteristics of the joints that are:
placement controlled test) was 2 mm/min. The temperature of
the room was 23 °C and the relative humidity was 50% for all the
 the initial sub-critical failure load: the load at the first sub-crit-
duration of the tests.
ical failure of the specimen;
After each test the damaged specimen was examined and the
 the initial sub-critical failure displacement: the displacement at
type of failure was identified.
the first sub-critical failure of the specimen;
The test results depend on Clearance Hole parameter (Cb). This
 the failure load: the maximum load attained in the test.
is the diameter of the plate that it is used in procedure B of the test.
In the tests performed Cb was taken as 30 mm. The dimensions of
The specimen shows a first failure mode (generally delamina-
all test specimens, the ratio of the Clearance Hole Diameter Cb to
tion) at a relatively low load. After delamination, the specimen is
Fastener Hole Diameter d, and the ratio of the Fastener Hole
able to support increasing loads. This point is identified in the
Diameter to the thickness of the specimen (h) are reported in the
curve by a load-drop, which is followed by a decrease of the stiff-
following points. The test specimens are square plates with a
ness of the test coupon.
length of 105 mm. Three specimens were tested for each configu-
ration and for each material: one with a 6 mm diameter hole and
2.4. GF-phenolic specimens

Table 3 reports the dimensions of the specimens and the geo-

metric parameters of the equipment used. Fig. 4 shows the load
vs. displacement curves for the phenolic specimens. The derived
properties for the specimens tested are reported in Tables 4 and 5.
It should be noted that some specimens (see Tables 4 and 5)
exhibited bolt failure. The bolts used for these tests were general
purpose bolts (D933 8.8).

2.5. GF-vinylester specimens

Table 6 reports the dimensions of the specimens and the geo-

metric parameters of the equipment used. Fig. 5 shows the load
vs. displacement curves for vinylester specimens. The derived
properties for the specimens tested are reported in Tables 7 and 8.

Table 3
GF-P specimens’ dimensions.

Specimen Diameter (mm) Cb/D h (mm) D/h

PT-P-6-1 6 5 6.80 0.882
PT-P-6-2 6 5 6.80 0.882
PT-P-6-3 6 5 6.80 0.882
PT-P-10-1 10 3 6.60 1.515
PT-P-10-2 10 3 6.80 1.471
PT-P-10-3 10 3 6.50 1.538
Fig. 2. ASTM D7332 – procedure B.
242 G. Catalanotti et al. / Composite Structures 94 (2011) 239–245

Fig. 4. Pull-through test; load vs. displacement for GF-phenolic specimens. Fig. 5. Pull-through test; load vs. displacement for GF-vinylester specimens.

Table 4 Table 7
Pull-through test, results for PT-P-6 specimens. Pull-through test, results for PT-V-6 specimens.
Specimen In. sub-crit. Failure load In. sub-crit. Specimen In. sub-crit. Failure load In. sub-crit.
failure load (N) failure displ. failure load (N) failure displ.
(N) (mm) (N) (mm)
PT-P-6-1a 12,340 17,348 1.228 PT-V-6-1a 11,200 17,184 1.127
PT-P-6-2a 11,752 17,236 1.301 PT-V-6-2 11,404 17,624 1.315
PT-P-6-3a 12,384 17,132 1.360 PT-V-6-3 11,340 17,204 1.463
Average 12,158 17,238 1.296 Average 11,314 17,337 1.302
STDV 352 108 0.066 STDV 104 248 0.168
Bolt failure. a
Bolt failure.

