Minerals Engineering 79 (2015) 239–260

Contents lists available at ScienceDirect

Minerals Engineering
journal homepage: www.elsevier.com/locate/mineng

Computational prediction of performance for a full scale

Isamill: Part 2 – Wet models of charge and slurry transport
Paul W. Cleary ⇑, Matt D. Sinnott
CSIRO Digital Productivity Flagship, Locked Bag 10, Clayton South, Vic 3169, Australia

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

Article history: The Isamill is a horizontal stirred media mill used for fine and ultrafine grinding of slurry transported rock
Received 3 January 2015 particles. The charge motion is analysed using two different approaches, (1) a fluid only model, and (2) a
Revised 14 April 2015 1-way coupled DEM + SPH model. The flow pattern when the classifier is closed is regular with a pair of
Accepted 15 April 2015
oppositely oriented vortices between each pair of grinding discs. A strong radial outflow from the middle
Available online 30 April 2015
of the classifier is generated by the high centrifugal force which creates a pair of toroidal vortices at the
discharge end of the mill. The classifier, when open, acts as a pump drawing slurry axially along the mill.
Keywords:
It enters the classifier through the holes in its end plate and is then forced radially outward by rotational
DEM
SPH
acceleration of the classifier cage. The enhanced outflow significantly strengthens the large toroidal
Isamill vortices on the outside of the classifier. This produces a strong retrograde annular flow along the mill
Comminution shell that penetrates a significant distance back into the grinding chamber. The effect of the classifier
Slurry transport is significant and strongly influences the flow over much of the mill and controls slurry (feed and
product) transport and discharge. The predictions of the different models are qualitatively similar but
with important differences including the fluid only model predicting higher flow speeds because it can-
not capture the strong slip between the media and the grinding discs. The strength of the axial transport
is strongly dependent on the slurry viscosity. A critical viscosity can be identified above which there is
insufficient axial transport to enable mill operation.
Crown Copyright Ó 2015 Published by Elsevier Ltd. All rights reserved.

1. Introduction 2003; de Kretser et al., 2009). In this paper, we will be using com-
putational models to help understand the operation of an Isamill
Milling is often operated as a wet process by addition of water. when operating wet.
This is done to improve the transport of fine material by enabling it For computational modelling of slurry there are two basic
to form a slurry. Milling of a slurry suppresses dust generation approaches that can be considered. Firstly, particles can be aver-
(which has environmental and health benefits and acts to reduce aged to give a continuum representation of the slurry which then
the loss of product material) and modifies the grinding conditions has a complex rheological behaviour. These governing equations
within the mill (by constraining the fine particles within the vis- are typically solved using a Finite Volume (FV) or Finite Element
cous slurry). Slurry transport is much more efficient than air trans- (FEM) method. This type of method can be highly effective when
port for removing ore progeny and finished product (sufficiently the particle sizes are small compared to other length scales in
comminuted particles) from a mill. Not all materials and locations the process and when inter-particle contact or collisions are not
support wet milling. Some materials, such as cement and refrac- a dominant manner of momentum transport. Examples of such
tory gold ores, need to be milled in dry conditions so air sweeping modelling include erosion of pipes during slurry flow (Brown,
of such mills is typically used. In some locations, scarcity of water 2002; Zhang et al., 2007), pneumatic and slurry flows (Solnordal
means that dry processing can be preferred for economic and envi- et al., 2013), flow around and consequent erosion of a cylinder
ronmental reasons even though it may be less efficient from a pro- (Lester et al., 2010) and impingement of a slurry jet on a flat plate
cess perspective and often the advantages of dry processing are (Gnanavelu et al., 2011).
negated by the requirement to use wet separation processes to The second approach is to resolve the individual particles in the
achieve the required efficiencies (Napier-Munn and Morrison, simulation and to calculate the effect of the collisions between par-
ticles and/or with the enclosing solid geometry. The Discrete
⇑ Corresponding author. Tel.: +61 3 9545 8005. Element Method (DEM) enables calculation of such flows
E-mail address: [email protected] (P.W. Cleary). (Cundall and Strack, 1979) and has been used extensively for mill

http://dx.doi.org/10.1016/j.mineng.2015.04.013
0892-6875/Crown Copyright Ó 2015 Published by Elsevier Ltd. All rights reserved.
240 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

modelling (Mishra and Rajamani, 1992, 1994; Rajamani and Jayasundara et al. (2010, 2011a, 2011b) used a 1.4 l, three disc per-
Mishra, 1996; Inoue and Okaya, 1996; Cleary, 1998a, 2001a, spex mill with 110 mm inner diameter and 90 mm diameter disks.
2001b, 2001c, 2004, 2009; Herbst and Nordell, 2001; Morrison Jayasundara et al. (2006, 2008) used a slightly larger mill with
and Cleary, 2004, 2008; Djordjevic, 2003, 2005; Kalala et al., 120 mm inner diameter and 100 mm diameter disks. Across this
2008; Lee et al., 2010; Powell et al., 2011; Carvalho and Tavares, series of studies, media of 2–5 mm glass beads were used and
2011; and papers in the special issue edited by Cleary and impellor speeds up to 2000 rpm and fill levels up to 80% were con-
Morrison, 2008). A recent review of the use of DEM in comminu- sidered. Cleary et al. (2008) used a DEM model with bi-modal
tion science is presented by Weerasekara et al. (2013). Coupling charge to explore the radial distribution of ore and media in an
of fluids to the motion of such solid particles can be performed Isamill. The ore was found to easily segregate towards the shell
in a range of ways using several different solution methods for because of the very strong centrifugal force filling the pore space
the fluid component. The most common is to use grid based meth- of the charge starting from the shell and moving inwards.
ods such as FV (as used by Tsuji et al., 1993; Xu and Yu, 1997; One of the first CFD models of a stirred mill was carried out by
Kawaguchi et al., 2000; Kafui et al., 2002; Hilton et al., 2010; Blecher et al. (1996) where they considered the axi-symmetric
Hilton and Cleary, 2011 and many since then often for fluidised flow around a solid disc. The fluid motion was followed and it
bed modelling). This is typically used when the fluid is a gas, the was shown that 80–90% of the energy dissipation occurred in
geometry is not complicated and there are no free surfaces to pre- 10–20% of the mill volume. Later, Lane (1999) confirmed these
dict. This type of coupled method is termed unresolved since the findings. Gers et al. (2010) considered the local hydrodynamics
fluid flow between the individual particles is not resolved, but by solving the Navier–Stokes equations for incompressible flow
rather the particle phase is locally averaged to give porosity and in a Couette flow. They demonstrated the fluid flow was composed
velocity distributions that can be used in continuum multiphase of a number of toroidal vortices and the most energetic collisions
models in conjunction with empirical drag laws that provide the occurred in the gap between the discs and the cylinder walls.
coupling force. This is a very effective approach for many systems Lane et al. (2002, 2005) have also recently modelled fluid motion
where the fine detail of the fluid flow is not critical to predicting in stirred tanks although their focus was on multiphase flow, i.e.,
the emergent behaviour of the system. how the gas phase is dispersed through the liquid in the stirred
A more recent alternative that has some specific advantages for tank.
modelling slurry in the presence of free surfaces and complex More recently, a coupled fluid dynamical and DEM study for the
geometries are particle methods. Again, a number of different Isamill has been carried out by Jayasundara et al. (2009). The solid
methods have been used by different researchers. For applications particles motion was modelled using DEM coupled to a multiphase
involving permeability calculation and for medium size computa- Navier–Stokes solution (finite volume based CFD) using CFX with a
tional problems, the Lattice Boltzmann method can be used. RANS turbulence model. The coupling of the phases was compli-
Examples for permeability calculation include Boek and Venturoli cated by the motion of the impellor which creates challenges for
(2010), Manwart et al. (2002), Ramstad et al. (2012) and for fluid grid based CFD. The coupling was achieved by first solving the solid
flow around bodies or particles see Feng et al. (2010), Phillips particle motion (using DEM) to give a porosity and volumetric fluid
and Roberts (2011) and Silva et al. (2015). The particle method that drag in each CFD cell and then the CFD solver uses this information
we will use here for the fluid solution is SPH (Smoothed Particle to calculate the fluid flow field. This fluid flow field was then fed
Hydrodynamics). This method was first used for free surface fluid back into the DEM model (as drag forces on solid particles) to cal-
flow by Monaghan (1994). Examples and details can be found in culate the particle motion. This procedure was reported as being
Monaghan (2005) and Cleary et al. (2007). The coupling of SPH very computationally intensive. Jayasundara et al. (2009) therefore
to DEM was first performed using a fully resolved method (mean- used a number of simplifying approximations so as to obtain
ing that the detailed fluid flow between the particles is resolved) results within a reasonable time frame, e.g., they simulated a thin
by Potapov et al. (2001). Prakash et al. (2007) gives a more recent axial slice of a 20 l pilot scale Isamill centred on one rotating disc as
application to particle fluid mixing using a fully resolved well as using a simplified coupling procedure. The presence of fluid
DEM + SPH coupled method. For most large scale systems, the res- flow on the slurry particles resulted in increasing flow in the axial
olution required to resolve the particle scale fluid flow is pro- direction due to fluid drag and also limited the particles motion
hibitive. In this case, an unresolved coupling of the DEM and SPH towards the outer walls. Nonetheless, this was the first wet DEM
methods is appropriate. Such a 1-way coupled unresolved method model of an Isamill and provides the first information on fluid flow
was proposed by Cleary et al. (2006). This has been used for full in this mill. Jayasundara et al. (2011c) then used this model to
scale three dimensional simulation of SAG mills (Cleary and compare to PEPT (Positron Emission Particle Tracking) experi-
Morrison, 2011, 2012); tower mill (Sinnott et al., 2011) and vibrat- ments. PEPT experiments for an M4 laboratory scale Isamill have
ing screens (Fernandez et al., 2011). The use of 1-way coupling is been reported by van der Westhuizen et al. (2011). Jayasundara
possible for such densely packed mills because the media domi- et al. (2012) later investigated the effect of media size for the
nates the motion of the charge with the slurry predominantly wet DEM mill model using a three disc 1.4 l laboratory scale
being a dependent flow driven by the media motion. This is not Isamill.
universally true and can become important in regions where the In a more spatially restricted study Gers et al. (2010) used DNS
media becomes less densely packed. simulation of fluid flow around the edge of a disc of a stirred media
The Isamill has been subject to investigation using computa- mill and investigated the effect of rheology, specifically Newtonian,
tional models for a decade, but most such modelling has been per- shear thinning and shear thickening on the presence of secondary
formed with the assumption of dry conditions (or that the fluid has vortices. More recently, Yamada and Sakai (2013) used a coupled
negligible influence). The largest body of such work is by the group DEM and MPS model of a bead mill with an LES approach to calcu-
of Yu and co-workers who have carried out a number of DEM stud- late an eddy viscosity.
ies on the IsaMill (Yang et al., 2006; Jayasundara et al., 2006, 2008, In this paper, we use SPH based models to explore the beha-
2010, 2011a,b) in which they tracked granular particle trajectories, viour of a full industrial scale M10000 Isamill operating as a wet
particle distributions throughout the mill, collision energies and mill. This is the second part of a study of the M10000 Isamill with
frequencies as a function of velocities as well as compared simula- the first part (Cleary et al., 2015) dealing with DEM only predic-
tion results to lab-scale experiments. Two different mill designs tions of dry operation of the Isamill. Here we use two types of
were used for these experiments. Yang et al. (2006) and model:
 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260 241

