See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/44003760

The ESyS_Particle: A New 3-D Discrete Element Model with Single Particle
Rotation

Article  in  Lecture Notes in Earth Sciences · January 2009

DOI: 10.1007/978-3-540-85879-9_6 · Source: OAI

CITATIONS READS

21 537

2 authors, including:

Yucang Wang
Central Queensland University
76 PUBLICATIONS   1,005 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Yucang Wang on 25 February 2014.

The user has requested enhancement of the downloaded file.

VI. The ESyS_Particle: A New 3-D Discrete
Element Model with Single Particle Rotation

Yucang Wang and Peter Mora

ESSCC, The University of Queensland, St. Lucia QLD 4072, Brisbane,

Australia

Abstract In this paper, the Discrete Element Model (DEM) is reviewed, and the
ESyS_Particle, our new version of DEM, is introduced. We particularly highlight
some of the major physical concerns about DEMs and major differences between
our model and most current DEMs. In the new model, single particle rotation is
introduced and represented by a unit quaternion. For each 3-D particle, six degrees
of freedom are employed: three for translational motion, and three for orientation.
Six kinds of relative motions are permitted between two neighboring particles, and
six interactions are transferred, i.e., radial, two shearing forces, twisting and two
bending torques. The relative rotation between two particles is decomposed into
two sequence-independent rotations such that all interactions due to the relative
motions between interactive rigid bodies can be uniquely determined. This
algorithm can give more accurate results because physical principles are obeyed.
A theoretical analysis about how to choose the model parameters is presented.
Several numerical tests have been carried out, the results indicate that most
laboratory tests can be well reproduced using our model.

1 Introduction: A Review of the Discrete Element Method

The Discrete Element Method (DEM), pioneered by Cundall (Cundall

et al., 1979), has been a powerful numerical tool in many scientific and
engineering applications. The basic idea behind DEM is to treat the sample
to be modeled as an assemblage of discrete particles interacting with one
another. At each time step, the calculations performed in DEM alternate
between integrating equations of motion for each particle, and applying the
force-displacement law at each contact.

H. Xing, Advances in Geocomputing, Lecture Notes in Earth Sciences 119,

DOI 10.1007/978-3-540-85879-9_6
© Springer-Verlag Berlin Heidelberg 2009
184 Y. Wang and P. Mora

One of the major advantages of DEM is that highly complex systems

can be modeled using basic methodologies without any assumptions on the
constitutive behaviors of the materials and any predisposition about where
and how cracks may occur and propagate. This advantage allows DEM to
study with great simplicity many problems which are highly dynamic with
large deformations and a large number of frequently changing contacts.
These simulations include blast and impact of brittle materials at high
strain rates (Donze et al., 1997; Magnier et al., 1998; Hentz et al., 2004a;
2004b), rock fractures (Donze et al., 1997; Young et al., 2000; Hazzard
et al., 2000b; 2000c; Wang et al., 2000; place et al., 2001; 2002; Boutt
et al., 2002; Chang et al., 2002; Matsuda et al., 2002; Hunt et al., 2003;
Potyondy et al., 1996; 2004), soil mechanics and shear bands (Bardet et al.,
1991; Anandarajah, 1994; Ng et al., 1994; Iwashita et al., 1998; 2000; Oda
et al., 1998; 2000; Thornton et al., 2000; 2003; 2006; McDowell et al.,
2002; Cheng et al., 2003; Tordesillas et al., 2004; Jiang et al., 2006),
earthquake faults and gouges (Mora et al., 1993; 1994; 1998; 1999; 2000;
2002a; 2002b; Scott 1996; Morgan, 1999; Morgan 2004; Place et al., 1999;
2000; 2001; Wang et al., 2000; 2004; Hazzard et al., 2000a; 2002; 2004;
Latham et al., 2006; Hu et al., 2004), frictional instability (Mora et al.,
1994; Guo et al., 2004), granular or powder flow (Schwarz et al., 1998;
Langston et al., 2004; Sheng et al., 2004), landslides and slope instability
(Chang, 1992; Cleary et al., 1993; Campbell et al., 1995; Eberhardt et al.,
2005) and mining and other industrial applications (Holst et al., 1999;
Huang et al., 1999; Cleary, 2000; Cleary et al., 2002; Morrison et al.,
2004; 2007; Prochazka, 2004).
In the past three decades, DEMs have been greatly developed and many
different types of DEMs have appeared. Although the basic concepts are
similar, these DEMs vary in the following aspects, all of which are the
major physical concerns in DEM simulations and affect the simulated
dynamics in various ways.

1.1 Dimensionality: 2-D or 3-D

Much of the early work on DEM was restricted to 2-D mainly because of
computational restrictions and the geometric simplicity of the 2-D case. In
some cases however 3-D simulations differ from their 2-D counterpart not
only quantitatively but also qualitatively. Numerical and laboratory tests
suggest that there are some significant differences regarding macroscopic
friction and wing crack extensions between the 2-D and 3-D cases
(Hazzard et al., 2003; Dyskin et al., 2003). As will be demonstrated later
in this paper, it is more difficult to theoretically deal with finite particle
 The ESyS_Particle 185

rotation in 3-D than in 2-D. In recent years 3-D DEM simulations have
become more common thanks to an increase in computer speeds.

1.2 Contact Laws: Linear or Non-Linear

The simplest contact law is the linear contact law, in which contact
stiffnesses are constants and independent of the relative displacement at
contacts. It is found that this law can not reproduce the realistic
phenomena in certain circumstances (such as cycling loading (Dorby et al.,
1992)). Non-linear contact laws, mainly based on Hertz and Mindlin
theory (Mindlin et al., 1953; Johnson, 1987) provide more successful
results in such circumstances (Morgan et al., 1999; Yang et al., 2000; Guo
et al., 2004; Li et al., 2005).

1.3 Particle Shapes: Disks/Spheres or

Polygons/Polyhedrons
Most DEMs employed disks (in 2-D) and spheres (in 3-D) due to their
simplicity. It has been noted that disks and spheres tend to roll or rotate
more easily. For this reason more complex shapes such as ellipse (Ting
et al., 1993), ellipsoid (Lin et al., 1997), polygons (Issa et al., 1992;
Matuttis et al., 2000; D’Addetta et al., 2002; Feng et al., 2004b), polyhedra
(Hart et al., 1988; Cundall, 1988; Ghaboussi et al., 1990) and superquadric
(Mustoe, 1992; Hogue, 1998) are used. However polygons and
polyhedrons introduce some disadvantages, these include difficulties in
detecting contacts and calculating forces and torque in cases of edge–edge,
edge–corner, corner–corner contact, and difficulties in extending from 2-D
to 3-D and bonding particles together. Another alternate may be
aggregates or clumps of disks and spheres bonded together or cluster
(Jensen et al., 1999), agglomerates (Ning et al., 1997; Cheng et al., 2003;
Lu et al., 2007). All of the disadvantages discussed above can be overcome
without too much effort and crushing and fracture of aggregates can be
easily modeled. The disadvantage is that this method requires a larger
number of particles. However with developing computing power and
paralleled techniques, this will not be a major problem in the future.

1.4 Single Particle Rotation: With or Without

In some DEMs, single particle rotations are prohibited. However, recent
studies suggest that in some cases unrealistic results are generated if single
186 Y. Wang and P. Mora

particle rotation is ignored (Wang et al., 2008a). The reason behind this is
that the rotation of a particle provides the mechanisms for local
transmission of shear forces and moments that are not present in non-
rotational DEM simulations.
Particle rotation in 2-D can be represented easily using a scalar
orientation and single angular velocity. In 3-D orientations require three
independent parameters and explicit representations for such orientations
are needed. By explicit, we mean that once the explicit variable is given,
the orientation of a 3-D particle is uniquely determined. Candidates of
such variants include matrix, quaternion, vector, Euler Angles.
In some existing DEMs, particle orientations are implicitly represented
using three angular velocities around three orthogonal axes. As will be
discussed in Sect. 6, this method has its shortcomings, because in 3-D one
can not extract the orientation of a particle by simply integrating three
angular velocities, limited by the physical principle. In the ESyS_Particle,
unit quaternion is used as an explicit representation of particle orientation
(Wang et al., 2006). To our knowledge, the few DEMs have employed
quaternion (Johnson et al., 2007; Fleissner et al., 2007).

1.5 Algorithm for Integrating the Equations of Motion

Typical integration methods used in DEMs are the Molecular Dynamic
algorithm, such as Velocity Verlet and leapfrog (Allen et al., 1987). If
rotation is involved, integration of rotational equations should also be
considered, including updating angular velocities and updating the
orientation degree of freedom (Dullweber et al., 1997; Kol et al., 1997;
Omelyan, 1998a; 1998b; Buss, 2000; Miller et al., 2002; Munjiza et al.,
2003; Krysl et al., 2005; Wang, 2008b).

1.6 Bonded or Not Bonded

In the bonded (cohesive) DEMs, particles are bonded together so that
tensile forces can be transmitted, but in the un-bonded (cohesionless)
DEMs, only repulsive forces can be transmitted between particles. The
former is often used to model wave propagation and fracture of intact
materials such as rocks (Mora et al., 1993; 1994; 1998; Place et al., 1999;
2002; Chang et al., 2002; Hazzard et al., 2000a; 2000b; Potyondy et al.,
1996; 2004; Donze et al., 1997; Hentz et al., 2004a; 2004b). While the
latter, or “real discrete” model, is used to model motions of power and
behaviors of particulate materials (Morgan et al., 1999; Sheng et al., 2004;
Sitharam, 2000).
 The ESyS_Particle 187

1.7 Interactions Between Particles: Complete

or Simplified
Theoretically, for 2-D bonded models, three interactions (normal force,
shear force and a bending moment) between two bonded particles should
be permitted; whereas in 3-D case, six interactions should be permitted
(i.e. normal force, two shear forces, two bending moments and a twisting
moment, see detail in Sect. 3.1.1). Under certain circumstances it is a good
approximation to use a simple force-displacement law and ignore certain
interactions, such as models in which only normal forces exist (Mora et al.,
1993; 1994; 1998; Toomey, et al., 2000), or models in which only normal
and shear forces are transmitted (Scott, 1996; Chang et al., 2002; Hazzard
et al., 2000a; 2000b). However studies from experiments and numerical
simulations suggest that rolling resistance at contacts has a significant
influence on the behaviors of granular media (Oda et al., 1982; 1998;
2000; Schlangen et al., 1997; Iwashita et al., 1998; 2000; Tordesillas et al.,
2002; Kuhn et al., 2004). When only radial forces are transmitted between
bonded particles, or rolling resistance is absent, the widely-observed
laboratory tests of wing-crack extension can not reproduced (Wang et al.,
2008a). Based on these analyses, DEM software, the ESyS_Particle, has
been extended to include the complete set of interactions in 3-D (Wang
et al., 2006; Wang, 2008b). We point out that the calculation of these
interactions is important to the accuracy of simulations.