Table 8
Table 5 Pull-through test, results for PT-V-10 specimens.
Pull-through test, results for PT-P-10 specimens.
Specimen In. sub-crit. Failure load In. sub-crit.
Specimen In. sub-crit. Failure load In. sub-crit. failure failure load (N) failure displ.
failure load (N) (N) displ. (mm) (N) (mm)
PT-P-10-1 19,140 32,460 1.462 PT-V-10-1 18,448 34,172 1.447
PT-P-10-2 18,528 32,780 1.403 PT-V-10-2 18,920 32,864 1.440
PT-P-10-3 17,976 31,608 1.492 PT-V-10-3 18,876 39,692 1.477
Average 18,548 32,282 1.452 Average 18,748 35,576 1.455
STDV 582 605 0.045 STDV 260 3624 0.020

2.6. Damaged zones

penetrated by the washer. The delamination occurs abruptly when
Fig. 6 shows the photos of a specimen tested up to failure. Both the sub-critical initial load is reached. It is not possible to exactly
sides of the specimen are visible. The bottom side, Fig. 6b, shows identify the location of the first delamination, even if it seems
delamination near the hole. As explained before, delamination oc- plausible that they occur at the interface with the thick 0° plies
curs at the begin of the test when the load is relatively low (initial due to the high interlaminar stresses promoted by the thick plies.
sub-critical failure). If the load increases, the intralaminar fracture
in the material starts and, at the end, the specimen is totally 2.7. Comparison of the PT test results

Fig. 7 shows the failure loads of the two materials tested. As ex-
Table 6 pected, increasing the hole diameter increases both the initial sub-
GF-V specimens’ dimensions. critical failure load and the final failure load of the specimens. The
Specimen Diameter (mm) Cb/D h (mm) D/h results indicate that the two materials tested exhibit only slight
PT-V-6-1 6 5 7.00 0.857
differences in the values of the two failure loads considered.
PT-V-6-2 6 5 7.15 0.839 It should be also noted that increasing the diameter also in-
PT-V-6-3 6 5 6.80 0.882 creases the ratio between the initial sub-critical failure load to
PT-V-10-1 10 3 7.00 1.429 the ultimate failure load. This means that larger holes exhibit an
PT-V-10-2 10 3 6.70 1.493
initial sub-critical failure load that is relatively lower when
PT-V-10-3 10 3 7.20 1.389
compared to the ultimate failure load.
G. Catalanotti et al. / Composite Structures 94 (2011) 239–245 243

(a) intralaminar fracture (b) delamination at sub-critical failure

Fig. 6. Pull-through specimens after loading.

The relation between the tractions transferred along a bridged

crack, s, and the corresponding displacement jumps, D, reads:
si ¼ ð1  dÞkDi  ½di3 dkhD3 i ð1Þ
where d is a scalar damage variable and k is the penalty stiffness.
The displacement jump D3 is related to mode I, whereas the dis-
placement jumps D1 and D2 are related to shear modes of loading.
The operator hxi is defined as hxi ¼ 12 ðx þ jxjÞ, and dij is the
Kroenecker delta.

Specimen Fastener

Fig. 7. Comparison of PT tests failure loads.

The load corresponding to the first load-drop will be taken as

the design load. This load corresponds to the onset of delamination
that is likely to propagate under fatigue loading.

3. Numerical model Steel plate

Fig. 8. FE model of the pull-through test (top view).

Based on the previous remarks, a numerical model, based on the
Finite Element Method, is developed to predict the delamination
onset load. This load is taken as the design load.

3.1. Numerical implementation

The finite element model was created using Abaqus 6.8 [15].
The mesh is shown in Fig. 8. The specimen and the bolt (screw
and washers) are represented as deformable bodies, while the steel
plate used for the pull-through test is modeled as an analytical ri-
gid surface. Frictionless contact is considered between the different
parts. Fig. 9 shows the FE model as viewed from the bottom.
The test specimen, shown in Fig. 10, was modeled using 8-
nodes linear brick reduced integration elements (CRD8R) with a
typical element size of 1 mm.
Cohesive finite elements, implemented as an Abaqus Users’sub-
routines (UMAT) [15], are used to predict delamination. These
elements are used in all interfaces between plies with different fi-
ber orientation angles.
The detailed definition of the cohesive model is presented in
[16]. For the sake of completeness, the main aspects of the consti-
tutive model are outlined in the following paragraphs. Fig. 9. FE model of the pull-through test (bottom view).
244 G. Catalanotti et al. / Composite Structures 94 (2011) 239–245

Table 9
Critical values of SERR (N/mm) [18].

