Hindawi

Advances in Materials Science and Engineering

Volume 2022, Article ID 2514023, 15 pages
https://doi.org/10.1155/2022/2514023

Research Article
A Study of Contact Detection between Noncircular Particles in
Discrete Element Method

Chengbo Li , Qianqian Gao, and Xianchang Li

College of Materials Science and Engineering, Anyang Institute of Technology, Anyang 455000, China

Correspondence should be addressed to Chengbo Li; [email protected] and Xianchang Li; [email protected]

Received 25 March 2022; Revised 13 July 2022; Accepted 20 July 2022; Published 31 August 2022

Academic Editor: Francesco Ruffino

Copyright © 2022 Chengbo Li et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

One of the key points of modeling noncircular particles in the discrete element method is the contact detection of particles. In this
study, a general contact detection algorithm of two-dimensional particles with analytic shape functions is provided. The contact
detection of particles with strictly convex shape function, such as ellipses and superellipses, is solved by Newton–Raphson method,
and a grid method is provided to deal with heart-shaped particles. The grid method can be generalized into a particle system, in
which the shape function is not convex. The accuracy and stability of the algorithm are verified by a series of tests. For the collision
of a pair of ellipses with an aspect ratio of α  a/b  1000, the efficiency is not worse than the Newton–Raphson method. For
random packing under gravity, no residual kinetic energy is observed, and the force that acts on the bottom is equal to the gravity,
that is, the system reaches a mechanical equilibrium state. After equilibrium, the process of hopper discharge is also simulated. The
present method is suitable for arbitrarily shaped particles with analytic shape functions in two-dimensional cases. In addition to
superellipses, the method can also be applied to other particle systems as long as the shape functions are given.

1. Introduction surfaces are described by a quadratic equation. If two var-

iables are found in the polynomial, then we have a conic
The granular matter is ubiquitous in nature. Various (curve) and a quadric surface for three variables [15].
amorphous structures can be found in aggregates of non- Several contact detection algorithms, such as intersec-
spherical or noncircular particles [1]. These aggregate tion strategies [16], curvature simplifications [17], Perram
models are usually used to study the structures of many and Wertheim (PW) overlap potential [18–22], and char-
materials, such as, glass [2], nanoparticles [3], and colloids acteristic polynomial methods [13–15, 23], have been in-
[4–6]. The discrete element method (DEM) can easily obtain troduced and developed [24].
the microscopic properties that are difficult to obtain in The main idea in PW overlap potential algorithm is
experiments, and has been an effective method to study the described as follows: Two particles are fixed at their cen-
process of the particle system. Scientific attention has troids. Then, one or both of them are rescaled so that they
broadened the study of spherical aggregates, which are the have only one common point. Next, the scaling factor and
simplest shapes in the past few decades [7–9]; moreover, the the contact point are obtained. The two particles are in
packings of nonspherical or noncircular particles have external tangency if the scaling factor is equal to 1. The
attracted considerable attention [10–12]. particles are separated if the scaling factor is higher than 1
In DEM, contact detection is the most time-consuming and overlapped if the factor is smaller than 1. A requirement
part. Moreover, the computational time is notably enhanced of particle shape being smooth and convex is important in
when determining the interaction of nonspherical or non- this algorithm because it makes the normal vector n(r) be a
circular particles whose mathematically rigorous treatment one-to-one function. A unique normal vector exists at any
is notably nontrivial [13, 14]. The simplest noncircular and point on the particle surface, and one can determine the
nonspherical particles are ellipses and ellipsoids, whose point on the surface that corresponds to normal vector as
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2 Advances in Materials Science and Engineering

