Efficient Implementation of Superquadric Particles in Discrete
Efficient Implementation of Superquadric Particles in Discrete
Efficient Implementation of Superquadric Particles in Discrete
DOI 10.1007/s40571-016-0131-6
Abstract Particle shape representation is a fundamental model and robustness of the implemented superquadric DEM
problem in the Discrete Element Method (DEM). Spherical code.
particles with well known contact force models remain pop-
ular in DEM due to their relative simplicity in terms of ease Keywords Discrete Element Method · Granular flow ·
of implementation and low computational cost. However, Contact detection · Superquadric · Non-spherical particles ·
in real applications particles are mostly non-spherical, and Angle of repose · Hopper discharge
more sophisticated particle shape models, like superquadric
shape, must be introduced in DEM. The superquadric shape
can be considered as an extension of spherical or ellipsoidal 1 Introduction
particles and can be used for modeling of spheres, ellip-
soids, cylinder-like and box(dice)-like particles just varying In many engineering applications different types of particles
five shape parameters. In this study we present an efficient have to be stored, transported, mixed, or segregated. Despite
C++ implementation of superquadric particles within the that, knowledge of the static and dynamic behavior of partic-
open-source and parallel DEM package LIGGGHTS. To ulate solids is still not fully understood. Such knowledge is of
reduce computational time several ideas are employed. In the major importance for a proper design of processing units of
particle–particle contact detection routine we use the mini- silos, rotating drums, and others [23]. The Discrete Element
mum bounding spheres and the oriented bounding boxes to Method (DEM), developed by Cundall and Strack [14], has
reduce the number of potential contact pairs. For the particle– proven to be an efficient tool for modeling granular materials.
wall contact an accurate analytical solution was found. We In DEM granular material is treated as a system of distinct
present all necessary mathematics for the contact detection interacting particles. Each particle has own mass, velocity,
and contact force calculation. The superquadric DEM code position, and contact properties; it obeys Newton’s second
implementation was verified on test cases such as angle of law and is tracked individually. Together with the rapidly
repose and hopper/silo discharge. The simulation results are increasing computational power available, DEM becomes
in good agreement with experimental data and are presented more and more popular among engineers and researchers.
in this paper. We show adequacy of the superquadric shape A comprehensive overview of major DEM applications can
be found in [56].
Many DEM codes still employ disks (in 2D) and spheres
B Alexander Podlozhnyuk (in 3D) to represent particle shapes due to their imple-
alexander.podlozhnyuk@dcs-computing.com
mentation simplicity and efficiency in speed of contact
1 DCS Computing GmbH, Industriezeile 35, 4020 Linz, Austria detection, which results in faster code development and lower
2 Christian-Doppler Laboratory on Particulate Flow Modelling,
computational time. The rolling friction correction can be
Johannes Kepler University, Altenberger Strasse 69, theoretically linked to various physical effects to model par-
4040 Linz, Austria ticle non-sphericity using spherical particles, as emphasized
3 DCS Computing GmbH, CFDEM Research GmbH, by Ai et al. [2]. Moreover, the contact force models that
Industriezeile 35, 4020 Linz, Austria include normal and tangential forces for a pair of interact-
123
Comp. Part. Mech.
123
Comp. Part. Mech.
for the practical range of blockiness parameters. Validation where sign “◦” denotes quaternion multiplication [22], ωi =
work along with several application examples is presented. (ωi x , ωi y , ωi z )T = Ai−1 Ω i is the angular velocity in the
particle-based (canonical) coordinate system, and Ω i is the
angular velocity in the observer-fixed coordinate system.
2 Numerical model Rotational motion of a particle is tracked by the following
equation in the observer-fixed coordinate system:
2.1 Motion of an arbitrarily shaped particle
L̇ i = T i , (6)
In DEM each particle i obeys Newton’s second law and is
tracked individually by solving explicitly their trajectories: where L i = I i · Ω i is the angular momentum of the particle
i, I i is the tensor of inertia, and T i is the total torque acting
m i Ẍ Ci = F i , (2)
on a particle i with respect to the particle center. Note that
the normal force can also produce torque and must be taken
where m i and X Ci are the mass and the position of a particle
into account while calculating T i . For a spherical particle
center, Fi is the total force acting on a particle i that is the
Eq. (6) above can easily be resolved with regard to Ω since
sum of normal particle–particle, tangential particle–particle
its tensor of inertia is always constant and only possesses non-
and external non-particle forces like gravity.