Mode mixity / ¼ GIGþG

Test G/Ic G/IIc Gc

0 DCB 1.25 0 1.25

0.25 MMB 1.14 0.38 1.52
0.5 MMB 0.97 0.97 1.94
0.75 MMB 0.6 1.81 2.41
1 ENF 0 3.6 3.6

Fig. 10. FE model, elastic and cohesive elements (in red). (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of
this article.)

Introducing k as the norm of the displacement jump:

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
k¼ hD3 i2 þ D2sh ; Dsh ¼ ð D1 Þ 2 þ ð D2 Þ 2 ð2Þ

The damage activation function for general mixed-mode loading is Fig. 11. Predicted load vs. displacement curves.
defined as:
F ðD; dÞ ¼ LðDÞ  d 6 0 ð3Þ
where To reduce the complexity of the model (and the time needed for the
( ) analysis) the cohesive elements were used only in the vicinity of the
Df ðk  Do Þ bolt as shown in Fig. 10.
LðDÞ ¼ min ;1 ð4Þ
kðDf  Do Þ The relevant parameters for the definition of the cohesive ele-
ments are obtained using experimental data previously measured
and: for a similar material [18]. The material properties used are shown
dt ¼ maxf0; max½LðDt Þg; 0 6 s 6 t; 8t P 0 ð5Þ in Table 9.
s Using these values the exponent for the B–K criterion [17] is cal-
The displacement jumps corresponding to delamination onset (Do) culated using the least-squares method as g = 1.98. The penalty
and to delamination propagation (Df) under mixed-mode condi- stiffness, k, is taken as 106 N/mm3. The interface strengths are cal-
tions are obtained using the Benzeggagh and Kenane criterion culated using the engineering solution proposed by Turon et al.
[17] for delamination propagation under mixed-mode loading, [19] resulting in so3 ¼ 28:8 MPa and sosh ¼ 48:8 MPa.
yielding [16]:
n h 3.2. Numerical results
2 2  2 i o1=2
Do ¼ Do3 þ Dosh  Do3 Bg ð6Þ
Taking into account that no major differences were found be-
1 h   i tween the delamination onset loads for the two materials tested,
Df ¼ o Do3 Df3 þ Dosh Dfsh  Do3 Df3 Bg ð7Þ the numerical simulations are conducted only for the GF-P
where g is the mixed-mode interaction parameter used in the Ben- Fig. 11 shows the load–displacement relation predicted by the
zeggagh and Kenane criterion [17] and B is a local mixed-mode ratio numerical model. The predictions for the initial sub-critical failure
defined as: load and the experimental values are reported in Table 10. A rea-
Gsh D2 sonable agreement between the numerical predictions and the
B¼ ¼ sh ð8Þ experimental data is obtained. The maximum error, 10%, is ob-
Gsh þ GI k2
tained for the PT-P-10 specimen. It is also observed that the
Do3 and Dosh are respectively the displacement jumps corresponding load-drop identified in the experiments is also captured by the
to delamination onset in mode I and in shear mode: numerical model.
so3 sosh
Do3 ¼ ; Dosh ¼ ð9Þ
k k
where so3 and sosh are the pure mode interlaminar strengths. Table 10
Df3 and Dfsh are respectively the displacement jumps correspond- Initial sub-critical failure load: experiments and predictions.
ing to delamination propagation under mode I and in shear mode: Specimens Experimental value (N) Numerical prediction (N) Error (%)
2GIc 2GIIc PT-P-6 12,158 11,922 2%
Df3 ¼ ; Dfsh ¼ ð10Þ PT-P-10 18,548 16,700 10%
so3 sosh
G. Catalanotti et al. / Composite Structures 94 (2011) 239–245 245

