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

net/publication/263472745

Calibration of discrete element properties and the modelling of packed rock

beds

Article in Powder Technology · September 2014

DOI: 10.1016/j.powtec.2014.05.063

CITATIONS READS

25 431

2 authors, including:

C.J. Coetzee
Stellenbosch University
63 PUBLICATIONS 2,387 CITATIONS

SEE PROFILE

All content following this page was uploaded by C.J. Coetzee on 11 May 2021.

The user has requested enhancement of the downloaded file.

 Powder Technology 264 (2014) 332–342

Contents lists available at ScienceDirect

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

Calibration of discrete element properties and the modelling of packed

rock beds
C.J. Coetzee 1, R.G. Nel
Department of Mechanical and Mechatronic Engineering, University of Stellenbosch, Private Bag X1, Matieland 7602, South Africa

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

Article history: The increased necessity to obtain energy from other sources than conventional fossil fuels has led to the growing
Received 28 September 2013 interest in solar energy. The problem with the proposed technology is that it can only provide power during the
Received in revised form 26 May 2014 day and therefore requires some sort of storage system, if power is to be supplied throughout the day and night.
Accepted 31 May 2014
A number of storage systems exist, but the one of particular interest is packed rock beds. Discrete Element
Available online 8 June 2014
Models (DEM) of rock beds were developed through both experimental and numerical procedures, by
Keywords:
conducting a series of sensitivity and calibration studies. Through these procedures, a set of microproperties
Discrete Element Modelling (DEM) could be determined to accurately model the rock particles. The properties were verified by modelling the
Property calibration discharge from a hopper with three different opening widths. Further, the research focused on the potential of
Hopper discharge constructing self-supporting tunnels within the rock beds in order to improve the air flow through the bed by
Packed rock beds minimizing the pressure drop. Horizontal and vertical tunnels were investigated, each with different diameters.
Self-sustaining tunnels It was observed that if the appropriate steps were followed, stable self-supporting tunnels could be formed using
particles of different scale.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction
There are three main types of thermal energy storage methods
available: sensible, latent and thermochemical heat storage. Unlike
latent and thermochemical heat storage systems, where energy is
stored or extracted by means of a phase change material or a chemical
reaction respectively, sensible heat storage systems store or extract
energy by heating or cooling a solid or liquid substance. There are a
number of different substances which can be used, but some of the
most common are rock pebbles, refractory, water, molten salts and
oils. The choice of substance depends mostly on the temperature level
of the required application [1]. Packed rock beds are considered in
this paper.
A typical packed bed storage system can either be designed for a
horizontal or a vertical flow regime. In the case of a vertical flow regime,
the storage system consists of an insulated container, loosely packed
rock pebbles on top of a metal grate and an air inlet and outlet port
[2]. The flow direction during the charging stage is from top to bottom,
where the warm air is introduced at the top and the cold air is removed
from the bottom. The air will therefore transfer its heat to the rocks as it
is forced downwards and will exit at more or less the same temperature
as the rocks in the bottom layer of the bed. On the other hand, during
the discharging stage the cold air is introduced at the bottom of
the bed and the hot air exits at the top at approximately the same
E-mail address: ccoetzee@sun.ac.za (C.J. Coetzee).
1
Tel.: +27 21 808 4239; fax: +27 21 808 4958.
temperature as the rocks in the top layer [3]. These flow directions
ensure that the forced and natural convection currents aid one another,
such that a stable temperature distribution is achieved and that the
natural convection currents do not de-stratify the bed [4].
The most economical and efficient particle size to use for rock bed
storage systems is still subject to investigation, but according to studies
done by Sanderson and Cunningham [5] a particle diameter greater
than 13 mm should be used in order to avoid high pumping power
requirements due to excessive pressure losses. Similar results were
obtained by Torab and Beasley [6] who stated that a particle diameter
larger than 12.7 mm should be used. However, the particles cannot be
too large, since larger particles reduce the volumetric heat transfer
coefficient as well as the amount of useful energy stored in the beds
[7]. Another aspect which influences the amount of useful energy
being stored in packed beds is the bed length.
One of the disadvantages of a packed bed is the large pressure drop
over the bed. One method to improve the air flow distribution as well as
reduce the pressure drop is to make use of tunnels inside the bed. The
best position, size and orientation of these tunnels can be determined
experimentally or using computational fluid dynamics. The aim of this
study, however, is to determine if it is possible to form self-supporting
tunnels. With self-supporting it is implied that the tunnels would be
constructed entirely out of the rock particles, without any adhesives
or supporting structures. It would therefore be cost effective and not
be subject to mechanical degradation during operation. However,
this is based on the assumption that the rocks themselves will be able
to withstand the high thermal cycles without degrading. Allen [4]

