The current issue and full text archive of this journal is available at

www.emeraldinsight.com/0264-4401.htm

EC
26,6
Industrial particle flow modelling
using discrete element method
Paul W. Cleary
CSIRO Mathematical and Information Sciences, Clayton South, Australia
698 Abstract
Purpose – The purpose of this paper is to show how particle scale simulation of industrial particle
flows using DEM (discrete element method) offers the opportunity for better understanding of the
flow dynamics leading to improvements in equipment design and operation.
Design/methodology/approach – The paper explores the breadth of industrial applications that
are now possible with a series of case studies.
Findings – The paper finds that the inclusion of cohesion, coupling to other physics such fluids, and
its use in bubbly and reacting flows are becoming increasingly viable. Challenges remain in
developing models that balance the depth of the physics with the computational expense that is
affordable and in the development of measurement and characterization processes to provide this
expanding array of input data required. Steadily increasing computer power has seen model sizes
grow from thousands of particles to many millions over the last decade, which steadily increases the
range of applications that can be modelled and the complexity of the physics that can be well
represented.
Originality/value – The paper shows how better understanding of the flow dynamics leading to
improvements in equipment design and operation can potentially lead to large increases in equipment
and process efficiency, throughput and/or product quality. Industrial applications can be
characterised as large, involving complex particulate behaviour in typically complex geometries.
The critical importance of particle shape on the behaviour of granular systems is demonstrated.
Shape needs to be adequately represented in order to obtain quantitative predictive accuracy for
these systems.
Keywords Particle size measurement, Fluid dynamics, Flow
Paper type Research paper

1. Introduction
Discrete element method (DEM) has been developed and used over the past 30 years
for modelling many applications, starting with small systems in simple geometries in
two dimensions, such as periodic cells and chute flows, small hoppers and shear cells
(Cundall and Strack, 1979; Walton, 1994; Campbell, 1990; Haff and Werner, 1986). This
lead to early industrial applications that can be characterised by small scales
(100-1,000s of particles) with idealised assumptions and representation in two

The author wishes to express his deep appreciation to all the past and present members of the
group who have contributed to the development of the DEM modelling over the past 15 years.
In particular, the author would like to thank Dr Matt Sinnott for geometry construction,
visualisations and general assistance with this work over many years, Nick Stokes for his work
on non-spherical contact detection, Mahesh Prakash for contributions to the SPH modelling,
Phil Owen for assistance in reviewing the literature and Mark Sawley. The author would also
Engineering Computations: like to thank the many collaborators whose interactions have helped improve the quality and
International Journal for Computer- realism of the industrial models studied. In particular, the author would like to thank Rob
Aided Engineering and Software
Vol. 26 No. 6, 2009 Morrison, Bill McBride, Madeleine Golding Matt Abel, Reuben Domike, Charles Cooney, Soon
pp. 698-743 Hyoung Pyo and Bruno Laurent. Financial support by the CRC for Sustainable Resource
# Emerald Group Publishing Limited
0264-4401
Processing for the banana screen modelling and some of the comminution modelling is
DOI 10.1108/02644400910975487 recognised.
dimensions. Examples include ball mills (Mishra and Rajamani, 1992, 1994) and Industrial
hoppers by Langston et al. (1995), Potopov and Campbell (1996a) and others. Similarly, particle flow
geophysical modelling using DEM started with small scales and idealised particles
(Hopkins et al., 1991; Cleary and Campbell, 1993). modelling
Through the 1990s, the problem sizes tackled increased into the 10,000-100,000 range,
but models generally were either two dimensional (Cleary, 1998a, b, c, 2000; Campbell
et al., 1995; Holst et al., 1999; Potapov and Campbell, 1996a; P}oschel and Buchholtz, 1995;
Thornton et al., 1996; Ristow, 1994 and many others) or three dimensional but with very
699
simple geometries (for example fracture by Potapov and Campbell, 1996b). More recently,
DEM has been able to be used for industrial applications in complex 3D geometries.
Mixing of particulates have been modelled by Kuo et al. (2002); Lemieux et al. (2007,
2008); Stewart et al. (2001a, b); Zhou et al. (2003); Cleary (2004); Kuo et al. (2004); Bertrand
et al. (2005); and Cleary and Sinnott (2008). Comminution processes have been modelled
by Rajamani and Mishra (1996); Cleary and Hoyer (2000); Cleary (2001a, b); Morrison
et al. (2001); Herbst and Nordell (2001); Cleary and Sawley (2002); Hlungwani et al. (2003);
and Morrison and Cleary (2004, 2008). Validation of DEM predictions for semi-
autogenous grinding (SAG) mills has appeared in Cleary et al. (2003) and Bwalya and
Moys (2003). Separations by screens have been modelled by Li et al. (2002, 2003) and
Cleary and Sawley (2002). Realistic excavation using dragline buckets were presented in
Cleary (2004). Hopper emptying has been explored by Potopov and Campbell (1996a, b),
Holst et al. (1999), Cleary and Sawley (2002), Langston et al. (2004), Cleary (2004),
Ketterhagen et al. (2007) and many others. Conveyor transfer chutes have been analysed
using DEM by Zhang and Vu-Quoc (2000), Dewicki and Mustoe (2002), Dewicki (2003)
and Gröger and Katterfeld (2006). Many other applications have also been modelled as
DEM usage has undergone explosive growth since 2000.
DEM has now progressed to the point where large scale industrial and geophysical
systems can be modelled with increasing realism. Quantitative prediction accuracy is
now feasible for dry cohesionless granular flows when the particle shape and
boundary geometry are well represented and realistic material properties are used.
The breadth of particulate applications that researchers have considered has led to
the inclusion of models for many other types of physics. Coupling with a gas phase
started with pioneering work on pneumatic conveying (Tsuji et al., 1992) and then gas-
particulate fluidised beds (Tsuji et al., 1993; Xu and Yu, 1997; Hoomans et al., 1996;
Kafui et al., 2002). Since then many researchers have used these coupled DEM-FV
methods to model gas driven particle flows. This has more recently been extended to
fluid-particulate systems using SPH (Cleary et al., 2007a) for systems where the fluid
exhibits complex free surface behaviour.
DEM has also been extended to include cohesive forces with simple pendular liquid
bridge models (Lian et al., 1993, 1998; Mikami et al., 1998 and many authors since then)
and van der Waals forces for fine powders (Thornton and Yin, 1991; Yen and Chaki,
1992; Thornton et al., 1996 and many other authors since). More recently, empirically
measured distributions of bond strengths using AFM have been used to characterise
combinations of mechanisms for fine powders (Cleary et al., 2005).
This paper, which describes modelling and results presented in a Keynote Lecture
at DEM 2007, focuses on four key themes:
(1) the importance of particle shape;
(2) case studies of industrial application representing many critical types of
processes;
EC (3) coupling with fluid using SPH including particulates and bubbly flow; and
26,6 (4) increasing scale of the models that are possible with steadily increasing
computer power.

2. DEM simulation technique

The general DEM methodology is now well established and is described in many papers
700 including older review articles Barker (1994), Campbell (1990) and Walton (1994). There
are a number of possible contact force models available in the literature, approximating
the collision dynamics to various extents, see Schäfer et al. (1996) and Walton (1994) for
details. There is little definitive evidence to support the use of any one of these models
over any of the others for the materials found in most industrial and geophysical flows
(predominantly rocks). There is little evidence to suggest that the choice of the contact
model actually affects the flow dynamics in large scale systems. Conversely, there is little
evidence to suggest that it is not sometimes important to the dynamics for specific
applications. In some areas, such as modelling of strongly elastoplastic materials, the
Hertz-Mindlin-Deresiewicz model is generally used (see Thornton and Yin, 1991;
Thornton, 1999 for theory and examples). In the modelling reported here we use a
classical linear-spring and dashpot collision model, which is described in more detail in
Cleary (1998a, 2004). Briefly, the particles are allowed to overlap and the amount of
overlap x, and normal vn and tangential vt relative velocities determine the collisional
forces via a contact force law. The normal force:

Fn ¼ kn x þ Cn vn ; ð1Þ

consists of a linear spring to provide the repulsive force and a dashpot to dissipate a
proportion of the kinetic energy. It is restricted to being positive to prevent unphysical
attractive forces at the end of collisions when the first term is small and the later term is
larger and negative. The maximum overlap between particles is determined by
the stiffness kn of the spring in the normal direction. Typically, average overlaps of 0.1-0.5
per cent are desirable, requiring spring constants of the order of 104–106 N/m in three
dimensions. The normal damping coefficient Cn is chosen to give the required coefficient of
restitution " (defined as the ratio of the post-collisional to pre-collisional normal component
of the relative velocity), and is given in many references including Cleary et al. (1998).
The tangential force is given by:
 ð 
Ft ¼ min Fn ; kt vt dt þ Ct vt ; ð2Þ

where the vector force Ft and velocity vt are defined in the plane tangent to the surface
at the contact point. The integral term represents an incremental spring that stores
energy from the relative tangential motion and models the elastic tangential
deformation of the contacting surfaces, while the dashpot dissipates energy from the
tangential motion and models the tangential plastic deformation of the contact. The
total tangential force Ft is limited by the Coulomb frictional limit Fn, at which point
the surface contact shears and the particles begin to slide over each other.
Particles in DEM are typically represented as spheres, which have strong
quantitative drawbacks. There are a range of choices for representing particle shape. A
number are described by Dziugys and Peters (2001). Rothenburg and Bathurst (1991)
used elliptical particles to explore shape effects and Cundall (1988) and Hopkins et al.
(1991) used polygonal particles to represent rocks and sea ice blocks, respectively. Industrial
Potapov and Campbell (1996b) used assemblies of this type of polygonal particle to particle flow
model brittle fracture during impact. Numerous authors have used the clustering
approach of gluing overlapping circular or spherical particles together to make simple modelling
shapes. This has the disadvantage of being very expensive in representing many
particles with high curvature and aspect ratios. Another approach is to represent the
particles as super-quadrics. These shapes were first used in two dimensions by
Williams and Pentland (1992) and related conference papers and in three dimensions
701
by Cleary (2004). In their principle reference frame, they are specified by:
xm ym zm
þ þ ¼1 ð3Þ
a b c
The ratios of the semi-major axes b/a and c/a are the aspect ratios of the particle and
the super-quadric power m determines the shape of the particle. For m ¼ 2 and then
aspect ratios are one then a spherical particle is obtained otherwise an elliptical particle
results. As m increases, the shape becomes progressively more cubic with the corners
becoming sharper and the particle more blocky or cubical. This is a very flexible class
of shapes, which varies continuously, can be dynamically changed during simulation
to evolve particle shape and allows plausible shape distributions to be well
represented. We favour this representation of particle shape as a good balance between
the need to include the broad aspects of shape and to minimise the computational cost,
see Cleary (2004) for more details. The most detailed representation of particle shape is
to use a detailed high resolution mesh taken from laser scans of the particles (Munjiza
et al., 2003; Latham and Munjiza, 2004; Latham et al., inpress). These are attractive
when detailed shape is required but large numbers of particles are not needed.
DEM is able to produce many types of quantitative output which can be used to
gain insight into particulate flow processes (Cleary 1998a, 2004). These include:
. transient flow visualization understanding of flow fundamentals;
. torque and power consumption;
. breakage rates, mill throughput and charge composition;
. collisional and cohesion force distributions;
. energy loss spectra/spatial and frequency distributions (for input into population
balance models);
. wear rates and distributions and the interaction of evolving boundary geometry
(e.g. mill liners) and the particle flows;
. dynamic boundary stresses (e.g. on lifters and liner plates);
. segregation and/or mixing rates;
. axial flows rates and residence time distributions; and
. sampling statistics and flow rates.
For many of our large scale DEM simulations, we typically have:
. 100,000 particles up to 14 million particles on single 3 GHz single core Pentium
style cpu computers;
. arbitrarily complex boundary geometry (any number of objects);
EC . objects undergo complex kinematic or dynamic motion;
26,6 . large size ranges (diameter ratios of largest to smallest of 10:1 to 100:1 or more)
with particle scales of 1 to 100 m;
. non-ideal shape particles, modelled here as super-quadrics, but also including
polyhedra, rods and other special shapes; and

702
. commonly in conjunction with other physics that is coupled to the flow including
particle breakage, cohesion from several physical sources, interstitial fluids,
bubbles and more.