long as it is provided. *e overlap potential can be deter- where ζ ≥ 1 is an exponent, a and b are the semiaxis lengths,
mined analytically for ellipsoids, and the maximum of the and α is the value of equipotential surface. An exponent 1/ζ
overlap potential can be easily found numerically [25]. is added above to properly normalize the convex function
In characteristic polynomial methods, the relative po- and define the particle shape, even though it is not strictly
sition of two ellipsoids is related to the root pattern of their necessary. It is a simple ellipse when ζ � 1 and f(x) � 1. It is
characteristic equation. Specifically, two ellipsoids are sep- a rectangle with sides 2a and 2b when ζ ⟶ ∞ and
arated if and only if their characteristic equation has two f(x) � 1. Another definition of superellipse is based on
distinct positive roots. Furthermore, they touched each other 􏼌􏼌x􏼌􏼌n 􏼌􏼌y􏼌􏼌m
􏼌 􏼌 􏼌 􏼌
externally if and only if one positive double root exists. *is f(x) � 􏼌􏼌􏼌 􏼌􏼌􏼌 + 􏼌􏼌􏼌 􏼌􏼌􏼌 , (2)
a b
result allows us to compute the configuration of two el-
lipsoids by simply determining the number of the positive where n and m are positive exponents, and f(x) � C defines
roots of a polynomial, which can be derived by the classical different equipotential surfaces. *e value of C determines
Descartes’ rule and the modified sign variation number of the size of a superellipse. Figure 1 shows several superellipses
the signed subresultant sequence. *e algebraic conditions with various equipotential surfaces at the same centroid and
given by Wang et al. [23] laid the theoretical foundation for orientation (θ � π/4), where the shape parameters are n � 6,
the follow-up practical applications in the collision detection m � 8, a � 2, and b � 4.
for ellipsoids. *e solution of the quartic polynomial *e equipotential surface value of C controls the size of a
equation for two-dimensional (2D) elliptical particles is particle. *at is, a larger value of C corresponds to a larger
complicated and unstable because of considerable round-off particle when the shape function is the same. *e main idea
errors caused by the limited computer accuracy. Details of of the PW overlap potential algorithm is that two particles
the corresponding analysis have been described in Reference can be rescaled to contact each other if they are separated or
[26]. Similarly, the solution of the sixth-order polynomial overlapped and to obtain the scaling factor (or the value of
equation for three-dimensional (3D) ellipsoidal particles is C) and the contact point position. Over many repetitions,
more complicated and unreliable because of several round- two particles are in external tangency if the scaling factor is
off errors [27]. To avoid directly solving a quartic polynomial equal to 1.
equation or a sixth-order polynomial equation, a parametric Here, a specific example is provided to explain the idea.
equation of ellipsoid is utilized to transform the sixth-order *e parameters of three particles, namely, i, j, and k, are
polynomial equation into a quartic polynomial equation by listed in Table 1, in which θ is the orientation, xc and yc are
two approximations [28]. Furthermore, a golden section the position coordinates of the centroid, vx and vy are the
search algorithm is used instead of a sixth-order polynomial translational velocities, and ω is rotational velocity.
equation to detect the overlapping of ellipsoids [29]. Particles i and j are separated, as shown in Figure 2(a). A
However, the algorithm is restricted to a narrow range of scaling factor is used to enlarge the particles simulta-
aspect ratios. neously. If particles are overlapped, as shown in
*is study attempts to present a universal method to Figure 2(b), they can be reduced to be in external tangency
determine the contact point of 2D particles that have an- by using the scaling factor. *erefore, whatever particles are
alytical shape functions for particles with strictly convex separated or overlapped, only one scaling factor guarantees
shape functions, such as ellipses and superellipses. *e that they are in external tangency. Two particles are in
Newton–Raphson (NR) method is used to obtain the scaling external tangency if the scaling factor is equal to 1. *ey are
factor and the contact point, and a grid method is used to separated if the scaling factor is higher than 1 and over-
deal with heart-shaped particles with concave shape func- lapped if the scaling factor is smaller than 1. *us, the
tion. *e grid method is also suitable for ellipses and problem of contact detection is translated into the solution
superellipses and is verified by simulations of random of the scaling factor. *e solving process is presented in
packing under gravity and hopper discharge. *e proposed detail as follows.
method never diverges without an initial iteration point as *e shape functions of particles i and j in the body-fixed
needed in the NR method. *e present method is suitable for coordinate frame are
arbitrarily shaped particles with analytic shape functions in n m
⎪
⎧
⎪ x i y i
2D cases. In addition to superellipses, the method can also be ⎪
⎪ f i (x) � 􏼠 􏼡 + 􏼠 􏼡 ,
⎪
⎪ ai bi
applied to other particle systems as long as the shape ⎪
⎨
functions are given. ⎪ (3)
⎪
⎪ n m
⎪
⎪ x j y j
⎪
⎪ f (x) � 􏼠 􏼡 + 􏼠 􏼡 .
⎩ j aj bj
2. Contact Detection of Superellipses
Compared with ellipses, studies on superellipse are infre- *eir shape functions in the space-fixed coordinate
quent [21]. For example, a superellipse is given by the shape frame can be transformed to
function (in the body-fixed coordinate frame)
􏼌􏼌x􏼌􏼌2ζ 􏼌􏼌y􏼌􏼌2ζ 1/ζ ⎨ Fi (X) � fi Ri X − Xi 􏼁􏼁,
⎧
􏼌 􏼌 􏼌 􏼌 ⎩ F (X) � f 􏼐R 􏼐X − X 􏼑􏼑, (4)
f(x) � 􏼠􏼌􏼌􏼌 􏼌􏼌􏼌 + 􏼌􏼌􏼌 􏼌􏼌􏼌 􏼡 � α, (1) j j j j
a b
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Advances in Materials Science and Engineering 3

5
4
3
2
1

Y (mm)
0
-1
-2
-3
-4
-5

-6 -4 -2 0 2 4 6
X (mm)
C=1 C = 13
C=5 C = 17
C=9 C = 21
Figure 1: Superellipses of various equipotential surfaces.

Table 1: Parameters of superellipses.

Particle a/mm b/mm n m θ/rad xc /mm yc /mm vx /mm·s− 1 vy /mm·s− 1 ω/rad·s− 1
i 2.5 20 2 6 2π/3 5 −9 1 30 20
j 3 10 4 6 π/4 7 15 20 − 50 1
k 3 10 4 6 π/4 −8 −5 — — —

Seperated super-ellipses Overlapped super-ellipses

30
20

20
10
j
10 0
Y (mm)

Y (mm)

k
i
0 -10

-10 i -20

-30
-20

-40
-30
-10 0 10 20 -10 0 10 20
X (mm) X (mm)
(a) (b)

Figure 2: Two kinds of relative positions of superellipses. (a) Separated superellipses. (b) Overlapped superellipses.

where Xβ is the centroid of particle β in the space-fixed cos θβ sin θβ

coordinate frame, and Rβ is the rotational matrix between Rβ � ⎝
⎛ ⎠
⎞. (5)
the body-fixed coordinate frame and the space-fixed coor- − sin θβ cos θβ
dinate frame, β � i or j. If the orientation of particle β is θβ ,
then the expression of the rotational matrix is
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4 Advances in Materials Science and Engineering

Enlarged super-ellipses Reduced super-ellipses

40 20

30
10
20
j
10 0

Y (mm) k

Y (mm)
0 i
-10
i
-10
-20
-20

-30 -30

-40
-10 0 10 20 -10 0 10
X (mm) X (mm)
(a) (b)

Figure 3: Contact detection of separated and overlapped particles. (a) Enlarged superellipses. (b) Reduced superellipses.

