MIT Open Access Articles

Bonded-cell model for particle fracture

The MIT Faculty has made this article openly available. Please share
how this access benefits you. Your story matters.

Citation: Nguyen, Duc-Hanh, Emilien Azéma, Philippe Sornay, and Farhang Radjai. “Bonded-Cell
Model for Particle Fracture.” Phys. Rev. E 91, no. 2 (February 2015) © 2015 American Physical
Society

As Published: http://dx.doi.org/10.1103/PhysRevE.91.022203

Publisher: American Physical Society

Persistent URL: http://hdl.handle.net/1721.1/94343

Version: Final published version: final published article, as it appeared in a journal, conference
proceedings, or other formally published context

Terms of Use: Article is made available in accordance with the publisher's policy and may be
subject to US copyright law. Please refer to the publisher's site for terms of use.
 PHYSICAL REVIEW E 91, 022203 (2015)

Bonded-cell model for particle fracture

Duc-Hanh Nguyen,1,2,* Emilien Azéma,1,† Philippe Sornay,2,‡ and Farhang Radjai1,3,§

1
Université de Montpellier, CNRS, LMGC, Place Eugène Bataillon, 34095 Montpellier, France
2
CEA, DEN, DEC, SPUA, LCU, F-13108 Saint Paul lez Durance, France
3
MultiScale Material Science for Energy and Environment, UMI 3466 CNRS-MIT, CEE, Massachusetts Institute of Technology,
77 Massachusetts Avenue, Cambridge 02139, USA
(Received 27 August 2014; published 9 February 2015)
Particle degradation and fracture play an important role in natural granular flows and in many applications of
granular materials. We analyze the fracture properties of two-dimensional disklike particles modeled as aggregates
of rigid cells bonded along their sides by a cohesive Mohr-Coulomb law and simulated by the contact dynamics
method. We show that the compressive strength scales with tensile strength between cells but depends also on
the friction coefficient and a parameter describing cell shape distribution. The statistical scatter of compressive
strength is well described by the Weibull distribution function with a shape parameter varying from 6 to 10
depending on cell shape distribution. We show that this distribution may be understood in terms of percolating
critical intercellular contacts. We propose a random-walk model of critical contacts that leads to particle size
dependence of the compressive strength in good agreement with our simulation data.

DOI: 10.1103/PhysRevE.91.022203 PACS number(s): 45.70.−n, 83.80.Fg, 62.20.M−, 82.30.Lp

I. INTRODUCTION towards the contact points and (2) the contact fracture
mode characterized by the propagation of cracks initiated
Particle breakage occurs commonly in natural granular
at the contact points toward the center of the particles. The
flows and industrial processes involving the transport, han-
experiments performed on a brittle material suggest that the
dling, and compaction of granular materials. The particle size
volume fracture mode is more likely to occur in large-enough
reduction is often undesirable or uncontrolled, and it is referred
particles.
to as the attrition process. In contrast, the fragmentation of
Several authors have used the finite element method (FEM)
particles under controlled conditions is used in comminution
to study single-particle fracture by incorporating the material
processes such as the milling of vegetal products or grinding of
behavior and an adequate damage or rupture criterion [23–26].
mineral materials. The evolution of particle size distribution
This method has the advantage of accounting for the true nature
and energy dissipation in such processes depend on many
of the material and provides access to the full stress field in
factors such as particle properties (shape, crushability), initial
a continuum framework. But it requires rather fine meshing
size distribution, loading history, and mobility of the grains
of the particle at its borders or at least around its contact
during the crushing process [1–9].
points with other particles and at crack tips. Its application to
Both single-particle crushing and fragmentation process of
an assembly of particles further requires a proper treatment
an assembly of particles subjected to shearing or compaction
of frictional contacts and large deformations, which make it
have been subject to experimental investigations in civil
computationally inefficient.
engineering and particle technology [10–22]. The compressive
Numerical simulations by the molecular dynamics (MD)
strength of a single particle, its variability, and size dependence
method or discrete element method (DEM) have been in-
are essential for understanding the collective response of a
creasingly employed in order to get a better understand-
granular material to applied loading. The fracture of a particle
ing of the particle-scale mechanisms of the comminution
inside a granular packing depends on the angular positions process [1,9,27–31]. Such methods combine the general
of its contact neighbors and the normal and tangential forces framework of the DEM, based on rigid-body dynamics and
exerted by them on the particle. For this reason, there is no frictional contact interactions, with a particle fracture model.
general analytical model for the fracture of a single particle. DEM numerical models have the advantage of allowing for the
The case of a particle subjected to diametrical compression treatment of frictional contact interactions and they provide
(also called the Brazilian test) has been more carefully detailed information about local particle environments and
considered in this respect. A detailed analytical model was force chains that control the breakup events.
developed by Tsoungui et al. using Weibull statistical flaw The most straightforward DEM-based approach consists
size distribution and compared to experiments [23]. Two in modeling the particles as aggregates of spherical sub-
different fracture modes were analyzed in this model: (1) the particles bonded together by cohesive forces. Such aggre-
volume fracture mode in which a transversal crack responsible gates may represent real aggregates such as pellets and
of particle fracture originates near the center and propagates ceramic compacts or simply be regarded as a toy model
for particle fracture. This bonded particle model (BPM)
has been employed to investigate the behavior of crushable
*
[email protected] soils, rocks, fault gouge, and other materials [7,22,32–41].
† An alternative method consists in replacing a circular or
[email protected]
‡ spherical particle at its fracture threshold by several smaller
[email protected]
§
[email protected] fragments of the same shape [42–46]. A major issue with