3. The importance of particle shape

Traditionally, particles are approximated by discs (in 2D) or spheres (in 3D). Such
particle assemblies frequently cannot reproduce the behaviour of real materials
because the shapes have been over-idealised. Circular (spherical) particles differ from
real particles in at least four major ways:
(1) material shear strength (resistance to shear forces and failure);
(2) dilation during shear (because of the volume of revolution);
(3) realistic voidage distributions (circular particles pack very efficiently but more
extreme shapes pack poorly); and
(4) partitioning of energy between linear and rotational modes is completely
different.
Depending on the flow, some combination of these is generally very important. Cleary
et al. (1998) showed that mixing rates predicted for mixing of non-round real material
in a rotating drum when using round particles was substantially in error by more than
an order of magnitude. This resulted from predicting the wrong flow pattern because
the model bed material was too weak and slumped continuously instead of
avalanching at the free surface.
For circular particles:
. the normal force (see Figure 1) is directed along the line of centres and generates
no torque; and
. the torque is generated entirely by the friction force which is tangential and
weaker.

Figure 1.
Direction of normal force
during collision for (left)
circular particles and
(right) non-spherical
particles
For non-circular particles: Industrial
. the normal force can directed far away from the centre of mass the particle; and particle flow
. the torque is generated by some combination of the normal and friction forces. modelling
The particle dynamics are substantially different for the two cases with a radical
difference in the partitioning and exchange of energy between rotational and
translational modes. This is now shown using some simple examples. 703
3.1 Vibrating plate
Figure 2 shows the motion of non-circular particles on a vibrating plate. If the particles
were circular they would all just bounce up and down with relatively simple motion.
When the particles have substantial non-sphericity, there are significantly larger
fluctuations and a rapid exchange of energy between linear and rotational modes
motion of motion. This is generated by the collision of particles near their ends which
converts linear motion into high spins. Particles with high spin collide and bounce
apart with little spin but large linear velocities.
3.2 Angle of repose and failure
A common characterisation process for granular materials is to use the angles of failure
and repose, McClung and Shaerer (1993) and Middleton and Wilcock (1994). When the
slope of a granular material is increased (such as in a rotating drum), the slope fails when
the angle of failure is reached. The material comes to rest forming a new slope at a lower
angle called the angle of repose. Here we summarise the dependence of these angles on
the size and shape variation of the particles (see Debroux and Cleary (2001) for more
details). These simulations were all performed with uniform size distributions centred on
15 mm. A constant fill level of 30 per cent was used in all cases.
The effect of size distribution is analysed first. The ranges of the size distributions
were varied from 5, 10, 20, 50 per cent to almost 100 per cent. Figure 3 shows
the particles used for these cases. The resulting angles for failure and repose are shown
in Figure 4. The angle of failure varies very little across the whole range, whilst the
angle of repose increased moderately by than 2 as the size range increased. This
shows that the angles of repose and failure are reasonably insensitive to the particle
size distribution (PSD) and that such variations cannot create the steep slopes observed
for many materials (angles exceeding 30 ).

Figure 2.
Motion of highly non-
circular particles on a
vibrating plate
EC The effect of the blockiness (shape) of the particles was examined using a sequence of
simulations with particles ranging from discs (N ¼ 2) to slightly rounded squares
26,6 (N ¼ 8) as shown in the top row of Figure 5. The maximum dimension of the particles
was unchanged (the diameter of a circular particle would equal the diagonal of a square
one). Figure 6(a) shows the matching failure angle, which consistently increases as the
blockiness increases, with the average value being more than 5 higher for square
particles compared to circular ones. The average value found for N ¼ 8 was about 29
704 with the angle of failure sometimes exceeding 30 . The main reason for the increase is
that blocky particles tend to pack better and interlock themselves, leaving less freedom
to spin and therefore to roll and flow. The repose angle increased strongly until N ¼ 6
where it seems to stabilise at a maximum value of about 20 .

Figure 3.
Sets of particles with
different PSD ranges: 5,
10, 20, 50 per cent and
approx. 100 per cent (left to
right)

Figure 4.
Angles of failure (upper
curve) and repose
(lower curve) as a
function of the particle

Figure 5.
Particles with different
shapes
 Industrial
particle flow
modelling

705

Figure 6.
Angles of failure (upper
curves) and repose (lower
curves) as a function of
the particle shape

Simulations were also performed using ellipses of aspect ratios varying from 1:1
(circular) up to 5:1 (surfboard like). The bottom row of Figure 5 shows the
corresponding shapes. Unlike the blockiness case where length scales were kept
identical, in this case the particles were scaled by the square root of their aspect ratio.
This enabled the particle volume to be kept constant. The resulting angles of failure
and repose are plotted in Figure 6(b). They reached a maximum value for 2:1 particles
with average values for the angles of failure and repose being respectively 4 and 6
higher than for 1:1 particles. As the aspect ratio increased from 2:1 towards 5:1, the
angles sharply decreased to less than 20 (failure) and 15 (repose). This behaviour is
not as simple as found for the blockiness case since the angles of failure and repose do
not vary monotically with the aspect ratio. The maximum angles were found to occur
for 2:1 particles. Two competing phenomena were observed which are affected by
aspect ratio in opposite ways. The particles are:
(1) Able to roll easily when they have small aspect ratios (since for 1:1 particles
there is little resistance to spin).
(2) Preferentially aligned with their long axes parallel to the slope and are able to
slide easily over each other when they have large aspect ratios.
Medium aspect ratios (such as 2:1) appear to be the strongest with the greatest
resistance to shear flow.

3.3 Extraction of a post

The effect of shape on the shear strength of a granular material is easily demonstrated
by examining the extraction of a post embedded in a packed bed of particles. We
consider four cases with varying degrees of idealisation:
(1) circular particles – flat sides on post;
(2) circular particles – rough post with small asperities;
(3) non-circular particles – flat sides on post; and
(4) non-circular particles – rough post with small asperities.
Figure 7 shows the final state of the particles after the post has been pulled out for the
four cases. The circular particle microstructure fails by in-situ rolling with the particle
EC
26,6

706

Figure 7.
Final state of the particles
after the post has
been removed

shear localised to within one or two particles of the post surface. Only very close
particles are affected by the post motion and they flow in a very fluid like way. The
circular particles have almost no ability to resist shear and this is not affected by the
choice of friction coefficient. When some roughness imperfections are added (Figure
7(b)), the particles have something to grip onto and a moderate amount of material
from lower down in the bed is pulled to the surface.
When non-circular particles are used with a flat sided post (Figure 7(c)), they
align themselves with their broader flat sides against the post. The volume of
revolution of these particles is not empty so they cannot simply rotate. The post
must actually slide against the particles and friction then provides strong resistance.
When both rough imperfections are added and non-circular particles are used, then
the inability of the particles to roll in combination with surface imperfections
catching and lifting them up, causes much larger triangular blocks of particles to be
lifted. These blocks fracture and shear, reform and fracture again. Much more of the
deeper material is pulled to the surface in this case. The fractures are observed
along bands of shear localisation produced by the much higher strength of this
material. This results in a much more realistic representation of the behaviour of
real materials.
Figure 8 shows the overall force needed to extract the post for all four cases. A
steady increase in force is observed as the particles and post are made less idealised.
The non-circular particle-rough wall case has around three times the resistance to flow
that the circular particles had on the flat wall. This demonstrates the critical
quantitative dependence of particle flows to the particle and object shape.
 Industrial
particle flow
modelling

707

Figure 8.
Resistance to post
removal for different
particle-post
combinations

4. Industrial particle processes

Industrial applications can be broadly classified according to the key physical
processes occurring:
. Separation
. Mixing
. Comminution
. Agglomeration
. Storage/unloading
. Transport
. Sampling
. Metering
. Excavation
. Fluid-particulate flow
. Fluid-bubble flow
Here we present ten industrial case studies from these application classes and one
landslide study.

5. Separation
5.1 Section of a screen
Particle separation using vibrating screens is a common method for dividing particles
into a product stream and a recycle stream. The efficiency of the screen is determined by
the size and shape of the screen openings compared to the size and shape of the particles
and the amplitude and frequency of the screen motion. DEM allows screen efficiency to
be easily evaluated. There is little DEM modelling of screening in the literature. Cleary
and Sawley (2002) first presented a 3D spherical model of a periodic section of a screen.
Li et al. (2002) presented a very simplified 2D circular particle model which was used to
explore the role of different particle sizes in Li et al. (2003). Here, we again study a batch
separation process using an 800 mm square section of a screen deck oscillating
EC diagonally upwards and slightly sideways at around 20 rad/s with an amplitude of
50 mm. The screen has an array of 12  12 openings which are 37.5 mm square holes in
26,6 a flat horizontal plate. The configuration is periodic in both the horizontal dimensions in
order to simulate a much larger screen area. The PSD is mass weighted and ranges from
10 to 65 mm and a density of 2,700 kg/m3. The coefficient of restitution used was 0.3
representing a reasonably inelastic rock material and the coefficient of friction was 0.5.
The initial bed depth was 200 mm and the initial load was 200 kg. This is the same
708 configuration as originally used in Cleary and Sawley (2002). Here, we compare two
different cases, one with spherical particles and one with super-quadric shaped particles
with a blockiness range of 2.0 to 5.0 and aspect ratios of 0.7 to 1.0.
Figure 9 shows the separation process for the two types of particles. This frequency
is enough to generate significant bulk motion of the bed, which separates from the
screen at the extreme points of its motion. The bed then crashes down onto the screen
producing a pulse of fine particles that are able to move through the holes in the screen.
The rate of separation is determined both by the ability of the smaller particles to pass
through the screen openings and by the rate of percolation of fine particles through the
bed. Figure 9(a) shows the initial state of the particle beds with the spherical case on
the left being just after flow through the screen commences and the super-quadric case
on the right being just before flow through the screen commences. The particle beds
are initially perfectly mixed and the particles are coloured/shaded by size with red/
dark grey being large, green/light grey mid-sized and blue/mid grey being the finest
particles.
After 2 s of operation the flow is well established. Dark blue/mid grey, light blue/mid
grey and some green/light grey particles are able to easily flow through the screen,
creating a dilute cloud of separated particles below it. There is little difference in the
distribution of the finer particles in the still well mixed beds. Over the next several seconds
the finer particles percolate through the bed and discharge through the grate. By 10 s
(Figure 9(c)) a substantial fraction of the fine spherical material has been separated and the
bed is reasonably well stratified with red/dark grey particles predominantly at the free
surface. There is little remaining visible fine material in the bed and the separation process
is well advanced. For the non-spherical material on the right, there are still clearly quite
large volumes of finer material trapped between the coarser particles in the particle bed.
The fine particle mobility is much lower for non-spherical particles leading to lower
vertical transport through the bed to the screen surface and therefore much lower screen
separation speed. By 20 s, there is little spherical material passing through the screen and
separation has long since been completed. The bed is much shallower due to the removal
of 50 per cent of the finer material (by volume). No blue/mid grey material is visible in
either the bed or in the trickle of particles occasionally passing through the screen. For the
non-spherical case, there are still light blue/mid grey particles visible in the bed and dark
blue/mid grey particles are still visible in the separated stream under the screen. The
particle shape has significantly affected the particle separation.
Another critical difference between the two types of particle shape is the
predisposition to peg. Pegging occurs when a particle becomes trapped in a screen
opening for a significant period of time and obstructs the flow of material through the
screen. Figure 10 shows the screen for the spherical particle case when the bed is at its
highest point and the screen surface is visible. Almost half the openings are blocked by
mid-sized yellow/pale grey particles. The row on the right edge of the screen has 8 of
the 12 holes blocked. When the particles have a non-spherical shape, these mid-sized
particles are able to pass through the screen by changing their orientations as they are
 Industrial
particle flow
modelling

709

Figure 9.
Panel of a vibrating
screen (darkest particles
have the maximum size)
EC jostled by the bed and so are able to work their way through the holes. Pegging still
occurs for non-spherical particles but at a significantly lower rate than when using
26,6 spherical particles. This shows the excess pegging that is predicted by DEM when
using spherical particles to model materials that are not really spherical.
Figure 11 shows the peak discharge rates of each of the size classes for each of the
particle shape types. For the spherical particles there is monotonic, almost linear
decrease in the discharge rate with size. The separation rate for the closest to screen
710 aperture size was 0.75 kg/s which is around a third of the 2.7 kg/s separation rate for
the finest particle class. For the non-spherical particles, the separation rate declines
very closely to linear and is consistently much lower than the corresponding rate for
spherical particles. The biggest difference is observed for the mid-size class of 21 to
26.5 mm which has less than half the discharge rate when the particles have a non-
spherical shape. Note that there is some separation of particles than are nominally
larger than the grate size. This occurs because the size has been specified by the
longest dimension of the particle but the size of the particle that can pass through a
screen aperture is determined by the cross-section perpendicular to the long axis, i.e.
by the intermediate and short axes. So some nominally larger discharge of larger
particles is possible as these particles become oriented with their long axis pointing

Figure 10.
Spherical particles peg
very easily in the square
holes of a vibrating
screen, substantially over-
representing the real
chances of pegging and
blocking any further
separation by those holes

Figure 11.
Separation performance
for different size fractions
of spherical and super-
quadric particles on a
periodic section of a
vibrating screen
down and can the slide down through the hole. As the particle intermediate axis length Industrial
approaches the opening dimension this takes longer and is less likely to occur giving
very low flow rates. The biggest particle size that can be separated is determined by
particle flow
the maximum aspect ratio of the particle, which in this case is 0.7. modelling
Wear patterns and relative rates of wear can also be predicted in order to estimate
the life span of the screen or the rate of damage to any protective coating (for example a
polyurethane or a ceramic liner). Rates of accretion of cohesive fines can also be
estimated. In both cases, the geometry of the screen can be evolved dynamically during 711
the simulation to predict the change in the open area of the screen and its effects on the
screen separation efficiency. Understanding both wear and blockage and their effect on
screen efficiency and it varies throughout the screen life are important predictions that
can allow designers to improve the optimization of their screens. The designs can also
be more easily matched to the expected PSDs.