If the two particles are tangent, then only one point XT zFn
satisfies fi (R(XT − Xi )) � fj (R(XT − Xj )) because this Jnm � . (9)
zXm
point is on both particles i and j. At this point, the gradients
of both surfaces are parallel (with opposite direction), *en, equation (8) can be rewritten as follows:
namely, ∇X Fi (X) + ∇X Fj (X) � 0. If the two particles are
separated, as shown in Figure 2(a), a scaling factor λ (or C) F(X + δX) � F(X) + J · δX + O􏼐δX2 􏼑. (10)
that is used to rescale the particles and guarantee one unique By neglecting terms of order δX2 and higher and setting
contact point between particles must be available [19, 30]. F(X + δX) � 0, we obtain a set of linear equations for the
Given λ ∈ (0, 1),λFi (X) � 1 means that particle i is enlarged corrections δX that move each function closer to zero si-
to the equipotential surface of Ci � 1/λ, and (1 − λ)Fj (X) � multaneously, namely,
1 means that particle j is enlarged. If the rescaled particles are
in external tangency, then the following equations exist: J · δX � − F(X). (11)
⎨ λ∇X Fi (X) + (1 − λ)∇X Fj (X) � 0,
⎧ (a), Equation (10) can be solved by the LU decomposition of
⎩ F (X) � F (X), (6)
i j (b), the matrix [32]. *e corrections are then added to the so-
lution vector,
Introducing α2 � (1 − λ)/λ and bringing the same
Xnew � Xol d + δX, (12)
equations obtained by Cleary et al. [31]
⎨ ∇X Fi (X) + α2 ∇X Fj (X) � 0,
⎧ (a), and the process is iterated to convergence.
⎩ F (X) � F (X), (7) *e aforementioned solution method is substituted into
i j (b). equation (7), F(X) � (F1 (X), F2 (X), F3 (X)) and variables
X � (X, Y, α) are obtained, where
Usually, these equations can be solved by the NR
method. ⎪
⎧ zF (X) zFj (X)
⎪
⎪
⎪ F1 (X) � i + α2 ,
*e NR method provides a very efficient way of con- ⎪
⎪ zX zX
verging to a multi-dimensional root. Let X denote the entire ⎪
⎪
⎪
⎪
vector of values Xn and F denote the entire vector of ⎨
⎪ zF (X) zFj (X) (13)
functions Fn , n � 1, ..., N. In the neighborhood of X, each of ⎪
⎪ F2 (X) � i + α2 ,
⎪
⎪ zY zY
the functions Fn can be expanded in Taylor series, ⎪
⎪
⎪
⎪
N ⎪
⎩
zFn F3 (X) � Fi (X) − Fj (X).
Fn (X + δX) � Fn (X) + 􏽘 δXm + O􏼐δX2 􏼑. (8)
m�1
zX m
After simple variable substitutions, the transition for-
*e matrix of partial derivatives that appear in equation mulas of gradients between the space-fixed coordinate frame
(8) is the Jacobian matrix J, which is described as and body-fixed coordinate frame are expressed as follows:
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Advances in Materials Science and Engineering 5

20

15
j
10

Y (mm)
0

-5
i
-10

-15

-20

-20 -10 0 10 20 30
X (mm)
Figure 4: Collision process of superellipses i and j.

∇X Fβ (X) � RTβ ∇x fβ (x), (14)

6
5
x � Rβ 􏼐X − Xβc 􏼑, (15)
4
where RTβ , Xβc are the transposed matrix of the rotational 3
matrix Rβ and centroid of particle β, β � i or j.
2
*e Jacobian matrix of particle β reads
1
Y (mm)

Jβ � RTβ Hx 􏼐fβ (x)􏼑Rβ , (16)

0
2 2
z f β /zx z f β /zxzy ⎞ 2 -1
⎝
where Hx (fβ (x)) � ⎛ ⎠ is the Hes-
z2 f β /zyzx z2 f β /zy2 -2

sian matrix of the shape function fβ (x) [33, 34]. -3

*e Jacobian matrix of equation (11) is -4

2 j 2αzFj
j -5
⎜
⎛
i
J11 + α J11 Ji12 + α2 J12 ⎟
⎞ -6 -4 -2 0 2 4 6
⎜
⎜
⎜
⎜ zX ⎟ ⎟
⎟
⎟
⎟ X (mm)
⎜
⎜ ⎟
⎟
⎜
⎜
⎜ ⎟
⎟
⎜
⎜ ⎟
⎟
⎟ α=1 α = 16
⎜
⎜ ⎟
⎜
⎜ 2αzF j⎟
⎟
⎟ α=4 α = 25
J �⎜
⎜
⎜
⎜
j
Ji21 + α2 J21 i 2 j
J22 + α J22 ⎟
⎟
⎟ . (17)
⎜
⎜
⎜
⎜ zY ⎟ ⎟
⎟
⎟
⎟
α=9
⎜
⎜
⎜ ⎟
⎟
⎜
⎜ ⎟
⎟
⎟ Figure 5: Varied sizes of heart-shaped particles.
⎜
⎜
⎜ ⎟
⎟
⎝ zF zFj zFi zF ⎟
⎠
i j
− − 0 If particles move and rotate with the translational and
zX zX zY zY angular velocities listed in Table 1, the collision process is
Substitute equations (13)–(17) into equations (11) and shown in Figure 4. *e collision time is t � 0.0197 s, and the
(12), iteration stops if either the sum of the magnitudes of collision point is at XT � (13.844, 6.3381). It takes 0.265 s
Fi (X) is less than a tolerance, or the sum of the absolute (440 time-steps) to calculate the process. *e blue curves
values of the corrections δX is less than a tolerance. illustrate the initial positions of particles, the green ones
*e particles in Figure 2 are rescaled by using the NR illustrate the trajectories, and the red ones illustrate the
method to find the external tangent point. Obviously, positions at the collision moment.
particles i and j need to be enlarged, and particles i and k NR method is a universal method used to determine the
should be reduced. *e value of the equipotential surface is contact point of particles with strictly convex shape func-
scaled from 1 to 6.2893 for particles i and j and to 0.3789 for tions. *e solution converges steadily to the unique contact
particles i and k. *e tangent points are located at XT1 � point through a scaling factor if the initial iteration point is
(14.6715, 3.7305) and XT2 � (− 2.8060, − 11.7405) with a provided properly. If the shape function is not strictly
tolerance of 10− 8, respectively (Figure 3). *e average convex and the solution is not unique, then it is likely to
computational time is 19.4 ms. cause the divergence of the NR method. For example, it will
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6 Advances in Materials Science and Engineering

Table 2: Parameters of heart-shaped particles.

