Powder Technology 352 (2019) 273–282

Contents lists available at ScienceDirect

Powder Technology

journal homepage: www.elsevier.com/locate/powtec

DEM particle upscaling for large-scale bulk handling equipment

and material interaction
Stef Lommen a, Mohammadjavad Mohajeri a, Gabriel Lodewijks b, Dingena Schott a,⁎
a
Delft University of Technology, department of Maritime and Transport Technology, section of Transport Engineering and Logistics, the Netherlands
b
UNSW Sydney, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Purpose: The development of bulk material handling equipment can be accelerated and made less expensive
Received 30 August 2018 when testing of virtual prototypes is adopted. However, the modelling of a grab unloader requires a large volume
Received in revised form 11 April 2019 (77 m3) of iron ore pellets, making the computational costs prohibitive. This paper investigates the extent to
Accepted 14 April 2019 which the original particles can be substituted by larger, coarser grains. It is crucial that this particle upscaling
Available online 25 April 2019
does not alter the realistic behaviour of the simulated bulk material, nor its interaction with the bulk handling
equipment.
Keywords:
Particle upscaling
Approach: First, our coarse graining technique is explained and set out for the particle system at hand. The mate-
Coarse graining rial behaviour is then characterized using three laboratory experiments (two angle of repose tests and a penetra-
Material equipment interaction tion test). Next, the results of simulations using two contact models with and without coarse graining with
Iron ore pellets different scale factors are compared with the measured material behaviour and material-equipment interaction.
Grab This includes a comparison of the macrobehaviour of the bulk material and the tool interaction of coarser grains
Penetration in a cutting and sliding process. After reaching a satisfactory verified solution on the laboratory scale, the material
Angle of repose behaviour and interaction behaviour of a large-scale experiment are modelled. A simulation model of a grab un-
Simulation
loader was used for validation of the chosen coarse graining approach.
Industrial scale
Findings: Using the scaling method presented, the macroscopic tests indicated consistent material behaviour, re-
gardless of the chosen particle scale for two contactmodels. Scaling of the tool interaction process produced
mixed results: the sliding process scaled consistently but the penetration process did not, most likely because
it is significantly harder for coarser grains to move since they have to move further to the sides before the tool
can pass, leading to higher normal forces and frictional forces on the tip. This inconsistency was compensated
for by adjusting the wall friction coefficient in the tip of the penetration tool. Once this adapted coarse graining
scheme was applied to the industrial-scale simulation of a grab unloader, it produced consistent particle-scale in-
variant results.
Originality/value: This research is the first to show how coarse graining schemes for DEM simulations can be ap-
plied to large-scale bulk handling equipment involving dominance of material equipment interaction through
penetration of the bulk material.
© 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction closing, the hoisting cables are also tensioned and the full grab is hoisted
out of the vessel towards the hopper on the quay.
Grabs are a type of equipment used for unloading dry bulk cargo ves- Fig. 1 shows that the computational costs of a single simulation of
sels as displayed in Fig. 1. They remove substances such as iron ore or the required 77 m3 of bulk material for grab operation are decreasing
coal from the ship and transfer it to a hopper on the quay. The grabbing for coarser element sizes. These costs have been determined for a
process consists of three steps: lowering the grab into the material, pen- 2014 hexa-core workstation PC and elements with a particle density
etrating the material on impact and then tensioning the closing cables of 4000 kg/m3 and a shear modulus of 300 MPa. It can be seen that
to commence closing of the grab. During closing, the two halves of the the computational costs are approximately 104 h for elements with
bucket further penetrate the material until the blades meet. The mate- the same size as the pellets of the material model, which is impractical
rial is pushed into the bucket, where it slides along its surface. After for the virtual prototyping of grabs. In fact, a computing time of 24 h
or less is desirable when simulating a large number of grab design var-
⁎ Corresponding author. iations. To achieve this desired reduction in computation costs, particle
E-mail address: d.l.schott@tudelft.nl (D. Schott). scaling techniques are investigated (Fig. 2).

https://doi.org/10.1016/j.powtec.2019.04.034
0032-5910/© 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
274 S. Lommen et al. / Powder Technology 352 (2019) 273–282

upscaling limit determined for the vertical velocity profile (2.5) were in-
dependent for the outlet opening to particle size ratios up to 21.7.
The approach in this paper is different in that we calibrate and vali-
date the DEM model explicitly with the aim to reduce computational
cost of modelling N107 particles in a real scale industrial application
where the equipment penetrates and excavates the bulk material. For
such applications particle upscaling in combination with calibration
and validation has never before been investigated. Moreover we explic-
itly address the interaction between material and equipment in
upscaling the particles in a penetration test with a constant realistic
tool size. Finally, we validate the scaling approach on the industrial
scale.
This paper aims to develop a coarse graining or particle upscaling ap-
proach capturing material-equipment interaction for large-scale bulk
handling equipment. Both the material model and the material interac-
tion models are scaled.