1539-3755/2015/91(2)/022203(9) 022203-1 ©2015 American Physical Society

NGUYEN, AZÉMA, SORNAY, AND RADJAI PHYSICAL REVIEW E 91, 022203 (2015)

these methods is that an aggregate of spherical subparticles

includes voids, so its breakup leads to considerable loss of
volume.
To circumvent the volume-loss issue, several authors have
used polygonal or polyhedral subparticles or cells generated
by Voronoi tessellation [47–50]. These cells pave the whole
volume of the particle so the volume is conserved during l>λ d 0
particle fracture and fragmentation. Such bonded cell models
(BCM) involve extended intercellular contacts that need to be
modeled differently from contacts between spherical particles.
In previous studies, the cells were interconnected by linear
(a)
springs with a breaking threshold [49,50]. The representation
of intercellular contacts by a linear force law as that between
spherical subparticles is, however, an unphysical approxima-
tion since the contacts extend along a line [in two dimensions
(2D)] or a surface (in 3D) between cells and thus their treatment
needs at least two or three displacement variables, respectively.
The cracks propagate along these contacts as intergranular
cracks in polycrystalline materials.
In this paper, we use the BCM approach with the contact
dynamics (CD) method to investigate the fracture properties
of disklike particles subjected to diametrical compression
[51–53]. Each particle is modeled as an aggregate of per- (b) (c)
fectly rigid cells with their common sides modeled as
frictional-cohesive contacts. The framework of the CD method FIG. 1. (Color online) Definition of the degree of regularity of
allows us to account for the perfectly rigid behavior of the mesh λ (a) and two examples of the discretization for λ = 0 (b)
the cells and the correct kinematics of side-side contacts. and λ = 0.8 (c).
Since the cells are treated as perfectly rigid elements, in
contrast to linear spring-dashpot models of contact, a crack respectively. We will see that λ affects the variability of fracture
is generated only when all critical intercellular contacts behavior.
(contacts at their tensile threshold) percolate across the A key issue in using DEM with crushable aggregates is the
particle. We are interested in the compressive strength of a statistical representativity of the particles and their fragments
single particle under uniaxial compression in this model with during crushing. In fact, the sizes of the initial aggregates
focus on its variability and size dependence. Our findings and cells are, respectively, the upper and lower bounds on
are of potential interest to intergranular fracture in poly- the size distribution of fragments in the debris. The statistical
cristalline materials, rock fracture, and quasibrittle fracture of representivity of particle size distribution in the process of
biomaterials. fragmentation is therefore determined by their ratio. Moreover,
In the following, we first describe the numerical model. the mechanical behavior and fracture of a particle depends on
Then, in Sec. III, we investigate the effects of cell size the number of cells. At the same time, since the cells are
distribution. Section IV is devoted to the variability of treated as subparticles interacting via cohesion forces, the
compressive strength. In Sec. V we study size dependence and computation time increases with their number. Hence, it is
introduce a simple model based on critical paths. We conclude essential to optimize the number of cells per particle in order
with a brief discussion of major findings of this work. to be able to include a large number of particles in the initial
configuration of the sample.
The cells are assumed to interact via cohesive frictional
II. BONDED CELL MODEL
contacts along their common sides. A side-side contact
Each particle is divided into nv cells by Voronoi tessellation, between two rigid cells involves two unilateral constraints. In
as shown in Fig. 1, each cell representing a rigid subparticle. other words, at least two repulsive forces along the common
For a particle of area S, the number of cells is given by side are necessary to prevent from their overlap. In practice,
nv = S/d02 , where d0 is the average cell size. nv points a side-side contact should therefore be represented by two
are distributed randomly on the surface of the polygon by contact points, as shown in Fig. 2, and the normal direction
imposing that the distance  between the points is larger than a is the normal to the common side. The cell motions are
minimum distance min = λd0 , as shown in the Fig. 1(a). The governed by equations of dynamics and at each side-side
parameter λ represents the degree of regularity of meshing contact two forces need to be calculated. The choice of the
or the span of cell size distribution. Indeed, for λ = 1 the two contact points representing a side-side contact is matter of
total surface may be meshed by squares of side d0 for S/d02 convenience. Only the resultant force at each side-side contact
points. Hence, in order to allow for cells of five sides and and its point of application are physically meaningful and
more, it is necessary to reduce λ. Figures 1(b) and 1(c) independent of the choice of the positions of contact points.
present two examples of meshing of a circle for λ = 0 and A side-side contact may open only at one of the two
λ = 0.8, corresponding to very irregular and regular meshes, points by pivoting around the other point. It may also open

022203-2
BONDED-CELL MODEL FOR PARTICLE FRACTURE PHYSICAL REVIEW E 91, 022203 (2015)

coefficient of friction and the coefficients of normal and

tangential restitution that control the rate of dissipation. An
n1 implicit time-stepping scheme makes the method uncondi-

f1
tionally stable. In contrast to the molecular dynamics method,
in the CD method tiny overlaps between particles are used for
contact detection but they do not represent an elastic deflection.