zero elements in the diagonal: I spher e = 25 m R 2 E, where E
The contact force and the contact point between two
non-spherical particles depend on particles’ orientation, and is the identity tensor. In general case the t A · Î · A−1 , where
hence accurate determination and integration of orientation Î is the principal tensor of inertia, i.e., the tensor of inertia
becomes critical. Accurate determination of a particle’s ori- in particle-based coordinate system which contains non-zero
entation is also critical for the determination of the angular entries only in the diagonal: Iˆx , Iˆy and Iˆz . Moving to the
velocity of a particle. Orientation of a non-spherical par- body-fixed coordinate system yields the following general
ticle is usually described as a rotation of the coordinate form of Eq. (6):
vectors {e1 , e2 , e3 } that define the global observer-fixed ref-
erence frame to the coordinate vectors {ê1 , ê2 , ê3 } that define Iˆi ω̇i + ωi × Iˆi ωi = t i ⇐⇒
⎧
the local body-fixed reference frame. This rotation can be ⎨ Iˆi x ω̇i x + ( Iˆi z − Iˆi y )ωi y ωi z = ti x
⎪
tracked by rotation vectors [7] or by quaternions [22] that ⇐⇒ Iˆi y ω̇i y + ( Iˆi x − Iˆi z )ωi z ωi x = ti y (7)
⎪
⎩ Iˆ ω̇ + ( Iˆ − Iˆ )ω ω = t ,
are singularity-free in contradistinction to the methods based iz iz iy ix ix iy iz
on Euler angles. The quaternion of rotation q can easily be
constructed from the unit axis of rotation e and the angle of
where t i = (ti x , ti y , ti z )T = Ai−1 T i is the total torque acting
rotation α around this axis:
on a particle in the particle-based coordinate system.
q = (q0 , q1 , q2 , q3 )T = cos (α/2) + esin (α/2). (3) Equation (7) in conjunction with Eq. (5) can be solved
by various methods, such as described by Miller et al. [38],
The rotation matrix A = A(q) is constructed from the quater- Walton and Braun [51] and Omelyan [42–45]. Corresponding
nion components: analytical expressions for volume and principal moments of
⎛ ⎞ inertia can be found in works by Jaklič et al. [25,26].
1 − 2(q22 + q32 ) 2(q1 q2 − q0 q3 ) 2(q1 q3 + q0 q2 )
A = ⎝ 2(q1 q2 + q0 q3 ) 1 − 2(q12 + q32 ) 2(q2 q3 − q0 q1 ) ⎠ .
2.2 Neighbor search
2(q1 q3 − q0 q2 ) 2(q2 q3 + q0 q1 ) 1 − 2(q12 + q22 )
(4) In this section we describe the broad phase of the contact
detection algorithm. To perform simulations of large-scale
systems, it is essential to optimize the computational strat-
By definition of the rotation matrix: A · e1 = ê1 , A · egy. The number of potential contact pairs can be minimized
e2 = ê2 , A · e3 = ê3 , A−1 = AT . The orientation of a by employing a neighbor list to exclude particle pairs that are
particle can be updated every time step using the following a priori too far from each other to be in contact [28]. Different
expression [32]: techniques for constructing neighbor lists have been pro-
1 posed in the literature. These include the Verlet-Neighbour
q˙i =q ◦ ωi ⇐⇒
2⎧i List [49], Linked Cell method [46], NBS algorithm [39],
⎪
⎪ q̇i0 = (−qi1 ωi x − qi2 ωi y − qi3 ωi z )/2 and MR linear contact detection algorithm [40]. The number
⎨
q̇i1 = (qi0 ωi x + qi2 ωi z − qi3 ωi y )/2 of potential contact pairs in a neighbor list can be reduced
⇐⇒ (5)
⎪
⎪ q̇ = (qi0 ωi y + qi3 ωi x − qi1 ωi z )/2 with the help of bounding volumes. A bounding volume is
⎩ i2
q̇i3 = (qi0 ωi z + qi1 ωi y − qi2 ωi x )/2, a simple volume that encapsulates a more complex body.
123
Comp. Part. Mech.
⎧ n /n −1
⎪
⎪ a x̃ + aλ x̃ n 2 −1 x̃ n 2 + (α x̃)n 2 1 2 = 0
⎪
⎨ αb x̃ + λ (α x̃)n 2 −1 x̃ n 2 + (α x̃)n 2 n 1 /n 2 −1 = 0
⎪
b
(10)
⎪
⎪ βc + λ
(β x̃)n 1 −1 = 0
⎪
⎪
x̃
⎩ n2 c n /n
x̃ + (α x̃)n 2 1 2 + (β x̃)n 1 − 1 = 0.
123
Comp. Part. Mech.
For practical use it is necessary to be able to define a applying the transition formulas:
superquadric particle with respect to a global coordinate
system which is usually observer-fixed. This is done by ∇ X F(X) = A · ∇ x f (x),
applying the usual translation and rotation operations. The
H X (F)(X) = A · H x ( f )(x) · AT , (18)
shape function F of a superquadric particle with center X C
and quaternion q in a global frame is given by the following x = A · (X − X C ).
T
expression:
Now we are ready to formulate the contact detection prob-
F(X) = f ( A · (X − X C )), A = A(q).
T
(13) lem in terms of an optimization problem. Consider two
superquadric particles with two times continuously differ-
The points X and X C are defined in the global frame. For a entiable (automatically fulfilled if n 1 ≥ 2 and n 2 ≥ 2) shape
contact detection algorithm it is also necessary to be able to functions F1 (X) and F2 (X) defined in global frame. Follow-
calculate 1st (gradient) and 2nd (Hessian matrix) derivatives ing Houlsby [24] and extending the algorithm for 3D case we
of the shape function. The gradient of the shape function seek a “midway” point P between the particles and “clos-
calculated at a point x = (x, y, z)T in the local frame is the est” to both. In other words, solve the following optimization
following: problem:
123
Comp. Part. Mech.