1. SPH only model where the SPH pseudo-fluid represent a mix- A quasi-compressible pressure formulation is used based on an
ture of solid media and slurry. This is a traditional CFD style equation of state which gives the relationship between particle
model but using a particle method for the numerical solution density and fluid pressure. A suitable one is:
process.  c 
2. The 1-way coupled DEM + SPH model of Cleary et al. (2006)
q
P ¼ P0 1 ; ð3Þ
which enables the particulate component of the charge to be q0
fully resolved and to then predict how this controls the slurry where q0 is the reference density and c = 7. This pressure is then
motion. used in the SPH momentum equation (Eq. (5)) to give the particle
motion. The pressure scale factor P0 is given by:
The aim of this study is to help better understand several criti-
cal aspects of the Isamill performance: cP0
¼ 100V 2 ¼ c2s ; ð4Þ
q0
1. Motion of slurry and charge in the grinding chamber. where V is the characteristic or maximum fluid velocity. This
2. Action of the classifier on the slurry.
ensures that the density variation is less than 1% and that the flow
3. Variation of the fluid flow and transport behaviour with can be regarded as incompressible.
viscosity.
2.2. DEM for particle modelling
Together, these provide an increased understanding of the
slurry motion and transport in an industrial scale mill which can-
The DEM method has been used extensively in mineral and pro-
not be studied in detail in any other way.
cessing industry research with suitable references given in the
introduction so details of the method are not included here. The
2. Computational methods software used in this modelling is that described and used in
Cleary (1998a, 1998c, 2004, 2009). It uses a linear spring-dashpot
2.1. SPH Method for charge and slurry modelling model, see Thornton et al. (2013) for a good summary of the con-
tact models and alternatives for inelastic collisions. An overview of
Smoothed Particle Hydrodynamics (SPH) is a powerful method the DEM model for the Isamill is given in Part 1 of this paper
for modelling complex fluid processes. It was initially developed (Cleary et al., 2015).
for astrophysical applications by Lucy (1977) and Gingold and
Monaghan (1977) with an example of early use given by Cleary 2.3. 1-way coupling of slurry flow to media
and Monaghan (1990). It was extended to model incompressible
free surface fluid flows by Monaghan (1994) and has been devel- The coupling of the fluid phase to the DEM phase is described in
oped and used extensively for industrial applications (Cleary, Cleary et al. (2006) and Cleary and Morrison (2012) and has also
1998b; Cleary et al., 2007 for details and application examples) used to predict slurry flow in tower mills (Sinnott et al., 2011)
and more generally for applications such as mixing of liquids and vibrating screens (Fernandez et al., 2011).
(Tartakovsky and Meakin, 2006; Robinson et al., 2008; Robinson The coupling of the slurry to the porous media is accomplished
and Cleary, 2011, 2012), die casting (Cleary et al., 2000; Cleary using a Darcy law drag:
and Ha, 2002), flow in mills (Cleary et al., 2006; Cleary and
Morrison, 2011, 2012; Lichter et al., 2011), dam flows (Ferrari, ðV a  V DEM Þ
F darcy ¼ e2DEM la ð5Þ
2010; Cleary et al., 2010), waterway/coastal engineering qa K DEM
(Gómez-Gesteira and Dalrymple, 2004), interfacial flows
where eDEM is the porosity (void fraction) of the porous media, la is
(Colagrossi and Landrini, 2003); porous media transport (Zhu
the fluid viscosity for SPH fluid particle a, Va is the velocity of SPH
and Fox, 2002); hydraulic Erosion (Krištof et al., 2009); fracture
fluid particle a, VDEM is the DEM solid velocity at that point and qa
(Tartakovsky and Meakin, 2005) and underwater explosion (Liu
is the fluid density of particle a.
et al., 2003).
This force from the Darcy law (Eq. (5)) is added to the standard
The SPH continuity equation, taken from Monaghan (1994), is:
SPH momentum equation (Eq. (4)). The SPH method has the advan-
dqa X tage that it can easily solve this coupled Navier–Stokes and Darcy
¼ mb ðv a  v b Þ  rW ab ; ð1Þ
dt b
law flow problem, since its pressure solution is quite robust across
the transition from free flowing fluid (where the Navier–Stokes
where qa is the density of particle a with velocity va and mb is the
terms dominate) to drag controlled flow in the porous media rep-
mass of particle b. We denote the relative position vector from par-
resenting the charge (where the Darcy drag dominates).
ticle b to particle a by rab ¼ ra  rb and the relative velocity by
The permeability of the porous media at each point of the
v ab ¼ v a  v b . The interpolation kernel with smoothing length h
charge is calculated from the porosity using the Koseny–Carman
evaluated for the distance jrab j is given as W ab ¼ Wðrab ; hÞ. This form
equation:
of the continuity equation is Galilean invariant (since the positions
and velocities appear only as differences), has good numerical con- e3
servation properties, and is not affected by free surfaces or density
K¼ : ð6Þ
CTð1  eÞ2 S2
discontinuities. It is a very useful formulation for predicting free
surface flows. where K is the permeability of the porous media, (1  e) is the
The momentum equation, converted to basic SPH form, (Cleary, solid fraction of the charge, C is the shape factor (typically 2–3),
1998b), becomes the acceleration for each particle a: T is the tortuosity of the fluid pathways through the charge and
X    S is the ratio of surface area to particle volume. The shape factor
dv a Pb Pa n 4la lb v ab rab
¼g mb þ  rW ; ð2Þ for spherical particles is 2.5. The tortuosity is reasonably well
dt qb q2a
2 qa qb ðla þ lb Þ r2ab þ g2 a ab
b known for packed relatively mono-sized spheres and is 2.0. The
where Pa and la are pressure and viscosity of particle a. n is a ratio of surface area to particle volume of the charge can be simply
calibration factor for the viscous term, g is a small parameter used calculated from the DEM particles and for the 15 mm media used
to smooth out the singularity at rab = 0 and g is the gravity vector. here is 400.22.
242 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

The SPH equations are formulated in terms of two variables: shaft. Fig. 1b shows a close up of one of the grinding discs. Each
disc contains a series of regularly spaced (kidney shaped) holes
 pore fluid velocity (or velocity of slurry in the pores of the through which the slurry and product can pass. The end of the
charge), and impeller is a rotary classifier (or product separator) which is shown
 bulk density of fluid (i.e. the fluid density  local porosity). in close-up in Fig. 1c. Its role is to prevent the flow of the centrifug-
ing grinding media into the discharge section of the mill whilst
The coupling process has three key steps: allowing slurry to enter the discharge region. It contains a series
of rectangular slots through which the slurry can carry the finer
 Perform a DEM simulation to predict the particle flow and aver- product out of the mill. The classifier is often operated with a finer
age the particle data onto a cylindrical grid to obtain steady screen across the slots. In the front face of the classifier there is also
state volume fraction and velocity distributions that then well a series of small holes aligned with those in the grinding discs that
characterise the charge distribution and motion. will permit the axial flow of slurry along the shaft and into the dis-
 Apply the continuum porosity and velocity information from charge region. For these SPH fluid simulations, we use an open slot
the DEM simulation in the SPH simulation. The mill geometry configuration for the rotary classifier so that the fluid slurry can
used by the two methods is the same but the charge represen- travel unimpeded through the openings in the classifier into the
tation changes from discrete to continuous. discharge section.
 Perform an SPH simulation to predict the motion and distribu- The grinding chamber of the M10000 Isamill is a cylinder with
tion of slurry in this charge using the slurry viscosity and a cou- internal radius 0.87 m and length of 4.3 m giving an enclosed vol-
pling based on the Darcy law for porous media. ume of the mill shell of 10,000 l. The grinding disc diameter is
1.4 m and they are 130 mm thick. The radial gap between the outer
3. M10000 Isamill configuration edges of the grinding discs and the inside of the shell is 170 mm.
The installed power on such a mill would usually be 3 MW. The
Here we investigate the wet charge and slurry behaviour in a net volume available for grinding is 6.95 m3. The media used is
full scale industrial M10000 Isamill. Fig. 1 shows the Isamill config- ceramic with a specific gravity 2.7. The disc-tip speed used is
uration. Fig. 1a shows the inside of the whole mill with the mill 22 m/s (from Burford and Clark, 2007), which corresponds to an
shell being transparent. The central impeller consists of eight impellor rotation rate of 300 rpm. Slurry inflow into the mill was
grinding discs which are uniformly distributed along a central neglected. Slurry discharge was allowed in some cases by having

Fig. 1. Isamill geometry used in the DEM simulations (Cleary et al., 2015), (top) impellor and classifier in place inside the mill shell, (bottom/left) close up of one grinding disc,
and (c) classifier.
 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260 243