Fig. 1. Unloading a vessel with a grab at TATA Steel, IJmuiden. Courtesy of Nemag B.V.
2. Particle upscaling method

Roessler and Katterfeld [23] give a brief overview of the scaling tech- Coarse graining is a technique whereby the original particles with
niques used in literature. They identify three general categories: exact radius R are substituted by larger, coarser grains with radius R′. The
scaling of grain and equipment scale; coarse graining or upscaling; coarse grain radius R′ can be expressed as a multiplication of the original
and scalping, where only the larger particles in a size distribution are grain radius R and a scaling factor s as shown in Eq. (1). For example the
considered. The exact scaling approaches are described by Feng et al. use of scaling factor s = 2 results in particles twice the original size.
[11,12]; Poschel et al. [23]. Schott et al. [25] scaled particles and equip- The accent' will be used for other parameters as well, denoting them
ment to experimentally investigate excavating motion characteristic at the coarse scale.
for grabs, bulldozers and reclaimers. The determined scale factors for
the torque resulting from exact scaling laws did not match the scale fac- R0 ¼ sR ð1Þ
tors found experimentally. Coarse graining or particle upscaling is a
technique where the original particles with radius R are substituted by
larger, coarser grains with radius R'. The main advantage of coarse One promising approach to coarse graining, by Bierwisch et al. [5] is
graining is reducing the number of elements in a simulation which in based on the idea that the system with the scaled grain size should in-
turn lowers the computational costs. Another benefit of using larger corporate the same energy density as the original system with unscaled
grains is that the critical timestep increases as well, reducing the num- grains. As the potential energy density is independent of the grain ra-
ber of iterations required to complete a simulation. Coarse graining al- dius if the void ratio and particle density are constant, it is necessary
lows for large-scale simulation involving large volumes of particles that the particle density of the scaled system ρp′ be identical to that of
while keeping computational expenses affordable. However, its effects the original system ρp (Eq. (2)). According to Bierwisch et al., the pro-
on accuracy still need to be assessed carefully. Coarse graining has posed scaling system does not affect the volume fraction, so the poten-
been investigated by several researchers, although their findings are tial energy density of the scaled system is comparable to the original
not always in agreement. For example, Baars [4] writes that the average one.
grain size only matters on the micro scale and “has absolutely no influ-
ence on the macro behaviour”, while Achmus and Abdel-Rahman [1] ρp ¼ ρ0p ð2Þ
state that “a remarkable scale effect exists” and that “the parameters
of the model have to be adapted by means of a new calibration proce-
dure if the scaling factor is changed.” Coarse graining techniques were Consequently, the mass of the coarse grained particles is scaled
applied to industrial applications of modelling material flow through according to Eq. (3), while the moment of inertia can be found using
transfer chutes by Xie et al. [28], for hopper discharge processes by Eq. (4).
Grima [13], for pneumatic conveying systems by Sakai and Koshizuka
[24] and industrial gas solid flows by Mori et al. [21] accounting for par-
ticle fluid interaction.
Little work has been done on calibration and validation of DEM pa-
rameters with upscaled particles. Roessler and Katterfeld [23] per-
formed simulations of the angle of repose of sand with an open ended
cylinder. They used a scale factor up to 12.5 for both the particle and cyl-
inder diameter to ensure the same number of particles in the simulation
to effectively observe the bulk behaviour. They found that upscaling of
both the particles and cylinder diameter had no significant effect on
the results of the angle of repose for a constant cylinder diameter to par-
ticle size ratio. With such approach the computing time is not reduced
since the number of particles remains the same. Coetzee [9] used a con-
stant drum size and upscaled non-spherical particles to investigate par-
ticle upscaling effects on the dynamic angle of repose using a linear
contact model. With the constant drum size the particles could be scaled
up to a factor 4 without compromising the bulk behaviour. The cali-
brated model was then used in a hopper setup where a maximum
scale factor of 1.3 could be obtained for the discharge rate. While the Fig. 2. Effect of particle size on computing time.
 S. Lommen et al. / Powder Technology 352 (2019) 273–282 275

4 03 0 rffiffiffi
m0 ¼ πR ρ pffiffiffiffiffi
3 ð3Þ 5 pffiffiffiffiffiffiffi
kt ¼ −8G R and ct ¼ −2 β m ð10Þ
¼ s3 m 6