http://dx.doi.org/10.1016/j.powtec.2014.05.063
0032-5910/© 2014 Elsevier B.V. All rights reserved.
 C.J. Coetzee, R.G. Nel / Powder Technology 264 (2014) 332–342
performed a preliminary thermal mechanical stress analysis on possible
rock candidates suitable for thermal storage applications. The rock
particles identified by Allen were used in this study [4].
Section 2 gives a brief overview of the Discrete Element Method
(DEM), the material calibration is presented in Section 3 and it includes
both experimental measurements and the numerical model. The cali-
brated model is verified in Section 4 by modelling material discharge
from a hopper while the packed rock bed with horizontal and vertical
tunnels is modelled in Section 5.
2. The Discrete Element Method
The Discrete Element Method (DEM) treats the interaction between
particles as a dynamic process and the movements of individual
particles are continuously traced [8]. These movements result due to
disturbances caused by particle motion and body forces as well as
disturbances caused by the geometric objects (walls) associated with
the system. These disturbances can then propagate through the particle
system and the rate of propagation depends on the physical properties
associated with the system.
In general the process is modelled in two stages. The first is a suitable
contact model, used to determine all the forces acting on each particle
within the system. A number of contact models have been developed
such as the Walton–Braun [9] and the Hertz–Mindlin [10] models,
which can model complex contact behaviour, but they are generally
computational intensive. A more basic contact model which is normally
the default contact model implemented in most of the commercially
available DEM software due to its efficiency, is the linear contact
model. In this study only the linear contact model was used for simplic-
ity and the calibration procedure used here was verified using this
model. The linear contact model can be considered as a soft particle
approach where the contact is modelled with a linear spring in both
the contact normal, kn, and the contact tangential direction, ks, (Fig. 1).
Particles in contact are thus allowed to overlap with one another and
the contact forces are calculated by combining the amount of overlap
R The bulk material selected for this study was coarse crushed
kn cn
ks
Friction µ
cs
Fig. 1. Linear contact model used in the Discrete Element Models.
333
between contacting particles with the spring stiffness values. The
second stage is then to apply Newton's second law of motion such
that the particle's acceleration can be calculated which can finally
be integrated to first update the particle's velocity and then its
position using the displacement increment [11].
In addition, frictional slip is also introduced in the tangential direc-
tion through a friction coefficient. Damping is accounted for by means
of a viscous damper in the normal and tangential contact directions
[10]. The walls follow a prescribed velocity and the equations of motion
are not applied.
3. Material calibration
DEM has been used by a number of researchers over the past few
decades, but according to Chung and Ooi [12], one crucial aspect
which has not adequately been addressed is how the input properties
were selected or determined such that satisfactory quantitative predic-
tions (bulk responses) could be produced. In many cases, the input
properties were assumed without the proper justification and often
not measured. It is therefore crucial that the correct input properties
are selected in order to ensure that the simulated bulk responses closely
resemble the real life physical system. According to Curry et al. [13], this
can be achieved by correctly defining the material properties affecting
the material behaviour at the particulate level (microproperties). How-
ever, it is not always possible to directly measure these microproperties
and even if in some cases it could be done, there is no guarantee
that they would provide accurate bulk behaviour when applied in the
simulations. A calibration process is therefore required, which can be
performed through an iterative approach where each microproperty is
varied individually until a specified or measured bulk response is
obtained. This process is then repeated until all the required bulk
responses can be simulated within acceptable accuracy levels [14]. The
calibration process followed here is based on that used by Coetzee and
Els [15].
3.1. Experimental measurements
3.1.1. Particle size and shape
aggregate with an average equivalent diameter of 32 mm (based on
volume) obtained from the Portland quarry located in the Western
Cape of South Africa. Fifty random rock particle samples were collected
from the bulk material and subdivided into four distinct shape classes
by means of visual inspection, similar to the approach used by Latham
et al. [16]. The four classes (Fig. 2) can be described as follows:
• Equant: particles which were close to cubical.
• Long: particles which had two orthogonal axes that were relatively
similar in magnitude and one that was much larger.
• Flat: particles which had one orthogonal axis that was much shorter
compared to the other two.
• Intermediate: all the particles which did not resemble any of the
above descriptions.
To obtain the particle volume, the fifty particles were scanned with a
NextEngine3D laser scanner [17]. The 3D image processing software
ScanStudio HD was then used to automatically construct the final 3D
images from the individual scans, after which they were imported into
Rapidworks [17] where the geometry could be meshed with tetrahedra.
The volume of all the individual tetrahedra within the mesh could then