Table 1 colouring or rendering velocity arrows in a range of ways to allow

Key attributes of a 10,000 l Isamill. the flow structures to be seen. This is in contrast to the usual visu-
Shell length 4.3 m long alisation of SPH flows (see Cleary et al., 2007 for examples) where
Shell diameter 1.8 m either the particles are drawn (which cannot easily be interpreted
Shaft diameter 525 mm in 3D) or rendering of the surface of the fluid (which for an
Grinding discs 8
Disc diameter 1400 mm
enclosed flow such as here is just the internal surfaces of the mill
Disc thickness 130 mm which is particularly uninformative).
Distance b/w disc surfaces 310 mm Fig. 2 shows the steady state combined charge and slurry
Radial gap b/w disc and shell 170 mm motion predicted by an SPH pseudo-fluid model for viscosity
Installed power 3 MW
1.0 Pa s in a vertical slice through the axis of the mill when there
is no flow through the classifier (which is closed) and no discharge
the discharge pipe open and not allowed in others by closing off from the mill. The fluid is coloured by speed, by axial speed and
this pipe. The details of the mill and operating conditions are sum- turbulent kinetic energy in the top, middle and bottom panels
marised in Table 1. respectively, so that the overall motion and the axial transport
For the SPH pseudo-fluid models of the whole charge, the rhe- can be independently observed. The highest speed is observed
ology was Newtonian and the viscosity used to represent the com- radially outside the classifier which is very effective at generating
bination of media and slurry was 1.0 Pa s. The speed of sound for a swirling motion for the fluid between it and the mill shell. The
the fluid solution was 300 m/s ensuring that the solution meets fluid speed is also high (red) between the discs and extending
the quasi-incompressible requirements. The spatial resolution for out along and around the perimeter of the discs. It decreases to
the fluid and the boundary geometry was either 20 mm or around 5 m/s (green) just beyond the ends of the discs and then
25 mm, giving numbers of particles between 600,000 and 1.1 mil- radially out towards the shell where the charge is relatively slowly
lion particles. moving (blue). There are moderate end wall effects from the feed
For the 1-way coupled DEM + SPH models, the SPH model fluid end wall that affect the flow around the first disc. The classifier
represents only the slurry of water mixed with fine particles with similarly affects the flow around the last disc while the middle
viscosities that varied from 0.01 to 10 Pa s. The spatial resolution six discs show very similar flow structures. The flow is cylindrically
for the fluid and the boundary geometry was 25 mm and the speed symmetric for the region radially outside the grinding discs (so is
of sound was 300 m/s. A single DEM configuration was used to the same at the top and bottom of the mill in the cross section
model the grinding media in all the coupled simulations reported shown). However, the flow between the discs varies circumferen-
here. It used a media-media friction coefficient of 0.4, a media-mill tially with proximity to the holes in the discs. In the cross-sectional
friction of 0.3 and a coefficient of restitution of 0.6 for media-media surface shown in Fig. 2, a row of disc holes is just approaching the
contacts and 0.5 for media-mill ones (as was used in Cleary et al., section plane at the top whilst at the bottom there is a row of holes
2015). The media fill level was 80% of the grinding chamber. The that have just passed the plane. The flow is weaker in advance of
media size used was 15 mm. The spring stiffness used was for the disc holes than behind so the charge speeds appear smaller
the contact model was 106 N/m. The SPH component of the cou- in the upper half of the image than in the lower half.
pled models contained 300,000 particles which is around half the Fig. 2b shows the axial velocity distribution in the same vertical
number in the pure fluid model because the fluid only fills the pore plane though the axis of the mill. The axial flow at radii beyond the
space between the media that is represented by a porosity field edges of the discs is quite weak compared to the flow between the
which has an average of around 0.5. discs. Again, the flow radially outside the discs is circumferentially
symmetric with a pair of weak oppositely rotating fluid vortices
(with the one just to the left of each disc coloured light blue and
4. Pseudo-fluid model of charge in the Isamill
showing fluid moving towards the feed end whilst on the other side
of the disc the fluid is coloured yellow and moves towards the dis-
With this model the charge (media and slurry) is considered
charge end). This results from outward flow of fluid from each disc
homogenised and represented as a viscous Newtonian fluid. This
which upon approaching the shell splits into streams moving in
is a similar approach to that of earlier CFD simulations of bead
opposite axial directions. The axial flow is significantly stronger
mills (Lane, 1999) except using a particle based method rather
on the left side of the first disc as the larger space to the feed end
than a traditional grid based method. Following Blecher et al.
wall allows larger recirculatory flow with fluid moving along the
(1996) and Lane (1999) we use a base case dynamic fluid viscosity
shell to the end wall and then radially inwards towards the shaft
of 1 Pa s.
with axial motion towards the holes in the discs. Adjacent to the
To enable us to understand the basic fluid flow generated by the
classifier (above and below in the view shown in Fig. 2b) are two
rotating impellor and the separate contribution of the classifier, we
pairs of oppositely directed axial flow (red and dark blue).
consider two variants
Directly beside the classifier fluid is sucked in from either side
and then forced radially outwards between the red/blue coloured
1. Classifier holes are blocked so that the contribution to the fluid
regions and upon reaching the shell then divides into two streams
swirl by the classifier is included but no flow through the clas-
– one travelling towards the discharge end wall (coloured red) and
sifier into or back from the discharge region is possible.
one back along the wall towards the discharge end. This latter ret-
2. Classifier holes are open so that the swirling flow can move
rograde flow strengthens significantly as it passes the last disc.
radially and axially in whatever way the local pressure gradi-
Between each pair of discs is again a double pair of small
ents indicate.
regions of oppositely moving fluid which correspond to a pair of
oppositely directed fluid circulations. Near the outer edge of the
4.1. Flow in the grinding chamber discs in each pair the fluid moves away from their surfaces meeting
at the mid-point between discs which creates a radially inward jet
The results presented in this sub-section for the basic flow gen- that travels towards the shaft where it then splits into streams
erated in the grinding chamber uses the classifier which is blocked moving towards the discs and the disc holes. Some fluid flows into
to through flow. For visualisation purposes the instantaneous flow the disc holes while the remainder flow radially outward to start
is averaged onto a co-rotating cylindrical grid and then shown by the recirculation process again. When there is no flow through
244 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

Fig. 2. Steady state combined charge and slurry motion predicted by an SPH only pseudo-fluid model with viscosity 1.0 Pa s, when the classifier is closed and there is no
discharge from the mill. The fluid is coloured by (a) speed, (b) axial velocity, and (c) turbulent kinetic energy. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)

the classifier (which is blocked) the flow pattern is very regular grinding potential of the charge, is moderate (green) between the
with symmetric pairs of recirculation vortices between discs. discs decreasing radially towards the shell. It is stronger around
There is reasonably strong interaction between the toroidal vor- disc 1 because of the feed end wall effects. It is strongest around
tices generated by the classifier and the vortices between the last the outside of the classifier with the peak (yellow) near to the bars
disc and the classifier. around the outside of the classifier cage.
Fig. 2c shows the matching prediction of the turbulent kinetic
energy. This is calculated from the instantaneous velocity field by 4.2. Effect of viscosity
calculating the fluctuating velocity distribution from the average
velocity field. The turbulent kinetic energy is the sum of the three The viscosity of the pseudo-fluid representing the media and
fluctuating velocity components. This shows that the turbulent the slurry has a strong affect on the flow pattern predicted using
kinetic energy, which could be considered as a proxy for the the fluid-only model. Fig. 3 shows the speed, axial velocity and
 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260 245

turbulent kinetic energy for a viscosity of 0.1 Pa s (still with the similar having nearly the same size and shape – which is due to the
classifier closed so that there is no flow through it). The high speed narrow space between classifier and shell and its relatively larger
(red) regions around and between the discs are sharply reduced in length compared to the width of the discs.
size compared to the more viscous flow (shown in Fig. 2) and are The axial speed distribution is much more similar for the two
now focussed around the disc holes. There is little motion in the viscosities (comparing Fig. 3b to Fig. 2b). This is due to the axial
fluid beyond the tips of the discs with this predominantly coloured flow being already predominantly confined to the region between
dark blue. This is a consequence of the reduced momentum diffu- the discs. The reduced momentum transfer beyond the discs there-
sion from the discs which is able to penetrate a much shorter dis- fore has little effect on the axial transport. The structure of the vor-
tance into the fluid and so is only able to mobilise fluid very close tex pairs between each pair of discs is unchanged and their
to the discs. In contrast, the swirling flow near the classifier is quite strengths are also very similar. The double pair of vortices between

Fig. 3. Steady state combined charge and slurry motion predicted by an SPH only pseudo-fluid model with viscosity 0.1 Pa s, when the classifier is closed and there is no
discharge from the mill. The fluid is coloured by (a) speed, (b) axial velocity, and (c) turbulent kinetic energy. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
246 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

the classifier and the mill shell are also little changed. The retro- the mill for a viscosity of 0.1 Pa s by showing velocity arrows at
grade (dark blue) flow outside the last disc has been strongly regularly spaced locations throughout the grinding chamber. The
damped as has the flow at the feed end of the mill near the shell. very strong circumferential flow generated by the rotation of the
In contrast, the turbulent kinetic energy (comparing Fig. 3c to impellor is very clear. The faster recirculation of the vortices
Fig. 2c) is strongly increased by the reduction in the viscosity. between the discs is also visible. The flow around the classifier is
This occurs because the viscous forces damp fluctuations in the very strong with the arrows all being red. Fig. 4b shows velocity
velocity field, so as the viscosity decreases, the velocity fluctua- arrows for the flow around and between two of the mid-mill grind-
tions and therefore the turbulent kinetic energy increases in mag- ing discs in a vertical plane with the arrows coloured by the fluid
nitude. This increase is around a factor of two for a one order of speed. This clearly shows the complex circulatory flow that occurs
magnitude reduction in viscosity. The distribution, however, in this mill. At the top, between the discs is a large clockwise vortex
remains very similar for the two viscosities with moderate and which strongly feeds fluid into the hole in the disc on the left. At
very even turbulent kinetic energy between each pair of discs the bottom of the mill between the same pair of discs is a large
decreasing radially to the shell and with a strong concentration anti-clockwise vortex which is fed by fluid coming from the out-
adjacent to the classifier. side of the hole in the disc on the left and which then returns fluid
Next we consider the flow behaviour when flow through the to the same hole near the shaft. Strong flow of fluid to the right out
classifier is allowed. Fig. 4a shows the full velocity field throughout of the hole at the bottom of the right disc is also visible. The flow is