Fig. 3 shows the spring-damper system in normal direction of a

2
I ¼ m0 R0
0 2 group of original particles as well as the coarse grain with s = 2. As
3 ð4Þ the coarse grain doubles in size, eight of the original particles are re-
¼ s5 I
placed by the coarse grain. In order to conserve the character of the con-
tact, the stiffness kn′ and damping cn′ of the coarse grained particle need
In order to create similarity for the kinetic energy density as well, the to be the same as the equivalent stiffness keq and damping ceq of the
particle velocities should be the same as in the original system. This group of original particles.
means that coarse graining should not affect the accelerations and de- The equivalent stiffness of the original system keq consists of s2 pairs
celerations within the system. of original springs and can be derived using Eq. (11). The first step is to
A system of coarse grained particles is assumed to move identically determine the stiffness kseries of each of the four pairs. This can be done
to the original system of particles (Eq. (5)), implying that the velocity based on δseries = sδn and the definition of kn in Eq. (9):
of the coarse grained particle is equal to the average of the group of orig-
inal particles it is representing [24]. This satisfies the demand for a sim- keq ¼ s2 kseries
ilar kinetic energy density shown in Eq. (6). 1
Fk
¼ s2 41:5
v0 ¼ v ð5Þ δseries
1
Fk
1 0 02 1 ¼ s2 4 1:5 ð11Þ
m v ¼ s3 mv2 ð6Þ ðsδn Þ
2 2 1:5
2 F k;series =δn
¼ s 1:5
s
The contact stiffness and damping need to be scaled accurately as kn
¼ s2 1:5
well, in order to maintain the same energy losses experienced in inelas- s
pffiffi
tic collisions. In this paper, normal and tangential contact models as per ¼ skn
Hertz Mindlin are used:
When the same equivalent Young's modulus is maintained for the
rffiffiffiffiffiffiffiffiffi coarse system (Eq. (12)),
3 3 1
Fn ¼ Fk;n þ Fd;n ¼ kn δn en þ cn vn
2
kn δ4n ð7Þ
2 
E0 ¼ E ð12Þ
pffiffiffiffiffi pffiffiffiffiffi 1
Ft ¼ Fk;t þ Fd;t ¼ kt δn δt et þ ct vt kt δ4n ð8Þ the stiffness of the coarse system kn′ in Eq. (13) is equal to the equiva-
lent stiffness of the original system keq. For the tangential stiffness kt,
the same scaling factor can be acquired. Therefore, the contact stiffness
kn and kt are the normal and tangential stiffness coefficients and is identical to the equivalent stiffness of the original system.
cn and ct are the normal and tangential damping coefficients as de-
fined in Eqs. (9) and (10), while δn and δt are the normal and tan- 0 4 0  pffiffiffiffiffiffi

kn ¼ E R0
gential overlaps and en and et the normal and tangential unit 3
vectors. pffiffi 4  pffiffiffiffiffi
¼ s E R ð13Þ
3 pffiffi
rffiffiffi ¼ skn
4  pffiffiffiffiffi 5 pffiffiffiffiffiffiffi ¼ keq
kn ¼ E R and cn ¼ −2 β m ð9Þ
3 6

Fig. 3. A coarse grain contact of s = 2 and the equivalent contact of original group of particles.
276 S. Lommen et al. / Powder Technology 352 (2019) 273–282

Eq. (14) shows that the equivalent damping coefficient ceq can be ob- similarity between the original system and the coarse grain one has
tained by computing the damping of the s2 series of springs. The been achieved.
damping of each series of springs cseries can be determined based on
the definition of cn in Eq. (9), δseries = sδn and the scaling factors for vn €θ0 ¼ T0
(Eq. (5)) and kn (Eq. (13)). I0
F 0t R0 þ μ r F 0n R0
¼
ceq ¼ s2 cseries I0 ð18Þ
s3 F t sR þ μ r s3 F n sR
1 ¼
Fd s5 I
¼ s2 q4ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1€
1:5 ¼ θ
kseries δseries
vseries s
1
F
¼ 4 d
s2 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi
This shows that the coarse grain system has the same energy compo-
kn nents as the original one. These identical components are the potential
vn 1:5 ðsδn Þ1:5 energy, the kinetic energy, the dissipation of energy through damping
s
1 ð14Þ and the rotational energy.
Fd
¼ s2 pffiffi q 4 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
svn kn ðδn Þ1:5 3. Model calibration on laboratory tests
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
F d;series = vn kn ðδn Þ1:5 DEM material models for iron ore pellets in interaction with a grab
¼ s2 pffiffi unloader have been developed by Lommen [20], using two angle of re-
s
cn pose tests (ledge and cone test) and a penetration test. These tests were
¼ s2 pffiffi
s selected to represent the characteristic material behaviour and the ma-
¼ s1:5 cn terial interaction behaviour in grabs. Fig. 5 shows the ledge and cone
angle of repose tests while the penetration test and its tools are
Eq. (15) then shows that the damping coefficient cn′ of the coarse shown in Figs. 4 and 6. For the penetration test the size and shape of
grain is identical to the equivalent damping ceq. For the tangential tool A are similar to the tool used by Asaf et al. [3]), while the two
damping ct, the same scaling factor can be calculated. This confirms other tools B and C are of equal length and have the same angle, but
that the energy losses due to inelastic collisions are the same for the both have a blunt tip. The experimentally measured angle of repose
original and the coarse grained system. with the ledge test and cone test were 41 and 26 degrees, respectively.
rffiffiffi The iron ore pellets are approximately spherical particles with a di-
5 pffiffiffiffiffiffiffi ffi ameter between 8 and 14 mm. The particle size distribution is shown
c0n ¼ 2 β m0 
r6ffiffiffi in Fig. 7 and can be approximated with a normal distribution with an av-
5 pffiffiffiffiffiffiffi ð15Þ erage diameter of 11 mm and a normalized standard deviation of 0.1.
¼ s1:5 2 β m
6 This approximation has an R2 of 0.9997.
¼ s1:5 cn
¼ ceq

Normal contact forces acting on the particles in the coarse grained

system are estimated to be s3 times larger than in the original system
[6]. This means that the entire group of original particles is considered
to be in the same collision as the coarse grained particle.
The tangential force Ft′ in the coarse grained system computed using
the same static friction coefficient μs, as shown in Eq. (16).