1.8 Criterion for Bond Breakage

For bonded DEMs, a criterion is required to judge the breakage of bonds.
The widely used criterion is that a bond breaks either when the tensile
strength or the shear strength is reached (either f r ≥ Fr0 or f s ≥ Fs0 )
(Potyondy et al., 1996; Donze et al., 1997; Magnier et al., 1998). A
combined criterion, which takes into account the effect of normal force on
shear failure, is used in the ESyS_Particle (Eq. (21)). A similar but
parabolic-type criterion is employed by Pelenne et al. (Pelenne et al.,
2004). Several different criteria are presented by Davie et al. (Davie et al.,
2003). More work is required to compare the different types of criteria.

1.9 Frictional Forces

Frictional interactions play important roles in DEM simulations,
particularly for granular flow. DEMs commonly employ a simple
188 Y. Wang and P. Mora

Coulomb friction model in which a shear stiffness is introduced for

touching particles that resists tangential movement between the particles
until shear force exceeds a threshold (Matuttis et al., 2000; Hazzard et al.,
2000c; Matsuda et al., 2002; Cheng et al., 2003; Hentz et al., 2004a;
2004b; Delenne et al., 2004; Potyondy et al., 2004).
In the earlier version of ESyS_Particle, a rate- and state- dependent
fictional law (motivated by laboratory studies of frictional sliding) was
implemented (Abe et al., 2002) but found to be too computationally
expensive. It is also unclear whether this law is the basic friction
mechanism at particle scale, or it is an emergent phenomenon which
should be reproduced at macroscopic scale. In the current version, a simple
stick-slip frictional algorithm is used which accurately captures the
transition between static and sliding frictional states. Besides, rolling
friction is considered by some modelers (Zhang et al., 1999; Zhou et al.,
1999; Feng et al., 2004a).

1.10 Parameter Calibration

Values for the micro-physical elastic stiffness parameters in DEM
simulations are typically chosen via empirical or trial-and-error methods
(Chang et al., 2002; Hazzard et al., 2000a; 2000b; Boutt et al., 2002;
Matsuda et al., 2002). Empirical calibration involves simulation of uni-
axial or tri-axial tests on a particle assembly and measurement of the
macroscopic elastic properties (Young’s modulus and Poisson’s ration) in
a numerical analogue for the laboratory studies with real materials.
Sometimes researchers outside DEM community have the wrong
impression that DEM parameters are chosen arbitrarily and DEM does not
generate the realistic macroscopic elasticity, as continuum models such as
the Finite Element Method do. For regular lattices of equal-sized spheres it
is possible to derive relations between the microscopic parameters (particle
scale stiffnesses) and macroscopic elastic properties (Wang et al., 2008c).
Choice of micro-physical fracture parameters is more difficult, requiring
empirical calibration against laboratory results.
Other major challenges in DEMs include coupling between different
physical processes such as heat, pore flow (Abe et al., 2000; Sakaguchi et
al., 2000) and coupling between DEM and Finite Element Method
(Munjiza et al., 1995; Owen et al., 2001).
In this paper, we will give a detailed description of the ESyS_Particle,
our DEM. The paper is arranged as follows: the basic features of the
model, and a detailed description, including equations and algorithm to
integrate the equations (rotational) are presented in Sect. 2; Sect. 3
 The ESyS_Particle 189

introduces three kinds of contacts and the ways to calculate the forces and
torques. In Sect. 4 model parameters are discussed; Sect. 5 gives some of
the simulations and Sect. 6 highlights the major differences between our
model and the most existing DEMs, followed by conclusions in Sect. 7.

2 The Model, Equations and Numerical Algorithms

to Integrate These Equations

2.1 A Brief Introduction to the ESyS_Particle

The ESyS_Particle, previously called Lattice Solid Model or LSMearth, is
similar to the Discrete Element Model (Cundall et al., 1979) and Molecular
Dynamics (MD, Allen et al., 1987; Rapaport, 1995) but involves a
different computational approach (Place et al., 1999). It has been applied
to the study of physical process such as rock fracture (Mora et al., 1993;
Place et al., 2002), stick-slip friction behavior and friction (Mora et al.,
1994; Place et al., 1999), granular dynamics (Mora et al., 1998; 1999),
heat-flow paradox (Mora et al., 1998; 1999), localization phenomena
(Place et al., 2000), Load-Unload Response Ratio theory (Mora et al.,
2002b; Wang et al., 2004) and Critical Point systems (Mora et al., 2002a).
In the current ESyS_Particle model, particles can be disks (2-D) or
spheres (3-D). Single particle rotations have been introduced recently and
are represented by unit quaternion (Wang et al., 2006), therefore each
particle brings three degrees of freedom in 2-D, and six in 3-D. There are
three kinds of basic contacts when two particles touch: bonded, Solely
normal repulsive and cohesionless frictional contact. In bonded contacts,
the full set of interactions are involved, that is, all three interactions
(normal, shearing forces and bending moment) are transmitted in 2-D and
six (normal, shearing forces, bending and twisting moment) in 3-D are
transmitted between each bonded particle pair. A new technique has been
developed to decompose the relative rotations between two rigid bodies in
such a way that the torques and forces caused by such relative rotations
can be uniquely determined (Wang, 2008b). In the case of the solely
normal repulsive contacts, only normal forces exist when two particle
contact, and for cohesionless frictional contacts, frictional forces appear in
tangential directions.
The source code is written in C++ with a Python script interface. Before
running the code, the initial conditions, physics parameters, integration
steps, types of particles (simple or rotational particle), types of loading
walls, the contact properties (elastic, frictional, bonded contacts), artificial
190 Y. Wang and P. Mora

viscosity, ways of loading (force controlled or displacement controlled)

and output fields are specified in the script.
Pre-processing includes a particle generation package, which can
generate regular or random sized particles (initial position, radius and a
unique identifier for each particle) and bond information (if needed).
Aggregates (grains), gouges and faults can also be made. Post-processing
includes Povray and VTK visualization packages, which can visualize the
article and fields (velocity, displacements). The code is Parallelized using
MPI, which significantly increases the computational capabilities.

2.2 Equations
Particle motion can be decomposed into two completely independent parts,
translational motion of the center of mass and rotation about the center of
mass. The former is governed by the Newtonian equation
r(t ) = f (t ) M (1)
where r (t ) , f (t ) and M are position of the particle, total forces acting on
the particle and the particle mass respectively. A velocity Verlat scheme
has been used to integrate the equation above (Allen et al., 1987; Mora
et al., 1994; Place et al., 1999).
The particle rotation depends on the total applied torque and usually
involves two coordinate frames, one is fixed in space, called space-fixed
frame, in which Eq. (1) is applied. The other is attached to the principal
axes of the rotation body, referred to as body-fixed frame. In the body-
fixed frame, the dynamic equations are (Goldstein, 1980)
τ xb ⎛I −I ⎞
ω xb = + ⎜⎜ yy zz ⎟⎟ω ybω zb
I xx ⎝ I xx ⎠
τ yb ⎛ I − I xx ⎞ b b
ω =
b
+ ⎜ zz ⎟ω z ω x
y
I yy ⎜ I ⎟
⎝ yy ⎠
τ zb ⎛ I xx − I yy ⎞ b b
ω =
b
z + ⎜⎜ ⎟⎟ω x ω y (2)
I zz ⎝ I zz ⎠
where τ xb , τ yb and τ zb b
are components of the total torque τ expressed in

body-fixed frame, ω x , ω y and ω zb

b b
are components of the angular
 The ESyS_Particle 191

b
velocities ω measured in the body-fixed frame, and I xx , I yy and I zz are
the three principle moments of inertia in body-fixed frame in which the
inertia tensor is diagonal. For 3-D spheres, I = I xx = I yy = I zz .
Due to the singularity caused by Euler angles, the unit quaternion
q = q0 + q1i + q2 j + q3 k is usually used to describe the orientation in
numerical simulations (Evans, 1977; Evans et al., 1977), where i, j, k
satisfy i 2 = j 2 = k 2 = ijk = −1 , ij = − ji = k , jk = − kj = i , ki = −ik = j and
3

∑q
i =0
2
i =1.

