Mixed Cohesive Model PDF

Engineering Fracture Mechanics 148 (2015) 145–179
Contents lists available at ScienceDirect
Engineering Fracture Mechanics

journal homepage: www.elsevier.com/locate/engfracmech
Coupled cohesive zone models for mixed-mode fracture:

A comparative study
R. Dimitri a,⇑, M. Trullo a, L. De Lorenzis b, G. Zavarise a
a
Dipartimento di Ingegneria dell’Innovazione, Università del Salento, via per Monteroni, 73100 Lecce, Italy
b
Institut für Angewandte Mechanik, Technische Universität Braunschweig, Bienroder Weg 87, 38106 Braunschweig, Germany
a r t i c l e i n f o a b s t r a c t
Article history: This paper checks the consistency of some published exponential and bilinear mixed-mode
Received 9 July 2015 cohesive zone models. The effect of coupling on traction-separation behavior and energy
Accepted 8 September 2015 dissipation is investigated and the path-dependence of the debonding work of separation
Available online 15 September 2015
and failure domain is evaluated analytically and numerically. All selected models present
several inconsistencies, except for the one by van den Bosch et al. (2006), which is, how-
Keywords: ever, not currently formulated within a thermodynamical framework but postulated in
Cohesive zone modeling
an ad-hoc manner. We thus propose a thermodynamically consistent reformulation of this
Contact
Debonding
model within damage mechanics, which holds monolithically for loading, unloading, deco-
Mixed-mode fracture hesion and contact.
Thermodynamics Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction
Increasing attention has been devoted in the last decades to the use of cohesive zone models (CZMs) to study mixed-
mode delamination, debonding, and, more generally, crack initiation and propagation within quasi-brittle materials or at
material interfaces. This is due to the computational efficiency of these models and to their versatility for numerical
implementation in many areas of computational mechanics. Cohesive models describe the traction-separation behavior of
interfaces before and during fracture, and are characterized by two phases, i.e. an increase of the traction up to a peak value
and a subsequent decrease to zero, which describe the crack initiation and the growth of cohesive surfaces until new
traction-free surfaces appear.
The basic concept of CZMs was proposed by Barenblatt [1,2] and Dugdale [3] as an alternative approach to singularity
driven fracture mechanics, and has been extensively used in the literature to analyze the fracture process in a number of
material systems such as concrete [4,5], polymers [6,7], ductile materials [8–10], ceramics [11], but also bimaterial systems
such as polymer matrix composites [12–15] and metal matrix composites [16]. CZMs have been also used to simulate
fracture under static [17–19], dynamic [11,20], and cyclic [21,22] loading conditions, delamination of layered composites
at the micro- and macro-scale [23–26], delamination between a coating and a substrate [27], debonding of fiber reinforced
polymer (FRP) sheets from concrete, masonry or steel substrates [28–34].
Despite the first models have been developed for single-mode fracture processes, cohesive fracture is expected to involve
mixed-mode conditions, as observed in practice by experimental investigations performed on various types of lap joints [35],
or interfaces between FRP sheets and flat [36,37] or curved substrates [38,39]. Mixed-mode CZMs can be classified as
⇑ Corresponding author. Tel.: +39 0832 297177.

E-mail address: rossana.dimitri@unisalento.it (R. Dimitri).
http://dx.doi.org/10.1016/j.engfracmech.2015.09.029
0013-7944/Ó 2015 Elsevier Ltd. All rights reserved.
146 R. Dimitri et al. / Engineering Fracture Mechanics 148 (2015) 145–179
Nomenclature
gN normal displacement
gT tangential displacement
g Nmax normal displacement corresponding to the normal cohesive strength
g Tmax tangential displacement corresponding to the tangential cohesive strength
g Nu critical separation in normal direction
g Tu critical separation in tangential direction
gm mixed-mode relative displacement
g m;u ultimate value of the mixed-mode relative displacement
g m;max maximum value of the mixed-mode relative displacement
pN normal pressure
pT tangential pressure
pNmax normal cohesive strength
pTmax tangential cohesive strength
/ mixed-mode fracture energy
/N fracture energy in mode I
/T fracture energy in mode II
W total work of separation
WN normal work of separation
WT tangential work of separation
dW C contact contribution to the virtual work
q ratio between the tangential and the normal work of separation
m mixed-mode parameter which controls the zone of influence of mode II
r traction-free normal separation following complete shear separation
k effective opening displacement
kP dimensionless deformation at which the softening behavior begins
kN initial (elastic) stiffnesses in mode I
kT initial (elastic) stiffnesses in mode II
kP penalty contact stiffness
b displacement-based mode-mixity
c material-dependent parameter for the PL criterion
g material-dependent parameter for the BK criterion
T traction vector
w Helmholtz energy
wþN elastic energy associated to the normal tensile contribution
wN elastic energy associated to the normal compressive contribution
wT elastic energy associated to the tangential contribution
ð jÞ
di mixed-mode scalar-valued damage parameter
u horizontal displacement
v vertical displacement
E elastic modulus of materials
m Poisson’s ratio of materials
lpz length of the fracture process zone
uncoupled or coupled. Based on the former approach, the normal (i.e. mode I) traction is independent of the tangential
(i.e. mode II) separation and vice versa, and a mixed-mode fracture criterion can be introduced or not. If a failure criterion
is met in a point, the interface becomes unable to bear any load in that point and the local tractions drop to zero (see e.g.
[35,40]). If no mixed-mode fracture criterion is defined, failure in a point is considered to occur when the energy release rates
in either mode I or mode II reach their respective maximum values (see e.g. [41–44]). Uncoupled CZMs are typically used
when interface separation is constrained to occur in a single predefined direction, i.e. when either mode I or mode II sepa-
ration take place. Cohesive models, however, are usually applied to engineering problems where the mode of interface sep-
aration is not predefined, and thus require the use of coupled formulations. In these approaches, all the components of the
traction vector depend on all the components of the interface separation.
Within the category of coupled models, cohesive models usually differ in several aspects, including the presence or not of
a potential, the shape of the curve, the unloading path, and the definition of the coupling parameters. Coupled cohesive mod-
els derived from a potential and based on dimensionless coupling parameters have been, for example, proposed by Beltz and
Rice [45], Needleman [46], Tvergaard [12], Tvergaard and Hutchinson [47], Xu and Needleman [18], or more recently, by Park
et al. [48], McGarry et al. [49]. With this approach, however, there are some cases where the fracture energy is the same for
varying mode mixities (see e.g. [47]) or where some limitations can emerge when the mode I fracture energy is different
R. Dimitri et al. / Engineering Fracture Mechanics 148 (2015) 145–179 147
from the mode II fracture energy (see e.g. [18,49]). This can be regarded as a drawback, since experimental results indicate
the fracture energy to be often different (typically significantly larger) in mode II than in mode I [21,50–54]. Due to the
high fracture energy in mode II a specimen can feature an higher loading capacity under certain loading conditions, as
demonstrated in the literature by Carpinteri et al. [55]. A potential function able to capture different fracture energies is,
therefore, necessary to simulate correctly a mixed-mode process of fracturing, as proposed recently by Park et al. [48],
and demonstrated in the present work.
In non-potential-based models, the traction-separation laws and their coupling are often directly postulated, see e.g.
Camanho et al. [23], Cornec et al. [56], Scheider and Brocks [57], Högberg [58], van den Bosch et al. [59]. Thus, despite having
been adopted for many practical applications, these models are not guaranteed to be thermodynamically consistent [60]. On
the other hand, they typically allow for different fracture energies in different mode mixities.
Despite the large number of cohesive models proposed, little attention has been devoted to ensuring their consistency for
mixed-mode conditions, primarily in a thermodynamical sense. Also, the relative performance of different models is receive-
ing an increasing attention in the scientific community for a comparative assessment of cohesive models. A comparison
between single-mode CZMs was performed by Chandra et al. [61], who employed two models (an exponential and a bilinear
one) to evaluate numerically the mechanical behavior of metal matrix composites subjected to a push-out test. Their numer-
ical analyses demonstrated the influence of the shape of single-mode relationships on the macroscopic mechanical response.
Similarly, Volokh [62] and Alfano [63] compared four different types of CZMs to study the behavior of a mode I rigid block
peel test [62], a mode I double cantilever beam (DCB) and a mode II pull-out test [63,64]. Volokh [62] showed that equal
cohesive strength and separation work values for different CZMs do not necessarily lead to the description of the same frac-
ture process. Alfano [63] also evidenced the influence of the specific interface law on the algorithmic numerical performance
and on the degree of approximation. The trapezoidal model was demonstrated to give the worst results both in terms of
numerical stability and convergence to the exact solution, while the exponential model was found to be the optimal com-
promise between computational cost and approximation. At the same time, both the bilinear and trapezoidal CZMs were
demonstrated to be quite appropriate for modeling the undamaged state of cracked adhesive joints by controlling the
pre-peak slope of the traction-separation model.
A consistency check was performed by van den Bosch et al. [27] for the exponential model by Xu and Needleman [18],
which was shown to describe unrealistically mixed-mode decohesion unless the fracture energies of both modes were
assumed equal. As already mentioned, this assumption goes generally against the experimental observations. In the same
paper, a modified, non-potential-based exponential model was proposed through a modification of the coupling between
normal and shear modes, to correct the unphysical behavioral features of the original model. A critical review on some
potential-based CZMs has been also provided, recently, in the literature by Park and Paulino [65], where two models based
on polynomial representations [14,46] have been discussed together with three models based on the concept of the universal
binding energy [17,18,45]. The main limitations of these potential-based models were highlithed when solving mixed-mode
problems, due to the boundary conditions associated with cohesive fracture. A single framework was suggested in the same
work to formulate more effective models leading to a positive stiffness (i.e. an increase of failure resistance for increasing
separations) under certain separation paths. A thorough investigation of new potential-based and non-potential-based CZMs
has been additionally performed in a recent work by McGarry et al. [49] under conditions of mixed-mode separation and
overclosure. A modified potential-based Xu and Needleman model was first proposed in order to control the non-physical
repulsive normal tractions and negative incremental energy dissipation under a monotonic mixed-mode separation, and
to provide a more correct penalization of mixed-mode closure. Two additional non-potential-based models were also
proposed in the same work to fully eliminate the problem of repulsive normal tractions.
Other mixed-mode models frequently used in numerical applications can be similarly checked for consistency, as here
performed for some additional potential-based and non-potential-based CZMs. A thorough analysis on the effect of the cou-
pling parameters on stress distributions and energy dissipation can be useful to evaluate possible physical inconsistencies of
these models, such as local peculiar features in the coupled elastic or softening mechanical response of the interface, incom-
plete dissipation of the fracture energy during decohesion, and residual load-carrying capacity in the normal or tangential
directions after complete failure. Advantages and/or shortcomings of each cohesive model, if clearly assessed, may also jus-
tify the adoption of one model or another for numerical implementations. The present work aims at carrying out such an
analysis, as partly illustrated in Dimitri et al. [66]. The improved Xu and Needleman mixed-mode exponential model pro-
posed by van den Bosch et al. [27] is considered as a benchmark, whereas the additional selected models are the exponential
one recently proposed by McGarry et al. [49], as well as the two linear mixed-mode CZMs by Camanho et al. [23] and
Högberg [58]. While the selected group of models is necessarily limited, the proposed analysis paradigm can be extended
to any other model in a straightforward manner. The basic question raised in this paper is whether these models realistically
predict a mixed-mode debonding process. A parametric analysis on the effect of the coupling parameters on tractions and
energy dissipation is performed. Analytical predictions are compared with results from a numerical finite element imple-
mentation, where the interface is described with zero-thickness generalized contact elements, which integrally incorporate
decohesion and contact. Three case studies are analyzed: a simple patch test, a bimaterial peel test under mixed-mode load-
ing conditions, and the standard mixed-mode bending test (MMB). The formulations are implemented and tested using the
finite element code FEAP.
In the second part of the paper, a thermodynamically consistent mixed-mode CZM is proposed based on a reformulation
of the improved model of Xu and Needleman [18] proposed by van den Bosch [27]. Starting from a predefined Helmholtz
energy, the interface model is derived by following the second law of thermodynamics through the Coleman and Noll pro-
cedure. The inelastic nature of the decohesion process is accounted for by means of damage variables and their evolution law
is determined in such a way as to reproduce the decohesion behavior of the model in van den Bosch [27]. The reformulated
model, unlike the original one, accounts monolithically for loading and unloading conditions, as well as for decohesion and
contact. Also in this case three computational examples are provided: a simple patch test, a matrix/particle debonding test,
and a mixed-mode DCB test.
The remainder of this paper is organized as follows: the selected coupled mixed-mode CZMs are briefly reviewed in Sec-
tion 2. They are investigated analytically in Sections 3 by evaluating the effect of coupling on cohesive tractions and energy
dissipation. Section 4 is concerned with numerical applications whose results are compared with the analytical predictions.
In Section 5, a thermodynamically consistent reformulation of the Xu and Needleman exponential model as modified by van
den Bosch et al. [27] is illustrated and tested. Finally, conclusions are drawn in Section 6.
2. Review of the selected cohesive zone models
A brief description of the mixed-mode coupled CZMs chosen for the investigation is provided in this section. As already
mentioned, the normal and the tangential stresses, pN and pT , respectively, are related with both the relative normal and
tangential displacements, g N and g T , whereas the main parameters defining the traction-separation laws affect the crack
initiation and propagation behavior.
2.1. Model by Xu and Needleman [18] modified by van den Bosch et al. [27]
We first analyze the exponential model by Xu and Needleman [18] as modified by van den Bosch et al. [27] (henceforth
briefly denoted as CZM1). This model was verified by van den Bosch et al. [27] to be consistent, and herein is considered as a
benchmark for the comparative assessment between CZMs. This model is not derived from a potential and corresponds
exactly to the original model by Xu and Needleman [18] when the ratio introduced therein between the tangential and
the normal work of separation, denoted as q, is set equal to one. The analytical expression of the interfacial stresses in
the normal and tangential directions is given by

