Theoretical and Applied Fracture Mechanics 103 (2019) 102246

Contents lists available at ScienceDirect

Theoretical and Applied Fracture Mechanics

journal homepage: www.elsevier.com/locate/tafmec

Finite element modeling strategies for 2D and 3D delamination propagation T

in composite DCB specimens using VCCT, CZM and XFEM approaches
⁎
Mohammad Heidari-Rarani , Mousa Sayedain
Department of Mechanical Engineering, Faculty of Engineering, University of Isfahan, 81746-73441 Isfahan, Iran

A R T I C LE I N FO A B S T R A C T

Keywords: Virtual crack closure technique (VCCT), cohesive zone modeling (CZM) and extended finite element method
Delamination (XFEM) are three well-known numerical methods frequently used for crack propagation modeling. It is often
VCCT questioned by new researchers and engineers: which method is more appropriate for modeling of delamination
CZM propagation in composites? In this study, advantages, limitations, and challenges of each method are discussed
XFEM
with the goal of finding a suitable and cost-effective solution for modeling of delamination propagation in
DCB specimen
laminated composites. To this end, a composite double cantilever beam (DCB) specimen as a benchmark ex-
ample is modeled in ABAQUS and delamination propagation is simulated using three above methods and the
combination of XFEM with VCCT and CZM. Two-dimensional plain strain and three-dimensional DCB models are
both considered. Finite element results are compared with experimental results available in the literature for
unidirectional DCB specimens. Finally, the accuracy, convergence speed, run-time and mesh dependency of each
method are discussed. The XFEM-CZM was found as a suitable method for simulation of delamination growth.

1. Introduction extended finite element method (XFEM) which can be used with finite
element software to simulate the crack propagation in different mate-
Delamination, which is called a crack that forms between adjacent rials. A brief description of the above techniques is given in the fol-
plies, is one of the most common failure modes in laminated composites lowing.
associated with decreasing the stiffness and strength. The main reason
for most delaminations is the material and structural discontinuity that 1.1. VCCT
increases the interlayer stress [1,2]. Due to the weakness of mechanical
properties of a delaminated composite structure, there are hidden The VCCT was first expressed by Irwin [4] who assumed that when
dangers that require attention to the design of the composite parts. a crack is extended, the energy required to open the crack is the same
Delamination can grow in three different modes: Mode I: opening, required to close it. It has been widely used to calculate the strain en-
Mode II: in-plane shearing and Mode III: out-of-plane shearing. In ergy release rates as a finite element computational tool. The VCCT
practice, a combination of these three modes may occur. Double can- analysis is usually carried out based on the computed displacements
tilever beam (DCB), End notched flexure (ENF), and Mixed-mode and nodal forces in local elements around the crack tip. As shown in
bending (MMB) have been introduced as the benchmark and standard Fig. 1, the delamination model is presented for the two-dimensional
specimens for characterizing of Mode I, Mode II and Mode I + II de- DCB model with four-node square elements. The VCCT uses the nodal
lamination propagation, respectively. However, in order to investigate forces at the crack tip (FYi, FXi), relates to displacements in nodes
the degradation of the combined modes, it is first necessary to carefully around the crack tip (Δνk, j, Δuk, j), and considers the length of dela-
examine the pure modes [3]. In this study, we have focused on the DCB mination (Δa). Additionally, crack growth self-similarity is invoked.
specimen. With the help of this assumption, the mode I and mode II components of
Finite element (FE) simulation of delamination propagation is of the strain energy release rate are easily obtained from the following
great interest for design engineers to understand the failure mechanism equations [1]:
of a structure. There are three well-known techniques, e.g., virtual 1
crack closure technique (VCCT), cohesive zone model (CZM) and GI = FYi ·Δνk, j Δνk, j = νk − νj
2Δa (1)

⁎
Corresponding author.
E-mail address: [email protected] (M. Heidari-Rarani).

https://doi.org/10.1016/j.tafmec.2019.102246
Received 26 January 2019; Received in revised form 8 April 2019; Accepted 9 May 2019
Available online 10 May 2019
0167-8442/ © 2019 Elsevier Ltd. All rights reserved.
M. Heidari-Rarani and M. Sayedain Theoretical and Applied Fracture Mechanics 103 (2019) 102246

t K 0 0 ⎤ ⎧ δn ⎫
→ ⎧ n⎫ ⎡ n →
T = ts = ⎢ 0 Ks 0 ⎥ δs = Kδ
⎨t ⎬ ⎢ ⎥ ⎨ ⎬
⎩ t ⎭ ⎣ 0 0 Kt ⎦ ⎩ δt ⎭ (3)
→ →
which T is the vector of the cohesive stresses, δ is the vector of the
displacement and K is the stiffness matrix of elements.
Following crack initiation, cohesive elements experience a linear
softening, which is illustrated by a scalar damage variable, D. This value
is zero initially and increases to unit monotonically when damage oc-
curs. If the damage values are specified in the software and the model
begins to damage, values of the contact stresses are affected by this
parameter. As shown in Fig. 2, the intersection of linear traction-se-
paration behavior with the slope of (1 − D)Kn by the softening line
gives normal (tn0 ) or tangential contact stresses (ts0 and tt0 ) at each
fracture mode. Eqs. (4)–(6) represents the softening response of the
cohesive elements [15]:

(1 − D) tn0, tn0 ⩾ 0
Fig. 1. Two-dimensional schema of VCCT for the four-node square elements tn = ⎧ 0
[1]. ⎨ tn ,
⎩ tn0 < 0 (4)