n2
For this reason, the time step can be larger than that in the
f2 molecular dynamics method. In CD, an iterative algorithm
based on nonlinear Gauss-Seidel iterations is used to determine
the contact forces and particle velocities simultaneously at
all potential contacts. The CD method has been extensively
employed for the simulation of granular materials in 2D and
3D [55–72].
The CD method is based on implicit time integration of
velocities but requires an explicit determination of the contact
network at the beginning of each time step [52,73]. The
FIG. 2. (Color online) Contact dynamics model of a side-side
contact detection between two bodies consists in looking
contact between two particles by two contact points with their normals the portions of space they occupy. The treatment of the
n1 and n2 and forces f1 and f2 . mechanical interaction requires additionally the identification
of a common tangent plane (a line in 2D). Of course, contacts
simultaneously at both points. The normal adhesion threshold may take place through a larger contact zone than a single
fc between two cells linearly depends on the length L of point. In 2D simulations of the present paper, the detection
the contact. Since the contact is represented by two points, of contact between two convex polygonal bodies was imple-
the tensile threshold is given by fc = σc L/2, where σc is the mented through the so-called shadow overlap method [73,74].
internal cohesion of the material. This means that a side-side
contact can lose its cohesion for normal force fn = −σc L/2 III. BOUNDARY CONDITIONS
reached at any of the two contact points representing the
side-side contact. But if none of the two normal forces f1n and In this paper, we use the BCM to study the breakup of
f2n at the two contact points is critical, then they continue to a single circular particle crushed between two platens. This
increase with no loss of cohesion and the particles remain glued “Brazilian” test will be used to investigate the effects of model
together until the total normal force f1n + f2n = 2fc = Lσc . parameters σc , μs , λ, and nv . The particle is crushed by
This situation is, however, very rare since the particles are applying stepwise displacements δy to the top and bottom
most of time subjected to force moments so the total torque platens. The total axial strain after N steps is given by N δy /d,
(f1n − f2n )L = 0 and thus a side-side contact opens mainly where d is particle diameter. Let F be the axial force exerted
when fc is reached at only one of the two contact points. on the particle; see Fig. 3. The average vertical stress σa acting
The shear strength along a side-side contact is given by on the particle is given by
τc = μs σc , where μs is the internal friction coefficient. A F
side-side contact may lose its cohesion only when the total σa = σyy  = . (1)
d
tangential force f1t + f2t at the two contact points representing
the side-side contact reaches the sliding threshold μs (f1n + We calculate σa directly from the forces between cells [75]:
f2n + 2fc ). The choice of a frictional material behavior is not σyy  = nc fy y , (2)
mandatory and τc may be defined independently of σc . In order
to limit the number of independent parameters, in this paper we where nc is the number density of contacts, fy is the y
used a Coulomb friction law. But the effect of local cracking component of the reaction force between two particles, and
criterion may be a subject of detailed investigation in the BCM
framework. F
When the cohesion between two cells is lost along a
side-side contact, the latter turns into a crack governed by
frictional contact behavior. The loss of cohesion is assumed to y
be irreversible. The cohesive state between cells is managed by
a matrix M. M[i,j ] = 1 if the cells i and j are connected by a x
cohesive contact. Otherwise, we set M[i,j ] = 0. This matrix
d
is updated at each time step according to the evolution of the
contacts.
The simulations were carried out by means of the CD
method, which is suitable for simulating large assemblies of
undeformable particles [51–54]. In this method, the rigid-body F
equations of motion are integrated by taking into account
the kinematic constraints resulting from contact interactions. FIG. 3. (Color online) Boundary conditions for Brazilian crush-
These interactions are characterized by three parameters: the ing of a particle.

022203-3
NGUYEN, AZÉMA, SORNAY, AND RADJAI PHYSICAL REVIEW E 91, 022203 (2015)

y is the y component of the branch vector joining their 2

centers. The averaging runs over all intercell contacts inside 50
σc=10 MPa
the particle. The average horizontal stress σxx  is zero. We σc=20 MPa
40

σa/σc
note that since the particle is rigid, in principle, no axial σc=30 MPa

σa (MPa)
1
deformation must occur until the particle breaks. But some 30 σc=40 MPa
deformation does occur numerically without causing fracture. σc=50 MPa
Such deformations are small and do not affect the stress values, 20
0
which are determined by contact dynamics calculations. Video 0 5 10 15
samples of the simulations analyzed below can be found by 10 εa/εp
following the link www.cgp-gateway.org/ref032.
0
0.000 0.002 0.004 0.006 0.008 0.010
IV. FRACTURE CHARACTERISTICS εa
Figure 4 displays snapshots of a cell-meshed particle
FIG. 5. (Color online) Axial stress versus axial strain for several
at incipient cracking together with compressive and tensile
values of tensile strength σc . The inset shows the axial stress
normalized by σc . The parameters are μs = 0.3, λ = 0.8, and
nv = 50.