δa i = (a0i − r i )/N
δbi = (b0i − r i )/N
δci = (c0i − r i )/N
δn i1 = (n i10 − 2)/N (25)
δn i2 = (n i20 − 2)/N
i = 1, 2
k = 1.
a i := r i + k · δa i
is the Jacobian matrix of the right-hand side term Φ: bi := r i + k · δbi
⎛ ⎞ ci := r i + k · δci
F1x
+ μ2 F2x
⎜ F ⎟ n i1 := 2 + k · δn i1 (26)
+ μ2 F2y
⎜ 1y ⎟
Φ= ⎜ ⎟. (23) n i2 := 2 + k · δn i2
⎝ F1z + μ F2z
2 ⎠
i := 1, 2
F1 − F2
k := k + 1.
For stability reasons it is necessary to find a scalar parameter (4) Calculate the contact point for particles with shape para-
α ∈ (0, 1] such that Z n+1 = Z n + αδ Z, ||Φ(Z n+1 )|| < meters (a 1 , b1 , c1 , n 11 , n 12 ) and (a 2 , b2 , c2 , n 21 , n 22 ) using
||Φ(Z n )|| at every iteration to ensure convergence of the the iterative algorithm described above and the last com-
algorithm. There are several methods to obtain such a scalar puted starting point. Use the found contact point as a new
parameter. One of them is first to check if α = 1 satisfies starting point for the next step.
||Φ(Z n+1 || < ||Φ(Z n )||. If not, let α := α/2 and repeat (5) Repeat steps 3 and 4 for N − 1 times
or use the Golden section algorithm [27] with termination
if any α satisfying ||Φ(Z n+1 )|| < ||Φ(Z n )|| is found. The Steps 1–5 of the iterative procedure listed above define the
solution from previous DEM time step can be used as a start- step-wise linear transition of the spherical shape parameters
ing point (initial guess). Usually a few Newton iterations to the shape parameters of the given particles 1 and 2 where
(1 − 3) are required to converge, depending on a user defined k is the iteration number. After k = N steps the shape and
tolerance ε 1 for termination criterion ||Φ(Z n+1 )|| blockiness parameters will be the same as initial ones by
< ε. construction of the procedure:
If there is no information on the contact point from the
previous step, let (a01 , b01 , c01 , n 110 , n 120 ) and (a02 , b02 , c02 , n 210 , a i := r i + N · δa i = r i + N · (a0i − r i )/N = a0i
n 220 ) be the shape and blockiness parameters of particle 1 and
2. The following approach is suggested: bi := r i + N · δbi = r i + N · (b0i − r i )/N = b0i
ci := r i + N · δci = r i + N · (c0i − r i )/N = c0i
(27)
(1) Find the contact point for a pair of volume equivalent n i1 := 2 + N · δn i1 = 2 + N · (n i10 − 2)/N = n i10
spheres with radii r1 and r2 and centers located at the
1 and X 2 . These n i2 := 2 + N · δn i2 = 2 + N · (n i20 − 2)/N = n i20
same points as for given particles, X C C
spheres defined as superquadrics have the following i = 1, 2.
shape and blockiness parameters: a 1 = b1 = c1 = r 1 ,
As a result, this procedure will give the contact point for the
a 2 = b2 = c2 = r 2 , n 11 = n 12 = n 21 = n 22 = 2. The
given pair of particles. This step-wise procedure ensures the
analytical solution for the sphere–sphere contact point is
convergence of the method and does not affect computational
time significantly, since it must be called only once per pair
X = (r2 X C
1
+ r1 X C
2
)/(r1 + r2 ). (24) of particles. We found that for the exponents n 1 ,n 2 ≤ 8 con-
vergence is guaranteed. In addition, Cleary and Sawley [11]
Use this point as a starting point. studied influence of the blockiness parameters on the mass
123
Comp. Part. Mech.
⎧
⎪ n 1 n −1 n n 1 /n 2 −1 = 0
flow rate from a hopper and found that the values n 1 , n 2 > 8 ⎪ n x + λ a x̃ 1 (1 + α 2 )
⎪
fail to make any difference to the nature of the flow. Thus, ⎪
⎨ n + λ n 1 α n 2 −1 x̃ n 1 −1 (1 + α n 2 )n 1 /n 2 −1 = 0
y b
we can safely recommend to use blockiness parameters in (31)
⎪ + λ n 1 n 1 −1 n 1 −1
⎪
⎪ n z c β x̃ =0
the range between 2 and 8. ⎪
⎩ x̃ n 1 (1 + α n 2 )n 1 /n 2 + β n 1 x̃ n 1 − 1 = 0.
