Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Noname manuscript No.

(will be inserted by the editor)

A combined SPH-DEM approach for extremely


deformed granular packings: Validation and
compression tests

F.J. Castro · S. Radl

Received: date / Accepted: date

Abstract We present a novel SPH-DEM approach for the simulation of de-


formable granular media. First, we show how the method converges to the
analytical solution in a simple contact mechanics problem, namely the Hertz
contact law for spherical grains. Second, we analyze the evolution of a 2D
packing of discs under uniaxial compression, displaying the evolution of key
metrics of the packing such as the coordination number and the vertical stress.
We show that the code produces data in quantitative agreement with what is
known from literature. Finally we demonstrate that our SPH-DEM coupling
can be used to study packings of deformable grains from the onset of jamming
to extremely compacted states, reaching packing fractions of ϕ ≃ 0.995.
Keywords SPH · DEM · Soft Deformable Grains · Compression
Mathematics Subject Classification (2020) 74S99 · 74A70 · 74E20

1 Introduction

Smoothed particle hydrodynamics (SPH) was introduced in 1977 in the field


of astrophysics [26, 14], and since then, the maturity, the fields where it is
used, and the applications of the method have increased significantly. Among
the number of applications the method was used for, one can mention the
following: water wave impact on coastal structures [1], flow around ships [28],
sloshing of liquids in carrier tanks [39], flooding [41], tsunami generation [47],
F.J. Castro
Institute of Process and Particle Engineering, Graz University of Technology,
Inffeldgasse 13/III, 8010 Graz, Austria
Tel.: +43 316 873 30407
E-mail: francisco.goiocastro@tugraz.at
S. Radl
Institute of Process and Particle Engineering, Graz University of Technology,
Inffeldgasse 13/III, 8010 Graz, Austria
2 F.J. Castro, S. Radl

flow in porous media [18, 34], fluid-solid elastic interaction [21, 48], or bird
strikes on aircraft components [15].
To extend the fields of application of SPH, many studies have coupled
it with other methods, addressing in this way multiphysics problems [10].
Discrete Element Method (DEM) is not an exception in this context, and in
literature, SPH-DEM couplings were presented. These algorithms were used
to study fluid-solid interaction [19, 5] or impact of two solid objects — recently
giving high importance to the contact models used for solid-solid interaction
[44, 43]— in a regime of high deformations.
Being able to simulate highly deformed solids is of paramount importance
in many fields, an extremely wide and interdisciplinary example is the study of
granular matter. The way the deformability of the constituents of a granular
packing can drastically change the macroscopic properties of the system has
been observed in several contexts. An important example is the discharge of
a silo, where the flow field and clogging statistics are dramatically different
for hard and soft grains [2, 45]. Also, often times softness is an indispensable
quality of particles. We can think to biology, where forces involved in cell-cell
interaction, i.e. force chains, have implications in the transmission of infor-
mation in cells [36]. Deformable grains are of main interest also in the food
industry, where the change of the properties can have direct consequences on
the quality of the product [42]. As a final example, from a more fundamen-
tal point of view, recent experiments on photoelastic discs have shown that
asperities play a relevant role in the transition from variable to persistent
force chains in a granular packing [23], opening question on the influences of
irregularities in the jamming transition.
In the last years, some mesh-free numerical approaches have been proposed
to explore the phenomenology of deformable granular packings [29, 31, 4, 30] as
alternatives to the mesh-based methods already used by the community [17,
35].
In spite of its wide range of applications, to the best of our knowledge, the
SPH paradigm has never been methodically applied for a quantitative study of
a granular packing composed by deformable grains. In this paper, we present an
SPH-DEM coupling aiming at taking the best from both methods: the ability
to simulate large deformations from SPH, and the specialized contact models
for granular systems developed by the DEM community. Our goal is twofold:
first overcoming one of the main limitations of DEM models, undeformable
particles, and second, extending the use of SPH to particle resolved simulations
in the field of granular materials.
Here, we demonstrate how our coupling of SPH-DEM provides a viable
tool to address the study of packings of deformable grains. This numerical
approach enables to extract detailed information about the number of contacts,
overlaps, grain shape, and contact forces. On top of this, the data produced
can be used to study in detail mesoscale systems but also to reach macroscale
systems. Moreover, this method can be used to calibrate existing DEM models,
that participated in taking into consideration particle deformation [13], thus
allowing to transfer our results on an industrially relevant scale. Furthermore,
Title Suppressed Due to Excessive Length 3

the great flexibility in the possible geometries allows to study the effects of
the shape on different levels, from aspect ratio, to asymmetric particles, to the
presence of irregularities on the surface of interest.

2 Methods

2.1 SPH-DEM

We use an in-house version of the software LIGGGHTS [22], where we coupled


an SPH package [22] to work with the variety of contact models available in
LIGGGHTS.
The SPH implementation we are using, was originally developed for the
simulation of continuum mechanics problems, with the idea to reproduce large
deformation events [25, 12]. The code considers a Total Lagrangian formalism,
with a viscous force for stabilization, and a correction force for the suppression
of zero energy modes, further details can be found in [12]. The DEM code [22]
was specifically developed for granular materials, and several contact models,
that can account from repulsion, to friction, to cohesion, are available. The
coupling is therefore a bridge between continuum mechanics and granular
matter, with the idea to study packings of deformable grains.