forces between cells for two different values of cell-shape

variability parameter λ. The main crack is nearly vertical and
corresponds on average to a mode I fracture as observed in
experiments [76]. However, deviations from the vertical and
zigzag aspect of the main crack, which reflect the coarse
meshing of the particle, indicate that cellular disorder and
friction forces between cells are important for the fracture. We
also observe secondary cracks and small fragments detached
from the particle.
Figure 5 shows axial stress σa as a function of axial
deformation εa for different values of tensile strength σc . The
stress sharply increases with strain and falls off abruptly when
fracture is triggered. The stress peak σp is the compressive
strength of the particle. Except for the lowest values of
cohesion, it scales with σc , as shown in the inset. The higher
value of σp /σc in the low-cohesion limit indicates that the
(a) compressive strength is determined by both the tensile strength
(normal strength) and shear strength, which is enhanced by
interlocking between cells.
In order to quantify the influence of local friction coefficient
μs , a series of crushing tests were conducted with 11 values
of μs from 0 to 1 with all other parameters kept at a fixed
value (σc = 10 MPa, λ = 0.8, and nv = 500). For each value
of μs , nine simulations with independent tessellations were
performed. Figure 6 shows the evolution of the average
compressive strength σp (peak stress) normalized by σc as
a function of μs . We see that, as expected, the compressive
strength increases with friction coefficient. Its dependence
with respect to the friction coefficient is, however, rather weak
since it increases from 0.4σc to 1.2σc as μs is increased from
0 to 1. This weak dependence may be attributed to the fact that
in Brazilian test the rupture occurs in tension. But the effect
of friction coefficient clearly shows that the mobilization of
friction forces along the crack path is an important factor for
compressive strength.
(b)
V. STRENGTH VARIABILITY
FIG. 4. (Color online) A single circular particle subjected to
diametral compression (Brazilian test) for λ = 0 (a) and λ = 0.8 (b). Another issue that we would like to address here is the
The red and green lines represent compressive and tensile contacts, statistical scatter of compressive strength and its possible scale
respectively. dependence. Such a scatter may be a consequence of the BCM.

022203-4
BONDED-CELL MODEL FOR PARTICLE FRACTURE PHYSICAL REVIEW E 91, 022203 (2015)

1.4 10
1

1.2
0
1.0 10

-log(Ps)
σp/σc

λ=0.0
0.8 λ=0.8
-1 m=6.4
10 m=9.7
0.6

0.4
-2
10 0.6 0.8 1.0 1.2 1.4
0.0 0.2 0.4 0.6 0.8 1.0 m
μs (σ/σw) (a)
11
FIG. 6. (Color online) Evolution of compressive strength σp as a
function of local friction coefficient μs for σc = 10 MPa, λ = 0.8, 10
and nv = 500. The error bars represent standard deviation calculated 9
from nine independent simulations with different tessellations.
8

m
7
In particular, it is important to assess the role of cell shape 6
distribution parameter λ. To clarify this point, we carried out
a series of Brazilian tests for five values of λ in the range 5
[0,0.8] with σc = 10 MPa, μs = 0.3, and nv = 1000. For 4
0.0 0.2 0.4 0.6 0.8
each value of λ, 30 simulations with independent tessellations
were performed, allowing us to obtain the variability of λ (b)
the compressive strength, expressed as cumulative survival
0.9
probability Ps of the particle as a function of σa . This is
the probability that the particle does not fail for all stresses σp/σc
σp/σc & σw/σc

below σa . 0.8 σw/σc

Figure 7(a) shows the cumulative survival probability Ps σwΓ{1+1/m}
in log-log scale as a function of the compressive stress σa for 0.7
λ = 0 and λ = 0.8. Despite fluctuations, the data are correctly
fitted by the cumulative Weibull function [77,78], 0.6

Ps (σa ) = e−(σa /σw ) ,

m
(3) 0.5
0.0 0.2 0.4 0.6 0.8

where σw is a stress scale and m is the Weibull modulus (or

λ (c)
shape factor). Both parameters can be extracted from the data. FIG. 7. (Color online) (a) Survival probability of a particle as a
Figure 7(b) shows that m is 6 at low values of λ and increases function of compressive stress σa in a Brazilian test for nv = 500 and
up to 10 for the largest values of λ. It is remarkable that this two values of mesh regularity parameter λ in log-log scale. The lines
range of values of m corresponds to the observed values in represent fits to the Weibull function with two different values of the
soils and powders although our system is purely 2D [78]. The Weibull modulus m. (b) The Weilbull modulus m as function of λ. (c)
increase of m indicates lower strength variability consistently Compressive strength σp and scale stress σw as a function of λ. The
with lower cell variability as λ increases. open symbols are predicted values of σp by Eq. (4).
We also note that the compressive strength σp and scale
stress σw increase with λ as observed in Fig. 7(c). For Weibull
distribution, it can be shown that σp and σw obey the simple VI. SCALE DEPENDENCE
relation The Weibull statistics for brittle materials is generally
  explained by the distribution of microcracks, which concen-
1
σp = σw  1 + , (4) trate stresses in the bulk of the material [79]. Their density
m determines the cracking state and leads to size dependence of
the strength [77]:
where  is the gamma function. This relation is in excellent  −1/m
agreement with our data in Fig. 7(c). Let us remark that  V
the overall Weibull fit to the simulation data reveals also σp = σp , (5)
V
fluctuations in the form of modes as observed in Fig. 7(a).
Such deviations indicate that the survival probability reflects where σp and σp are the strength for samples of volumes
not only the overall disorder but also some fine details of the V and V  , respectively. The applicability of this distribu-
microstructure. tion to quasibrittle materials has been questioned by some