Particles α θ/rad xc /mm yc /mm vx /mm·s− 1 vy /mm·s− 1 ω/rad·s− 1
ih 1 2π/9 2.5 5 5 40 2
jh 2 π/6 5 7 − 20 − 50 −3
kh 2 π/6 3 7 — — —

Separated heart-shaped particles Overlapped heart-shaped particles

11
9
10
8.5
9 8

8 7.5
kh
jh 7
Y (mm)

Y (mm)
7
6.5
6
6
ih
5 5.5
ih
4 5
4.5
3
4
2
2 3 4 5 6 2 3 4
X (mm) X (mm)
(a) (b)

Figure 6: Two kinds of relative positions of heart-shaped particles. (a) Separated heart-shaped particles. (b) Overlapped heart-shaped
particles.

Enlarged heart-shaped particles Reduced heart-shaped particles

11
9
10

9
8
8
jh kh
7
Y (mm)

Y (mm)

7
6
ih
5 6
4
ih
3 5

1 4
2 4 6 2 3 4
X (mm) X (mm)
(a) (b)

Figure 7: Contact detection of separated and overlapped heart-shaped particles. (a) Enlarged heart-shaped particles. (b) Reduced heart-
shaped particles.
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Advances in Materials Science and Engineering 7

8.5

7.5
jh
7

Y (mm)
6.5

5.5
ih
5

4.5

1 2 3 4 5 6
X (mm)
Figure 8: Collision process of heart-shaped particles.

Table 3: Parameters of ellipses.

Particle a/mm b/mm n m θ/rad xc /mm yc /mm vx /mm·s− 1 vy /mm·s− 1 ω/rad·s− 1
i 2.5 0.1 2 2 2π/3 5 −9 1 30 20
j 100 0.1 2 2 π/4 7 15 20 − 50 1
k 100 0.1 2 2 π/4 4 −8 — — —

Separated ellipses Overlapped ellipses Enlarged ellipses Reduced ellipses

10
-4 8 -7
10
Y (mm)

Y (mm)
Y (mm)

Y (mm)

j
-6 6 j
-8 k
0 i
-8 4 -9
i k i
-10 i
-10 2 -10

-20 -10 0 10 20 0 2 4 6 8 -6 -4 -2 0 2 4 2 4 6
X (mm) X (mm) X (mm) X (mm)
(a) (b) (c) (d)

Figure 9: Contact detection of ellipses by using the grid method. (a) Separated ellipses. (b) Overlapped ellipses. (c) Enlarged ellipses. (d)
Reduced ellipses.

fail to determine the contact point between heart-shaped *e size of heart-shaped particles is controlled by a
particles. subtracted constant (here, the subtracted constant is “1”) in
equation (18) instead of the value of the shape function.
*us, the generalized shape function can be written as
3. Contact Detection of Heart-Shaped Particles
3
f(x, α) � 􏼐x2 + y2 − α􏼑 − x2 y3 � 0. (20)
*e shape functions of heart-shaped particles are defined in
different ways, such as equations (18) and (19) Particles with different values of α are shown in Figure 5,
2
f(x) � 􏼐x + y − 1􏼑 − x y ,2 3 2 3
(18) and the shape functions are not strictly convex. *e normal
vector n(x) of each point on the particle is not a one-to-one
function, which may be the reason for the failure of using the
3 √3 ��2 2
f(x) � x2 + 2􏼒 x − y􏼓 − 1. (19) NR method.
5
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8 Advances in Materials Science and Engineering

Initial positions Final positions

100 100

80 80
60
60
40
40

Y (mm)
Y (mm)

20
20
0
0
-20
-20
-40
-40 -60
-60 -80
-80 -60 -40 -20 0 20 40 60 80 -60 -40 -20 0 20 40 60 80
X (mm)
X (mm)
(a) (b)
Process of collision Local zoom
80 35
30
60
25
40
20
20 15
Y (mm)

Y (mm)

10
0
5
-20
0
-40 -5

-60 -10
-15
-50 0 50 100 -30 -20 -10 0 10 20 30
X (mm) X (mm)
(c) (d)

Figure 10: Collision process of two ellipses. *e smaller ellipse is particle i and the bigger one is particle j. (a) Initial positions. (b) Final
positions. (c) Process of collision. (d) Local zoom.

Table 4: Forces and torques between particles. Table 4: Continued.

Properties Equations
Properties Equations
Tangential spring ⇀ ⇀ ∧
Normal fct � − μs |fcn |(1 − (1 − δt /δt,max )3/2 )δt
δn force
overlapping ⇀ ⇀ 􏽱����������
Tangential ⇀
Tangential ∧ ⇀ ⇀ fdt � − ct (6μs mij |fcn | 1 − δt /δt, max /δt, max )1/2 v t,ij
viscous force
displacement δt � δ t /|δ t | ⇀ ⇀ ⇀ ⇀
vector Torque qi � r ic × (fct + fdt )
Tangential Relative sliding ⇀ ⇀ ⇀ ⇀ ⇀ ⇀ ⇀
maximum δt, max � μs (2 − v)/(2 − 2v)δn v ij � v i − v j +ωi × r ic − ωj × r jc
velocity
displacement Normal relative ⇀ ⇀ ∧ ∧
Poisson’s ratio v v n,ij � ( v ij ∙n)n
velocity
Effective Young’s Tangential ⇀ ⇀ ⇀
E∗ � E/(2 − 2v2 ) v t,ij � v ij − v n,ij
modulus relative velocity
Effective mass mij � mi mj /(mi + mj ) Relative rolling → ∧ → →
v r � − (ricn rjcn /rijn )n × ( ω i − ω j )
velocity
Effective
R∗ � Ri Rj /(Ri + Rj )
curvature
√��� ∧ →r ∧ r
⎨→
Normal spring ⇀ ⎧
⎪
fcn � 4/3E∗ R∗ δ3/2
n n Rolling force f r � min (μr fn , | f 0 |)f0
force ⎪→ r →
􏽰���� ⇀ ⎩ →
Normal viscous ⇀ f 0 � − kr ξ r − cr v r
fdn � − cn (8mij E∗ R∗ δn )1/2 v n,ij Rolling torque rolling →
force ⇀
q
∧
� (ricn rjcn /rijn )n × f r
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Advances in Materials Science and Engineering 9

Table 5: Parameters of simulations.