2.1.1 SPH interaction

The SPH code in our simulations is used for the computation of the physics
at the single grain scale. This means that only the discretization points be-
longing to the same grain interact within the SPH framework. The role of
this part of the code is the calculation of the strain tensor, and using a given
material model — in this work restricted to linear elasticity — computes its
corresponding stress tensor. The stress tensor will define the force applied on
each discretization point, which is then used for time integration. Technical
specifications of the SPH code are reported in appendix B.1. How two different
grains in contact interact and deform each other is mediated by the DEM part
of the code.

2.1.2 DEM interaction

The inter grain interaction is mediated by the DEM code. When two grains
come into contact, the points used to discretize their boundaries come in con-
tact too. These interact via one of the contact models available in LIGGGHTS.
This means that the discretization points have associated with them also a
contact radius rc . For this reason, we therefore place the surface discretization
points at a distance rc from the surface we want to discretize.
We point out that the DEM interactions are active only between discretiza-
tion points belonging to different grains. This means that points belonging to
4 F.J. Castro, S. Radl

the same grain can be much closer than rc , without this causing any internal
stress because of repulsive forces.
In the following, we only use repulsive forces, in this case, the specific
contact law should not be critical since, for a clean simulation of grains in
contact, we want as less interpenetration as possible. We underline that in
this scheme the interpenetration between the boundaries of two grains is a
necessary ingredient for the interaction to happen. In order to avoid the effects
of an unphysical volumetric interpenetration of the grains, the stiffness of the
contact model — that we express here in terms of the associated Young’s
modulus — Yc needs to be higher than the stiffness defined in the material
properties of the grain. To be more precise, we consider two discretization
points with identical properties rc , Yc , and Poisson’s ratio νc , in contact. If
√ 3
they interact via a normal force of the type Fij = k r2c δij2 , where k = 3(1−ν
4Yc
2 ,
c)
we found that the simulations produce good results, is to say small grain
interpenetration, already for Yc = 5E, where E is the Young’s modulus of the
material of the grain. More details on the DEM interaction are described in
the appendix B.2.

3 3D uniaxial compression

3.1 Simulation setup

We consider here a well-known system to benchmark the simulation results


against an analytical solution. Henceforth the problem we tackle is the contact
force between linear elastic, frictionless spheres. Which can be analytically
solved for small overlaps [24], finding a relation between the overlap δ and the
contact force F , for two spheres of radius R1 and R2 :

1 R1 R2 3
F = δ2 .
D R1 + R2
Given that r is the distance center-to-center of the two spheres, the overlap
( 1−ν 2 1−ν 2 )
is δ = R1 + R2 − r. The quantity D = 43 E1 1 + E2 2 , where ν1 , ν2 are the
Poisson’s ratios and E1 , E2 the Young’s moduli of the two bodies. In our
simulations we always √
used identical particles, so the above expression of the
2R 3
force becomes F = 3(1−ν 2 ) Eδ . Using the undeformed radius of the particle R
2

as unit of length, the non dimensional force becomes F ∗ = 3(1−ν 2 3
2 ) δ . We will
2

present our analysis in terms of this force, hence the only relevant parameter
is the Poisson’s ratio, that we chose to be ν = 0.4.
To produce the data to compare, the implemented numerical setup con-
sists of three aligned purely elastic spheres discretized as in figure 1. The two
external spheres are slowly pushed against the one in the middle, then the
center-to-center distance and the contact force are measured.
To push the spheres we define, on the external side of the two lateral
spheres, a spherical cap. The thickness of the cap is 3.5rc , so the size and the
Title Suppressed Due to Excessive Length 5

number of particles involved is varying with the resolution of the simulation.


The thickness decreases with an increasing number of particles used in the
discretization. Similarly does the fraction of particles included in the cap,
going from Ncaps /Ntot ≃ 0.07 for the coarser simulation with Nsphere = 107,
to Ncaps /Ntot ≃ 0.006 for the case with Nsphere = 3731. The spherical caps
will move as a rigid body, at constant velocity, pushing the two spheres against
the central one. The movement is parallel to the line connecting the centers of
the three spheres so that no shear components are introduced. The magnitude
of the compression velocity is orders of magnitude slower than the speed of
sound in the material, and the compression can be considered quasi-static.
Since the simulation can be considered in equilibrium at every moment, the
force applied to the cap is equal to the contact force between the particles. For
this reason, we measure the force acting on the solid caps. At the center of each
sphere, there is an SPH particle which is considered the center of the deformed
sphere. These particles are used to compute the distance center-to-center and
the overlap between the grains.

3.2 Convergence

The arrangement of discretization points can have effects on the computation


of physical observables such as stress and strain [9, 7]. Some of these problems
are caused by an uneven arrangement of the discretization points, some others
by the intrinsic instability of the lattice used to place these points. We decided
to use an arrangement that can accurately reproduce the shape of a sphere.
All the external points are placed on the surface of a sphere of radius R − rc ,
this reduces the shape irregularities that will be observed if, instead, we were
choosing a lattice arrangement. In the last case, the points would be placed
on the surface of a sphere only when the lattice sites coincide with it, an
eventuality that can be met only for a small number of lattice points, and
along a few directions.
The internal volume of the sphere is discretized by a superposition of con-
centric spherical shells. More details on discretization can be found in the
appendix A.
We performed simulations for different levels of discretization to study the
convergence of the simulation to the solution. In figure 2 we report the results
obtained for the force-overlap relation using Nsphere = 107, 195, 3731 dis-
cretization points per sphere. The qualitative behavior is always captured, also
for the coarser discretization, although eventually showing a significant quan-
titative disagreement with the data obtained from finer simulations. Starting
from Nsphere = 195 the solution is satisfactory for high levels of compres-
sion. For δ/R > 0.15, the measured relative force difference with respect to
the full resolution simulation is ∆F/F < 0.1 at most. For small overlaps the
differences are larger, this can be expected since: i) the details of the surface
discretization play a dominant role at the onset of the contact; ii) the SPH
approximation is less accurate on the surface where discretization points have
6 F.J. Castro, S. Radl