The physical meaning of a quaternion is that it represents a one-step

rotation around the vector q1iˆ + q 2 ˆj + q 3 kˆ with a rotation angle of
2 arccos ( q 0 ) (Kuipers, 1998). It is a very convenient tool in that sequences
of rotations can be represented as a quaternion product (Kuipers, 1998).
A quaternion for each particle satisfies the following equations of
motion (Evans, 1977; Evans et al., 1977),
1
Q& = Q0 (q ) Ω , (3)
2
⎛ q&0 ⎞ ⎛ q0 − q1 − q2 − q3 ⎞ ⎛ 0⎞
⎜ ⎟ ⎜ ⎟ ⎜ b⎟
⎜ q&1 ⎟ ⎜ q1 q0 − q3 q2 ⎟ ⎜ ωz ⎟
where Q& = ⎜ ⎟ , Q0 ( q) = ⎜ ⎟ , Ω = ⎜ b⎟.
q& q q3 q0 − q1 ω
⎜ 2⎟ ⎜ 2 ⎟ ⎜ zb ⎟
⎜ q& ⎟ ⎜q − q2 q1 q0 ⎟⎠ ⎜ω ⎟
⎝ 3⎠ ⎝ 3 ⎝ z⎠

2.3 Algorithms to Integrate the Equations of Rotation

Now Eqs. (2) and (3) can be solved numerically and the numerical method
we use is outlined below (Fincham, 1992).
In the numerical time integration scheme we obtain the quaternion
q(t + dt ) at the next time step using:
dt 2
q (t + dt ) = q(t ) + dt q& (t ) + q&&(t ) + O ( dt 3 )
2 (4)
= q (t ) + dt q& (t + dt / 2) + O ( dt 3 )
192 Y. Wang and P. Mora

Hence, the quaternion derivative at mid-step ( q ( t + dt / 2 ) ) is re-

b
quired. Equation (3) indicates that q(t + dt / 2) and ω (t + dt / 2) are
also required, where the former can be easily calculated using
q(t + dt / 2) = q(t ) + q (t ) dt / 2 (5)
b
where q (t ) again is given by Eq. (3), and ω (t ) can be calculated using

ωb (t ) = ωb (t − dt / 2) + I −1 τ b (t )dt / 2 (6)

and ωb (t + dt / 2) can be obtained using

ωb (t + dt / 2) = ωb (t − dt / 2) + I −1 τ b (t )dt (7)
b
In this algorithm, only ω (t − dt / 2) and q(t ) need to be stored,
b
whereas the other quantities, such as q(t + dt / 2) , ω (t ) , q(t ) ,
q (t + dt / 2) etc. are treated as temporary and auxiliary values.
To avoid build-up errors, it is a common practice to renormalize quater-
nion at frequent intervals (usually done at every time-step). The whole al-
gorithm proceeds as follows:
b
Step 1: calculate the force f and torque τ (t ) at time t according to
Eqs. (11), (12), (13), (14), (15), (16), (17) and (18).
b
Step 2: update ω (t ) using the stored ωb (t − dt / 2) according to Eq.
(6).
Step 3: obtain q(t ) using Eq. (3).
b
Step 4: calculate ω (t + dt / 2) using the stored ωb (t − dt / 2) accord-
ing to Eq. (7).
Step 5: compute q(t + dt / 2) using Eq. (5).
Step 6: evaluate q (t + dt / 2) using Eq. (3).
Step 7: calculate q(t + dt ) using Eq. (4).
Step 8: renormalize quaternion q(t + dt ) .
 The ESyS_Particle 193

3 Contact Laws, Particle Interactions and Calculation of

Forces and Torques

When particles come into contact, three kinds of interactions can exist in
the current ESyS_Particle model: bonded, solely normal repulsive and
cohesionless frictional interaction. Bonded interaction permits tensile
forces to be transmitted between particles and can be used to model
behaviors of continuum or intact materials. The breakage of bonds
provides an explicit mechanism for microscopic fracture. While the solely
normal repulsive and cohesionless frictional interaction do not allow
tensile forces to be transmitted between particles, and therefore suitable for
modeling the motions and behaviors of “real discrete” or granular
materials (such as powders, sands and earthquake gouges).

3.1 Bonded Interaction

For bonded interactions, the three important issues that need to be
specified are types of interactions being transmitted between each particle
pair, the algorithm to calculate the interactions between bonded particles
due to the relative motion and the criterion for a bond to break.

3.1.1 The Bonded Model

In the ESyS_Particle, three parameters are used to represent the centre
position of each 3-D particle, and three parameters to represent rotations
about the center, the complete set of interactions requires six kinds of
interactions being transmitted between each particle pair in the 3-D case,
and three in the 2-D case.
Under small deformations, the relationship between interactions and
relative displacements between two bonded particles can be written in the
linear form (Fig. 1)
fr = Kr Δ r ,
f s1 = Ks1 Δ s1 ,
f s2 = Ks 2 Δ s2 ,
τt = Kt Δαt ,
τb1 = Kb1 Δα b1 ,
τb2 = Kb 2 Δα b2 , (8)
194 Y. Wang and P. Mora

where Δ r , Δ s1 ( Δ s2 ) are the relative displacements in normal and

tangent directions. Δαt and Δα b1 ( Δα b2 ) are the relative angular
displacements caused by twisting and bending. fr , f s1 , f s2 , τt , τb1
and τb2 are forces and torques, Kr , Ks1, Ks 2 , K t , Kb1 and Kb 2 are
relevant stiffnesses. If the bond is identical in every direction,
K s = K s1 = K s 2 and Kb = Kb1 = Kb2 .

fr

τt
f s2
f s1

τb1 τb2

Y
X

Fig. 1 Six kinds of interactions between bonded particles. fr is normal force, f s1

and f s2 are shear forces , τt is twisting torque, τb1 and τb2 are bending torque

3.1.2 Calculation of Interactions due to Relative Motion

The calculation of the interactions between bonded particles due to relative
motion requires decomposition of the relative motions. Suppose in the
space-fixed frame, the initial positions of particle 1 and 2 at time t = 0 are
r10 = x10 i + y10 j + z10 k and r20 = x20 i + y20 j + z20 k, the initial
 The ESyS_Particle 195

orientations of particle 1 and 2 are p 0 = 1 + 0i + 0 j + 0k and

0
q = 1 + 0i + 0 j + 0k . At time t, the current positions are
r1 = x1i + y1 j + z1k and r2 = x2 i + y2 j + z2 k, the current orientations
are p = p0 + p1i + p2 j + p3k and q = q0 + q1i + q2 j + q3k . Let
X 1Y1Z1 and X 2Y2 Z 2 represent the body-fixed frame of particle 1 and 2, the
relative translational displacements of particle 1 with respect to X2Y2Z2 is
(Fig. 2)
Δr = rb − r0 = q −1rq − r0 , (9)
where r0 = r10 − r20 is initial relative position of particle 1 with respect to
X2Y2Z2 , and
rb = q −1rq = x12 i + y12 j + z12 k (10)
is the current position of particle 1 with respect to X2Y2Z2, in which the
possible rotation of particle 2 has been taken into account, and
r = r1 − r2.

Z2 G Z 2'
rb
ŝ
G
Δr
X 2' γ
G
r0
Y2
X2
Y2'
Fig. 2 The relative translational displacements of particle 1 relative to particle 2
with respect to X2 Y2 Z2 co-ordinates is Δr . r0 and rb are initial and current
relative position of particle 1 in X2 Y2 Z2 .

In case of Δr << rb and Δr << r0 , we have normal force

fr = Kr ( rb − r0 ) rb rb (11)
196 Y. Wang and P. Mora

Shearing force has two contributors

f s = f st + f sr (12)
t
where f , caused by translational displacement relative to X2Y2Z2 without
s
relative rotation, can be expressed as
f st = Ks rb γ s (13)
r × ( rb × r0 )
where cos γ = r0 ⋅ rb and unit vector s = b .
r0 rb rb × ( rb × r0 )
t
Since f s is exerted on surface, rather than the center of mass, it gener-
ates a torque
1
τ st = rb f st t (14)
2
r0 × rb
where unit vector t = . Note that Eqs. (11), (13) and (14) are evalu-
r0 × rb
ated in the second particle’s body-fixed frame X2Y2Z2.
f sr in Eq. (12), caused by relative rotation of particle 1 with respect to
X2Y2Z2, is given later.
The relative rotation of particle 1 over particle 2, or rotation from
o −1 *
X 2Y2 Z 2 to X 1Y1Z1 , is represented by the quaternion r21 = q p = q p .
Suppose X 2′Y2′Z 2′ is an auxiliary frame, obtained by directly rotating
X 2Y2 Z 2 such that its Z 2′ -axis is pointing to particle 1. It is the X 2′Y2′Z 2′
system in which the relative rotation between two particles should be
evaluated (Fig. 1). Such relative rotation makes X 2′ Y2′Z 2′ rotate to
X 1′Y1′Z1′ (Fig. 3, only the Z1′ -axis is drawn), and can be decided by the
−1 o
quaternion r21 = m r21 m = r0 + r1i + r2 j + r3k (expressed in X 2′ Y2′Z 2′
system), where the quaternion m specifies the rotation
from X 2Y2 Z 2 to X 2′ Y2′Z 2′ , and is given by

2 x122 + y122 + z122 + z12

m0 = ,
2 x122 + y122 + z122
 The ESyS_Particle 197

2 x122 + y122 + z122 − z12 y12

m1 = − ,
2 x122 + y122 + z122 x122 + y122

2 x122 + y122 + z122 − z12 x12

m2 = ,
2 x122 + y122 + z122 x122 + y122
m3 = 0 (15)

By using quaternion algebra, it has been proved that (Wang et al.,

2008a) an arbitrary rotation between two rigid bodies or two coordinate
systems cannot be decomposed into three mutually independent rotations
around three orthogonal axes. However it can be decomposed into two
rotations, one pure axial rotation of angle ψ around its Z 2′ -axis, and one
rotation of Z 2′ -axis over θ on certain plane controlled by another
parameter φ. Figure 3 describes such a decomposition. These two
rotations, corresponding to the relative twisting and bending between two
bodies in our model, are sequence-independent. Such a two-step rotation is
controlled by three independent parameters ψ, θ and φ, which can be
decided as follows,
ψ r0
cos = ,
2 r02 + r32
ψ r3
sin = ,
2 r0 + r32
2

cos θ = r02 − r12 − r22 + r32 ,

r1r3 + r0 r2
cosϕ = ,
( r02 + r32 )( r12 + r22 )
r2 r3 − r0 r1
sin ϕ = . (16)
( r02 + r32 )( r12 + r22 )
198 Y. Wang and P. Mora

ψ
Z 2′
Z1′

O
Y2′

X 2′ ϕ
Z ′′

Fig. 3 An arbitrary rotation between two rigid bodies or two coordinate systems
can be decomposed into two-step rotations, one pure axial rotation of angle ψ
around its Z 2′ -axis, and one rotation of Z 2′ -axis over θ on certain plane controlled
by another parameter φ

After applying this decomposition, the bending, twisting torque and

shear forces exerted on particle 2 are (expressed in X 2′Y2′Z 2′ co-ordinates)
τ b′ = Kbθ ( − sinϕ i + cosϕ j )
τ ′t = K tψ k
rb θ
f s′ = −Ks (cos ϕ i + sinϕ j )
2
2
rb θ
τ ′s = K s ( sinϕ i − cosϕ j ) (17)
4
τ ′s is the torque generated by f s′ . It should be pointed out that Eqs.
(14) and (17) are based on assumption of two equal-sized spheres. In case
of different sized spheres, torques should be changed accordingly. For
example, the torque in Eq. (14) should be replaced by
τ s1t = rb f st tˆ R1 (R1 + R2 ) and τ s2t = rb f st tˆ R2 (R1 + R2 ) , where R1 and
R2 are radii of two particles.
 The ESyS_Particle 199

Torques and forces can easily be transformed back to X2Y2Z2 frame.