022203-5
NGUYEN, AZÉMA, SORNAY, AND RADJAI PHYSICAL REVIEW E 91, 022203 (2015)

authors and deviations from the Weibull statistics have been

observed [80–83].
Strictly speaking, there are no cracks in our system, and
stress concentration is a consequence of cell-induced disorder.
In this respect, the fracture is analogous to intergranular failure
in polycrystalline materials. Another feature, which makes our
system specific, is that the cells are perfectly rigid and thus
their displacements are subject to compatibility with steric
exclusions between cells. Hence, when the tensile threshold
−σc is reached at a contact between two cells, it will generally
not open unless collectively with other contacts. Since the force δx
at such a critical contact cannot exceed its tensile threshold,
new force increments are redistributed to neighboring contacts.
The loading of the contacts gets therefore accelerated every
time a new contact becomes critical. This stress redistribution
continues until the critical contacts percolate across the
particle, in which case a crack occurs in the sense that all
critical contacts open at the same time. Hence, the fracture
mode differs from both volume fracture and contact fracture
modes described in Ref. [23]. In the simulations, as described
in Sec. II, the force at a critical contact is fc = −σc L/2. The
tensile threshold σc being the same for all contacts, the weakest
contact is the one having shortest length L. However, the FIG. 8. (Color online) Normal force network before rupture. The
above loading process of contacts implies that the cracking dashed lines represent the contours of the stress concentration zone.
of the particle is not controlled by the weakest contact but
requires, on the contrary, the strongest contact (with the longest
length) to become critical on the path of a potential crack. may take a different value depending on the nature of the
Otherwise, none of the critical contacts can open due to network. The number of steps from the poles to the center is
1/2 α/2
kinematic incompatibility. This situation radically contrasts simply proportional to nv , so δx ∝ d0 nv . This length may
with the assumption of weakest link [84]. Nevertheless, the be interpreted as the width of the zone at the center of the disk
above argument also indicates that the “flaws” in our system in which the stresses are concentrated, as illustrated in Fig. 8.
are equivalent to potential crack paths rather than single critical Hence, the compressive stress at the center of the particle is
α/2 (1−α)/2
contacts between cells. In other words, considering only the ∝F /δx ∝ σa d/nv ∝ σa nv .
lengths of the contacts, the weakest “link” in our system is The intercellular forces are actually larger in the vicinity
the path of intercell contacts having the shortest length. In of the poles and one might expect failure to be initiated at the
this sense, the variability of compressive strength reflects poles. But, as argued above, the crack is not effective as long as
the statistics of all paths (with their different lengths and the cohesion threshold is not reached inside the whole range
directions), on one hand, and the heterogeneous distribution defined by the random walk. Hence, in this “traction zone,”
of forces, as generally observed in granular materials, on the failure occurs when the weakest compressive stress becomes
other hand [56,69,85,86]. critical. The weakest compressive stress occurs in the center
The statistics of critical contacts and crack paths in our where the compressive forces are dispersed. This stress with
system may naturally lead to size effect but its description σa = σp is balanced by a tensile stress equal to σc in the y
cannot be based on the elasticity of the material. Here we direction, so F /δx ∝ σc , thereby
introduce a simple model and compare the results with those σp
of Ref. [23]. The number of potential paths of critical contacts ∝ n−(1−α)/2
v ∝ d −(1−α) . (6)
σc
increases with the number nv ∝ (d/d0 )2 of cells as a result of
either the increase of particle diameter d for a constant cell This model predicts a size dependence with exponent
size d0 or the decrease of d0 for a given particle diameter d. −(1 − α).
In the continuum-mechanics limit, one expects the particle to Figure 9(a) shows σp as a function of nv for λ = 0.8 and
fracture into two equal fragments along its diameter. In the 16 different combinations of the values of d and d0 with 11
presence of the cells, the crack follows a more complex path independent tests for each combination. Within our statistical
around this mean orientation imposed by the correlations, but precision, we observe a power law σp ∝ σc n−b v with b 
the fluctuations around the mean may well be assumed to result 0.24. This yields α  1/2, which corresponds to a normal
from a stochastic process reflecting disorder. Let us assume random walk. The scale stress σw follows the same behavior
that the cracks are random walks of equal length d0 from as a function of nv . Figure 9(b) shows σp as a function of nv
the top and bottom poles towards the center of the particle. for λ = 0 and λ = 0.8. The exponent for λ = 0 is b  0.3,
Then the mean path is a straight line joining the poles and the which corresponds to α  0.4. This “subdiffusive” feature of
standard deviation from this line at the center of the particle crack paths may simply be attributed to the fact that, for λ = 0,
varies as δx ∝ nαs , where ns is the number of steps. For a there is no constraint on the cell sizes and their shapes so the
normal random walk we have α = 1/2. But the exponent α critical contacts are more likely to occur in more complex

022203-6
BONDED-CELL MODEL FOR PARTICLE FRACTURE PHYSICAL REVIEW E 91, 022203 (2015)

5.0 strength properties of circular particles subjected to axial

compression. The particles are modeled as aggregates of
d constant polygonal rigid cells bonded by the action of cohesion forces
d0 constant along their common sides. Our approach allows for both the
conservation of volume and rigorous treatment of unilateral
σp/σc

1.0 constraints at intercellular contacts.