pN gN g g 2
¼ exp ð1Þ exp N exp 2 T ð1Þ
pNmax g Nmax g Nmax g Tmax

pT pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi g T g g g 2
¼ 2 exp ð1Þ 1þ N exp N exp 2 T ð2Þ
pTmax g Tmax g Nmax g Nmax g Tmax
where g Nmax and g Tmax are the normal and tangential displacements corresponding to the normal and tangential cohesive
strengths pNmax and pTmax , respectively (see the 3D domains in Fig. 1). The model is defined by four independent parameters,
e.g. the fracture energies in the normal (mode I) and tangential (mode II) directions, /N and /T , respectively, and the cohesive
strengths in the same directions, pNmax and pTmax . For pure modes I and II, the uncoupled laws pN ¼ pN ðg N ; g T ¼ 0Þ and
pT ¼ pT ðg N ¼ 0; g T Þ are depicted in Fig. 2. /N and /T are given by
Fig. 1. Coupled 3D domain for CZM1.

Fig. 2. CZMs 1 and 2.
Z 1
/N ¼ pN ðg N ; g T ¼ 0Þdg N ¼ exp ð1ÞpNmax g Nmax ð3Þ
0
Z 1 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
/T ¼ pT ðg N ¼ 0; g T Þdg T ¼ exp ð1ÞpTmax g Tmax ð4Þ
0 2
In the original formulation of the model, unloading was not considered. As visible in Fig. 2a, in pure mode I contact is
described using the branch of the mode I law in the negative range of g N . This branch corresponds to a regularization of
the exact contact non-penetration conditions, similar to a non-linear penalty formulation. Under simultaneous tangential
loading, the shape of this branch changes depending on the mode II loading component, so that the contact penetration error
is influenced by the tangential loading. More observations on this aspect will be reported later (Section 5).
2.2. Model by McGarry et al. [49]
The second evaluated exponential law, recently proposed by McGarry et al. [49], is again based on a modification of the
Xu and Needleman model. This law, henceforth indicated as CZM2, was devised with the objective to avoid unphysical repul-
sive normal tractions and instantaneous negative incremental energy dissipation under displacement controlled monotonic
mixed-mode separation when the work of tangential separation exceeds the work of normal separation. In addition, the
model provides the advantage of a correct penalization of mixed-mode penetration in contrast to the original model. The
modified form of the Xu and Needleman potential function is defined as

f ðg T Þg N f ðg T Þg N 1q g 2 f ðg T Þg N r q
/ðg N ; g T Þ ¼ /N 1 þ exp 1rþ exp 2 T qþ ð5Þ
g Nmax g Nmax r1 g Tmax g Nmax r 1
where

g 2T
f ðg T Þ ¼ 1 þ m m exp ð6Þ
g 2Tmax
Coupling between the normal and tangential tractions is here defined by three coupling parameters: the parameter m which
controls the zone of influence of mode II behavior for mixed-mode conditions, the parameter q already defined by Xu and
Needleman [18] as the ratio between mode II and mode I fracture energies, and the traction-free normal separation following
complete shear separation, denoted as r. More details about the physical meaning of r are given in Appendix. The interface
relationships, obtained from pðgÞ ¼ @/ðgÞ=@g, with p ¼ ðpN ; pT Þ as the interface traction vector and g ¼ ðg N ; g T Þ as the relative
displacement vector, are expressed as follows (see Fig. 3)

pN f ðg T Þg N f ðg T Þg N g 2 1q g 2 f ðg T Þg N
¼ f ðg T Þ exp ð1Þ exp exp 2 T þ 1 exp 2 T r ð7Þ
pNmax g Nmax g Nmax g Tmax r1 g Tmax g Nmax
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pT 2 1 g g 2 f ðg T Þg N
¼ expð1Þ T exp 2 T exp
pTmax q 2 g Tmax g Tmax g Nmax
r q f ðg Þg
T N mg N f ðg T Þg N g 2 1q g 2 f ðg T Þg N
qþ þ exp 2 T þ 1 exp 2 T r ð8Þ
r 1 g Nmax g Nmax g Nmax g Tmax r1 g Tmax g Nmax
Six characteristic parameters are necessary to define the model, e.g. the fracture energies /N and /T (always defined by
Eqs. (3) and (4)), the cohesive strengths pNmax ; pTmax , and the coupling parameters r and m.
Also this model does not consider unloading. As in Xu and Needleman [18], contact is dealt with using the same relations
valid for the debonding range. Thus the initial contact stiffness (at zero penetration) is dictated by the debonding parameters,
albeit in a different way than with the original model.
2.3. Model by Högberg [58]
The chosen bilinear models are those proposed by Högberg [58] and Camanho et al. [23]. The cohesive model proposed by
Högberg [58], herein indicated as CZM3, is not derived from a potential. Here the coupling parameter is an effective opening
displacement k, defined as follows
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2
gN gT
k¼ þ 1 ð9Þ
g Nu g Tu
g Nu and g Tu being the critical separation values in mode I and II, respectively, after which the interfacial stresses become
equal to zero, i.e. the surfaces are locally no longer held together. The analytical relations in both directions can be expressed
as (see Fig. 4)
( 1 gN
pN kP g Nu
for 0 < k 6 kP
¼ gN ð10Þ
pNmax 1k
kð1kP Þ g Nu
for kP < k 1
( 1 gT
pT kP g Tu
for 0 6 k 6 kP
¼ gT ð11Þ
pTmax 1k
kð1kP Þ g Tu
for kP 6 k 1

where kP is the dimensionless deformation at which the softening behavior of the adhesive layer begins, and is given by
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2ffi
gN
g Nu
þ ggTuT
kP ¼ r
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ð12Þ
2 2
gN gT
g Nmax
þ g Tmax
The fracture energies for pure modes I and II are given by

1
/N ¼ p g ð13Þ
2 Nmax Nu
1
/T ¼ pTmax g Tu ð14Þ
2
while the uncoupled models in pure modes are depicted in Fig. 5. This cohesive model is globally defined by six characteristic
parameters of the interface, e.g. the fracture energies, /N and /T , the cohesive strengths, pNmax and pTmax , and the correspond-
ing relative displacements g Nmax and g Tmax .
This formulation does not include unloading paths, and controls contact conditions independently from debonding-
related parameters.
2.4. Model by Camanho et al. [23]
The last analyzed model is the bilinear cohesive one proposed by Camanho et al. [23], denoted as CZM4. This model is not
derived from a potential. The authors define a mixed-mode relative displacement, g m , which is related to the single-mode
relative displacement as follows
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
gm ¼ hg N i2 þ g 2T ð15Þ
where the Macauley operator implies that negative normal relative displacement are assumed not to generate mixed-mode
effects. For g N > 0, a displacement-based mode-mixity b ¼ g T =g N is introduced. Single-mode loading conditions are defined
as
8
> K igi for g i 6 g imax
>
>
<
pi ¼ ð16Þ
>
> ð1 di ÞK i g i for g imax < g i < g iu
>
:
0 for g i > g iu
(see Fig. 5) with
g iu ðg i g imax Þ
di ¼ ; i ¼ N; T; di 2 ½0; 1 ð17Þ
g i ðg iu g imax Þ
and K i as the initial (elastic) stiffnesses in mode I or II. The mixed-mode relative displacement corresponding to the onset of
softening is computed as
8 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
<g 1þb2
for g N > 0
Nmax g Tmax
g m;max ¼ g 2Nmax þb2 g 2Tmax ð18Þ
:
g Tmax for g N 0
Fig. 5. CZMs 3 and 4.

based on a quadratic failure criterion of the type

2 2
pN pT
þ ¼1 ð19Þ
pNmax pTmax
while the debonding propagation under mixed-mode loading is predicted by applying the power law (PL) criterion [67] or
the Benzeggagh and Kenane (BK) criterion [52], defined in terms of the mode I and mode II works of separation W N and W T
and of the fracture energies /N and /T . For simplicity, CZM4 combined with the PL and BK criteria will be indicated as
CZM4-PL (see the 3D domains in Fig. 6) and CZM4-BK (see the 3D domains in Fig. 7), respectively. The expressions for
the two criteria are given respectively by
c c
WN WT
þ ¼1 ð20Þ
/N /T
for the PL criterion, with c as a material-dependent parameter, or
g
WT
/N þ ð/T /N Þ ¼ / with W ¼ W N þ W T ð21Þ
W
for the BK criterion, with / as the mixed-mode fracture energy, and g as a parameter obtained experimentally for varying
materials and mixed-mode ratios.
Fig. 6. Coupled 3D domain for CZM4-PL.
Fig. 7. Coupled 3D domain for CZM4-BK.