τ br = mτ b′ m−1 , τ tr = mτ ′t m−1
f sr = m f s′m−1 , τ sr = m τ ′s m−1 . (18)
The total forces and torques of each particle are expressed as sums of
contributions from all particles bonded to it

f = ∑ AT ( f r + f st + f sr ) (19)
τ = ∑ (τ + τ + τ + τ )
b t
s
r
s
r
b
r
t (20)
where

⎛ q02 + q12 − q22 − q32 2(q1q2 + q0q3 ) 2(q1q3 − q0q2 ) ⎞⎟

⎜
A = ⎜ 2(q1q2 − q0q3 ) 2(q2q3 + q0q1 ) ⎟
2 2 2 2
q0 − q1 + q2 − q3
⎜⎜ 2⎟
q0 − q1 − q2 + q3 ⎟⎠
2 2 2
⎝ 2(q1q3 + q0q2 ) 2(q2q3 − q0q1 )

is the rotation matrix, representing the rotation specified by quaternion

q = q0 + q1i + q2 j + q3k .
Since f ni Eq. (1) is expressed in space-fixed frame, but f r , f st
r
and f s in Eqs. (11), (13) and (18) are evaluated in the body-fixed frame, a
second transfer is required in Eq. 19. Once forces and torques are gained,
Eqs. (1, 2 and 3) can be integrated for each particle.
To test our algorithm described above, we performed simulations using
a simple model which consists of seven bonded cubes. Slow constant ve-
locities or angular velocities are exerted on the first and the last cube to
make the whole chain deform (i.e. stretching, twisting, bending or combi-
nation of deformations). We then compare the relative errors incurred us-
ing the incremental method (which is used by the most DEMs) and using
our algorithm for different time step sizes dt (Wang, 2008b). The relative
error is defined as (kinetic energy + potential energy – external
work)/(external work). We found that the errors in the incremental method
increased much faster with increasing time step than those in our algo-
rithm, indicating that our algorithm is more accurate and stable over a
greater range of time step sizes (Fig. 4).
 200 Y. Wang and P. Mora

0.04
dt = 0.002
dt = 0.0012
dt = 0.0006
0.03 dt = 0.0002
dt = 0.0001
Errors

0.02

0.01

0.00
0 50 100 150 200 250 300
Time steps(X1000)

0.04 dt = 0.002
dt = 0.0012
dt = 0.0006
0.03 dt = 0.0002
dt = 0.0001
Errors

0.02

0.01

0.00
0 50 100 150 200 250 300
Time steps(X1000)

Fig. 4 Evolution of Relative errors with time in the incremental method (above)
and in our algorithm (down) for different integration time steps dt (from 0.0001 to
0.002). The errors increase much faster in the incremental method than that in our
algorithm with increasing integration time step dt

3.1.3 Criterion for Bond Breakage

In the ESyS_Particle model, a bond breaks under either of the following
conditions:
If the pure extensional force exceeds the threshold f r ≥ Fr0 (but it
does not break under pure compression)
If the pure shear force f s ≥ Fs0
If the pure twisting torque τ t ≥ Γ t 0
 The ESyS_Particle 201

If the pure bending torque τ b ≥ Γb0 .

When all the interactions exist at the same time, the following empirical
criterion is used to judge whether or not a bond is going to break
fr | f s | | τt | | τb |
+ + + ≥1 , (21)
Fr0 Fs0 Γ t 0 Γ b0
where f s = f s1 + f s2 , τ b = τ b1 + τ b2 (in 3-D). In the 2-D case the third
term in Eq. (21) drops. We set f r positive under extension and negative
under compression such that it is more difficult for a bond to break under
compression than under extension, therefore the effects of normal force on
breakage of the bond has been taken into account.

3.2 Solely Normal Repulsive Interaction

When two particles contact elastically, only the normal force fr exists,
and Eq. (11) holds only when d < R1 + R2 , here d is the distance between
the two particles, R1 and R2 are radii of two particles.

3.3 Cohesionless Frictional Interaction

In the case of the cohesionless frictional interaction, forces are transmitted
in both normal and tangential directions when d < R1 + R2 . The normal
force f r is dealt exactly the same way it is dealt in case of the solely
normal repulsive interaction. In the current ESyS_Particle model, frictional
forces are updated in an incremental fashion, similar to the most DEMs.
The reason we do not use a Finite Deformation Scheme (such as bonded
interactions) is that particles may be frequently in and out of contact due to
the change of normal forces. It is not convenient to store the positions and
orientations of the particle pairs each time they contact.
Stick-slip style frictional forces are employed in tangential directions.
At time t, if it is in a stick phase ( f s < fr μs , here μs is the static
frictional coefficient), the frictional force at time t+dt can be calculated as
follows
f s (t + dt) = f s (t ) + Δ f s , (22)
where the increment Δ f s = Δ f s1 + Δ f s2 .
202 Y. Wang and P. Mora

The first term, Δ f s1

= Ks Δ us , is the contribution from incremental
shear displacements Δ us = vs dt , where
vs =ω2× rp2 − ω1× rp1 + v2 − v1 −((v2 − v1 ) ⋅ n)n
is the relative transverse velocity due to relative motion at contact point
(Fig. 5), where v1 and v2 are the velocity vectors of the two particles.
The second term has two contributors
Δ f s2 = dt ω s1× f s + dt ω s2 × f s , (23)
where
ω s1 = (r2 − r1 ) × (v2 − v1 ) (r2 − r1 )
2
(24)
and
ω s2 = (((ω 1+ ω 2) ⋅ n) n) 2 (25)
are the angular velocity due to rotation of the line between two center and
the angular velocity due to spin of the line between two centers
respectively. Here r1 and r2 are the position vectors of the two particles,
and ω 1 and ω 2 are the angular velocities.
Since the shear force f s is stored in the global coordinate system, the
rotation of the contact plane caused by movements of two particles as a
rigid body change the direction of the stored shear forces. Therefore the
two terms specified in Eqs. (24) and (25) have to be taken into account in
the 3-D case. It should be pointed out that in the most literatures, the
second term in Eq. (23) is missing (Hart et al., 1988; Monterio-Azevedo
et al., 2005). In the 2-D case that term drops automatically, but in 3-D this
term can not be ignored theoretically.
 The ESyS_Particle 203

ω2
fs
ω1
rp2 n
rp1

r2
Z
r1
Y
X
Fig. 5 Incremental method to update shear forces

At time t, if it is in slip phase, f s = −μd fr vs vs , here μd is the

sliding (or dynamic) frictional coefficient. Equation (22) still holds, but the
increment only includes the contribution from the rotation of the contact
plane caused by movements of two particles as a rigid body, or
Δ f s = Δ f s2 , here Δ f s2 is still calculated using Eqs. (23, 24 and 25).
The transition between slip and stick is captured by the following
criteria:
stick to slip: when f s ≥ μs fr ,
slip to stick: when v s = 0 .
Generally we choose μd < μs such that unstable slip can be modeled. If
μs = μd , it is similar to the Coulomb friction implementation.
204 Y. Wang and P. Mora

4 Parameter Calibration

When starting a simulation, one important thing is to set up the input

parameters. DEM input parameters include the particle mass M, radius R,
rigidity parameters K r , K s , Kb and Kt , fracture parameters Fr0 , Fs0 , Γ t 0
and Γ b0 , artificial damping parameters, integration time step and loading
rate etc. We derive the relationship between the macro-scopic elastic
constants and particle scale stiffness in the case of equal sized particles and
regular lattices, which will provide a theoretical basis for DEMs.

4.1 Elastic Parameters: Spring Stiffness

4.1.1 2-D Triangular Lattice
For the 2-D triangular lattice, Eq. (8) is reduced to f r = Kr Δ r ,
f s = Ks Δ s , and τb = Kb Δα b . The derivation of the spring constant is
based on a comparison between the discrete lattice model and the
continuum model (Bathurst et al., 1988; Griffiths et al., 2001; Ostoja-
Starzewski, 2002). First we let the displacement of every lattice point
equal that of a corresponding point in the continuum model, then we
choose a unit cell which is a periodically repeating part of the network in
both models, lastly we let the energy stored in the unit cell of lattice equal
the strain energy stored in the unit cell of the continuum model. If the
model is arranged in Fig. 6 and the strain is uniform (no rotation
components), we have (Wang et al., 2008c),
3E
Kr = ,
3(1 − ν )
1 − 3ν
Ks = Kr . (26)
1 +ν
The result (Eq. (26)) is invariant with respect to the orientional
angle α (Fig. 6), in other words, the model arranged in Fig. 6 yields
macroscopic isotropic elasticity.
To determine the bending stiffness, we consider the special lattice
orientation α = 0 , in which both the continuum model and hexagonal
arrangement are subject to the stress field: σ 11 = Hy, σ 22 = 0 and
σ 12 = 0, where H is a small constant. This stress field corresponds to
macroscopic bending (in this case, rotation is involved). Similarly, by
 The ESyS_Particle 205

comparing the strain energy density in the continuum model with the
equivalent energy density in discrete grids, we get (Wang et al., 2008c)
3 (1 + ν )(1 − 2ν ) ER 2
Kb = , (27)
36(1 − ν )
where R is the radius of the particles.

y′( sˆ )

B x′( nˆ )
A α

Fig. 6 2D close-packed hexagonal lattice and unit cell (the area enclosed by
dashed lines)

4.1.2 3-D Lattices: HCP and FCC

In the 3-D case, the densest packing of equal-sized spheres are hexagonal
closed packing (HCP) lattice (Fig. 7) and face-centered cubic (FCC) lattice
(Fig. 8). However, both HCP and FCC structures are intrinsically
anisotropic, or direction variant. When particles are arranged in HCP
lattice (Fig. 7), we have