1 ts = (1 − D) ts0, (5)
GII = FXi ·Δuk, j Δuk, j = uk − uj
2Δa (2)
tt = (1 − D) tt0, (6)
where FYi and FXi are the forces at the coincident node i, which are
calculated from the FE solution. Additionally, Δνk, j and Δuk, j are the Elices et al. [16] reviewed the advantages and limitations of CZM
relative displacements between the nodes k and j. and examined the predictive capability of the CZM when applied to
Krueger [5] presented a comprehensive report about the history, different materials such as concrete, PMMA and steel. CZMs have been
approach, and applications of VCCT. Afterward, VCCT has been used by widely considered by researchers for simulating delamination propa-
many researchers for modeling of delamination propagation in two- gation in the last decade [17–20]. For example, Camanho et al. [21]
dimensional (2D) and three-dimensional (3D) simulations. For ex- investigated the numerical simulation of delamination propagation in
ample, de Morais et al. [6], Shokrieh et al. [7], Adrian and Krueger [8], composite material using CZM. They presented a method for simulation
Shokrieh et al. [9] and Pereira and de Morais [10] calculated strain of delamination propagation based on decohesion elements placed be-
energy release rate in unidirectional and multidirectional DCB speci- tween the layers of solid elements that crack probably extends. Gutkin
mens using VCCT. Robinson et al. [11] presented an alternative data et al. [22] implemented a tri-linear cohesive law in FE software and
reduction method that does not need to measure delamination length showed that this shape of law could simulate the R-curve effects for
and the critical failure energy is determined by using the VCCT. Sam- DCB and compact-tension specimens accurately. Mollón et al. [23]
borski [12] studied the distribution of mode I strain energy release rate compared the effect of implicit and explicit solvers in the simulation of
along the delamination front using VCCT. Miravete and Jiménez [13] delamination growth in DCB specimens using two-step extension model
reviewed the application of FEM for prediction of onset of delamination and CZM. Heidari-Rarani et al. [24] investigated finite element mod-
growth by VCCT. Valvo [14] presented a revised VCCT which furnishes eling of delamination propagation in a DCB specimen with considering
a physically consistent partitioning of fracture modes by associating the the damage effects by CZM and finally compared the results with ex-
mode I and II contributions to the amounts of work done in a suitably perimental results. Fan et al. [25] simulated delamination growth in
defined two-step process of closure of the virtually extended crack. DCB specimens using cohesive elements in front of delamination. They
compared FE results with experimental ones, and concluded that CZM
could predict delamination growth well. Lu et al. [26] investigated the
1.2. CZM effect of cohesive element parameters on systematic delamination
modeling. They studied the sensitivity of the overall process of damage
Modeling of crack growth by the CZM has provided one of the great progression and failure loading for strength and stiffness of the cohesive
changes in the fracture mechanics and has made it possible to easily elements and concluded that the penalty stiffness parameter mainly
define the growth path in the region ahead of the crack tip. In CZM, the affects computational efficiency and modeling accuracy.
equations needed to analyze the fracture process are simplified and
focused only on the tip region of the crack, which considers the char- 1.3. XFEM
acteristics of this region with a traction-separation law. In the finite
element modeling, the traction-separation law is assigned to cohesive Modeling stationary discontinuities, such as a crack, with the con-
elements defined in the crack path. The constitutive response of the ventional FEM requires that the mesh conforms to the geometric dis-
cohesive elements used in this paper is defined by bi-linear traction- continuities. Therefore, considerable mesh refinement is needed in the
separation law as shown in Fig. 2 for individual modes. neighborhood of the crack tip to capture the singular asymptotic fields
For separate mode, traction-separation law depends on at least three adequately. Modeling a growing crack is even more cumbersome be-
parameters, i.e., initial stiffness (penalty stiffness), maximum traction cause the mesh must be updated continuously to match the geometry of
and steady-state fracture energy. Under mixed-mode fracture, the the discontinuity as the crack progresses. XFEM is an extension of the
properties required to define the bilinear traction-separation law are conventional FEM, with the difference being that it allows the presence
three penalty stiffnesses (Kn, Ks, Kt), three critical fracture energies (GIc, of discontinuities in an element by enriching degrees of freedom with
GIIc, GIIIc), and the interfacial strengths (tn0 , ts0 , tt0 ). special displacement functions, e.g., the discontinuous jump function H
The elastic constitutive model that relates the normal and shear (x) and near-tip asymptotic functions Fα. Based on the partition of unity,
tractions to the normal and shear separations of a cracked element is the approximation of a displacement vector function u around a 2D
given by (without considering the coupling effects) crack can be written as [15]:

2
M. Heidari-Rarani and M. Sayedain Theoretical and Applied Fracture Mechanics 103 (2019) 102246

Fig. 2. Bilinear traction-separation response of the cohesive elements.