Based on Eqs. (20), or (21), the ultimate value of the mixed-mode relative displacement is computed as
" c !c #1=c
2 1 þ b2 KN b2 K T
g m;u ¼ þ ð22Þ
g m;max /N /T
for the PL criterion, or

8 h 2 g i
< 2ð1þb2 Þ
/N þ ð/T /N Þ K bþbK2TK for g N > 0
g m;u ¼ ðK N þb2 K T Þgm;max N T ð23Þ
:
g Tu for g N 0
for the BK criterion. As can be inferred from Eqs. (18)–(23), this model requires seven parameters, e.g. the fracture energies
/N and /T , the cohesive strengths pNmax ; pTmax , the relative separations g Nmax ; g Tmax and the material parameters c or g
depending on the debonding propagation criterion.
This model defines the unloading behavior by means of the irreversibility conditions on the damage variables. The
interpenetration of the crack surfaces is avoided by imposing the following condition in mode I under compression
pN ¼ K N g N ; gN 0 ð24Þ
Thus contact is enforced independently from the debonding formulation.
3. Analytical assessment of the coupling effects
This section analyzes the mixed-mode behavior of an interface as predicted by the selected coupled CZMs. A detailed
assessment of the influence of the coupling on stresses and energy dissipation is performed to more deeply understand
the performance of each model. As mentioned earlier, an adequate coupling between the normal and tangential directions
is necessary in a CZM to describe the physically occurring interface behavior realistically. The coupling parameters, if not
well defined, may lead to unphysical predictions during the debonding process such as, for example, the existence of a resid-
ual non-zero traction or energy dissipation, when the load bearing capacity of the interface is lost [27]. It is also observed
that CZMs may or may not present inconsistencies depending on the selected input fracture parameters, which probably
explains why these inconsistencies have been mostly overlooked in the existing literature. A summary of the main features
of each CZM is provided in Table 1.
Table 1
Main features of the CZMs.
Advantages Disadvantages
CZM1 – Convex traction-relative displacement curves – Contact penetration error dependent on tangential separation
– Monotonic variation of the work of separation – Unloading not considered
– Path-independent failure domain – Inability to reach a complete failure under a critical energy
separation, due to the uncontrolled general shape
CZM2 – Absence of unphysical repulsive normal tractions – Non-convex traction-relative displacement curves
– Non monotonic variation of the work of separation
– Path-dependent failure domain
– Inability to reach a complete failure under a critical energy
separation, due to the uncontrolled general shape
– Unloading not considered
CZM3 – Ability to reach a complete failure under a critical energy – Non convex traction-relative displacement curves
separation, due to the controlled general shape
– Non monotonic variation of the work of separation
– Unloading not considered
CZM4 – Monolithic handling of loading and unloading – Non convex traction-relative displacement curves
– Ability to reach a complete failure under a critical energy – Non monotonic variation of the work of separation
separation, due to the controlled general shape
Proposed – Convex traction-relative displacement curves – Inability to reach a complete failure under a critical energy
CZM separation, due to the uncontrolled general shape
– Monotonic variation of the work of separation
– Path-independent failure domain
– Contact penetration error independent on tangential separation
– Monolithic handling of loading and unloading
3.1. Effect of coupling on the traction-separation curves
Figs. 8–15 show the dimensionless traction-separation curves of each cohesive model in the normal and tangential direc-
tions for varying degrees of mode mixity expressed by means of the quantities g T pTmax =/T and g N pNmax =/N . Pure mode I is
obtained by imposing g T pTmax =/T ¼ 0 while pure mode II corresponds to g N pNmax =/N ¼ 0.
The exponential CZMs 1 and 2 provide smooth traction-separation curves with continuous derivatives. However, they do
not allow for control of the initial stiffness in the ascending branch without affecting the cohesive strength. By using these
two models, convex domains are always obtained for varying mode mixities, as visible in Figs. 8 and 9b. It is reasonable to
expect a gradual reduction in the maximum of the pN =pNmax and pT =pTmax curves for increasing mode mixities, due to the
increasing non-zero separation in the other debonding mode. Conversely, CZM2 obtains for the mode I strength in mixed
mode values higher than the single-mode cohesive strength (Fig. 9a), which is physically inconsistent. The parameters used
in the figure are m ¼ 2; q ¼ 0:5; r ¼ 0, but the subsequent Figs. 10–12 show that a similar behavior is obtained for other
values. Therein, six other parameter sets have been chosen, based on the usual range of values considered in the literature
for m; q; r. In particular, q is varied from 0:43 to 2 following Abdul-Baqui and Van der Giessen [68], Xu and Needleman [18],
and Hattiangadi and Siegmund [69], whereas the combinations with m ¼ 1; m ¼ 2, and m ¼ 5 as well as r ¼ 0:3 and r ¼ 0:7
are chosen following McGarry et al. [49]. As can be seen in Fig. 12c, an unphysical repulsive normal traction can occur during
the initial normal separation when m ¼ 2; q ¼ 0:5 and r ¼ 0:3, whose magnitude increases for increasing mixed-mode con-
ditions. As already mentioned by McGarry et al. [49], the r parameter can cause only a shift of the normal traction curve
Fig. 8. Coupling effect on stresses for CZM1.
Fig. 9. Coupling effect on stresses for CZM2. m ¼ 2; q ¼ 0:5, r ¼ 0.

Fig. 10. Coupling effect on stresses for CZM2: effect of m. m ¼ 1; q ¼ 0:5; r ¼ 0 (a and b). m ¼ 5, q ¼ 0:5; r ¼ 0 (c and d).
without avoiding possible source of repulsive tractions. Non-zero values of r are not appropriate, and should be avoided in
any case, for numerical implementations.
The bilinear CZMs 3 and 4 enable a flexible control of the initial stiffness independently of the cohesive strength. On the
other hand, the derivative of the traction-separation curve may be discontinuous at the onset of the softening stage, depend-
ing on the degree of mode mixity (Figs. 13–15), which may create numerical issues in some circumstances. The traction-
separation curves of CZM3 are not always convex but show concave portions and/or hardening effects (Fig. 13a and b) under
some mixed-mode conditions. This also represents a physical inconsistency, as it is reasonable to expect a monotonic
decrease in the initial (elastic) stiffness as the degree of mode mixity increases. Concave regions are also obtained in the soft-
ening branch of CZM4-PL. This behavior is influenced by the cohesive parameters (i.e. g Nu =g Nmax , g Tu =g Tmax ), as visible from the
comparison between Fig. 14a, b and c, d. A further inconsistency can be noticed for CZM4-BK. In the advanced softening
stage, most mixed-mode curves in the normal and tangential directions feature stress values higher than those of the cor-
responding pure mode curve (see Fig. 15a, c and d). This leads to residual stresses when g N ¼ g Nu or g T ¼ g Tu in mode I or II
respectively, (i.e. when g N pNmax =/N or g T pTmax =/T reach the respective maximum values in pure modes), whose entity varies
with the material parameter g, as well as with the cohesive parameters g Nu =g Nmax and g Tu =g Tmax .
3.2. Effect of coupling on the energy dissipation
The behavior of the coupled CZMs is now studied by analyzing the normal, the tangential and the total work of separation,
W N , W T , and W, respectively, for different loading paths. A similar analysis was performed by van den Bosch [27] when
evaluating the consistency of the exponential model by Xu and Needleman [18]. Two non-proportional loading paths are
taken into account. In path 1, the interface is loaded first in the normal direction up to a predetermined normal displacement
Fig. 11. Coupling effect on stresses for CZM2: effect of q. m ¼ 2; q ¼ 0:43; r ¼ 0 (a and b). m ¼ 2, q ¼ 2; r ¼ 0 (c and d).
g N , and subsequently in the tangential direction up to failure (Fig. 16a). In path 2, the interface is first loaded in the tangential
direction up to g T , and then completely broken in the normal direction (Fig. 16b). In each case, the total work of separation is
computed as
Z Z
W ¼ WN þ WT ¼ pN ðg N ; g T Þdg N þ pT ðg N ; g T Þdg T ð25Þ
C C
where C is the selected separation path.

All the models are assumed to be characterized by the same fracture energy, cohesive strength, and peak relative
displacement in both modes. Three cases are analyzed, which differ for the relative magnitude of the fracture energies
(i.e. /N < /T ; /N ¼ /T ; /N > /T ). The fracture energies are set to 100 N/m or 200 N/m, as also assumed by van den Bosch
et al. [27], and Park et al. [48] for their parametric analyses. The cohesive strengths are both set to 6 N=mm2 . The values
of g Nmax and g Tmax in CZMs 3, 4 are set in such a way as to maintain the same secant stiffness at the peak defined by
CZM1. This leads to g Nmax ¼ 0:006 mm; g Tmax ¼ 0:020 mm when /N ¼ 100 N=m; /T ¼ 200 N=m (case 1); g Nmax ¼ 0:006 mm;
g Tmax ¼ 0:010 mm when /N ¼ /T ¼ 100 N=m (case 2); g Nmax ¼ 0:012 mm, g Tmax ¼ 0:010 mm when /N ¼ 200 N=m,
/T ¼ 100 N=m (case 3). In CZM4, a coupling parameter c ¼ 0:5 or g ¼ 1:75 is assumed for debonding propagation dictated
by the PL or the BK criterion, respectively.
3.2.1. Case 1 (/N < /T )

Here we consider /N ¼ 100 N=m and /T ¼ 200 N=m. The normal, the tangential and the total work of separation com-
puted under the two non-proportional loading paths in Fig. 16 are shown in Figs. 17 and 18. With CZM1, the limiting cases
Fig. 12. Coupling effect on stresses for CZM2: effect of r. m ¼ 2; q ¼ 0:5; r ¼ 0:7 (a and b). m ¼ 2, q ¼ 0:5; r ¼ 0.3 (c and d).
pffiffiffi
Fig. 13. Coupling effect on stresses for CZM3: g Nu =g Nmax ¼ 2e for pure mode I, g Tu =g Tmax ¼ 2 e for pure mode II.
pffiffiffi pffiffiffi
Fig. 14. Coupling effect on stresses for CZM4-PL: g Nu =g Nmax ¼ 2e, g Tu =g Tmax ¼ 2 e for pure modes I and II (a), (b); g Nu =g Nmax ¼ e, g Tu =g Tmax ¼ e for pure
modes I and II (c), (d). c ¼ 0:5.
of pure mode I and mode II failure are consistently captured for each loading path and the transition in between is smooth
and monotonic (Figs. 17a, and 18a). For non-proportional path 1, this means that W ¼ W T ¼ /T when g N ¼ 0 (i.e. loading is
completely driven by shear), and W ¼ W N ¼ /N when g N ! 1 (i.e. loading is completely driven by normal separation). A
monotonic decrease of W from /T to /N is correctly observed for intermediate values of g N (Fig. 17a). Conversely, for path
2, W ¼ W N ¼ /N when g T ¼ 0 and W ¼ W T ¼ /T when g T ! 1, while W monotonically increases from /N to /T (Fig. 18a).
These results were expected due to the corrections applied by van den Bosch et al. [27] to the exponential model by Xu
and Needleman [18].
When the same investigation is repeated for CZMs 2, 3, and 4, the limiting cases are consistently captured by the three
models, but the transition in between is not always monotonic (see Figs. 17c and 18c for CZM3, and Figs. 17d and 18d for
CZM4-PL). Monotonic variations of W are obtained for CZM2 under path 1 (Fig. 17b), as well as for CZM4-BK under both
paths (Figs. 17e and 18e).
For CZM2 under path 2, W N changes from /N to zero as g T increases from zero to g T , becoming negative after this value in
presence of repulsive tractions, while W T monotonically increases from zero to /T for increasing values of g T from zero to g Tu
(Fig. 18b). As a result, W is always constant at /N for all values of g T =g Tu .
3.2.2. Case 2 (/N ¼ /T )

Here /N ¼ /T ¼ 100 N=m. The normal, the tangential and the total work of separation as computed by means of each
model under non-proportional loading paths 1 and 2 are shown in Figs. 19 and 20, respectively. For both CZMs 1 and 2,
W N and W T vary by equal and opposite amounts with increasing values of g N (path 1), or g T (path 2), so that W remains
pffiffiffi pffiffiffi
Fig. 15. Coupling effect on stresses for CZM4-BK: g Nu =g Nmax ¼ 2e, g Tu =g Tmax ¼ 2 e for pure modes I and II (a), (b); g Nu =g Nmax ¼ e, g Tu =g Tmax ¼ e for pure
modes I and II (c), (d). g ¼ 1:75.
Fig. 16. Analyzed loading paths for the debonding process.
constant. The curves do not depend on the loading path for CZM1 (Figs. 19a and 20a) and are path-dependent for CZM2
(Figs. 19b and 20b).
Once again, CZMs 3 and 4 lead to inconsistent results, due to the non-monotonic transition of W between the limiting
cases (see Figs. 19c and 20c for CZM3, Figs. 19d and 20d for CZM4-PL, and Figs. 19e and 20e for CZM4-BK).
3.2.3. Case 3 (/N > /T )