123
Comp. Part. Mech.
real but intends to model the deformation of the interacting X 0 and the contact direction ni j define the contact line. We
particles at a contact point in an indirect way. seek points X i and X j as the nearest (with respect to X 0 )
Di Renzo and Di Maio [15] suggested using the linear intersection points between the contact line and the particles’
spring Hertzian model without cohesion for the particle– surfaces. In other words, we solve the following non-linear
particle and particle–wall contacts. This model employs the algebraic equations separately with respect to the scalars
following formula for the interparticle contact force, acting αi > 0 and α j < 0 at every DEM time step for each pair of
from a spherical particle i with radius Ri and center at point overlapping particles:
X Ci on a spherical particle j with radius R j and center at
point X C j : Fi (X i ) = 0, where X i = X 0 + αi ni j ,
(36)
F j (X j ) = 0, where X j = X 0 + α j ni j .
F n,i j + F t,i j , if δn,i j ≥ 0
Fi j = (35)
0, if δn,i j < 0, Then the normal overlap vector is δ n ≡ X i − X j = (αi −
α j )ni j .
where These equations with respect to αi and α j are easier to
solve if moved to their own local reference frames:
• F n,i j = kn,i j δ n,i j + γn,i j v n,i j is the normal force com-
ponent, f i (x i0 + αi n̂ii j ) = 0,
j
• F t,i j = kt,i j δ t,i j + γt,i j v t,i j is the tangential force com- f j (x j0 + α j n̂i j ) = 0,
ponent,
• δ n,i j = δn,i j ni j is the normal overlap vector, where x l0 = AlT · (X 0 − X Cl ), n̂li j = AlT · ni j , l = {i, j}.
• ni j = (X C j − X Ci )/||X C j − X Ci || is the normal overlap Note that in both reference frames scalars αi and α j are the
direction, same for each particle. The equations can be easily solved
• δn,i j = Ri + R j − di j is the normal overlap distance, by Newton’s iterations:
• di j = ||X C j − X Ci || is the distance between particles’
centers, f i (x i0 + αim n̂ii j )
αim+1 = αim − ,
• v n,i j = ((v j − v i ) · ni j )ni j is the normal component of ∇ f i (x i0 + αim n̂ii j ) · n̂ii j
the relative velocity, j (37)
• v t,i j = v j − v i − v n,i j is the tangential component of the f j (x j0 + α mj n̂i j )
α m+1
j = α mj − j j
.
relative velocity, ∇ f j (x j0 + α mj n̂i j ) · n̂i j
T
• δ t,i j = T0 v t,i j dτ is the tangential overlap [28].
Calculation of the coefficients kn , kt , γn , γt in(35) requires
Both normal and tangential forces contain a spring force knowledge of the equivalent radius Req [15]:
and a damping force. The coefficients kn , kt , γn , γt are cal-
culated from the material properties (density, coefficient of Ri R j
Req = , (38)
restitution, Poisson ratio, Young’s modulus, shear modulus), Ri + R j
overlap, and radii of the particles. The corresponding expres-
sions can be found in [4]. where Ri and R j are the radii of particles i and j. We will use
However, the models above are only applicable for spher- R = 1/K as the particle radius, where K = K mean = 21 (κ1 +
ical particles. Feng and Owen [20] proposed theoretical κ2 ) is the mean local curvature coefficient [21] calculated at
framework for developing energy-conserving normal contact X i and X j for each particle correspondingly, and κ1 and κ2
models for arbitrarily shaped particles. It has been estab- are the principal curvature coefficients:
lished that the normal force must be a potential field vector
associated with a potential function φ that is a function of the K mean = (∇ F T · H(F) · ∇ F
overlap volume [20]. However, the accurate calculation of the (39)
− |∇ F|2 (FX X + FY Y + FZ Z ))/2|∇ F|3 .
overlap volume may be computationally expensive and thus
become not applicable for a case with millions of particles. Alternatively, one can use the Gaussian curvature coefficient
Previously, Zheng et al. [54] modified the Hertzian model √
K = K Gauss = κ1 κ2 [21]:
taking into consideration two principal radii of curvature
and applied it to ellipsoidal particles. They showed good A = FZ (FX X FZ − 2FX FX Z ) + FX2 FZ Z ,
agreement with the results calculated by means of the finite
B = FZ (FY Y FZ − 2FY FY Z ) + FY2 FZ Z ,
element method (FEM). However, the extension of the model
to superquadric particles becomes quite complex. In this C = FZ (FX Y FZ − FX Z FY − FX FY Z ) + FX FY FZ Z ,
study we propose the following approach. The contact point D = FZ2 (FX2 + FY2 + FZ2 )2 ,
123
Comp. Part. Mech.
√
AB − C 2
K Gauss = . (40)
D
Obviously, K mean = K Gauss = 1/R for a spherical particle
of radius R.
One of the disadvantages of the proposed methodology
(and hence of the possible extension of the method by Zheng
et al. [54] to superquadrics) is that the curvature coeffi-
cients may become zero leading to infinite curvature radii if
superquadric exponents n 1 and n 2 are more than 2, especially
in face-to-face contact. The mean curvature radius becomes
infinite if both principal curvature coefficients are zero. This
occurs at 6 points on the particle surface: x = y = 0, z = ±c,
y = z = 0, x = ±a and x = z = 0, y = ±b (in the local
reference frame). The Gaussian curvature radius becomes
infinite if any of the principal curvature coefficients is zero.
