Leclerc 2017

International Journal of Solids and Structures 108 (2017) 98–114
Contents lists available at ScienceDirect
International Journal of Solids and Structures

journal homepage: www.elsevier.com/locate/ijsolstr
On the suitability of a Discrete Element Method to simulate cracks

initiation and propagation in heterogeneous media
W. Leclerc a,∗, H. Haddad b, M. Guessasma a
a
Université de Picardie Jules Verne, LTI, EA 3899, F-02100 Saint-Quentin, France
b
Université de Bordeaux 1, I2M, CNRS UMR 5295, F-33405 Talence, France
a r t i c l e i n f o a b s t r a c t
Article history: The present paper investigates the suitability of a Discrete Element Method (DEM) to simulate cracks ini-
Received 24 June 2016 tiation and propagation in heterogeneous media. We focus our studies on the cohesive hybrid-particulate
Revised 3 November 2016
model in which the continuous medium is modeled using a cohesive beam model. Previous works ex-
Available online 18 November 2016
hibited the ability of such an approach to simulate the mechanical behavior of continuous materials un-
Keywords: der various solicitations. The DEM is actually well-suited to take into account local heterogeneities and
Discrete Element Method complex fracture phenomena occurring at very fine scales. In a first step, several tests are performed
Crack in the context of a 2D homogeneous medium in order to better quantify micro-macro scale transition
Composite materials effects related to the DEM on cracks initiation and propagation. Then, in a second step, unidirectional
Stress field fibre-reinforced composites are modeled using 2D models taking into account a brittle matrix failure and
Numerical modeling interfacial debonding effects. Numerical tensile tests are set up for two main configurations: a single-
fiber composite constituted of a single fiber embedded in an alumina matrix and the case of multi-fiber
composites constituted of parallel fibers also embedded in an alumina matrix. The results exhibit the
suitability of the DEM to yield suitable stress fields and crack patterns in the investigated heterogeneous
media. Scale transition effects are noticeable in terms of stress fields but turn out to be relatively limited
in the mechanism leading to cracks initiation and propagation. Furthermore, the competition between
debonding and failure is also well captured whatever the fiber arrangement.
© 2016 Elsevier Ltd. All rights reserved.
1. Introduction a scientific challenge of main interest. These aim to better charac-

terize the mechanical behavior of composites and the phenomena
Characterized by an excellent stiffness-to-weight ratio, a good arising in the material such as crack initiations, interfacial debond-
flexibility, a potential high strength and the ability to resist to cor- ing, defect effects, local variability and heterogeneity.
rosion, composite materials have become widely used in many in- From a numerical standpoint, several methods exist for simulat-
dustrial sectors such as aeronautic, aerospace, automotive, building ing cracks initiation and propagation. Most of them are based on
and marine. Besides, the development of biocomposites composed the Finite Element Method (FEM) which can be extended to em-
of a polymer matrix potentially derived from renewable resources bed the discontinuity. Thus, the eXtended Finite Element Method
and natural fibres as flax or hemp has paved the way to new ma- (XFEM) (Moës et al., 1999) was initially set up to tackle the diffi-
terials able to respond to current sustainability and environment culties related to the mesh updating and is now used for a larger
issues. However, in order to make them able to serve specific pur- scope of applications. However, such an approach is poorly adapted
poses in various situations, their performance and safety have to to model multiple fractures with joining and bifurcating cracks
be thoroughly monitored. Thus, the composite has to resist to time due to costly crack path enforcement. Other methods only allow
degradation and fatigue under a large scope of solicitations. This cracks to propagate between the elements. For example, the Vir-
requires an accurate multi-scale characterization of the material to tual Crack Closure Technique (VCCT) is often considered for mod-
better understand the process leading to damaging. For this pur- eling and predicting the delamination (Krueger, 2004). Other ap-
pose, in order to replace long and costly experimental measure- proaches such as the Cohesive Zone Model (CZM) (Xu and Needle-
ments, the development of robust numerical methods has become man, 1994; Zhang and Paulino, 2005) and its variants the Dis-
crete Cohesive Zone Model (DCZM) (Xie and Waas, 2006) and
the Discrete Damage Zone Model (DDZM) (Liu et al., 2012) give
∗
Corresponding author. Fax: +33 323 503 698. the possibility to consider random crack initiation and to manage
E-mail address: willy.leclerc@u-picardie.fr (W. Leclerc).
http://dx.doi.org/10.1016/j.ijsolstr.2016.11.015
0020-7683/© 2016 Elsevier Ltd. All rights reserved.
W. Leclerc et al. / International Journal of Solids and Structures 108 (2017) 98–114 99
multiple cracks propagation in the material. These methods are bending test and an indentation test are modeled and studied. In
mesh-dependent and require a costly remeshing to alleviate these each case, the variability is observed using stress fields determined
effects. Other approaches which can be qualified as meshless using Zhou’s formulation (Zhou, 2003) at the scale of the parti-
methods enable to consider crack propagation without represen- cle and using Love-Weber formulation (Nicot et al., 2013; Love,
tation of the crack’s topology. Thus, the cracking particles method 1927) at a mesoscale introduced with the help of a tessellation of
introduced by Rabczuk and Belytschko (20 04); 20 07) and later im- the domain in specimen (Nicot et al., 2013; Haddad et al., 2015).
proved to avoid local enrichment (Rabczuk et al., 2010) uses a con- In a second step, we investigate the case of composite materials.
tinuous description of the medium coupled to a discrete repre- Two main models are considered: a composite composed of a sin-
sentation of the discontinuity. Its advantage is to more naturally gle metallic fiber embedded in an alumina matrix and the case of
treat complex crack propagation but this is a priori less accurate parallel metallic fibers randomly embedded in an alumina matrix.
than methods based on embedded elements. Peridynamics (Silling, In this latter example, effects related to the arrangement of fibres
20 0 0) and its variants as the dual-horizon peridynamics (Ren et al., are also considered using closely-packed and dilute fiber arrays. In
2016) use a continuous description based on an integral formula- each configuration, tensile tests are modeled and the positive part
tion in order to avoid classical issues related to partial differen- of the hydrostatic stress field is observed before, during and after
tial equations. In this approach, forces applied to a particle are the crack initiation. For sake of clarity, only the matrix failure is
estimated by summing the contributions of a set of surrounding considered and fibers are supposed undamageable for each inves-
particles located in an area called horizon. Another class of meth- tigated case. Cracks propagation, variability effects, and the compe-
ods based on a particulate system also exists. Originally used in tition between matrix failure and interfacial debonding are finally
the field of rock mechanics (Cundall and Strack, 1979; Cleary and discussed.
Campbell, 1993), the DEM is an ideal tool for solving mechanical The paper is organized as follows. Section 2 is dedicated to
problems in which multiple scales and discontinuities arise. This the DEM based on the cohesive beam model which is designated
approach exhibits several similarities with the peridynamics since as the Cohesive Discrete Element Method (CDEM) in the sequel
the motion of a particle is also computed by summing the forces for more conciseness. Section 3 introduces the RDEF criterion de-
of neighboring particles. However, the DEM uses a discrete repre- scribing the matrix failure and the DDZM simulating the interfacial
sentation in the sense that all particles have to be modeled as in- debonding. Section 4 deals with the three numerical tests carried
dividual entities and requires to calibrate microscopic parameters. out in the context of a homogeneous medium. Finally, Section 5 is
In the last decade, the DEM was used in various domains dedicated to the modeling of tensile tests carried out on represen-
among them the simulation of tribological granular flows (Fillot tative patterns of composite materials.
et al., 2007), silo discharge (Nicot et al., 2013), impact effects on
concrete (Shiu et al., 2008), vickers indentation (Jebahi et al., 2013)
2. CDEM
and ball bearing loading (Machado et al., 2015). Besides, the DEM
was also considered for simulating cracks initiation and propaga-
The present section describes the CDEM as developed by
tion in continuous materials (Hentz et al., 2004; Tan et al., 2009;
Haddad et al. (2015) in 2D and André et al. (2012) in 3D. In this
André et al., 2013). In this domain, the most important challenge
approach a continuous medium is modeled by a granular pack-
was to develop a numerical model able to quantitatively repro-
ing consisted of disks in point contact in 2D (spheres in 3D) in
duce the multi-scale mechanical behavior of a continuous medium.
which the cohesion between two particles is itself modeled by a
Based on the works of Schlangen and Garboczi (1997), it was
beam element. Under several assumptions of size and isotropy, the
established that the cohesive beam model (André et al., 2012)
obtained hybrid particulate-lattice model can be designated as an
is more suitable than spring elements to model continuous me-
Equivalent Continuous Domain (ECD) the mechanical properties of
dia and its damaging under several mechanical loadings. Further-
which are related to those of beam elements using a calibration
more, recent works highlighted the ability of the cohesive beam
process.
model to simulate the mechanical behavior of heterogeneous ma-
terials throughout the examples of a unidirectional fibre compos-
ite (Maheo et al., 2015) and random media (Haddad et al., 2015). 2.1. ECD
The issue of cracks initiation and propagation in composites was
partially treated by Le et al. (2016) in the field of composite fibres. The CDEM is based on the discretization of a continuous
However, the authors only considered the cohesive beam model for medium by a granular packing composed of disks in point contact
modeling the matrix phase and more in-depth studies have to be in 2D. In the present work, granular packings are generated by the
performed to relate the mechanical behavior to the process leading efficient Lubachevsky–Stillinger Algorithm (LSA) (Lubachevsky and
to damaging and cracks propagation. Stillinger, 1990) which enables an accurate control of intrinsec pa-
In the present work, we aim to investigate the suitability of the rameters such as the compacity, the polydispersity, the coordina-
cohesive beam model to simulate cracks initiation and propagation number and the randomness of contact angles. Under several
tion in heterogeneous media. Besides, we intend to better under- assumptions related to these parameters, the granular domain can
stand the influence of variability effects related to the micro-macro be then considered as an ECD in that this is enough representative
scale transition in the Discrete Element (DE) approach on the dam- of the continuous medium. First, the compacity in terms of volume
aging initiation and the crack propagation. For this purpose, we fraction has to be closed to 0.85 which corresponds to the Random
consider the Removed DE Failure (RDEF) criterion introduced by Close Packing (RCP) (Torquato et al., 20 0 0; Donev et al., 20 04) for a
André et al. (2013). We suppose a brittle fracture with a mode I random granular packing composed of circular particles in 2D. Sec-
crack propagation so that damaging occurs when the hydrostatic ond, the coordination number which represents the average num-
stress is greater than a given stress limit. Interfacial debonding ef- ber of particles in contact with one given particle has to be close
fects are also taken into account by the DDZM introduced by Liu to 4.5. Third, a slight polydispersity of particle size must be intro-
et al. (2012) and here adapted for discrete simulations. In a first duced in order to avoid undesirable directional effects. Typically,
step, several tests are performed in the context of a 2D homo- the particle’s radius follows a Gaussian distribution law and the
geneous material in order to better exhibit the micro-macro scale dispersion is characterized by the coefficient of variation which is
transition effects on the damaging process and their influence on the ratio between the standard deviation and the average radius.
crack propagation. Thus, a plate with a centered hole, a 3-point For information purposes, this is set to 0.3 in the present work.
100 W. Leclerc et al. / International Journal of Solids and Structures 108 (2017) 98–114
where θ i and θ j are respectively the rotations of particles i and j.