F 0t ¼ μ s F 0n
ð16Þ
¼ μ s s3 F n

For the contact model with rolling motions, the rotational energy of
the coarse grained system also needs to be comparable with the original
system (Eq. (17)). When rotation is considered, both the original parti-
cles and the coarse grain rotate around their respective centre of mass.
Rotation around other particles is only taken into account when this
_ By ensuring that rotational energy is sim-
leads to a rotational velocity θ.
ilar, the coarse grain system can be compared to the original system.

1 0 _ 02 1 2
I θ ¼ s3 Iθ_ ð17Þ
2 2

The torques on the coarse grains depend on the contact forces and
the radius of the particle, and accordingly the torque is now s4 times
higher than the original torque. This is also true for the rolling friction
torque. Eq. (18) shows the rotational velocity of the coarse grained par-
ticle €θ0 when the coarse grain's torque and moment of inertia (Eq. (4))
are taken into account. The resulting angular velocity of the coarse
grain satisfies Eq. (17), confirming that the coarse grain has the same ro-
tational energy as the group of original particles and consequently that Fig. 4. Penetration test.
 S. Lommen et al. / Powder Technology 352 (2019) 273–282 277

Fig. 5. The ledge (left) and cone (right) angle of repose tests.

Two contact models have been used in this research; the Hertz- simulation results for both AoR tests leading to the set of parameters
Mindlin model with angular movements restricted and the Hertz- in Table 1.
Mindlin model with rolling enabled. The rolling-restricted model has The DEM input parameters are able to predict the penetration resis-
been selected because as Bierwisch [6] suggests that this can be useful tance of three different tools in iron ore pellets very well. Figs. 8 and 9
for spherical particles. The rolling-enabled model follows the recom- show the resistance as the three tools penetrate the iron ore pellets.
mendation by Ai et al. [2] to use model C [2] for quasi-static simulations. The mean and 95% confidence intervals are plotted based on 10 repeti-
Following the suggestion of Wensrich and Katterfeld [26], the rolling tions of experiments and 10 repetitions for simulations.
stiffness of Iwashita and Oda [14] is used and the viscous rolling
damping torque is disabled.
4. Particle upscaling applied to calibrated model
The length, width and height of the ledge test measured 300 mm,
200 mm and 300 mm, respectively. The width of the container was 18
The following subsections focus on testing this coarse graining
times the average diameter, which is lower than the minimum ratio
scheme. The coarse graining is assessed with an angle of repose test, a
prescribed by Zhou et al. [30] and Derakhshani et al. [10] to exclude
wall interaction and a penetration test.
wall effects. However, here possible wall effects were omitted from
the results by excluding particles near the walls. The free cone test has
been simulated in a similar rectangular set-up as the ledge test, al- 4.1. Angle of repose
though the filling process differs. Instead of filling the volume of the
start of the simulation and allowing the material to settle before open- Angle of repose simulations were performed to investigate the effect
ing the ledge, the ledge is open and particles are generated with a con- of coarse graining on the simulated shearing behaviour. Fig. 10a shows
stant (scale independent) mass flow rate. This method of creating an the angle of repose of particles with a diameter of 7.5 mm and coarse
angle of repose resembles the free cone setup as the flowing particles grains with s = 2 (diameter of 15 mm) and s = 4 (diameter of
create a slope by settling down. The drop height in the simulation is 30 mm), all without rolling. For the results displayed in Fig. 10a a
comparable to the experimental drop height making the kinetic energy fixed box was used. It can be observed that the angles produced are
in the system similar. No significant effect of the rectangular setup was comparable, although for very gentle angles the coarse grains with d
expected as possible wall effects were omitted from the results by ex- = 30 mm show a steeper angle than the original system. It is possible
cluding particles near the walls. that this is an effect of the limited number of elements in the coarse
The DEM input parameters are summarized in Table 1. Two addi- grain simulations, disturbing the accuracy of the angle measurements.
tional input parameters are required with the rolling model: the rolling For the course grained simulations of the ledge and cone test the di-
friction coefficients μr and ϕr. These two coefficients influence the mensions of the box were scaled with s, and for the free cone test the
rolling behaviour of interparticle contacts and of contacts between par- mass flow per unit width was kept constant. The results for the material
ticles and equipment, respectively. Both material models were cali- model with rotational motions are shown in Fig. 10b, where it can
brated by minimising the difference between experimental and be observed that the scaling factor does not significantly affect the out-
come of the simulations. These results demonstrate that the proposed

Fig. 6. Three tools for the penetration test.