(a) (b)

Fig. 1: Discretization technique. a) To simulate a disc, discretization points


are placed following concentric rings. The blue area is representing the disc to
be discretized. The contact radius, here not represented, of the most external
points touches exactly the border of the original disc. b) A sphere is composed
of a sequence of discs, of different radii, along the direction perpendicular to
the ring plane. For major visual clarity, the radius of the discretization points
is smaller than their contact radius.

fewer neighbors. For the simulation with Nsphere = 3731, we see that the data
is in good agreement with the analytical solution until δ/R ≃ 0.1, which is
satisfactory since we do not expect the Hertz law to hold for large overlaps.

Fitting a function of the form y = A′ δ γ to the simulation data, where A′
and γ ′ are fitting parameters, we want to verify that the Hertz contact law
is recovered with increasing accuracy by simulations with finer discretization.
The fit is done excluding the values of δ > 0.15, and the results are plotted in
fig 3.
We report the results for one additional simulation with Nsphere = 1028,
previously not reported for the sake of visual clarity. We observe that the
relative deviation from the analytical solution is converging to 0 with the
− 12
number of discretization points as ∼ Nsphere . The error in the evaluation of the
exponent in the contact force law, for the finer discretization, is ∆γ/γHertz ≃
0.06. Reaching a good agreement also according to this metric.

4 2D compression

4.1 Simulation setup

The system studied in this section is composed by Nmol = 300 2D discs,


enclosed by walls. The grains radii were randomly extracted in the interval
[1, 2] mm, following a uniform distribution. The simulations were run fore
nearly incompressible grains with Poisson’s ratio ν = 0.45. The packing is adi-
abatically compressed, at constant velocity, by a moving piston. The evolution
of quantities such as coordination number, overlaps, and stress is studied.
Title Suppressed Due to Excessive Length 7

Fig. 2: Force against overlap, data from different discretization, N is the num-
ber of discretization points per sphere in the simulation.

Fig. 3: Fitted exponent on data from simulations with different resolution. N


is the number of discretization points per sphere.

To generate the initial configuration we first start from a purely DEM


simulation performed in LAMMPS [40]. Inspired by the procedure described
by Herrmann [16] to produce a granular packing, a mixture of grains is poured
from a height between 150 mm and 200 mm in a box of length 62.7 mm. The
energy is dissipated via inelastic collisions, and the grains settle under the
effect of the gravitational acceleration g0 . Once the configuration is relaxed,
we drastically reduce gravity to g1 = 0.005g0 , keeping the stiffness of the
particles fixed, to reach a state where the overlaps are as small as possible.
Since the considered interaction potential is linear elastic, we expect the
overlap to linearly decrease with the gravitational acceleration. We can esti-
mate the overlap considering the balance between the elastic force and the
force caused by the pressure of the packing. Considering the equivalent hydro-
8 F.J. Castro, S. Radl

static pressure at the bottom of the packing PH = ρg0 h, hence a typical value
of the respective force FH ≃ πR2 PH , and equating with the elastic normal
force Fel = kn δ the expected relative overlap is δ/R ≃ πRρgh/kn . Following
this approximation, we can estimate the overlap to change from δ/R ≃ 10−4
with the initial gravitational acceleration to δ/R ≃ 10−6 when the gravity is
reduced. The packing we are obtaining has then minimal overlaps between the
different grains, this is important because in the next step we restart the simu-
lation from the configuration thus obtained, considering undeformed particles.
Every DEM particle is replaced by a disc, discretized by SPH particles,
which center and radius are identical to the starting DEM particle.
The discretization is done following the strategy mentioned in appendix
A.1, the average density of discretization points is 25.7 mm−2 , correspond-
ing to a discretization length a = 0.2 mm. This implies that the number of
(min)
discretization points per grain varies, from a minimum of Ndisc = 81 to a
(max) (tot)
maximum of Ndisc = 321, using a total of Nat = 52172 discretization
points for the grains. In this simulation, also the walls are discretized using
point particles, reaching a total of N tot = 54288 points.
The initial packing is contained in a box of length 62.7 mm and height
larger than the maximum height of the particles. To obtain a more compact
configuration before sampling, we apply some cycles of compression to reach a
more compact configuration, reaching the maximum packing fraction ϕmax ≃
0.995. Considering the speed of sound in the medium to be cs , the following
protocol is followed:
(i) first compression to reach a compact configuration at packing fraction
ϕ ≃ 0.80 with velocity v = 5 × 10−4 cs ;
(ii) the piston is kept fixed and the system is relaxed. During this relax-
ation, the system dissipates energy through collisions. The packing fraction,
although high, has not yet reached a value where the contact network prevents
any movements of the particles;
(iii) then compression at v = 10−4 cs to ϕmax and again the piston is fixed
in this position to let the system relax. In this state, particles are not free to
move anymore, and they are highly deformed;
(iv) the system is decompressed until reaching again the state at ϕ ≃ 0.81;
(v) repetition of the compression with v = 10−4 cs to ϕmax and again
relaxation, keeping the strain constant;
(vi) again decompression to reach ϕ ≃ 0.81 but now with v = 1.4 × 10−5 cs ,
followed by a relaxation;
(vii) final compression during which we collect the data, the piston is moved
with v = 5 × 10−5 cs until reaching ϕmax .