be summed to give the total volume of the scanned particle [16]. The
proposed shape and size distributions are summarised in Table 1.
3.1.2. Particle density, bulk density and porosity
The particles were weighed using an electronic scale (accuracy
of 0.1 g) and together with the volume data, the average particle density
could be determined as 2610 kg/m3. The bulk density was measured by
334 C.J. Coetzee, R.G. Nel / Powder Technology 264 (2014) 332–342
Fig. 2. The four identified classes used to describe the particle shape.
filling a known volume (300 × 300 × 300 mm) with material and simply
finding the weight of the material. The measurement was repeated
three times and the average bulk density was found to be 1357 kg/m3.
The average porosity was found to be 48.0%.
3.1.3. The angle of repose
The dependence on various microproperties such as friction, the
particle shape and size distribution as well as the way in which the
angle of repose test is executed, has led to numerous debates among
researchers on the usefulness of the angle of repose test as a flow
property [18]. Due to all these factors it is very challenging to accu-
rately define and therefore interpret the results from such tests
[19]. However, because it is relatively simple and inexpensive to ob-
tain the angle of repose, it is a useful test to obtain a basic under-
standing of the behaviour of a bulk material sample with regard to
flow or frictional characteristics.
The angle of repose was measured by removing the four sides of the
300 × 300 × 300 mm container one by one with the bottom of the
container away from the ground so that particles could slide down
and form a free-standing pile. The angle of repose was measured using
a digital inclinometer that resulted in an average angle of 35.6°. Digital
photographs of the slope were also taken from various angles and
image processing tools in Matlab used to determine the angle of repose.
This method resulted in an average angle of 36.0°.
3.1.4. Bulk stiffness
A confined compression test was conducted. The setup consisted out
of a cylindrical steel container with a diameter of 590 mm, a height of
370 mm, and a cylindrical lid with a diameter of 580 mm and a lever
mechanism to apply a normal load to the lid through a 50 kN HBM S9
load cell. The displacement was measured using a HBM 200 mm linear
variable differential transformer (LVDT).
The container was filled by pouring the material from a fixed height,
after which the sample was carefully levelled such that the initial bulk
height corresponded to the height of the container wall. The lid was
added and the normal load was then increased incrementally by adding
Table 1
Particle shape, size and equivalent radius.
Particle shape % of sample Rmin (mm) Rmax (mm)
Equant 16 10.4 17.0
Long 20 13.9 21.2
Flat 24 9.8 19.3
Intermediate 40 10.4 21.3
10 kg weights to the lever arm until a total of 50 kg was reached, each
time recording the normal load as well as the displacement resulting
from the compression. The process was then reversed by removing
all the weights in 10 kg increments, again recording the normal load
and displacement. This represented one compression cycle and was
repeated three times to complete the confined compression test.
A typical load-displacement graph is shown in Fig. 3. The slope of the
second and third loading cycle was measured and the average defined
as the bulk stiffness. The entire experiment was repeated three times
and the average bulk stiffness was found to be 1.44 MN/m.
3.1.5. Particle–wall friction coefficient
The particle–wall friction coefficient can be determined from what
is known as a sliding friction tester [20]. In such a test, several of the
granular particles are glued to a flat base plate and then placed onto a
test plate such that the glued particles are between the two plates.
The test plate is then slowly lifted at one end, while the opposite end
is allowed to rotate until the base plate with the glued particles start
to slide. The inclination angle, θ, is measured. The particle to wall friction
coefficient can then be estimated using μp = tan θ. For this study a
coated mild steel plate was used and several tests were conducted to
obtain an average particle–wall friction coefficient of 0.62.
3.2. Numerical model
The simulations were performed with the DEM package, PFC3D [10].
At the start of the calibration process an initial set of material properties,
based on the available experimental data and material properties from
literature, was used. A more detailed calibration procedure was then
used to fine tune the material properties one at a time by comparing the
simulated bulk material behaviour to the experimental measurements.
The particle density was set equal to the measured density of
2610 kg/m3. Horn [14] measured the angle of repose of a specific
crushed aggregate to be 40°. If it assumed that the angle of repose is
an indication of the internal friction angle [21] and thus the particle–
particle friction coefficient, the particle–particle friction coefficient in
Horn's [14] case should be in the order of tan(40 °) = 0.84. This was
close to the value of 0.86 that he obtained through shear box calibration.
Thus, in our case the initial estimate of the particle–particle friction
coefficient was set to tan(36.0 °) = 0.73. The particle–wall friction
Rave (mm) coefficient was set equal to the measured value of 0.62. Based on the
13.2 average coefficient of restitution, COR = 0.85, calculated from a range
15.6 of rock particles studied by Imre et al. [22], the critical damping ratio
16.3 was estimated to be 0.05 based on the relation between the critical
16.1
damping ratio and the coefficient of restitution [10]. The particle
 C.J. Coetzee, R.G. Nel / Powder Technology 264 (2014) 332–342 335