i, j i, j
un and ut are respectively the normal and tangential displace-
ments. Kn and Kt are the classical normal and tangential parame-
ters respectively given by:
Eμ Aμ
Kn = (2)
Lμ
12Eμ Iμ
Kt = (3)
L3μ
The normal component acts as an attractive force, the tangen-

tial component allows to resist to the tangential relative displace-
ment and the moment component counteracts the bending motion
(André et al., 2012; Haddad et al., 2015). Practically, Aμ and Iμ are
related to a dimensionless parameter rμ ∈ ]0, 1] which reads:
Fig. 1. Example of a granular packing constituted of 500 particles. Links belong- 2h
ing to two adjoining triangles are depicted in green, those belonging to one ad- rμ = (4)
Ri + R j
joining triangle are depicted in blue and those added by the Delaunay process are
depicted in red. (For interpretation of the references to colour in this figure legend, where h is the height of the rectangular cross section and Ri and
the reader is referred to the web version of this article.)
Rj are respectively the radius of connected particles i and j. Thus,
Aμ and Iμ only depend on rμ and the size of particles, and the me-
chanical behavior of the beam is described by the only parameters
Eμ , rμ and the size of particles. However, this latter is connected to
the number of particles and has no effect on the macroscopic me-
chanical behavior for granular packings respecting the hypotheses
of the RCP which are checked out for a minimum of about 70 0 0
particles in the case of a 2D homogeneous domain (Haddad et al.,
2015). In the present approach, equations of motion for a parti-
cle i are derived from internal cohesive forces and the numerical
resolution is carried out by an explicit time integration with a for-
mulation based on a Verlet scheme.
Fig. 2. Beam model.
2.3. Calibration process

These three first parameters ensure the randomness of the gran-
ular packing and consequently the isotropy of the ECD. The net-
The main issue of the CDEM is that on the contrary of FE cal-
work of contacts is finally densified using the Delaunay triangula-
culations for which local properties at the scale of the element are
tion carried out in specific locations where the granular packing
identical to the macroscopic properties for a homogeneous mate-
exhibits some flaws such as voids or areas with a low coordination
rial, microscopic properties of the beam element (Eμ , rμ ) have to
number. These locations are detected considering the labelling il-
be correlated to macroscopic mechanical properties as done in pre-
lustrated on Fig. 1. A green link belongs to two adjoining triangles
vious works (André et al., 2012; Haddad et al., 2015). Such a pro-
and a blue one belongs to only one triangle (very rare links be-
cess is easily carried out in the hypothesis of a linear isotropic
longing to no triangle are deleted). Researched spots are delimited
elastic medium when parameters related to the arrangement of
by a loop of blue links in which the application of the Delaunay
particles as the compacity and the coordination number are set.
triangulation is limited. Such a process respects the initial network
In the present study, all granular packings meet the assumptions
of contacts since this is reduced to generate additional links within
of an RCP which is a suitable benchmark to ensure constant ar-
the loops. Besides, for the same reason, it turns out to be really ef-
rangement parameters and avoid directional effects. For more de-
ficient in comparison with a triangulation applied from the initial
tails, we invite the interested reader to have a look at the works of
cloud of particle’s centers. Thus, the coordination number comes
Kumar et al. (2016) about the effects of arrangement parameters
from about 4.5 to about 5.9 and about 10% of new links are gener-
on the macroscopic response. Thus, in the present study, two tests
ated.
are sufficient to fully characterize the mechanical behavior which
only depends on two independent coefficients. Typically, quasi-
2.2. Beam model static tensile and shear tests are performed using suitable bound-
ary conditions and the Young’s modulus EM and the Poisson’s ra-
The cohesive link between two particles is modeled by a beam tio ν M of the medium are extracted of numerical simulations for
element the intrinsec parameters of which are the length Lμ , the a given pair of microscopic parameters (Eμ , rμ ). The process is re-
Young’s modulus Eμ , the cross-section Aμ and the quadratic mo- peated for a full range of investigated configurations so that the
ment Iμ (see Fig. 2). Each cohesive link is controlled by a vector of evolution of microscopic properties allows us to choose the de-
three-component generalized forces acting as internal forces which sired macroscopic ones. In the present work, we consider a mi-
are given by the following system: croscopic Young’s modulus in the interval [2 GPa, 10 0 0 GPa], and
⎡ ⎤ ⎡ ⎤⎡ ⎤ an rμ parameter in the interval [0.1, 0.9]. Evolution curves are plot-
Fnj→i Kn 0 0 0 uin − unj ted in Fig. 3-a and -b. We notice that ν M does not depend on Eμ
⎢ ⎥ ⎢
⎢ j→i ⎥ ⎢ Kt Lμ Kt Lμ ⎥ ⎢ ⎥
⎥⎢ uti − utj ⎥ but quadratically depends on the dimensionless radius rμ . EM lin-
⎢F ⎥=⎢0 Kt
⎢ t ⎥ ⎣ 2 2 ⎥⎢ ⎥ (1) early depends on rμ and quadratically depends on Eμ . These con-
⎣ ⎦ ⎦⎣ θi ⎦
Kt Lμ Kt L2μ Kt L2μ clusions are in good agreement with those obtained by André et al.
0 θj
M j→i 2 3 6 (2012) in the context of spheres in 3D.
Fig. 3. Influence of the microscopic parameters Eμ and rμ on (a) EM and (b) the Poisson’s ratio ν M .
3. Failure criterion and interfacial debonding Due to their strong ability to resist in compression, the rupture of
fragile materials almost always occur in traction even when sub-
The crack propagation is modeled using the failure criterion in- mitted to external compression solicitations as in the case of an
troduced by André et al. (2013) which is based on the stress tensor indentation test. This prevents the authors to consider a classical
given by Zhou’s formulation (Zhou, 2003). For heterogeneous me- Von-Mises stress criterion since this is positive and do not discrim-
dia, an interfacial model is derived from the works of Liu et al. inate traction and compression areas. That is why our choice is to
(2012) which consider a discrete damage zone for the FEM. consider that the fracture occurs when the hydrostatic stress which
is only positive for local tensile solicitations is greater than a given
3.1. Stress tensor stress limit σ lim :
Trace(σi ) = 2σlim (7)
Stress fields within the ECD are obtained using the Zhou’s for-
mulation which relates the equivalent stress tensor σ i of a particle Practically, the hydrostatic stress field is estimated at the scale
i to the internal cohesion forces as follows: of the particle i. When the criterion is reached, cohesive links be-
1 tween the particle i and the set of particles Zi are deleted and the
σi = fi j di j (5)
2 i particle i does not interact with the granular system anymore.
j∈Zi
where fij is the cohesion force vector applied to the particle i by 3.3. Interfacial debonding
a particle j, dij is the relative position vector between particles i
and j and Zi is the set of particles linked to the particle i. i is the Interfacial debonding occurring at the interface between two
area of representation of the particle i which is approximated by phases of a heterogeneous medium is taken into account using the
the following expression: DDZM described and used by Liu et al. (2012) for FE calculations.
Ai During DE simulations some links connect two particles belonging
i = (6) to two different phases and consequently make a bridge between
φ
each domain throughout . The main idea consists in replacing the
where Ai is the real area of the particle and φ is the volume frac- beam element describing such a link by a discrete interface ele-
tion of the granular system which is 0.85 in the present work ment introduced as a spring-like element of normal stiffness K n.
where granular packings meet the assumptions of an RCP. Thus, In this model, the displacement u n is related to the normal bond-
the area of representation is easily given by the multiplication of ing force Fn as follows:
the real area by a constant coefficient equal to 1.176.
F
n = Kn un (8)
3.2. Failure criterion where u
n is the normal displacement of the spring element. Two
regimes are considered: a linear elastic loading and an exponential
In the present work, we consider the RDEF process developed softening one which are illustrated in Fig. 4. In the first regime,
,c
by André et al. (2013) which is based on the deletion of a DE when when u n < un where u
n
,c
is the critical displacement, K
n is con-
a given criterion is reached. This choice is motivated by previous stant and equal to a given K n
,0
stiffness:
studies (André et al., 2013; Maheo et al., 2015) which highlighted
the suitability of this approach to model complex crack patterns K ,c