Fig. 3. XFEM using phantom nodes [15].

only for nodes whose shape function is separated by the crack tip [17].
The phantom points that are located on the main points are presented
for discontinuity elements. In Fig. 3, when an element is healthy, every
phantom point is entirely on the main points. When the element is
broken by crack, the cracked element is divided into two parts. Each
part consists of some phantom points and the main points that depend
on the direction of crack. Each phantom point to its original point is not
Fig. 4. Dimensions of DCB specimens. tied to each other and can move.
Here, some references which applied XFEM for modeling of dela-
N 4 mination propagation are reviewed. Sosa and Karapurath [27] simu-
u= ∑ NI (x ) ⎡⎢uI + H (x ) aI + ∑ Fα (x ) bIα⎤⎥ lated delamination propagation in GLARE fiber metal laminates using
I=1 ⎣ α=1 ⎦ (7) XFEM. The results showed that XFEM could be a promising method for
analyzing the failure of composite structures. Motamedi and Milani
where NI (x) are nodal shape functions, and uI is the displacement [28] presented a 3D XFEM approach for numerical simulation of de-
vector associated with the continuous part of the finite element solu- lamination in unidirectional composites under mode I fracture. To
tion. aI generates the vector of the enriched degree of freedom, H(x) is avoid instability during simulation, a zone ahead of the crack tip is
the discontinuous jump function between crack surfaces, bIα is the en- embedded, so that the user-defined XFEM elements are activated along
riched nodes degree of freedom vector, and Fα(x) are the crack tip the tip of this zone. The XFEM results were compared with experi-
functions. The first part of the Eq. (7) applies to all nodes within the mental results. Li and Chen [29] proposed an extended cohesive da-
model, and the second part is applicable for nodes that are functionally mage model for simulating multi-crack propagation in fibrous
shaped by internal cracks, and the third part of the relationship is used

3
M. Heidari-Rarani and M. Sayedain Theoretical and Applied Fracture Mechanics 103 (2019) 102246

Fig. 5. Definition of connection nodes zone and the initial crack length.

90
Experiment [24]
• Maximum principal stress criterion (Maxps):
80 σ ⎫
VCCT Gc=Gini f = ⎧ max
0
70 ⎨
⎩ σmax ⎬
⎭ (8)
VCCT Gc=Gprop
0
60 where σmax represents the maximum allowable principal stress.
Load, N

50
• Maximum principal strain criterion (Maxpe):
40
ε ⎫
f = ⎧ max
0
30 ⎨
⎩ εmax ⎬
⎭ (9)
0
20 where εmax represents the maximum allowable principal strain.

10
• Maximum nominal stress criterion (Maxs):
0
〈t 〉 t t
0 2 4 6 8 10 12 14 16 18 20 f = max ⎧ n0 , s0 , t0 ⎫
Displacement, mm ⎨
⎩ tn ts tt ⎬ ⎭ (10)
Fig. 6. Comparison of FE and experimental results for 2D modeling of dela- where tn is the component normal to the likely cracked surface, and ts
mination propagation in a DCB specimen by VCCT. and tt are the two shear components on the likely cracked surface. Here,
tn0 , ts0 and tt0 represent the peak values of the nominal stress.
Table 1
Mechanical properties of UD DCB specimen [24]. • Maximum nominal strain criterion (Maxe):
E1 (GPa) E2 (GPa) G12 (GPa) ν12 Kp (N/ Gini (J/ Gss (J/ σmax ε ε ε
mm3) m2) m2) (MPa) f = max ⎧ n0 , s0 , t0 ⎫
⎨
⎩ εn εs εt ⎬ ⎭ (11)
33.5 10.23 4.26 0.27 104 98.2 559 15
where εn0 , εs0 and εt0 represent the peak values of the nominal strain.

composites. They combined the cohesive damage model (CDM) in • Quadratic nominal stress criterion (Quads):
2 2 2
conjunction with XFEM. It eliminates degrees of enriched freedom that 〈t 〉 t t
exists in the growth path of cracks in the composites without a pre- f = ⎧ n0 ⎫ + ⎧ s0 ⎫ + ⎧ t0 ⎫
⎨ t
⎩ n ⎭ ⎬ ⎨ t
⎩ s⎭ ⎬ ⎩ t ⎬
⎨ t ⎭ (12)
determined path. They used a UEL subroutine to implement this ap-

• Quadratic nominal strain criterion (Quade):

proach in ABAQUS.
For traction-separation laws, both stress and strain-based criteria for
2 2 2
damage initiation and energy-based criteria for damage evolution are ε ε ε
available in ABAQUS. The damage initiation criteria in ABAQUS are: f = ⎧ n0 ⎫ + ⎧ s0 ⎫ + ⎧ t0 ⎫
⎨ ε
⎩ n⎭ ⎬ ⎨ ε
⎩ s⎭ ⎬ ⎨
⎩ t ⎬
ε ⎭ (13)

Fig. 7. A DCB specimen with a thin layer of cohesive elements.

4
M. Heidari-Rarani and M. Sayedain Theoretical and Applied Fracture Mechanics 103 (2019) 102246

Fig. 8. Damage of cohesive elements in the wake of delamination front.

80 suggested by Benzeggagh and Kenane [31]:

Experiment [24]
70 CZM Gc=Gini (Surface-based) GT
CZM Gc=Gprop (Surface-based)
⩾1
GII + GIII η
60 CZM Gc=Gini (Element-based)
GIc + (GIIc − GIc ) ( GT ) (15)
CZM Gc=Gprop (Element-based)
50 where η is the B-K material parameter. Reeder [32] showed that it fits
well the mixed-mode I + II data.
Load, N

40

30 • Reeder law
20 The Reeder's fracture criterion is the development of 2D B-K cri-
terion as follows [32]:
10
GT
⩾1
GII + GIII η GII + GIII η
0
0 2 4 6 8 10 12 14 16 18 20 22
GIc + (GIIc − GIc ) ( GT ) + (GIIIc − GIIc ) ( GIII
GII + GIII )( GT )
Displacement, mm (16)