Fig. 3. A typical stress–strain load curve from the axial compression test.

stiffness was initially set to 1.72 MN/m based on the value used
by Horn [14] for a crushed aggregate. The initial microproperties
are summarised in Table 2 and the resulting bulk properties
are summarised in Table 3 (Initial estimate). Comparing the bulk
properties to the measured properties, the porosity and the angle
of repose were accurate within 10% but the bulk density and stiffness
were not accurate.
3.2.1. Particle size and shape
PFC3D can only model spherical particles. If one is therefore interested
in irregular shaped particles, such as the crushed aggregate used here,
several spherical particles have to be clustered or clumped together in
order to create the desired shape [23]. The spheres comprised in these
clumps can overlap one another without generating additional contact
forces and their centres remain at fixed relative locations [10]. Even
though research on the validity of the clumped approach is very limited,
it is arguably the most efficient and robust method to model irregular
shaped particles with DEM, since the contact detection and force
calculation are still based on simple algorithms [23]. With the particle
shape modelled as accurately as possible, the implementation of rolling
resistance was deemed unnecessary.
Clumps were implemented with the aid of a commercially available
automatic clump generator called ASG3D [24]. The surface mesh of each
of the scanned particles was imported into ASG3D. The automatic clump
generator would then randomly fill the surface mesh with a user
defined number of spherical particles, after which the radii and centre
positions of all the spheres would be optimised in order to minimise
the distance between mesh vertices and the surface of their nearest
sphere, [24] as shown in Fig. 4.
The volume error is defined as the percentage error in the clumped
volume compared to the mesh (particle) volume. Even though the
volume error can be decreased by increasing the number of spheres,
this would have a negative impact on the DEM computational time.
Care should therefore be taken when deciding on the accuracy to
which the clumps should be reconstructed. With this in mind each
sample was reconstructed with 10, 35 and 75 spheres respectively
which resulted in the volume errors shown in Fig. 5. It is clear that
the respective volume errors decreased significantly as the number of
spheres per clump was increased. The most noticeable decrease
corresponded to the clump representing the flat particles, simply
because the minor axis was much smaller compared to the other two
axes and the spheres had to become relatively large in order to
completely fill the surface mesh.
It has been suggested by Chung and Ooi [20] and Grima and
Wypych [25] that one can get away with a rather crude geometrical
representation of the granular material and still obtain satisfactory
results, since a very accurate representation does not necessarily
lead to more accurate predictions of the bulk behaviour. This was
confirmed by investigating the effect of the clump accuracy on the
various bulk responses by repeating the simulations with a varying
number of spheres per clump: 10, 35 and 75. All the particle (sphere)
microproperties were kept the same and the results showed that
the clump accuracy had no significant effect on the bulk stiffness
(2.15, 2.08 and 2.34 MN/m respectively), the angle of repose
(38.4°, 38.4° and 37.6° respectively) or the porosity (51.8%, 50.7%
and 50.0% respectively).
However, with an increase in the number of spheres per clump,
there was a significant increase in the bulk density (2117.1, 3228.8
and 4290.8 kg/m3, respectively). The reason for the large change in
bulk density was that the sphere density was not adjusted and with
more sphere–sphere overlaps, the weight of the clumped particle
increased. The correct procedure would be to adjust the sphere density
so that the clumped particle weighs the same as the physical particle.
However, the sphere–sphere overlap was taken into account for the
volume calculations and had no influence on the porosity calculations.

Table 3
Table 2
Bulk properties from measurements, initial DEM estimate and calibrated DEM.
Particle microproperties used in the initial estimate and values after calibration.
Bulk property Measured value Initial estimate Final calibrated
Microproperty Initial estimate Final calibrated
Value Error Value Error
Particle–particle friction 0.73 0.9
Particle–wall friction 0.62 0.85 Bulk density (kg/m3) 1357 2117 56.0% 1311 3.4%
Particle density (kg/m3) 2610 1610 Porosity (%) 48.0 52.2 8.8% 46.8 2.5%
Particle stiffness (MN/m) 1.72 0.62 Angle of repose (°) 36.0 38.7 7.5% 35.6 1.1%
Critical damping ratio 0.05 0.23 Bulk stiffness (MN/m) 1.44 3.10 115.3% 1.56 8.3%
336 C.J. Coetzee, R.G. Nel / Powder Technology 264 (2014) 332–342

Fig. 4. Examples of the scanned particles and the clump representation using 10, 35 and 75 spheres, respectively.