In this case the values of /N and /T are set equal to 200 N=m and 100 N=m, respectively. As in Case 1, a path-dependency
of the energy dissipation is observed for each CZM. Figs. 21 and 22 illustrate W N ; W T and W as computed by means of each
250 250 250

WN WN WN
WT WT WT
W W W
200 200 200
WN_num WN_num WN_num
WN, WT, W [Jm-2]

WN, WT, W [Jm-2]
WT_num WT_num
WN, WT, W [Jm-2]
WT_num
W_num W_num W_num
150 150 150
100 100 100
50 50 50
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
gN/gNu gN/gNu gN/gNu
250 250
WN WN
WT WT
W W
200 200
WN_num WN_num
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]

WT_num WT_num
W_num W_num
150 150
100 100
50 50
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
gN/gNu gN/gNu
Fig. 17. Work of separation under non-proportional Path 1 (/N < /T ).
250 250 250

WN
WT
200
W
200 WN_num 200
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]
150
WN, WT, W [Jm-2]
WT_num
W_num
150 WN 150 WN
100
WT WT
W W
WN_num 50 WN_num
100 100
WT_num WT_num
W_num 0 W_num
0 0.2 0.4 0.6 0.8 1
50 50
-50
0 -100 0
0 0.2 0.4 0.6 0.8 1 gT/gTu 0 0.2 0.4 0.6 0.8 1
gT/gTu gT/gTu
250 250
200 200
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]
WN WN
150 WT 150 WT
W W
WN_num WN_num
WT_num WT_num
100 100
W_num W_num
50 50
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
gT/gTu gT/gTu
Fig. 18. Work of separation under non-proportional Path 2 (/N < /T ).
model under the two loading paths. The observations made earlier about the non-monotonic evolution of the work of
separation are applicable in this case to CZM2 (Fig. 21b) and CZM4-PL (Figs. 21d and 22d). Unlike in the first two cases,
here CZM3 is always consistent (Figs. 21c and 22c), whereas CZM4-BK does not dissipate all the fracture energy
(Fig. 22e). The other cases are correctly captured, as shown in Figs. 21a and 22a and b.
120 120 120
100 100 100

WN
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]

WN
WN, WT, W [Jm-2]

80 WT 80 WT 80
W W WN
WN_num WN_num WT
60 WT_num 60 WT_num 60 W
W_num W_num WN_num
WT_num
40 40 40 W_num
20 20 20
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
gN/gNu gN /gNu gN/gNu
120 120
100 100
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]

80 WN 80 WN
WT WT
W W
60 WN_num 60 WN_num
WT_num WT_num
W_num W_num
40 40
20 20
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
gN/gNu gN/gNu
Fig. 19. Work of separation under non-proportional Path 1 (/N ¼ /T ).
120 120 120
100 100 100

WN, WT, W [Jm-2]
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]
80 WN 80 WN 80 WN
WT WT WT
W W W
60 60 WN_num 60 WN_num
WN_num
WT_num WT_num WT_num
40 W_num W_num W_num
40 40
20 20 20
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
gT/gTu gT/gTu gT/gTu
120 120
100 100
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]
80 80 WN
WN WT
WT W
60 W 60 WN_num
WN_num WT_num
WT_num W_num
40 W_num 40
20 20
0 0
0 0.2 0.4 0.6 0.8 1 0 0.5 1 1.5 2
gT/gTu gT/gTu
Fig. 20. Work of separation under non-proportional Path 2 (/N ¼ /T ).
3.3. Effect of coupling on the failure domains
In the following, we present the analytical results in terms of W N –W T failure domains. As in the previous section, three
different combinations of fracture energies (i.e. /N < /T ; /N ¼ /T , /N > /T ) are considered and the path-dependency of the
failure domains is analyzed for each CZM.
200 250 200
200
150 WN WN 150 WN
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]

WT WT WT
W 150 W W
WN_num WN_num WN_num
100 WT_num WT_num 100 WT_num
W_num W_num W_num
100
50 50
50
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
gN/gN,u gN/gNu gN/gNu
250 250
WN WN
WT WT
W W
200 WN_num 200 WN_num
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]

WT_num WT_num
W_num W_num
150 150
100 100
50 50
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
gN/gNu gN/gNu
Fig. 21. Work of separation under non-proportional Path 1 (/N > /T ).
250 250 250

WN WT WN
WT W WT
W WN W
200 200 WN_num 200
WN_num WN_num
WT_num
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]
WT_num
WN, WT, W [Jm-2]
WT_num
W_num W_num
150 150 W_num
150
100 100 100
50 50 50
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.5 1 1.5 2
gT/gTu gT/gTu gT/gTu
250 250
WN WN
WT WT
W W
200 200 WN_num
WN_num
WT_num
WN, WT, W [Jm-2]
WN, WT, W [Jm-2]
WT_num
W_num W_num
150 150
100 100
50 50
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
gT/gT,u gT/gTu
Fig. 22. Work of separation under non-proportional Path 2 (/N > /T ).
These domains can be regarded as debonding failure criteria for mixed-mode loading, and can be obtained from the
computation of W N and W T along a predefined loading path with Eq. (25). E.g. with path 1, we can compute
Z g N

W N g N ¼ pN ðg N ; g T ¼ 0Þdg N ð26Þ
0
Z
1
W T g N ¼ pT ðg N ; g T Þdg T ð27Þ
0
and then eliminate g N among the two. In an analogous way for path 2 we can compute
Z g T

W T g T ¼ pT ðg N ¼ 0; g T Þdg T ð28Þ
0
Z
1
W N g T ¼ pN ðg N ; g T Þdg N ð29Þ
0
and then eliminate g T . The same procedure can be repeated for a more complicated non-proportional loading path such as
those shown in Fig. 23. For example, let us consider a mixed non-proportional loading path for which the interface is first
loaded in the normal direction up to a maximum normal displacement g N , then is loaded in the tangential direction up to
a maximum tangential displacement g T , and finally is completely broken in the normal direction (i.e. g N ! 1, see path 3
in Fig. 23a). Thus W N and W T read
Z g N

W N;1 g N ¼ pN ðg N ; g T ¼ 0Þdg N ð30Þ
0
Z g T

W T g N ; g T ¼ pT g N ; g T dg T ð31Þ
0
Z
1
W N;2 g T ¼ pN g N ; g T dg N ð32Þ
g N
with W N ¼ W N;1 þ W N;2 .

The above computation for path 1 applied to CZM1, also considering Eqs. (3) and (4), leads to
/N
W N ¼ /N WT ð33Þ
/T
which clearly corresponds to a linear failure domain. An identical result is obtained from the computations for path 2, as well
as (with some additional manipulations) for paths 3 and 4 in Fig. 23. The detailed calculations are here omitted for brevity.
Thus a linear and path-independent failure criterion is determined by applying CZM1 (see Fig. 24).
Conversely, all the other CZMs lead to failure domains which depend on the loading path (see Figs. 25–28). Fig. 25a–c
display the failure domains as given by CZM2 for the three cases /N < /T ; /N ¼ /T ; /N > /T . For the case /N ¼ /T mixed-
mode conditions are always limited by a linear failure domain (Fig. 25b), however, under non-proportional path 2, only
about one-half of the fracture energy in pure mode I can be dissipated. In addition, the failure domains under path 1 are
not convex for all mode mixities (see Fig. 25a) and feature other inconsistencies such as kinks due to the abrupt variation
in the normal work of separation W N when W T reaches the tangential fracture energy /T (Fig. 25c for loading path 1), or
values of W T larger than the pure mode II fracture energy /T and thus unrealistic (Fig. 25c for loading path 2). These mis-
matches are due to the energetic effects already shown in Fig. 18b for W T , whose value drops to zero for g T < g Tu , and in
Fig. 22b for W N , whose value never attains zero in pure mode conditions.
A similarly significant influence of the loading paths and cohesive parameters on the local failure domains is also obtained
from the bilinear CZMs 3 and 4. Once again, the failure domains are not generally convex, but can present more or less pro-
nounced concavities for wide ranges of mode mixities (see Figs. 26–28). While in the limit cases of single-mode conditions it
Fig. 23. Additional non-proportional loading paths for the debonding process.
200 200 200

Analytical path 1 Analytical path 1 Analytical path 1
Numerical path 1 Numerical path 1 Numerical path 1
150 Numerical path 2 150 Numerical path 2 150 Numerical path 2
WN [Jm-2]
WN [Jm-2]
WN [Jm-2]
100 100 100
50 50 50
0 0 0
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200
WT [Jm-2] WT [Jm-2] WT [Jm-2]
Fig. 24. Fracture domains for CZM1.
200 200 200

WN [Jm-2]
WN [Jm-2]
WN [Jm-2]
100 100 100
50 50 50
0 0 0
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200
200 200 200

WN [Jm-2]
WN [Jm-2]
WN [Jm-2]
100 100 100
50 50 50
0 0 0
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200
is always W N ¼ /N or W T ¼ /T , close to pure mode II conditions there are instances where W T is larger than /T depending on
the loading path.
4. Numerical examples
In this section we present some examples where the selected coupled CZMs have been implemented numerically with the
finite element method (FEM). Several different ways exist to incorporate a CZM in a finite element setting, e.g. using interface
elements [11,70,71] or generalized contact elements [72]. More recently, cohesive zone approaches have been proposed
within extended FEM [73–75] and embedded mesh formulations [76,77], or using isogeometric discretizations [78–81].
The latter have shown several advantages, due to the higher order continuity conditions achieved.
However, the main focus of this work is not on the numerical aspects but on the performance of different models when
simulating mixed-mode fracture processes. Thus, a simple contact formulation based on the node-to-segment (NTS) strategy
and generalized to handle cohesive forces in both the normal and tangential directions is considered, as employed in
200 200 200

Analytical Path 2 Analytical path 2 Analytical Path 2
WN [Jm-2]
WN [Jm-2]
WN [Jm-2]
100 100 100
50 50 50
0 0 0
0 50 100 150 200 0 20 40 60 80 100 0 50 100 150 200
Fig. 27. Fracture domains for CZM4-PL.
200 200 200