Fig. 9. Comparison of FE and experimental results for 2D modeling of dela- where GT is the sum of the strain energy release rate components, i.e.,
mination propagation in a DCB specimen by CZM. GI + GII + GIII. The factor η is determined by a curve fit. The Reeder law
is best applied when GIIC ≠ GIIIC; when GIIC = GIIIC, the Reeder law
For all above criteria, damage is assumed to initiate when f = 1. reduces to the B-K law.
Relations (10) and (12) are usually applied to predict damage initiation The above damage initiation and propagation criteria are in general
in cohesive elements used for delamination modeling. forms and they can be used for damage modeling in various types of
Also, ABAQUS provides the following energy-based damage evolu- materials. That’s why the effect of different fracture modes and their
tion criteria for 3D mixed-mode delamination propagation: interactions are observed in these relations. About the three-dimen-
sional effects on the fracture toughness, it is comprehensively discussed

• Power law in the literature, e.g. see [32–36]. Since this study has focused on the
unidirectional symmetric DCB specimens under mode I loading, the
This criterion is one of the early and more popular criteria and it is above laws are simplified so that they only need to normal traction for
established in terms of an interaction between the strain energy release crack initiation and GIc for crack propagation. However, it is often
rates as follows [30]: questioned by new engineers what are the advantages, limitations, and
challenges of VCCT, CZM and XFEM for delamination modeling in
α β χ composite laminates? Or which type of model, two-dimensional (2D) or
⎛ GI ⎞ + ⎛ GII ⎞ + ⎛ GIII ⎞ ⩾ 1
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
three-dimensional (3D) gives more accurate results? This study pro-
⎝ GIc ⎠ ⎝ GIIc ⎠ ⎝ GIIIc ⎠ (14)
vides a comprehensive comparison of different simulation methods of
where α, β, χ, GIc, GIIc, and GIIIc are six fitting parameters that they delamination growth in multilayered composites and the advantages
describe the fracture critical surfaces. and limitations of each method are finally addressed.

• B-K law 2. Finite element modeling

A somewhat different approach to represent the mode interaction is In this section, the numerical procedures for the simulation of

Fig. 10. Initial crack and mesh size for XFEM

analysis.

5
M. Heidari-Rarani and M. Sayedain Theoretical and Applied Fracture Mechanics 103 (2019) 102246

Fig. 11. Damage at the elements of crack edges in XFEM.

100 90

90 Experiment [24]
Experiment [24 ] 80
VCCT (3D-shell)
80 XFEM (VCCT behavior) 70 VCCT (3D-solid)
70 XFEM (CZM behavior)
60
60
Load, N

50

Load, N
50
40
40
30
30
20
20
10
10

0 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 22
Displacement, mm
Displacement, mm
Fig. 14. Comparison of FE and experimental results for 3D modeling of dela-
Fig. 12. Comparison of FE and experimental results for 2D modeling of dela-
mination propagation in a DCB specimen by XFEM. mination propagation in a DCB specimen via VCCT.

delamination propagation using the VCCT, CZM and XFEM are de- displacement control employing the linear loading approach.
scribed. All the simulations were conducted in the commercial finite Therefore, the displacement boundary conditions are applied to the
element package ABAQUS [15]. A DCB specimen as a standard spe- upper and lower left-end corners of the beam (according to Fig. 5, it is
cimen is selected and all geometrical and mechanical properties and 10 mm) in the y-direction, while the other end point of the beam is fixed
experimental results are provided from Ref. [24]. Both 2D and 3D finite in × and y directions.
element models are considered.
A DCB specimen consists of two rectangular beams with a pre- 2.1.1. VCCT
cracked surface introduced in the mid-plane. The specimen has a In order to define the VCCT in ABAQUS, a fracture criterion with the
nominal dimension of 150 × 25 × 4.2 mm and a pre-crack length of definite fracture energy (Gini or Gss) should be defined in the interaction
35 mm as shown in Fig. 4. The material parameters for the finite ele- module. The initial crack is defined by debonding of two parts of the
ment simulation are summarized in Table. Kp is penalty stiffness, Gini DCB specimen. In other words, a node set is defined in front of crack tip
and Gss are initiation and steady-state toughness, and σmax is maximum along the path which crack will extend and bonding contact between
traction. the edge nodes is introduced. Fig. 5 shows the node set defined in front
of the crack tip. After the mesh sensitivity analysis, the element size and
2.1. Two-dimensional DCB modeling the number of elements through-the-thickness are 0.2 mm and 20, re-
spectively.
The 2D model is created and meshed by four-node bilinear plane The load-displacement curve of DCB specimen obtained from the
strain (CPE4) elements. The numerical analysis is performed under finite element analysis is compared with the experimental results in

Shell model (S4) Solid model (C3D8R)

Fig. 13. 3D modeling of DCB specimen.

6
M. Heidari-Rarani and M. Sayedain Theoretical and Applied Fracture Mechanics 103 (2019) 102246

Fig. 15. Damage of cohesive elements in 3D modeling of delamination growth.

80 90
Experiment [24] Experiment [24]
80
70 CZM-3D-shell VCCT 2D
CZM-3D-Element based 70 CZM 2D
60 CZM-3D-Surface-based XFEM-CZM 2D
60
50

Load, N
50
Load, N

40 40

30 30

20 20

10
10
0
0 0 2 4 6 8 10 12 14 16 18 20
0 2 4 6 8 10 12 14 16 18 20 22 Displacement, mm
Displacement, mm
Fig. 18. Comparison of FE and experimental results of 2D modeling of dela-
Fig. 16. Comparison of FE and experimental results for 3D modeling of dela- mination propagation in a DCB specimen.
mination propagation in a DCB specimen by CZM.
80
80 Experiment [24]
70 VCCT 3D
Experiment [24]
70
XFEM (VCCT behavior) CZM 3D
60
60 XFEM (CZM behavior) XFEM-CZM 3D

50 50
Load, N
Load, N

40 40

30 30

20
20
10
10
0
0 2 4 6 8 10 12 14 16 18 20 0
0 2 4 6 8 10 12 14 16 18 20
Displacement, mm
Displacement, mm
Fig. 17. Comparison of FE and experimental results for 3D modeling of dela-
mination propagation in a DCB specimen by XFEM. Fig. 19. Comparison of FE and experimental results of 3D modeling of dela-
mination propagation in a DCB specimen.