With the above and computational times taken into account,
it was decided to make use of 10 spheres per clump in all subsequent
simulations.
3.2.2. Particle density and bulk density
A container, with the same dimensions of that used in the
experiment, was filled with the clumped particles and the particle
density varied from 500 kg/m 3 to 5000 kg/m 3 in five increments.
The relation between the particle density and the bulk density
was found to be linear. Using linear interpolation, a particle density
of 1610 kg/m3 was selected and this resulted in a bulk density of
1286 kg/m3.
Note that the measured particle density was much higher at
2610 kg/m3. In the model, a lower particle density will always be
used due to the sphere–sphere overlap within the clump.
3.2.3. Particle–particle friction
The experimental angle of repose test was duplicated numerically
and the particle friction coefficient varied from 0.1 to 1.0. The relation
between the particle friction coefficient and the angle of repose is
shown in Fig. 6. This response was similar to the response obtained by
[15,26]. Based on these results, a value of 0.9 was assigned to the particle
friction coefficient. This resulted in an angle of repose of 37.2°, a bulk
density of 1283 kg/m3, a porosity of 47.5% and a bulk stiffness of
2.733 MN/m.
3.2.4. Damping ratio
The calibration of the density and the particle–particle friction
(angle of repose) was repeated for five critical damping ratios ranging
between 0.0 and 1.0. Since these processes were all quasi-static in

Fig. 5. The average clump volume error using 10, 35 and 75 spheres per clump for each of the identified shapes.
 C.J. Coetzee, R.G. Nel / Powder Technology 264 (2014) 332–342 337

Fig. 6. The effect of the particle friction coefficient on the angle of repose.

nature, the damping had little influence on the results. However,
from the results the critical damping ratio was set to 0.23. This resulted
in an angle of repose of 36.9°, a bulk density of 1313.3 kg/m3, and a
porosity of 46.5%.
3.2.5. Particle–wall friction
The calibration of the density and the particle–particle friction
(angle of repose) was repeated for five particle–wall friction coefficients
between 0.2 and 1.0. The particle–wall friction coefficient had a small
influence on the porosity and the angle of repose. Based on these results
the particle–wall friction coefficient was set to 0.85. This resulted in an
angle of repose of 36.1°, a bulk density of 1318.7 kg/m3, and a porosity
of 47.3%.
experimental results. The experimental hopper had a rectangular
outlet with two horizontal trapdoors, which could be pulled apart
in order to control the outlet size (Fig. 7).
The hopper was filled with approximately 122 kg of the rock particles.
The hopper was then suspended 1 m above the ground using a 50 kN
HBM S9 load cell. Three outlet sizes were considered by manually pulling
the trapdoors 0.15 m, 0.25 m and 0.40 m apart. The weight of the material
inside the hopper was plotted versus time and the numerical and
experimental results for an outlet opening of 0.15 m are shown
in Fig. 8a. It can be seen that the calibrated properties resulted in an

3.2.6. Particle stiffness

The particle stiffness was varied between 0.17 MN/m and 9.0 MN/m
and used in a model of the confined compression test. The relation
between the particle stiffness and the bulk stiffness was found to be
linear. Coetzee and Els [15] obtained similar results. Based on the re-
sults, the particle stiffness was set to 0.62 MN/m, resulting in a bulk
stiffness of 1.56 MN/m.

3.2.7. Summary of material properties

The final set of material microproperties after one iteration was
summarised in Table 2 with the resulting bulk responses summarised
in Table 3. One could repeat the process several times until the cali-
brated properties converge. However, this would require numerous
additional simulations and since the relative errors for all the bulk
responses (except for the stiffness) were under 3.5%, the calibration
process was deemed finished. The modelled bulk stiffness had the
highest error of 8.3%. However, from a sensitivity study it could
be shown that the bulk behaviour modelled in this study (hopper
discharge and static packed rock bed) is not very sensitive to the
particle stiffness. According to Chung and Ooi [12], one can reduce
the particle stiffness in order to gain computational advantage,
while still obtaining accurate and reliable bulk responses. The
calibrated microproperties summarized in Table 2 were used in all
the subsequent simulations.

4. Model validation

In order to validate the calibrated properties, material discharge

from a hopper was modelled and the results were compared to Fig. 7. The hopper used in model validation.
338 C.J. Coetzee, R.G. Nel / Powder Technology 264 (2014) 332–342

Hopper aer 4.5 seconds

(a)

Hopper aer 4.5 seconds

(b)

Hopper aer 4.5 seconds

(c)