The compressive strength was shown to scale well with
tensile threshold between cells. However, due to the Mohr-
Coulomb plastic criterion and interlocking between rigid
cells, the strength is also an increasing function of the
0.2 friction coefficient. By means of extensive simulations, we
10 100 1000
nv also performed a detailed parametric analysis of strength
variability. The statistical scatter of the data is well described
(a) by the Weibull distribution function. This distribution provides
5.0 a good fit to our data independently of cell shape distribution
but the Weibull modulus varies from 6 to 10 as the cell shape
λ=0.0
λ=0.8 span is reduced.
b=0.24 The Weibull distribution of the data in our system was
discussed and modeled in terms of critical intercellular
σp/σc

1.0 contacts and their percolation across the particle. The fracture
b=0.30 is controlled by the population of weakest paths from the
force application point to the center of the particle. Assuming
that those paths are composed of random walks through
intergranular contacts, they define a stress concentration zone
0.2
10 100 1000 with a high density of critical contacts. This leads to a power-
nv law particle size dependence of the compressive strength, in
(b) close agreement with our simulation data. The value of the
exponent in the case of nearly regular cells is consistent with a
FIG. 9. (Color online) (a) Compressive strength of a particle normal random-walk feature of the crack, whereas for irregular
normalized by internal cohesion as a function of the number of cells cells it is anomalous.
for constant particle diameter d and varying cell size d0 (green circle) The Brazilian compression test is often used for indirect
and for constant d0 and varying d (red square). The error bars represent measurement of tensile strength in brittle materials [87,88].
standard variation for 11 independent simulations. (b) Compressive But the tensile strength measured from diametral compres-
strength as a function of the number of cells for two extreme values sion tests are usually lower compared to other uniaxial
of mesh variability parameter λ. tests. In contrast to theoretical prediction, the cracks do
not always propagate from the center to the periphery of
configurations. The exponent predicted by elastic theory is the sample as a result of surface defects, which lead to
b = 1 − 2/m for volume fracture [23]. This implies 1 − α = failure by shearing at the contact points with platens. Our
2/m, which is consistent for λ = 0 and λ = 0.8 for which simulations are consistent with this picture although the cells
we have m  6 and m  10, thus yielding 1 − α  1/3 and are rigid and the crack opens only when critical contacts
1 − α  1/5, respectively. percolate across the particle. For this reason, we also have
These observations show clearly that the fracture of cell- a good scaling of the compressive strength with internal
structured materials is scale dependent. The model based cohesion.
on crack-path statistics around the mean, as briefly outlined A detailed description of single-particle fracture in this
above, provides quantitative prediction of the exponent. What paper was made possible by extensive 2D simulations. It is
is more, it generalizes the weakest-link assumption to the straightforward to extend this work to investigate the role of the
more general “weakest-path” mechanism governed by the contact law such as non-Coulomb friction and damage for the
percolation of critical contacts. Size effect in single-particle scaling of compressive strength. Another possible extension
fracture suggests that in an assembly of crushable particles is the fracture of noncircular particles. We applied the BCM
the largest particles are most susceptible to break. However, approach to the fragmentation of an assembly of polygonal
particle size affects the local distribution of contact forces. In particles. Our simulations reproduce correctly and efficiently
particular, large particles have more contacts and sustain for the nonlinear and inhomogeneous features of the comminution
this reason lower deviatoric stresses. Due to such competing process such as the shattering instability and survival of many
effects, the fragmentation of a granular packing is a complex large particles. Size effect in single-particle fracture suggests
process. that in an assembly of crushable particles the largest particles
are most susceptible to breakage. However, particle size affects
the local distribution of contact forces. In particular, large
VII. CONCLUSION
particles have more contacts and for this reason sustain lower
In this paper, we introduced a BCM in the framework of deviatoric stresses. The results of this work will be reported
the contact dynamics method for the investigation of fracture elsewhere.

022203-7
NGUYEN, AZÉMA, SORNAY, AND RADJAI PHYSICAL REVIEW E 91, 022203 (2015)