4.2 Results

During the compaction, we studied the evolution of the coordination num-


ber and the stress. To compute the coordination number Z, we exclude all
the discs in contact with the wall, which naturally have a lower number of
Title Suppressed Due to Excessive Length 9

contacts. Above jamming, we expect to observe an increase of the contact


1
number of the type Z − Z0 ∝ (ϕ − ϕ0 ) 2 , where Z0 is the coordination number
at jamming. The scaling law of the excess contact number has not only been
derived on theoretical grounds [46, 11] but has also been consistently observed,
numerically and in experiments across a variety of systems [27, 32, 33, 8]. As
a phenomenon, is a solid and universal signature characterizing a granular
system close to jamming.
In figure 5 we show the evolution of the excess contact number. Since Z0
varies for several reasons, such as friction [38] and preparation of the packing
[20], we do not have a priori knowledge of what is the exact value of Z0 for
our system. We expect it to be 3 < Z0 < 4, since for two dimensional discs
the coordination number varies smoothly between these two values depending
on the friction coefficient [37]. We estimate Z0 and its corresponding packing
fraction ϕ0 from fitting on the data Z − Z0′ = b′ (ϕ − ϕ′0 ) 2 , being b′ , Z0′ and ϕ′0
1

fit parameters. We obtain Z0 = 3.3, ϕ0 = 0.84, in line with values expected


from the theory. Furthermore we obtain b = 5.3, in close agreement with was
observed in [6] for a similar system (b = 5.1). Keeping Z0 , ϕ0 to the values

found above, we then perform a second fit Z − 3.3 = b′ (ϕ − 0.84)γ , with γ ′
and b′ fit parameters. We find γ = 0.501 and b = 5.4, in perfect agreement
with the theory.
In figure 6 we show the measured vertical stress against the packing frac-
tion for two systems. The initial spatial configuration is the same, but in one

case the particles as two times stiffer. σy,0 , a small but finite value, is the
measured vertical stress for the configuration at ϕ0 . As it is expected, the
data collapse on the same curve. At the grain scale, the systems are follow-
ing different evolutions. It is possible to observe non-monotonic behavior in
each dataset. These are caused by local rearrangements of the particles, shear
band formation, which release internal stress from the grains. We notice also a
small systematic difference comparing the two datasets, this is caused by the

uncertainty in the determination of ϕ0 , and hence σy,0 .
Finally in figure 7 we focus on the very beginning of the compression,
stopping at strains (ϕ/ϕ0 − 1) < 0.02. Considering a system of soft grains,
interacting through repulsive normal forces of the type f ∝ δ α , we know
that the pressure, above a critical packing fraction ϕc , scales with a power law
σ ∝ (ϕ−ϕc )α [32]. Where ϕc indicates the onset of jamming. For small strains,
the grains composing the system behave as linear springs. In figure 7 we show
that the measured vertical stress linearly scales with the distance from ϕ0 , the
best estimation we have for ϕc . From a linear fit y = α′ x + c on the data in the
figure, we find that the slope of the line is α = 1.07. Meaning that the code is
correctly reproducing an affine response of the packing to deformations.

4.3 Computational Time

Being both the DEM and the SPH codes originally developed in LAMMPS, as
for most of the packages the computational time scales as O(N/P ) [40] where
10 F.J. Castro, S. Radl

(a) Configuration for packing fraction (b) Compressed state, packing fraction
ϕ ≃ 0.81. ϕ ≃ 0.99.

Fig. 4: Snapshot of configuration around jamming.

N is the total number of particles and P is the processor count. Varying


some parameters can radically change the computational time. One of the
most important is the SPH kernel length h. This parameter is crucial for the
number of neighbors each particle is interacting with. This number will scale as
(h/a)D where a and h are the discretization pace and the kernel length, while
D is the dimension of the system. This was confirmed by direct inspection of
the CPU time in a 2D simulation, varying only h. We found that the CPU
time is scaling approximately as (h/a)1.9 , indicating that the DEM part of the
code has a minor impact on the total computational time. To give an idea of
the computational time needed by the method, on a regular workstation Xeon
E3-1270 V2 3.50GHz 4 Cores, a 2D simulation of compaction of one grain,
running for P = 4, with 2785 discretization points, for 40000 timesteps takes
about 600 s.
This time is to be considered as a starting point, since in the coupling our
main effort was not dedicated to the optimization of the code. In particular,
great improvement can be done in the optimization of the neighbor lists.