Fig. 8. The hopper discharge curves for openings of (a) 150 mm, (b) 250 mm and (c) 400 mm. The insets show the material remaining behind in the hopper after 4.5 s.
 C.J. Coetzee, R.G. Nel / Powder Technology 264 (2014) 332–342
accurate discharge rate. It should be noted that the hopper did not
fully discharge, since there was some material build up against the
inside walls.
The numerical result for an outlet opening of 0.25 m also correlated
well with the experimental results, as shown in Fig. 8b and similar to the
0.15 m case, the hopper did not fully discharge due to material build up
at the inside walls. Finally, for an outlet opening of 0.4 m the numerical
results correlated well with the experimental results as seen in Fig. 8c. In
this case, the doors were fully open and no material build up occurred.
Even though there were some local variations between the numeri-
cal and experimental results, the overall discharge rates for all the
outlet openings were accurate. Thus, from a practical perspective the
numerical models provided satisfactory results and the selected set of
microproperties could therefore be implemented to investigate packed
rock beds.
5. Modelling of packed rock beds
This section focuses on the construction of self-supporting tunnels
within packed rock beds to potentially improve the flow characteristics
and high pressure drop factors commonly associated with such beds.
Two very simple tunnel construction approaches were adopted. The
first was to place a tube with a size corresponding to the desired tunnel
geometry at the required location within the containment vessel, after
which the rock particles were introduced and allowed to settle under
gravity. The idea was that the tube would induce the formation of stable
arches in the surrounding area, which would act as a naturally formed
support structure once the tube was removed. The second approach
resembled an actual excavation process where the particles were
removed from the bed with the aim to induce the formation of
stable tunnels.
Only horizontal and vertical tunnels were considered and in each
case, three bed layouts were used: the real particle bed, a scaled
particle bed and a layered/annular particle bed. In all cases, the
same 300 × 300 × 300 mm container was used. Where the tube was
Tube removal process
Excavaon process
Fig. 9. Steps followed to model and study the horizontal self-supporting tunnels using the tube removal process and the excavation process.
339
used, its centre was placed 100 mm from the bottom and it was
removed at a speed of 0.5 m/s (Fig. 9). The status of the resulting self-
supporting tunnel was recorded in terms of the visibility through the
tunnel. The state of visibility was based on visual judgement and
could be categorized as either full visibility, partial visibility or zero vis-
ibility (Fig. 9). Three reference tubes were considered with diameters
corresponding to 60 mm, 30 mm (equal to the average hydraulic
particle diameter) and 15 mm. For each tube five different rock beds
were modelled to ensure repeatability.
In the second tunnelling approach particles with their centre
points falling within a specified excavation zone were removed.
The rectangular excavation zone had a height of 100 mm, stretching
upwards from the bottom of the container and an initial width of
15 mm. The state of the excavated tunnel was then recorded, after
which the width of the excavation zone was increased by 15 mm
and the process repeated for several more steps. The height of the
excavation zone was kept constant (Fig. 9).
5.1. Results from horizontal tunnels
5.1.1. Results from real particle bed
In this model, the particles had the same size distribution as the
physical particles.
5.1.1.1. Tube removal. Once the tube was removed it was observed
that in most instances the resulting tunnels corresponded to a state of
zero visibility with only one or two tunnels being partially visible. The
location at which the particles collapsed also varied, with no noticeable
trend. However the number of collapsed particles did seem to increase
slightly as the tube diameter increased. The collapsed particles were
then removed, after which it was observed that for some instances the
resulting tunnels were still stable, whereas others completely collapsed.
Furthermore, no distinction could be made between the different tube
diameters considered.
340 C.J. Coetzee, R.G. Nel / Powder Technology 264 (2014) 332–342

Diameter = 15 mm Diameter = 30 mm Diameter = 60 mm

Inial condion Tube removed Inial condion Tube removed Inial condion Tube removed

Fig. 10. Horizontal tunnels resulting from the removal of the tubes from the scaled particle bed.

Fig. 11. Horizontal tunnels resulting from the consecutive steps of the excavation process for the scaled particle bed.

5.1.1.2. Excavation process. During the excavation process it was
observed that for all the beds modelled the excavated tunnels would
simply collapse. There were one or two instances where partial
segments of an excavated tunnel remained, but these tunnels were
very unstable and occurred at random instances during the excavation
process. It can therefore be concluded that the proposed tunnelling
method would not result in stable self-supporting tunnels suitable for
practical applications.
5.1.2. Results from scaled particle bed
In this model the particle size, across the size distribution, was in-
creased by a factor of 3.2.
5.1.2.1. Tube removal. It was observed that there were tunnels which
were partially visible or in one or two cases fully visible (Fig. 10).
However, most of the tunnels, specifically the ones constructed
with the two larger tube diameters, still resulted in a state of zero
visibility. The number of collapsed particles increased as the tube
diameter increased, however, relatively fewer particles collapsed as
compared to the real particle bed.
5.1.2.2. Excavation process. During the excavation process (Fig. 11), it
was observed that a significant number of stable tunnels were obtained.
However, it should be noted that some of these tunnels only formed
after a substantial number of collapsed particles were removed, which
was clearly visible in the reduction in the bed's height.
5.1.3. Results from layered particle bed
In this model, the bed consisted of three different particle layers,
each with an approximate height of 100 mm. In the bottom and top
layer, the real particle size was used while in the middle layer, the scaled
particles were used (Fig. 12). This particle arrangement was selected
specifically with the aim that some of the smaller particles in the top
layers would fall into the void spaces between the larger particles,
potentially increasing the particle interlocking. In theory, the increased
interlocking would then significantly reduce the number of loose
particles associated with the middle rock layer. The investigation
could be conducted by placing the reference tubes directly beneath
the middle layer, during the construction of the bed, whereas the
excavation process could simply be repeated as for the previous cases.
5.1.3.1. Tube removal. The results showed a significant improvement and
were more consistent (Fig. 13). The tunnels resulting from the 15 mm
tubes for instance were all fully visible, while most of the remainder
of the tunnels were either fully visible or partially visible, with only a
few cases of zero visibility. Furthermore, the collapsed particles typically
corresponded to some of the smaller particles from the upper layer,
which suggests that an insufficient number of large particles were