Parameters Values SI units
Length-unit 1 1 mm
Mass-unit 1 1 mg
Time-unit 1 1 ms
Particle density ρ 5 5 × 103 kg/m3
Young’s modulus E 106 1 GPa
Friction coefficient μs 0.50 0.50
Poisson’s ratio v 0.25 0.25
Normal viscosity coefficient cn 0.4 0.4
Tangential viscosity coefficient ct 0.4 0.4
Rolling stiffness kr 106 106 N/m
Rolling viscosity cr 103 1.0 kg/s
Rolling friction coefficient μr 0.1 0.1
Integration time-step 1 × 10− 3 1 × 10− 6 s

Initial state
300

240
Final state
200
180
Y (mm)

160

120
Y (mm)

120

80
60
40

0 0
-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150
X (mm) X (mm)
(a) (b)

Figure 11: Random packing process of superellipses. (a) Initial state and (b) equilibrium state of the system.

A new algorithm, the so-called grid method, is intro- Fj 􏼐X, αj 􏼑 � f􏼐Rj 􏼐X − Xj 􏼑, αj 􏼑 � 0. (21)
duced to address the contact detection of heart-shaped
particles. *e main idea of the grid method is simple. *en, search the solutions of equation (21) to find the
Similarly, a specific example is provided to illustrate this smallest value Fmin of
method. *ree heart-shaped particles are denoted by ih , jh ,
and kh with the parameters as listed in Table 2, and their Fi (X, 0) � f Ri X − Xi 􏼁, 0􏼁. (22)
relative positions and orientations are shown in Figure 6. In
Table 2, θ is the orientation, xc and yc are the position *us, the corresponding point Xmin is the contact point
coordinates of the centroid, vx and vy are the translational when particle ih is rescaled by αmin , and αmin satisfies
velocities, and ω is rotational velocity. f(Ri (Xmin − Xi ), αmin ) � 0.
Particles ih and jh are separated, but ih and kh are For convenience, the reason to determine Xmin and αmin
overlapped in Figure 6. *erefore, two scaling factors must in the body-fixed coordinate frame is only a rotation in the
be provided to guarantee that particle ih is in external space-fixed coordinate frame with the relative position
tangency with particles jh and kh . *e key problem is to unchanged.
determine the smallest value of the scaling factor that is used Given a function in the body-fixed coordinate frame of
to rescale a particle to touch with another one. Generally, particle ih ,
particle jh is fixed and particle ih is rescaled to find the 3
smallest α and the tangent point. First, one should calculate g(x) ≡ 􏼐x2 + y2 􏼑 − x2 y3 , (23)
the points on the surface of particle jh with the centroid
position Xj , orientation Rj and αj , that is, the following denote A � x2 + y2 ,B � x2 y3 , and substitute equation (23)
equation should be solved. into the shape function f(x, α) � (A − α)3 − B � 0, we
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10 Advances in Materials Science and Engineering

105

100
Et/10-6J

Er/10-6J
100

10-5 10-5
0 0.5 1 1.5 0 0.5 1 1.5
time (s) time (s)
(a) (b)

×105 ×105
6.5 6.5
Eg/10-6J

Es/10-6J
6 6

5.5 5.5
0 0.5 1 1.5 0 0.5 1 1.5
time (s) time (s)
(c) (d)

Figure 12: Energies change with time. (a) Translational energy, (b) rotational energy, (c) gravity potential energy, and (d) total energy.

t=0s t=0.11s t=0.32s

450 450 450

400 400 400

350 350 350

Y (mm)

Y (mm)

Y (mm)

300 300 300

250 250 250

200 200 200

150 150 150

-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150
X (mm) X (mm) X (mm)

(a) (b) (c)

Figure 13: Snapshots of particle flow at different times. (a) t � 0 s. (b) t � 0.11 s. (c) t � 0.32 s.

obtain g � A3 − (A − α)3 , which means that the minimal g *e aforementioned process is used to deal with the
corresponds to the minimal α. A point xmin is in a series of example as shown in Figure 7. For the case of separation,
points on the surface of particle jh which makes the value of particle ih is enlarged to the surface of α � 2.5532 and the
g in equation (22) be the smallest gmin . In other words, when contact point is located at X � (3.3989, 6.6597). For the
particle ih is separated from particle jh , particle ih is enlarged case of overlapping, particle ih is reduced to the surface of
slowly, and the first common point xmin is on the equipo- α � 0.3674, and the contact point is located at
tential surface of αmin , thereby meeting the condition X � (2.5675, 5.9296). *e average computational time
gmin � A3 − (A − αmin )3 . (approximately 19.2 ms) to detect the tangent point and
Another way is to directly solve the equation the scaling factor is almost the same as that of
f(Ri (X − Xi ), α) � 0 for the points on the surface of superellipses.
particle jh and finds xmin and αmin . However, computa- If particles move and rotate with the translational and
tional efficiency and time have not been significantly angular velocities listed in Table 2, then the collision process
improved in this way. is shown in Figure 8. *e collision time is t = 0.0088 s, and
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Advances in Materials Science and Engineering 11

25

20

15 Q=cpbg1/2 (D-kd)3/2

Q (g/s)
10

0
0 10 20 30
D (mm)

Simulated
Fit (k=2.21)
Figure 15: Mass flow rate versus the outlet sizes for particles with
aspect ratios about 3 : 1, the outlet sizes range from 6dmaj to 16dmaj .
Symbol: simulated data, line: fitted by using the Beverloo’s law.

illustrate the trajectories, and the red ones illustrate the

positions at the collision moment.
When the grid method is used to determine the contact
point of superellipses, the computational time is 18 ms and
slightly lower than that of using the NR method. Never-
theless, the grid method is suitable for ellipses with a very
large aspect ratio of 1000 and a size dispersion of 40 times, in
which case the NR method will fail. *e parameters of el-
lipses are listed in Table 3. *e contact detection is shown in
Figure 9.
Figures 9(a) and 9(c) illustrate the contact detection
between particles i and j by enlarging particle i, and
Figures 9(b) and 9(d) illustrate the contact detection be-
tween particles i and k by reducing particle i. If particles i and
j move and rotate with the translational and angular ve-
locities listed in Table 3, then the collision process is shown
in Figures 10. Figures 10(a) and 10(b) are the initial and final
positions of the two particles, respectively. In Figure 10(c),
the green curves illustrate the initial positions of particles,
the blue ones illustrate the trajectories, and the red ones
illustrate the positions at the collision moment. A local zoom
of the collision process is shown in Figure 10(d).