Fig. 4. Fluid flow field in the Isamill represented by velocity arrows for viscosity 0.1 Pa s for the case when the classifier flow and mill discharge are included. The arrows are
coloured and also have lengths proportional to their speed. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this
article.)
 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260 247

very coherent with clear vortices between each pair of discs and velocity for both direction and magnitude. In Fig. 5b, the arrows
strong flow into and out of the disc holes which play a critical role show the velocity components within the vertical slice (since the
in axial fluid circulation within the mill. The axial flow is critical to swirl component is much larger then it can mask the smaller radial
mill performance since this is the flow component that is ulti- and axial components). The flow pattern is very complex with sig-
mately responsible for transporting feed material towards the dis- nificant spatial variation, however there is some very clear struc-
charge end and which is then responsible for transporting ground ture that has important performance contributions. The classifier
product along the mill enabling its removal from the discharge end is located just to the right of the region shown. There is a very
of the mill. The transport paths for the slurry are convoluted and strong retrograde axial flow along the wall of the mill shell in
not at all intuitive. The flow behaviour between the discs is quite the annular region between the outer edges of the discs and the
different when flow through the classifier is permitted. The sym- shell as shown by the long dark blue band at the top and bottom
metric pairs of vortices found for the blocked classifier are replaced of these images. This extends backwards to disc 5 meaning that
by single vortices when classifier flow is permitted with the direc- it extends nearly halfway along the mill. This is significantly stron-
tion of each vortex seemingly random. ger than the retrograde flow found when there was no flow permit-
Fig. 5 shows the fluid flow in the vertical plane through the axis ted by the classifier. There is now strong axial flow towards the
of the mill with fluid coloured by axial speed with white velocity feed end through the discs with large red coloured regions
arrows showing components of the flow for a fluid viscosity of between each pair of discs. In this case, the retrograde flow along
1.0 Pa s. In Fig. 5a the arrows indicate the swirl (azimuthal) the wall is interrupted at disc 5 with fluid flowing radially outward

Fig. 5. Fluid only flow in the Isamill including the classifier flow for viscosity 0.1 Pa s. The fluid is coloured by the axial speed. Velocity arrows (with length dependent on the
flow speed) are shown in white. The arrows show (a) tangential (swirl) velocity, and (b) transverse flow within the vertical plane. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)
248 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

along its left side, then travelling around the outside of the disc and 4.3. Action of the classifier
then radially inwards along its right face. The flow shown is neither
symmetric (circumferentially) nor is it consistent from one disc to The purpose and effect of the classifier is not immediately obvi-
the next. It is characterised as mostly retrograde flow in the ous based on information in the public domain. Earlier, we have
annular region between discs and mill shell and forward flow seen that the rotation of the classifier cage generates very strong
within the region bounded by the discs via flow through the disc swirl near the mill shell at the discharge end of the mill and this
holes with a complex pattern of vortices and recirculations flow affects the nature and symmetry of the flow in the main part
superimposed. of the grinding chamber. When flow was allowed through the clas-
The only published results for a CFD fluid solution in a full bead sifier then retrograde flow was observed along the mill shell out-
mill geometry is by Lane (1999). This was for a 4 l Netzsch stirred side the discs. To better understand the action of the classifier,
mill (122 mm internal diameter and 392 mm length with 9 discs we identify the fluid volume that is initially within the classifier
and five holes per disc). The simulations were performed using and visualise this fluid only. In this model we use the open classi-
CFX with a finite volume method. Lane reported very coarse grid fier geometry that allows flow both into and out of the discharge
solutions of the full mill interior, finer resolution solutions in a region inside the classifier. The surface of this sub-volume of fluid
periodic 72 degree slice around both single disc and the full length is rendered to show the manner in which it flows and deforms. This
of the mill, and axi-symmetric 2D simulations in the radial/axial approach has previously been used to successfully represent the
plane (ignoring the azimuthal variations such as the disc holes). flow in viscous liquid mixers (Robinson and Cleary, 2011, 2012).
The most comparable case to our full mill solutions is with his full The identification of the sub-volume at each time is simple and
length steady state axi-symmetric flow case. This showed the accurate when using the SPH method because of its Lagrangian
existence of two opposing vortices in between each pair of discs. nature and the fact that the particles continuously represent the
This shows agreement on the most critical fluid flow structure in same elements of fluid. As a consequence, there is no numerical
a multi-disc bead mill, albeit that the mills are of quite different diffusion of the fluid volume interface relating to surface tracking
physical scales. as would occur in a grid based method.

Fig. 6. Slurry discharge and back flow along the mill shell produced by the classifier for a viscosity of 0.1 Pa s.
 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260 249

Fig. 6 shows the slurry discharge and back flow along the mill The turbulent kinetic energy (Fig. 7c) has been sharply
shell produced by the classifier for a viscosity of 0.1 Pa s with the increased with the area shown as very high (red) substantially
initial volume shown in Fig. 6a. After 0.2 s (Fig. 6b), the flow has increasing in the region around the classifier and in an annular
become established with the fluid being rapidly accelerated in region extending back past disc 8 and reaching disc 7. It is also very
the circumferential direction leading to a very strong centrifugal high for the regions of flow between the last disc and the classifier
force that forces the fluid radially outwards reaching the mill shell. end plate and the flow through the plate into the classifier cage.
The fluid speed is low for the fluid to the right of the end of the The turbulent kinetic energy between the discs is less even (due
classifier but exceeds 25 m/s for fluid ejected radially from the to the irregular vortex structure seen best in Fig. 5 and Fig. 7b)
classifier. By 0.4 s (Fig. 6c), the flow is fully established with all but is of similar magnitude. It has increased strongly in the region
the fluid that is axially in line with the classifier moving at above between the outside of the discs and the shell over the full length
25 m/s (red). Fluid can now be seen moving in a retrograde direc- of the grinding chamber due to the effect of the retrograde axial
tion in a thin annular layer adjacent to the mill shell. The leading flow along the shell. For all the flow quantities shown, the change
fluid is coloured green1 indicating that it is moving at between 11 arising from the classifier being open has been significant.
and 15 m/s. At 0.6 s (Fig. 6d) the leading surface of the visualised
fluid has just reached disc 7. Otherwise the flow pattern is very sim-
ilar to Fig. 6c. By 0.8 s, (Fig. 6e) the leading fluid has reached halfway 5. One way coupled DEM-SPH model of the M10000 Isamill
between disc 7 and disc 6. Fluid can now be seen flowing from the
retrograde moving outer annular layer into the space between discs 5.1. DEM solution for media only flow
7 and 8. In Fig. 6f, (at 1.0 s) the leading fluid has now just passed disc
6. The fluid that has previously flowed into the space between discs A detailed investigation of the media behaviour in the Isamill
7 and 8 is now being forced through the holes in disc 8 back towards using DEM was reported in Cleary et al. (2015). The mill is rotating
the classifier. So the action of the classifier is clear – it acts as a pump with a tip speed of 22 m/s, has a fill level of 80% by volume and is
drawing slurry axially through the holes of the end classifier disc and using 9 mm grinding media. Fig. 8a shows the distribution of
forcing fluid radially outward by rotational acceleration of the slurry media and its speed around one representative central grinding
inside the classifier cage to generate centrifugal flow through the disc viewed along the axis of the mill. There are two areas of signif-
classifier holes and then forming an annular retrograde flow along icant voidage and one minor one. The largest void is in the upper
the mill shell. Some of this retrograde moving fluid is sucked into half of the mill starting directly above the impellor shaft and
the space between each pair of discs and then pumped axially along extending to the edge of the disc in the direction to which the disc
the mill back into the classifier forming a large axial recirculation rotates. The second large void is a region below and to the right of
pattern. The axial distance that the retrograde flow can finally travel the shaft. The small void region is near the top on the left where
towards the feed end is controlled by the balance of its strength as it the motion of the media is not sufficiently fast to centrifuge the
leaves the classifier region and the amount of the fluid that is drawn particles against the shell and they are able to separate for a brief
off by the recirculatory flow between each pair of grinding discs. For period. The flow is relatively slow in the region radially beyond the
this viscosity the retrograde flow extends roughly halfway back discs in the lower half of the mill. Flow of media from the rising
towards the feed end of the mill. disc holes forms radial jets that cause the upward moving stream
to form two parts, one travelling around the mill shell and the
other flowing over the shaft. The main voids separate these differ-
4.4. Comparison of flow with open and closed classifier ent flow streams. Fig. 8b shows the flow in a vertical slice through
the axis of the mill with media also coloured by speed. In the lower
The impact of the classifier being open can be clearly seen in half of the mill, the charge below the level of the discs and near the
Fig. 7 (which can be directly compared to Fig. 3). There is substan- feed end wall are coloured mid-blue indicating that this material is
tial change in all three flow attributes shown. The most marked relatively slowly moving at between 0.4 and 0.7 m/s. The hydro-
difference is the significant expansion of the high speed region static pressure gradient generated by gravity means that the
between the classifier and the shell (Fig. 7a) which has extended charge presses downwards and is more densely packed and less
significantly back into the grinding chamber in the form of a strong able to be sheared by the motion of the discs. In contrast, in the
retrograde axial flow (dark blue region extending from the middle upper half of the mill, the lower bed pressures allow the charge
of the classifier back to the seventh disk in Fig. 7b). This leads to the to move at 2.5 m/s or above (with the particles all coloured green
substantive elimination of the very slow regions (the dark blue up to red). On each side of each of the discs there is a nearly vertical
outer region between the outside of the discs and the shell in band of fast moving particles (coloured red). These correspond to
Fig. 3a) with this material now predominantly green (both at the jets of media being ejected at high speeds from the holes in the
top and bottom of the mill). The vortex structure between the discs discs. The classifier produces an annulus of fast moving material
has strengthened and become less symmetric (Fig. 7b compared to swirling with the classifier cage.
Fig. 3b) due to strong axial flow towards the discharge end through The flow of media only as predicted by DEM has many features in
the disc holes. This has led to the expansion of the red regions with common with the fluid only solution. The most obvious difference is
matching decreases in the blue regions as the vortex structure is that the media distribution is more spatially variable with macro-
now superimposed with the axial flow to the right within the vol- scopic voids whereas the fluid fills the entire grinding chamber.
ume of the impellor. In particular, the strong axial flow through the The fluid flow radially outside the discs is circumferentially sym-
end plate of the classifier is visible as fluid is pumped by the clas- metric for the fluid model whereas the hydrostatic pressure gradient
sifier from around the end disk into the classifier and then radially has quite a strong effect on the media speed creating importance dif-
outward through the classifier cage towards the shell. This gener- ferences from the DEM model. Aspects such as the radial variation of
ates the strong matching retrograde flow along the shell (Fig. 7b) charge speed and the very strong swirling flow generated by the
which leads to the complex flow paths and vortex structure classifier are common to both. The speeds for the fluid model are
reported earlier for Fig. 5. approximately double those of the media only model indicating that
the fluid has a stronger shear interaction with the discs. This accel-
1
For interpretation of colour in Figs. 6 and 9–11, the reader is referred to the web erates the flow more strongly than is the case for just media by itself
version of this article. which is pushed away from the discs by the centrifugal force limiting
250 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