Fig. 6. The effect of using the initiation fracture toughness (Gini) and the
steady-state fracture toughness (Gss) as the fracture energy have been and element-based cohesive. In surface-based cohesive, a traction-se-
investigated in this figure. The same initial slope of the load-displace- paration contact with cohesive properties is defined between two lines
ment curves in FE and experimental results verifies the FE model. The of upper and lower arms of DCB. While in element-based cohesive
overall trend of load-displacement curve is obtainable using steady- model, a thin layer (about 0.02 mm) of cohesive elements with traction-
state toughness. The VCCT cannot model the nonlinear part as well as separation behavior is defined between two arms as shown in Fig. 7. A
the peak load of load-displacement curve accurately. bilinear traction-separation law is assumed as shown in Fig. 2 for co-
hesive behavior.
In this study, both methods are applied and their efficiencies are
2.1.2. CZM discussed. The size and numbers of elements, especially at the cohesive
As in the VCCT, a DCB specimen is modeled with dimensions of region are very important. If the elements size increases, it causes many
Fig. 4 and material properties of Table 1. There are two ways for convergence errors during the analysis, and if it refines too much, the
modeling of delamination propagation via CZM: surface-based cohesive computational time will increase significantly. From mesh sensitivity

7
M. Heidari-Rarani and M. Sayedain Theoretical and Applied Fracture Mechanics 103 (2019) 102246

80
element local 1-direction (default) or parallel to the element local 1-
Experiment [24]
direction. As mentioned before, XFEM models crack propagation more
70 VCCT 3D convenient than conventional methods by enriching the elements.
CZM 3D However, element size is important but it is not as sensitive as the
60 XFEM-CZM 3D conventional FE methods. Element size affects the run-time and accu-
XFEM-CZM 2D racy of the results. Mesh sensitivity analysis was performed for both
50 techniques separately. Element size and the numbers of elements
through-the thickness was selected respectively 0.2 mm and 21 for 2D
Load, N

40 XFEM-VCCT modeling as shown in Fig. 10; 0.5 mm and 7 for XFEM-

CZM.
30 Fig. 11 shows the damage at the elements of the crack edges. The
damage value varies from 0 to 1. The damage variable of bisected
20 elements change to 1, when crack grows through the element.
The load-displacement curves of 2D XFEM simulation are compared
10
with experimental results in Fig. 12. The XFEM-CZM simulates the
overall trend of load-displacement curve more accurate than XFEM-
0
VCCT. It should be noted that the initial slope of load-displacement
0 2 4 6 8 10 12 14 16 18 20 curve cannot be modeled precisely by XFEM-VCCT.
Displacement, mm

Fig. 20. Comparison of FE and experimental results various modeling of dela- 2.2. Three-dimensional DCB modeling
mination propagation in a DCB specimen.
Two configurations, i.e., solid and shell are considered for 3D
modeling of DCB specimens in this section. The types of elements used
analysis, the element size and the number of elements along the
for meshing the solid and shell models are C3D8 and S4, respectively.
thickness are 0.5 mm and 20, respectively. Fig. 8 shows the damage in
Dimensions and mechanical properties are the same as the 2D model in
cohesive elements during the delamination propagation. Red and blue
the previous section. Three elements are considered through the
colors show damage variable 1 and 0, respectively.
thickness of each arm. End of the specimen is entirely fixed in all di-
The load-displacement curve of 2D modeling of delamination pro-
rections and the line load is coupled to a reference point (RP), and the
pagation using cohesive elements is compared with experimental re-
displacement is applied to this RP. Fig. 13 represents the 3D finite
sults in Fig. 9. Both the initiation fracture toughness (Gini) and steady-
element models. FE results are compared with experimental ones. In the
state fracture toughness (Gss) has been applied as the fracture energy.
following, the results of each technique are discussed in detail. Ac-
The results show that steady-state fracture toughness should be used for
cording to the conclusions given for the 2D modeling in previous sec-
delamination propagation in the case of the difference between the
tion, only results of the steady-state toughness are reported here.
initiation toughness and steady-state toughness, it is so-called R-curve
effects, are dominant.
FE results show that both surface-based and element-based ap- 2.2.1. VCCT
proaches of CZM give the same results. The main advantage of element- The element size and the number of elements through-the-thickness
based cohesive approach is that damage of cohesive elements is visible are respectively obtained 0.5 mm and 4 after mesh sensitivity. Fig. 14
during the crack propagation. The disadvantages of element-based compares the load-displacement curves of solid and shell models for 3D
method are problems in convergence and more run-time. Also, cohesive DCB specimens. Results of both models are approximately the same.
layer thickness is a crucial parameter may significantly affect the FE The advantage of shell model is the use of fewer elements and conse-
results. Most of the recent investigations have suggested a value be- quently reduction of computational time.
tween 0.01 and 0.02 mm for delamination propagation [37–39]. The
disadvantage of surface-based method is its high dependency to the 2.2.2. CZM
element size in the contact area and its effect on the convergence. The element size and the number of elements along the thickness for
the solid model are selected 0.8 mm and 6 respectively, and for the shell
model it is chosen 2 mm after mesh sensitivity analysis. The damage of
2.1.3. XFEM cohesive elements during the delamination propagation is shown in
The XFEM can be combined with VCCT and CZM. Here, it is named Fig. 15.
XFEM-VCCT and XFEM-CZM. The maximum nominal stress criterion, The FE and experimental load–displacement curves of the DCB
Eq. (10), was used for damage initiation and B-K law with mode in- specimens are illustrated in Fig. 16. Simulations with the solid and shell
dependent assumption for damage evolution in XFEM-CZM. Also, the elements show the same results. The initial slopes of the curves of both
crack propagation direction can be chosen when the damage initiation models are perfectly coincided with the initial slope of the experimental
criterion is satisfied. The crack can extend at a direction normal to the curve. Since the bilinear CZM is used in this study, simulation of the