5 Conclusions

As a first result, we presented, in a 3D simulation, how the code can, given


the geometry, the material model, and the contact model, reproduce an ana-
lytically known solution for a contact mechanics problem. The Hertzian law,
relating the force to the overlap of two linear elastic, frictionless spheres, was
recovered with increasing precision enhancing the level of discretization. We
found also that the error on the estimate of the exponent of the power law is
monotonically decreasing with the number of discretization points as N − 2 .
1

We also tackled a more complex system, in a 2D setup, studying the be-


havior of 300 discs under compression, and comparing the results to what
is known from the literature. The code successfully reproduced the complex
phenomenology of a granular packing beyond jamming. We observed, as ex-
pected, a linear scaling of the pressure for small strains[32]. Furthermore,
Title Suppressed Due to Excessive Length 11

Fig. 5: Evolution of coordination number during the compression. The dashed


1
line is a fit where the scaling (ϕ − ϕ0 ) 2 is imposed. Via fit we obtain Z0 = 3.3,
ϕ0 = 0.84, and b1 = 5.3. The dotted line is the result from a fit where the
exponent is a fit parameter while we fixed ϕ0 = 0.84 and Z0 = 3.3. We obtain
γ = 0.501, and b2 = 5.4.

Fig. 6: Evolution of the vertical stress σy∗ = σy /E with the packing fraction
during compression. The results of simulation with different Young’s mod-
uli are plotted. The data collapse on the same curve. The small systematic

difference is caused by the uncertainty in the determination of ϕ0 and σy,0 .

Fig. 7: Detail from the initial part of the compression, σy∗ = σy /E. For very
small strains we observe a linear response of the system, as expected if the
particles were interacting via harmonic springs. We obtained the dashed line
from a linear fit on the log-log scale.
12 F.J. Castro, S. Radl

the evolution of the coordination number correctly reproduced the typical


1
Z − Z0 = b(ϕ − ϕ0 ) 2 behavior [27, 33]. Using this law, we were able to ap-
proximately evaluate the coordination number and the packing fraction at the
onset of jamming, obtaining results fully compatible with the literature[27, 32,
8, 3].
The SPH-DEM approach here used shows itself to be a promising tool for
the study of compressed granular materials. Among the many problems we
imagine it could be used for, there is the understanding of how macroscopic
properties of granular materials are influenced by the underlying deformabil-
ity of their grains. We are indeed able to produce in detail information on
the evolution of the number of contacts and their orientation, the stress dis-
tribution in the packing, and the shape of the grains. Most important, this
is even possible at extreme deformation that were computationally inaccessi-
ble (or only with extreme computational effort), or shadowed by the inability
of classical DEM to predict contact forces at extremely large grain deforma-
tion. Furthermore, although in this work we presented only grains of regular
shape, the method can be used for arbitrary complex geometries, extending
its applications to the investigation of the role of anisotropies in packings of
deformable particles. Also, the extension of our current study to situations in-
volving cohesion is straight forward: this is due to the implementation into the
LIGGGHTS-framework that allows us to access a variety of cohesion models.
We believe that foams, emulsions, hydrogel packings, and cells, to mention
a few, are good candidates as object of study with this method. The applica-
tions could range from the engineering of batteries in automotive industry to
the optimization of silo flow, to the food and pharma industry.

Acknowledgements FC thanks Nazanin Ghods for the numerous motivating discussions.


We acknowledge support from the European Union’s Horizon 2020 Marie Skłodowska-Curie
grant ‘CALIPER’ (No. 812638), and from NAWI Graz (via access to its HPC computing
resource dcluster.tugraz.at).

Conflict of interest

On behalf of all authors, the corresponding author states that there is no


conflict of interest.

A Discretization algorithm

A.1 Discretization of a disc

We start by describing the discretization of a circumference, the extension to a circle is


straightforward. The algorithm and the idea of the discretization are simple. In this study,
we are interested in reproducing as accurately as possible the shape of the particles. For
this reason, we choose an algorithm that respects as much as possible the geometry of the
system we want to discretize. It is important to keep in mind that the external discretization
points are interacting through a DEM potential, meaning that the points have an associated
Title Suppressed Due to Excessive Length 13

contact radius rc . Hence, if we want to discretize a circumference of radius R′ , we place


the discretization points at distance R = R′ − rc . In the following, to keep the notation
lighter, we will use R as the relevant radius. After the radius of the circumference R and
the discretization pace a are defined, the code will determine the regular polygon, inscribed
in the circumference of radius R − rc , which side length a′ approaches a from above. The
discretization points are then placed on the vertices of the polygon, hence at distance exactly
R from the center of the ring. Obtaining as a discretization for a circumference the regular
polygon inscribed in it.
A circle is then discretized as a series of concentric regular polygons. The vertices are
placed at distance R from the center for the first one, R − a for the second, R − 2a for the
next one, and so on until the center of the circle is reached, where a particle is placed.
Eventually, the symmetry group Dn of the polygon can be enforced, as it is the case
in the simulations presented. For the compression of the discs, the symmetry of the rings is
imposed to be D8n so that the points in the external circumference vary between 32 and 64,
following multiples of eight, according to the size of the disc. In figure 1 is shown the case of a
disc, discretized with symmetry D8 , where the approximation of the external circumference
is made with a 64-sided polygon.

A.2 Discretization of a sphere

We describe first the discretization of a spherical shell, the algorithm for the sphere naturally
follows from it. After the radius of the shell R and a discretization pace a are defined, the
code places a series of circumferences, of different radius, parallel to each other, on the surface
of the shell. Each couple of consecutive circumferences has distance which approaches a from
below, and each circumference is discretized as described above. In this way we obtain a
set of points, at fixed distance from the center, which distance with the nearest neighbor is
always less than a.
To discretize a sphere of radius R, a series of shells are used. The first one of radius
R, the second one R − a, the third one R − 2a and so on until the center of the sphere is
reached.