278 S. Lommen et al. / Powder Technology 352 (2019) 273–282

Fig. 7. Particle size distribution of iron ore pellets.

coarse graining scheme is able to predict the shearing behaviour of par- small difference might be explained by the smaller number of repeti-
ticles regardless of the chosen scale. tions performed for tool A simulations. Splitting the bulk material into
two parts to slide along the shaft is significantly harder for coarser
grains since they have to move further to the sides before the tool to
4.2. Penetration resistance pass, leading to higher normal forces and frictional forces on the tip
and requires more movement underneath the tool. It is likely that the
The effect of coarse graining was also investigated using a penetra- assumption that the coarse grain moves in a similar way to the original
tion test, as this is key to modelling material-equipment interaction in system no longer holds during penetration. In short, resistance gains as
grab unloading and other operations involving penetration. This pene- grain size increases, due to the dependence of the penetration process
tration test is similar to those described in the calibration, although on the particle size.
there are some differences: it uses (blunter) tool shape C from Fig. 6, a The resistance on the shaft is 3–5% of the tip resistance, and does not
penetration rate of 100 mm =s and a maximum penetration depth of display such a strong effect as at the tip. The small difference between
300 mm in order to approximate grab penetration more closely. The the original system and the coarser ones is possibly caused by the
penetration test has been modelled in such a way that the resistances shaft area close to the tip experiencing slightly higher compressive
of the tip and shaft are measured separately. The tip is defined as the forces due to the additional particle movements required to split
part of the tool where the cross is still expanding until it reaches the the coarser systems. The difference in shaft resistance between tool A
maximum width of 40 mm. For tool A and C this defines the shaft height and C for the coarser grains might be explained by the lesser repetitions
as 225 mm and 265 mm, respectively. The used tool length was 200 mm and difference in effective shaft height for tool A (225 mm) and tool C
for s = 1. For the course grained simulations the tool length, depth and (265 mm). All the other contacts between the bulk material and the
width of the box was scaled with s. The width of the box was large shaft are scaled without affecting the resistance, confirming the validity
enough to exclude wall effects. The tool length was also increased of coarse graining scheme except for tip forces.
with scalefactor s to have a comparable number of particle contacts As the reduction of computational expenses is essential for the vir-
over the length of the tool. The penetration resistance measured in the tual prototyping of grabs, ways of mitigating the effects on the splitting
simulations of s = 2 to s = 5 were related to the tool length of s = 1 of the bulk material have been investigated. Since the coarse grains be-
as to compare the results. For tool C with a tip size of 20 mm the ratio have adequately in the angle of repose test and for the shaft friction, a
of the tip and particle diameter is between 1.8 (s = 1) and 0.36 (s = solution that incorporates changing behaviour in these tests is ruled
5), while for the sharp tool (tool A) the ratio is about 0 regardless of out. This leads to a solution that changes the contact behaviour of the
the particle scale. tip; for example, assigning different contact properties to the tip-
The penetration resistance of the calibrated particles (s = 1) and of particle contacts. The tip-particle contacts can be altered in several
four coarser materials is shown in Fig. 11 for tool C and in Fig. 12 for tool ways; for example, in their stiffness, damping and sliding behaviour.
A. It can be observed that for both tools the resistance on the tip in- Lowering the stiffness of the particle-tip contacts would make the over-
creases when the grains become larger, averaging an additional 16% lap higher than obtained from Lommen et al. [19] or the value recom-
and 18% for each step in grain size for tool C and A respectively. This mended by Cleary [8] and so compress the bulk material, resulting in
undesirable local density concentrations around the tip. Also, the trans-
fer of a contact from the tip to the shaft would lead to problems since a
Table 1
Calibrated material models with rotation of particles restricted (NR) and enabled (RC)
jump in stiffness would result in undesired jumps in overlap. Damping
(Lommen [20]). of the contacts did not show any effect on the sensitivity analysis from
the previous chapter since contact velocities are low and are therefore
Property Symbol NR RC Unit
not expected to aid in reducing the penetration resistance. Adjusting
Shear Modulus G 1e10 1e10 Pa the sliding coefficient of the tip is in fact the most promising solution,
Poisson's ratio ν 0.3 0.3 –
since it reduces the additional friction forces caused by the additional
Density ρ 3700 3700 kg/m3
Coefficient of restitution CR 0.6 0.6 – travel of the coarse grains to normal proportions. Consequently, the
Static Friction μs 0.21 0.41 – sliding behaviour of the tip-particle contacts is selected to resolve the
Rolling Friction μr – 0.145 – increase in tip forces caused by the coarse graining scheme.
Wall Friction ϕw 0.41 0.36 – Influencing the penetration by lowering the wall friction coefficient
Rolling friction wall-particle ϕr – 0.13 –
of the tip on tool C is investigated by assigning a different value to the tip
 S. Lommen et al. / Powder Technology 352 (2019) 273–282 279