This occurs if x = 0, y = 0 or z = 0 (in the local refer- Fig. 6 Normal force as a function of the overlap distance for different
ence frame). For this reason, we limit the curvature radius: models of the equivalent radius. Results are compared with the FEM
Rcur vatur e = min(Rcur vatur e , q Rvol ), where Rvol is radius simulation results by Zheng et al. [54]
of the volume equivalent sphere, q is the limiting coefficient
that must be chosen in advance. In the current implementation ered to be elastic and frictionless. The overlap distance was
of LIGGGHTS q = 10 is used. The influence of the choice varied in the range between 0 and 5μm. Three radius models
of the curvature radius and the limiting coefficient q on the were tested: the mean curvature radius, the Gaussian curva-
simulation results is to be studied in the future publications. ture radius, and radius of the volume equivalent sphere. The
normal contact force between overlapping ellipsoidal parti-
cles is plotted as a function of the overlap and presented in
3 Validation Fig. 6. Calculated results are compared with the FEM analy-
sis carried out by Zheng et al. [54] and added to Fig. 6.
3.1 Contact force between two ellipsoidal particles It can be observed that for elastic contacts the normal
force can be well described by the Hertz theory using the
Here we bring in contact two ellipsoidal particles (Table 1) Gaussian curvature radius for the particle radii Ri and R j in
oriented parallel to each other (Fig. 5). Particles are consid- Eq. (38) since it gives minimal discrepancy with respect to
the FEM results. Despite this, only the mean curvature radius
Table 1 Particle parameters for ellipsoids is used in the further test cases in this paper since it becomes
Parameter Value infinite only at 6 points of the non-ellipsoidal (superquadric
exponents n 1 , n 2 > 2) particle surface rather than on the
Three semi-axes, a, b, c (mm) 5, 2.5, 2.5 infinite number of points if the Gaussian curvature is used,
Young’s modulus (GPa) 10 as described in the previous section. A more comprehensive
Poisson ratio 0.3 comparison of different curvature radii using different par-
Density (kg/m3 ) 2500 ticle shapes at different orientations is to be discussed in a
future paper.
123
Comp. Part. Mech.
123
Comp. Part. Mech.
cylinder’s face and the line joining the contact point and the
center of the particle, θ is the angle between the cylinder’s
face and the wall, and I yy is the moment of inertia around the
y-axis. The parameter r is the distance between the cylinder’s
center and the corner point C (Fig. 9), which is assumed to
be fixed. Particle parameters are listed in Table 3.
The post-impact angular and translational velocities were
calculated for various orientation angles θ for the DEM sim-
ulations and compared with analytical expressions in Fig. 10.
The wall was removed immediately after collision to prevent
the secondary contact that occurs in reality at low and high
impact angles and is not taken into account in Eqs. (41) and
(42).
For angles 5◦ θ 86◦ the superquadric DEM gives
Fig. 9 Scheme of cylinder–wall impact, X Z projection relative good agreement with analytics; however, for other
angles there is small disagreement mainly due to the error
Table 3 Simulation parameters in particle–wall impact simulation in shape approximation, and, as a result, because of the
corner point C, that is non-static with respect to the particle-
Parameter Value
based coordinate system. This corner point is always static
Cylinder diameter, D (m) 8 × 10−3 at impact angles θ = 0◦ and θ = 90◦ for true cylinders.
Cylinder length, L (m) 5.3 × 10−3 According to [29], multisphere simulations (with 54 prime
Volume (m3 ) 2.49 × 10−7 spheres) show significant errors over most of the orientation
Density (kg/m3 ) 1245 range.
Mass (kg) 3.1 × 10−4
Moment of inertia I x x (kg/m2 ) 1.834 × 10−9 3.4 Piling of particles
Moment of inertia I yy (kg/m2 ) 1.834 × 10−9
Moment of inertia Izz (kg/m2 ) 2.362 × 10−9 For the second validation test case superquadric particles
Shear modulus (GPa) 1.15
with the following shape parameters were used: a = 2.0 mm,
b = 2.0 mm, c = 1 mm, n 1 = n 2 = 4. Particle parameters
Poisson’s ratio 0.35
along with simulation setup data are listed in Table 4. They
Coefficient of friction 0.0
were compared with volume equivalent spherical particles of
Coefficient of restitution 0.85
radius R = 1.836 mm. Domain boundaries are represented
Time step
t (s) 5 × 10−7
by rigid walls of the same material as the particles. In both
123
Comp. Part. Mech.
Table 4 Simulation parameters used in pilling test case particles fall and leave the computational domain, while other
Parameter Value particles remain on the plate and form a heap as a result.
For this test case a set of experiments was conducted and
Density (kg/m3 ) 957 compared with the simulations. The corresponding shape
Young’s modulus (Pa) 2.5 × 107 parameters for each sort of the particles were found and are
Poisson’s ratio 0.25 listed in Table 5. Corresponding superquadric approxima-
Coefficient of friction 0.5 tions of the particles in question are presented in Fig. 12.