B Code technical specifications

B.1 SPH technical specifications

To define the behavior of a given material, is to say how the stress tensor is computed
from the strain tensor, two models need to be specified: the material strength model, that
computes the off diagonal elements of the stress tensor; and an equation of state (EOS) that
computes its diagonal elements. This two are chosen to be linear elastic and the stress strain
relation is ( )
1
σ ij = KTr(ϵ)δij + 2G ϵij − Tr(ϵ)δij
3
Where σ is the stress tensor, ϵ the strain tensor, K and G are the bulk and shear modulus
of the material.
To fully determine the behavior of the SPH interaction, some additional parameters
need to be specified: q1 , that defines the strength of a linear viscous force between SPH
discretization points; q2 , a quadratic viscous term that we always set to zero; khg that
defines the strength of an hourglass correction force. All the parameters, together with the
kernel type and its length h, are specified in the table 1.
Further details on the SPH code can be found in [12,25] while the source code is available
in LAMMPS [40] and LIGGGHTS [22].
14 F.J. Castro, S. Radl

Table 1: Details of the SPH implementation

Simulation Material model EOS q1 khg h∗ Kernel w∗∗

3D compression Linear elastic Linear 0.08 10 3 15


πh6
(h − r)3

2D compression Linear elastic Linear 0.1 10 3 10


πh5
(h − r)3

* In units of the discretization length a


** For r < h, else w(r) = 0

Table 2: Details of the DEM implementation

Simulation rc∗ Yc∗∗ νc Restitution coefficient


3D compression 1 5.5 0.4 0.9

2D compression 1 10 0.45 0.9

* In units of the discretization length a


** In units of the Young’s modulus of the material model E

B.2 DEM technical specifications

In all the simulations the contact interaction is only normal, hence there is no frictional
interaction directly acting through the DEM model. It is worth noting that in spite of
this, when two grains are in contact, some friction is introduced as a consequence of the dis-
cretization itself. Is indeed impossible, with our method, to obtain perfectly smooth surfaces,
although the irregularities can be reduced increasing the number of discretization points of
the surface.

References

1. Altomare, C., Crespo, A.J., Domínguez, J.M., Gómez-Gesteira, M., Suzuki, T., Ver-
waest, T.: Applicability of smoothed particle hydrodynamics for estimation of sea wave
impact on coastal structures. Coastal Engineering 96(Complete), 1–12 (2015). DOI
10.1016/j.coastaleng.2014.11.001
2. Ashour, A., Trittel, T., Börzsönyi, T., Stannarius, R.: Silo outflow of soft frictionless
spheres. Physical Review Fluids 2(12), 123302 (2017)
3. Bolton, F., Weaire, D.: Rigidity loss transition in a disordered 2d froth. Phys. Rev.
Lett. 65, 3449–3451 (1990). DOI 10.1103/PhysRevLett.65.3449. URL https://link.
aps.org/doi/10.1103/PhysRevLett.65.3449
4. Boromand, A., Signoriello, A., Ye, F., O’Hern, C.S., Shattuck, M.D.: Jamming of de-
formable polygons. Physical review letters 121(24), 248003 (2018)
5. Canelas, R.B., Crespo, A.J., Domínguez, J.M., Ferreira, R.M., Gómez-Gesteira,
M.: Sph–dcdem model for arbitrary geometries in free surface solid–fluid flows.
Computer Physics Communications 202, 131–140 (2016). DOI https://doi.org/10.
1016/j.cpc.2016.01.006. URL https://www.sciencedirect.com/science/article/pii/
S0010465516000254
6. Cantor, D., Cárdenas-Barrantes, M., Preechawuttipong, I., Renouf, M., Azéma, E.:
Compaction model for highly deformable particle assemblies. Physical Review Letters
124 (2020). DOI 10.1103/PhysRevLett.124.208003
7. Colagrossi, A., Bouscasse, B., Antuono, M., Marrone, S.: Particle packing algorithm
for sph schemes. Computer Physics Communications 183, 1641–1653 (2012). DOI
10.1016/j.cpc.2012.02.032
Title Suppressed Due to Excessive Length 15