⎛ σ 11 ⎞ ⎛ C11 C12 C13 0 0 0 ⎞ ⎛ ε11 ⎞

⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ σ 22 ⎟ ⎜ C12 C11 C13 0 0 0 ⎟ ⎜ ε 22 ⎟
⎜σ ⎟ ⎜ C C13 C33 0 0 0 ⎟ ⎜ ε 33 ⎟
⎜ 33 ⎟ = ⎜ 13 ⎟⎜ ⎟ and
⎜ σ 23 ⎟ ⎜ 0 0 0 C44 0 0 ⎟ ⎜ 2ε 23 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ σ 13 ⎟ ⎜ 0 0 0 0 C44 0 ⎟ ⎜ 2ε13 ⎟
⎜σ ⎟ ⎜ 0 0 0 0 0 C66 ⎟⎠ ⎜⎝ 2ε12 ⎟⎠
⎝ 12 ⎠ ⎝
206 Y. Wang and P. Mora

⎛ ε11 ⎞ ⎛ 1 E x −ν yx E y −ν zx E z 0 0 0 ⎞ ⎛ σ 11 ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ ε 22 ⎟ ⎜ −ν xy E x 1 Ey −ν zy E z 0 0 0 ⎟ ⎜ σ 22 ⎟
⎜ ε ⎟ ⎜ −ν E −ν yz E y 1 Ez 0 0 0 ⎟ ⎜ σ 33 ⎟
⎜ 33 ⎟ = ⎜ xz x ⎟⎜ ⎟
⎜ 2ε 23 ⎟ ⎜ 0 0 0 1 μ yz 0 0 ⎟ ⎜ σ 23 ⎟
⎜ ⎟ ⎜ 1 μ xz
⎟
0 ⎟ ⎜ σ 13 ⎟
⎜ 2ε13 ⎟ ⎜
0 0 0 0
⎜ ⎟
⎜ 2ε ⎟ ⎜ 0 0 0 0 0 1 μ xy ⎟⎠ ⎜⎝ σ 12 ⎟⎠
⎝ 12 ⎠ ⎝

(28)

where

C11 = 2 ( 29 K r2 + 89 K s2 + 170 K r K s ) 48R( K r + 5K s ) ,

C12 = 2 ( K r − K s )(11K r + 49 K s ) 48 R( K r + 5K s ) ,

C13 = 2 ( K r − K s ) 6 R ,

C33 = 2 ( 2 K r + K s ) 3R ,

C44 = 2 ( K r + 2 K s ) 6 R ,

C66 = 2 (3K r2 + 23K s2 + 22 K r K s ) 16 R( K r + 5 K s ) = (C11 − C12 ) 2 .

This is transversely isotropic, which is a special case of an orthotropic

solid, with x1 x2 plane of isotropy and generally 5 independent constants
(Daniel et al., 1994). Here the Young’s modulus, Poisson’s ratio and shear
modulus are respectively

3 2 K r ( K r + K s )(3K r2 + 23K s2 + 22 K r K s )
Ex = E y = ,
R (18K r3 + 119 K r2 K s + 128 K r K s2 + 23K s3 )

3 2K r ( K r + K s ) ,
Ez =
R (5 K r + K s )
 The ESyS_Particle 207

( K r − K s )(6 K r2 + 23K s2 + 31K r K s ) ,

ν xy = ν yx =
18K r3 + 119 K r2 K s + 128K r K s2 + 23K s3

Kr − Ks ,
ν zx = ν zy =
5K r + K s

( K r − K s )(3K r2 + 23K s2 + 22 K r K s )
ν xz = ν yz = ,
18 K r3 + 119 K r2 K s + 128K r K s2 + 23K s3

2 (3K r2 + 23K s2 + 22 K r K s )
μ xy = ,
16 R( K r + 5 K s )

2 ( K r + 2K s )
μ xz = μ yz = . (29)
6R

In the case of FCC (see Fig. 8), we have

⎛ σ 11 ⎞ ⎛ C11 C12 C12 0 0 0 ⎞ ⎛ e11 ⎞

⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ σ 22 ⎟ ⎜ C12 C11 C12 0 0 0 ⎟ ⎜ e22 ⎟
⎜σ ⎟ ⎜ C C12 C11 0 0 0 ⎟ ⎜ e33 ⎟
⎜ 33 ⎟ = ⎜ 12 ⎟⎜ ⎟ and
⎜ σ 23 ⎟ ⎜ 0 0 0 C44 0 0 ⎟ ⎜ 2e23 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ σ 13 ⎟ ⎜ 0 0 0 0 C44 0 ⎟ ⎜ 2e13 ⎟
⎜σ ⎟ ⎜ 0 0 0 0 0 C44 ⎟⎠ ⎜⎝ 2e12 ⎟⎠
⎝ 12 ⎠ ⎝

⎛ ε11 ⎞ ⎛ 1 E −ν E −ν E 0 ⎞ ⎛ σ 11 ⎞
0 0
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ ε 22 ⎟ ⎜ − ν E 1 E −ν E 0 0 0 ⎟ ⎜ σ 22 ⎟
⎜ ε ⎟ ⎜ −ν E −ν E 1 E 0 0 0 ⎟ ⎜ σ 33 ⎟
⎜ 33 ⎟ = ⎜ ⎟⎜ ⎟ (30)
⎜ 2ε 23 ⎟ ⎜ 0 0 0 1μ 0 0 ⎟ ⎜ σ 23 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ 2ε13 ⎟ ⎜ 0 0 0 0 1 μ 0 ⎟ ⎜ σ 13 ⎟
⎜ 2ε ⎟ ⎜ 0 0 0 0 0 1 μ ⎟⎠ ⎜⎝ σ 12 ⎟⎠
⎝ 12 ⎠ ⎝

where
C11 = 2 ( K r + K s ) 2 R ,
208 Y. Wang and P. Mora

C12 = 2 ( K r − K s ) 4 R ,
C44 = 2 ( K r + K s ) 4 R = C11 2 .
From Eq. (30), we know that this is cubic material, the simplest case for
an orthotropic solid. The 3 independent macroscopic elastic constants are
2 ( K r + 3K s ) K r ,
E=
R (3 K r + K s )
K − Ks
ν= r ,
3K r + K s
2(Kr + Ks )
μ= . (31)
4R
K r and K s can be expressed in terms of E and ν as follows:
2 ER
Kr = ,
2(1 − 2ν )
1 − 3ν
Ks = Kr . (32)
1 +ν

z
y
x

Fig. 7 3D HCP packing, the central particle is placed on the origin of coordinate
 The ESyS_Particle 209

13
10 12
11

2 3
5 z
1 4
y
x
9
6 8
7

Fig. 8 FCC lattice and coordinate system. The center of the FCC structure is
placed on the origin of the coordinates

To have an estimation of K b and K t , we consider an infinite FCC

continuum subjected to the pure bending stress: σ 11 = Hz , σ 22 = σ 33 = 0 ,
σ 12 = σ 23 = σ 13 = 0 . Similarly, by comparing the strain energy
distribution and the distribution of moments with respect to the y axis
between the continuum and discrete lattice, we have (Wang et al., 2008c),
2 ER3 (1 − 2ν ) R 2
Kb = = Kr ,
48(1 − ν ) 24(1 − ν )
1 − 3ν (1 − 2ν ) R 2
Kt = Kb = Ks . (33)
1 +ν 24(1 − ν )

4.2 Fracture Parameters

It is very difficult to have a theoretical analysis on how to choose fracture
parameters Fr0 , Fs0 , Γ t 0 and Γ b 0 . However this problem can be
investigated numerically. Some numerical tests have been done to
reproduce the fracture of brittle rocks under uni-axial compression (Wang
et al., 2006; 2008a). We find that k = Fs 0 Fr 0 is an important parameter
to control the fracture behaviors and macro-scopic strength, and k = 1 to 4
gives reasonable brittle behaviors. Our preliminary studies also suggest
that Γ b 0 = Fr 0 R 3 in 2-D, or Γ t 0 = Fs 0 R 2 , Γ b 0 = Fr 0 R 4 in 3-D will
210 Y. Wang and P. Mora

generate satisfactory results. Further comparisons between the different

criteria and detailed investigations about how these parameters affect the
macroscopic strength and fracture patterns are required.

4.3 Other Parameters

4.3.1 Time Step

Generally the time step can be decided according to dt = α M min K max ,

here α ~ 0.1 gives a satisfactory and safe result. In the 2-D case, the
mass M ∝ R 2, stiffness K r and K s do not change with R (Eq. (26)),
therefore dt ∝ R . If the particle rotation is involved, similarly we have
dt = α I min K max
r
. Since the moment of inertia I ∝ MR 2 ∝ R 4 ,
K r ( K b and K t ) ∝ R 2 (Eq. (27)), therefore dt ∝ R. In the 3-D case,
M ∝ R 3 , K r and K s ∝ R (Eq. (32)), we have dt ∝ R in the translational
case. The rotational parameters I ∝ MR 2 ∝ R 5 , K b and K t ∝ R 3 (Eq.
(33)), we still have dt ∝ R in the rotational case. This means that if we
keep the same macroscopic Young’s modulus E, it means that a smaller
time step is required for smaller particle sizes.

4.3.2 Artificial Damping

In the ESyS_Particle model, two types of damping are employed. The first
one is global damping, f1ν = − ν 1 Mv , here v is the absolute velocity of
the particle, and ν 1 is the damping coefficient. ν 1 can not be significantly
large such that in one time step a particle changes its direction of motion
purely by the damping force, that is, f1ν dt < Mv , then we have ν 1 < 1 dt
as the upper limit.
ν
The second kind of the damping force is f 2 = − ν 2 K vij , here vij ,
K and ν 2 are the relative velocity between two particles, stiffness at the
contact and damping coefficient. This kind of damping encourages the
 The ESyS_Particle 211

particles to move as rigid bodies, but discourages relative motion between

the particles (such as wave propagation). Similarly we have f sν dt < Mvij ,
or ν 2 < dt α , here α ~ 0.1 .

4.3.3 Loading Rate

Within one time step, particles should not be moved a distance larger than
its size, or Vload dt << R . This principle should also hold true for highly
dynamic processes such as the impact or blast simulations, in which case
time steps are required be very small due to the large velocities.