Fig. 7. Steady state combined charge and slurry motion predicted by an SPH only pseudo-fluid model with viscosity 0.1 Pa s, when the classifier is open and there is flow from
the inside of the classifier radially out into the grinding chamber and discharge from the mill. The fluid is coloured by (a) speed, (b) axial velocity, and (c) turbulent kinetic
energy. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

the traction that the discs can apply. In contrast, fluid (when trying of the media flow. So whilst the two charge phases have similarities
to move away from the discs) generates strong suction pressures in some behaviour there are critical differences as well. This means
that balance the centrifugal force and prevent the fluid from moving that models representing the solid media as a pseudo-fluid will be
away radially. Such differences should be expected as the fluid missing key elements of the solid charge behaviour, whilst DEM only
model has only a very simple rheology and cannot capture key models that assume that the slurry moves only along with the media
aspects of the media behaviour and its strong spatial variation. do not well represent the slurry component. These differences moti-
However, they also indicate that the fluid wants to behave differ- vate the use of the 1-way coupled DEM + SPH model with DEM being
ently to the media, particularly in the strong axial flow through used to represent the particulate media behaviour and the SPH
the disc holes towards the discharge end and the retrograde flow model being used to represent the slurry flow through the media.
along the shell generated by the classifier, which were not features The results of such a model are presented in the next section.
 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260 251

Fig. 8. Media distribution and velocity distribution in (a) an axial slice around one of the impellor discs, and (b) vertical slice of the Isamill for 9 mm media with a fill level of
80%. Particles coloured by speed with dark blue shows stationary particles, green mid-speed particles around 2.5 m/s and red shows high speed particles of 5 m/s or higher.
(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

5.2. Averaged DEM media distributions classifier. It is therefore essential that these flow structures be well
represented and so the averaging needs to be performed in the
The DEM flow is averaged into a cylindrical grid co-rotating frame of the impellor. Jayasundara et al. (2009), in developing their
with the impellor. This allows the solid charge component to be coupled DEM-CFD model of flow around a single disc, also used a
represented as a dynamic porous media which has a solid fraction CFD computational domain that co-rotated with the disc in order
distribution and velocity distribution that are consistent with the to resolve the flow through its holes.
steady state particle flow. The porosity and velocity data are Fig. 9 shows the solid fraction and axial speed distribution of
defined in a co-rotating frame with the impellor so that all the flow the dynamic porous media representing the solid charge from
structure relating to the critical holes in the grinding discs and the the averaging of the DEM media predictions. These are shown on
classifier are captured in the porous media model. If the data col- vertical plane sections along the axis of the mill and between each
lection is performed in a stationary frame then the very strong pair of discs and between the last disc and the classifier. The media
swirling flow resulting from the very high impellor rotation rate volume fraction is fairly uniform within the grinding chamber
means that all the flow structures are averaged over the azimuthal (coloured mainly yellow through red corresponding to 0.5 to
direction. The resulting data then contains only radial and axial 0.6). This level of packing is only modestly below the packing limit
variations with all the azimuthal flow structure eliminated by for mono-disperse spheres of 0.64. The packing is quite uniform
the averaging. The particle flows around and through the disc holes radially in regions between discs 2–7. The media is more densely
were shown in Cleary et al. (2015) to be complex and critical to packed around the shell between discs 1 and 2 and is less dense
mill operation. The earlier fluid solutions also show that the disc between the end of the classifier plate and disc 8 and also between
holes provide a critical path for axial transport of slurry that the shell and disc 1. The large region coloured dark blue is the
enables a large scale axial recirculation to be generated by the empty area within the classifier cage. There is also a thin layer of
252 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

Fig. 9. Media distribution from the DEM component of the simulation averaged into a cylindrical grid co-rotating with the impellor, (a) solid fraction (VF), and (b) axial speed
of media.

low media density around the periphery of the classifier which centrelines. These are the vertical axial sections that were shown
reflects the combination of the strong dilative pressure arising in Fig. 9b which showed little axial flow between most of the discs.
from the media collisions with the axial classifier struts and the This highlights the challenges in visualising and interpreting com-
high centrifugal force. plex flow fields around complex geometries. These strong recircu-
The average media speed is dominated by the circumferential lations generated by the passing of the holes in the discs vary
swirl (azimuthal flow) which is very uniform and so not useful to strongly with the azimuthal position. This can be seen best in
show. Fig. 9b shows the axial velocity of the media on the same Fig. 10b which shows the axial speed rendered on an isosurface
sections as for Fig. 9a. The axial velocity is very uniform and near of constant solid fraction surrounding one of the discs. There is
zero in most of the cross-sections of the grinding chamber shown. strong flow in the positive direction (red) from the disc holes and
There is a high axial velocity between discs 7 and 8 corresponding strong axial flow back towards the disc in an adjacent azimuthal
to flow between and through the axially adjacent holes in these location. The orientation of these structures, compared to the ver-
discs. There is also strong media flow between disc 8 and the tical section in Fig. 10a, determines their relative strength in the
end wall of the classifier (with small red and dark blue regions upper and lower half of the mill. The axial flow looks stronger in
located axially between the disc holes). There is some very weak the lower half only because this part of the section plane passes
axial flow between disc holes between discs 1–3 (with yellow through the structures where they are strong whereas in the upper
and faint light blue regions). Between discs 3–7 there is very little part the intersection is through areas where the structures are less
sign of any axial motion in the sections shown at the mid-points strong.
between discs. Comparing the average axial flow of media from the DEM model
Fig. 10 shows the average media flow field in two additional (Fig. 10) and the charge from the SPH fluid only model (Fig. 2), we
ways. Fig. 10a shows the axial velocity on just the vertical plane observe considerable similarities but also some important differ-
through the axis of the mill. This reveals significant axial transport ences. Between the discs, both methods predict oppositely ori-
between each pair of discs. These consist of symmetric pairs of ented pairs of vortices (giving the red/dark blue pattern) but the
regions of oppositely directed flow (dark blue and red) which cor- shape and structure of these depends on the method. The flow
respond to pairs of oppositely directed vortices between each disc. radially outside the classifier has two toroidal recirculation zones
The centre-plane of the region between each pair of discs acts as a (as indicated by the two pairs of oppositely directed axial flow)
flow separatrix with the oppositely directed flows from the discs and its prediction is common to both methods but the details of
cancelling so that there is no axial movement on most of these the shape of these zones differ. In both cases there is also a regular
 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260 253

Fig. 10. Axial velocity of media from the DEM component of the simulation averaged on a co-rotating cylindrical grid shown on: (a) a vertical slice through the axis of the
mill, and (b) on an isosurface of constant solid fraction around one of the impellor discs.

pattern of weak alternating axial flow between the outer edges of viscosity of 0.1 Pa s with the fluid coloured by speed and axial
the discs and the shell. From this we deduce that the media and the velocity on both the axial centre-plane and on the intersecting sec-
slurry have broadly similar flow behaviour even though they have tions used previously. The predicted slurry flow has qualitative
very different rheological properties, but there are important dif- similarities with the fluid only flow field (Fig. 2) but also quite
ferences in the details of their response to the rotation of the mill important differences. The flow into and around the classifier is
impellor. very similar with the same structure for the flow from the holes
of disc 8 through the space to the start of the classifier, through
5.3. Slurry motion in the Isamill the holes in this plate, within the classifier and the outward
radial discharge flow from the classifier cage. Since this region is
The 1-way coupled SPH model using the porous media data devoid of media in the DEM + SPH model there is only fluid present
shown in Figs. 9 and 10 provides predictions of the slurry motion and so this part of the solution matches closely to the fluid only
in the Isamill. This is different to the earlier SPH model because model.
now the SPH fluid is representing only the fine slurry with the In the region around the grinding discs, the flow field is struc-
media now represented by the coupled DEM solution. To allow turally similar but the slurry speeds are all lower by around a half.
interpretation of these SPH results, the fluid particle motion is also This occurs because the viscous interaction of the fluid with the
averaged into a cylindrical grid co-rotating with the impellor (in discs is stronger than the traction with the particulate media
the same way as was performed for the DEM data). This data can (which are pushed away from the discs) and so the fluid only
then be visualised to show the nature of the slurry flow. charge accelerates more whereas the particulate charge has signif-
Fig. 11 shows the steady state fluid flow in the M10000 Isamill icant slip relative to the disc. So the fluid only model overstates the
with open classifier using the 1-way coupled model for a slurry charge speeds because the traction with the disc from a viscous
254 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