material models can be coarse grained while maintaining the penetra-

tion characteristics of the original system. It is expected that due to
the lower tip resistance slightly lower reduced wall frictions are ex-
pected to compensate for tool A, this suggest that slightly higher scale
factors can be used for sharp tool shapes. The reason for the difference
in the observed maximum scale factors has not yet been fully under-
stood, and could be further investigated by analysing the particle behav-
iour in close vicinity to the penetration tool and thereby linking particle
level behaviour to bulk response.
The coarse graining scheme leads to a considerable reduction in
computational costs and cuts simulation time for a large-scale grab.
For the material model excluding rotation, the maximum scaling factor
of s = 5 reduces the computing time from an estimated 104 hours to just
under 18 h. The material model including rolling model C can be coarse
grained with s = 3, resulting in a computing time of five days. With the
help of these reductions, the coarse grain material model can now
be used to complete full-scale grab simulations within days, instead of
Fig. 8. Calibrated penetration resistance with restricted rotation of particles.
years, without significant consequences. Achieving a reasonable com-
putational time remains important specifically for application of optimi-
while the wall friction coefficient of the shaft remains unaffected. Fig. 13 sation algorithms.
shows that, for coarse grains with s = 4, the wall friction needs to be re-
duced to ϕw′ = 0.05 in order to achieve the same the penetration char- 5. Large-scale experiments
acteristics as the original system. It appears that the limit for coarse
graining of penetration is reached at s = 5, since the wall friction cannot For a large-scale simulation of a grab unloader a model describing
be reduced further and it is no longer possible to compensate for the the movements of the grab needs to interact with the DEM model. Ac-
coarser grains. For the material model with rotation restriction both cording to Wittenburg [27], multibody dynamics (MBD) deals with sys-
tip friction coefficients ϕw′ and ϕr′ have been adjusted. Fig. 14 shows tems composed of rigid bodies interconnected by joints and force
that the material model including rolling needs more compensation, elements. The technique allows the behaviour of these systems to be de-
resulting in an upper limit of s = 3 for the coarse graining of the pene- scribed in mass-, damping- and stiffness matrices, offering the possibil-
tration. By applying reduced tip-particle friction coefficients both ity to solve complex systems with a large number of bodies numerically.
Research by Brans [7]; Park et al. [22]; Yoo et al. [29] and Langerholc
et al. [15] shows that MBD is a suitable tool for the modelling of complex
large-scale rigid body mechanisms such as grab unloaders. In this paper
the grab is modelled with MBD based on Lommen et al. [17,18].
The previous section presented a general coarse graining approach
as well as an adapted approach for penetration processes, which con-
sists of adapted wall friction coefficients for the tip of the blades with
toolshape C. This adaptation is applied to the tip of the blades of the
grab geometry. The MBD model is coupled with the DEM model as per
Lommen et al. [16], resulting in the co-simulation shown in Fig. 15.
The maximum grain size for the rolling-C model has been deter-
mined in (Section 4.2) at s = 3, which is lower than the s = 5 for the
no-rolling model. As a result, a co-simulation with 77 m3 requires 4.6
times more particles, increasing the computational costs.
Fig. 16 shows the effect of coarse graining on the filling of the grab,
with and without compensation for the penetration behaviour. Clearly,
the increase in grain size results in a decline in the mass of material
grabbed when there is no compensation for the increase in penetration
Fig. 9. Calibrated penetration resistance with rolling friction model. resistance. When the compensation discussed in Section 4.2 is taken

(a)

(b)

Fig. 10. Coarse graining of angle of repose. (a) No-rotation material model, (b) rolling particles. The errorbars denote the 95% confidence interval of 10 repetitions.
280 S. Lommen et al. / Powder Technology 352 (2019) 273–282

(a) (b)

Fig. 11. Effect of coarse graining on the penetration resistance of tool shape C. (a) On the tip, (b) on the shaft. The errorbars denote the 95% confidence interval of 10 repetitions.

(a) (b)

Fig. 12. Effect of coarse graining on the penetration resistance of tool shape A (a) On the tip, (b) on the shaft. The errorbars denote the 95% confidence interval of 3 repetitions.

(a) (b)

Fig. 13. Penetration resistance with adjusted tip friction coefficient, (a) s=4, (b) s=5.

into account, this effect disappears and the predicted amount becomes 6. Conclusions
independent on the selected particle scale - an independence which
demonstrates that the adapted coarse graining technique can be used In this work the coarse graining technique was used to demonstrate
in large-scale grab simulations to save considerable amounts of compu- that a group of particles can be replaced by a single, coarse particle as
tation effort without compromising the results. In the case of the grab long as the particle process is not inherently dependent on the grain
discussed in this paper, the limit for particle upscaling is equal to five size. This work focussed on the particle upscaling in relation to material
times the original size, while the length scale of the particles remains behaviour, as well as the effect of particle upscaling on material equip-
much smaller than the length scales of the grab. ment interactions for tool penetration and sliding regimes.
 S. Lommen et al. / Powder Technology 352 (2019) 273–282 281