5 Some Recent Simulation Results

Here we give some results from recent simulations using the

ESyS_Particle model.

5.1 2-D Tests

5.1.1 Uni-Axial Tests
Figure 9a, 9b and 9c show simulations of fractures of brittle rock-like
material under uni-axial compression. Random sized particles are arranged
into rectangles of different aspect ratios (from 1 to 3). In the beginning, the
neighbouring particles are bonded together to model the intact material.
Two loading walls at the top and bottom of the model move slowly to
compress the sample. On the right of each figure are the laboratory tests
(Andreev, 1995). The figures suggest that the simulations are quite similar
to the laboratory tests.
212 Y. Wang and P. Mora

b
 The ESyS_Particle 213

c
Fig. 9 2D simulation of brittle fracture under uni-axial compression for different
aspect rations (3 in Fig.9a, 2 in Fig. 9b and 1 in Fig. 9c). The photos from the right
of each picture are laboratory tests (Andreev, 1995)

5.1.2 Wing Crack Extension

Figure 10 and Movie 1 show the 2-D wing crack extensions. In this figure
there is a pre-existing oblique crack (the bonds are removed) in the
beginning. Both regular and random particle arrangements are used. Stable
tensile cracks are reproduced with an increase of loading. The left of Fig.
10 is a sketch of the laboratory test (Brace, 1960).
It is found that when only normal stiffness exists between two bonded
particles, or when the single particle rotation is prohibited, or when single
particle rotation exists but rolling stiffness is not permitted between two
particles, the realistic pattern of crack propagation observed in laboratories
can not be reproduced. Only when normal, shear and rolling stiffness exist
and particle rotation is permitted, is it possible to reproduce laboratory
tests (Wang et al., 2008a). We conclude that particle rotation and rolling
resistance play a significant role and can not be neglected while modeling
such phenomenon.
214 Y. Wang and P. Mora

Fig. 10 2D wing crack extensions, left: a sketch of the laboratory test (Brace,
1960); middle: same sized particles; right: random sized particles

5.1.3 Shearing and Crushing of Aggregates

Figure 11 and Movie 2 show the crushing of soil grains. The simulation
models circular aggregates consisting of small particles bonded together,
sheared by two elastic layers of bonded particles. The constant pressure is
applied while shearing is modeled by moving the two elastic blocks. A
periodical boundary is used in the horizontal direction. Rotations of the
aggregates as rigid bodies can be observed, suggesting that frictional
forces between the particles from the different aggregates take into effect.
The fracture of some aggregates can be seen clearly.
 The ESyS_Particle 215

Fig. 11 Simulation of crushing of grains under shearing and the constant pressure

5.1.4 Simulation of Brittle Fracture by Dynamic Impact

Figure 12 and Movie 3 show the simulation of cracks generated in a brittle
rock after an impact. The brittle rock is modeled by bonded particles of
different sizes. The large ball falls under gravity and hits the rock and
cracks are formed. The propagation of brittle cracks and elastic waves are
reproduced.
216 Y. Wang and P. Mora

Fig. 12 Simulation of brittle crack extension under impact

5.2 3-D Tests

5.2.1 Uniaxial Test
Figure 13 and Movie 4 show the progressive fractures in a 3-D brittle rock.
In this example, equal sized particles are arranged into FCC lattice which
is subjected to uni-axial compression. The colors represent vertical
displacement. Discontinuities in colors mean the formation of fractures
which is difficult to be captured in laboratory tests since this process
always occurs very fast. After the peak stress, the main faults are formed
and two intact cores can be clearly observed with more fragile parts
shattering away from four sides, which is always observed in rock fracture
tests.
 The ESyS_Particle 217

Fig. 13 Progressive fractures in 3D brittle rock under uni-axial compression

5.2.2 Wing Crack

Figure 14 Movie 5 present 3-D simulations of the extension of wing cracks
in a brittle rock under slow uni-axial compression. The sample consists of
equal-sized particles arranged into FCC cubic. Similarly, at the beginning
of the simulation an oblique pre-existing circular crack is generated by
removing the bonds in the cracked regions (small particles representing the
centers of the removed bonds). The bigger particles in the plot are the
218 Y. Wang and P. Mora

centers of the fractured bonds, representing the modeled events. The

laboratory tests are also given (Dyskin et al., 2003) in Fig. 15. Not only the
basic shapes in laboratory tests, the curved wings (wrapping), are well
reproduced, but also another interesting fact is also observed in our
simulation: 3-D wing crack growth under uni-axial compression is limited,
opposite to the 2-D case (Dyskin et al., 2003).
As far as we know, this is the first effort to model wing cracks using the
Discrete Element Model.

Fig. 14 Simulated 3D wing crack extension. Small particles are the centers of the
removed bonds (pre-existing fault). The bigger particles are the centers of the
fractured bonds, representing the modeled events.
 The ESyS_Particle 219

Fig. 15 Laboratory tests of 3D wing crack extension under uni-axial compression

(Dyskin et al., 2003).

6 Discussion: Major Differences of the ESyS_Particle

Compared with the Other Existing DEMs

Compared with the other DEMs, the ESyS_Particle is different and

advantageous in the following aspects:
Firstly we make use of unit quaternion to explicitly represent the
orientation of a particle. In the most existing DEMs, particle orientations
are implicitly represented using three angular velocities around three
orthogonal axes. In 2-D cases this does not present a problem, since an
angle can be easily integrated from angular velocities for each rigid body
at each time step, thereby defining the orientation. However, in 3-D, this is
not the case. Strictly speaking, one can not extract the exact orientation of
a particle by simply integrating three angular velocities. The physical
principle which limits this is: finite rotations in 3-D are order
dependent. Consequently, for the same three finite angles integrated from
angular velocities, different orders of rotations result in different final
orientations (there are 12 possibilities). This salient feature of finite
rotations in 3-D is one which is ignored by most modelers, and in this
paper we attempt to address and highlight this discrepancy.
220 Y. Wang and P. Mora

Secondly, for bonded contacts, our model permits six kinds of

independent relative motions between two particles and therefore six kinds
of interactions being transmitted (three in 2-D). Our algorithm to calculate
forces and torques is based on the Finite Deformation Method in which
total displacements are calculated instead of incremental displacements
(Eq. (8)). At each time step t, we only need the initial, current position and
orientation (i.e. at t = 0 and t). In nearly all other algorithms, shear forces
and torques are computed in an incremental fashion. This means that the
three incremental angles from time t-dt to t are computed using three
angular velocities, then three incremental torques are calculated and added
to the torques at time t-dt to get the final torques at time t. It should be
noted that when twisting and bending co-exist, bending changes the axis of
twisting, and thus, the incremental method fails to decouple the twisting
and bending. In our model, a new technique is used to decompose the
relative rotation between two particles such that twisting and bending are
completely decoupled, strictly distinguished and sequence-independent.
Therefore our algorithm is physically more reliable. It is extremely useful
to apply our method when twisting and bending stiffnesses are different.
Thirdly, from a theoretical standpoint, only infinitesimal rotations in 3-
D are order independent. Hence, in order to gain accurate results, the
incremental method requires very small time steps. This computational
inefficiency is highlighted by Wang (Wang, 2008b), where the errors in
the incremental method are shown to increase much faster than those in
our algorithm with increasing time step (Fig. 4).
Lastly, other differences include criterion to judge breakage of a bond,
stick-slip friction and theoretical studies on how to choose model
parameters.

7 Conclusions

What has been presented in this paper is a detailed description of the

ESyS_Particle model. Particular focus is on the basic physical ingredients
or mechanisms which a DEM should have.
We highlight the major differences between our model and most
existing DEMs, which include the representation of particle rotations, fully
decoupled and order-independent twisting and bending, a state-of-the-art
algorithm to calculate forces and torques between bonded particles, unique
criterion to judge breakage of a bond, stick-slip friction and detailed
theoretical studies on how to choose model parameters. Specifically, single
 The ESyS_Particle 221

particle rotation and full interactions between particles make our model
geometrically and theoretically complete, and studies suggest that the
algorithm to decouple twisting and bending is physically reliable and
numerically accurate because the basic physical principle, finite rotations
in 3-D are order dependent, which is ignored by most modelers, is
respected and obeyed in our model.
The problem of how to choose model parameters is discussed. These
parameters include elastic stiffnesses, fracture parameters, integration time
step, artificial damping parameters and loading rate.
The applicability of the ESyS_Particle is illustrated through several
numerical simulations. Most qualitative features of rock fracture observed
in laboratory tests are well reproduced, these include, fracture of brittle
materials under compression, wing crack extension both in 2-D and 3-D,
crushing of aggregates and fracture caused by dynamic impact. Although
these examples have mostly focused on the simulations of fractures of
intact materials, the ESyS_Particle is also very suitable to model “discrete”
material (Latham et al., 2006).
It is found that single particle rotation and rotational stiffnesses play
very important roles in reproducing realistic laboratory tests. The reason is
that particle rotation effects local elasticity, especially stress fields near the
crack tips, which is an important factor for crack extension.
The future direction of the ESyS_Particle will include the coupling with
other physical processes. Currently, the model has been extended to in-
clude heat transfer, thermal expansion, friction generated heat, pore flow
(or Darcy flow) and full coupling between mechanical and pore effects. In
future Smooth Particle Hydrodynamics (SPH) will be implemented to
model the interaction of liquids and solids. However, if multi-physics are
involved, dimensionless analysis and scaling laws are required to deter-
mine the different physical parameters. More applications, either from en-
gineering or scientific fields, are required to further validate and improve
the model.
Acknowledgements Funding support is gratefully acknowledged by the Australian
Computational Earth Systems Simulator Major National Research Facility, The
University of Queensland and SGI. The ACcESS MNRF is funded by the
Australian Commonwealth Government and participating institutions (Univ. of
Queensland, Monash U, Melbourne U., VPAC, RMIT) and the Victorian State
Government.
The first author would like to thank Mr. William Hancock for checking the
manuscript.
222 Y. Wang and P. Mora