WN [Jm-2]
WN [Jm-2]
WN [Jm-2]
100 100 100
50 50 50
0 0 0
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200
Fig. 28. Fracture domains for CZM4-BK.
Wriggers et al. [72]. Three numerical examples are analyzed: a simple patch test, a peel test under mixed-mode loading con-
ditions, and the standard mixed-mode bending test.
4.1. Numerical model
In the numerical model, all the interface laws analytically described in the previous sections have been implemented into
the generalized NTS contact element. In the normal direction under compression the non-penetration condition is enforced
using the penalty method. Depending on the contact status, an automatic switching procedure is used to choose between
cohesive and contact models. Each element contribution for the cohesive and contact forces is added to the global virtual
work equation as follows
dW C ¼ pN dg N þ pT dg T ð34Þ
where dW C is the contact contribution to the virtual work. The non-linear problem is solved with an iterative Newton–
Raphson procedure, where the global tangent stiffness matrix is obtained through the consistent linearization of the
Eq. (34). The model is implemented in the finite element code FEAP (courtesy of R. L. Taylor).
4.2. Mixed-mode patch test
A mixed-mode patch test (see Fig. 29) was first carried out to check the correctness of the numerical implementation.
A 10 10 mm2 square is pulled apart by applying uniform vertical and horizontal displacements (v and u) to the top and
right sides, respectively, until complete failure. Plane-stress 4-node isoparametric elements are used, with an elastic
modulus E ¼ 150 GPa and a Poisson’s ratio m ¼ 0. W N ; W T , and W are evaluated once again for the non-proportional loading
paths 1 and 2, by varying the mixed-mode conditions between pure modes I and II in terms of u and v. Very good agreement
between the analytical and numerical results is obtained in terms of tangential, normal and total work of separation, as
visible from Figs. 17–22 and 24–28.
4.3. Mixed-mode peel test
The second example is the classical peel test, where a thin elastic adherend with thickness t bonded to a substrate for a
length L (t ¼ 0:165 mm;L ¼ 100 mm) is peeled away at a certain angle h, named ‘‘peel angle”, and the peeling force needed to
Fig. 29. Geometry of mode I (a), mode II (b) and mixed-mode (c) patch test.
produce debonding is measured (Fig. 30). In this configuration the interface is subjected to shear and normal stresses, leading
to mixed-mode fracture. The main objective is to evaluate the performance of the models and the effect of their inconsistent
features on the macroscopic behavior in terms of load–displacement curves for different mixed-mode conditions. The adher-
end is modeled with two-dimensional, finite deformation, linearly elastic beam elements (E ¼ 250 GPa; m ¼ 0), while the
substrate is discretized with plane-stress 4-node isoparametric elastic elements. The substrate elements are characterized
by a very large elastic modulus so that the effect of the substrate compliance on results is negligible. The test is performed
in displacement-control mode. The peel angle is maintained constant during the whole loading process, as also described in
De Lorenzis and Zavarise [29]. After an initial transient, a steady-state peeling phase is reached whereby the adherend grad-
ually debonds from the substrate. Once again, three possible combinations of normal and tangential single-mode fracture
energies are considered, adopting the following values: /N ¼ 100 N=m; /T ¼ 200 N=m; /N ¼ /T ¼ 100 N=m; /N ¼ 200 N=m,
/T ¼ 100 N=m. The cohesive strenghts are pNmax ¼ pTmax ¼ 6 MPa. The peel angle is varied between 0 and 10 , in order to
induce different mixed-mode conditions. Figs. 31–33 report the numerical load–displacement curves for different peel
angles. Both load and displacement refer to the direction given by the peel angle. The curves present an ascending branch
up to the peak force, followed by a plateau representing the steady-state peeling phase. The corresponding force represents
the steady-state peeling force, F peel , whose difference with respect to the peak load is more pronounced for larger peel angles.
In addition to some differences in the elastic slope and peak force stemming from the different shapes of the various models,
significant differences can be observed in the steady-state peeling force. This is also evidenced in Fig. 34a–c, where F peel is
plotted vs. the peel angle.
As expected from analytical predictions based on linear elastic fracture mechanics (LEFM) [29], F peel should always
decrease monotonically with the increase of the peel angle from mode II (for h ¼ 0 ) to prevailing mode I conditions (for
h ¼ 10 , the largest angle analyzed). This is verified for CZMs 1, 3 and 4-BK but not for CZMs 2 and 4-PL (Fig. 34a–c). In these
cases the peeling load is not always monotonic but remains constant or increases for increasing peel angles between h ¼ 0
and h ¼ 4 (Fig. 34c). It is interesting to note that CZM1 predictions are always in excellent agreement with analytical LEFM
results, while more or less significant discrepancies are observed with the other CZMs. Independently from the mode I to
mode II fracture energy ratios, the largest differences are obtained with CZM4-PL. The numerical values of the peeling load
are lower than expected from LEFM for all the peel angles, thus reflecting the local inconsistencies in the energy dissipation
observed earlier.
Fig. 30. Schematic representation of the peel test.

Fig. 31. Mixed-mode peel test: load–displacement response for /N < /T . r ¼ 0; m ¼ 2; c ¼ 0:5; g ¼ 1:75.
4.4. Mixed-mode bending test
The third example is a standard mixed-mode bending (MMB) test, originally developed by Reeder and Crews [51] in order
to measure the energy dissipation during debonding over a wide range of mode I/mode II ratios. In this test, the DCB test for
mode I loading is combined with the end-notch flexure test for mode II loading, and any desired mixed-mode ratio can be
produced by varying the length of the lever c (Fig. 35). Pure mode II is obtained when the applied load is directly above the
beam midspan (c ¼ 0 mm), while pure mode I is obtained by removing the beam and pulling on the hinge [51]. The analyses
herein consider an MMB specimen with length 2L ¼ 102 mm, width B ¼ 25:4 mm, thickness h ¼ 1:56 mm and precrack
length a0 ¼ 33:7 mm, in accordance with the standard [82], see Fig. 35. The cohesive zone where energy is dissipated, named
as ‘‘process zone”, is defined by its length, lpz , as shown in Fig. 35. Plane-stress 4-node isoparametric elements are used to
model the two adherends, which are characterized by an elastic modulus E ¼ 122 GPa and a Poisson’s ratio m ¼ 0:25. Once
again /N is taken smaller, equal or larger than /T . A sufficiently fine discretization of the fracture process zone is needed to
limit the iterative convergence problems due to possible unphysical (numerical) snap-through or snap-back branches in the
load–deflection response. Moreover, the mode mixity at the debonding front is varied by increasing the distance c from 0
(pure mode II) to 130 mm (pure mode I) by steps of 10 mm. The main differences in the load–displacement curves obtained
with the four selected CZMs are assessed. The numerical results are also compared to the analytical solution based on elastic
beam theory (during initial elastic loading) and LEFM (during debonding) as provided by Mi et al. [83]. More details about
the closed-form analytical formulation can be found in Park et al. [48].
The results corresponding to the different cohesive models in terms of load vs. displacement curves are shown in
Fig. 36a–c for four sample mode mixities (c ¼ 50; 70; 90; 110 mm). The curves present an ascending elastic branch up to
the peak force, followed by a descending and by a new ascending branch corresponding to the debonding phases for
a < L and a > L, respectively. Note that it is reasonable to expect that the ascending elastic branch has a stiffness correspond-
ing to the perfectly bonded beams, whereas the second ascending branch during the debonding phase should asymptotically
approach the straight line corresponding to the stiffness of the upper beam only.
For the selected mode mixities and cohesive parameters, the elastic range of the curves is almost unaffected by the initial
stiffness of the CZM, whereas the peak load and the softening evolution are very sensitive to the choice of the model. As
Fig. 32. Mixed-mode peel test: load–displacement response for /N ¼ /T . r ¼ 0; m ¼ 2; c ¼ 0:5; g ¼ 1:75.
shown in Fig. 36a–c, many numerical curves present unexpected features. CZM2 leads to hardening branches in the
ascending part of the load–displacement curves, which in turn lead to failure of the Newton–Raphson iterative scheme
(Fig. 36a–c). For CZM4-PL, an oscillatory behavior is registered in the descending branch of the curves (Fig. 36a and b for
c ¼ 50; 70; 90 mm), or in some cases convergence is lost (Fig. 36a and b for c ¼ 110 mm). For CZM4-BK with /N > /T
(Fig. 36c), an overestimation of the debonding load is visible from the significant distance between the numerical load–
displacement curve and the asymptotic line at complete debonding, which reflects the incomplete dissipation of the work
of separation already shown in Fig. 22. The curves obtained with CZM3 match quite well the reference curves of CZM1, at
least for the selected mode mixities. Results of this example illustrate quite clearly that local inconsistencies of cohesive
models can significantly affect the global description of the mechanics of interfaces for varying mode mixities, which
underlines the importance of using a consistent model in numerical applications.
5. Thermodynamically consistent reformulation of the exponential model by Xu and Needleman [18] as modified by
van den Bosch et al. [27]
5.1. Introduction
As shown in the previous sections, CZM1 is the only one of the selected models not leading to inconsistencies in the
behavior of interfaces subjected to mixed-mode loading. This model allows for different values of the fracture energy in
the normal and tangential directions, as obtained experimentally. However, the model as formulated originally from van
den Bosch et al. [27] is not based on a potential. As pointed out in Mosler and Scheider [60], non-potential-based models
do not necessarily satisfy the symmetry requirements for the tangent matrix which have to be verified at least for elastic
unloading between interfaces, and introduce independent formulations for loading and unloading (in fact, as mentioned
earlier, unloading was not explicitly dealt with in [27]). The last feature is in contrast with thermodynamical requirements,
for which loading and unloading are uniquely defined by means of the same Helmholtz energy [84]. As follows, CZM1 is
reformulated within a thermodynamical framework through the introduction of damage variables. Through this reformula-
tion, the model becomes thermodynamically consistent. Moreover, the reformulation is convenient for the numerical
implementation, as it allows for the introduction of a unique model for loading and unloading, as well as for debonding
Fig. 33. Mixed-mode peel test: load–displacement response for /N > /T . r ¼ 0; m ¼ 2; c ¼ 0:5; g ¼ 1:75.
and contact conditions, which does not come naturally with the original formulation. Through the appropriate choice of the
evolution laws for the damage variables, the reformulated model gives a debonding behavior identical to the model by van
den Bosch et al. [27]. Conversely, a very simple modification is made to the contact formulation, which prevents the inac-
curate penalization of crack penetration under simultaneous tangential loading mentioned earlier (Section 2.1).
Note that, although the numerical applications reported in this work are limited to a 2D setting, the following formulation
is valid in general 3D conditions if we take g T :¼ kgT k, gT being the tangential relative displacement vector.
5.2. The formulation
A predefined Helmholtz energy is introduced, and the inelastic nature of the decohesion process is accounted for by
means of damage variables. The postulated Helmholtz energy contains the decomposition of decohesion and closure of
the crack surfaces, as follows
ðNÞ ðTÞ ðNÞ ðTÞ
w ¼ ð1 dN Þð1 dN ÞwþN ðg N Þ þ wN ðg N Þ þ ð1 dT Þð1 dT ÞwT ðg T Þ ð35Þ
Here the first two contributions are associated with mode I and the last one with mode II. wþ
N ; wN and wT are the elastic ener-
gies associated to the normal tensile, normal compressive and tangential contributions, respectively. Based on a quadratic
assumption, these can be expressed as
1
wþN ðg N Þ ¼ kN hg N i2 ð36Þ
2
1
wN ðg N Þ ¼ kP ðg N hg N iÞ2 ð37Þ
2
1
wT ðg T Þ ¼ kT g 2T ð38Þ
2
The split of the normal component of the energy into a positive and a negative part allows for monolithic handling of
decohesion and contact. kN in the Eq. (36) refers to the undamaged stiffness in the normal direction for g N > 0, while kP
140
LEFM
120 CZM1
CZM2
100
CZM3
Fpeel (N/mm)
80 CZM4-PL
CZM4-BK
60
40
20
0
0 2 4 6 8 10
Peel Angle θ (deg)
140
LEFM
120 CZM1
CZM2
100
CZM3
Fpeel (N/mm)
80 CZM4-PL
CZM4-BK
60
40
20
0
0 2 4 6 8 10
Peel Angle θ (deg)
140
LEFM
120 CZM1
CZM2
100
CZM3
Fpeel (N/mm)
80 CZM4-PL
CZM4-BK
60
40
20
0
0 2 4 6 8 10
Peel Angle θ (deg)
Fig. 34. Steady-state force vs. peel angle. r ¼ 0; m ¼ 2; c ¼ 0:5; g ¼ 1:75.
in Eq. (37) is the penalty contact stiffness for g N < 0. The degradation by means of the damage variables acts only on the
separation part wþ N such that the resistance to crack closure is maintained during interface failure.
ðjÞ
The material damage is modeled through a set of scalar-valued damage parameters di 2 ½0; 1, with i ¼ N; T and j ¼ N; T,
where different damage mechanisms are coupled multiplicatively. The damage variables increase monotonically from zero
in the undamaged case, up to one, which corresponds to the fully damaged case. By imposing that the damage variables are
non-decreasing as conventionally done in damage models, loading and unloading conditions are automatically dealt with
monolithically.
By applying the Coleman and Noll procedure the traction vector T becomes
ðNÞ ðTÞ @wþN ðg N Þ @wN ðg N Þ ðNÞ ðTÞ @w ðg Þ