Coefficient of rolling friction 0 Material properties were chosen the same for each particle
Coefficient of restitution 0.5 shape and are listed in Table 6.
Sizes of the simulation domain 0.31 × 0.03 ×0.24 m
Time step
t (s) 5 × 10−6
Table 5 Superquadric shape parameters for particles in the “Angle of
Number of time steps 600, 000
repose” test case
Particle a (mm) b (mm) c (mm) n1 n2
Fig. 11 Piling problem, simulation snapshots at the final time step, superquadrics (left) and volume equivalent spheres (right) without rolling
friction, 4680 particles
123
Comp. Part. Mech.
Fig. 13 Angle of repose, chewing gum, experiment (left), and simulation (right)
Fig. 14 Angle of repose, sugar cubes, experiment (left), and simulation (right)
From the pictures (Figs. 13, 14 and 15) it can be seen Table 7 Static packing of cylinders, experimental and simulation data
that there is only a qualitative agreement between the exper- Parameter Experiment Simulation
iments and the simulations mainly due to relatively big
size of the particles used in comparison to the cylinder Cylinder diameter, D (m) 8 × 10−3 8 × 10−3
diameter. Cylinder length, L (m) 5.3 × 10−3 5.3 × 10−3
Volume (m3 ) 2.664 × 10−7 2.49 × 10−7
Density (kg/m3 ) 1160 1245
4.2 Static packing of cylinders Mass (kg) 3.1 × 10−4 3.1 × 10−4
Shear modulus (GPa) 1.15 1.15
In this validation test we simulate a static packing of cylin- Poisson’s ratio 0.35 0.35
ders, defined as superquadric particles. The particles were Coefficient of friction 0.5 0.5
dropped into a cylindrical container, and compared to the Coefficient of restitution 0.85 0.85
experimental data provided by Kodam et al. [29]. The DEM Container diameter (mm) 50.6 50.6
material properties along with particle properties are listed Container height (mm) 130 130
in Table 7. Particle size parameters, a, b, and c, were chosen
Time step
t(s) − 10−5
such that particles in the simulation have the same diam-
eter and length as in the experiment: a = b = D/2,
123
Comp. Part. Mech.
Fig. 16 Packing of cylinders after dropping 250 particles, superquadric DEM (left), and experiment [29] (right)
Fig. 17 Comparison of experimental (Liu et al. [33], top) and simulation (bottom) snapshots of the discharging candies from a hopper at different
moments of time: t = 0 s, t = 0.9 s, t = 1.8 s, t = 2.7 s, t= 3.6 s, and t = 6.6 s (from left to right)
c = L/2 (Table 7). Volume and principal moments of iner- which is in good agreement with the experiment. Kodam et al.
tia of the cylinders defined as superquadrics have values [29] simulated packing of the cylinders with the multisphere
smaller than those for true cylinders due to the rounded edges. approach and found that 9-sphere particles underpredict the
The difference between a true cylinder and its superquadric bed height by 21 %, while 54-sphere particles underpredict
approximation decreases with the increase of the blocki- the fill height by 8 %.
ness/roundness superquadric shape parameter n 1 , however,
leading to less stability of the method. Hence, a compro- 4.3 Hopper discharge
mise value must be chosen. The superquadric cylinder in
this simulation was set to have the same mass as the true We also use the method in simulating the discharge of Smar-
cylinder by increasing density by 7 % and setting blockiness ties®chocolate candies from a flat bottom hopper. These
parameters to n 1 = 6.0, n 2 = 2.0. DEM time step was set candies have ellipsoidal shape with the following shape para-
t = 10−5 s. meters: 2a = 13.56 mm, 2b = 2c = 7.19 mm, n 1 = n 2 =
The image of the final state from the simulation is shown 2.0, and density 1338 kg/m3 . The hopper is 290 mm along
in Fig. 16. The final experimental fill height according to the X-direction, 55mm in Y-direction, and has an orifice
Kodam et al. [29] is 53.3±2.0 mm, while superquadric DEM of 54 mm in X-direction and 55mm in Y-direction. Grav-
simulation gives the fill height of roughly 52.0 ± 3.0 mm ity is oriented along the Z-direction. The particle–particle
123
Comp. Part. Mech.
123
Comp. Part. Mech.
123
Comp. Part. Mech.
19. Favier J, AbbaspourFard M, Kremmer M, Raji A, Abbaspour-Fard 35. Lu G, Third JR, Müller CR (2012) Critical assessment
M, Kremmer M, Raji A (1999) Shape representation of axisymmet- of two approaches for evaluating contacts between super-
rical, nonspherical particles in discrete element simulation using quadric shaped particles in DEM simulations. Chem Eng
multielement model particles. Eng Comput 16(4):467–480. doi:10. Sci 78:226–235. doi:10.1016/j.ces.2012.05.041, http://linkinghub.