4. Contact Detection Used in DEM

*e grid method is used in the following superellipse system
to further verify its validity. *e contact forces and torques of
simulation and parameters are listed in Tables 4 and 5, more
details can be found in previous studies [7, 35]. *e particle
system contains 8100 superellipses with shape function
f1 (x) ≡ |x/1.0|10 + |y/1.0|10 � 1. *e initial body-fixed co-
ordinate frame coincides with the space-fixed coordinate
frame. In the beginning, particles are packed cubically, and
Figure 14: Snapshot of granular flow in the hopper.
no overlapping occurs between them. *e rotational and
translational energies are zero. *en, the particles fall under
gravity onto the bottom, which consists of particles with
the collision point is at XT � (3.1465, 6.3711). It takes shape function f2 (x) ≡ |x/1.5|16 + |y/1.5|16 � 1. *e peri-
0.173 s to calculate the collision process. *e blue curves odic boundary condition is applied in the x direction. *e
illustrate the initial positions of particles, the green ones initial and final states are shown in Figure 11.
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
12 Advances in Materials Science and Engineering

Initial size Local zoom of initial size

80
-78.5
60
-79
40
-79.5
20
-80
Y (mm)

Y (mm)
0 -80.5
-20 -81
-40 -81.5
-60 -82

-80 -82.5

-100 -50 0 50 100 51 52 53 54 55 56 57

X (mm) X (mm)

Particle i Particle i
Particle j Particle j
(a) (b)
Contact aer enlargement Local zoom of contact

200 -77

100 -78
Y (mm)

Y (mm)

0 -79

-80
-100

-81
-200
-82
-300
-300 -200 -100 0 100 200 300 50 51 52 53 54 55 56 57
X (mm) X (mm)
Particle i Particle i
Particle j Particle j
(c) (d)

Figure 16: Contact detection between two ellipses with large eccentricity. (a) Initial size. (b) Local zoom of the initial size. (c) Contact after
enlargement. (d) Local zoom of contact.

*e energy changes with time are shown in Figure 12. At *e contact detection proposed in this work can also be
the final equilibrium state, the kinetic energy is almost 0, and used to study the granular flow in a hopper similar to
the force that acts on the bottom is equal to the gravity of the Reference [37]. *e granular flows of the elliptical particle
system. *us, the final state is a stable equilibrium state. system shown in Figure 14 are simulated with the grid
After reaching the equilibrium state, a hole is opened at method under different outlet sizes. *e relationships be-
the bottom, and its width is 24 mm, which is 12 times the tween the discharge rate and the outlet size are calculated
width of the particle. Figure 13 shows the snapshots of and according to Beverloo’s law [38], as shown in Figure 15.
particle flow at different times. After 0.11 s, 67 particles drop *e two-dimensional hopper (Figure 14) consists of a
through the hole, a semicircular structure of force arch rectangular channel of height (H � 800 mm) and width
comes into being, and the particle flow stops. *e stable arch (W � 80 mm). *e outlet is situated at the center of the
is shown in Figure 6, and the results are consistent with that bottom and has a length (D), which is varied at will during
of a previous study [36]. the simulation process. *e major principal axis and minor
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Advances in Materials Science and Engineering 13

t=0 t = 0.02ms t = 0.04ms t = 0.06ms

-5 -5 -5

-10 -10 -10

-10

Z (mm)
Z (mm)

Z (mm)
Z (mm)
-15 -15 -15
-15
-20 -20 -20
-20
-10 -10
-15 -5 -15 -5 -15
-10 Y -10 -20 -10 -15
Y ( -20 -15 (m -20 -15 Y( -15 -20 -10
mm m mm -25
) -25 -20 m) ) -25 -20 m) -20 m) -15
X (m X (m ) X (m Y (mm) -25 -20 m)
X (m

(a) (b) (c) (d)

t = 0.08ms t = 0.10ms t = 0.12ms t = 0.1924ms

-8
-10 -10 -10
-10
Z (mm)

Z (mm)

-12

Z (mm)
-15 -15

Z (mm)
-15 -14
-20 -16
-20
-20 -18
-10 -20 -10
-15 -15
-10 -10 -15 -15 )
-20 -20 -15 -15 -24 -22 -20 m
-15 -20 ) -18 -16 -20 (m
-25 m) -25 -20 m) mm -14 -12 X
Y (mm) -20
X (m Y (mm) X (m Y (mm) -25 -20 X( Y (mm)

(e) (f ) (g) (h)

Figure 17: Collision process of two superellipsoids. (a) t � 0 s. (b) t � 0.02 ms. (c) t � 0.04 ms. (d) t � 0.06 ms. (e) t � 0.08 ms. (f ) t � 0.10 ms.
(g) t � 0.12 ms. (h) t � 0.1924 ms.