[1] C. Thornton, K. K. Yin, and M. J. Adams, J. Phys. D: Appl. [33] D. Potyondy and P. Cundall, Int. J. Rock Mech. Min. Sci. 41,
Phys. 29, 424 (1996). 1329 (2004).
[2] D. Fuerstenau, O. Gutsche, and P. Kapur, in Comminution 1994, [34] Y. P. Cheng, M. D. Bolton, and Y. Nakata, Géotechnique 54,
edited by K. Forssberg and K. Schönert (Elsevier, Amsterdam, 131 (2004).
1996), pp. 521–537. [35] N. Cho, C. Martin, and D. Sego, Int. J. Rock Mech. Min. Sci.
[3] C. Couroyer, Z. Ning, and M. Ghadiri, Powder Technol. 109, 44, 997 (2007).
241 (2000). [36] M. Khanal, W. Schubert, and J. Tomas, Miner. Engin. 20, 179
[4] A. V. Potapov and C. S. Campbell, Powder Technol. 120, 164 (2007).
(2001). [37] S. Abe and K. Mair, Geophys. Res. Lett. 36, L23302 (2009).
[5] Y. Nakata, M. Hyodo, A. F. Hyde, Y. Kato, and H. Murata, Soils [38] J. Wang and H. Yan, Soils Found. 52, 644 (2012).
Found. 41, 69 (2001). [39] G. Timár, F. Kun, H. A. Carmona, and H. J. Herrmann, Phys.
[6] P. Cleary, Miner. Eng. 14, 1295 (2001). Rev. E 86, 016113 (2012).
[7] M. D. Bolton, Y. Nakota, and Y. P. Cheng, Géotechnique 58, [40] M. J. Metzger and B. J. Glasser, Powder Technol. 217, 304
471 (2008). (2012).
[8] C. Hosten and H. Cimilli, Int. J. Miner. Process. 91, 81 [41] T. Ueda, T. Matsushima, and Y. Yamada, Granul. Matter 15, 675
(2009). (2013).
[9] L. Liu, K. Kafui, and C. Thornton, Powder Technol. 199, 189 [42] J. Astrom and H. Herrmann, Eur. Phys. J. B 5, 551 (1998).
(2010). [43] O. Tsoungui, D. Vallet, and J. Charmet, Powder Technol. 105,
[10] J. Jaeger, in International Journal of Rock Mechanics and 190 (1999).
Mining Sciences & Geomechanics Abstracts (Elsevier, [44] O. Ben-Nun, I. Einav, and A. Tordesillas, Phys. Rev. Lett. 104,
Amsterdam, 1967), Vol. 4, pp. 219–227. 108001 (2010).
[11] K. L. Lee and I. Farhoomand, Can. Geotech. J. 4, 68 (1967). [45] L. Elghezal, M. Jamei, and I.-O. Georgopoulos, Granul. Matter
[12] B. O. Hardin, J. Geotech. Eng. 111, 1177 (1985). 15, 685 (2013).
[13] M. Hagerty, D. Hite, C. Ullrich, and D. Hagerty, J. Geotech. [46] V. Esnault and J.-N. Roux, Mech. Mater. 66, 88 (2013).
Eng. 119, 1 (1993). [47] F. Kun and H. J. Herrmann, Comput. Methods Appl. Mech. Eng.
[14] M. Eriksson and G. Alderborn, Pharm. Res. 12, 1031 (1995). 138, 3 (1996).
[15] P. V. Lade, J. A. Yamamuro, and P. A. Bopp, J. Geotech. Eng. [48] B. Van de Steen, A. Vervoort, and J. Napier, Int. J. Fract. 108,
122, 309 (1996). 165 (2001).
[16] M. R. Coop, K. K. Sorensen, T. B. Freitas, and G. Georgoutsos, [49] G. D’Addetta, F. Kun, and E. Ramm, Granul. Matter 4, 77
Géotechnique 54, 157 (2004). (2002).
[17] H. Arslan, G. Baykal, and S. Sture, Granul. Matter 11, 87 (2009). [50] S. Galindo-Torres, D. Pedroso, D. Williams, and L. Li, Comput.
[18] B. Imre, J. Laue, and S. M. Springman, Granul. Matter 12, 267 Phys. Commun. 183, 266 (2012).
(2010). [51] J. Moreau, European J. Mech. A Solids 13, 93 (1994).
[19] V. Bandini and M. R. COOP, Soils Found. 51, 591 (2011). [52] F. Radjaı̈ and V. Richefeu, Mech. Mater. 41, 715 (2009).
[20] A. Ezaoui, T. Lecompte, H. Di Benedetto, and E. Garcia, Granul. [53] F. Radjaı̈ and F. Dubois, Discrete Numerical Modeling of
Matter 13, 283 (2011). Granular Materials (Wiley-ISTE, New York, 2011).
[21] F. Casini, G. M. Viggiani, and S. M. Springman, Granul. Matter [54] M. Jean, Comput. Methods Appl. Mech. Eng. 177, 235 (1999).
15, 661 (2013). [55] J. Moreau, Eur. J. Mech. A Solids Supp. 13, 93 (1994).
[22] J. Huang, S. Xu, and S. Hu, Mech. Mater. 68, 15 (2014). [56] F. Radjai, M. Jean, J.-J. Moreau, and S. Roux, Phys. Rev. Lett.
[23] O. Tsoungui, D. Vallet, J. Charmet, and S. Roux, Granul. Matter 77, 274 (1996).
2, 19 (1999). [57] L. Staron, J.-P. Vilotte, and F. Radjai, Phys. Rev. Lett. 89, 204302
[24] H. Liu, S. Kou, and P.-A. Lindqvist, Mech. Mater. 37, 935 (2002).
(2005). [58] A. Taboada, K. J. Chang, F. Radjaı̈, and F. Bouchette, J. Geophys.
[25] W. Schubert, M. Khanal, and J. Tomas, Int. J. Miner. Process. Res. 110, 1 (2005).
75, 41 (2005). [59] M. Renouf and P. Alart, Comput. Methods Appl. Mech. Eng.
[26] A. Bäckström, J. Antikainen, T. Backers, X. Feng, L. Jing, 194, 2019 (2005).
A. Kobayashi, T. Koyama, P. Pan, M. Rinne, B. Shen et al., [60] E. Azéma, F. Radjaı̈, R. Peyroux, F. Dubois, and G. Saussine,
Int. J. Rock Mech. Min. Sci. 45, 1126 (2008). Phys. Rev. E 74, 031302 (2006).
[27] R. Moreno, M. Ghadiri, and S. Antony, Powder Technol. 130, [61] E. Azéma, F. Radjaı̈, R. Peyroux, V. Richefeu, and G. Saussine,
132 (2003). Eur. Phys. J. E 26, 327 (2008).
[28] S. Antonyuk, M. Khanal, J. Tomas, S. Heinrich, and L. Mörl, [62] N. Estrada, A. Taboada, and F. Radjaı̈, Phys. Rev. E 78, 021301
Chem. Eng. Process. 45, 838 (2006). (2008).
[29] F. Wittel, H. Carmona, F. Kun, and H. Herrmann, Int. J. Fract. [63] E. Azéma and F. Radjaı̈, Phys. Rev. E 81, 051304 (2010).
154, 105 (2008). [64] E. Azéma and F. Radjaı̈, Phys. Rev. E 85, 031303 (2012).
[30] J. Wang and H. Yan, Int. J. Numer. Anal. Methods Geomech. [65] N. Estrada, E. Azéma, F. Radjai, and A. Taboada, Phys. Rev. E
37, 832 (2013). 84, 011306 (2011).
[31] G. Ma, W. Zhou, and X.-L. Chang, Comput. Geotech. 61, 132 [66] V. Visseq, A. Martin, D. Iceta, E. Azéma, D. Dureisseix, and
(2014). P. Alart, Comput. Mech. 49, 709 (2012).
[32] Y. Cheng, Y. Nakata, and M. Bolton, Geotechnique 53, 633 [67] B. Saint-Cyr, J.-Y. Delenne, C. Voivret, F. Radjai, and P. Sornay,
(2003). Phys. Rev. E 84, 041302 (2011).