T ¼ ð1 dN Þð1 dN Þ þ þ ð1 dT Þð1 dT Þ T T ð39Þ
@g @g @g
Fig. 35. Schematic representation of the MMB specimen.
Eqs. (36)–(38) lead to
@wþN ðg N Þ
¼ kN hg N in ð40Þ
@g
@wN ðg N Þ
¼ kP ðg N hg N iÞn ð41Þ
@g
@wT ðg T Þ
¼ kT ðg g N nÞ ¼ kT g T t ð42Þ
@g
where n and t are the (outward) normal and tangential unit vectors to the master surface at the current projection point
from the slave surface, respectively. The combination of Eqs. (39)–(42) gives
ðNÞ ðTÞ ðNÞ ðTÞ
T ¼ ð1 dN Þð1 dN ÞkN hg N in þ kP ðg N hg N iÞn þ ð1 dT Þð1 dT ÞkT g T t ð43Þ
or equivalently
T ¼ pN n þ pT t ð44Þ
From a comparison between Eqs. (43) and (44) the normal and tangential tractions are expressed as
N T
pN ¼ ð1 dN Þð1 dN ÞkN hg N i þ kP ðg N hg N iÞ ð45Þ
ðNÞ ðTÞ
pT ¼ ð1 dT Þð1 dT ÞkT g T ð46Þ
Based on the above general formulation, the thermodynamically consistent version of the exponential model by van den
Bosch et al. [27] can be obtained straightforwardly from the comparison between Eqs. (1), (2), (45), and (46). The same cohe-
sive models given by van den Bosch are recovered within the current framework if the evolution of the damage variables in
the normal and tangential directions is governed by the following equations
/N 2/T
kN ¼ ; kT ¼ ð47Þ
g 2Nmax g 2Tmax

ðNÞ g ðNÞ g g
dN ¼ 1 exp N ; dT ¼ 1 1 þ N exp N ð48Þ
g Nmax g Nmax g Nmax

ðTÞ ðTÞ g 2
dT ¼ dN ¼ 1 exp 2 T ð49Þ
g Tmax
Thus the reformulated model is identical to the model by van den Bosch et al. [27] in the debonding range. For negative
g N , a linear penalty branch with constant slope kP substitutes in the present model the variable non-linear branch used by Xu
and Needleman [18] and van den Bosch et al. [27]. This modification avoids the very large penetration errors arising in the
original models for simultaneous negative g N and tangential loading, and keeps the geometry restraint of non-penetration of
the crack surfaces decoupled from the mixed-mode debonding behavior. The only drawback of the present modification is
that a discontinuous slope is obtained at the boundary between debonding and contact regimes (for g N ¼ 0). However, this
feature induced no numerical difficulties in all the presented examples. The features of the proposed model are summarized
in Table 1.
100
c=50 LEFM
Numerical – CZM1
90 c=70
Numerical – CZM2
c=90
Numerical – CZM3
80 c=110 Numerical – CZM4-PL
Numerical – CZM4-BK
70
Load P [N]
60
50
40
30
20
10
0
0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4
Displacement Δ [mm]
100
c=50 LEFM
Numerical – CZM1
90 c=70
Numerical – CZM2
c=90
Numerical – CZM3
80 c=110
Numerical – CZM4-PL
Numerical – CZM4-BK
70
Load P [N]
60
50
40
30
20
10
0
0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4
LEFM
100 Numerical – CZM1
c=50 Numerical – CZM2
90 c=70 Numerical – CZM3
c=90 Numerical – CZM4-PL
80 Numerical – CZM4-BK
c=110
70
Load P [N]
60
50
40
30
20
10
0
0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4
Fig. 36. Load–displacement curves for /N < /T (a), /N ¼ /T (b) and /N > /T (c). r ¼ 0, m ¼ 2; c ¼ 0:5; g ¼ 1:75.
5.3. Numerical implementation and results
The thermodynamically consistent cohesive model proposed above has been implemented with the same numerical
setup illustrated earlier. Some numerical examples are presented as follows to demonstrate the performance of the reformu-
lated model, namely, a patch test, a matrix/particle debonding test, and a DCB test with uneven bending moments.
5.3.1. Patch test

As first example, we reconsider the patch test in Fig. 29 and use it to check the model performance, especially upon
unloading and in the transition between contact and debonding regions. A 10 10 mm2 square is subjected to imposed dis-
placements at its top and right sides. Subsequently the displacements are applied in opposite directions, leading first to
unloading and then to reversed loading until complete failure. Both applied displacements are equal to 0:05 mm in pure
modes I and II, whereas proportional values between horizontal and vertical displacement components are considered for
varying mode mixities. An elastic isotropic behavior is assumed for the plate, with material properties E ¼ 200 GPa and
m ¼ 0. The cohesive strengths pNmax and pTmax are assumed equal to 0:1 N=mm2 and 0:233 N=mm2 , respectively, whereas
kN ¼ pNmax =g Nmax ¼ 100 N=mm3 and kT ¼ pTmax =g Tmax ¼ 233 N=mm3 , which corresponds to maximum separations in the
normal and tangential directions g Nmax ¼ g Tmax ¼ 103 mm. The square is discretized with plane-stress 4-node isoparametric
elements, and the cohesive elements are introduced along the diagonal direction. Fig. 37a shows pN as a function of g N for
mode I. On the blue curve of Fig. 37a, the specimen is first loaded following the exponential law up to a value
pN ¼ 0:045 N=mm2 , then linearly unloaded along the secant stiffness to the origin, continuing in the compressive regime
up to g N ¼ 0:038 mm along the virgin stiffness to the curve. This corresponds to taking kP equal to the tangent stiffness
to the cohesive curve at the origin (in this case kP ¼ 271:8 N=mm3 ). The specimen is finally reloaded up to complete failure.
Local pressures in mode II and varying mode mixities (i.e. varying u=v ratios) are evaluated in a similar fashion by consid-
ering a complete cycle of loading, unloading up to compression, and reloading up to failure (see Fig. 37b, c and d). The red
curves of Fig. 37a and b as well as the black curves in Fig. 37c and d refer to a pure loading process in mixed mode conditions,
without any unloading or reloading steps, here taken as reference curves.
0.1 0.25
0.08
0.15
0.06
0.04
0.05
0.02
pN [MPa]
pT [MPa]
-0.004 -0.003 -0.002 -0.001 0 0.001 0.002 0.003 0.004
0 -0.05
-0.001 0.0005 0.002 0.0035 0.005 0.0065
-0.02
-0.15
-0.04
-0.06
-0.25
-0.08
-0.1 -0.35
gN [mm] gT [mm]
0.1 0.16
u=2v u=2v
0.08 u=1.8v u=1.8v
u=1.6v u=1.6v
0.06 u=1.4v u=1.4v
u=1.2v u=1.2v 0.08
u=v u=v
0.04
u=0.8v u=0.8v
u=0.6v u=0.6v
0.02 u=0.4v u=0.4v
pN [MPa]
pT [MPa]
u=0.2v u=0.2v
0 0
-0.0025 -0.001 0.0005 0.002 0.0035 0.005 0.0065 -0.0045 -0.003 -0.0015 0 0.0015 0.003 0.0045
-0.02
-0.04
-0.08
-0.06
-0.08
-0.1 -0.16
gN [mm] gT [mm]
Fig. 37. Computational results in mode I (a), mode II (b) and mixed-mode (c,d).
Fig. 38. Computational scheme.

Fig. 39. Homogenized stress–strain curve (a) and numerical shear stress contours at different points of the curve (b).
As also shown in the traction-separation curves of Fig. 37a and b, the maximum normal and tangential tractions pNmax and
pTmax are attained at the characteristic separations g Nmax and g Tmax , respectively, whereas a clear reduction in pure-mode
interface strengths is visible in Fig. 37c and d due to the non-zero separation history value in the other debonding mode.
5.3.2. Matrix/fiber mixed-mode debonding

The second example considers a fiber–matrix debonding process caused by transverse loads under the assumption of
plane stress. The problem geometry reduces, in the plane perpendicular to the particle axis, to a circular fiber with radius
a ¼ 2:5 mm surrounded by a matrix square with side b ¼ 10 mm, see Fig. 38. Both matrix and fiber are assumed to be lin-
early elastic, with material properties Ep ¼ 5 GPa; mp ¼ 0 for the fiber, and Em ¼ 500 MPa; mm ¼ 0:25 for the matrix. The cohe-
sive strengths pNmax and pTmax are assumed equal to 2 N=mm2 and 1 N=mm2 , respectively, whereas kN ¼ pNmax =g Nmax ¼
500 N=mm3 and kT ¼ pTmax =g Tmax ¼ 250 N=mm3 . The geometry shown can be thought of as the smallest element of a com-
posite in case of periodicity of the material microstructure. Fiber–matrix interface debonding is modeled by the reformulated
exponential CZM. The cohesive parameters are the same as in the previous example. Horizontal displacements are imposed
on the vertical boundaries of the cell as shown in Fig. 38, and the homogenized stress–strain response is numerically com-
puted. The average stress r is determined as the sum of the horizontal reactions divided by the length of the vertical cell side
b, while the average strain e is evaluated as the imposed relative displacement 2u divided by the original length of the hor-
izontal side of the cell b. The stress–strain history curve is reported in Fig. 39a, whereas Fig. 39b depicts some snapshots at
different time steps, including the shear stress contours as well as the reaction forces. In the initial loading phase, the particle
and the matrix are perfectly bonded throughout the interface (point A in Fig. 39) and the slope of the curve corresponds to
the elastic stiffness of the whole volume (matrix + fiber). Soon afterwards debonding starts taking place, leading to a strongly
nonlinear behavior. This is associated to the progressive decohesion of the interface, leading to two separate debonding
regions which are symmetric with respect to the x2 -axis (point B in Fig. 39). A gradual unloading is then applied by reversing
the boundary displacements with a subsequent behavior returning to linearity along the secant stiffness to the origin, for
g N > 0, and the virgin tangent stiffness to the origin, for g N < 0 (point C in Fig. 39). A further reloading is finally applied until
complete debonding between the particle and the surrounding matrix. In the final steps a linear behavior is observed with a
stiffness corresponding to that of the matrix alone, now fully detached from the particle (point D in Fig. 39).
5.3.3. A DCB test with uneven bending moments

As last example a DCB specimen is loaded in mixed-mode conditions through uneven bending moments (DCB-UBM), M 1
and M 2 (Fig. 40), on the two beams. The DCB-UBM test is often employed in experimental investigations on mixed-mode
fracture of sandwich specimens, e.g. specimens where two skin layers are joined by a core which is much thinner than
the other specimen dimensions [85]. A precrack of length a0 ¼ 20 mm is assumed on the left side of the sample between
the layers (Fig. 40). Each layer is discretized by 120 4 plane-stress 4-node elements in the x1 x2 plane. This level of refine-
ment gives a sufficiently accurate approximation of the bending behavior of the bi-layer sample during the debonding pro-
cess. An elastic isotropic behavior is assumed for both bodies, with an elastic modulus E ¼ 122 GPa and a Poisson’s ratio
m ¼ 0:25. The reformulated exponential CZM with cohesive parameters pNmax ¼ pTmax ¼ 4 N=mm2 , /N ¼ 100 N=m, and
/T ¼ 200 N=m is assigned to the interface. A full range of mode mixities can be realized by varying the M 1 =M 2 ratio between
1 (pure mode I) and 1 (pure mode II).
Fig. 40. DCB-UMB: Computational scheme.
150
M1/M2=-1
M1/M2=0.8
100
M1/M2=1
50
Moment
0
-0.3 -0.2 -0.1 0 0.1 0.2 0.3
-50
-100
-150
Rotation
Fig. 41. DCB-UMB: Moment-rotation curves.