n un < un = K
n
,0
(9)
in homogeneous (André et al., 2013) and heterogeneous (Maheo ,c
In the second regime, when u
n ≥ un , Kn explicitly depends on
et al., 2015) media in comparison with the classical bond-based ,c
un and un according to an exponential decrease so that K

n tends
Rankine criterion for beam elements. One must keep in mind that
to zero for high displacements:
the RDEF criterion do not respect the conservation of mass due to
the deletion of particles. Nevertheless, the mass loss is generally
1
K ,c
n un ≥ un = ,c
Kn ,0 (10)
quite limited with a number of deleted particles less than 1%. This u
n − un
point is particularly true for denser granular packings constituted exp
u
n
,c
of smaller particles. In the present work, we limit our studies to
the brittle fracture of fragile materials such as silica and alumina From a practical standpoint, the normal stiffness K
n only de-
with a mode I crack propagation. However, the present discrete ap- pends on two parameters, namely the linear stiffness Kn
,0
and the
,c
proach could also be envisaged for shear or mixed fracture mode. critical displacement un which is a priori unknown. However, this
Thus, from a theoretical standpoint, the maximum tensile stress

σxxmax which is located at top (A) and bottom (B) positions of the
hole is equal to 4.32 MPa. Before starting calculations, we first aim
to estimate the influence of the number of particles on the me-
chanical response. For this purpose, several granular packings con-
stituted from 500 to 50,0 0 0 particles are generated using the LSA
and the reaction force on the right edge as well as the potential
energy of the system are estimated. DE calculations are performed
using an adjustable time step which is determined as a function of
critical time steps tcrit
c associated to each beam element for an
explicit formulation as follows:
t = Dt min(tcrit
c
) (15)
c∈ζ
where ζ designates the set of contacts within the granular pack-

ing and Dt is a security ratio which is set to 0.04. Thus, for ex-
Fig. 4. F
n -un relationship. amples, t=3.5e−9 s for a granular packing composed of 10,0 0 0
particles and t=1.6e−8 s for 500 particles. Fig. 5a illustrates the
evolution of both the reaction force and the potential energy as a
latter can be determined knowing the interfacial stress limit σlim
function of the real time for a given granular packing. One can no-
using the following relation:
tice that the reaction force is perfectly equal to the applied loading.
σlim
A
This result is independent of the number of particles and the time
u ,c
n = (11) step. The potential energy follows the evolution of the force but
K
n
,0
the results depend on the number of particles. For example, the
where A is the area associated to the spring element which is potential energy tends to a value of 4.76e−5 J for 10,0 0 0 particles
,c
chosen equal to Aμ . Practically, the microscopic stress limit σlim when this is only 4.35e−5 J for 500 particles. Thus, the denser the

has to be calibrated to match the expected macroscopic value σlim . granular packing is and the higher the energy is. Besides, one can
,c notice that denser packings lead to oscillations clearly noticeable
For that purpose, several values of σlim are preliminary tested until
. for 50,0 0 0 particles. In fact, the greater the number of particles is
a good agreement is found between the observed result and σlim
and the lower Dt has to be chosen to avoid unstabilities. For infor-
Besides, u
n
,c
can also be determined using the critical fracture en- mation purposes, the kinetic energy which is not illustrated here
ergy release rate in mode I propagation GcI as follows: is constant and close to 1e−8 J which is negligible in comparison
GcI with the potential energy. In a second step, we consider the influ-
u ,c
n = (12) ence of the time step for a number of particles set to 10,0 0 0 par-
σlim
exp(1 )
ticles. This is sufficiently dense to well describe the macroscopic
mechanical behavior of a homogeneous medium according to re-
4. Validation tests for a homogeneous medium cent works of Haddad et al. (2015) and leads to acceptable energy
outputs. Fig. 5b illustrates the evolution of the potential energy as
The present section is dedicated to validation tests for a homo- a function of the real time for several time steps which are directly
geneous medium. We consider three cases, namely a plate with related to Dt parameter. Please notice that due to instabilities aris-
a hole, a beam in bending and an indentation test. Numerical re- ing for Dt > 0.25, the time step was knowingly limited to 2.19e−8 s
sults are compared to analytic solutions in terms of stress fields which corresponds to Dt = 0.25. One can observe that a smaller
and cracks propagation direction is discussed. time step leads to more unstable results but all results tend to the
same mean value of 4.76e−5 J.
4.1. Plate with hole According to previous results, DE calculations are now per-
formed using a time step set to 3.5e−9 s and a granular pack-
We consider the case of a rectangular thin plate with a centered ing constituted of 10,0 0 0 particles which are connected by about
hole. The diameter d of the hole is equal to 25 mm, the length l 29,0 0 0 cohesive links. Fig. 6a illustrates the σ xx stress field ob-
is 100 mm, the width b is 50 mm and the thickness h is 1 mm. tained at convergence for about 450,0 0 0 steps. From a qualitative
The plate consists of silica of Young’s modulus 72.5 GPa and Pois- standpoint, numerical outputs exhibit high stresses at the top and
son’s ratio 0.27 and the macroscopic stress limit is 64 MPa. In a bottom of the hole what is in accordance with our expectations.
first step, a quasi-static tensile strength of 50 N is applied in the From a quantitative standpoint, the CDEM seems to underestimate
x-direction according to a linear ramp of 20 0,0 0 0 steps to both stresses at A and B positions since the numerical estimates are
lateral sides of the plate. In such a configuration, the maximum 4.34 MPa and only 3.30 MPa respectively. Please notice that these
tensile stress σxx
max in the cross-section is given by the following values are not extracted from the exact positions of A and B but
expression: from the nearest DE. Besides, the numerical maximum stress value
which is 6.62 MPa is higher than the theoretical one. In fact, this
Ct F
σxxmax = (13) is not located at one of extremal positions but at a point of co-
( b − d )h ordinates (51.2 mm; 38.1 mm) close to A. Other calculations were
where Ct is the stress concentration factor which describes the ra- performed for several time steps and numbers of particles. Fig. 7a
tio between the nominal and the maximum tensile stress. This is shows the influence of the number of particles on both the maxi-
function of the geometry and is here given by the following rela- mum stress and the values extracted from A and B positions. One
tion (Pilkey, 2005): can notice that stress values are all clearly affected by the density
2 3 of the granular packing. A kind of convergence is noticeable from
d d d 10,0 0 0 particles even though some disruptions still remain for a
Ct = 3 − 3.14 + 3.667 − 1.527 (14)
b b b denser packing. However, the maximum stress value and this as-
Fig. 5. Influence of (a) the number of particles and b) the time step on both the reaction force and the potential energy.
Fig. 6. σ xx stress field of the plate with hole (a) without triangular specimen b) with triangular specimens.
Fig. 7. Influence of (a) the number of particles and b) the time step on both the maximum stress and the values extracted from A and B positions.
sociated to B position do not tend to theoretical estimates. Fig. 7b of the particle and only stress values are estimated at the scale of
depicts the evolution of both the maximum stress and the values the specimen using the Love-Weber formulation (Nicot et al., 2013;
extracted from A and B positions as a function of the time step. Love, 1927). This is very similar to Eq. (5) except the area of repre-
Results exhibit that stress values do not depend on this parameter. sentation is now the area of the triangular specimen and the cohe-
The gaps observed in terms of positions and stress values are sion forces are now exerted by external particles linked to one or
due to microscopic scale effects which lead to local variabilities more particles located within the specimen. The number of trian-
in a DE approach. These effects are noticeable whatever the time gles has to be chosen in order to sufficiently smooth the solution
step and the number of particles. For comparison purposes, a third and avoid a loss of informations due to an excessive smoothing.
mesoscopic scale is introduced using triangular specimens (Haddad Fig. 8 shows the influence of the number of specimens on both
et al., 2015) which can be seen as representative elements of the the maximum stress and the values extracted from A and B posi-
medium. The idea is to tessellate the domain in triangular ele- tions. In the present study, all numerical outputs are obtained for
ments, each of them containing several DE and to evaluate the a number of particles set to 10,0 0 0 particles and a time step set
stress at the scale of the specimen rather than the scale of the par- to 3.5e−9 s. Two conclusions can be done. A too small number of
ticle. Please notice that the model remains computed at the scale specimens leads to underestimated σ xx stress values. Conversely, a
the stress level is exactly the same for a bunch of particles belong-
ing to the same specimen which affects the continuity of the prop-
agation. A second crack initiates for a strength of about 880 N at
a point of coordinates (47.4 mm, 12.6 mm) located near B position
and propagates vertically. These findings exhibit that variability ef-
fects related to micro-macro scale transition in the DE approach
leads to only small variations on the position of cracks initiation.
However, the results also show that only the formulation of the
stress at the scale of the particle is able to reproduce a suitable
crack pattern in accordance with the assumption of brittle fracture
with a mode I crack propagation.(Fig. 10)
4.2. Three point bending test
We investigate the case of a three point bending test. We con-