022203-8
BONDED-CELL MODEL FOR PARTICLE FRACTURE PHYSICAL REVIEW E 91, 022203 (2015)

[68] J. C. Quezada, P. Breul, G. Saussine, and F. Radjai, Phys. Rev. [78] G. R. McDowell and M. D. Bolton, Géotechnique 48, 667
E 86, 031308 (2012). (1998).
[69] C. Voivret, F. Radjaı̈, J.-Y. Delenne, and M. S. El Youssoufi, [79] A. A. Griffith, Phil. Trans. Roy. Soc. 221, 163 (1920).
Phys. Rev. Lett. 102, 178001 (2009). [80] S. van der Zwaag, ASTM J. Test. Eval. 17, 292 (1989).
[70] D. Kadau, G. Bartels, L. Brendel, and D. E. Wolf, Comput. Phys. [81] X. Gao, R. Dodds, R. Tregoning, J. Joyce, and R. Link, Fatigue
Commun. 147, 190 (2002). Fract. Eng. Mater. Struct. 22, 481 (1999).
[71] I. Bratberg, F. Radjai, and A. Hansen, Phys. Rev. E 66, 031303 [82] Y. Nakata, Y. Kato, M. Hyodo, A. F. HYDE, and H. Murata,
(2002). Soils Found. 41, 39 (2001).
[72] D.-H. Nguyen, E. Azéma, F. Radjai, and P. Sornay, Phys. Rev. [83] Z. Bertalan, A. Shekhawat, J. P. Sethna, and S. Zapperi, Phys.
E 90, 012202 (2014). Rev. Appl. 2, 034008 (2014).
[73] E. Azéma, N. Estrada, and F. Radjaı̈, Phys. Rev. E 86, 041301 [84] S. Batdorf and H. Heinisch, J. Am. Ceram. Soc. 61, 355
(2012). (1978).
[74] G. Saussine, C. Cholet, P. Gautier, F. Dubois, C. Bohatier, and [85] R. P. Behringer, K. E. Daniels, T. S. Majmudar, and M. Sperl,
J. Moreau, Comput. Methods Appl. Mech. Eng. 195, 2841 R. Soc. A 366, 493 (2008).
(2006). [86] V. Richefeu, Moulay Saı̈d El Youssoufi, and F. Radjaı̈, Phys.
[75] L. Staron, F. Radjaı̈, and J.-P. Vilotte, Eur. Phys. J. E 18, 311 Rev. E 73, 051304 (2006).
(2005). [87] M. K. Fahad, J. Mater. Sci. 31, 3723 (1996).
[76] K. Schönert, Powder Technol. 143-144, 2 (2004). [88] A. T. Procopio, A. Zavaliangos, and J. C. Cunningham, J. Mater.
[77] W. Weibull, J. Appl. Mech. 18, 293 (1951). Sci. 38, 3629 (2003).

022203-9