Table 2
A general comparison among all methods of simulation of delamination propagation in a unidirectional DCB.
Number and type of elements Accuracy Run-time (s) Mesh dependency Convergence speed (increments/run-time)

VCCT (2D) 15,000 (CPE4) Good 2136 High 0.05

VCCT (3D) 15,000 (C3D8R) Good 7025 High 0.018
CZM (2D) 2630 (COH2D4, CPE4R) Very good 142 Moderate 0.79
CZM (3D) 46,282 (C3D8R, COH3D8) Very good 10,837 High 0.01
XFEM-VCCT (2D) 4845 (CPE4) Good 1365 Moderate 0.3
XFEM-VCCT (3D) 30,000 (C3D8R, RNODE2D) Good 13,034 Moderate 0.017
XFEM-CZM (2D) 2100 (CPE4R) Excellent 835 Moderate 0.56
XFEM-CZM (3D) 30,000 (C3D8R, RNODE2D) Excellent 40,928 Moderate 0.02

8
M. Heidari-Rarani and M. Sayedain Theoretical and Applied Fracture Mechanics 103 (2019) 102246

Table 3
Advantages and limitations of different methods of delamination modeling.
Method Advantages Limitations

VCCT • Simple • Very sensitive to mesh size

• Gives G values at separate modes • Needs to crack path a priori
• Cannot model fiber bridging
• Needs to initial crack
• Defined only as surface-based interaction
CZM (Bilinear) • Simple • Sensitive to cohesive parameters
• Accurate results for initiation and propagation • Needs to crack path a priori
• Relatively less run-time • Cannot model fiber bridging
• Defined as both element-based and surface-based • Needs to initial crack
XFEM-CZM • Relatively simple • More run-time
• No need to crack path a priori • Difficulties in convergence
• Available several criteria for damage initiation and evolution in software • Significance of crack tip location within the element
• No need to initial crack
nonlinear part of load-displacement curve which is due to fiber bridging paper. Readers are referred to Refs. [24,44–46].
cannot be modeled. This point is completely depicted in Ref. [24] and a Some factors such as accuracy, run-time, numbers of elements and
new trilinear traction-separation law is proposed for modeling of the convergence rate for each method are compared in Table 2. All analyses
bridging effect. Other shapes of traction-separation laws like what de- are performed at same conditions by a PC with Intel Core i5 CPU and
veloped in Refs. [40–43] are not in the scope of this paper. The use of 8 GB RAM. From Table 2, the convergence rate of the CZM is higher
surface-based cohesive can be more effective than element-based one in than other methods, and the accuracy of the XFEM-CZM is more than
reduction of computational time. others.
Table 3 summarizes the advantages and limitations of VCCT, bi-
2.2.3. XFEM linear CZM, and XFEM-based cohesive zone approach.
In 3D modeling of delamination propagation by XFEM, it is not
possible to use shell elements. Therefore, DCB specimens are only me- 4. Conclusions
shed by solid elements (C3D8) with the size of 1 mm and the number of
element along the thickness was selected 6 from mesh sensitivity ana- In this study, different techniques of delamination modeling in a
lysis. Fig. 17 compares the FE and experimental results for 3D XFEM. unidirectional DCB composite specimen are compared with experi-
According to this figure, XFEM-CZM provides more accurate results in mental results. Results showed that the VCCT does not have ability to
comparison to the XFEM-VCCT. In general, the XFEM-VCCT is not a accurately model delamination propagation in multilayered DCB com-
suitable method for simulating delamination propagation in unidirec- posites. But it is a simple and effective way to predict the delamination
tional DCB specimens. Therefore, in order to reduce the computational initiation. VCCT is very mesh sensitive because it works with nodal
time, 3D modeling of delamination via the XFEM-CZM method is pre- force and displacements. The main feature of VCCT is determination of
ferable. strain energy release rate at individual modes. Implementing bilinear
CZM via element-based and surface-based methods provides the same
results, while run-time and convergence of these approaches are com-
3. Comparison of results and discussion pletely different. The main shortcoming of VCCT and CZM is that the
crack path should be determined prior to propagation. Therefore, XFEM
In this section, results of 2D and 3D modeling of delamination is combined with CZM and VCCT and these techniques are applied for
propagation via different methods are compared and discussed to fi- modeling of delamination propagation in 2D and 3D models. CZM and
nally find the appropriate approach for simulation of delamination XFEM-CZM methods showed more accurate results rather than other
growth in unidirectional DCB specimens. The load-displacement curves methods. But the main feature of XFEM-CZM method is that it doesn’t
of 2D and 3D models are compared with each other in Figs. 18–20. An need to determine the crack growth path before. Thus, the XFEM-CZM
oscillating behavior is observed in 3D models rather than 2D ones. This method is suggested as a robust method for modeling of delamination
is perhaps due to stick-slip behavior during the delamination propa- propagation.
gation. For example in 2D VCCT, only one node is located at the crack
tip and this is split into two nodes after crack propagation. While a References
number of nodes are located at the delamination front in 3D models.
Due to Poisson’s ratio effect, nodes at the middle of DCB specimen tend [1] I.S. Raju, T.K. O’Brien, Fracture mechanics concepts, stress fields, strain energy
to grow sooner than others or in other words delamination front release rates, delamination initiation and growth criteria, in: S. Sridharan (Ed.),
changes from straight to curve shape. Delamination Behaviour of Composites, Woodhead Publishing, Cambridge, 2008,
pp. 3–27.
None of these methods could model the nonlinear part of load-dis- [2] H.Y. Sarvestani, A. Naghashpour, M. Heidari-Rarani, Bending analysis of a general
placement curve that it is due to fiber bridging. In fiber bridging phe- cross-ply laminate using 3D elasticity solution and layerwise theory, Int. J. Adv.
nomenon, a softening behavior is observed. After delamination initia- Struct. Eng. 7 (4) (2015) 329–340.
[3] M.M. Shokrieh, M. Salamat-talab, M. Heidari-Rarani, Effect of interface fiber angle
tion, the bridging happens at a low traction and large separation. VCCT on the R-curve behavior of E-glass/epoxy DCB specimens, Theoret. Appl. Fract.
only needs one parameter, i.e., critical strain energy release rate. It Mech. 86 (2016) 153–160.
means that this model reaches to a maximum stress after a linear elastic [4] G.R. Irwin, Analysis of stresses and strains near the end of a crack traversing a plate,
J. Appl. Mech. 24 (1957) 361–364.
behavior. Therefore, this technique cannot model fiber bridging.
[5] R. Krueger, Virtual crack closure technique: history, approach, and applications,
Bilinear CZM needs two parameters, i.e., the critical energy release rate Appl. Mech. Rev. 57 (2004) 109–143.
and maximum interface stress. Therefore, the fiber bridging cannot be [6] A.B. de Morais, M.F. Moura, A.T. Marques, J.P.M. Goncalves, P.P. Camanho,
modeled by linear softening at bilinear CZM but it can predict the onset Analysis of crack propagation in double cantilever beam tests of multidirectional
laminates, Mech. Mater. 35 (2003) 641–652.
and growth of the delamination well. Normally, trilinear or multilinear [7] M.M. Shokrieh, M. Heidari-Rarani, S. Rahimi, Influence of curved delamination
CZMs can model the bridging well that it is out of the scope of this front on toughness of multidirectional DCB specimens, Compos. Struct. 94 (2012)