sider a beam of effective length l = 72 mm and a rectangular sec-
Fig. 8. Influence of the number of specimens on both the maximum stress and the tion S = h × b = 4 × 1 mm2 composed of alumina of Young’s
values extracted from A and B positions. modulus E1 = 350 GPa and Poisson’s ratio ν 1 = 0.1. The gravity is
neglected and the beam is loaded in bending by a punctual force
F applied at the middle of the beam and simply supported on the
ends which are let free to rotate. Numerical simulations are car-
ried out using the CDEM and comparisons are performed in terms
of normal stress with Euler-Bernoulli theory. The supports and the
solicitation are modeled with the help of three large DE composed
of diamond of Young’s modulus E2 = 1050 GPa and Poisson’s ratio
ν 2 = 0.17. These have a radius of 4mm and are in point contact
with the 2D domain so that the total length L of the beam is cho-
sen equal to 80 mm. In this configuration, the punctual force is
applied directly to one large DE located at the middle of the upper
face of the beam and the solicitation is equally transmitted to two
Fig. 9. Fracture of the material (a) without triangular specimen b) with triangular others playing the role of supports. The contact between a large DE
specimens. and a small one is assumed to be elastic and ruled by the linear
Hertz theory for a cylinder-cylinder contact:
Fcn = Kcn ucn (16)
too dense tessellation leads to an overestimated maximum stress
value. Consequently, a good compromise has to be found in order where Fcn
is the normal reaction force, ucn is the interpenetration
to smooth the local perturbations observed for a dense tessellation between a large DE and a small one and Kcn is the normal stiffness
and avoid the loss of informations noticeable for a too small num- which is chosen as follows:
ber of specimens. Fig. 6b shows the σ xx stress field obtained using π E∗ b
a tessellation in 10 0 0 specimens which is a good compromise be- Kcn = 3 ∗ (17)
b πE
tween smoothing and accuracy. One can notice that the maximum ln
stress is now located at A position and its value is 4.30 MPa which FR∗
is very close to the theoretical reference. This highlights a good where E∗ is the equivalent Young’s modulus defined such that:
adequation between the numerical calculations and the analytical
formulation. 1 1 − ν12 1 − ν22
∗ = + (18)
In a second step, we investigate cracks initiation and propaga- E E1 E2
tion within the plate with a centered hole. A tensile strength is ap- and R∗ is the equivalent radius:
plied to both vertical sides of the plate according to the x-direction
2R1 R2
using a linear ramp of one million steps from 0 to 1,0 0 0 N. DE cal- R∗ = (19)
culations are performed using a time step of 3.5e−9 s and we con- R1 + R2
sider the failure criterion described in Section 3.2 with a hydro- In a first step, DE calculations are performed using a punctual
static stress limit of 50 MPa. In a first study σ xx stress is only es- force F=10N applied using a linear ramp of 10 0,0 0 0 time steps.
timated at the scale of the particle without triangular specimen. A In such a configuration, the Euler-Bernoulli beam theory gives the
crack initiates at the upper part of the domain near the top of the theoretical maximum normal stress σxx max = 3Fl/(2bh2 ) = 6.75e7 Pa.
hole at a position (51.1 mm, 38.6 mm) and propagates vertically up Before starting more in-depth investigations, we intend to deter-
to the upper edge of the plate. At this juncture, the corresponding mine the influence of dynamic effects which are more prone to oc-
tensile strength is close to 850 N. Then a second crack initiates cur in the present case due to cylinder-cylinder contacts between
from the bottom of the hole at the position (51.4 mm; 12.4 mm) large and small DE. For this purpose, the beam is discretized by
for a strength of about 870 N and propagates vertically up to lower a granular packing constituted of about 3200 DE and 9300 links.
edge of the plate (see Fig. 9a). In a second study σ xx stress is now The time step is set to 2.5e−9 s which corresponds to a Dt param-
estimated at the scale of the triangular specimen. We consider a eter equal to 0.05 which is quite close to the value considered in
tessellation of the domain in 1500 triangular elements. A first crack the case of the plate with hole. Fig. 11a illustrates the influence
initiates at a point of coordinates (49.8 mm, 37.3 mm) located near of the real time on both applied and reaction forces. Please notice
A position for a strength of about 870 N which is quite close to the that the applied force is estimated by summing the contributions
previous value. However, a branching arises and two cracks finally of small particles in contact with the large DE located on the up-
propagate as we can observe on Fig. 9b. In fact, in the present case, per face of the beam. Besides, the reaction force is estimated by
Fig. 10. Three point bending test.
Fig. 11. Influence of the real time on (a) both applied and reaction forces and b) both potential and kinetic energies.
Fig. 12. σ xx stress field of the beam (a) without triangular specimen b) with triangular specimens.
summing the contribution of each support. One can notice that the ical estimate given by Euler–Bernoulli beam model. However, the
applied force oscillates around a mean value of 10 N even though stress value at A position remains underestimated with a value of
this directly applied to the large DE is constant. A Rayleigh damp- 4.91e7 Pa. In the second case, the maximum σ xx stress is largely
ing model is used in the present approach in order to reduce these overestimated with a value of 8.85e7 Pa but the stress extracted
effects so that oscillations decrease in amplitude over time. The at A position which is 6.65e7 Pa is in good agreement with the
reaction force follows the same behavior with a larger amplitude. theoretical reference. Besides, the maximum value remains located
Fig. 11b shows the evolution of both potential and kinetic energy near A position in both configurations. These results highlight the
as a function of the real time. The potential energy follows the evo- variability effects related to micro-macro scale transition in a DE
lution of the applied force with the same decrease in amplitude approach which have been already observed in the previous case
over time so that it tends to a mean value of 1.81e−4 J. The kinetic of the plate with a hole. In order to smooth out these fluctua-
energy decreases over time and becomes negligible, i.e. less than tions, we consider a mescoscopic scale described by a tessellation
1e−8 J, for a real time higher than 5e−3 s. This value is the total of the domain in 400 triangular specimen each of them containing
duration necessary to avoid dynamic effects which could affect our about 8 particles. Fig. 12b illustrates the σ xx stress field given by
numerical outputs. DE calculations with the set of specimens and the granular packing
Fig. 12 a illustrates the σ xx stress field given by DE calculations composed of 3200 particles. Results exhibit lower extrema since
at the scale of the particle. Results exhibit an expected antisym- maximum and minimum values are respectively equal to 6.90e7 Pa
metric field in the cross-section of the beam with maximum and and −7.08e7 Pa which are quite close to the theoretical maximum
minimum values respectively equal to 7.58e7 Pa and −7.50e7 Pa stress given by Euler-Bernoulli beam theory. However, small fluc-
which slightly overestimate the theoretical reference. Besides, the tuations still remain since the normal stress at A position is now
maximum value is not located at A position but at another one 5.96e7 Pa and the maximum stress is obtained for a DE the center
of coordinates (40.56 mm; 0.19 mm) near A position where the of which has coordinates (38.85 mm; 0 mm). Other calculations
normal stress is only 4.68e7 Pa. For comparison purposes, other performed for 600 and 800 specimens exhibit similar results with
calculations are performed using two other granular packings re- a normal stress at A position respectively equal to 6.18e7 Pa and
spectively composed of 1600 and 8000 particles with a set Dt pa- 6.44e7 Pa. The maximum value is even higher, namely 9.40e8 Pa
rameter equal to 0.05. In the first case, the estimated maximum and 9.83e8 Pa respectively. In order to more reduce these scale
σ xx stress value is 6.67e7 Pa which is quite close to the theoret- effects, a larger number of particles within each specimen has to
Table 1
Properties of each phase.
Material Diamond Silica
Young’s modulus (GPa) 1050 72.5