References

Abe S, Mora P, Place D (2000) Extension of the Lattice Solid Model to incorpo-
rate temperature related effects. Pure Appl Geophys 157:1867–1887
Abe S, Dieterich J, Mora P, Place D (2002) Simulation of the influence of rate-
and state- dependent friction on the macroscopic behavior of complex fault
zones with the lattice solid model. Pure Appl Geophys 159:1967–1983
Allen MP, Tildesley DJ (1987) Computer simulation of liquids. Oxford Science
Press, Oxford
Anandarajah A (1994) Discrete-Element Method for simulating behavior of
cohensive soil. J Geotech Engrg 120:1593–1613
Andreev GE (1995) Brittle failure of rock materials: test result and constitutive
models. Rotterdam AA Balkema
Bardet JP, Proubet J (1991) A numerical investigation of the structure of persistent
shear bands in granular media. Geotechnique 41:599–613
Bathurst RJ, Rothenburg L (1988) Micromechanical aspects of isotropic granular
assemblies with linear contact interactions. J Appl Mech 55:17–23
Boutt DF, Mcpherson BJOL (2002) Simulation of sedimentary rock deformation:
lab-scale model calibration and parameterization. GRL 29:10.1029/
2001GL013987
Brace WF (1960) An extension of the Griffith theory of fracture to rocks. J G R
65:3477–3480
Buss SR (2000) Accurate and efficient simulation of rigid-body rotations. J Comp
Phys 164:377–406
Campbell CS, Cleary PW Hopkins MA (1995) Large-scale landslide simulations:
global deformation, velocities, and basal friction. J G R 100:8267–83
Chang CS (1992) Discrete Element Method for slope stability analysis. J Geotech
Eng 118:189–1905
Chang SH, Yun KJ, Lee CI (2002) Modeling of fracture and damage in rock by
the bonded-particle model. Geosys Eng 5:113–120
Cheng YP, Nakata Y, Bolton MD (2003) Discrete element simulation of crushable
soil. Geotechnique 53:633–641
Cleary PW, Campbell CS (1993) Self-lubrication for long run-out landslides: ex-
amination by computer simulation. J G R 98:21911–21924
Cleary PW (2000) DEM simulation of industrial particle flows: case studies of
dragline excavators, mixing in tumblers and centrifugal mills. Powder Tech-
nol 209:83–104
Cleary PW, Sawley ML (2002) DEM modeling of industrial granular flows: 3D
case studies and the effect of particle shape on hopper discharge. Appl Math
Model J 26:89–111
Cundall PA, Strack O (1979) A discrete element model for granular assemblies.
Geotechnique 29:47–65
Cundall PA (1988) Formulation of a three-dimensional distinct element model –
Part I, A scheme to detect and represent contacts in a system composed of
many polyhedral blocks. Int J Rock Mech 25:107–116
 The ESyS_Particle 223

D’Addetta GA, Kun F, Ramm E (2002) On the application of a discrete model to

the fracture process of cohesive granular materials. Granul Matter 4:77–90
Daniel IM, Ishai O (1994) Engineering mechanics of composite material. Oxford
University Press
Davie CT, Bicanic N (2003) Failure criteria for qusi-brittle materials in lattice-
type models. Commun Numer Meth Eng 19:703–713
Delenne JY, Youssoufi MSE, Cherblanc F, Benet JC (2004) Mechanical behavior
and failure of cohesive granular materials. Int J Numer Anal Mech Geomech
28:1577–1594
Donze FV, Bouchez J, Magnier SA (1997) Modeling fractures in rock blasting. Int
J Rock Mech Min Sci 34:1153–1163
Dorby R, Ng T-T (1992) Discrete modeling of stress-strain behaviour of granular
media at small and large strains. Eng Computation 9:129–143
Dullweber A, Leimkuhler B, McLachlan R (1997) Symplectic splitting methods
for rigid-body molecular dynamic. J Chem Phys 107:5840–5851
Dyskin AV, Sahouryeh E, Jewell RJ, Joer H, Ustinov KB (2003) Influence of
shape and locations of 3-D cracks on their growth in uniaxial compression.
Engrg Fract Mech 70:2115–2136
Eberhardt E, Thuro K, Luginbuehl M (2005) Slope instability mechanisms in dip-
ping interbedded conglomerates and weathered-the 1999 Rufi landlide, Swit-
zerland. Eng Geol 77:35–56
Evans DJ (1977) On the representation of orientation space. Mol Phys 34:317–325
Evans DJ, Murad S (1977) Singularity free algorithm for molecular dynamic
simu1lation of rigid polyatomice. Mol Phys 34:327–331
Feng YT, Han K, Owen DRJ (2004a) Discrete element simulation of the dynamics
of high energy planetary ball milling processes. Mater Sci Engrg A
375–377:815–819
Feng YT, Owen DRJ (2004b) A 2D polygon/polygon contact model: algorithmic
aspects. Engrg Comput 21:265–277
Fincham D (1992) Leapfrog rotational algorithm. Molec Simul 8:1165
Fleissner F, Gaugele T, Eberhand P (2007) Applications of the discrete element
method in mechanical engineering. Multibody Syst Dyn 18:81–94
Ghaboussi J, Barbosa R (1990) Three-dimensional discrete element method for
granular materials. Int J Numer Anal Meth Geomech 14:451–472
Goldstein H (1980) Classical mechanics. 2nd edn., Addison-Wesley
Griffiths DV, Mustoe GGW (2001) Modeling of elastic continuum using a grillage
of structural elements based on discrete element concepts. Int J Numer Meth
Eng 50:1795–1775
Guo YG, Morgan JK (2004) Influence of normal stress and grain shape on granu-
lar friction: results of discrete element simulations. J G R 109:B12305
Hart R, Cundall PA, Lemos J (1988) Formulation of a three-dimensional distinct
element model – Part II. Mechanical calculations for motion and interaction of
a system composed of many polyhedral blocks J Rock Mech Min Sci Ge-
omech Abstr 25:117–125
224 Y. Wang and P. Mora

Hazzard JF, Collins DS, Pettitt WS, Young RP (2000a) Simulation of unstable
fault slip in granite using a bonded-particle model. Pure Appl Geophys
159:221–245
Hazzard JF, Young RP (2000b) Simulation acoustic emissions in bonded-particle
models of rock. Int J Rock Mech Min Sci 37:867–872
Hazzard JF, Young RP, Maxwell SC (2000c) Micromechanical modelling of
cracking and failure in brittle rock. J G R 105:16683–16697
Hazzard JF, Young RP (2002) Moment tensors and micromechanical models. Tec-
tonophysics 356:181–197
Hazzard JF, Mair K (2003) The importance of the third dimension in granular
shear. G R L 30:doi:10.1029/2003GL017534
Hazzard JF, Young RP (2004) Dynamic modeling of induced seismicity. Int J
Rock Mech 41:1365–1376
Hentz S, Daudeville L, Donze FV (2004a) Identification and validation of a dis-
crete element model for concrete. J Eng Mech 130:709–719
Hentz S, Donze FV, Daudeville L (2004b) Discrete Element modeling of concrete
submitted to dynamics loading at high strain rates. Comput Struct 82:
2509–2524
Hogue C (1998) Shape representation and contact detection for discrete element
simulations of arbitrary geometries. Eng. Comput 15:374–390
Holst JMFG, Rotter JM, Ooi JM, Rong GH (1999) Numerical modeling of silo
filling. II: Discrete element analysis. J Engrg Mech 125:104–110
Hu JC, Angelier J (2004) Stress permutations: three-dimensional distinct element
analysis accounts for a common phenomenon in brittle tectonics J G R
109:doi:10.1029/2003JB002616
Huang HY, Detournay E, Bellier B (1999) Discrete element modeling of rock cut-
ting. Rock mechanics for industry. Amadei, Kranz, Scott and Smeallie (eds)
Balkema Rotterdam
Hunt SP, Meyers AG, Louchnikov V (2003) Modelling the Kaiser effect and de-
formation rate analysis in sandstone using the discrete element method. Com-
put Geotech 30:611–621
Issa JA, Nelson RB (1992) Numerical analysis of micromechanical behaviour of
granular materials. Engrg Comput 9:211–223
Iwashita K, Oda M (1998) Rolling resistance at contacts in simulation of shear
band development by DEM. J Eng Mech 124:285–292
Iwashita K, Oda M (2000) Micro-deformation mechanism of shear banding proc-
ess based on modified distinct element method. Powder Technol 109:192–205
Jensen RP, Bosscher PJ, Plesha ME, Edil TB (1999) DEM simulation of granular
media – structure interface: effects of surface roughness and particle shape.
Int J Numer Anal Meth Geomech 23:531–547
Jiang MJ; Yu HS, Harris D (2006) Discrete element modelling of deep penetration
in granular soils. Int J Numer Anal Meth Geomech 30:335–361
Johnson KL (1987) Contact mechanics. Cambridge University Press, Cambridge
Johnson SM, Williams JR, Cook BK (2007) Quaternion-based rigid body rotation
integration algorithms for use in particle methods. Int J Numer Meth Engrg
74:1303–1313
 The ESyS_Particle 225