Fig. 11. Time averaged steady SPH velocity field and fluid fraction within the mill with open classifier using the 1-way coupled DEM and SPH model for viscosity 0.1 Pa s, (a)
slurry speed in a vertical cross-section, (b) axial speed in the same vertical section, and (c) fluid speed in axial slices and the vertical slice viewed obliquely from the side.

fluid is much stronger than the traction experienced between the enable representation as a complex rheological fluid that requires
discs and the particulate media. The Newtonian continuum model impractical spatial resolution to capture the critical flow behaviour
is unable to capture the high level of slip experienced by the media. near the disc surfaces.
In principle, this behaviour may be recoverable using a shear thin- This effect also leads to a significant difference in the extent of
ning rheology model however the extremely thin layers of pseudo- the high speed flow around the outside of the classifier. In the fluid
fluid over which this behaviour exhibits would require exceedingly only model, significant motion is generated between the classifier
fine spatial resolution in order to sufficiently accurately capture and the shell which extends back past disc 8. In contrast, when
the very high controlling velocity gradients – which is likely to using the coupled DEM + SPH model, the high speed fluid travels
be computationally impractical. This highlights the advantage of radially only halfway to the shell before being slowed by the dissi-
using a DEM model component to represent the particulates and pation arising from being forced through the densely packed media
to well capture their behaviour rather than averaging them to layer that is found along the shell.
 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260 255

Fig. 12. Axial flow of slurry in the Isamill with an open classifier (using the 1-way coupled DEM and SPH model) as a function of viscosity, (a) 0.1 Pa s, (b) 1 Pa s, and (c)
10 Pa s. The slurry is coloured by the axial speed as shown in the colour bar. (For interpretation of the references to colour in this figure legend, the reader is referred to the
web version of this article.)

The detailed structures of the vortices generated in the slurry lower viscosity. The structure of the flow above the shaft (which
are best shown when colouring by axial speed (Figs. 11b and 2b). is for a slice of the solution away from the disc holes) is very sim-
For the coupled solution, the scale of detail in the structures is ilar with the dark blue (retrograde) flow on the right sides of the
higher and is the result of the slurry in coupled solution having a discs and flow toward the discharge end (red coloured fluid) beside
256 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

the left faces of the discs. In contrast, the flow below the shaft the fluid only case is predominantly in the radial–axial direction
shows that it interacts with the disc holes more and has a broadly but in the coupled case it is predominantly in the circumferential
different structure. In particular, the distribution of dark blue and direction with quite different flow paths between the discs and is
red are generally reversed but also indicates a different pattern. more three dimensional in nature. This is a consequence of the
For the fluid only solution, near the shaft the fluid flows axially more complex three dimensional flow structure of the media and
towards the closest disc, then flows radially outward near the faces their interactions with the disc holes as shown in Fig. 10b. For
of the discs and then axially back along the mill shell towards the the coupled flow, the slurry behaviour is controlled by a combina-
centre of the space between the discs before returning radially tion of the forces experienced by the fluid directly but is even more
inward to the starting position beside the shaft. This creates a pair strongly influenced by the effect of the particulate media which
of oppositely circulating vortices. For the coupled solution, the imposes aspects of its structure onto the slurry motion.
fluid flows axially into the disc holes (in this slice of the solution) The axial flow in the classifier region is quite similar in
as seen by the small red/blue pairs of regions beside the disc hole. Figs. 11b and 2b. There is again a pair of toroidal vortices located
Oppositely moving fluid can then be seen in a U shaped region on either side of the outward radial flow from the centre of the
around each of these. So within the region between each disc pair classifier, but there have been some changes in the shapes of the
the fluid flows axially away from each of the discs (in the solution structures and the relative magnitudes of the flow in each area.
slice shown) with some of the fluid splitting from this stream and Specifically, the regions of very high speed (dark blue and red)
diverting into the disc holes. This fluid travels into this region by are enlarged and the moderate speed (green) areas in between
flowing in the circumferential direction. So the recirculation in are reduced. The dark blue retrograde flow on the left of the

Fig. 13. Axial flow of slurry in the Isamill with an open classifier (using the 1-way coupled DEM and SPH model) as a function of viscosity, (a) 0.1 Pa1 s, (b) 1 Pa1 s, and (c)
10 Pa1 s. The slurry is coloured by the axial speed as shown in the colour bar. (For interpretation of the references to colour in this figure legend, the reader is referred to the
web version of this article.)
 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260 257

classifier is particularly stronger. These differences occur because this region. This type of effect cannot be captured by the 1-way
the classifier was closed in the case shown in Fig. 2 whereas it is coupled model used here. This question will be explored in future
open in the case in Fig. 11. When open this allows strong antegrade work using a fully 2-way coupled model.
flow into the classifier (indicated by the red coloured regions pass-
ing through the holes in the face plate of the classifier) and then 5.4. Effect of viscosity on slurry flow in the Isamill
generates stronger radial outward flow from the classifier towards
the shell, strengthening the toroidal vortices and producing strong Figs. 12–14 show the steady slurry flow in the M10000 Isamill
retrograde flow along the shell back to disc 8. So the high speed with open classifier as a function of viscosity with the flow
rotation of the classifier cage generates the basic double toroidal coloured by speed, axial velocity and vorticity, respectively. The
flow driven by the high centrifugal force. Any flow through the three viscosities considered are 0.1 Pa s, 1 Pa s, and 10 Pa s which
classifier simply enhances the strength of this basic classifier flow. cover the plausible maximum range of viscosities that could be
These differences are not a consequence of the different model encountered within such a mill. The basic flow pattern is similar
used but reflect the operation of the classifier when closed and in all three cases indicating that the basic flow structure does not
open. In both cases there are pairs of vortices predicted between change with viscosity but the strength of the slurry flow decreases
each pair of discs, but the details of the structures are different. strongly with increasing viscosity. The flow between the discs and
One might anticipate that in the region directly adjacent to the into the disc holes and the flow into and around the classifier are
classifier that the back flow shown in Figs. 6 and 11 could be suf- all still present but all weaken with the speed of the slurry
ficiently strong so as to influence the motion of media particles in (Fig. 12) consistently decreasing with viscosity. For 1 Pa s much

Fig. 14. Slurry vorticity distribution in the Isamill with an open classifier (using the 1-way coupled DEM and SPH model) as a function of viscosity, (a) 0.1 Pa s, (b) 1 Pa s, and
(c) 10 Pa s. The slurry is coloured by the vorticity as shown in the colour bar. (For interpretation of the references to colour in this figure legend, the reader is referred to the
web version of this article.)
258 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

of the vortex structure between the discs is strongly reduced with The flow pattern within the mill using a fluid only model has
peak speeds being halved with respect to the 0.1 Pa s case. The flow many similarities with DEM predictions of flow of media within
directly adjacent and within the holes remains strong because of the grinding chamber but with velocities around double the mag-
the direct impact of the disc motion. The regions of high shear gen- nitude. This occurs because the viscous fluid has a stronger interac-
erated by the motion of the discs (as shown by the high vorticity in tion with the discs than does the media by itself. This increased
Fig. 14) shrink with increasing viscosity showing that the effect of traction results in stronger acceleration, more energy transfer from
the discs on the slurry becomes increasingly more localised to the the discs and higher predictions of flow speeds. This suggests that a
immediate proximity of the discs. strongly shear thinning rheology is needed for a fluid model to cap-
The flow around the classifier is also strongly damped with the ture the media behaviour but that this would required pro-
amount of flow into the classifier through its end plate being shar- hibitively fine spatial resolution for the numerical discretisation.
ply reduced. The retrograde flow from the classifier back down The flow field in the absence of flow through the classifier consists
along the shell (as shown in Fig. 13) is also strongly damped. For of pairs of recirculation vortices between each pair of disc and a
a viscosity of 10 Pa s, almost all axial flow is suppressed and the return flow towards the shaft in between. This flow pattern is very
flow into and out of the classifier is very weak. At this viscosity, regular in the axial direction.
the slurry has almost no capacity to move independently of the The model with open classifier has allowed us to understand
media and simply flows around the mill circumferentially with the impact that it has on the flow within the mill. The effect is sig-
the grinding media. This means that the classifier pumping perfor- nificant and strongly influences the flow over at least the discharge
mance and its ability to circulate slurry within the mill and the half of the mill and controls slurry (feed and product) transport and
axial flow of slurry through the disc holes towards the classifier discharge. The classifier acts as a pump drawing slurry axially
are strongly viscosity dependent. It is arguable that from a slurry through the holes in its end disc and forcing it radially outward
transport perspective (which includes the ability to deliver fresh by rotational acceleration of the slurry inside the classifier cage
feed slurry to the grinding regions between discs and the ability to generate centrifugal flow through the long rectangular gaps
to remove ground product from the mill) that a viscosity of between the axial bars of the cage. The resulting strong radial out-
10 Pa s is the upper limit that is possible for an M10000 Isamill flow generates a large toroidal vortex on either side of classifier.
using 15 mm media. These produce a strong annular retrograde flow along the mill shell
The critical viscosity is dependent on the media size being used that penetrates a significant distance back into the grinding cham-
in the mill. The media size controls the dimensions of the pore ber. When the classifier is open and pumping then there is strong
spacing through which the slurry needs to flow. As the media axial flow of slurry through the holes in the discs and very strong
becomes smaller, viscous forces are more effective in generating flow from the region around discs 7 and 8 through the classifier
resistance to flow on these smaller scales and so a smaller critical end wall and into centre of the classifier cage. The strong centrifu-
viscosity is needed to halt flow. Eq. (5) gives the porous media gal force enhances the outward flow from the classifier with fluid
force on the slurry. The velocity and porosities for the media flowing from inside the cage though the classifier and out to the
change only very weakly with decreasing media size for 15 mm shell where it is forced towards the feed end as a strong retrograde
and smaller (Cleary et al., 2015) so the only terms that affect the flow between the outer edges of the discs and the shell. This can
slurry force significantly are the viscosity with a linear dependence extend 3–4 discs back into the grinding chamber. When the classi-
and the permeability (given by Eq. (6)). Given that the porosity var- fier is open the regular pattern of vortices between the discs is
ies only weakly with media size then the principle dependence of modified with some reversals of the vortex motion as the basic disc
the permeability is on S (the Surface to Volume ratio of the charge) vortices interact with the axial cross flows.
which varies inversely with media size. This means that the critical The 1-way coupled DEM + SPH model uses a DEM solution to
viscosity varies with the square of the media size. So a critical vis- represent the media. The averaging of the media flow is performed
cosity of 10 Pa s for 15 mm media is equivalent to 0.71 Pa s for in a frame co-rotating with the impellor in order to resolve the
4 mm media. detailed flow through the holes in the discs. The SPH component
Whilst the concept of a critical viscosity appears to be new for represents the slurry phase and predicts the differential movement
the Isamill, similar issues have previously been reported for ball of the slurry relative to the media. This coupled model, which bet-
mills by Fuerstenau et al. (1985). For wet grinding with high solids ter represents the particulate media behaviour than a fluid only
concentrations in a ball mill, it was observed that when the pulp model, predicts qualitatively similar flow behaviour but with
viscosity exceeded a critical viscosity that the media began to cen- important differences. These include significantly lower flow
trifuge, leading to obvious reductions in grinding (with less speeds which are comparable to the media flow speeds predicted
cataracting and cascading flow) and declines in power consumption by the DEM component, reduced strength in the outward classifier
energy. The inclusion of a polymeric additive that can reduce pulp flow because of the strong damping produced by the packed media
viscosity can therefore lead to better grinding performance and into which the slurry is flowing and more complex three dimen-
increases in energy consumption. Here we have demonstrated that sional flow in the vicinity of the disc holes. The media behaviour,
similar behaviours also occur for stirred mills such as the Isamill. with flow into and out of the disc holes is more complex than pre-
dicted by the fluid only model and drives the slurry motion. In
6. Conclusions some areas the slurry is dragged along with the media while in
others it is forced to flow in the opposite direction due to require-
Slurry and charge motion in a full industrial scale M10000 ments of conservation of volume and strongly changing media
Isamill have been explored computationally using porosity near the discs. This means that near the disc holes the
slurry circulates in the opposite direction to the fluid only model
 a fluid only approximation solved by the SPH method, and with a significant circumferential component to the motion
 a 1-way coupled DEM + SPH model that uses DEM to predict the whereas between disc holes it flows in a similar manner to the
basic behaviour of the particulate media which is then repre- fluid with the vortex pairs between discs. The fluid only model is
sented as a dynamic porous media in an SPH model to predict unable to capture some of the complex features revealed by the
the resulting slurry motion through the charge. 1-way coupled model that relate to significant variations in media
 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260 259