Poisson’s ratio 0.1 0.17
Density (kg/m3 ) 3500 2300
σ lim (MPa) ∞ 50
Fig. 13. Rupture of the beam with vertical crack.
ness b = 40 μm. The displacement in the y-direction is imposed

be considered either by increasing the number of particles within to zero on the bottom surface of the silica sample and the gravity
the granular packing which is potentially costly or by reducing the is neglected. Mechanical properties of both materials are given in
number of specimens which reduces the discretization level. Table 1.
In a second step, we investigate cracks initiation and propaga- During DE calculations, the indenter contacts several particles
tion within the beam in bending. The vertical loading is applied ac- belonging to the silica medium. Such elementary contacts are ruled
cording to a linear ramp of one million steps from 0 to 100 N in or- by the Hertzian contact law the formulation of which is given in
der to reduce dynamics effects. Numerical calculations are carried Eq. (16). One possible choice is to consider the normal stiffness
out using the same time step as previously used, namely 2.5e−9 s. Kcn given in Eq. (17). However, due to the very large scale differ-
The fracture is modeled using the RDEF criterion with a hydrostatic ence between the size of the indenter and the size of the particle
stress limit of 200 MPa. Please notice that we limit our investiga- such an expression provides a too small stiffness which leads to
tions to the case of a stress field estimated at the scale of par- very important deformation and interpenetration. That is why Kcn
ticle in view of results obtained at the scale of the specimen in is calibrated in order to ensure that the macroscopic contact be-
the previous example of the plate with hole. A crack initiates at tween the indenter and the silica verifies the Hertz theory. From
the position (41.04 mm; 0.21 mm) near A position for a loading a practical standpoint, a large range of Kcn values are tested and
of about 48 N and propagates quasi-vertically up to upper edge the interpenetration given by the numerical approach is estimated
cp
of the beam. Fig. 13 illustrates the rupture of the beam with the and compared to un given by the Hertz theory for a cylinder-plane
vertical crack. Results exhibit suitable cracks initiation and propa- contact:
gation which are in accordance with the hypotheses of RDEF cri-
b π E∗
3
F
terion, namely a brittle fracture with a mode I crack propagation. ucp
n = ∗ 1 + ln (20)
As for the case of the plate with a hole, micro-macro scale transi-
πE b FR
tion effects slightly impact on the crack initiation position but the In a first step, a load of 100 mN is applied to the indenter in the
propagation is not affected. y-direction using a linear ramp of 20 0,0 0 0 time steps with a time
step set to 1.4e−12 s. The silica sample is discretized by a gran-
4.3. Indentation tests ular packing composed of 15,0 0 0 particles and 44,0 0 0 links. For
this loading, no crack occurs but tensile stresses are already no-
Based on the works of André et al. (2013), indentation tests ticeable. Fig. 15a illustrates the positive hydrostatic stress within
are simulated using the CDEM and cracks propagation is investi- the silica sample. For sake of clarity, the stress is only estimated
gated using the RDEF criterion. The present study aims to high- at the scale of the particle and no specimen is considered in the
light the suitability of the CDEM to model such a process and to present study case. One can notice that tensile stresses appear at
check out the ability of the RDEF criterion to yield suitable cracks both sides of the indenter with a maximum value of 32.7 MPa less
initiation and propagation. Fig. 14 illustrates the 2D model used in than the stress limit of 50 MPa. Less important stresses also appear
the present study. A cylindrical diamond indenter is modeled by a at both lateral sides of the sample. These are due to boundary ef-
unique DE of radius R = 10 μm in contact with a parallelepipedic fects related to the finite size of the silica sample. Fig. 15b shows
silica sample of length L = 40 μm, width h = 20 μm and thick- the Von-Mises stress field for the same configuration. Results ex-
Fig. 14. Model of indentation test.

Fig. 15. (a) Positive hydrostatic stress field and (b) Von-Mises stress field within the silica sample.
hibit a maximum value located near the area of contact between that local links loaded in traction could be located in compression
the indenter and the silica sample. This is 2.09 GPa which is much areas and affect the results. Fig. 16a and b illustrate the network of
higher than the maximum positive hydrostatic stress value and links obtained after cracking for the bond-based model proposed
highlights the important compression stresses undergone by the in the present work and the RDEF criterion without particle ejec-
medium. We remind that the failure of the silica due to a compres- tion. Please keep in mind that red links are simple contacts ruled
sion loading is rarely observed which is explained by a very high by Hertz’s law and black links are cohesive contacts. One can no-
compressive strength which is not well-known. Comparisons are tice that both approaches lead to satisfying locations of crack ini-
carried out with the Hertz theory for a cylinder-plane contact in tiation even though the second model yields initiation positions
order to validate these results. According to this theory, the maxi- quite remote of the contact area. Besides, cracks initiate for the
mum pressure Pmax reads: same reaction force of 270 mN which is in good agreement with
2F numerical and experimental measurements given by André et al.
Pmax = (21) (2013). For information purposes, the reaction force is here esti-
π ab
mated by summing the contributions of particles in contact with
where a is the half-width of the rectangular area in contact with
the indenter. However, both models do not yield a suitable crack-
the indentor which reads:
ing pattern. Thus, the bond-based approach leads to vertical cracks
4FR and the RDEF criterion without particle ejection leads to a kind of
a= (22)
π bE∗ Hertzian cone with undesirable branchings.
Fig. 17 a illustrates the cracking pattern obtained using the
Thus, in the present configuration, Pmax and a values given by the
RDEF criterion with particle ejection for the same granular packing
Hertz theory are respectively equal to 2.44 GPa and 6.53e−7 m.
composed of 6600 DE. Results exhibit the initiation of two cracks
During DE calculations, 22 elementary contacts occur between the
at both sides of the indenter on the upper face of the silica sample
indenter and the sample. The half-width given by the numerical
for a reaction force of 270 mN. Cracks propagate obliquely up to a
approach can be approximated by summing the diameters of the
reaction force of 285 mN for a total duration of 2.1e−8 s. One can
set of DE in contact with the indenter and dividing by two the re-
observe a conical geometry the parameters of which are:
sult. This leads to a numerical half-width of 1.08e−6 m which is
slightly higher than the theoretical reference. The maximum pres- • the conical angle α = 28.6◦ /31.4°
sure given by the numerical approach is extracted of σ yy stress • the length of the crack c = 8.7 μm/9.4 μm
field on the upper edge of silica sample. This is 2.26 GPa which • the cone width d = 10.9 μm
is quite close to the theoretical value. These findings exhibit the • the half-width of the contact area a = 2.3 μm
good adequation between the numerical approach and the theory.
A constant velocity v = 0.3 m s−1 is now imposed to the inden- The obtained 2D conical geometry can be related to the 3D
ter. Our objective is to investigate the influence of several parame- Hertzian cone well-known in quasi-static indentation of brittle ma-
ters related to the DEM on the cracking pattern. For that purpose, terials. The set of parameters as well as the reaction force leading
the silica sample is discretized by a granular packing composed of to cracks initiation are quite close to numerical and experimen-
6600 particles and 19,500 links. The time step is set to 2.1e−12 s tal values given by Andrè et al. for the same mechanical parame-
which corresponds to a Dt parameter equal to 0.05. In a first step, ters and an equivalent geometrical configuration. Even though the
we aim to study the influence of the failure criterion. Three models cylindrical geometry of the indenter reduces the relevance of such
are considered. The first approach is a bond-based criterion based a comparison, we can nevertheless conclude that the RDEF crite-
on the deletion of a bond rather than a particle when a given hy- rion leads to suitable qualitative crack patterns in the present in-
drostatic stress level is reached. In this model, the stress associated dentation test. In order to investigate the influence of the size of
to the bond is obtained as the geometric mean of values given by the particle, the same simulation is now performed for a gran-
both particles in contact. The second approach is the RDEF crite- ular packing composed of 11,0 0 0 DE and 32,400 cohesive links.
rion for which the particle is not ejected but kept in the simula- Fig. 17 illustrates the corresponding cracking pattern. One can ob-
tion as debris after deleting the bonds. The third approach is the serve a 2D conical geometry the characteristics of which are quite
RDEF criterion as described in Section 3.2. Please notice that for similar to the previous ones except for the cone width d which is
each study case the stress is estimated at the scale of the DE using now 15.4 μm and closer to the experimental value. Cracks initiate
the Zhou’s formulation (Zhou, 2003). Moreover, a fourth approach, for the same reaction force of 240 mN. Thus, the particle size has
the bond-based Rankine criterion has also been investigated in the little effect on geometric parameters of the cracking pattern. Other
works of André et al. (2013). The authors only considered the rup- tests have also been performed to determine the influence of the
ture of cohesive links when the beam is loaded in traction and not applied velocity and the time step using the granular packing com-
in compression. Thus, their simulations highlighted suitable loca- posed of 6600 particles. To this end, the Dt was varied from 0.005
tions of crack initiation but vertical propagations. We hypothesize to 0.25 and the velocity was varied from 0.03 to 10 m s−1 . Re-
Fig. 16. Network of links after cracking: (a) case of a bond-based failure mode and (b) case of the RDEF criterion without particle ejection.
Fig. 17. Cracking pattern obtained using the RDEF criterion with particle ejection (a) for a granular packing composed of 6600 DE and (b) for a granular packing composed
of 11,0 0 0 DE.
sults exhibit no influence of both parameters on the cracking pat-

tern and the initiation force. Consequently, combined to previous
studies, we can conclude that the RDEF criterion turns out to be
well-suited to model cracks initiation and propagation within a 2D
elastic and homogeneous ECD. The next session aims to investigate
the case of heterogeneous media.
5. Heterogenous media
We consider the case of heterogeneous media. Two main con-

figurations are investigated, namely a single-fiber composite and
a multi-fiber one. Studies are performed in terms of stress field,
cracks propagation and interfacial debonding under a tensile load-
ing.
Fig. 18. 2D square pattern of the single-fiber composite.
5.1. Single-fiber composite
We investigate the case of a single metallic fiber embedded exist between a DE belonging to the matrix phase and a DE be-
in a brittle alumina matrix. Fig. 18 illustrates the heterogeneous longing to the fiber one. These are associated to the interface .
medium modeled by a 2D square pattern of length L = 0.1 m Both phases follow a linear elastic behavior and are modeled by
constituted of one centered inclusion of radius R = L/4. The dis- the CDEM in which DE are connected by cohesive beams. This is
cretization level of the ECD describing the square pattern is ob- consistent with the stress field predicted by DE calculations within
tained according to preliminary studies performed in a previous the fiber the maximum value of which always remains lower than
work (Haddad et al., 2015). Thus, a granular packing constituted the elasticity limit of a classical metal such as aluminum. Elastic
of 80 0 0 particles and about 2320 links is generated which ensures properties of each phase are given in Table 2. In present studies,
the mass conservation and the isotropy of the domain. Each parti- the stress limit of the metal is assumed infinite so that no damage
cle is associated to one phase according to the position of its cen- occurs within the fiber. The rupture of brittle alumina matrix is
ter outside or inside the inclusion. The repartition is as follows: modeled using the RDEF criterion based on the hydrostatic stress
about 6400 particles belong to the matrix and about 1600 parti- field evaluated at the scale of the particle. The behavior of the in-
cles to the single fiber. About 1740 cohesive links connect two DE terface is ruled by the DDZM discussed in Section 3.3. In other
belonging to the matrix phase and about 440 cohesive links con- words, a link associated to is modeled by a spring-like element
nect two DE belonging to the fiber phase. Besides, about 140 links the normal stiffness Kn
,0
of which is chosen equal to the geometric
Fig. 19. Positive hydrostatic stress field obtained using (a) DE calculations and (b) FE Calculations: case of the single fiber.
Table 2 order to better understand the influence of this variability on the