Kol A, Laird BB, Leimkuhler BJ (1997) A symplectic method for rigid-body mo-
lecular simulation. J Chem Phys 107:2580–2588
Krysl P, Endres L (2005) Explicit Newmark/Verlet algorithm for time integration
of the rotational dynamics of rigid bodies. Int J Numer Meth Engrg 62:
2154–2177
Kuhn MR, Bagi K (2004) Contact rolling and deformation in granular media. Int J
Solid Stru 41:5793–5820
Kuipers JB (1998) Quaternion and rotation sequences. Princeton University Press,
Princeton, New Jersey
Langston PA, Al-Awamleh MA, Fraige FY, Asmar BN (2004) Distinct element
modeling of non-spherical frictionless particle flow. Chem Eng Sci 59:
425–435
Latham S, Abe S, Mora P (2006) Parallel 3D simulation of a fault gouge using the
lattice solid model. Pure Appl Geophys 163:1949–1964
Li Y, Xu Y, Thornton C (2005) A comparison of discrete element simulations and
experiments for “sandpile” composed of spherical particles. Powder Tech
160:219–228
Lin X, Ng TT (1997) A three-dimensional discrete element model using arrays of
ellipsoids. Geotechnique 47:319–329
Lu M, McDowell GR (2007) The importance of modelling ballast particle shape
in the discrete element method. Granul Matter 9:69–80
Magnier SA, Donze FV (1998) Numerical simulations of impacts using a discrete
element method. Mech Cohes-Frict Mater 3:257–276
Matsuda Y, Iwase Y (2002) Numerical simulation of rock fracture using three-
dimensional extended discrete element method. Earth Planets Space 54:
367–378
Matuttis HG, Luding S, Herrmann H (2000) Discrete element simulations of dense
packings and heaps made of spherical and non-spherical particles. Powder
Tech 109:278–293
McDowell GR, Harireche O (2002) Discrete element modelling of soil particle
fracture. Geotechnique 52:131–135
Miller III TF, Eleftheriou M, Pattnaik P, Vdirango, Newns AD (2002) Symplectic
quaternion scheme for biophysical molecular dynamic. J Chem Phys
116:8649–8659
Mindlin RD, Deresiewicz H (1953) Elastic spheres in contact under varying
oblique forces. J Appl Mech 20:327–344
Monterio-Azevedo N, Lemos JV (2005) A generalized rigid particle contact
model for fracture analysis. Int J Numer Anal Meth Geomech 29:269–285
Mora P, Place D (1993) A lattice solid model for the nonlinear dynamics of earth-
quakes. Int J Mod Phys C4:1059–1074
Mora P, Place D (1994) Simulation of the frictional stick-slip instability. Pure
Appl Geophys 143:61–87
Mora P, Place D (1998) Numerical simulation of earthquake faults with gouge:
towards a comprehensive explanation for the heat flow paradox. J G R
103:21067–21089
Mora P, Place D (1999) The weakness of earthquake faults. G R L 26:123–126
226 Y. Wang and P. Mora

Mora P, Place D, Abe S, Jaume S (2000) Lattice solid simulation of the physics of
earthquakes: the model, results and directions, in GeoComplexity and the
Physics of Earthquakes (Geophysical Monograph series; no. 120),Rundle JB,
Turcotte DL and Klein W (eds) American Geophys Union, Washington DC,
pp 105–125
Mora P, Place D (2002a) Stress correlation function evolution in lattice solid
elasto-dynamic model of shear and fracture zones and earthquake prediction.
Pure Appl Geophys 159:2413–2427
Mora P, Wang YC, Yin C, Place D (2002b) Simulation of the Load-Unload Re-
sponse Ratio and critical sensitivity in the Lattice Solid Model. Pure Appl
Geophys 159:2525–2536
Morgan JK (1999) Numerical simulations of granular shear zones using the dis-
tinct element method: II. The effect of particle size distribution and interparti-
cle friction on mechanical behavior. J G R B 104:2721–2732
Morgan JK, Boettcher MS (1999) Numerical simulations of granular shear zones
using the distinct element method: I. Shear zone kinematics and microme-
chanics of localization. J G R B 104:2703–2719
Morgan JK (2004) Particle dynamics simulations of rate and state dependent fric-
tional sliding of granular fault gouge. Pure Appl Geophys 161:1877–1891
Morrison RD, Cleary PW (2004) Using DEM to model ore breakage within a pilot
scale SAG mill. Min Engrg 17:1117–1124
Morrison RD, Shi F, Whyte R (2007) Modelling of incremental rock breakage by
impact-for use in DEM models. Min Engrg 20:303–309
Munjiza A, Owen DRJ, Bicanic N (1995) A combined finite/discrete element
method in transient dynamics of fracturing solids. Engrg Comput 12:145–174
Munjiza A, Latham JP, John NWM (2003) 3D dynamics of discrete element sys-
tems comprising irregular discrete element-integration solution for finite rota-
tions in 3D. Int J Numer Meth Eng 56:35–55
Mustoe G (1992) A generalized formation of the discrete element method. Engrg
Comput 9:181–190
Ng TT, Dobry R (1994) A nonlinear numerical model for soil mechanics. J Geo-
tech Engrg 120:388–403
Ning Z, Boerefijn R, Ghadiri M, Thornton C (1997) Distinct element simulation
of impact breakage of lactose agglomerates. Adv Powder Tech 8:15–37
Oda M, Konishi J, Nemat-Nasser S (1982) Experimental micromechanical evalua-
tion of the strength of granular materials: effects of particle rolling. Mech Ma-
ter 1:269–283
Oda M, Kazama H (1998) Microstructure of shear bands and its relation to the
mechanisms of dilatancy and failure of dense granular soils. Geotechnique
48:465–481
Oda M, Iwashita K (2000) Study on couple stress and shear band development in
granular media based on numerical simulation analyses. Int J Eng Sci
38:1713–1740
Omelyan IP (1998a) Algorithm for numerical integration of the rigid-body equa-
tions of motion. Phys Rev E 58:1169–1172
 The ESyS_Particle 227

Omelyan IP (1998b) On the numerical integration of motion for rigid polyatomics:

the modified quaternion approach. Comp Phys 12:97–103
Ostoja-Starzewski M (2002) Lattice models in micromechanics. Appl Mech Rev
55:35–60
Owen DRJ, Feng YT (2001) Parallelised finite/discrete element simulation of
multi-fracturing solids and discrete systems. Engrg Comput 18:557–576
Place D, Mora P (1999) A lattice solid model to simulate the physics of rocks and
earthquakes: incorporation of friction. J Comp Phys 150:332–372
Place D, Mora P (2000) Numerical simulation of localization in a fault zone. Pure
Appl Geophys 157:1821–1845
Place D, Mora P (2001) A random lattice solid model for simulation of fault zone
dynamics and fracture process, in Bifurcation and Localization Theory for
Soil and Rock’99, Muhlhaus HB, Dyskin AV, Pasternak E, and Balkema AA
(eds) Rotterdam/Brookfield
Place D, Lombard F, Mora P, Abe S (2002) Simulation of the micro-physics of
rocks using LSM earth. Pure Appl Geophys 159:1933–1950
Potyondy D, Cundall P, Lee CA (1996) Modelling rock using bonded assemblies
of circular particles. Rock Mechanics, Aubertin M, Hassani F, Mitri H (eds)
Balkema, Rotterdam
Potyondy D, Cundall P (2004) A bonded-particle model for rock. Int J Rock Mech
Min Sci 41:1329–1364
Prochazka PP (2004) Application of discrete element methods to fracture mechan-
ics of rock bursts. Eng Fract Mech 71:601–618
Rapaport DC (1995) The art of molecular dynamic simulation. Cambridge Uni-
versity Press
Sakaguchi H, Muhlhaus H (2000) Hybrid modeling of coupled pore fluid-solid de-
formation problems. Pure Appl Geophys 157:1889–1904
Schlangen E, Garboczi EJ (1997) Fracture simulations of concrete using lattice
models: computational aspects. Engrg Fract Mech 57:319–332
Schwarz OJ, Horie Y, Shearer M (1998) Discrete element investigation of stress
fluctuation in granular flow at high strain rates. Phys Rev E 57:2053–2061
Scott D (1996) Seismicity and stress rotation in a granular model of the brittle
crust. Nature 381:592–595
Sheng Y, Lawrence CJ, Briscoe BJ, Thornton C (2004) Numerical studies of uni-
axial powder compaction process by 3D DEM. Engrg Comp 21:304–317
Sitharam TG (2000) Numerical simulation of particulate materials using discrete
element modeling. Curr Sci 78:876–886
Thornton C (2000) Numerical simulations of deviatoric shear deformation of
granular media. Geotechnique 50:43–53
Thornton C, Zhang L (2003) Numerical simulations of the direct shear test. Chem
Eng Technol 26:153–156
Thornton C, Zhang L (2006) A numerical examination of shear banding and sim-
ple shear non-coaxial flow rules. Phil Mag 86:3425–3452
Ting JM, Khwaja M, Meachum LR, Rowell JD (1993) An ellipse-based discrete
element model for granular materials. Int J Num Anal Meth Geomech
17:603–623
228 Y. Wang and P. Mora

Toomey A, Bean CJ (2000) Numerical simulation of seismic waves using a dis-

crete particle scheme. G J I 141:595– 604
Tordesillas A, Walsh DCS (2002) Incorporating rolling resistance and contact ani-
sotropy in micromechanical models of granular media. Powder Tech
124:106–111
Tordesillas A, Peters J (2004) Role of particle rotations and rolling resistance in a
semi-infinite particulate solid indented by a rigid flat punch. The 12th Bien-
nial Computational Techniques and Applications Conference, 27 Sep–1 Oct
2004. The University of Melbourne, Victoria, Australia
Wang YC, Yin XC, Ke FJ, Xia MF, Peng KY (2000) Numerical simulation of
rock failure and earthquake process on mesoscopic scale. Pure Appl Geophys
157:1905–1928
Wang YC, Mora P, Yin C, Place D (2004) Statistical tests of load-unload response
ratio signals by Lattice Solid Model: implication to tidal triggering and earth-
quake prediction. Pure Appl Geophys 161:1829–1839
Wang YC, Abe S, Latham S, Mora P (2006) Implementation of particle-scale rota-
tion in the 3-D Lattice Solid Model. Pure Appl Geophys 163:1769–1785
Wang YC, Mora P (2008a) Modeling wing crack extension: implications to the
ingredients of Discrete Element Model. Pure Appl Geophys, 165: 609–620
Wang YC (2008b) A new algorithm to model the dynamics of 3-D bonded rigid
bodies with rotations. Acta Geotechnica, in press
Wang YC, Mora P (2008c) Elastic properties of regular lattices and calibration of
3-D Discrete Element Model. J Mech Phys Solids, in press
Yang RY, Zou RP, Yu AB (2000) Computer simulation of the packing of fine par-
ticles. Phys Rev E 62:3900–3908
Young RP, Hazzard JF, Pettitt WS (2000) Seismic and Micromechanical Studies
of Rock Fracture. G R L 27:1667–1670
Zhang D, Whiten WJ (1999) A new calculation method for particle motion in tan-
gential direction in discrete element simulations. Powder Tech 102:235–243
Zhou YC, Wright BD, Yang, RY Xu BH, Yu AB (1999) Rolling friction in the
dynamic simulation of sandpile formation. Physica A 269:536–553