principal axis of the elliptical particles are dmaj � 2 mm and particles i and j are in Figures 16(a) and 16(b) have the same
dmin � 0.66 mm. *e particles are initially randomly placed shape function f(x) � x2 /0.12 + y2 /1002 � 1. *eir centroid
in the closed hopper, and then they settle under the gravity to coordinates are Xi � (1.2, 3.4) and Xj � (2.1, − 19.5) and
minimize the gravitational potential and are allowed to leave angular coordinates are θi � π/5.6 and θj � π/4.5, respec-
the hopper from the outlet centered at the bottom. During tively. When using the NR method to calculate the contact
the steady discharge process, the number of elliptical par- point, singular matrices often appear when solving the in-
ticles as a function of time is recorded every 200 time-steps, verse Jacobian matrix. *is fails correct solution because the
from which we measure globally the instantaneous flow rate eccentricity is too large. But by using the grid method and
as ΔN/Δt, where ΔN is the number of particles passing enlarging particle j to satisfy fj (x) � x2 /0.12 + y2 /1002 �
through the exit during the time interval Δt. *e mean flow 13.6177 in the contact detection, the contact point is de-
rate Q is obtained by temporal averaging the instantaneous termined at (54.4032, − 81.2724) as shown in Figures 16(c)
flow rate during the steady flow state. Based on the above and 16(d).
calculation, the dependence of discharge rate on the outlet *e grid method is not only suitable for 2D cases but also
size is examined, as shown in Figure 15. for 3D superellipsoids. In Figure 17, two superellipsoids i
and j have different shape functions of
f i (x) � |x/4|6 + |y/4|6 + |z/4|12 � 1 and
5. Discussion and Conclusion f j (x) � |x/2|8 + |y/3|8 + |z/2|8 � 1, their centroid coordi-
*e iterative convergence direction and step length are nates are Xi � (− 10, − 15, − 10) and Xj � (− 20, − 25, − 20),
difficult to control when solving the contact point between and the corresponding Euler angles are
ellipses with large eccentricity by using the NR method, and (ψ i , θi , φi ) � ((π/9), (π/8), (π/7)) and
singular matrices often appear when solving the inverse (ψ j , θj , φj ) � ((π/6), (π/5), (π/4)), respectively. *e initial
Jacobian matrix, which makes the solution fail. *erefore, it velocities and angular velocities are
is difficult to use the NR method in discrete element sim- (vix , viy , viz ) � (− 20, − 10, − 15), (vjx , vjy , vjz ) � (10, 20, 15),
ulation of the particle system. However, solving the inverse (ωix , ωiy , ωiz ) � (20, − 30, 2), and
matrix is not needed in the grid method which is not affected (ωjx , ωjy , ωjz ) � (− 10, 20, − 2), respectively. By running the
by the major and minor axes of particles. For instance, program in the MATLAB compiling environment on a PC
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
14 Advances in Materials Science and Engineering

with a CPU of Intel(R) Core (TM), I9-8950HK 2.9 GHz, and packings,” Physical review e-statistical physics, plasmas, fluids,
32 GB of RAM, the NR method is employed to simulate the and related interdisciplinary topics, vol. 91, no. 2, Article ID
collision process of these two particles as shown in Figure 17. 022311, 2015.
In the simulation, the time-, length-, and mass-unit [7] L. Liu, Z. Zhang, and A. Yu, “Dynamic simulation of the
values are ms, mm, and mg, respectively. If the time-step is centripetal packing of mono-sized spheres,” Physica A: Sta-
set as 1 × 10− 7 s, the collision contact point will be located at tistical Mechanics and Its Applications, vol. 268, no. 3-4,
pp. 433–453, 1999.
(− 16.5993, − 20.3341, − 14.2559) after 1924 time-steps which
[8] S. Li and K. L. Ng, “Monte Carlo study of the sphere packing
means that there are 1924 contact detections of arbitrary
problem,” Physica A: Statistical Mechanics and Its Applica-
positions and angular coordinates. *e calculation process tions, vol. 321, no. 1-2, pp. 359–363, 2003.
of the grid method takes 0.986460 s, and the time of the NR [9] T. Aste, M. Saadatfar, and T. Senden, “Geometrical structure
method is 1.865421 s. Under the same solution accuracy of disordered sphere packings,” Physical Review E-Statistical
( < 10− 3 ), the time consumption of grid method is less than Physics, Plasmas, Fluids, and Related Interdisciplinary Topics,
that of NR method. vol. 71, no. 6, Article ID 061302, 2005.
In addition to dealing with superellipses, the use of grid [10] M. Moradi, S. Hashemi, and F. Taghizadeh, “Monte Carlo
method in detecting the contact of heart-shaped particles is a simulation of a confined hard ellipse fluid,” Physica A: Sta-
universal approach and is suitable for all particles with tistical Mechanics and Its Applications, vol. 389, no. 21,
analytical shape functions. *e advantage of the method is pp. 4510–4519, 2010.
that the inverse of the Jacobian matrix and the initial it- [11] A. Donev, F. H. Stillinger, P. M. Chaikin, and S. Torquato,
eration point are unnecessary. *e grid method will never “Unusually dense crystal packings of ellipsoids,” Physical
diverge, and the problem wherein singular matrices appear Review Letters, vol. 92, no. 25, Article ID 255506, 2004.
[12] A. Donev, S. Torquato, and F. H. Stillinger, “Pair correlation
in the NR method is avoided.
function characteristics of nearly jammed disordered and
ordered hard-sphere packings,” Physical review e-statistical
Data Availability physics, plasmas, fluids, and related interdisciplinary topics,
vol. 71, no. 1, Article ID 011105, 2005.
*e data used to support the findings of this study are [13] X. Jia, Y. K. Choi, B. Mourrain, and W. Wang, “An algebraic
available from the corresponding author upon request. approach to continuous collision detection for ellipsoids,”
Computer aided geometric design, vol. 28, no. 3, pp. 164–176,
Conflicts of Interest 2011.
[14] W. Wang, Y. K. Choi, B. Chan, M. S. Kim, and J. Wang,
*e authors declare no conflicts of interest. “Efficient collision detection for moving ellipsoids using
separating planes,” Computing, vol. 72, no. 1-2, pp. 235–246,
Acknowledgments 2004.
[15] K. Chan, Multiscale Optimization Methods and Applications,
*is work was supported by the Training Program for pp. 271–298, Springer, Boston, MA, USA, 200.
University Key Teachers of Henan Province (2019GGJS237) [16] L. Rothenburg and R. J. Bathurst, “Numerical simulation of
and the National Natural Science Foundation of China idealized granular assemblies with plane elliptical particles,”
(11804005). Computers and Geotechnics, vol. 11, no. 4, pp. 315–329, 1991.
[17] S. Johnson, J. R. Williams, and B. Cook, “Contact resolution
algorithm for an ellipsoid approximation for discrete element
References modeling,” Engineering computations, vol. 21, no. 2/3/4,
pp. 215–234, 2004.
[1] S. R. Williams and A. P. Philipse, “Random packings of
[18] A. Donev, I. Cisse, D. Sachs et al., “Improving the density of
spheres and spherocylinders simulated by mechanical con-
jammed disordered packings using ellipsoids,” Science,
traction,” Physical review e-statistical physics, plasmas, fluids,
vol. 303, no. 5660, pp. 990–993, 2004.
and related interdisciplinary topics, vol. 67, no. 5, Article ID
[19] J. W. Perram, J. Rasmussen, E. Præstgaard, and J. L. Lebowitz,
051301, 2003.
[2] D. He, N. N. Ekere, and L. Cai, “Computer simulation of “Ellipsoid contact potential: t,” Physical Review E-Statistical
random packing of unequal particles,” Physical review Physics, Plasmas, Fluids, and Related Interdisciplinary Topics,
e-statistical physics, plasmas, fluids, and related interdisci- vol. 54, no. 6, pp. 6565–6572, 1996.
plinary topics, vol. 60, no. 6, pp. 7098–7104, 1999. [20] A. Donev, R. Connelly, F. H. Stillinger, and S. Torquato,
[3] Y. Sun and Y. Xia, “Shape-controlled synthesis of gold and “Underconstrained jammed packings of nonspherical hard
silver nanoparticles,” Science, vol. 298, no. 5601, particles: ellipses and ellipsoids,” Physical Review E-Statistical
pp. 2176–2179, 2002. Physics, Plasmas, Fluids, and Related Interdisciplinary Topics,
[4] A. Kuijk, A. Van Blaaderen, and A. Imhof, “Synthesis of vol. 75, no. 5, Article ID 051304, 2007.
monodisperse, rodlike silica colloids with tunable aspect ra- [21] A. Donev, J. Burton, F. H. Stillinger, and S. Torquato, “Tetratic
tio,” Journal of the American Chemical Society, vol. 133, no. 8, order in the phase behavior of a hard-rectangle system,”
pp. 2346–2349, 2011. Physical Review B: Condensed Matter, vol. 73, no. 5, Article ID
[5] L. Rossi, S. Sacanna, W. T. M. Irvine, P. M. Chaikin, D. J. Pine, 054109, 2006.
and A. P. Philipse, “Cubic crystals from cubic colloids,” Soft [22] F. M. Schaller, M. Neudecker, M. Saadatfar, G. W. Delaney,
Matter, vol. 7, no. 9, pp. 4139–4142, 2011. G. E. Schröder-Turk, and M. Schröter, “Local origin of global
[6] S. I. R. Castillo, D. M. E. *ies-Weesie, and A. P. Philipse, contact numbers in frictional ellipsoid packings,” Physical
“Formation and liquid permeability of dense colloidal cube Review Letters, vol. 114, no. 15, Article ID 158001, 2015.
 5928, 2022, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2022/2514023 by Universidad De Concepcion, Wiley Online Library on [06/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Advances in Materials Science and Engineering 15