Properties of each phase.
cracks initiation and propagation. Previous studies dedicated to ho-
Material Alumina Metal mogeneous media provided some responses since suitable cracking
Young’s modulus (GPa) 350 70 pattern were observed. We now aim to study the case of cracks
Poisson’s ratio 0.25 0.3 initiation and propagation within heterogeneous media.
Density (kg/m3 ) 3900 2700 Matrix failure and interfacial debonding are now taken into ac-
σ lim (MPa) 200 ∞ count. An imposed velocity of 0.1 m s−1 is imposed at both vertical
edges of the 2D square pattern. In a first step, the ratio C is cho-
sen equal to 0.25 so that the stress limit of the interface strength
f is lower than this of the matrix. DE calculations are carried out
average of normal stiffnesses Kn and Kmn of beam elements respec-
tively associated to the fiber and the matrix: considering the same discretization and time step than previously.
Fig. 20a illustrates the final cracking pattern within the heteroge-
,0 neous medium and the debonding area in red around the fiber.
Kn = Knf Km
n (23)
First debonding occurs at the interface around the fiber at time
is defined as a ratio C
The macroscopic interfacial stress limit σlim t≈ 2e−4 s then cracks initiate at its top and bottom and propa-

of the matrix stress limit σlim
m so that: gate vertically at time t≈ 2.2e−4 s. These findings are in accordance
with a ratio C lower than 1 for which the debonding is prone to

σlim = C σlim
m
(24) occur before damaging. Besides, vertical cracks are typical of a brit-
We first consider an outer displacement of 12 μm imposed at both tle rupture in mode I. In a second step, the ratio C is chosen equal
lateral edges of the 2D square in x-direction. Thus, the medium to 4 so that the stress limit of the interface strength is greater than
is under a tensile stress lower than stress limits of both phases this of the matrix. Fig. 20b illustrates the final cracking pattern.
and the interface. In other words, no debonding and failure arise Please notice that in the present case no debonding arises except
in the present study case. DE calculations are performed using a for local spots. First cracks initiate at top and bottom positions
time step of 1.2e−8 s and comparisons are performed with FE ones of the fiber and propagate vertically at time t≈ 2.5e−4 s. Then a
which are based on a structured mesh composed of 360,0 0 0 nodes third crack appears and bypasses the fiber and vertical cracks ini-
and 720,0 0 0 3-node triangles. Fig. 19a and b show the positive hy- tiate from this latter at time t≈ 3.6e−4 s. These results are in good
drostatic stress field predicted using DE calculations and FE ones. agreement with a ratio C greater than 1. Due to the strong inter-
From a qualitative standpoint, stress fields are very similar with face strength, debonding does not occur and vertical cracks initi-
maximum values located at top and bottom of the fiber and mini- ated from the fiber are finally connected by a third one.
mum values located at its left and right. From a quantitative stand- Fig. 21 illustrates the strength-displacement relationship for
point, DE and FE calculations lead to different maximum stress C = 0.25 and C = 4. One can observe a linear loading typi-
values respectively equal to 123 MPa and 69.9 MPa and located cal of an elastic medium up to a displacement respectively equal
at M(45.8 mm; 77.6 mm) and A(50 mm; 75 mm) positions. How- to 1.8e−5 m and 2.4e−5 m and a strength respectively equal to
ever, DE calculations are prone to local variabilities which could 920 N and 1210 N. This phase is a little bit shorter in the first
explain such a discrepancy. In order to verify this hypothesis, the case due to the interfacial debonding. Then the strength starts to
hydrostatic stress field is determined using a tessellation of the do- decrease and quickly converges to zero. A small reloading is no-
main in 500 triangular specimens and local values are extracted ticeable for C = 0.25. This is probably related to the transition
and compared to FE predictions. Table 3 provides the stress val- step between the interfacial debonding and the matrix failure. Both
ues predicted at A, B(50 mm; 50 mm) and C(25 mm; 50 mm) po- cases, C = 0.25 and C = 4 lead to suitable cracks propaga-
sitions respectively located at the top, the center and the left of tion and initation. This highlights the consistency of the CDEM-
the fiber. The hydrostatic stress given by DE calculations without based approach to simulate such phenomena using RDEF criterion
specimen at A position is only 53.2 MPa while this given by speci- in the present configuration. Furthermore, variability effects related
mens is 67.6 MPa and quite close to the FE prediction. Besides, the to the micro-macro scale transition have seemingly little impact on
maximum stress is decreased to 74.5 MPa which is also in good cracks initiation and propagation. This point could be nevertheless
agreement with FE results. Even though small fluctuations still re- more in-depth investigated.
main, results given by DE calculations with specimen are globally
in good agreement with FE ones. This highlights the consistency 5.2. Multi-fiber composite
of the DE approach to yield a suitable stress field. However, we
remind that the failure criterion is applied at the scale of the par- We now consider a composite constituted of parallel metallic
ticle where variabilities arise. Thus, a great care has to be taken in fibers embedded in an alumina matrix. Two cases are investigated:
Table 3
Hydrostatic stress values at A, B and C positions.
A position B position C position Maximum value
DE predictions without specimen 53.2 MPa 13.5 MPa 3.47 MPa 123 MPa
DE predictions with specimens 67.6 MPa 13.2 MPa 3.47 MPa 74.5 MPa
FE predictions 69.9 MPa 14.7 MPa 1.95 MPa 69.9 MPa
Fig. 20. Cracking pattern and debonding area for (a) C = 0.25 and (b) C = 4.
rion so that no damage occurs in the present study. Comparisons

are carried out with FE calculations based on a structured mesh
composed of 491,0 0 0 nodes and 980,0 0 0 elements. Fig. 23a and b
show the positive hydrostatic stress field predicted using DE and
FE calculations. One can notice that stress fields are very similar
with hot spots located close to contact points. From a quantita-
tive standpoint, as previously observed in the case of the single
inclusion, DE predictions yield a maximum stress higher than the
one given by FE simulations, namely 128 MPa versus 85.7 MPa.
However, the maximum value is located at the same position M
of coordinates (90.4 mm; 20.6 mm). In order to better appreciate
the difference between both approaches, hydrostatic stress values
are compared at several positions of the square pattern. Compar-
isons are also performed with DE predictions obtained using a tes-
sellation of 80 0 0 triangular specimens. Table 4 provides the stress
values predicted at A(21.1 mm; 39.3 mm), B(62.6 mm; 70.5 mm),
Fig. 21. Strength-displacement relationship: case of the single fiber. C(40.3mm;88.6mm) and D(90.6 mm; 59.7 mm) positions. Glob-
ally, considering the difference of discretization level, DE predic-
tions are quite close to FE ones whatever the investigated position
a closely-packed fiber composite and a randomly dilute fiber one. except for the maximum stress location. This maximum value is
The first configuration is modeled by a 2D square pattern of length nevertheless much closer to the FE prediction using a set of speci-
L = 0.1 m composed of 25 monodisperse closely-packed fibers of mens, namely 76.1 MPa versus 85.7 MPa. These results exhibit the
radius R = 9.6 mm in point contact. This is generated by LSA suitability of the CDEM to predict the hydrostatic stress field in the
without periodicity assumptions so that the compacity of the final present configuration of a complex material. DE’s inherent variabil-
packing is close to 0.73 due to wall effects. The 2D square pattern ity effects still impact on the maximum value but its position is a
related to the dilute configuration is generated from the previous priori not affected. Consequently, the cracks initiation should be
one by reducing the size of particles so that the final compacity is not affected but this point has to be verified.
set to 0.5 and the fiber radius R is set to 6.25 mm. In this latter Interfacial debonding effects and matrix failure are now taken
model, a minimum interstice of 3.7 mm is respected between two into account. A tensile test is simulated until the material failure
given fibers. Fig. 22a and b show the 2D square pattern associated by applying an outer velocity of 0.1 m s−1 at both lateral edges
to each configuration. In present investigations, material properties of the 2D square pattern in the x-direction. In the present study,
and stress limits are the same than in the previous study. Based C ratio is set to 0.25 so that the interfacial debonding is prone
on the works of Haddad et al. (2015), the domain is discretized in to arise before cracking. Besides, the hydrostatic stress limit in the
a granular packing constituted of 10 0,0 0 0 particles which ensures RDEF criterion remains set to 200 MPa. DE simulations are per-
the representativity of each medium. Tensile tests are set up and formed using an initial time step of 1.6e−9 s. Fig. 24a, b and c
the mechanical behavior of each configuration is studied in terms illustrate the positive hydrostatic stress field and the interfacial
of hydrostatic stress field and cracks initiation and propagation. debonding at three given times t = 1.33e−4 s, t = 1.44e−4 s and
t = 1.60e−4 s. In a first step, the interfacial debonding appears
5.2.1. Closely-packed fiber composite at a position of coordinates (54.4 mm; 18.2 mm) which is encir-
In a first step, an outer displacement of 1.2e−5 m is imposed cled in blue on Fig. 24a. Then, the debonding propagates verti-
at both lateral edges of the pattern in the x-direction. DE calcu- cally from fiber to fiber throughout the contact points. The max-
lations are performed using a time step of 1.6e−9 s. Under such imum hydrostatic stress is 4.32e8 Pa which is higher than the fail-
a solicitation, resulting stresses are lower than the failure crite- ure criterion. However, no matrix rupture occurs except for local
Fig. 22. 2D square patterns of (a) a closely-packed fiber composite and (b) a randomly dilute fiber composite.
Fig. 23. Positive hydrostatic stress field obtained using (a) DE calculations and (b) FE Calculations: case of the closely-packed fiber composite.
Table 4
Hydrostatic stress values at A, B, C and D positions.
A position B position C position D position Maximum value
DE predictions without specimen 18.4 MPa 14.6 MPa −1.27 MPa 71.0 MPa 128 MPa
DE predictions with specimens 23.8 MPa 14.6 MPa −0.665 MPa 68.6 MPa 76.1 MPa
FE predictions 22.8 MPa 17.8 MPa −1.03 MPa 60.7 MPa 85.7 MPa
Fig. 24. Positive hydrostatic stress field and interfacial debonding within the closely-packed fibre composite at time (a) t = 1.33e−4 s, (b) t = 1.44e−4 s and (c) t = 1.60e−4 s.
a tensile testing. One must keep in mind that the displacement