5.2 Twin deck banana screen

Banana screens are a type of screen typically used for high capacity separation of iron
ore, coal and aggregates into different size fractions. They consist of one or more curved
decks that are fitted with screen panels with arrays of square or rectangular holes. The
screen structure is vibrated at high frequency to generate around 2-20 g of acceleration.
A dense stream of particles is loaded onto the upper end of the screen, which accelerates
down the steeper early panels of the screen and then slow as the panel angle decreases
towards the discharge end. The material discharging from the top deck is typically an
oversize stream requiring further processing. The material falling through the deck can
be further separated by additional decks below. Each lower deck returns a product
stream and the material passing out through the bottom deck is the undersize. Here, we
model a full industrial scale twin deck banana screen. This work is done in collaboration
with JKMRC and the contribution of Rob Morrison to this modelling and financial
support by the CRC for Sustainable Resource Processing are recognised.
A typical configuration is shown in Figure 12(a). The frame is spring mounted on a base
and is free to move in response to the balance of forces created by the motors, gravity and
the transient weight of the bed of material flowing through the machine. The driving
motors are the complicated structure located at the top of the frame half way along the
length of the screen. They consist of four heavy, off centre weights shaped as circular sectors
located at the ends of two parallel shafts. The shafts are driven by the motors and rotate in
opposite directions. The changing centre of mass of these weights then forces the screen to
move in a specific direction. As the weights rotate to the other side, the screen is forced to
move in the other direction. The phase difference between the two sets of rotating weights
controls the shape of the motion generated by the screen motors. In the case modelled here
we use a linear motion directed at 45 upwards and towards the discharge direction. This
screen is 2.4 m wide by 6.1 m long and has two decks which allow the incoming particle
stream to be separated into an oversize stream from the top deck, a product stream from the
bottom deck and an undersize passing out the bottom of the machine. The decks consist of
a series of linear sections which start with a steeper inclination and which are then
decreased progressively along the screen. This gives each deck the banana like shape which
gives this machine its name. Three product streams are created:
(1) Top deck material travels via a chute onto conveyor 1.
(2) Bottom deck material travels via a chute onto conveyor 2.
(3) Material passing through the bottom deck falls via a chute on conveyor 3.
EC
26,6

712

Figure 12.
Twin deck banana screen
used in the DEM model

The cloth details for the top and bottom decks are summarised:
(1) Upper deck (see Figure 12(b)):
. 8  5 panels;
. 2  7 holes for most panels;
. Total of 592 holes;
. Each hole is 70  130 mm; and
. Panels inclination varies from 33 (top) to 10 (bottom)
(2) Lower deck (see Figure 12(c)):
. 4  5 panels;
. 10  13 holes for most panel;
 . Total of 2,720 holes; Industrial
. Each hole is 35  65 mm; and particle flow
. Panels inclination varies from 33 (top) to 10 (bottom) modelling
The key geometric and operating conditions of the screen are given in Table I. The feed
material has a top size of 200 mm and bottom size of 20 mm. The feed size distribution
is given in Table II. Particle shape is critical for this separation process so the particles 713
are again modelled as super-quadrics. Their attributes are:
. Blockiness 2.1-4.0 (round to moderately blocky); and
. Aspect ratios (intermediate axis 0.85-1.0 and short axis 0.7-1.0)
The particle size is specified and measured using the intermediate axis length of the
particles rather than using the maximum length as was used in the previous case.
Figure 13 shows the particle flow on the vibrating screen for a high peak acceleration
of 14 g. The top panel shows the particles coloured/shaded by size and the bottom panel
shows the particles coloured/shaded by speed. The simulation has been run until the
system is in equilibrium at which time there is 2.85 tonnes of material consisting of
152,000 particles. Feed particles flow into the front of the rock box from the vibrating
feeder above. The material at the back of the rock box is stagnant (blue/mid grey) whilst
the material at the front flows down and onto the screen at about 2 m/s. The combination
of the strong forward vibration of the screen and the steep angle of the first panel causes
the particles to accelerate down along the top deck. Peak speeds of 4-5 m/s are reached in
the middle of the screen (over panels 2 and 3). The speed then declines during passage
over the shallower panel 4 and slow further to only 2 m/s across the fifth panel and
discharge from the top deck into the receiving chute at this speed. The top deck receiving
chute captures the particles and drops then onto the higher of the orthogonal product
conveyors. This is the oversize stream of material that is nominally too coarse to pass
through the top deck of the screen.

Screen width 2.4 m

Screen width 6.1 m

Vibration frequency 1,000 rpm Table I.
Vibration amplitude 14 mm Key geometric and
Vibration type Linear at 45 operational parameters
Feed rate 1,000 tph for the banana screen

Size range (mm) Composition (%) (by mass)

200 þ 140 5.0

140 þ 100 5.0
100 þ 70 10.0
70 þ 50 20.0
50 þ 35 20.0 Table II.
35 þ 21 20.0 Feed PSD for the
21 þ 20 20.0 banana screen
EC
26,6

714

Figure 13.
Particle distribution on
the screen vibrating with
14 g peak acceleration at
two times during a
vibration cycle for a
particle bed that is in
equilibrium for feed 1;
coloured/shaded (top) by
particle size and
(bottom) by speed

Particles start to flow through the top deck of the screen once they have passed onto its
first panel. The amount of material passing through the second panel is larger and a bed
begins to form on the second deck. The high entry speed of the particles combined with
the collisional interaction with the rapidly vibrating lower deck gives the particles high
speeds of 2-5 m/s and produces a strongly dilated bed. The bed depth increases along the
screen with the largest amount passing through the top deck from the last panel. This is
due to the top panel bed only becoming well enough stratified by then and because of the
lower flow speed on the shallow panel which allows the particles more time to pass Industrial
through the screen apertures. The bottom deck is only well loaded by the last panel. The
slow build up of the bed and its dilated nature strongly inhibits the bottom deck from
particle flow
working effectively. Most of the separation occurs for the bottom deck around the last modelling
panel. This can be seen by the density of the underflow under each panel of the bottom
deck. The underflow is collected by the large lower chute which funnels the particles
through a bottom opening onto the bottom belt. The remaining particles on the middle
deck form the product stream which discharges into the middle chute and flows onto
715
another, the lower of the two orthogonal conveyors. For these operating conditions the
separation performance is visibly poor with a quite small undersize stream. The upper end
of the lower deck screen is seriously under-utilised because of the very low bed depth on
the upper panels. The poor flow through the early panels of the top deck results from the
high particle speed which gives very short capture periods for the openings and because of
the need for the fine material to percolate down through the flowing bed to reach the
openings in the cloth.
The lower panel of Figure 13 shows the particles coloured/shaded by diameter with red/
dark grey being large, greens/light greys being intermediate and blues/mid greys being fine
material. It takes some time for the bed flowing down the screen to become vertically
stratified with the finer material near the deck and the coarser material near the free surface.
This stratification process occurs by percolation of the finer material through the shearing
bed. There is a visible reduction of blue/mid grey fine material along the length of the top
screen deck. The upper half of the deck is dominated by blue/mid grey colouration but the
lower half is dominated by yellow/pale grey and green/light grey shades. The very large but
sparsely spaced red/dark grey particles are distributed along the length of the bed and are
predominantly at the free surface by the time the bed material reaches the fourth panel.
The particles flowing through the top deck to form the bed on the lower deck are entirely
dark blue/mid grey, light blue/mid grey and green/light grey. The dilation of the bed in the
upper half of the screen lower deck makes it difficult to observe composition variation along
the screen. The underflow consists of just dark and some light blue/mid grey particles. The
flow through the early panels of the lower deck is very weak, with the majority of the
separation occurring in the last two panels when a reasonable depth bed has finally formed.
The quality of the separation is shown by the colours/shades of the material on each of the
product stream belts. Clearly, all the coarser material has reported to the oversize stream (as
it should have) but these is also quite a lot of light blue/mid grey material that should have
passed through the top deck screen. The middle stream ranges in colour/shade from dark
blue/mid grey to yellow/pale grey. The dark blue/mid grey component should have reported
to the undersize stream. The discharge rates for the oversize were 190 kg/s, the product
stream was 68 kg/s and the undersize was 36 kg/s. This shows that the separation of the
screen operating at 14 g peak acceleration is well short of optimal.

6. Mixing
Mixing of particles is performed using a wide variety of devices. One class of mixers uses
mechanical stirring by agitator, to generate motion and therefore mixing in a particle
bed. DEM has been used to analyse the flow patterns in laboratory scale mixers of
various types, such as V-blenders (Kuo et al., 2002; Lemieux et al., 2007, 2008) and bladed
high shear mixers (Stewart et al., 2001a, b; Zhou et al., 2003; Kuo et al., 2004; Bertrand et
al., 2005). DEM can also be used to measure the progress of batch and continuous mixing
(see Cleary and Sinnott, 2008) and to assess the effect of operating and equipment design
changes on mixing performance. Here, we study mixing in a laboratory scale plough
EC share mixer. It consists of a horizontal cylindrical shell of diameter 250 mm and length
450 mm. One or more plough blades (shown in Figure 14) are mounted on a shaft of
26,6 diameter 30 mm located along the centerline of the shell. In this case, the plough blade
stirs a bed consisting of rice particles of approximately 2 mm by 4 mm. The DEM
plough share blade and containment vessel were constructed to exactly match the
specifications of the laboratory mixer used in PEPT experiments (Laurent et al., 2000).
This configuration has previously been studied using DEM with the rice approximated
716 as spherical particles (Cleary et al., 2002; Cleary 2004; Cleary and Sinnott, 2008).
Since rice is poorly represented as spheres we repeat the analysis using super-
quadric particles. This shape description is particularly well suited to grains. Here the
rice is modelled as particles with diameter 4-5 mm, super-quadric power from 2.1 to 3.0
(reflecting their very ellipsoidal nature), intermediate and short axis aspect ratios from
0.4 to 0.5. The coefficient of restitution used was 0.5 and the friction coefficient was 0.4.
A spring stiffness of 1,000 N/m is required and gives average overlaps of 0.5 per cent of
the diameter. The mixer is filled to 25 per cent by volume giving a bed made up of
280,000 particles. The particles are coloured/shaded according to their initial position,
so that we can see the extent and nature of the mixing. In each radial cross section the
particles in the left half and right half have different colours/shades. These colours/
shades are also different on either side of the blade in the axial direction.
Figure 15 shows the mixing in a single blade plough share mixer operated at 4 Hz and
using the realistic rice shaped particles. The blade starts on the near side of the mixer.
Figure 15(a) shows the mixer after 0.56 revs when the blade has risen above the initial
surface level of the particles. It has pushed and lifted all the material that was in the path of
the blade, leaving a clear trench behind the blade that is empty of particles. The particles
near the periphery of the lifted mass, under the influence of gravity, are starting to flow
down and away from the blade. The part of the bed lifted is very coherent and has a clearly
defined free surface. This material is behaving like a deforming solid. There are very few
particles splashed about by the impact of the blade with the bed. The motion is best
characterised as a high speed bulldozing of the bed rather than as a high speed collision of
the blade with individual particles. The very strong convective mixing effect of the blade is
clearly visible with substantial amounts of blue/mid grey and green/light grey material
pushed from the front of the mixer into the rear. Similarly large amounts of yellow/pale
grey and red/dark grey material are lifted into the air by the blade. In Figure 15(b), (at
0.84 revs) the mass of particles that was lifted up have almost entirely lost contact with the
blade. The ones closest to the blade maintain contact for the longest period and follow
trajectories that are closest to the blade trajectory. Particles that are further away flow
earlier from the mass being carried by the blade, tend to have slower velocities and move on