Fig. 14. Penetration resistance for coarse grain rolling material model.
Fig. 16. Effect of coarse graining on the grabbed amount of iron ore pellets for two different
contactmodels: particle rotation restricted (NR) and enabled (RC). Two approaches are
tested: the general one from Section 2 and an adapted one for penetration processes
• The particle upscaling relationships have been theoretically derived from Section 4.2.
and tested for two contact models; Hertz-Mindlin (no-slip) with rolling
model C and with rolling of the particles restricted. The coarse grained course graining scaling factors are limited by the dominant material
system contains identical quantities of potential, kinetic, rotational and equipment interaction.
dissipated energy as the original system. This was confirmed in simula- • The particle upscaling was succesfully demonstrated and confirmed for
tions of the angle of repose, where the shearing behaviour of coarse an industrial setting for both contact models; Hertz-Mindlin (no-slip)
grains matched that of the original material model. with rolling model C and with rolling of the particles restricted. The
• The material equipment interaction tested by means of two tool obtained scalingfactors are 3 and 5 respectively and are limited
shapes showed that the resistance on the shaft is 3 − 5% of the by the above mentioned particle scale variance of the tip resistance
tip resistance. of the penetrating tool. With the maximum scale factors the length
• The sliding resistance on the shaft of the penetration tool is constant scales of the particles remain much smaller than the length scales of
for the tested grain sizes and can be concluded to be scale invariant the grab.
for quasi static sliding interactions. This scale invariance is likely to • The coarse graining technique resulted in a reduction of the duration of
hold for other flow regimes with sliding contacts as well. the iron ore pellet grab simulation of 55 times, while the performance
• The penetration resistance of the tip is dependent on the grain size. and calibrated behaviour including penetration resistance remained
Coarse graining of the penetration resulted in an increase of 16% in re- constant.
sistance on the tip of the blunt shaped penetration tool when the grain
size was doubled. In conclusion, particle upscaling can be applied succesfully although
• A compensation was succesfully applied by lowering the sliding fric- specific grain size dependent interactions should be identified, evaluated
tion of the penetrating tip for coarser grains, bringing the resistance and adapted. The representative selected calibration tests can be used to
back to normal levels. The maximum compensation possible is apply- identify specific particle size variant material equipment interactions and
ing a particle wall friction coefficient of 0 at the tip. adapt them accordingly. When coarse graining schemes are carefully ap-
• For industrial processes involving penetration of tools, such as bucket plied depending on the case at hand, this technique can then help users to
elevators, bucket reclaimers and grabs, the maximum applicable undertake large-scale DEM simulations without affecting results.

Fig. 15. Co-simulation of a scissors grab, consisting of the MBD model (left) and the DEM model (right).
282 S. Lommen et al. / Powder Technology 352 (2019) 273–282