porosity and the significant amount of slip between the media and interactions with experimental assessment. Int. J. Numer. Methods Eng. 81 (2),
229–245.
the driving discs.
Fernandez, J.W., Cleary, P.W., Sinnott, M.D., Morrison, R.D., 2011. Using SPH one way
Increasing slurry viscosity (as predicted by the coupled model) coupled to DEM to model wet industrial banana screens. Miner. Eng. 24, 741–
leads to a reduction in the strength of the recirculation of slurry 753.
between discs and the disc vortices and weakens the outward Ferrari, A., 2010. SPH simulations of free surface flow over a sharp crested weir. Adv.
Water Resour. 33, 270–276.
radial flow from the classifier and the retrograde axial flow along Fuerstenau, D.W., Venkataraman, K.S., Velamakanni, B.V., 1985. Effect of chemical
the shell that it generates. This suggests that there is a critical max- additives on the dynamics of grinding media in wet ball mill grinding. Int. J.
imum viscosity for which the mill can produce acceptable slurry Miner. Process. 15 (4), 251–267.
Gers, R., Clement, E., Legendre, D., Anne-Archard, D., Frances, C., 2010. Numerical
transport. From the current model results this would appear to modelling of grinding in a stirred media mill: hydrodynamics and collision
be at most 10 Pa s for 15 mm media, which is equivalent to a characteristics. Chem. Eng. Sci. 65, 2052–2064.
0.71 Pa s limit for 4 mm media. Gingold, R.A., Monaghan, J.J., 1977. Smoothed particle hydrodynamics: theory and
application to non-spherical stars. Mon. Not. R. Astron. Soc. 181, 375–389.
Gnanavelu, A., Kapur, N., Neville, A., Flores, J.F., Ghorbani, N., 2011. A numerical
References investigation of a geometry independent integrated method to
predict erosion rates in slurry erosion. Wear 271, 712–719.
Gómez-Gesteira, M., Dalrymple, R.A., 2004. Using a three-dimensional smoothed
Blecher, L., Kwade, A., Schwedes, J., 1996. Motion and stress intensity of grinding
particle hydrodynamics method for wave impact on a tall structure. J.
beads in a stirred mill. Part 1: energy density distribution and motion of single
Waterway, Port, Coastal, Ocean Eng. 130 (2), 63–69.
grinding beads. Powder Technol. 86, 59–68.
Herbst, J.A., Nordell, L., 2001. Optimization of the design of sag mill internals using
Boek, E.S., Venturoli, M., 2010. Lattice-Boltzmann studies of fluid flow in porous
high fidelity simulation. In: Barratt, Allan, Mular (Eds.), Proc. of the SAG
media with realistic rock geometries. Comput. Math. Appl. 59 (7), 2305–2314.
Conference, vol. IV, pp. 150–164.
Brown, G.J., 2002. Erosion prediction in slurry pipeline tee-junctions. Appl. Math.
Hilton, J.E., Cleary, P.W., 2011. The influence of particle shape on flow modes in
Model. 26 (2), 155–170.
pneumatic conveying. Chem. Eng. Sci. 66, 231–240.
Burford, B.D., Clark, L.W., 2007. IsaMill™ technology used in efficient grinding
Hilton, J.E., Mason, L., Cleary, P.W., 2010. Dynamics of gas–solid fluidised beds with
circuits. In: VIII International Conference on Nonferrous Ore Processing, Poland.
non-spherical particle geometry. Chem. Eng. Sci. 65, 1584–1596.
Carvalho, R.M., Tavares, L.M., 2011. Leaping forward in SAG and AG Mill simulation
Inoue, T., Okaya, K., 1996. Grinding mechanism of centrifugal mills – a simulation
using a mechanistic model framework. In: SAG2011 Conference, Vancouver.
study based on the discrete element method. Int. J. Miner. Process. 44–45, 425–
Cleary, P.W., 1998a. Predicting charge motion, power draw, segregation, wear and
435.
particle breakage in ball mills using discrete element methods. Miner. Eng. 11,
Jayasundara, C.T., Yang, R.Y., Yu, A.B., 2006. Discrete particle simulation of particle
1061–1080.
flow in the Isamill process. Ind. Eng. Chem. Res. 45, 6349–6359.
Cleary, P.W., 1998b. Modelling confined multi-material heat and mass flows using
Jayasundara, C.T., Yang, R.Y., Yu, A.B., Curry, D., 2008. Discrete particle simulation of
SPH. Appl. Math. Modell. 22, 981–993.
particle flow in Isamill – effect of grinding media properties. Chem. Eng. J. 135,
Cleary, P.W., 1998c. The filling of dragline buckets. Math. Eng. Ind. 7, 1–24.
103–112.
Cleary, P.W., 2001a. Recent advances in DEM modelling of tumbling mills. Miner.
Jayasundara, C.T., Yang, R.Y., Guo, B.Y., Yu, A.B., Rubenstein, J., 2009. Effect of slurry
Eng. 14, 1295–1319.
properties on particle motion in Isamills. Miner. Eng. 22 (11), 886–892.
Cleary, P.W., 2001b. Charge behaviour and power consumption in ball mills:
Jayasundara, C.T., Yang, R.Y., Yu, A.B., Rubenstein, J., 2010. Effects of disc rotation
sensitivity to mill operating conditions, liner geometry and charge composition.
speed and media loading on particle flow and grinding performance in a
Int. J. Miner. Process. 63, 79–114.
horizontal stirred mill. Int. J. Miner. Process. 96 (1), 27–35.
Cleary, P.W., 2001c. Modelling comminution devices using DEM. Int. J. Numer. Anal.
Jayasundara, C.T., Yang, R.Y., Yu, A.B., Curry, D., 2011a. Prediction of the disc wear in
Meth. Geomech. 25, 83–105.
a model IsaMill and its effect on the flow of grinding media. Miner. Eng. 24 (14),
Cleary, P.W., 2004. Large scale industrial DEM modelling. Eng. Comput. 21, 169–
1586–1594.
204.
Jayasundara, C.T., Yang, R.Y., Yu, A.B., 2011b. Discrete particle simulation of particle
Cleary, P.W., 2009. Industrial particle flow modelling using DEM. Eng. Comput. 26
flow in a stirred mill: effect of mill properties and geometry. Ind. Eng. Chem.
(6), 698–743.
Res. 51 (2), 1050–1061.
Cleary, P.W., Ha, J., 2002. Flow modelling in casting processes. Appl. Math. Modell.
Jayasundara, C.T., Yang, R.Y., Guo, B.Y., Yu, A.B., Govender, I., Mainza, A., Van der
26, 171–190.
Westhuizen, A., Rubenstein, J., 2011c. CFD-DEM modelling of particle flow in
Cleary, P.W., Monaghan, J.J., 1990. Towards a realistic three body problem.
IsaMills – comparison between simulations and PEPT measurements. Miner.
Astrophys. J. 349, 150–162.
Eng. 24 (3), 181–187.
Cleary, P.W., Morrison, R.D., 2008. Editorial, Miner. Eng. 21, 743.
Jayasundara, C.T., Yang, R.Y., Yu, A.B., 2012. Effect of the size of media on grinding
Cleary, P.W, Morrison, R.D., 2011. Predicting patterns of slurry flow within and
performance in stirred mills. Miner. Eng. 33, 66–71.
discharge from a 3D pilot SAG mill. In: Proc. SAG 2011, Paper #105, 25–28th
Kafui, D.K., Thornton, C., Adams, M.J., 2002. Discrete particle-continuum fluid
September 2011, Vancouver, Canada.
modelling of gas–solid fluidised beds. Chem. Eng. Sci. 57, 2395–2410.
Cleary, P.W., Morrison, R.D., 2012. Prediction of 3D slurry flow within the grinding
Kalala, J.T., Breetzke, M., Moys, M.H., 2008. Study of the influence of liner wear on
chamber and discharge from a pilot scale SAG mill. Miner. Eng. 39, 184–195.
the load behaviour of an industrial dry tumbling mill using the Discrete
Cleary, P.W., Ha, J., Ahuja, V., 2000. High pressure die casting simulation using
Element Method (DEM). Int. J. Miner. Process. 86 (1), 33–39.
Smoothed Particle Hydrodynamics. J. Cast Metals Res. 12, 335–355.
Kawaguchi, T., Sakamoto, M., Tanaka, T., Tsuji, Y., 2000. Quasi-three dimensional
Cleary, P.W., Sinnott, M.D., Morrison, R.D., 2006. Prediction of slurry transport in
numerical simulation of spouted beds in cylinders. Powder Technol. 109, 3–12.
SAG mills using SPH fluid flow in a dynamic DEM based porous media. Miner.
Krištof, P., Beneš, B., Křivánek, J., Št’ava, O., 2009. Hydraulic erosion using smoothed
Eng. 19, 1517–1527.
particle hydrodynamics. In: Computer Graphics Forum, vol. 28, no. 2. Blackwell
Cleary, P.W., Prakash, M., Ha, J., Stokes, N., Scott, C., 2007. Smooth particle
Publishing Ltd.
hydrodynamics: status and future potential. Prog. Comput. Fluid Dyn. 7, 70–90.
Lane, G.L., 1999. CFD modelling of a stirred bead mill for fine grinding. In: Second
Cleary, P.W., Sinnott, M.D., Morrison, R.D., 2008. DEM prediction of particle flows in
International Conference on CFD in the Minerals and Process Industries, CSIRO,
grinding processes. Int. J. Numer. Methods Fluids 58, 319–353.
Melbourne, Australia, 6–8 December.
Cleary, P.W., Prakash, M., Rothauge, K., 2010. Combining photogrammetric digital
Lane, G.L., Schwarz, M.P., Evans, G.M., 2002. Predicting gas–liquid flow in a
terrain and surface textures with large scale particle based computational
mechanically stirred tank. Appl. Math. Model. 26, 223–235.
models to predict three dimensional dam collapse and landslide events. Int. J.
Lane, G.L., Schwarz, M.P., Evans, G.M., 2005. Numerical modelling of gas–liquid flow
Image Data Fusion 1, 337–357.
in stirred tanks. Chem. Eng. Sci. 60, 2203–2214.
Cleary, P.W., Sinnott, M.D., Pereira, G.G., 2015. Computational prediction of
Lee, H., Cho, H., Kwon, J., 2010. Using the discrete element method to analyze the
performance for a full scale Isamill: Part 1 – media motion and energy
breakage rate in a centrifugal/vibration mill. Powder Technol. 198, 364–372.
utilisation in a dry mill. Miner. Eng. (in press). http://dx.doi.org/10.1016/j.
Lester, D.R., Graham, L.A., Wu, J., 2010. High precision suspension erosion modeling.
mineng.2015.04.005.
Wear 269, 449–457.
Colagrossi, A., Landrini, M., 2003. Numerical simulation of interfacial flows by
Lichter, J.K., Suazo, M., Noriega, R., Murariu, V., 2011. The application of
smoothed particle hydrodynamics. J. Comput. Phys. 191 (2), 448–475.
multiphysics models for the design of mill discharge systems. In: Major, K.,
Cundall, P.A., Strack, O.D., 1979. A discrete numerical model for granular assemblies.
Flintoff, B., Klein, B., McLeod, K. (Eds.), Int. Autogenous Grinding,
Geotechnique 29 (1), 47–65.
Semiautogenous Grinding and High Pressure Grinding Roll Technology 2011
de Kretser, R.G., Powell, M., Scales, P.J., Lim, J., 2009. The Water Efficient Plant of the
(SAG 2011), Vancouver, BC, Canada.
Future – Towards a Holistic Process Chain Approach Water in Mining
Liu, M.B., Liu, G.R., Lam, K.Y., Zong, Z., 2003. Smoothed particle hydrodynamics for
Conference Perth, WA, 15–17 September.
numerical simulation of underwater explosion. Comput. Mech. 30 (2), 106–118.
Djordjevic, N., 2003. Discrete element modelling of the influence of lifters on power
Lucy, L.B., 1977. A numerical approach to the testing of the fission hypothesis.
draw of tumbling mills. Miner. Eng. 16, 331–336.
Astron. J. 82, 1013–1024.
Djordjevic, N., 2005. Influence of charge size distribution on net-power draw of
Manwart, C., Aaltosalmi, U., Koponen, A., Hilfer, R., Timonen, J., 2002. Lattice-
tumbling mill based on DEM modelling. Miner. Eng. 18, 375–378.
Boltzmann and finite-difference simulations for the permeability for three-
Feng, Y.T., Han, K., Owen, D.R.J., 2010. Combined three-dimensional lattice
dimensional porous media. Phys. Rev. E 66 (1), 016702.
Boltzmann method and discrete element method for modelling fluid–particle
260 P.W. Cleary, M.D. Sinnott / Minerals Engineering 79 (2015) 239–260