Figure 14.
Sketches of the top view
and side view of the
plough (dimensions in
mm) on the left and
two views of the CAD
model used in the DEM on
the right
 Industrial
particle flow
modelling

717

Figure 15.
One blade plough share
mixer operating at 4 Hz
with ellipsoidal
particles at times

trajectories that are lower and with more axial displacement. The band of particles flowing
from each side of the blade remains quite coherent with a well defined structure which is
wide near the bed and narrowing progressively as one moves above the bed and towards
the tip of the blade. There is significant transport of the different colours/shades to new
parts of the mixer space but very little actual mixing between them. The sides of the trench
behind the blade have collapsed and particles have flowed back into the trench partially
filling it, but the bed surface level remains lower than the initial bed surface location.
After 1.08 revs (shown in Figure 15(c)) material from the first blade pass has become
much more dilute as the divergent trajectories of the particles cause them to be spread over
progressively larger volumes. The leading material which is very close to the blade is
sliding on the mixer shell and creates a coherent arc leading back towards the bed at the
EC rear of the mixer. Significant axial displacement has now occurred for most of the ballistic
particles. The blade tip is just about to enter the bed for its second revolution. Reasonable
26,6 variation in the colours/shades of the particles in the forward part of the ballistic bands
indicates some mixing has occurred. There is little mixing of the colours/shades in the
wider but slower moving parts of the lifted mass closer to the bed surface. Figure 15(d)
shows the situation after 1.88 revs. The ballistic material from the first pass of the blade
has now fallen onto the surface of the bed at various distances from the plane of the blade
718 passage. There is some modest mixing in the material that has come to rest on the shell at
the front of the mixer. The passage of the blade has now lifted a second mass of particles
into the air. The structure of the material flow around and from the blade is similar to that
of the first pass, but in comparing Figures 15(b) and (d) we can observe that there is
significantly less material being lifted on the second pass. The lifted bands are much
narrower reflecting the much lower bed surface level in the path of the blade after the
trench from the first blade pass only partially refills this region with particles. The
material in these narrower bands is close to the blade and is thrown on the highest and
widest trajectories. The parts of the bands from the first pass that are missing from the
second are the slower material furthest from the blade, which falls from the mass much
earlier and was less influenced by the blade. The amount of colour/shade variation in the
lifted mass is much higher for the second blade pass reflecting the presence of both
colours/shades on each side from the stretchingof the bed produced by the first blade pass.
The structure of the bed around the trench is very similar to the state at the comparable
time of the previous blade pass (Figure 15(b)) indicating that the bed flow back into the
trench is similar foreach of the blade passes.
Figure 15(e) shows the flow at 3.88 revs as the blade is starting its fourth pass through
the bed. The mass of material to be lifted is just being accelerated by the blade. After a few
blade passes the shape of the free surface of the bed stabilises, with the bed material
removed by the blade being balanced by the material flow back along the free surface
(which sits at the angle of repose) back into the trench left by the blade. Once this
equilibrium is reached then each blade pass has the same effect on the bed and the amount
of material lifted, the structure of the lifted bands of particles and their trajectories are the
same. Each blade pass then makes an incremental change to the mixing state. The colour/
shade variation of material on the bed surface shows that there is already quite good re-
distribution of the blue/mid grey and green/light grey material across the top of the bed
and that the surface material towards the front of the view is well mixed. This partially
disguises a key problem with this mixer, which is the presence of large dead regions to
either side of the blade induced trench. This was previously observed for spherical particles
(Cleary, 2004). Here the rice grain shaped material locks together better and steeper angles
of repose mean that the trench is steeper and narrower and that the dead regions are now
actually expanded in size. Figure 15(f) shows the flow just before four revs are complete.
The ballistic stream of particles flowing from the blade to the right now contains a good
mix of green/light grey and red/dark grey with some blue/mid grey material. On the other
side there is a good mix of blue/mid grey and yellow/pale grey material with some green/
light grey. So in the agitated regions where the blade is able to transport the particles, the
mixing is quite fast and works well. This leads to good mixing near the blade path and in
surface regions further away, but not in the dead regions to either side.
Compared to the earlier predictions using spherical particles (Cleary et al., 2002;
Cleary, 2004) we observe that the streams of material lifted and thrown by the blade are
more coherent and larger. The greater strength of the non-spherical material allows the
blade effect to be felt further from the blade. So more particles can be agitated, which is
positive for mixing. The trench is also more coherent in shape and is longer lived Industrial
reflecting the stronger bed material taking longer to fail and flow back into the trench.
This is not expected to affect the mixing. The sides of the trench are steeper though
particle flow
which reduces the distance from which material can flow back in and therefore modelling
increases the size of the dead regions which is a negative for mixing performance.

7. Excavation
Draglines are used to remove overburden in open cut mining. A dragline is a giant crane-
719
like structure (see Figure 16(a)) from which hangs a huge bucket (Figure 16(b)) with
volume up to 100 m3. This is dragged up the side of the pit (with depths of order 50 m)
and fills with rock. Dragline performance is affected by the bucket design, the mode of
operation, the attachment of the cables, and the material properties of the overburden.
Here, we examine the filling of an ESCO bucket. It is 5 m long, 2.5 m high and about 2 m
wide with a curved back. The lip/teeth section is 0.5 m long and is inclined with an angle
of attack of at 40 (to the bottom of the bucket). This bucket has previously been modelled
with discs in 2D (Cleary, 1998c), as super-quadrics in 2D (Cleary, 2000) and as spheres in
3D (Cleary, 2004). Now we include the particle shape in 3D, modelling the particles as a
broad mixture of different super-quadrics reflecting the varied shapes produced by
blasting of the rock over-burden. The bucket is fully dynamic, moving in response to the
influence of the particle collisions, gravity and the forces from the cables in the rigging.
Figure 17 shows the filling sequence for this ESCO bucket with particles ranging in
size from 100 to 300 mm. As the bucket moves towards the dragline, the lip and teeth
bite into the overburden, producing an initially thin stream of particles flowing into the
bucket. As it fills, the resistance to shear of the material already in the bucket needs to
be overcome in order for the material be pushed up into the back of the bucket. A pile of
particles forms in front of the bucket that is of comparable height to the particles in the
bucket. This pile is bulldozed along in front of the bucket. Its size increases until the
resistance to shear of this pile exceeds the resistance of the material inside the bucket
causing it to flow and increase in volume and therefore height. The piles inside and
outside the bucket are closely linked and grow together. For this material, the flow is
fairly steady with no large or abrupt changes in force. The bucket motion is therefore
quite stable. At 7 s, the bucket is substantially filled and the front cables shorten in
preparation for the lifting of the bucket and completion of the drag process. Shortening
these cables change the angle of the bucket tilting it backwards and allowing the teeth
to disengage from the ground. All cables are then shortened to lift the bucket and by

Figure 16.
Dragline excavation,
(left) dragline in an open
pit mine, (right)
dragline bucket
EC
26,6

720

Figure 17.
Progress in the filling
and lifting of an
ESCO dragline bucket
for non-spherical
rock overburden with a
100 mm bottom
particle size

12 s, then it has separated from the ground. A small amount of material flows down
the front surface of the spoil and falls from the bucket.
Comparing these results with the earlier 3D results in Cleary (2004), we observe that
there is much more resistance to flow of these particles compared to spheres. The
maximum bucket fill level is moderately reduced and the fill time is increased by one
second. Further increases in the drag period do not lead to further increases in filling.
The filling is limited by the resistance of the material already in the bucket. Higher fill
levels would require an increase in the pushing force from the bull dozed pile in front of
the bucket. This is principally controlled by the cut depth of the teeth below the
surface. So the bucket rigging would need to be changed in order to dig deeper and fill
the bucket further. Once the bucket is filled, there is only a small amount of material
loss during lifting. This contrasts with the spherical case, where the bucket filled
substantially higher due to the higher particle mobility, but much of this extra material
was subsequently lost after lifting as the very mobile material with its very low angle Industrial
of repose flowed back out of the bucket. The filling and retention behaviour of this
model including non-sphericity of the particles is significantly more realistic.
particle flow
Many draglines are operated in areas with significant rainfall, meaning that the modelling
granular material is often significantly wet and also often contains significant sticky fine
material or clay materials. It is important to understand the effect of cohesion on the
dragline fill process. Here, we use a simple linear cohesion force in addition to the
collision force. The maximum force (often termed ‘‘the pull-off’’ force is chosen to have a
721
Bond number of one based on the largest particle size. The Bond number is the ratio of
the maximum cohesion force on a particle to its weight. In this case, the force is 400 N
for the particles with diameters from 150 to 300 mm. Figure 18 shows the filling process
for the same dragline bucket configuration using this coarse cohesive non-spherical
material. The early part of the fill process is similar, but the filling rapidly stalls. The

Figure 18.
Filling of a dragline
bucket for coarse rock
overburden with
moderate cohesion
EC increased resistance of the cohesive rock pushes the bucket backwards. In this position,
the cables and the rock create a greater torque on the bucket which rotates slightly with
26,6 its lip and teeth lifting. This reduces the depth of the cut and leads to a much smaller
bulldozed pile in front of the bucket. This smaller pile exerts a much weaker force on the
material in the bucket, which cannot overcome the cohesion or the inter-locking of the
non-spherical particles. Consequently, there is no flow into the bucket, which is dragged
along pushing a small pile in front of it but never filling up. So the cohesion changes the
722 resistance of the spoil material leading to reduced filling. Sticky material is harder to dig.
This significantly affects the final spoil load in the dragline at the end of a drag. Longer
drag times do not increase the fill levels because the resistance of the material in the
bucket reaches a balance with the weight of material bulldozed in front of the bucket.
Longer dragging in this equilibrium configuration does not change the force distribution
and so cannot produce filling. The only solution is to change the rigging of the cables to
increase the angle of attack of the teeth or to position the bucket initially lower for a
deeper cut (which produces a bigger pile in front of the bucket leading to higher forces
and more flow into the bucket). This means that for this bucket the optimal rigging is
dependent on the material properties of the over-burden which changes with rock and
clay composition throughout the mine and with weather and seasons depending on rain.

8. Transfer
Conveyor belts are commonly used to transfer particulates, particularly in large scale
bulk materials applications. They are cost effective and can be used to transport
materials for very long distances. Unfortunately, they are not easily able to
accommodate sharp bends or sharp changes in elevation. In these circumstances a
transfer chute is used to capture the material as it leaves in conveyor and to load this
properly onto a second conveyor. Previously, Zhang and Vu-Quoc (2000) used a DEM
simulation of an inclined flat floor with periodic boundary conditions and particles that
are clusters of four spheres to model chute flow of soybeans. Dewicki and Mustoe
(2002) and Dewicki (2003) evaluated the design of full scale transfer chutes and rock
boxes using spherical particles. Gröger and Katterfeld (2006) also used clusters of
overlapping spheres to model somewhat more realistic shapes in a chute scenario. To
illustrate DEM modelling of industrial scale conveyor chutes with realistic particles,
we use an in-line transfer designed originally for coal transfer between two conveyors
with a significant vertical height change. The feed belt is 1.8 m wide and moves at
4.05 m/s. The receiving belt is slightly narrower at 1.6 m and moves lightly faster at
4.4 m/s. We will examine two variants whose details are given in Table III. The first is a
coarse coal transferred at 1,500 tph and the second is a fine material (which could be
either fine coal or iron ore fines) at a ship loader rate of 4,500 tph.
Figure 19(a) shows the flow through the inline conveyor transfer chute for coarse coal
which is coloured/shaded by size. The particles start in a uniformly mixed bed along the

Case 1 2

Flow rate 1,500 tph 4,500 tph

Particle diameter 20-50 mm 5-7 mm
Particle shape SQ Sphere
Table III. Blockiness 2.5-6.0 2.0
Conveyor transfer rates Aspect ratios 0.7-1.0 1.0
and particle details Number of particles 200,000 11.1 million
 Industrial
particle flow
modelling