100 mm

100 mm

100 mm

Fig. 12. Particle layers used to improve conditions for the formation of self-supporting horizontal tunnels.
 C.J. Coetzee, R.G. Nel / Powder Technology 264 (2014) 332–342 341

Diameter = 15 mm Diameter = 30 mm Diameter = 60 mm

Inial condion Tube removed Inial condion Tube removed Inial condion Tube removed

Fig. 13. Horizontal tunnels resulting from the removal of the tubes from the layered particle bed.

Fig. 14. Horizontal tunnels resulting from the consecutive steps of the excavation process for the layered particle bed. In step 5 the tunnel completely collapsed.

used to form the second layer. Care should therefore be taken during the
bed construction, to ensure that the smaller rock particles fall effectively
between the larger particles, without reaching the tubes. The initially
collapsed particles were removed after which it was observed that all
the tunnels remained stable.
5.1.3.2. Excavation process. During the excavation process stable tunnels
formed as shown in Fig. 14 (steps 2 to 4) and there was also a noticeable
consistency in the shape of the tunnels.
5.2. Results from vertical tunnels
The tubes were placed at the centre of the container, perpendicular
to the container floor. The excavation process was not used for the
annular particle bed due to the bed layout.
5.2.1. Results from real and scaled particle bed
Similar to the horizontal tunnels, it was observed that most of the
tunnels using the real particle bed and the scaled particle bed were
either partially visible or they corresponded to a state of zero visibility,
with only one or two tunnels being fully visible. The location of the
collapsed particles varied between the bottom, middle and top section
of the tunnels, while the number of collapsed particles seemed to
increase as the tube diameter increased.
5.2.2. Results from annular particle bed
This bed was constructed by allowing the large particles to settle at the
centre of the container within an annulus created by two tubes, while the
smaller particles were allowed to settled around the outside of the annulus
(Fig. 15). The inner diameter corresponded to the desired tube diameter,
and the outer diameter was selected accordingly to ensure that a sufficient
number of large particles could settle around the tube. The outer tube
was then removed by pulling it vertically upwards to allow the smaller
particles to displace into the void spaces between the larger particles.
The results showed a significant improvement (over the real and
scaled particle beds) and were more consistent (Fig. 16). Most of the
resulting tunnels were fully visible, with only a few being partially
visible. In addition, simulations were also repeated for tube diame-
ters of 90 mm and 120 mm (Fig. 17). It should be noted that for
practical reasons the container dimensions had to be increased
(400 × 400 × 400 mm) to be able to perform the simulations. During
this process it was observed that relatively stable tunnels could be
obtained from both the 90 mm and 120 mm diameter tubes and
only small number of particles collapsed.
6. Conclusions
A calibration process was used to determine the DEM microproperties.
The material bulk density, porosity, angle of repose and the bulk stiffness