1108/02644409910271894, http://www.emeraldinsight.com elsevier.com/retrieve/pii/S0009250912003223
20. Feng YT, Han K, Owen DRJ (2012) Energy-conserving con- 36. Marigo M, Stitt EH (2015) Discrete Element Method (DEM) for
tact interaction models for arbitrarily shaped discrete elements. Industrial applications: comments on calibration and validation
Comput Methods Appl Mech Eng 205–208(1):169–177. doi:10. for the modelling of cylindrical pellets. KONA Powder Part J
1016/j.cma.2011.02.010, http://linkinghub.elsevier.com/retrieve/ 32:236–252. doi:10.14356/kona.2015016, https://www.jstage.jst.
pii/S0045782511000454 go.jp/article/kona/32/0/32_2015016/_article
21. Goldman R (2005) Curvature formulas for implicit curves and sur- 37. Markauskas D, Kačianauskas R, Džiugys A, Navakas R (2009)
faces. Comput Aided Geomet Desig 33(7):632–658. doi:10.1016/ Investigation of adequacy of multi-sphere approximation of ellipti-
j.cagd.2005.06.005 cal particles for DEM simulations. Granular Matter 12(1):107–123.
22. Hamilton WR (1847) LXIX. On quaternions; or on a new system of doi:10.1007/s10035-009-0158-y, http://link.springer.com
imaginaries in algebra. Philos Mag Ser 3 30(203), 458–461. doi:10. 38. Miller TF, Eleftheriou M, Pattnaik P, Ndirango A, Newns D, Mar-
1080/14786444708645426, http://www.tandfonline.com tyna GJ (2002) Symplectic quaternion scheme for biophysical
23. Höhner D, Wirtz S, Scherer V (2014) A study on the influence molecular dynamics. J Chem Phys 116(20):8649. doi:10.1063/1.
of particle shape and shape approximation on particle mechan- 1473654, http://scitation.aip.org/content/aip/journal/jcp/116/20/
ics in a rotating drum using the discrete element method. Powder 10.1063/1.1473654
Technol 253:256–265. doi:10.1016/j.powtec.2013.11.023, http:// 39. Munjiza, A, Andrews K (1998) NBS contact detection algorithm
linkinghub.elsevier.com/retrieve/pii/S0032591013007067 for bodies of similar size. Int J Numer 149(2), 131–149.
24. Houlsby G (2009) Potential particles: a method for modelling doi:10.1002/(SICI)1097-0207(19980915)43:1<131::AID-
non-circular particles in DEM. Comput Geotech 36(6):953–959. NME447>3.0.CO;2-S. ftp://crack.seismo.unr.edu/downloads/
doi:10.1016/j.compgeo.2009.03.001, http://linkinghub.elsevier. russell/partitioning_space/munjiza_1998_nbs.pdf
com/retrieve/pii/S0266352X09000469 40. Munjiza A, Rougier E, John NWM (2006) MR linear contact detec-
25. Jaklič, A, Leonardis, A, Solina F (2000) Segmentation and recovery tion algorithm. Int J Numer Methods Eng 66(1):46–71. doi:10.
of superquadrics, computational imaging and vision, 20. Springer, 1002/nme.1538
Dordrecht. doi:10.1007/978-94-015-9456-1, http://link.springer. 41. Nezami, E, Hashash Y (2004) A fast contact detection algorithm
com for 3-D discrete element method. Comput Geotech 31(7): 575–587.
26. Jaklic, A, Solina F (2003) Moments of superellipsoids and their doi:10.1016/j.compgeo.2004.08.002, http://linkinghub.elsevier.
application to range image registration. IEEE Trans Syst Man com/retrieve/pii/S0266352X04001016, http://www.sciencedirect.
Cybernet Part B Cybernetics 33(4), 648–57. doi:10.1109/TSMCB. com/science/article/pii/S0266352X04001016
2003.814299, http://www.ncbi.nlm.nih.gov/pubmed/18238214 42. Omelyan IP (1998) Algorithm for numerical integration of the
27. Kiefer J (1953) Sequential minimax search for a maximum. rigid-body equations of motion. Phys Rev E 58(1):1169–1172.
Proc Am Math Soc 4(3):502–502. doi:10.1090/S0002-9939-1953- doi:10.1103/PhysRevE.58.1169, http://arxiv.org/abs/physics/
0055639-3, http://www.ams.org/jourcgi/jour-getitem?pii=S0002- 9901027, http://link.aps.org
9939-1953-0055639-3 43. Omelyan IP (1998) Numerical integration of the equa-
28. Kloss C, Goniva C, Hager A, Amberger S, Pirker S (2012) Models, tions of motion for rigid polyatomics: the matrix method.