723

Figure 19.
Flow through an inline
conveyor transfer chute

incoming conveyor with the highest point along the middle of the belt. The belt shape is
flat on the bottom and inclined at 30 on the sides. As the head pulley is approached the
belt flattens out and the pile broadens. The particles therefore depart from the head pulley
with slightly divergent trajectories. In this design, a pair of wing fillets capture the falling
stream and gently narrow it, eliminating the divergence of the flow and focusing it slightly
to form a very coherent stream that in falling, makes contact with the lower spoon section.
This is a gently curved chute that first captures the stream by making contact with a
surface almost parallel to the particle trajectories which then becomes shallower. This
reduction in the chute angle slows the falling stream leading to compaction as the dilation
produced during the acceleration caused by falling is reversed. The slowed stream then
flows gently onto the receiving conveyor. If the chute is well designed, the exit speed from
EC the chute should match the receiving conveyor belt speed to ensure that there is minimal
disturbance to the flow and minimum damage to the belt and particles in the transfer
26,6 process. For the 1,500 tph flow rate, this chute operates very well with little evidence of
any flow disturbance or of any flow related problems. Figure 19(b) shows the flow for the
second case of a much finer, round material at a ship loader volumetric transfer rate of
4,500 tph, with the particles coloured/shaded by speed. This simulation uses 11.1 million
particles with diameters of 5-7 mm and took around 14 days on a single core Pentium
724 CPU. The tripling of the flow rate leads to a much higher load on the feed conveyor and in
the chute but this design is easily able to accommodate the flow rate increase. The
reduction in the particle size has not had an appreciable effect on the flow. The particles
are initially green/light grey and accelerate to red/dark grey as they fall under gravity. The
speed reduction from the frictional interaction with the wing fillets is observable. The
particle stream slows from red/dark grey to orange/medium light grey as it passes onto
the receiving belt. This speed change is small enough that there is little splashing or
turbulent boiling at the contact point on the belt.
Figure 20 shows the effect of flow of the fine particles through the in-line conveyor
transfer for the 4,500 tph flow rate. The impact damage (Figure 20(a)), as measured by the
absorption rate from energy from the normal component of the particle collisions, is very
weak across the entire chute surface. The abrasion damage, as measured by the energy
absorption from the tangential particle interactions, (Figure 20(b)) is evenly distributed
over most of the area of the receiving chute. The first red/dark grey band is the location of
the initial contact of the stream with the chute. Since the trajectory of the stream is almost
parallel to the chute surface there is little direct impact explaining the very weak impact
damage measure. The particles slide upon impact. There is a short region of lower
abrasion as the gentle rebound of the particle stream from the initial contact region leads
to a reduction in the normal pressure and therefore a reduction in the frictional dissipation.
The stream then re-contacts the chute and maintains constant pressure and speed
thereafter leading to a very even distribution of abrasion. This design demonstrates two
key requirements of a good chute design. Firstly, the chute should operate with only one
damage mechanism active so that a suitable chute material can be chosen. In this case the
abrasion is dominant and an abrasive resistant material, such as ceramic tiles, can be
chosen to line the chute. If the contact was dominated by impact then an impact resistant
material, such as steel could be chosen. The second requirement is that the damage should
be uniformly distributed since the time to failure is governed by the first location to wear
out, it is important to make the wear as uniform as possible. The DEM modelling shows
that the damage is almost entirely abrasive and is very uniformly distributed in the lower
half of the chute so this design has been well optimised. Figure 20(c) shows the average
contact speed of particles with the chute. The high speed contacts with the wing fillets can
be easily observed. The speed on the lower half of the chute is very even which contributes
to the evenness of the abrasion damage. The high speed contacts above this are produced
by much smaller numbers of particles that are near the underneath free surface of the
falling stream and which collide earlier and therefore higher up the chute than the main
bulk of the stream. The particles have higher speed, but their low numbers and the low
pressures in this region mean that they contribute little to the damage measures.

9. Comminution
9.1 Crushing
Crushing is typically the first step of comminution processing of bulk materials in mineral
processing, aggregate and cement industries. Many types of crushers have been developed
 Industrial
particle flow
modelling

725

Figure 20.
Effect of particle flow on
an in-line conveyor
transfer
EC using different principles for breakage. Previous modelling of crushers is very limited.
Djordjevic et al. (2003) modelled single particle breaking under impact and Lindqvist and
26,6 Evertsson (2003) predicted wear in a cone crusher. Another crusher type is a jaw crusher.
Particles are fed under gravity into the gap between a vertical fixed jaw and an inclined
moving jaw. The motion of this jaw is elliptical and has a cyclic motion component along
the plate to assist with transport of particles within the machine and motion orthogonal to
this that leads to cyclic compression of the particles. As particles flow down into the
726 machine, they reach a point where there are trapped between the jaws or between a jaw
and another particle and are compressed sufficiently that they fracture. The daughter
fragments then are free to flow further down into the crusher where they can be either
trapped again and crushed further or discharged from the bottom opening as product.
DEM modelling of such crushers requires the particle breakage to be included.
Otherwise particles become wedged in the device, cannot be transported through the
machine and therefore block it up. We use an approach first introduced by Cleary
(2001c) where the energy absorption of stress on the particle is monitored and the
particle is then instantaneously broken when the threshold breakage energy or stress is
exceeded. The particle is then replaced by a geometrically packed set of daughter
particles that fill the space occupied by the original. This model is reasonable because
the timescale of the fracture process is orders of magnitude faster than the normal
collision timestep and the timescales for the particles to move through the crusher. In
this section we will describe DEM models of three different styles of crusher.
A jaw crusher breaks particles by cyclic compression. The one modelled here is
1.92 m high and 1.01 m deep. The feed opening is 800 mm wide while the discharge
opening is 200 mm giving a product top size of 200 mm. The inclination of moving jaw
is 20.6 and the angular speed of jaw about its pivot point is 200 rpm. The feed size
distribution was uniformly mass weighted from 100 to 600 mm and the feed specific
gravity is 2.8. The very high pressures in the crusher require a very high spring stiffness
to produce acceptable deformations of 0.5 per cent. We use a stiffness of 108 N/m which
is close the actual stiffness of real rock. So unlike many other applications where we can
treat the stiffness of the material as a numerical parameter and can use much lower
values to accelerate the simulations, here we need to use a very high value. For the rock,
we use a coefficient of restitution of 0.3 and a friction coefficient of 0.5. The rock breakage
threshold force ranges from 1 MN for the largest (and most fragile particles) to 3 MN for
the smallest fragments. The minimum fragment size included in the model is 10 mm.
Figure 21(a) shows the jaw crusher system in a view from above. Feed with a very
broad size distribution flows onto a vibrating grizzly feeder. This has long slots that
allow finer material to fall through and then travel down a chute to the receiving belt,
thereby bypassing the crusher. This is commonly done to prevent degradation of crusher
performance by fine material blocking up the pore space between the large particles and
providing a cushioning effect that reduces compressive fracture. Coarse material that is
bigger than the grizzly grate size flows off the end of the feeder and into the jaw crusher.
Figures 21(b)-(d) shows the progress of particles through the crusher during one
compression cycle. The largest particle in b is compressed and has fractured in c and the
fragments then flow down lower into the crusher and are waiting for the opening part of
the compression cycle so that they can flow down and discharge from the crusher. The
crusher charge when full is predicted to be around 1.5 tonnes with a discharge speed of
4 m/s. The power draw is 44 kW and throughput is predicted to be 900 tph. Of the total
energy consumption in the crusher 22 per cent is absorbed by the crusher working
surfaces and will lead to wear with 12.5 per cent found for the fixed jaw and 9.6 per cent
 Industrial
particle flow
modelling

727

Figure 21.
Breakage and transport
through a jaw crusher

for the moving jaw. The rest of the energy is expended in doing surface damage to the
particles (the energy dissipation by the contact model) or in particle breakage.
Another very efficient method of crushing is derived from firing particles at high
speeds into obstacles, causing breakage through the impact forces generated. One such
machine is called a vertical shaft impactor (VSI). A previous single particle DEM model
by Djordjevic et al. (2003) has shown that DEM can well predict the fragment size
distribution. Figure 22 shows a model of an entire VSI. A continuous stream of
particles is centrally choke fed from a hopper down into an impeller that rotates at
1,000 rpm. The particles at the bottom of the feed are accelerated sideways by the
centrifugal force produced by the driving surfaces of the impeller. This produces a high
speed stream of particles from each of the three blocks in the impeller. Figure 22(c)
shows the particles coloured/shaded by speed. A clearly defined, coherent curved
stream of particles can be seen leading from the front block and curving back around
to the left. These three streams then collide with the ring of anvils and are broken by
EC
26,6

728

Figure 22.
Particle distribution in a
VSI
impact with the anvils and/or other particles. The breakage rule used here is based on Industrial
the energy absorption by each particle whose specific comminution energy can be
determined from drop weight or other breakage characterization tests. The particles
particle flow
are coloured by size in Figures 22(a) and (b). The feed material is coloured/shaded modelling
green/light grey through yellow/pale grey to red/dark grey. This material is thrown
outward and fractures at the anvils. The fine blue/mid grey product rebounds with
high speed and produces an increasingly dense granular gas. On the millisecond
timescale of the operation of this device there is very little gravitational settling. This
729
system is often thought to be a dilute system, but this is not really the case. It takes the
product material quite some time to discharge and leads to the build up of a reasonable
resident volume of material which can then be further broken by subsequent collisions
with new feed material. In future work, we will include the gas flow which is critical to
the classification of the fine product and its removal from the VSI system.

9.2 Grinding – SAG mill

Typically in a comminution process, there is at least one grinding process following on
from the initial crushing of the particles. There are many different devices used for
grinding in mineral processing, cement, industrial minerals, pharmaceuticals, food
processing and nanoparticle generation. DEM is ideally suited to predicting the
behaviour of the charge in the mills and is increasingly able to shed light on the
grinding mechanisms being exploited and how to understand the breakage occurring
in mills. For a recent summary of DEM prediction of comminution processes, see
Morrison and Cleary (2008).
Tumbling mills are large cylindrical vessels that rotate about their axes at
reasonable fractions of the speed required to centrifuge the particulate material that is
ground within. The grinding is produced by a range of mechanisms including high
speed impact, abrasion, chipping and nipping of small particles between large ones.
Grinding is often augmented by adding various types of grinding media. SAG mills are
the largest tumbling mills used (at up to 400 or 12.2 m diameter) and are used for
primary grinding, mainly in mineral processing, using a coarse feed (typically up to
250 mm diameter rock) coming from some arrangement of crushers. There is
insufficient coarse rock to produce large impacts so the charge is augmented by steel
grinding balls. These can make up between 20-50 per cent of the volume of the charge.
SAG mills have previously been modelled by Rajamani and Mishra (1996), Cleary
(2001a, c), Morrison et al. (2001), Herbst and Nordell (2001), Hlungwani et al. (2003) and
Morrison and Cleary (2004, 2008). Validation of DEM predictions for SAG mills in
Cleary et al. (2003) and Bwalya and Moys (2003) demonstrates very good agreement
between full 3D DEM simulations and laboratory scale model SAG mill experiments.
Previously, we have studied the charge motion in a pilot scale Hardinge
60 (D)  20 (L) mill, see Cleary (2004) and Morrison and Cleary (2004). An advantage of
modelling such a mill is that the entire mill along with around 95 per cent of the
measured PSD can be included in the DEM model. The model contains the complete
mill geometry including the feed end, the main/belly/circumferential lifters, the end
wall lifters, the grate, the pulp lifters, the discharge cone and the external mill shell.
This configuration has 16 belly lifters with 10 face angles, 16 lifters on the feed end
cone and 8 lifters on the discharge end. A representative SAG mill charge rock and ball
size distribution was used, with a rock top size of 122 mm and a ball size range from 15
to 105 mm. The mill rotation rate used was 24 rpm corresponding to 76 per cent
critical. In the earlier work, we examined the charge flow behaviour, energy utilisation
EC and grate discharge of this mill with a fill level of 39 per cent (by volume), with a 5 per
cent ball load and a 6 mm bottom size cutoff for the PSD. Here, we look at what is now a
26,6 more common set of operating conditions with a fill level of 30 and 10 per cent ball
charge. In addition, we have reduced the minimum with particle size to 3.5 mm. This
means that >95 per cent of the charge was correctly represented in the model. The total
number of particles used was 485,000. The material properties for the steel and rock
are given in Table IV.
730 Figure 23(a) shows a cutaway view of the mill. The sectioned cylindrical opening on
the left is the feed. The discharge grate can be seen on the far side. The particles are
coloured/shaded by their size. Figure 23(b) and c shows the mill, sectioned in half by an
axial plane and opened up so that the viewer can look along the axis towards the feed
end (Figure 23(b)) and the discharge end of the mill (Figure 23(c)). The particles are
shaded by size in the feed end view and by speed in the discharge end view. A typical
mill flow pattern is visible, with the particles carried upwards by the motion of the mill
liner until they reach the shoulder position and then flow down the cascading free
surface of the charge. For these lifters and mill speed, there are significant coherent
cataracting streams that are composed almost entirely of fines due to strong radial
segregation. The space between lifters acts like a bucket being tipped producing a
coherent stream. This leads to significant numbers of impacts on the liner above the toe
position. The blue/mid grey band along the centre of the charge in Figure 23(c) is
slowly moving and is sandwiched between the upward moving layers of charge below
and the downward moving layers above. This generates significant shear and much of
the energy dissipation occurs as a result of this sliding of the particles over reach other.
This leads to the abrasion, rounding and chipping of particles (see Morrison and
Cleary, 2008 for details of these mechanisms). The upper sections of the cascading
stream and the cataracting stream produce stronger impacts that can break particles
either with one impact or with the accumulation of damage over many successful weak
impacts (again see Morrison and Cleary, 2008 for details).
Comparing the flow predicted here to the previously published case with a fill level
of 39 per cent (Cleary, 2004 and Morrison and Cleary, 2004) we observe that the surface
level is much lower (as one would expect) and adopts the bi-linear shape that is
commonly observed in SAG mills. The volume of material within the cataracting
streams is also substantially reduced – visually appearing to be around half the
volume. The curved band of slowly moving (blue/mid grey colour) material is around
halfway between the liner and the free surface for a fill level of 30 per cent whereas it
was located much closer to the free surface for the 39 per cent fill case. The thickness of
the cascading layer above this is quite similar for the two cases, so the increase in free
surface level for the 39 per cent fill occurs entirely below the slow moving band. This
means that the shear rates are reduced in the lower regions of the charge by a factor of
2-3. Since this is where much of the abrasive reduction in size occurs as a result of the
shear and the high loads, the reduction in shear in this region is unlikely to be positive
for grinding performance. The amount of material resting on the end wall lifters is also
reduced when the fill level is reduced.