8. Cárdenas-Barrantes, M., Cantor, D., Barés, J., Renouf, M., Azéma, E.: Compaction of
mixtures of rigid and highly deformable particles: a micro-mechanical model (2020).
DOI 10.1103/PhysRevE.102.032904. URL http://arxiv.org/abs/2006.02867http://
dx.doi.org/10.1103/PhysRevE.102.032904
9. Das, R., Cleary, P.W.: Evaluation of accuracy and stability of the classical sph method
under uniaxial compression. Journal of Scientific Computing 64, 858–897 (2015). DOI
10.1007/s10915-014-9948-4
10. Domínguez, J.M., Fourtakas, G., Altomare, C., Canelas, R.B., Tafuni, A., García-Feal,
O., Martínez-Estévez, I., Mokos, A., Vacondio, R., Crespo, A.J., Rogers, B.D., Stansby,
P.K., Gómez-Gesteira, M.: Dualsphysics: from fluid dynamics to multiphysics problems.
Computational Particle Mechanics (2021). DOI 10.1007/s40571-021-00404-2
11. Franz, S., Parisi, G., Urbani, P., Zamponi, F.: Universal spectrum of normal modes in
lowtemperature glasses. Proceedings of the National Academy of Sciences of the United
States of America 112, 14539–14544 (2015). DOI 10.1073/pnas.1511134112
12. Ganzenmüller, G.C.: An hourglass control algorithm for lagrangian smooth particle
hydrodynamics. Computer Methods in Applied Mechanics and Engineering 286, 87–
106 (2015). DOI 10.1016/j.cma.2014.12.005
13. Ghods, N., Poorsolhjouy, P., Gonzalez, M., Radl, S.: Discrete element modeling of
strongly deformed particles in dense shear flows. Powder Technology 401 (2022). DOI
10.1016/j.powtec.2022.117288
14. Gingold, R.A., Monaghan, J.J.: Smoothed particle hydrodynamics: theory and applica-
tion to non-spherical stars. Monthly Notices of the Royal Astronomical Society 181,
375–389 (1977). DOI 10.1093/mnras/181.3.375
15. Guida, M., Marulo, F., Belkhelfa, F., Russo, P.: A review of the bird impact process
and validation of the sph impact model for aircraft structures. Progress in Aerospace
Sciences 129, 100787 (2022). DOI 10.1016/j.paerosci.2021.100787
16. Herrmann, H.: Simulation of granular media. Physica A: Statistical Mechan-
ics and its Applications 191(1), 263–276 (1992). DOI https://doi.org/10.1016/
0378-4371(92)90537-Z. URL https://www.sciencedirect.com/science/article/pii/
037843719290537Z
17. Jean, M.: The non-smooth contact dynamics method. Computer Methods in
Applied Mechanics and Engineering 177(3), 235–257 (1999). DOI https://doi.
org/10.1016/S0045-7825(98)00383-1. URL https://www.sciencedirect.com/science/
article/pii/S0045782598003831
18. Jiang, F., Oliveira, M.S., Sousa, A.C.: Mesoscale sph modeling of fluid flow in isotropic
porous media. Computer Physics Communications 176(7), 471–480 (2007). DOI https:
//doi.org/10.1016/j.cpc.2006.12.003. URL https://www.sciencedirect.com/science/
article/pii/S001046550700029X
19. Joubert, J.C., Wilke, D.N., Govender, N., Pizette, P., Tuzun, U., Abriak, N.E.: 3d gra-
dient corrected sph for fully resolved particle–fluid interactions. Applied Mathematical
Modelling 78, 816–840 (2020). DOI https://doi.org/10.1016/j.apm.2019.09.030. URL
https://www.sciencedirect.com/science/article/pii/S0307904X19305608
20. Khalili, M.H., Roux, J.N., Pereira, J.M., Brisard, S., Bornert, M.: Numerical study of
one-dimensional compression of granular materials. i. stress-strain behavior, microstruc-
ture, and irreversibility. Physical Review E 95(3), 032907 (2017)
21. Khayyer, A., Gotoh, H., Falahaty, H., Shimizu, Y.: An enhanced isph–sph cou-
pled method for simulation of incompressible fluid–elastic structure interactions.
Computer Physics Communications 232, 139–164 (2018). DOI https://doi.org/10.
1016/j.cpc.2018.05.012. URL https://www.sciencedirect.com/science/article/pii/
S0010465518301759
22. Kloss, C., Goniva, C., Hager, A., Amberger, S., Pirker, S.: Models, algorithms and
validation for opensource dem and cfd-dem. Progress in Computational Fluid Dynamics,
An International Journal 12, 140 (2012). DOI 10.1504/PCFD.2012.047457
23. Kool, L., Charbonneau, P., Daniels, K.E.: Gardner-like transition from variable to per-
sistent force contacts in granular crystals (2022). URL http://arxiv.org/abs/2205.
06794
24. Landau, L.D., Lifshitz, E.M., Kosevich, A.M., Pitaevskii, L.P.: Theory of elasticity,
vol. 7 (1986)
16 F.J. Castro, S. Radl