For the excavation process studied in this paper we found different [13] Andrew P. Grima, P.W. Wypych, On improving the calibration and validation of
computer simulations for bulk materials handling systems, Aust. Bulk Hand. Rev.
scale factors than previous research aiming at the calibration of DEM pa- (October 2009) 84–91.
rameters for upscaled particles [9,23] which confirms the dependence [14] K. Iwashita, M. Oda, Rolling resistance at contacts in simulation of shear band devel-
on process, characteristic equipment size and, possibly, contact models. opment by DEM, J. Eng. Mech. 124 (3) (1998) 285–292, http://ascelibrary.org/doi/
abs/10.1061/%28ASCE%290733-9399%281998%29124%3A3%28285%29.
Further work is required to outline the limits and applicability range for [15] M. Langerholc, M. Česnik, J. Slavič, M. Boltežar, Experimental validation of a com-
particle upscaling. plex, large-scale, rigid-body mechanism, Eng. Struct. 36 (Mar. 2012) 220–227,
http://www.sciencedirect.com/science/article/pii/S0141029611004962.
[16] S. Lommen, G. Lodewijks, D.L. Schott, Co-simulation framework of discrete element
method and multibody dynamics models, Eng. Comput. 35 (3) (2018) 1481–1499,
Acknowledgements https://doi.org/10.1108/EC-07-2017-0246.
[17] S. Lommen, D. Schott, G. Lodewijks, Multibody dynamics model of a scissors grab for
cosimulation with discrete element method, FME Trans. 40 (4) (2012) 177–180.
The authors would like to thank Nemag B.V., The Netherlands for
[18] S. Lommen, D. Schott, G. Lodewijks, Validation of a multibody dynamics model of a
their support in enabling this research. scissors grab for DEM modelling, Proceedings of the 20th International Conference
on Material Handling, Constructions and Logisitics. Belgrade, Serbia 2012,
References pp. 61–65 , Sep.
[19] S. Lommen, D. Schott, G. Lodewijks, DEM speedup: stiffness effects on behavior of
[1] M. Achmus, K. Abdel-Rahman, The influence of up-scaling on the results of particle bulk material, Particuology 12 (Feb. 2014) 107–112, http://www.sciencedirect.
method calculations of non-cohesive soils, Proceedings of 1st International PFC com/science/article/pii/S1674200113001387.
Symposium on Numerical Modelling in Micromechanics Via Particle Methods. [20] S.W. Lommen, Virtual Prototyping of Grabs: Co-Simulations of Discrete Element and
Balkema, Lisse, The Netherlands 2003, pp. 183–187. Rigid Body Models, http://resolver.tudelft.nl/uuid:418996d9-9b48-4349-823e-
[2] J. Ai, J. Chen, J. Rotter, J. Ooi, Assessment of rolling resistance models in discrete ele- b78d8349af25 2016.
ment simulations, Powder Technol. 206 (3) (Jan. 2010) 269–282. [21] Y. Mori, C.-Y. Wu, M. Sakai, Validation study on a scaling law model of the demin in-
[3] Z. Asaf, D. Rubinstein, I. Shmulevich, Determination of discrete element model pa- dustrial gas-solid flows, Powder Technol. 343 (2019) 101–112 , cited By 0. URL ,
rameters required for soil tillage, Soil Tillage Res. 92 (1–2) (Jan. 2007) 227–242, https://www.scopus.com/inward/record.uri?eid=2-s2.0-85056253424&doi=10.1016%
http://www.sciencedirect.com/science/article/B6TC6-4JVSWGD-1/2/3062571bb2c6 2fj.powtec.2018.11.015&partnerID=40&md5=a450607c677af424dbae59d694a728f0.
13af5821a10a1fef9a2a. [22] J. Park, W. Yoo, H. Park, Matching of flexible multibody dynamic simulation and ex-
[4] S.V. Baars, Discrete Element Analysis of Granular Materials, PhD Thesis Delft Univer- periment of a hydraulic excavator, Proceedings of the Second Asian Conference on
sity of Technology, Delft, The Netherlands, 1996, http://resolver.tudelft.nl/uuid: Multibody Dynamics. Seoul, Korea Aug. 2004, pp. 459–463.
bc642899-842e-4593-88ef-9aa1afcc5e41. [23] T. Poschel, C. Saluea, T. Schwager, Can we scale granular systems, 2018 in: Balkema
[5] C. Bierwisch, T. Kraft, H. Riedel, M. Moseler, Three-dimensional discrete element Sendai, T. Roessler, A. Katterfeld (Eds.),Scaling of the Angle of Repose Test and its In-
models for the granular statics and dynamics of powders in cavity filling, J. Mech. fluence on the Calibration of Dem Parameters using Upscaled Particles. Powder
Phys. Solids 57 (1) (Jan. 2009) 10–31, http://www.sciencedirect.com/science/arti- Technol, 330, 2001, pp. 58–66, http://www.sciencedirect.com/science/article/pii/
cle/pii/S0022509608001750. S0032591018300500.
[6] C.S. Bierwisch, Numerical Simulations of Granular Flow and Filling, PhD Thesis Uni- [24] M. Sakai, S. Koshizuka, Large-scale discrete element modeling in pneumatic convey-
versity of Freiburg, Freiburg, Sep. 2009, http://www.freidok.uni-freiburg.de/ ing, Chem. Eng. Sci. 64 (3) (Feb. 2009) 533–539, http://www.sciencedirect.com/sci-
volltexte/6498/. ence/article/pii/S0009250908005228.
[7] T. Brans, Simulation of a Four Rope Grab. Computeropdracht 2000.TT.5331, TU Delft, [25] D.L. Schott, S.W. Lommen, R. van Gils, J. de Lange, M.M. Kerklaan, O.M. Dessing, W.
Delft, 2000. Vreugdenhil, G. Lodewijks, Scaling of particles and equipment by experiments of
[8] P.W. Cleary, DEM prediction of industrial and geophysical particle flows, an excavation motion, Powder Technol. 278 (2015) 26–34.
Particuology 8 (2) (Apr. 2010) 106–118, http://www.sciencedirect.com/science/ar- [26] C. Wensrich, A. Katterfeld, Rolling friction as a technique for modelling
ticle/pii/S1674200109001308. particle shape in DEM, Powder Technol. 217 (Feb. 2012) 409–417, http://www.
[9] C. Coetzee, Particle upscaling: calibration and validation of the discrete element sciencedirect.com/science/article/pii/S0032591011006000.
method, Powder Technol. 344 (2019) 487–503, http://www.sciencedirect.com/sci- [27] J. Wittenburg, Dynamics of Multibody Systems, 2nd edition Springer, Berlin; New
ence/article/pii/S0032591018310593. York, Nov. 2007.
[10] S.M. Derakhshani, D.L. Schott, G. Lodewijks, Micro macro properties of quartz sand: [28] L. Xie, W. Zhong, H. Zhang, A. Yu, Y. Qian, Y. Situ, Wear Process During Granular Flow
experimental investigation and dem simulation, Powder Technol. 269 (2015) Transportation in Conveyor Transfer vol. 288 (2016) 65–75.
127–138, http://www.sciencedirect.com/science/article/pii/S0032591014007888. [29] W. Yoo, K. Kim, H. Kim, J. Sohn, Developments of multibody system dynamics: com-
[11] Y. Feng, K. Han, D. Owen, Discrete element modelling of large-scale systems-i: exact puter simulations and experiments, Multibody Syst. Dyn. 18 (1) (2007) 35–58.
scaling laws, Comput. Part. Mech. 1 (2) (2014). [30] Y. Zhou, B. Xu, A. Yu, P. Zulli, An experimental and numerical study of the angle of
[12] Y. Feng, K. Han, D. Owen, J. Loughran, On upscaling of discrete element models: sim- repose of coarse spheres, Powder Technol. 125 (1) (2002) 45–54, http://www.
ilarity principles, Eng. Comput. 26 (6) (2009) 599–609. sciencedirect.com/science/article/pii/S0032591001005204.