Material combinations Steel–steel Steel–rock Rock–rock

Table IV.
Material properties for Coefficient of restitution 0.8 0.5 0.3
the charge in a SAG mill Coefficient of friction 0.5 0.5 0.5
 Industrial
particle flow
modelling

731

Figure 23.
Charge behaviour in a
Hardinge pilot mill for a
fill level of 30 per cent,
ball loading of 10 per cent
rotating at 76 per cent of
critical speed
EC The power draw for the original charge was 9.7 kW which is larger than the 8.7 kW
found for the new lower ball charge. This reduction is due to the reduction in the charge
26,6 mass from 1,188 kg to 918 kg that occurs with the reduction in fill level but partially
offset by the increase in the ball load. This means that the energy intensity of the
charge is 8.2 W/kg of charge for the original configuration. This is increased to 9.5 W/
kg of charge for the lower fill level. So the reduction in fill level has reduced the power
draw but increased the intensity of energy use. As observed earlier, the extra material
732 is located near the free surface which moves close to the centre of the mill. This extra
material does not consume a lot of energy but contributes a lot of extra mass (Cleary,
1998b). In addition, the lower shear rates observed below the centre of circulation of the
charge mean that the energy consumption is spread out more which is consistent with
the lower energy intensity predicted. This reduction will be negative for grinding
performance since there are many more particles experiencing very weak collisions
that do not lead to useful particle size reduction.
Table V shows the energy utilisation for the two fill level/ball charge combinations
of charge in the Hardinge pilot SAG mill for each of the types of collisions. Overall the
change in the fill level has had little influence on the distribution of energy
consumption. The doubling of the ball load has lead to a 3 per cent reduction in the
fraction of rock-rock dissipation, a matching 3 per cent increase in the rock-ball
dissipation and an increase in ball-ball energy dissipation from 0.2 to 0.3 per cent. The
energy dissipation in the rock-liner and ball-liner collisions is unchanged.

10. Landslides
Landslides are an important class of natural disaster that can lead to significant loss of
life and significant property damage. Understanding when and where landslides may
occur and under what circumstances is a key part of risk assessment and protective
strategy formation. The path that any specific landslide may take and how far it will
run from the initiation point are also important considerations. One class of large
landslides is particularly dangerous because of the long distances that they can travel
(more than several kilometers is quite possible) while exhibiting very low effective
friction coefficients. Cleary and Campbell (1993) showed that this was possible using a
very simple two dimensional DEM model with periodic boundaries because the shear
in the landslide was restricted to a narrow layer at the base with most of the landslide
mass travelling undisturbed over the top. This resulted in lower energy dissipation and
the low effective friction that was required to explain this long run-out. Campbell et al.
(1995) showed, using large scale 2D DEM simulations of the mountain slope and valley
flow that the phenomena was more complex with the basal shear region being more
blurred and continuous, but still restricted to the lower portions of the landslide. More
recently, we extended the modelling to three dimensions and used real topography to
explore the effect of the additional realism on the DEM predictions of landslides (Cleary

Collision type 39% fill and 5% ball charge 30% fill and 10% ball charge

Rock–rock 88.7 85.9

Table V. Rock–ball 6.7 9.4
Energy utilisation in the Rock–liner 4.3 4.3
Hardinge pilot SAG mill Ball–ball 0.2 0.3
for two different charges Ball–liner 0.1 0.1
and Prakash, 2004; Cleary, 2004). Here, we re-visit the first of these landslide scenarios Industrial
and explore the changes that occur when using more realistic non-spherical particle
shapes.
particle flow
In Figure 24, we show the collapse of the side of a steep escarpment into a valley in modelling
Northern California. On the left is the spherical case shown from above with
progressive rotation of the camera. On the right is shown the case using super-quadric
particles with a camera that is in a fixed orientation. The particles are coloured/shaded
by their speed with blue/mid grey being stationary and red/dark grey being >50 m/s.
733
In both cases, the landslide mass is mobilised and able to move all at once accelerating
rapidly downhill. By 10 s, the mass has divided into two streams with the larger one on
the right descending quickly into the main valley. The smaller stream on the left passes
behind the small peaks in front of the initial landslide position. For both streams of
particles, it is already clear that the leading non-spherical particles are travelling more
slowly than the leading spherical ones and the leading particles have consequently
travelled less far down the mountain side. By 20 s, the spherical particle landslide has
reached the steeply inclined central valley floor and the two streams have merged with
the leading material reaching the bottom of the rendered region. The non-spherical
particle landslide is now lagging significantly behind with the leading material having
travelled only about two-thirds of that distance. The stream on the left has almost come
to rest in one of the valleys well above the central valley. This stream is unable to reach
and merge with the main landslide branch. At 26 s, the spherical particle landslide fills
much of the valley and moves at its highest speed and has just encountered the large
ridge that blocks the valley and dissipates much of the kinetic energy of the landslide.
In contrast, for the non-spherical particle case, the volume of material in the central
valley is considerably smaller, is still far short of the ridge and is travelling more
slowly. These significant differences demonstrate the critical importance of the particle
shape and realistic topography on landslide motion. The non-spherical particles have
much higher angles of repose meaning that the mobilised volume of rock is lower and
the dissipation experienced during flow is higher leading to faster deceleration of the
landslide. The combination results in predicted landslides that are smaller and stop
more quickly.

11. Fluid-particulate flows

In all the previous application areas the particulates were dry, but many applications
exist where the particulates are surrounded by a fluid. The first example of the
coupling of DEM to CFD was for pneumatic conveying by Tsuji et al. (1992). Particles
coupled to gas motion have subsequently been modelled by many authors in
applications such as fluidised beds (Tsuji et al., 1993; Xu and Yu, 1997; Hoomans et al.,
1996; Kafui et al., 2002 and many other authors since then) using DEM coupled to finite
volume solutions of the fluid dynamic equations. For coupling to fluids, this type of
fluid modelling method is less feasible because of the presence of dynamic free surfaces
and the complex moving geometries used in these applications. A meshless modelling
method that is highly suited for modelling fluids coupled to particulates is SPH. For
details of this method and examples of its application to industrial fluid flows, see
Cleary et al. (2007a).
To demonstrate the utility of SPH coupled to DEM for modelling fluid-particulate
systems, we examine the motion of buoyant pellets in a mixing tank. The mixer consists of
a 1 m diameter by 1 m high cylindrical vessel with a single baffle on the side and a four
bladed downward-pumping impeller rotating at 200 rpm. The pellets were made from
EC
26,6

734

Figure 24.
Landslide from a
collapsing mountain peak
cylindrical sections of wood with a specific gravity of 0.5 with dimensions of 16 mm Industrial
(diameter)  22.5 mm (height). Pellet loadings of 1.5 kg. Detailed comparison with
experiment in Prakash et al. (2007) demonstrated a high degree of accuracy of the
particle flow
predictions for the distribution and rate of transport of the pellets using this model. In this modelling
variant of the coupled SPH-DEM model, the particulates were discretised into hundreds of
SPH particles allowing the detailed pressure and stress distributions over the particle
surfaces from the fluid to be calculated. This is similar to the approach used by Potapov
et al. (2001) except performed here in 3D and on a much larger scale. This approach is more
735
accurate and avoids the need to use an empirical drag law to couple the fluids and solids as
is done conventionally for FV-DEM coupling. Its drawback is the greater expense of
modelling all the SPH sub-elements of the particulates. This is feasible if the particulates
are not very small and there are moderate numbers of them (1,000 are currently feasible).
Figure 25 shows the progress of the mixing and submergence of the particulates
into the fluid. At 1.0 s, the highly buoyant pellets have begun to be pulled down into
the fluid just below the free surface. The impeller starts to generate a strong central
downdraft of water leading to a bulk recirculation flow within the tank. This sucks
solid pellets down towards the impeller. By 1.5 s, the pellets have started to cluster
near the centre of the tank as a result of impeller starting to generate this recirculation
pattern. At 2.0 s, the pellets are being dragged down into the fluid with the leading

Figure 25.
Mixing and submergence
of buoyant particulates
in a tank of water driven
by a central impeller
rotating at 200 rpm
EC pellets have reached half way down to the impeller. The recirculating flow pattern
becomes fully established around 2.5 s and a significant proportion of the pellets are
26,6 being drawn down towards the impeller. At 3.0 s, pellets are being flung outwards by
the impeller and are now being re-circulated within the fluid. Very good agreement is
obtained with experiment for the distribution of solids, the critical speed submergence
and the rate of submergence. This demonstrates that coupled DEM-SPH is very viable
as a method for modelling particulate-fluid systems.
736
12. Discrete bubble – fluid flows
In all the previous application areas, the discrete entities have represented solid
particulates. It is also possible to represent bubbles using the same methods since bubble-
bubble collision is a key element of the physics that needs to be modelled for such systems.
Discrete bubbles are a good representation when the shape of the bubble is simple and does
not vary much with time. Small bubbles of less that a few millimetres diameter are
spherical due to the dominance of surface tension at these scale and do not vary in shape as
the move. These are ideally suited for modelling using discrete bubble models (Delnoij et al.,
1997, 1998). Bubbles only exist in a fluid which necessarily must be also be modelled. The
same advantages of using the SPH method for simulating the fluid coupled to the bubbles
exist as for particulates. Space limitations here preclude the description of the theory used
for the SPH model and its coupling to the bubbles. A detailed exposition is contained in
Cleary (2007b). Briefly each bubble is coupled to the fluid and vice versa using an empirical
drag law in much the same was as is conventionally used for FV-DEM models of gas-
particulate systems. Bubble collisions with each other and solids are modelled using a soft
particle linear spring model and cohesive forces are added for the bubbles when they form
dry foam (i.e rise above the free surface of the fluid).
Figure 26 shows a fairly simple example of gas bubble motion coupled to fluid being
agitated by a moving solid block. Bubbles are generated at the bottom of the tank by a
row of nozzles and rise towards the free surface. A solid block oscillates sideways
generating a sloshing motion in the tank. The bubbles are entrained by the fluid
motion, are captured by the wake where they swirl and collide, concentrating and then
separating in dense plumes which rise upwards. The dense whitish region towards the