9
M. Heidari-Rarani and M. Sayedain Theoretical and Applied Fracture Mechanics 103 (2019) 102246

1359–1365. unidirectional composite laminates: a sensitivity analysis of modeling parameters,

[8] C.O. Adrian, R. Krueger, Benchmark assessment of automated delamination pro- J. Compos. Mater. 3 (2013) 113–126.
pagation capabilities in finite element codes for static loading, Finite Elem. Anal. [29] X. Li, J. Chen, An extended cohesive damage model for simulating multicrack
De. 54 (2012) 28–36. propagation in fiber composites, Compos. Struct. 143 (2016) 1–8.
[9] M.M. Shokrieh, H. Rajabpour-Shirazi, M. Heidari-Rarani, M. Haghpanahi, [30] J.R. Reeder, An evaluation of mixed-mode delamination failure criteria. NASA/TM-
Simulation of mode I delamination propagation in multidirectional composites with 1992-104210, 1992.
R-curve effects using VCCT method, Comput. Mater. Sci. 65 (2012) 66–73. [31] M.L. Benzeggagh, M. Kenane, Measurement of mixed-mode delamination fracture
[10] A.B. Pereira, A.B. de Morais, Mode I interlaminar fracture of carbon/epoxy multi- toughness of unidirectional glass/epoxy composites with mixed-mode bending ap-
directional laminates, Compos. Sci. Tech. 64 (2004) 2261–2270. paratus, Compos. Sci. Technol. 56 (1996) 439–449.
[11] P. Robinson, F. Javidrad, D. Hitchings, Finite element modelling of delamination [32] J.R. Reeder, 3D mixed-mode delamination fracture criteria – an experimentalist’s
growth in the DCB and edge delaminated DCB specimens, Compos. Struct. 32 perspective, in: Proceedings of American Society for Composites, 21st annual
(1995) 275–285. technical conference, Dearborn, MI, United States, 17–20 September, 2006.
[12] S. Samborski, Numerical analysis of the DCB test configuration applicability to [33] L.P. Pook, 50-year retrospective review of three-dimensional effects at cracks and
mechanically coupled fiber reinforced laminated composite beams, Compos. Struct. sharp notches, Fatigue Fract. Eng. Mater. Struct. 36 (8) (2013) 699–723.
152 (2016) 477–487. [34] M.R.M. Aliha, H. Saghafi, The effects of thickness and Poisson’s ratio on 3D mixed-
[13] A. Miravete, M.A. Jiménez, Application of the finite element method to prediction mode fracture, Eng. Fract. Mech. 98 (2013) 15–28.
of onset of delamination growth, Appl. Mech. Rev. 55 (2002) 89–105. [35] F. Berto, L. Pook, A. Campagnolo, Corner point singularities under in-plane and out-
[14] P.S. Valvo, A revised virtual crack closure technique for physically consistent of-plane loading: a review of recent results, Eng. Solid. Mech. 5 (3) (2017) 167–176.
fracture mode partitioning, Int. J. Fract. 173 (1) (2012) 1–20. [36] M.R.M. Aliha, A. Bahmani, S. Akhondi, Numerical analysis of a new mixed mode I/
[15] Abaqus 6.14 user’s manual, Dassault systems, Providence, RI, USA, 2014. III fracture test specimen, Eng. Fract. Mech. 134 (2015) 95–110.
[16] M. Elices, G.V. Guinea, J. Gómez, J. Planas, The cohesive zone model: advantages, [37] A. Turon, C.G. Dávila, P.P. Camanho, J. Costa, An engineering solution for mesh
limitations and challenges, Eng. Fract. Mech. 69 (2002) 137–163. size effects in the simulation of delamination using cohesive zone models, Eng.
[17] E. Abdullah, J.F. Ferrero, J.J. Barrau, J.B. Mouillet, Development of a new finite Fract. Mech. 74 (2007) 1665–1682.
element for composite delamination analysis, Compos. Sci. Tech. 67 (2007) [38] J.L. Curiel-Sosa, B. Tafazzolimoghaddam, Ch. Zhang, Modelling fracture and dela-