4. Concluding remarks [4] Thoppul SD, Finegan J, Gibson RF. Mechanics of mechanically fastened joints in
polymer–matrix composite structures: a review. Compos Sci Technol
An experimental and a numerical study of the pull-through [5] Hart-Smith LJ. Mechanically-fastened joints for advanced composites –
damage in GFRP laminates is presented in this paper. Two different phenomenological considerations and simple analysis, Douglas paper
6748. McDonnell Douglas Corporation; 1978. p. 1–32.
material system (GF-phenolic and GF-vinylester) and two geome-
[6] Hart-Smith LJ. Design and analysis of bolted and riveted joints in fibrous
tries (diameter of the hole 6 mm and 10 mm) are investigated. It composite structures, Douglas paper 7739. McDonnell Douglas Corporation;
is concluded that: 1986. p. 1–15.
[7] Camanho PP, Lambert M. A design methodology for mechanically fastened
joints in laminated composite materials. Compos Sci Technol
 increasing the diameter of the bolt increases both the sub-crit- 2006;66:3004–20.
ical initial failure load and the failure load; [8] Banbury A, Kelly DW. A study of fastener pull-through failure of composite
 larger holes exhibit a sub-critical initial failure load that is rel- laminates. Part 1: Experimental. Compos Struct 1999;45:241–54.
[9] Banbury A, Kelly DW, Jain LK. A study of fastener pull-through failure of
atively lower when compared to the ultimate failure load; composite laminates. Part 2: Failure prediction. Compos Struct 1999;45:255–70.
 the interlaminar damage is the predominant phenomenon for [10] Kelly G, Hallström S. Strength and failure mechanisms of composite laminates
damage onset; subject to localised transverse loading. Compos Struct 2005;69:301–14.
[11] Elder DJ, Verdaasdonk AH, Thomson RS. Fastener pull-through in a carbon fibre
 using three-dimensional finite element models using linear epoxy composite joint. Compos Struct 2008;86:291–8.
elastic elements together with cohesive elements it is possible [12] French Norm NF F 16-101:1988. All vehicles – the burning behavior –
to predict the value of the sub-critical initial failure load with materials to choose, October; 1998 [original in French].
[13] French Norm NF T 57-107:1986. Glass–fibre-reinforced plastics –
a reasonable accuracy. measurement of the change of characteristics during hot water treatment,
December; 1986 [original in French].
[14] ASTM D7332/D7332M – 09. Standard test method for measuring the fastener
pull-through resistance of a fiber-reinforced polymer matrix composite; 2009.
[15] Abaqus 6.8 documentation. Dessault Systèmes; 2008.
[16] Turon A, Camanho PP, Costa J, Dávila CG. A damage model for the simulation of
The first author acknowledges the financial support of the Euro- delamination in advanced composites under variable-mode loading. Mech
Mater 2006;38:1072–89.
pean Commission under Contract No. MRTN-CT-2005-019198.
[17] 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;56:439–49.
References [18] Tumino D, Catalanotti G, Cappello F, Zuccarello B. Experimental tests on
fatigue induced delamination in GFRP and CFRP laminates. In: Experimental
[1] Catalanotti G. Multi-material joints for high-speed train structures. PhD thesis, analysis of nano and engineering materials and structures: proceedings of
University of Porto; 2011. ICEM13, Alexandropulis, Greece, July 1–6; 2007.
[2] Comparaison entre caisses à technologie différent. Alstom Transport; 2008 [in [19] Turon A, Camanho PP, Costa J, Renart J. Accurate simulation of delamination
French]. growth under mixed-mode loading using cohesive elements: definition of
[3] Camanho PP, Matthews FL. Stress analysis and strength prediction in FRP: a interlaminar strengths and elastic stiffness. Compos Struct 2010;92:1857–64.
review. Composites – Part A 1997;28:529–47.