Figure 26.
Gas bubble motion
coupled to fluid
being agitated by a
moving block
top represents dry foam with concentrated bubbles supported by bubble-bubble Industrial
collisions rising above the fluid level.
A more complex and interesting application is the modelling of beer. In this case we
particle flow
simulate the pouring of beer into a glass. This is a more complex model that also modelling
includes prediction of the transport of dissolved gas in the liquid, the nucleation of
discrete bubbles at nucleation defects on the glass surface, the ongoing diffusion of gas
from dissolved phase into the discrete phase leading to bubble growth, bubble
coalescence and surface tension induced sticking of small bubbles to their nucleation
737
sites. The details of the model are given in Cleary et al. (2007b). The filling of a beer
glass using the model with these components is shown in Figure 27.
Beer (which is a gas super-saturated liquid) starts to pour into a glass. Contact with
some bubble nucleation sites on the glass leads to the formation of microscale bubbles,
typically 30-60 microns. These grow by absorption of gas from the dissolved phase in the
fluid, but remain trapped on the surface until the combination of fluid dynamic and
buoyancy forces overcome the resisting surface tension. This is a rapid process and a
strong plume of millimetre scale bubbles surges from the initial contact point (Figures
27(b) and (c)). Once separated, the bubbles rise and continue to grow by absorbing gas
from the liquid through which they rise. Upon reaching the free surface, the bubbles form
a dense slightly sticky foam (Figure 27(c)). The high speed and asymmetric nature of the
pouring means that there is substantial internal flow of the fluid to which the bubble
plumes are coupled via the inter-phase drag. As the fluid level increases (Figure 27(e)) the
effect of the jet diminishes and the bubble plume becomes increasingly vertical as the
bubble buoyancy dominates. A thick stable foam has formed and is floating on the beer
surface. Around one third of the bubbles are beneath the surface and remain wetted with
the remaining being supported above the surface and being considered dry. The foam raft

Figure 27.
Simulation of beer, with
nucleation of bubbles
which are coupled to the
fluid flow and creating
foam at the free surface
EC is strongly disturbed by the entry of the fluid jet into the system (Figures 27(d) and (e)).
Shortly after the pouring ceases and the system comes almost to rest. The rate of bubble
26,6 generation declines sharply as the fluid adjacent to the nucleation sites on the glass
become gas poor. The jet led to rapid bubble generation both because it added new gas
rich liquid and because it strongly mixing the existing fluid bringing high gas liquid in
contact with the nucleation sites. Without this stirring the gas transport rapidly becomes
purely diffusive, which is comparatively very slow, leading to a significant decline in the
738 rate of bubble generation and of bubble growth. Over time the foam coarsens by
coalescence of bubbles and some bubbles burst leading to a gradual decay of the foam
layer. This process was modelled using a combination of SPH for the fluid and the
dissolved gas transport and DEM for the discrete bubble phase. Qualitative validation of
this application is straight forward for the reader to perform.

13. Conclusions
A consistent theme of this paper is the demonstration that the particle shape matters! It
cannot be properly treated using fixes such as artificial torsional resistance. It is a real and
important component of the physics and controls the strength of granular materials, when
it will fail and flow and when it will remain stationary. It also controls the shear and
dilation in flowing regions, the void fraction in granular solids and the interaction of the
linear and rotational dynamics of the granular material. Two dimensional DEM using
circular particles and idealised surface representations is easy and often gives qualitatively
plausible flow predictions but these generally possess significant quantitative errors.
Modelling in 3D and using accurate representations of boundary geometry has been easy
to do for several years and should be the minimum type of model used from this time. But
to achieve quantitative predictive accuracy at least the largest scale aspects of particle
shape, namely its blockiness and the aspect ratios of the particles need to be included.
There is a growing body of evidence that if the particles and boundary geometry are well
represented then good quantitative predictive accuracy can be obtained by DEM for
cohesionless particle flows. This is essential in order to use DEM to explore processes and
flows that cannot be measured or performed in the laboratory and for DEM to be used as
an effective tool to design new equipment long before a prototype is even fabricated.
Increasingly, DEM can be used for modelling cohesive particle systems, particle-gas
and particle-fluid system, bubbly flows and gas-liquid-particulate systems including
complex physics for bubble nucleation, growth, coalescence, interaction with particulates,
heat transfer and chemical reactions. The challenges in these problems include the
characterization of the complex physics that occurs, typically on small spatial scale and the
development of suitable models that allow these behaviours to be predicted but which are
computationally feasible. Substantial challenges also exist in developing material
characterization tests that are needed to provide all the input data for these models. How
does one measure the cohesive force distribution on half clay-half blasted rock overburden
when the particle sizes are up to meters and the shear rates during processing are many
meters per second? The quantitative success of DEM for systems with these additional
forces and physics will depend not only on the quality of the models being developed but
also on the quality of the data that is needed to drive them. Overall, DEM is well advanced
in being able to be used for understanding complex particle and particle-fluid systems.
Substantial progress has been made in understanding and modelling industrial
processes. Examples of separation, mixing, comminution, storage and unloading, transport,
excavation, fluid-particulate flow, fluid-bubble flow and gas-liquid-particulate flows have
been described in this paper. Increasingly, DEM is being used as a predictive tool for these
systems with significant amounts of quantitative data being available to help create insights Industrial
into the complex physical processes that occur and how they can be modified and optimised. particle flow
A key driver for the increasing scale and complexity of the DEM based models that
are possible is the continuous increase in computer power. Our current practical upper modelling
limit for DEM is 12-16 million particles on a single core single desktop cpu. For coarse
systems (cm-m scale) this can be sufficient to reach full process scale. For fine materials
(micron-mm) this is still small compared to that required to reach the process scale. 739
With each year the limits increase, but there will be large scale systems of fine particles
that remain beyond the traditional DEM approach. So scaled down DEM, hybrid
discrete-continuum models and multiscale models are needed for these systems.
The rate of increase in computer capacity indicates the likely scales of modelling
that will be possible in the future, particularly with the advent of multi-core and
commodity multi-cpu servers. From our direct experience over the past 15 years we
have seen model size and complexity grow:
. 1995 (1,000 circular particles in 2D);
. 2001 (250,000 spheres in 3D);
. 2005 (1 million non-spherical particles in 3D);
. 2007 (15 million particles for some applications); and
. 2015 (1 billion? with high quality physics, complete geometry . . .)
With a billion or more particles there are a lot of systems that can be modelling
completely and many interesting questions that can be answered.