value is willingly chosen lower than a given threshold leading to
matrix failure and debonding effects. Thus, no damage occurs in
the present study. DE calculations are performed using a time step
set to 1.6e−9 s and the granular packing composed of 10 0,0 0 0 par-
ticles is reused in the present study. Again, comparisons are carried
out with FE calculations based on a structured mesh composed
of 491,0 0 0 nodes and 980,0 0 0 elements. Fig. 26a and b show the
positive hydrostatic stress field predicted using DE and FE calcula-
tions. Both stress fields are quite similar with hot and cold spots
located close to matrix-fiber interfaces. From a quantitative stand-
point, the maximum hydrostatic stress estimated at the scale of
the particle by the CDEM is 140 MPa. Again, this is higher than
the prediction given by the FEM, namely 72.3 MPa. Nevertheless,
each one is located quite close to each other, namely the position
M of coordinates (28.8 mm; 20.3 mm) for DE calculations and the
Fig. 25. Strength-displacement relationship: case of the closely-packed fiber com- position M’ of coordinates (29.7 mm; 18.8 mm) for FE calculations.
posite. Table 5 gives the hydrostatic stress values predicted at A(21.3 mm;
49.1 mm), B(90.6 mm; 59.7 mm) and C(76.4 mm; 23.3 mm) posi-
tions using the DE approach without specimen, the DE approach
points since hot spots are in fact located on the debonded inter-
with a set of 80 0 0 specimen and FE calculations. As previously
face. Finally, the debonding area reaches the edges of the square
seen for the case of closely-packed fibers, DE predictions are quite
pattern and the ruin of the material starts. These findings are in
close to FE ones whatever the investigated position except for the
good agreement with a C ratio less than 1 for which the debond-
maximum value which requires a set of specimens to be in suit-
ing is predominent. Besides, as expected, the fibre-matrix debond-
able agreement with the FE measurement. These findings confirm
ing propagates in the perpendicular direction of the tensile solici-
the consistency of the CDEM to yield the hydrostatic stress field of
tation. Consequently, the DDZM leads to suitable debonding effects
a complex heterogeneous medium. Again, variability effects do not
in the present configuration.
affect the position of the maximum stress value which should lead
Fig. 25 illustrates the strength-displacement relationship. In a
to suitable cracks initiation position.
first step, the loading is linear elastic up to a critical displacement
DE simulations are now carried out until the material failure
equal to 1.33e−5 m and a reaction force equal to 270 N. In a sec-
by applying an outer velocity of 0.1 m s−1 at both lateral edges of
ond step, debonding effects arise and the force quickly decreases
the 2D square pattern in the x-direction. Again, interfacial debond-
to zero till the ruin of the material. The critical displacement de-
ing is ruled by the DDZM using a C ratio set to 0.25 and the
duced from this curve is in perfect agreement with the initiation
hydrostatic stress limit is set to 200 MPa. The time step is set
of debonding effects observed on Fig. 24a at time t = 1.33e−4 s.
to 1.6e−9 s. Fig. 27a and b show the positive hydrostatic stress
However, even though the present example enables us to better
field within the randomly dilute fibre composite when cracks ini-
appreciate the suitability of the DDZM for modeling debonding ef-
tiate at t = 1.44e−4 s and when the ruin of the material starts at
fects in complex heterogeneous materials, none can be concluded
t = 1.58e−4 s respectively. One can notice that hot spots are lo-
as regards cracks initiation and propagation. Indeed, only small and
cated close to debonding areas and cracks. Conversely to the pre-
local matrix failures occur in the present configuration and cracks
vious configuration, the maximum hydrostatic stress is a little bit
consequently play a very minor role in the ruin of the material.
lower than the stress limit which is 200 MPa. In a first step, at
The next study is dedicated to the randomly dilute fiber compos-
time t = 1.42e−4 s, cracks initiate at top and bottom positions of a
ite in which cracks phenomena are more prone to occur and affect
fiber the center of which has the coordinates (45.9 mm; 39.2 mm).
the mechanical behavior.
Additional tests performed on several other random patterns ex-
hibit that cracks always appear at the middle of square pattern in
5.2.2. Randomly dilute fiber composite the x-direction which denotes no boundary effect. In a second step,
We now consider the case of the randomly dilute fiber compos- each crack propagates vertically up to the nearest fiber which is
ite. In a first approach, as previously done for the closely-packed bypassed by a debonding area linking the tip of the crack and the
fibers, an outer displacement of 1.2e−5 m is imposed at both lateral initiation location of a new one. The process is repeated from fiber
edges of the 2D square pattern in the x-direction so that to model
Fig. 26. Positive hydrostatic stress field obtained using (a) DE calculations and (b) FE Calculations: case of the randomly dilute fiber composite.
Table 5
Hydrostatic stress values at A, B and C positions.
A position B position C position Maximum value
DE predictions without specimen 1.90 MPa 49.5 MPa 26.2 MPa 140 MPa
DE predictions with specimens 2.64 MPa 55.6 MPa 22.9 MPa 87.2 MPa
FE predictions 2.23 MPa 56.3 MPa 25.4 MPa 75.7 MPa
Fig. 27. Positive hydrostatic stress field within the randomly dilute fibre composite at time (a) t = 1.42e−4 s and (b) t = 1.58e−4 s.
6. Conclusions and prospects
The present paper aimed at investigating the suitability of the

CDEM to model cracks initiation and propagation in heterogeneous
media. In a first step, several tests were carried out in a homo-
geneous context using the RDEF criterion. Results exhibited some
micro-macro scale effects on the hydrostatic stress field but suit-
able cracks initiation and propagation. In a second step, two het-
erogeneous configurations were considered, namely a single metal-
lic fiber embedded in an alumina matrix and the case of parallel
metallic fibers embedded in an alumina matrix. Hydrostatic stress
field as well as cracks initiation and propagation were studied un-
der a tensile solicitation. Main conclusions are listed below:
• Micro-macro scale effects mainly impact on the maximum hy-