25. Leroch, S., Varga, M., Eder, S., Vernes, A., Rodriguez Ripoll, M., Ganzenmüller, G.:
Smooth particle hydrodynamics simulation of damage induced by a spherical indenter
scratching a viscoplastic material. International Journal of Solids and Structures 81,
188–202 (2016). DOI https://doi.org/10.1016/j.ijsolstr.2015.11.025. URL https://www.
sciencedirect.com/science/article/pii/S0020768315004874
26. Lucy, L.B.: A numerical approach to the testing of the fission hypothesis. The Astro-
nomical Journal 82, 1013 (1977). DOI 10.1086/112164
27. Majmudar, T.S., Sperl, M., Luding, S., Behringer, R.P.: Jamming transition in granular
systems. Physical Review Letters 98 (2007). DOI 10.1103/PhysRevLett.98.058001
28. Marrone, S., Bouscasse, B., Colagrossi, A., Antuono, M.: Study of ship wave breaking
patterns using 3d parallel sph simulations. Computers & Fluids 69, 54–66 (2012). DOI
10.1016/j.compfluid.2012.08.008
29. Mollon, G.: A multibody meshfree strategy for the simulation of highly deformable gran-
ular materials. International Journal for Numerical Methods in Engineering 108(12),
1477–1497 (2016)
30. Mollon, G.: The soft discrete element method. Granular Matter 24 (2022). DOI
10.1007/s10035-021-01172-9
31. Nezamabadi, S., Nguyen, T.H., Delenne, J.Y., Radjai, F.: Modeling soft granular ma-
terials. Granular Matter 19(1), 1–12 (2017)
32. O’Hern, C.S., Langer, S.A., Liu, A.J., Nagel, S.R.: Random packings of frictionless
particles. Physical Review Letters 88, 4 (2002). DOI 10.1103/PhysRevLett.88.075507
33. O’Hern, C.S., Silbert, L.E., Liu, A.J., Nagel, S.R.: Jamming at zero temperature and
zero applied stress: The epitome of disorder. Physical Review E - Statistical Physics,
Plasmas, Fluids, and Related Interdisciplinary Topics 68, 19 (2003). DOI 10.1103/
PhysRevE.68.011306
34. Osorno, M., Schirwon, M., Kijanski, N., Sivanesapillai, R., Steeb, H., Göddeke, D.: A
cross-platform, high-performance sph toolkit for image-based flow simulations on the
pore scale of porous media. Computer Physics Communications 267, 108059 (2021).
DOI https://doi.org/10.1016/j.cpc.2021.108059. URL https://www.sciencedirect.
com/science/article/pii/S0010465521001715
35. Procopio, A.T., Zavaliangos, A.: Simulation of multi-axial compaction of granular media
from loose to high relative densities. Journal of the Mechanics and Physics of Solids
53(7), 1523–1551 (2005). DOI https://doi.org/10.1016/j.jmps.2005.02.007. URL https:
//www.sciencedirect.com/science/article/pii/S0022509605000529
36. Ruiz-Franco, J., van Der Gucht, J.: Force transmission in disordered fibre networks.
Frontiers in Cell and Developmental Biology 10 (2022)
37. Shundyak, K., van Hecke, M., van Saarloos, W.: Force mobilization and generalized
isostaticity in jammed packings of frictional grains. Physical Review E 75(1), 010301
(2007)
38. Silbert, L.E., Ertaş, D., Grest, G.S., Halsey, T.C., Levine, D.: Geometry of frictionless
and frictional sphere packings. Phys. Rev. E 65, 031304 (2002). DOI 10.1103/PhysRevE.
65.031304. URL https://link.aps.org/doi/10.1103/PhysRevE.65.031304
39. Souto-Iglesias, A., Delorme, L., Pérez-Rojas, L., Abril-Pérez, S.: Liquid moment ampli-
tude assessment in sloshing type problems with smooth particle hydrodynamics. Ocean
Engineering 33, 1462–1484 (2006). DOI 10.1016/j.oceaneng.2005.10.011
40. Thompson, A.P., Aktulga, H.M., Berger, R., Bolintineanu, D.S., Brown, W.M., Crozier,
P.S., in ’t Veld, P.J., Kohlmeyer, A., Moore, S.G., Nguyen, T.D., Shan, R., Stevens,
M.J., Tranchida, J., Trott, C., Plimpton, S.J.: Lammps - a flexible simulation tool for
particle-based materials modeling at the atomic, meso, and continuum scales. Computer
Physics Communications 271, 108171 (2022). DOI 10.1016/j.cpc.2021.108171
41. Vacondio, R., Mignosa, P., Pagani, S.: 3d sph numerical simulation of the wave generated
by the vajont rockslide. Advances in Water Resources 59, 146–156 (2013). DOI 10.1016/
j.advwatres.2013.06.009
42. Vego, I., Tengattini, A., Andò, E., Lenoir, N., Viggiani, G.: The effect of high relative
humidity on a network of water-sensitive particles (couscous) as revealed by in situ x-ray
tomography. Soft Matter 18, 4747–4755 (2022). DOI 10.1039/d2sm00322h
43. Vyas, D.R., Cummins, S.J., Delaney, G.W., Rudman, M., Cleary, P.W., Khakhar,
D.V.: Elastoplastic frictional collisions with collisional-sph. Tribology International 168,
107438 (2022)
Title Suppressed Due to Excessive Length 17

44. Vyas, D.R., Cummins, S.J., Rudman, M., Cleary, P.W., Delaney, G.W., Khakhar, D.V.:
Collisional sph: A method to model frictional collisions with sph. Applied Mathematical
Modelling 94, 13–35 (2021)
45. Wang, J., Fan, B., Pongó, T., Harth, K., Trittel, T., Stannarius, R., Illig, M., Börzsönyi,
T., Hidalgo, R.C.: Silo discharge of mixtures of soft and rigid grains. Soft Matter 17,
4282–4295 (2021). DOI 10.1039/D0SM01887B
46. Wyart, M., Silbert, L.E., Nagel, S.R., Witten, T.A.: Effects of compression on the vibra-
tional modes of marginally jammed solids. Physical Review E - Statistical, Nonlinear,
and Soft Matter Physics 72 (2005). DOI 10.1103/PhysRevE.72.051306
47. Xenakis, A., Lind, S., Stansby, P., Rogers, B.: Landslides and tsunamis predicted by in-
compressible smoothed particle hydrodynamics (sph) with application to the 1958 lituya
bay event and idealized experiment. Proceedings of the Royal Society A: Mathematical,
Physical and Engineering Science 473, 20160674 (2017). DOI 10.1098/rspa.2016.0674
48. Zhang, C., Rezavand, M., Zhu, Y., Yu, Y., Wu, D., Zhang, W., Wang, J., Hu,
X.: Sphinxsys: An open-source multi-physics and multi-resolution library based
on smoothed particle hydrodynamics. Computer Physics Communications 267,
108066 (2021). DOI https://doi.org/10.1016/j.cpc.2021.108066. URL https://www.
sciencedirect.com/science/article/pii/S0010465521001788

You might also like