The moment vs. rotation response of the composite specimen is depicted in Fig. 41 within a complete loading–unloading-
reloading cycle. Numerical simulations are performed for three different values of M 1 =M 2 indicating conditions of mode I
ðM1 =M 2 ¼ 1Þ, mode II ðM 1 =M 2 ¼ 1Þ and mixed-mode loading ðM 1 =M 2 ¼ 0:8Þ. The observed behavior in Fig. 41 displays
increasing moments in the early stages for small rotations, constant or decreasing moments during the debonding stage
for different mode mixities, and a return to linearity in the fully debonded stage. Unloading and reloading conditions are well
captured, which confirms once again the ability of the proposed exponential model to combine loading and unloading in a
monolithic fashion.
6. Conclusions
In this paper, we performed a consistency check for some widely used exponential and bilinear mixed-mode CZMs,
through a parametric analysis on the effect of the coupling parameters on cohesive stress-interface relative displacement
behavior, energy dissipation and mixed-mode failure domains corresponding to different loading paths. With the exception
of the model by Xu and Needleman [18] modified by van den Bosch [27], all the analyzed mixed-mode CZMs were found to
produce inconsistencies at the local level (i.e. at each point of the interface), including non-convex cohesive stress-relative
displacement curves, non-monotonic energy dissipation during non-proportional monotonic loading paths, incomplete
energy dissipation upon interface failure, and non-convex and loading path-dependent failure domains. The models were
implemented numerically within a finite element setting. Some numerical examples showed that the local inconsistencies
of the models may translate into undesirable features of the global behavior, including unrealistic hardening branches,
oscillations, loss of iterative convergence, and inability to reach the expected asymptotic branches at complete failure of
the interface. This underlines the importance of model consistency for an accurate prediction of interface phenomena in
challenging cases. While this paper necessarily dealt with a limited number of models, its analysis paradigm can be straight-
forwardly extended to any mixed-mode cohesive model.
Additionally, the model by Xu and Needleman [18] modified by van den Bosch [27] was reformulated herein within a
thermodynamically consistent framework, starting from an appropriate choice of the Helmholtz free energy function. The
chosen energy features additive normal and tangential quadratic contributions. A tension–compression split enables the
monolithic handling of decohesion and contact. Normal, tangential and coupling damage variables are introduced and the
irreversibility constraint on the damage variables enables the monolithic treatment of loading and unloading conditions,
not included in the original model. The evolution law is chosen in such a way to identically reproduce the original model
under monotonic loading conditions in the debonding range. For negative normal separations (i.e. for penetration of the
crack faces), a simple modification introduced herein prevents a penetration error dependent on the amount of simultaneous
tangential separation as observed in the original model. The proposed model is simple, thermodynamically consistent, and
therefore recommended for the solution of challenging mixed-mode interface problems.
Acknowledgements
The authors have received funding for this research from the European Research Council under the European Union’s
Seventh Framework Programme (FP7/2007-2013), ERC Starting Researcher Grant ‘‘INTERFACES”, Grant Agreement No.
279439.
Appendix A. Physical meaning of the r coupling parameter
Herein we give additional details about the physical meaning of r. This coupling parameter was introduced the first time
by Xu and Needleman [18], to control the traction-free normal separation bg N after a complete shear separation. Based on the
Xu and Needleman model, the normal tractions are defined as

/N g gN g2 1q g g2
pN ¼ exp N exp 2 T þ r N 1 exp 2 T
g Nmax g Nmax g Nmax g Tmax r1 g Nmax g Tmax
which becomes