algorithms and validation for opensource DEM and CFD-DEM. Comput Phys Commun 109(2–3):171–183. doi:10.1016/S0010-
Progress Comput Fluid Dyn Int J 12(2/3):140. doi:10.1504/PCFD. 4655(98)00024-1, http://arxiv.org/abs/physics/9901026, http://
2012.047457, http://www.inderscience.com/link.php?id=47457 linkinghub.elsevier.com/retrieve/pii/S0010465598000241
29. Kodam M, Bharadwaj R, Curtis J, Hancock B, Wassgren C 44. Omelyan IP (1998) On the numerical integration of motion for
(2010) Cylindrical object contact detection for use in discrete ele- rigid polyatomics: the modified quaternion approach. Computn Phy
ment method simulations. Part II-Experimental validation. Chem 12(1):97–103. doi:10.1063/1.168642
Eng Sci 65(22):5863–5871. doi:10.1016/j.ces.2010.08.007, http:// 45. Omelyan IP (1999) A new leapfrog integrator of rotational
linkinghub.elsevier.com/retrieve/pii/S0009250910004744 motion. The Revised angular-momentum approach. Mol Simul
30. Kremmer M, Favier JF (2001) A method for representing bound- 22(3):213–236. doi:10.1080/08927029908022097, http://arxiv.
aries in discrete element modelling-part I: Geometry and contact org/abs/physics/9901025, http://www.tandfonline.com
detection. Int J Numer Methods Eng 51(12):1407–1421. doi:10. 46. Plimpton S (1995) Fast parallel algorithms for short-range
1002/nme.184, http://doi.wiley.com/10.1002/nme.184 molecular dynamics. J Comput Phy 117(1):1–19. doi:10.
31. Kruggel-Emden H, Rickelt S, Wirtz S, Scherer V (2008) A study 1006/jcph.1995.1039, http://linkinghub.elsevier.com/retrieve/pii/
on the validity of the multi-sphere Discrete Element Method. Pow- S002199918571039X
der Technol 188(2):153–165. doi:10.1016/j.powtec.2008.04.037, 47. Portal R, Dias J, de Sousa L (2010) Contact detection between
http://linkinghub.elsevier.com/retrieve/pii/S0032591008002143 convex superquadric surfaces. Arch Mech Eng LVI I(2):165–186.
32. Langston PA, Al-Awamle MA, Fraige FY, Asmar BN (2004) Dis- doi:10.2478/v10180-010-0009-8, http://www.degruyter.com/
tinct element modelling of non-spherical frictionless particle flow. view/j/meceng.2010.lvii.issue-2/v10180-010-0009-8/v10180-
Chem Eng Sci 59(2):425–435. doi:10.1016/j.ces.2003.10.008 010-0009-8.xml
33. Liu SD, Zhou ZY, Zou PR, Pinson D, Yu AB (2014) Flow charac- 48. Rougier, E, Munjiza, A, Latham J (2004) Shape selec-
teristics and discharge rate of ellipsoidal particles in a flat bottom tion menu for grand scale discontinua systems. Eng Com-
hopper. Powder Technol 253:70–79. doi:10.1016/j.powtec.2013. put 21(2/3/4), 343–359. doi:10.1108/02644400410519820, http://
11.001 www.emeraldinsight.com
34. Lu G, Third J, Müller C (2015) Discrete element models for 49. Verlet L (1967) Computer “Experiments” on classical fluids. I.
non-spherical particle systems: From theoretical developments Thermodynamical properties of Lennard-Jones molecules. Phys
to applications. Chem Eng Sci 127:425–465. doi:10.1016/j. Rev 159(1):98–103. doi:10.1103/PhysRev.159.98
ces.2014.11.050, http://linkinghub.elsevier.com/retrieve/pii/ 50. Vu-Quoc L, Zhang X, Walton OR (2000) A 3-D discrete-element
S0009250914007040 method for dry granular flows of ellipsoidal particles. Comput
123
Comp. Part. Mech.
Methods Appl Mech Eng 187(99):483–528. doi:10.1016/S0045- 54. Zheng QJ, Zhou ZY, Yu AB (2013) Contact forces between vis-
7825(99)00337-0 coelastic ellipsoidal particles. Powder Technol 248:25–33. doi:10.
51. Walton, OR, Braun RL (1993) Simulation of rotary-drum and 1016/j.powtec.2013.03.020
repose tests for frictional spheres and rigid sphere clusters. http:// 55. Zhu H, Zhou Z, Yang R, Yu AB (2007) Discrete particle sim-
www.grainflow.com/index_files/Rotary_Drum_Simulation_ ulation of particulate systems: theoretical developments. Chem
DOE-NSF-1993.pdf Eng Sci 62(13):3378–3396. doi:10.1016/j.ces.2006.12.089, http://
52. Williams JR, Pentland AP (1992) Superquadrics and modal dynam- linkinghub.elsevier.com/retrieve/pii/S000925090700262X
ics for discrete elements in interactive design. Eng Comput 56. Zhu H, Zhou Z, Yang R, Yu AB (2008) Discrete particle simulation
9(2):115–127. doi:10.1108/eb023852, http://www.emeraldinsight. of particulate systems: a review of major applications and findings.
com Chem Eng Sci 63(23):5728–5770. doi:10.1016/j.ces.2008.08.006,
53. Xu WX, Chen HS, Lv Z (2011) An overlapping detection algo- http://linkinghub.elsevier.com/retrieve/pii/S0009250908004168
rithm for random sequential packing of elliptical particles. Phys A
390(13):2452–2467. doi:10.1016/j.physa.2011.02.048
123