[23] W. Wang, J. Wang, and M. S. Kim, “An algebraic condition

for the separation of two ellipsoids,” Computer Aided Geo-
metric Design, vol. 18, no. 6, pp. 531–539, 2001.
[24] X. Lin and T. T. Ng, “Contact detection algorithms for three-
dimensional ellipsoids in discrete element modelling,” In-
ternational Journal for Numerical and Analytical Methods in
Geomechanics, vol. 19, no. 9, pp. 653–659, 1995.
[25] A. Donev, S. Torquato, and F. H. Stillinger, “Neighbor list
collision-driven molecular dynamics simulation for non-
spherical hard particles. I. Algorithmic details,” Journal of
Computational Physics, vol. 202, no. 2, pp. 737–764, 2005.
[26] A. Džiugys and B. Peters, International Journal for Numerical
and Analytical Methods in Geomechanics, vol. 25, p. 1487,
2001.
[27] P. Schneider and D. Eberly, Geometric Tools for Computer
Graphics, Elsevier, Amsterdam, Netherlands, 2002.
[28] H. Ouadfel and L. Rothenburg, “An algorithm for detecting
inter-ellipsoid contacts,” Computers and Geotechnics, vol. 24,
no. 4, pp. 245–263, 1999.
[29] Y. Lee, C. T. Yang, and C. S. Chien, “A 3D ellipsoid-based
model for packing of granular particles,” International Journal
of Computer Applications in Technology, vol. 17, no. 3, p. 148,
2003.
[30] J. W. Perram and M. Wertheim, “Statistical mechanics of hard
ellipsoids. I. Overlap algorithm and the contact function,”
Journal of Computational Physics, vol. 58, no. 3, pp. 409–416,
1985.
[31] P. Cleary, N. Stokes, and J. Hurley, Proceedings of 8th In-
ternational Computational Techniques and Applications
Conference CTAC97, Adelaide, 1997.
[32] W. Press, S. Teukolsky, W. Vetterling, and B. Flannery, Ge
Art of Scientific Computing, Cambridge University Press, 3rd
edition, 2007.
[33] G. Houlsby, “Potential particles: a method for modelling non-
circular particles in DEM,” Computers and Geotechnics,
vol. 36, no. 6, pp. 953–959, 2009.
[34] A. Podlozhnyuk, S. Pirker, and C. Kloss, “Efficient imple-
mentation of superquadric particles in Discrete Element
Method within an open-source framework,” Computational
Particle Mechanics, vol. 4, no. 1, pp. 101–118, 2017.
[35] W. Zhong, A. Yu, X. Liu, Z. Tong, and H. Zhang, “DEM/CFD-
DEM modelling of non-spherical particulate systems: theo-
retical developments and applications,” Powder Technology,
vol. 302, pp. 108–152, 2016.
[36] I. Zuriguel, “Invited review: c,” Papers in Physics, vol. 6, Article
ID 060014, 2014.
[37] X. Diao, H. Yan, T. Li, H. Zhang, and H. Bo, “Pressure
fluctuation instability in vertical plug formation of coarse
particles with a non-mechanical feeder,” Exp. Comput.
Multiph. Flow, vol. 2, pp. 79–88, 2020.
[38] W. A. Beverloo and H. A. Leniger, Journal van de Velde.
Chemical Engineering Science, vol. 15, pp. 85030–85036, 1961.