/N g 1q g
pN ¼ exp N r N
g Nmax g Nmax r1 g Nmax
under a complete shear separation, i.e. for g T ! 1.
The coupling parameter r is then determined by imposing the previous equation equal to zero, which gives
b
gN
r¼
g Nmax
It was observed by Abdul-Baqui and Van der Giessen [86] that realistic penalization of normal over-closure from a phys-
ical point of view can be determined only when r P q. However, many studies that use the Xu and Needleman model set
r ¼ 0 [6,18,65,87] which results to be a reasonable choice as demonstrated numerically as follows by analyzing a diagonal
Fig. 42. Numerical determination of the r value.
patch test under a mode II loading process (see the scheme in Fig. 29b). The same mechanical parameters are herein assumed
to define the material as previously done in the work for the same example, while the cohesive fracture energies defining the
Xu and Needleman model are /N ¼ 100 N=m, /T ¼ 200 N=m (i.e. q ¼ 1) for cohesive strenghts pNmax ¼ pTmax ¼ 6 MPa. As
shown in Fig. 42 and its close-up, a vanishing value of g N =g Nmax is attained after a complete shear separation, i.e. when
pT ¼ 0, which means that r ¼ 0 after a complete detachment in mode II conditions and confirms the feasibility of this
assumption from a physical point of view.
References
[1] Barenblatt GI. The formation of equilibrium cracks during brittle fracture: general ideas and hypotheses: axially-symmetric cracks. J Appl Math Mech
(PMM) 1959;23(3):622–36.
[2] Barenblatt GI. The mathematical theory of equilibrium cracks in brittle fracture. Adv Appl Mech 1962;7:55–129.
[3] Dugdale DS. Yielding of steel sheets containing slits. J Mech Phys Solids 1960;8:100–4.
[4] Hillerborg A, Modéer M, Petersson P. Analysis of crack formation and crack growth in concrete by means of fracture mechanics and finite elements.
Cem Concr Res 1976;6:773–82.
[5] Palmieri V, De Lorenzis L. Multiscale modeling of concrete and of the FRP-concrete interface. Engng Fract Mech 2014;131:150–75.
[6] Rahulkumar P, Jagota A, Bennison SJ, Saigal S. Cohesive element modeling of viscoelastic fracture: application to peel testing of polymers. Int J Solids
Struct 2000;37:1873–97.
[7] Jiang L, Nath C, Samuel J, Kapoor SG. Estimating the cohesive zone model parameters of carbon nanotube–polymer interface for machining simulations.
J Manuf Sci Engng 2014;136(3):031004.
[8] Siegmund T, Brocks W. A numerical study on the correlation between the work of separation and the dissipation rate in ductile fracture. Engng Fract
Mech 2000;67:139–54.
[9] Li H, Chandra N. Analysis of crack growth and crack-tip plasticity in ductile materials using cohesive zone models. Int J Plast 2003;19:849–82.
[10] Chen X, Deng X, Sutton MA, Zavattieri P. An inverse analysis of cohesive zone model parameter values for ductile crack growth simulations. Int J Mech
Sci 2014;79:206–15.
[11] Camacho GT, Ortiz M. Computational modeling of impact damage in brittle materials. Int J Solids Struct 1996;33:2899–938.
[12] Tvergaard V. Effect of fibre debonding in a whisker-reinforced metal. Mater Sci Engng A 1990;125:203–13.
[13] Espinosa HD, Dwivedi S, Lu HC. Modeling impact induced delamination of woven fiber reinforced composites with contact/cohesive laws. Comp
Methods Appl Mech Engng 2000;183(3–4):259–90.
[14] Freed Y, Banks-Sills L. A new cohesive zone model for mixed mode interface fracture in bimaterials. Engng Fract Mech 2008;75:4583–93.
[15] Yang QD, Schesser D, Niess M, Wright P, Mavrogordato MN, Sinclair I, Spearing SM, Cox BN. On crack initiation in notched, cross-plied polymer matrix
composites. J Mech Phys Soldis 2015;78:314–32.
[16] Foulk JW, Allen DH, Helms KLE. Formulation of a three-dimensional cohesive zone model for application to a finite element algorithm. Comp Methods
Appl Mech Engng 2000;183:51–66.
[17] Needleman A. An analysis of tensile decohesion along an interface. J Mech Phys Solids 1990;38:289–324.
[18] Xu XP, Needleman A. Void nucleation by inclusions debonding in a crystal matrix. Model Simul Mater Sci Engng 1993;1:111–32.
[19] Erarslan N, Williams DJ. Mixed-mode fracturing of rocks under static and cyclic loading. Rock Mech Rock Engng 2013;46(5):1035–52.
[20] Xu XP, Needleman A. Numerical simulation of fast crack growth in brittle solids. J Mech Phys Solids 1994;42:1397–434.
[21] Yang B, Mall S, Ravi-Chandar K. A cohesive zone model for fatigue crack growth in quasibrittle materials. Int J Solids Struct 2001;38:3927–44.
[22] Li H, Yuan H, Li X. Assessment of low cycle ftigue crack growth under mixed-mode loading conditions by using cohesive zone model. Int J Fatigue
2015;75:39–50.
[23] Camanho PP, Dávila CG, De Moura MF. Numerical simulation of mixed-mode progressive delamination in composite materials. J Compos Mater
2003;37(16):1415–38.
[24] Pantano A, Averill RC. A mesh-independent interface technology for simulation of mixed-mode delamination growth. Int J Solid Struct
2004;41:3809–31.
[25] Dávila CG, Rose CA, Camanho PP. A procedure for superposing linear cohesive laws to represent multiple damage mechanisms in the fracture of
composites. Int J Fract 2009;158:211–23.
[26] de Moura MFSF, Gonçalves JPM. Cohesive zone model for high-cycle fatigue of composite bonded joints under mixed-mode I + II loading. Engng Fract
Mech 2015;140:31–42.
[27] van den Bosch MJ, Schreurs PJG, Geers MGD. An improved description of the exponential Xu and Needleman cohesive zone law for mixed-mode
decohesion. Engng Fract Mech 2006;73:1220–34.
[28] Chen JF, Teng JG. Anchorage strength models for FRP and steel plates bonded to concrete. ASCE J Struct Engng 2001;127(7):784–91.
[29] De Lorenzis L, Zavarise G. Modeling of mixed-mode debonding in the peel test applied to superficial reinforcements. Int J Solid Struct
2008;45:5419–36.
[30] De Lorenzis L, Zavarise G. Cohesive zone modelling of interfacial stresses in plated beams. Int J Solid Struct 2009;46(24):4181–91.
[31] De Lorenzis L, Fernando D, Teng JG. Coupled mixed-mode cohesive zone modeling of interfacial debonding in plated beams. Int J Solid Struct
2013;50:2477–94.
[32] de Moura MFSF. Application of cohesive zone modeling to composite bonded repairs. J Adhes 2014;91:71–94.
[33] Fernando D, Yu T, Teng JG. Behavior and modeling of CFRP-strengthened rectangular steel tubes subjected to a transverse end bearing load. Int J Str
Stab Dyn 2015. http://dx.doi.org/10.1142/S0219455415400313.
[34] Shahbazpanahi S, Ali AAA, Kamgar A, Farzadnia N. Fracture mechanic modeling of fiber reinforced polymer shear-strengthened reinforced concrete
beam. Compos Part B – Engng 2015;68:113–20.
[35] Kafkalidis MS, Thouless MD. The effect of geometry and material properties on the fracture of single lap-shear joints. Int J Solid Struct
2002;39:4367–83.
[36] Yao J, Teng JG, Chen JF. Experimental study on FRP-to-concrete bonded joints. Composites: Part B 2005;36:99–113.
[37] Pan J, Leung CKY. Debonding along the FRP-concrete interface under combined pulling/peeling effects. Engng Fract Mech 2007;74:132–50.
[38] De Lorenzis L, Zavarise G. Interfacial stress analysis and prediction of debonding for a thin plate bonded to a curved substrate. Int J Non-Linear Mech
2009;44:358–70.
[39] De Lorenzis L, Zavarise G. Debonding analysis of thin plates from curved substrates. Engng Fract Mech 2010;77:3310–28.
[40] Li S, Thouless MD, Waas AM, Schroeder JA, Zavattieri PD. Mixed-mode cohesive-zone models for fracture of an adhesively bonded polymer–matrix
composite. Engng Fract Mech 2006;73:64–78.
[41] Tijssens MGA, Van der Giessen E, Sluys LJ. Modeling of crazing using a cohesive surface methodology. Mech Mater 2000;32:19–35.
[42] Tijssens MGA, Sluys LJ, Van der Giessen E. Simulation of fracture of cementitious composites with explicit modeling of microstructural features. Engng
Fract Mech 2001;68:1245–63.
[43] Hanson JH, Bittencourt TN, Ingraffea AR. Three-dimensional influence coefficient method for cohesive crack simulations. Engng Fract Mech
2004;71:2109–24.
[44] Li S, Thouless MD, Waas AM, Schroeder JA, Zavattieri PD. Use of cohesive-zone model to analyze the fracture of a fiber-reinforced polymer–matrix
composite. Compos Sci Technol 2005;65:537–49.
[45] Beltz GE, Rice JR. Dislocation nucleation versus cleavage decohesion at crack tips. In: Lowe TC, Rollett AD, Follansbee PS, Daehn GS, editors. Modeling
the deformation of crystalline solids presented. The minerals, metals & materials society (TMS). Cambridge, MA, USA: Havard University; 1991. p.
457–80.
[46] Needleman A. A continuum model for void nucleation by inclusion debonding. ASME J Appl Mech 1987;54:525–31.
[47] Tvergaard V, Hutchinson JW. The influence of plasticity on mixed mode interface toughness. J Mech Phys Solids 1993;41(6):1119–35.
[48] Park K, Paulino GH, Roesler JR. A unified potential-based cohesive model of mixed-mode fracture. J Mech Phys Solids 2009;57:891–908.
[49] McGarry JP, Máirtín ÉÓ, Parry G, Beltz GE. Potential-based and non-potential-based cohesive zone formulations under mixed-mode separation and
over-closure. Part I: Theoretical analysis. J Mech Phys Solids 2014;63:336–62.
[50] Banks-Sills L, Bortman Y. A mixed-mode fracture specimen: analysis and testing. Int J Fract 1986;30(3):181–201.
[51] Reeder JR, Crews Jr JH. Mixed-mode bending method for delamination testing. AIAA J 1990;28(7):1270–6.
[52] 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.
[53] Dollhofer J, Beckert W, Lauke B, Schneider K. Fracture mechanical characterisation of mixed-mode toughness of thermoplast/glass interfaces. Comput
Mater Sci 2000;19:223–8.
[54] Warrior NA, Pickett AK, Lourenço NSF. Mixed-mode delamination – experimental and numerical studies. Strain 2003;39:153–9.
[55] Carpinteri A, Ferrara G, Melchiorri G. Single edge notched specimen subjected to four point shear. An experimental investigation. In: Shah SP, Swartz
SE, Barr B, editors. International Conference on Recent Developments in the Fracture of Concrete and Rock. Elsevier Applied Science; 1989. p. 605–14.
[56] Cornec A, Scheider I, Schwalbe KH. On the practical application of the cohesive model. Engng Fract Mech 2003;70:1963–87.
[57] Scheider I, Brocks W. Simulation of cup-cone fracture using the cohesive model. Engng Fract Mech 2003;70:1943–61.
[58] Högberg JL. Mixed mode cohesive law. Int J Fract 2006;141:549–59.
[59] van den Bosch MJ, Schreurs PJG, Geers MGD. Identification and characterization of delamination in polymer coated metal sheet. J Mech Phys Solids
2008;56:3259–76.
[60] Mosler J, Scheider I. A thermodynamically and variationally consistent class of damage-type cohesive models. J Mech Phys Solids 2011;59:1647–68.
[61] Chandra N, Li H, Shet C, Ghonem H. Some issues in the application of cohesive zone models for metal–ceramic interfaces. Int J Solids Struct 2002;39
(10):2827–55.
[62] Volokh KY. Comparison between cohesive zone models. Commun Numer Methods Engng 2004;20:845–56.
[63] Alfano G. On the influence of the shape of the interface law on the application of cohesive-zone models. Compos Sci Technol 2006;66:723–30.
[64] Alfano M, Furgiuele F, Leonardi A, Maletta C, Paulino GH. Mode I fracture of adhesive joints using tailored cohesive zone models. Int J Fract
2009;157:193–204.
[65] Park K, Paulino GH. Cohesive zone models: a critical review of traction-separation relationships across fracture surfaces. Appl Mech Rev 2013;64(6).
http://dx.doi.org/10.1115/1.4023110. pp. 060802–060802-20. AMR-11-1036.
[66] Dimitri R, Trullo M, De Lorenzis L, Zavarise G. A consistency assessment of coupled cohesive zone models for mixed-mode debonding problems.
Frattura ed Integrità Strutturale 2014;29:266–83. http://dx.doi.org/10.3221/IGF-ESIS.29.2.
[67] Wu Jr EM, Reuter RC. Crack extension in fiberglass reinforced plastics. T. & AM report no. 275, University of Illinois; 1965.
[68] Abdul-Baqui A, Van Der Giessen E. Numerical analysis of indentation-induced cracking of brittle coatings on ductile substrates. Int J Solids Struct
2002;39:1427–42.
[69] Hattiangadi A, Siegmund T. An analysis of the delamination of an environmental protection coating under cyclic heat loads. Eur J Mech – A/Solids
2005;24:361–70.
[70] Schellekens JCJ, de Borst R. On the numerical integration of interface elements. Int J Numer Meth Engng 1993;36:43–66.
[71] Dogan F, Hadavinia H, Donchev T, Bhonge PS. Delamination of impacted composite structures by cohesive zone interface elements and tiebreak
contact. Cent Eur J Engng 2012;2(4):612–26.
[72] Wriggers P, Zavarise G, Zohdi TI. A computational study of interfacial debonding damage in fibrous composite materials. Comput Mater Sci
1998;12:39–56.
[73] Moës N, Belytschko T. Extended finite element method for cohesive crack growth. Engng Fract Mech 2002;69(7):813–33.
[74] Campilho RDSG, Banea MD, Pinto AMG, da Silva LFM, de Jesus AMP. Strength prediction of single-and double-lap joints by standard and extended finite
element modelling. Int J Adhes Adhes 2011;31:363–72.
[75] Benvenuti E, Tralli A. Simulation of finite-width process zone in concrete-like materials by means of a regularized extended finite element model.
Comp Mech 2012;50(4):479–97.
[76] Dolbow J, Harari I. An efficient finite element method for embedded interface problems. Int J Numer Methods Engng 2009;78(2):229–52.
[77] Gálvez JC, Planas J, Sancho JM, Reyes E, Cendón DA, Casati MJ. An embedded cohesive crack model for finite element analysis of quasi-brittle materials.
Engng Fract Mech 2013;109:369–86.
[78] Verhoosel CV, Scott MA, de Borst R, Hughes TJR. An isogeometric approach to cohesive zone modeling. Int J Numer Methods Engng 2011;87(1–
5):336–60.
[79] Dimitri R, De Lorenzis L, Wriggers P, Zavarise G. NURBS- and T-spline-based isogeometric cohesive zone modeling of interface debonding. Comput
Mech 2014;54:369–88.
[80] Dimitri R. Isogeometric treatment of large deformation contact and debonding problems with T-splines: a review. Curved Layer Struct 2015;2:59–90.
[81] Dimitri R, Zavarise G. T-splines discretizations for large deformation contact problems. PAMM (Proc Appl Math Mech), Gesellschaft für Angewandte
Mathematik und Mechanik, Wiley-VCH Verlag 2015;15 [in press].
[82] ASTM International. Standard test method for mixed mode I–mode II interlaminar fracture toughness of unidirectional fiber reinforced polymer matrix
composites. Technical report ASTM D 6671/D 6671M; 2006.
[83] Mi Y, Crisfield MA, Davies GAO, Hellweg HB. Progressive delamination using interface elements. J Compos Mater 1998;32(14):1246–72.
[84] Simo JC. Numerical analysis of classical plasticity. In: Ciarlet PG, Lions JJ, editors. Handbook for numerical analysis, vol. 4. Amsterdam: Elsevier; 1998.
[85] Sorensen FB, Jorgensen K, Jacobsen TK, Østergaard RC. DCB-specimen loaded with uneven bending moments. Int J Fract 2006;141(1–2):163–76.
[86] Abdul-Baqui A, Van der Giessen E. Indentation-induced interface delamination of a strong film on a ductile substrate. Thin Solid Films
2001;381:143–54.
[87] Zavattieri P, Hector Jr LG, Bower AF. Cohesive zone simulations of crack growth along a rough interface between two elastic–plastic solids. Engng Fract
Mech 2008;75(15):4309–32.

Mixed Cohesive Model PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Mixed Cohesive Model PDF

Caricato da

Copyright:

Formati disponibili

Engineering Fracture Mechanics 148 (2015) 145–179

Contents lists available at ScienceDirect

Engineering Fracture Mechanics

Coupled cohesive zone models for mixed-mode fracture:

⇑ Corresponding author. Tel.: +39 0832 297177.

2. Review of the selected cohesive zone models

Fig. 1. Coupled 3D domain for CZM1.

Fig. 2. CZMs 1 and 2.

2.2. Model by McGarry et al. [49]

Fig. 3. Coupled 3D domain for CZM2.

2.3. Model by Högberg [58]

Fig. 4. Coupled 3D domain for CZM3.

The fracture energies for pure modes I and II are given by

2.4. Model by Camanho et al. [23]

(see Fig. 5) with

Fig. 5. CZMs 3 and 4.

based on a quadratic failure criterion of the type

Fig. 6. Coupled 3D domain for CZM4-PL.

Fig. 7. Coupled 3D domain for CZM4-BK.

for the PL criterion, or

Thus contact is enforced independently from the debonding formulation.

3. Analytical assessment of the coupling effects

3.1. Effect of coupling on the traction-separation curves

Fig. 8. Coupling effect on stresses for CZM1.

Fig. 9. Coupling effect on stresses for CZM2. m ¼ 2; q ¼ 0:5, r ¼ 0.

3.2. Effect of coupling on the energy dissipation

where C is the selected separation path.

3.2.1. Case 1 (/N < /T )

3.2.2. Case 2 (/N ¼ /T )

Fig. 16. Analyzed loading paths for the debonding process.

3.2.3. Case 3 (/N > /T )

250 250 250

WN, WT, W [Jm-2]

100 100 100

WN, WT, W [Jm-2]

Fig. 17. Work of separation under non-proportional Path 1 (/N < /T ).

250 250 250

WN, WT, W [Jm-2]

Fig. 18. Work of separation under non-proportional Path 2 (/N < /T ).

120 120 120

100 100 100

WN, WT, W [Jm-2]

WN, WT, W [Jm-2]

WN, WT, W [Jm-2]

Fig. 19. Work of separation under non-proportional Path 1 (/N ¼ /T ).

120 120 120

100 100 100

WN, WT, W [Jm-2]

WN, WT, W [Jm-2]

WN, WT, W [Jm-2]

Fig. 20. Work of separation under non-proportional Path 2 (/N ¼ /T ).

3.3. Effect of coupling on the failure domains

200 250 200

WN, WT, W [Jm-2]

WN, WT, W [Jm-2]

WN, WT, W [Jm-2]

Fig. 21. Work of separation under non-proportional Path 1 (/N > /T ).

250 250 250

WN, WT, W [Jm-2]

100 100 100

WN, WT, W [Jm-2]

Fig. 22. Work of separation under non-proportional Path 2 (/N > /T ).

with W N ¼ W N;1 þ W N;2 .