References
Barker, G.C. (1994), ‘‘Computer simulations of granular materials’’, in Anita M. (Ed.), Granular
Matter: An Interdisciplinary Approach, Springer-Verlag, New York, NY.
Bertrand, F., Leclaire, L.A. and Levecque, G. (2005), ‘‘DEM based models for the mixing of
granular materials’’, Chemical Engineering Science, Vol. 60, pp. 2517-31.
Bwalya, M.M. and Moys, M.H. (2003), ‘‘The use of DEM in predicting grinding rate’’, in Lorenzen,
L. and Bradshaw, D.J. (Eds), Proceedings of the XXII International Mineral Processing
Congress, pp. 1612-17.
Campbell, C.S. (1990), ‘‘Rapid granular flows’’, Annual Review of Fluid Mechanics, Vol. 22, pp. 57-92.
Campbell, C.S., Cleary, P.W. and Hopkins, M.A. (1995), ‘‘Large scale landslide simulations: global
deformation, velocities and basal friction’’, Journal of Geophysical Research, Vol. 100 No. B5,
pp. 8267-83.
Cleary, P.W. (1998a), ‘‘Discrete element modelling of industrial granular flow applications’’, Task
Quarterly: Scientific Bulletin, Vol. 2, pp. 385-416.
Cleary, P.W. (1998b), ‘‘Predicting charge motion, power draw, segregation, wear and particle breakage
in ball mills using discrete element methods’’, Minerals Engineering, Vol. 11, pp. 1061-80.
Cleary, P.W. (1998c), ‘‘The filling of dragline buckets’’, Mathematical Engineering in Industry,
Vol. 7, pp. 1-24.
Cleary, P.W. (2000), ‘‘DEM simulation of industrial particle flows: case studies of dragline excavators,
mixing in tumblers and centrifugal mills’’, Powder Technology, Vol. 109, pp. 83-104.
Cleary, P.W. (2001a), ‘‘Modelling comminution devices using DEM’’, International Journal for
Numerical and Analytical Methods in Geomechanics, Vol. 25, pp. 83-105.
EC Cleary, P.W. (2001b), ‘‘Charge behaviour and power consumption in ball mills: sensitivity to mill
operating conditions, liner geometry and charge composition’’, International Journal of
26,6 Mineral Processing, Vol. 63, pp. 79-114.
Cleary, P.W. (2001c), ‘‘Recent advances in DEM modelling of tumbling mills’’, Minerals
Engineering, Vol. 14, pp. 1295-319.
Cleary, P.W. (2004), ‘‘Large scale industrial DEM modelling’’, Engineering Competition, Vol. 21,
pp. 169-204.
740
Cleary, P.W. and Campbell, C.S. (1993), ‘‘Self-lubrication for long run-out landslides: examination
by computer simulation’’, Journal of Geophysical Research, Vol. 98 No. B12, pp. 21911-24.
Cleary, P.W. and Hoyer, D. (2000), ‘‘Centrifugal mill charge motion and power draw: comparison of
DEM predictions with experiment’’, International Journal of Mineral Processing, Vol. 59,
pp. 131-48.
Cleary, P.W. and Prakash, M. (2004), ‘‘Smooth particle hydrodynamics and discrete element
modelling: potential in the environmental sciences’’, Philosophical Transactions A, Vol. 362,
pp. 2003-30.
Cleary, P.W. and Sawley, M.L. (2002), ‘‘DEM modelling of industrial granular flows: 3D case
studies and the effect of particle shape on hopper discharge’’, Applied Math Modeling,
Vol. 26, pp. 89-111.
Cleary, P.W. and Sinnott, M.D. (2008), ‘‘Assessing mixing characteristics of particle mixing and
granulation devices’’, Particuology, Vol. 6 No. 6, pp. 419-44.
Cleary, P.W., Laurent, B.F.C. and Bridgewater, J. (2002), ‘‘DEM prediction of flow patterns and
mixing rates in a ploughshare mixer’’, Proceedings of the World Congress Particle
Technology 4, CD paper 715.
Cleary, P.W., Metcalfe, G. and Liffman, K. (1998), ‘‘How well do discrete element granular flow
models capture the essentials of mixing and segregation processes?’’, Applied
Mathematical Modelling, Vol. 22, pp. 995-1008.
Cleary, P.W., Morrison, R. and Morrell, S. (2003), ‘‘Comparison of DEM and experiment for a scale
model SAG mill’’, International Journal of Mineral Processing, Vol. 68, pp. 129-65.
Cleary, P.W., Pyo, S.H., Prakash, M. and Koo, B.K. (2007b), ‘‘Bubbling and frothing liquids’’, ACM
Transaction on Graphics, Vol. 26, Article No. 97.
Cleary, P.W., Domike, R.D., Ortiz, C., Robinson, G. and Cooney, C.L. (2005), ‘‘The application of
force microscopy measurements to predict bulk flow of a fine powder using DEM’’,
Proceedings of the 7th Congress Chemical Engineering 2005.
Cleary, P.W., Prakash, M., Ha, J., Stokes, N. and Scott, C. (2007a), ‘‘Smooth particle hydrodynamics:
status and future potential’’, Progress in Computational Fluid Dynamics, Vol. 7, pp. 70-90.
Cundall, P.A. (1988), ‘‘Formulation of a three-dimensional distinct element model. Part I.A.
Scheme to detect and represent contacts in a system composed of many polyhedral
blocks’’, International Journal of Rock mechanics and Mining Sciences, Vol. 25, pp. 107-16.
Cundall, P.A. and Strack, O.D.L. (1979), ‘‘A discrete numerical model for granular assemblies’’,
Geotechnique, Vol. 29, pp. 47-65.
Debroux, F. and Cleary, P.W. (2001), ‘‘Characterising the angles of failure and repose of
avalanching granular material using the discrete element method’’, Proceedings of the 6th
World Congress of Chemical Engineering, paper 1537.
Delnoij, E., Kuipers, J.A.M. and van Swaaij, W.P.M. (1997), ‘‘Computational fluid dynamics
applied to gas-liquid contactors’’, Chemical Engineering Science, Vol. 52, pp. 3623-38.
Delnoij, E., Lammers, F.A., Kuipers, J.A.M. and van Swaaij, W.P.M. (1998), ‘‘Dynamic simulation
of dispersed gas-liquid two-phase flow using a discrete bubble model’’, Chemical
Engineering Science, Vol. 53, pp. 1927-40.
Dewicki, G. (2003), ‘‘Bulk material handling and processing – numerical techniques and Industrial
simulation of granular material’’, Bulk Solids Handling, Vol. 23.
particle flow
Dewicki, G. and Mustoe, G. (2002), ‘‘Bulk material belt conveyor transfer point simulation of material
flow using DEM’’, Third International Conference on DEM, Santa Fe, New Mexico, NY. modelling
Djordjevic, N., Shi, F.N. and Morrison, R.D. (2003), ‘‘Applying discrete element modelling to
vertical and horizontal shaft impact crushers’’, Minerals Engineering, Vol. 16, pp. 983-91.
Dziugys, A. and Peters, B. (2001), ‘‘An approach to simulate the motion of spherical and non- 741
spherical fuel particles in combustion chambers’’, Granular Matter, Vol. 3, pp. 231-65.
Gröger, T. and Katterfeld, A. (2006), ‘‘On the numerical calibration of discrete element models for
the simulation of bulk solids’’, 16th European Symposium on Computer Aided Process
Engineering and 9th International Symposium on Process Systems Engineering, pp. 533-8.
Haff, P.K. and Werner, B.T. (1986), ‘‘Computer simulation of the mechanical sorting of
grains’’,Powder Technology, Vol. 48 No. 3, pp. 239-45.
Herbst, J.A. and Nordell, L. (2001), ‘‘Optimization of the design of sag mill internals using high
fidelity simulation’’, in Barratt, A. and Mular, IV (Eds), Proceedings of the SAG
Conference,pp. 150-64.
Holst, J.M., Rotter, J.M., Ooi, J.Y. and Rong, G.H. (1999), ‘‘Numerical modelling of silo filling. II:
discrete element analysis’’, Journal of Mechanical Engineering, Vol. 125, pp. 94-110.
Hoomans, B.P.B., Kuipers, J.A.M., Briels, W.J., Van Swaaij, P.M. (1996), ‘‘Discrete particle
simulation of bubble and slug formation in a two-dimensional gas-fluidised bed: a hard-
sphere approach’’, Chemical Engineering Science, Vol. 51, pp. 99-118.
Hopkins, M.A., Hibler, W.D. and Flato, G.M. (1991), ‘‘On the numerical simulation of the sea ice
ridging process’’, Journal of Geophysical Research, Vol. 96, C3, pp. 4809-20.
Hlungwani, O., Rikhotso, J., Dong, H. and Moyes, M.H.,(2003), ‘‘Further validation of DEM modelling
of milling effects of liner profile and mill speed’’, Minerals Engineering, Vol. 16, pp. 993-8.
Kafui, K.D., Thornton, C. and Adams, M.J. (2002), ‘‘Discrete particle-continuum fluid modelling of
gas-solid fluidized beds’’, Chemical Engineering Science, Vol. 57, pp. 2395-410.
Ketterhagen, W.R., Curtis, J.S., Wassgren, C.R., Kong, A., Narayan, P.J., Hancock, B.C. (2007),
‘‘Granular segregation in discharging cylindrical hoppers: a discrete element and
experimental study’’, Chemical Engineering Science, Vol. 62, pp. 6423-39.
Kuo, H.P., Knight, P.C., Parker, D.J., Adams, M.J. and Seville, J.P.K. (2004), ‘‘Discrete element
simulations of a high-shear mixer’’, Advanced Powder Technology, Vol. 15, pp. 297-309.
Kuo, H.P., Knight, P.C., Parker, D.J., Tsuji, Y., Adams, M.J. and Seville, J.P.K. (2002), ‘‘The influence
of DEM simulation parameters on the particle behaviour in a V-mixer’’, Chemical
Engineering Science, Vol. 57, pp. 3621-38.
Langston, P.A., Tuzun, U. and Heyes, D.M. (1995), ‘‘Discrete element simulation of granular flow
in 2D and 3D hoppers: dependence of discharge rate and wall stress on particle
interactions’’, Chemical Engineering Science, Vol. 50, p. 967.
Langston, P.A. Al-Awamleh, M.A., Fraige, F.Y. and Asmar, B.A. (2004), ‘‘Distinct element modelling
of non-spherical frictionless particle flow’’, Chemical Engineering Science, Vol. 59, pp. 425-35.
Latham, J.P. and Munjiza, A. (2004), ‘‘The modelling of particle systems with real shapes’’,
Philosophical Transactions of the Royal Society, Vol. 362, pp. 1953-73.
Latham, J.P., Munjiza, A., Garcia, X., Xiang, J. and Guises, R. (2008), ‘‘Three dimensional particle
shape acquisition and use of shape library for DEM and FEM/DEM simulation’’, Minerals
Engineering, Vol. 21 No. 11, pp. 797-805.
Laurent, B.F.C., Bridgwater J. and Parker, D.J. (2000), ‘‘Motion in a particle bed agitated by a single
blade’’, AIChE Journal, Vol. 46, pp. 1723-34.
EC Lemieux, M., Bertrand, F., Chaouki, J. and Gosselin, P. (2007), ‘‘Comparative study of the mixing of
free-flowing particles in a V-blender and a bin-blender’’, Chemical Engineering Science,
26,6 Vol. 62, pp. 1783-802.
Lemieux, M., Leonard, G., Doucet, J., Leclaire, L.A., Viens, F., Chaouki, J. and Bertrand, F. (2008),
‘‘Large-scale numerical investigation of solids mixing in a V-blender using the discrete
element method’’, Powder Technology, Vol. 181, pp. 205-16.
Li, J., Webb, C., Pandiella, S.S. and Campbell, G.M. (2002), ‘‘A numerical simulation of separation
742 of crop seeds by screening – effect of particle bed depth’’, Transactions in Chemical
Engineering, Vol. 80, pp. 109-17.
Li, J., Webb, C., Pandiella, S.S. and Campbell, G.M. (2003), ‘‘Discrete particle motion on sieves – a
numerical study using the DEM simulation’’, Powder Technology, Vol. 133, pp. 190-202.
Lian, G., Thornton, C. and Adams, M.J. (1993), ‘‘A theoretical study of the liquid bridge forces
between two rigid spherical bodies’’, Journal of Colliod and Interface Science, Vol. 161, pp.
138-47.
Lian, G., Thornton, C. and Adams, M.J. (1998), ‘‘Discrete particle simulation of agglomerate
impact coalescence’’, Chemical Engineering Science, Vol. 53, pp. 3381-91.
Lindqvist, M. and Evertsson, C.M. (2003), ‘‘Prediction of worn geometry in cone crushers’’,
Minerals Engineering, Vol. 16, pp. 1355-61.
McClung, D. and Shaerer, P. (1993), The Avalanche Handbook, 2nd ed., The Mountaineers,
Seattle, WA.
Middleton, G.V. and Wilcock, P.R. (1994), Mechanics in the Earth and Environmental Sciences,
Cambridge University Press, Cambridge.
Mikami, T., Kamiya, H. and Horio, M. (1998), ‘‘Numerical simulation of cohesive powder
behaviour in a fluidized bed’’, Chemical Engineering Science, Vol. 53, pp. 1927-40.
Mishra, B.K. and Rajamani, R.J. (1992), ‘‘The discrete element method for the simulation of ball
mills’’, Applied Mathematical Modelling, Vol. 16, pp. 598-604.
Mishra, B.K. and Rajamani, R.K. (1994), ‘‘Simulation of charge motion in ball mills. Part 1:
experimental verifications’’, International Journal of Mineral Processing, Vol. 40, pp. 171-86.
Morrison, R.D. and Cleary, P.W. (2004), ‘‘Using DEM to model ore breakage within a pilot scale
SAG mill’’, Minerals Engineering, Vol. 17, pp. 1117-24.
Morrison, R.D. and Cleary, P.W. (2008), ‘‘Towards a virtual comminution machine’’, Minerals
Engineering, Vol. 21, pp. 770-81.
Morrison, R., Cleary, P.W. and Valery, W. (2001), ‘‘Comparing power and performance trends from
DEM and JK modelling’’, Proceedings of the SAG 2001, pp. 4-284.
Munjiza, A. and Latham, J.P. (2004), ‘‘Comparison of experimental and numerical results for
gravitational deposition of identical cubes’’, The International Journal of Engineering
Computation, Vol. 21, pp. 249-65.
Munjiza, A., Latham, J.P. and John, N.W.M. (2003), ‘‘3D dynamics of discrete element systems
comprising irregular discrete elements’’, International Journal for Numerical Methods in
Engineering, Vol. 56, pp. 35-5.
Pöschel, T. and Buchholtz, V. (1995), ‘‘Complex flow of granular material in a rotating cylinder’’,
Chaos, Solitons and Fractals, Vol. 5, pp. 1901-12.
Potapov, A.V. and Campbell, C.S. (1996a), ‘‘Computer simulation of hopper flows’’, Physics of
Fluids, Vol. 8, pp. 2884-94.
Potapov, A.V. and Campbell, C.S. (1996b), ‘‘A three dimensional simulation of brittle solid
fracture’’, International Journal of Modern Physics, Vol. C7, pp. 717-29.
Potapov, A.V., Hunt, M.L. and Campbell, C.S. (2001), ‘‘Liquid–solid flows using smoothed particle
hydrodynamics and the discrete element method’’, Powder Technology, Vol. 116, pp. 204-13.
Prakash, M., Cleary, P.W., Noui-Mehidi, M.N., Blackburn, H. and Brooks, G. (2007), ‘‘Simulation of Industrial
suspension of solids in a liquid in a mixing tank using SPH and comparison with physical
modeling experiments’’, Progress in Computational Fluid Dynamics, Vol. 7, pp. 91-100. particle flow
Rajamani, R.K. and Mishra, B.K. (1996), ‘‘Dynamics of ball and rock charge in sag mills’’, modelling
Proceedings of the SAG 1996, Department of Mining and Mineral Process Engineering,
University of British Columbia.
Ristow, G.H. (1994), ‘‘Granular dynamics: a review about recent molecular dynamics simulations’’,
Annual Reviews of Computational Physics, Vol. 1, pp. 275-308. 743
Rothenburg, L. and Bathurst, R.J. (1991), ‘‘Numerical simulation of idealised granular assemblies
with plane elliptical particles’’, Computers and Geotechnics, Vol. 11, pp. 315-29.
Schäfer, J., Dippel, S. and Wolf, D.E. (1996), ‘‘Force schemes in simulation of granular material’’,
Journal of Physique I France, Vol. 6, p. 5.
Stewart, R.L., Bridgewater, J. and Parker, D.J. (2001a), ‘‘Granular flow over a flat-bladed stirrer’’,
Chemical Engineering Science, Vol. 56, pp. 4257-71.
Stewart, R.L., Bridgewater, J., Zhou, Y.C. and Yu, A.B. (2001b), ‘‘Simulated and measured flow of
granules in a bladed mixer – a detailed comparison’’, Chemical Engineering Science, Vol. 56,
pp. 5457-71.
Thornton, C. (1999), Mechanics of Granular Materials – An Introduction, in Oda and Iwashita
(Eds), Balkema, pp. 207-17.
Thornton, C. and Yin, K.K. (1991), ‘‘Impact of elastic spheres with and without adhesion’’, Powder
Technology, Vol. 65, pp. 153-66.
Thornton, C., Yin, K.K. and Adams, M.J. (1996), ‘‘Numerical simulation of the impact fracture and
fragmentation of agglomerates, Journal of Physics D: Applied, Vol. 29, pp. 424-35.
Tsuji, Y., Kawaguchi, T. and Tanaka, T. (1993), ‘‘Discrete particle simulation of two-dimensional
fluidized bed’’, Powder Technology, Vol. 77, pp. 79-87.
Tsuji, Y., Tanaka, T. and Ishida, T. (1992), ‘‘Lagrangian numerical simulation of plug flow of
cohesionless particles in a horizontal pipe’’, Powder Technology, Vol. 71, pp. 239-50.
Walton, O.R. (1994), ‘‘Numerical simulation of inelastic frictional particle-particle interaction’’, in
Roco, M.C. (Ed.), Chapter 25, Particulate Two-Phase Flow, pp. 884-911.
Williams, J.R. and Pentland, A. (1992), ‘‘Super-quadrics and modal dynamics for discrete
elements in interactive design’’, International Journal of Computer Aided Engineering
Computations, Vol. 9, p. 115.
Xu, B.H. and Yu, A.B. (1997), ‘‘Numerical simulation of the gas-solid flow in a fluidized bed by
combining discrete particle method with computational fluid dynamics’’, Chemical
Engineering Science, Vol. 52, pp. 2785-809.
Yen, K.Z.Y. and Chaki, T.K. (1992), ‘‘A dynamic simulation of particle rearrangement in powder
packings with realistic interactions’’, Journal of Applied Physics, Vol. 71, pp. 3164-73.
Zhang, X. and Vu-Quoc, L. (2000), ‘‘Simulation of chute flow of soybeans using an improved
tangential force-displacement model’’, Mechanics of Materials, Vol. 32, pp. 115-29.
Zhou, Y.C., Yu, A.B. and Bridgewater, J. (2003), ‘‘Segregation of binary mixture of particles in a
bladed mixer’’, Journal of Chemical Technology & Biotechnology, Vol. 78, pp. 187-93.

Corresponding author
Paul W. Cleary can be contacted at: [email protected]

To purchase reprints of this article please e-mail: [email protected]

Or visit our web site for further details: www.emeraldinsight.com/reprints