Fig. 15. Annular arrangement of particles used to improve conditions for the formation of self-supporting vertical tunnels.
342 C.J. Coetzee, R.G. Nel / Powder Technology 264 (2014) 332–342
Diameter = 15 mm Diameter = 30 mm
Inial condion Tube removed Inial condion
Fig. 16. Vertical tunnels resulting from the removal of the tubes from the annular particle bed. The top view is showed.
Diameter = 15 mm Diameter = 30 mm Diameter = 60 mm
Fig. 17. Vertical tunnels resulting from the removal of the tubes from the annular particle bed. This is a different bed from the one used in Fig. 16 and the tube diameters go up to 120 mm.
could be accurately modelled. The accuracy of the properties was further
verified by comparing the discharge from a hopper to experimental
results. It was shown that the DEM model could accurately predict the
average discharge rate for three different hopper openings. In the future,
aspects such as the inclusion of rolling resistance and the use of other
contact models should be investigated.
The calibrated properties were then used in DEM models of packed
rock beds. The potential of constructing self-supporting tunnels of
different diameter within the rock bed was investigated. Horizontal
and vertical tunnels were modelled. Attempts to create self-supporting
tunnels using particles of the same size scale proved to be ineffective
as most of the tunnels would collapse. However, when the bed was con-
structed of layered particles, using a different size scale in each layer,
most of the tunnels were stable. The construction process, either by
removing an embedded tube or by an excavation process, had little
effect on the results.
It is concluded that if particles of different size scale are used,
horizontal and vertical self-supporting tunnels can be formed in
rock beds. Temperature changes will result in expansion and con-
traction of the rock particles which may influence the stability of
the tunnels. Vibration, such as seismic activity, may also result in
tunnel collapse and needs to be investigated.
Acknowledgements
The authors want to thank the Solar Thermal Energy Research Group
(STERG), Department of Mechanical and Mechatronic Engineering,
Stellenbosch University, for financial and other support.
References
[1] O.E. Ataer, Storage of thermal energy, in: Y.A. Gogus (Ed.), Energy Storage Systems,
EOLSS Publishers, Oxford, UK, 2006.
[2] SRCC, Solar Design Manual, Available at www.solar-rating.org 2012.
[3] H. Garg, S. Mullick, A. Bhargava, Solar Thermal Energy Storage, vol. 1D. Reidel
Publishing Company, Dordrecht, Holland, 1985.
[4] K. Allen, Performance Characteristics of Packed Bed Thermal Energy Storage for
Solar Thermal Power Plants, (Master's thesis) University of Stellenbosch, 2010.
View publication stats
Diameter = 60 mm
Tube removed Inial condion Tube removed
Diameter = 90 mm Diameter = 120 mm
[5] T. Sanderson, G. Cunningham, Packed bed thermal storage systems, Appl. Energy 51
(1995) 51–67.
[6] H. Torab, D. Beasley, Optimization of a packed bed thermal energy storage unit, J.
Sol. Energy Eng. 109 (1987).
[7] T. Sanderson, G. Cunningham, Performance and efficient design of packed bed
thermal storage systems: part 1, Appl. Energy 50 (1995) 119–132.
[8] P. Cundall, O. Strack, A discrete numerical model for granular assemblies,
Geotechnique 29 (1979) 47–65.
[9] O.R. Walton, R.L. Braun, Viscosity, granular-temperature, and stress calculations for
shearing assemblies of inelastic, frictional disks, J. Rheol. 30 (1986) 949–980.
[10] Itasca, PFC3D, Particle Flow Code in 3 Dimensions, User's Guide, 4th edition, 2008.
(www.itascacg.com).
[11] K. Hanley, C. O'Sullivan, J. Oliveira, K. Cronin, E. Byrne, Application of taguchi
methods to DEM calibration of bonded agglomerates, Powder Technol. 210 (2011)
230–240.
[12] Y. Chung, J. Ooi, Influence of discrete element model parameters on bulk
behavior of a granular solid under confined compression, Part. Sci. Technol. 26
(2007) 83–96.
[13] D. Curry, J. Favier, R.D. LaRoche, A Systematic Approach to DEM Material Model
Calibration, The American Institute of Chemical Engineers (AIChE), Tennessee:
Nashville, 2009.
[14] E. Horn, The Calibration of Material Properties for Use in Discrete Element Models,
(Master's thesis) Stellenbosch University, 2011.
[15] C. Coetzee, D. Els, Calibration of discrete element parameters and the modelling of
silo discharge and bucket filling, Comput. Electron. Agric. 65 (2009) 198–212.
[16] J. Latham, A. Munjiza, X. Garcia, J. Xiang, R. Guises, Three-dimensional particle
shape acquisition and use of shape library for DEM and FEM/DEM simulation,
Miner. Eng. 21 (2008) 797–805.
[17] NextEngine, 3D Scanner HD, Available at www.nextengine.com 2000.
[18] Y. Zhou, B. Xu, A. Yu, P. Zulli, An experimental and numerical study of the angle of
repose of coarse spheres, Powder Technol. 125 (2002) 45–54.
[19] K. Ileleji, B. Zhou, The angle of repose of bulk corn stover particles, Powder Technol.
187 (2008) 110–118.
[20] Y. Chung, J. Ooi, Confined compression and rod penetration of a dense granular
medium: discrete element modelling and validation, Mod. Trends Geomech. 106
(2006) 223–239.
[21] T.W. Lambe, R.V. Whitman, Soil Mechanics, John Wiley & Sons, Inc., New York, 1969.
[22] H. Kruggel-Emden, S. Rickelt, S. Wirtz, V. Scherer, A study on the validity
of the multi-sphere discrete element method, Powder Technol. 188 (2008)
153–165.
[23] Cogency, ASG Software, Available at www.cogency.co.za.
[24] J. Hartl, J. Ooi, Experiments and simulations of direct shear tests: porosity, contact
friction and bulk friction, Granul. Matter 10 (2008) 263–271.
[25] A.P. Grima, P.W. Wypych, Discrete element simulations of granular pile
formation: method for calibrating discrete element models, Eng. Comput. 28
(2011) 314–339.
[26] B. Imre, S. Räbsamen, S. Springman, A coefficient of restitution of rock materials,
Comput. Biomed. Res. Geosci. 34 (2008) 339–350.