2208–2218. mination in composite laminates: energy release rate and interface stress, Compos.
[18] D. Yang, J. Ye, Y. Tan, Y. Sheng, Modeling progressive delamination of laminated Struct. 189 (2018) 641–647.
composites by discrete element method, Comput. Mater. Sci. 50 (2011) 858–864. [39] T. Yamanaka, M. Heidari-Rarani, L. Lessard, V. Feret, P. Hubert, A new finite ele-
[19] M. Meo, E. Thieulot, Delamination modelling in a double cantilever beam, Compos. ment method for modeling delamination propagation without additional degrees of
Struct. 71 (2005) 429–434. freedom, Compos. Struct. 147 (2016) 82–98.
[20] C.T. Sun, Z.H. Jin, Modeling of composite fracture using cohesive zone and bridging [40] E. Farmand-Ashtiani, D. Alanis, J. Cugnoni, J. Botsis, Delamination in cross-ply
models, Compos. Sci. Tech. 66 (2006) 1297–1302. laminates: identification of traction separation relations and cohesive zone mod-
[21] P.P. Comanho, C.G. Dávila, M.F. de Moura, Numerical simulation of mixed-mode eling, Compos. Sci. Tech. 119 (2015) 85–92.
progressive delamination in composite materials, J. Compos. Mater. 35 (2003) [41] L. Zhao, J. Zhi, J. Zhang, Z. Liu, N. Hu, XFEM simulation of delamination in
641–652. composite laminates, Compos. Part A: Appl. Sci. Manuf. 80 (2016) 61–71.
[22] R. Gutkin, M.L. Laffan, S.T. Pinho, Modelling the R-curve effect and its specimen- [42] N. Vajragupta, V. Uthaisangsuk, B. Schmaling, S. Münstermann, A. Hartmaier,
dependence, Int. J. Solid Struct. 48 (2011) 1767–1777. W. Bleck, A micromechanical damage simulation of dual phase steels using XFEM,
[23] V. Mollón, J. Bonhomme, A.M. Elmarakbi, A. Argüelles, J. Viňa, Finite element Comput. Mater. Sci. 54 (2012) 271–279.
modelling of mode I delamination specimens by means of implicit and explicit [43] M. Heidari-Rarani, A.R. Ghasemi, Appropriate shape of cohesive zone model for
solvers, Polym. Tes. 31 (2012) 404–410. delamination propagation in ENF specimens with R-curve effects, Theor. Appl.
[24] M. Heidari-Rarani, M.M. Shokrieh, P.P. Camanho, Finite element modeling of mode Fract. Mech. 90 (2017) 174–181.
I delamination growth in laminated DCB specimens with R-curve effects, Compos. [44] C.G. Dávila, C.A. Rose, P.P. Camanho, A procedure for superposing linear cohesive
Part B: Eng. 45 (2013) 897–903. laws to represent multiple damage mechanisms in the fracture of composites, Int. J.
[25] C. Fan, P.Y. Ben, J.J.R. Cheng, Cohesive zone with continuum damage properties Fract. 158 (2) (2009) 211–223.
for simulation of delamination development in fiber composites and failure of ad- [45] E. Farmand-Ashtiani, J. Cugnoni, J. Botsis, Specimen thickness dependence of large
hesive joints, Eng. Fract. Mech. 75 (2008) 3866–3880. scale fiber bridging in mode I interlaminar fracture of carbon epoxy composite, Int.
[26] X. Lu, M. Ridha, B.Y. Chen, V.B.C. Tan, T.E. Tay, On cohesive element parameters J. Sol. Struct. 55 (2015) 58–65.
and delamination modelling, Eng. Fract. Mech. 206 (2019) 278–296. [46] B.D. Manshadi, E. Farmand-Ashtiani, J. Botsis, A.P. Vassilopoulos, An iterative
[27] J.L. Curiel Sosa, N. Karapurath, Delamination modelling of GLARE using the ex- analytical/experimental study of bridging in delamination of the double cantilever
tended finite element method, Compos. Sci. Tech. 72 (2012) 788–791. beam specimen, Compos. Part A 61 (2014) 43–50.
[28] D. Motamedi, A.S. Milani, 3D nonlinear XFEM simulation of delamination in

10