drostatic stress which is up to two times this given by the FEM.
Fig. 28. Strength-displacement relationship: case of the randomly dilute fiber com- However, its position is barely affected and stress values esti-
posite.
mated at different positions of the domain are in quite good
agreement with FE calculations.
• For an interface strength lower than this of the matrix, crack
to fiber till cracks reach the edges of the 2D square pattern. The fi- initiation always occurs at the matrix-fiber interface and no
nal rupture pattern is perpendicular to the direction of solicitation preferential side were observed.
which is indicative of a predominent role of the brittle failure. This • Whatever the configuration, single- or multi-fiber composite, a
is emphasized by the relatively poor influence of debonding areas vertical crack propagates perpendicularily to the tensile solici-
on the rupture mechanism. Indeed, these only appear in local spots tation which is in good adequation with a mode I propagation.
to make a bridge between two cracks separated by a fiber. • The fiber arrangement affects the competition between inter-
Fig. 28 shows the strength-displacement relationship related facial debonding and matrix failure. The role of the failure is
to the present configuration. First, a linear elastic loading is ap- predominent in the case of dilute fibers when this is quite neg-
plied up to a critical displacement of 1.42e−5 m and a maximum ligible for closely-packed fibers.
force of 440 N. For comparison purposes, the maximum force was
only 270 N in the previous configuration. This difference is due to Our prospects are as follows. Currently, we are working on
fiber volume which is 50% for the dilute model versus 73% for the the thermomechanical coupling based on a dilatation model in-
closely-packed one. More precisely, the decrease of fiber volume troduced at the scale of the particle. In a next future, we expect
in the composite leads to a higher Young’s modulus and higher to consider 3D configurations such as a random fiber composite.
forces in the present case. Then, in a second step, matrix failure This point will require to adapt our numerical codes to handle a
and debonding effects arise and the force quickly drops to zero. very large number of particles (probably more than one million
These findings exhibit a quasi brittle behavior of the composite DE). Besides, we would like to investigate some defects, local het-
which confirms the predominent role played by the matrix in a erogeneities and variation effects on the macroscopic behavior of
dilute system. This scope of results exhibit the ability of the RDEF composite materials.We are also interested in better characterizing
criterion to yield cracks pattern in agreement with the imposed the influence of packings characteristics using sensitivity analysis
solicitation and the observed mechanical behavior in the case of a (Vu-Bac et al., 2015; 2016) which could enable to better under-
complex heterogeneous material. stand the micro-macro relation in the calibration process.
Conflicts of interest Maheo, L., Dau, F., André, D., Charles, J.L., Iordanoff, I., 2015. A promising way to
model cracks in composite using discrete element method. Compos. Part B 71,
193–202.
None declared. Moës, N., Dolbow, J., Belytschko, T., 1999. A finite element method for crack growth
without remeshing. Int. J. Numer. Methods Eng. 46 (1), 131–150.
References Nicot, F., Hadda, N., Guessasma, M., Fortin, J., Millet, O., 2013. On the definition of
the stress tensor in granular media. Int. J. Solids Struct. 50, 2508–2517.
Love, A.E.H., 1927. A Treatise of Mathematical Theory of Elasticity. Cambridge Uni- Pilkey, W.D., 2005. Formulas for Stress, Strain and Structural Matrices, Second ed.
versity Press, Cambridge. John Wiley & Sons.
André, D., Iordanoff, I., Charles, J.C., Neauport, J., 2012. Discrete element method Rabczuk, T., Belytschko, T., 2004. Cracking particles: a simplified meshfree method
to simulate continuous material by using the cohesive beam model. Comput. for arbitrary evolving cracks. Int. J. Numer. Methods Eng. 61, 2316–2343.
Methods Appl. Mech. Eng. 213, 113–128. Rabczuk, T., Belytschko, T., 2007. A three-dimensional large deformation mesh-
André, D., Jebahi, M., Iordanoff, I., Charles, J.C., Neauport, J., 2013. Using the discrete free method for arbitrary evolving cracks. Comput. Methods Appl. Mech. 196,
element method to simulate brittle fracture in the indentation of a silica glass 2777–2799.
with a blunt indenter. Comput. Methods Appl. Mech. Eng. 265, 136–147. Rabczuk, T., Zi, G., Bordas, S., Nguyen-Xuan, H., 2010. A simple and robust three-
Cleary, P.W., Campbell, C.S., 1993. Self-lubrication for long run-out landslides: ex- -dimensional cracking-particle method without enrichment. Comput. Methods
amination by computer simulation. J. Geophys. Res. 98 (B12), 21911–21924. Appl. Mech. 199, 2437–2455.
Cundall, P.A., Strack, O.D.L., 1979. Discrete numerical model for granular assemblies. Ren, H., Zhuang, X., Cai, Y., Rabczuk, T., 2016. Dual-horizon peridynamics. Int. J. Nu-
Geotech 29, 47–65. mer. Methods Eng.. Accepted paper
Donev, A., Cisse, I., Sachs, D., Variano, E.A., Stillinger, F.H., Connelly, R., et al., 2004. Schlangen, E., Garboczi, E.J., 1997. Fracture simulations of concrete using lattice
Improving the density of jammed disordered packings using ellipsoids. Science models: computational aspects. Eng. Fract. Mech. 57 (2), 319–332.
303, 990–993. Shiu, W., Donzé, F.V., Daudeville, L., 2008. Penetration prediction of missiles with
Fillot, N., Iordanoff, I., Berthier, Y., 2007. Modelling third body flows with a discrete different nose shapes by the discrete element numerical approach. Comput.
element method-a tool for understanding wear with adhesive particles. Tribol Struct. 86, 2079–2086.
Int. 40 (6), 973–981. Silling, S.A., 20 0 0. Reformulation of elasticity theory for discontinuities and
Haddad, H., Leclerc, W., Guessasma, M., 2015. Application of DEM to predict the long-range forces. J. Mech. Phys. Solids 48, 175–209.
elastic behavior of particulate composite materials. Granul. Matter 17, 459–473. Tan, Y., Yang, D., Sheng, Y., 2009. Discrete element modelling of fracture and damage
Hentz, S., Donzé, F.V., Daudeville, L., 2004. Discrete element modelling of concrete in the machining of polycristalline sic. J. Eur. Ceram. Soc. 29 (6), 1029–1037.
submitted to dynamic loading at high strain rates. Comput. Struct. 82 (29–30), Torquato, S., Truskett, T.M., Debenedetti, P.G., 20 0 0. Is random close packing of
2509–2524. spheres well defined ? Phys. Rev. Lett. 84, 2064–2067.
Jebahi, M., André, A., Dau, F., Charles, J.L., Iordanoff, I., 2013. Simulation of vickers Vu-Bac, N., Lahmer, T., Zhuang, X., Nguyen-Thoi, T., Rabczuk, T., 2016. A software
indentation of silica glass. J. Noncryst. Solids 378, 15–24. framework for probabilistic sensitivity analysis for computationally expensive
Krueger, R., 2004. Virtual crack closure technique: history, approach, and applica- models. Adv. Eng. Softw. 100, 19–31.
tions. Appl. Mech. Rev. 57 (2), 109–143. Vu-Bac, N., Rafiee, R., Zhuang, X., Lahmer, T., Rabczuk, T., 2015. Uncertainty quan-
Kumar, R., Rommel, S., Jauffrès, D., Lhuissier, P., Martin, C., 2016. Effect of packing tification for multiscale modeling of polymer nanocomposites with correlated
characteristics on the discrete element simulation of elasticity and buckling. Int. parameters. Compos. Part B 68, 446–464.
J. Mech. Sci. 110, 14–21. Xie, D., Waas, A.M., 2006. Discrete cohesive zone model for mixed-mode fracture
Le, B.D., Dau, F., Charles, J.L., Iordanoff, I., 2016. Modeling damages and cracks using finite element analysis. Eng. Fract. Mech. 73, 1783–1796.
growth in composite with a 3d discrete element method. Compos. Part B 91, Xu, X.P., Needleman, A., 1994. Numerical simulations of fast crack growth in brittle
615–630. solids. J. Mech. Phys. Solids 42 (9), 1397–1434.
Liu, X., Duddu, R., Waisman, H., 2012. Discrete damage zone model for fracture ini- Zhang, Z., Paulino, G.H., 2005. Cohesive zone modeling of dynamic failure in homo-
tiation and propagation. Eng. Fract. Mech. 92, 1–18. geneous and functionally graded material. Int. J. Plasticity 21 (6), 1195–1254.
Lubachevsky, B.D., Stillinger, F.H., 1990. Geometric properties of random disk pack- Zhou, M., 2003. A new look at the atomic level virial stress: on continuum–
ings. J. Stat. Phys. 60, 561–583. molecular system equivalence. Proc. R. Soc. London Math. Phys. Eng. Sci. 2037,
Machado, C., Guessasma, M., Bellenger, E., 2015. Electromechanical modeling by 2347–2392.
DEM for assessing internal ball bearing loading. Mech. Mach. Theory 92,
338–355.

Leclerc 2017

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Leclerc 2017

Uploaded by

Copyright:

Available Formats

International Journal of Solids and Structures 108 (2017) 98–114

Contents lists available at ScienceDirect

International Journal of Solids and Structures

On the suitability of a Discrete Element Method to simulate cracks

1. Introduction a scientiﬁc challenge of main interest. These aim to better charac-

where θ i and θ j are respectively the rotations of particles i and j.

The normal component acts as an attractive force, the tangen-

2.3. Calibration process

the suitability of this approach to model complex crack patterns K   ,c

Thus, from a theoretical standpoint, the maximum tensile stress

where ζ designates the set of contacts within the granular pack-

4.2. Three point bending test

We investigate the case of a three point bending test. We con-

Fig. 10. Three point bending test.

Material Diamond Silica

Young’s modulus (GPa) 1050 72.5

Fig. 13. Rupture of the beam with vertical crack.

ness b = 40 μm. The displacement in the y-direction is imposed

Fig. 14. Model of indentation test.

sults exhibit no inﬂuence of both parameters on the cracking pat-

We consider the case of heterogeneous media. Two main con-

5.1. Single-ﬁber composite

Table 2 order to better understand the inﬂuence of this variability on the

A position B position C position Maximum value

rion so that no damage occurs in the present study. Comparisons

A position B position C position D position Maximum value

a tensile testing. One must keep in mind that the displacement

A position B position C position Maximum value

6. Conclusions and prospects

The present paper aimed at investigating the suitability of the

• Micro-macro scale effects mainly impact on the maximum hy-

You might also like

the suitability of this approach to model complex crack patterns K ,c