Mishra, B.K., Rajamani, R.K., 1992. The discrete element method for the simulation Robinson, M., Cleary, P.W., Monaghan, J.J., 2008. Analysis of mixing in a twin-cam
of ball mills. Appl. Math. Modell. 16, 598–604. mixer using Smoothed Particle Hydrodynamics. AIChE J. 54, 1987–1998.
Mishra, B.K., Rajamani, R.K., 1994. Simulation of charge motion in ball mills. Part 1: Silva, R., Garcia, F.A., Faia, P.M., Rasteiro, M.G., 2015. Settling suspensions flow
exp. verifications. Int. J. Miner. Process. 40, 171–186. modelling: a review. KONA Powder Particle J. 32, 41–56.
Monaghan, J.J., 1994. Simulating free surface flows with SPH. J. Comput. Phys. 110, Sinnott, M.D., Cleary, P.W., Morrison, R.D., 2011. Slurry flow in a tower mill. Miner.
399–406. Eng. 24, 152–159.
Monaghan, J.J., 2005. Smoothed particle hydrodynamics. Rep. Prog. Phys. 68 (8), Solnordal, C.B., Wong, C.Y., Zamberi, A., Jadid, M., Johar, Z., 2013. Determination of
1703. erosion rate characteristic for particles with size distributions in the low Stokes
Morrison, R., Cleary, P.W., 2004. Using DEM to model ore breakage within a pilot number range. J. Wear 305, 205–215.
scale SAG mill. Miner. Eng. 17, 1117–1124. Tartakovsky, A.M., Meakin, P., 2005. Simulation of unsaturated flow in complex
Morrison, R.D., Cleary, P.W., 2008. Towards a virtual comminution machine. Miner. fractures using smoothed particle hydrodynamics. Vadose Zone J. 4 (3), 848–855.
Eng. 21, 770–781. Tartakovsky, A.M., Meakin, P., 2006. Pore scale modeling of immiscible and miscible
Napier-Munn, T.J., Morrison, R.D., 2003. The potential for the dry processing of ores. fluid flows using smoothed particle hydrodynamics. Adv. Water Resour. 29,
In: Proceedings Water in Mining 2003, The Role of Water in a Sustainable 1464–1478.
Mining Industry, Brisbane, Australia, 3–6 October, 2003, AusIMM, pp. 247–250. Thornton, C., Cummins, S., Cleary, P.W., 2013. An investigation of the comparative
Phillips, T.N., Roberts, G.W., 2011. Lattice Boltzmann models for non-Newtonian behaviour of alternative contact force models during inelastic collisions.
flows. IMA J. Appl. Math. 76 (5), 790–816. Powder Technol. 233, 30–46.
Potapov, A.V., Hunt, M.L., Campbell, C.S., Technology, Powder., 2001. Liquid–solid Tsuji, Y., Kawaguchi, T., Tanaka, T., 1993. Discrete particle simulation of two
flows using smoothed particle hydrodynamics and the discrete element dimensional fluidized bed. Powder Technol. 77, 79–87.
method. Powder Technol. 116, 204–213. van der Westhuizen, A.P., Govender, I., Mainza, A.N., Rubenstein, J., 2011. Tracking
Powell, M.S., Weerasekara, N.S., Cole, S., LaRoche, R.D., Favier, J., 2011. DEM the motion of media particles inside an IsaMill™ using PEPT. Miner. Eng. 24,
modelling of liner evolution and its influence on grinding rate in ball mills. 195–204.
Miner. Eng. 24, 341–351. Weerasekara, N.S., Powell, M.S., Cleary, P.W., Tavares, L.M., Evertsson, M., Morrison,
Prakash, M., Cleary, P.W., Noui-Mehidi, M.N., Blackburn, H., Brooks, G., 2007. R.D., Quist, J., Carvalho, R.M., 2013. The contribution of DEM to the science of
Simulation of suspension of solids in a liquid in a mixing tank using SPH and comminution. Powder Technol. 248, 3–24.
comparison with physical modeling experiments. Prog. Comput. Fluid Dyn. 7, Xu, B.H., Yu, A.B., 1997. Numerical simulation of the gas–solid flow in a fluidized
91–100. bed by combining discrete particle method with computational fluid dynamics.
Rajamani, R.K., Mishra, B.K., 1996. Dynamics of ball and rock charge in SAG mills. In: Chem. Eng. Sci. 52, 2785–2809.
Proc. SAG 1996, University of British Columbia. Yamada, Y., Sakai, M., 2013. Lagrangian–Lagrangian simulations of solid–liquid
Ramstad, T., Idowu, N., Nardi, C., Øren, P.E., 2012. Relative permeability calculations flows in a bead mill. Powder Technol. 239, 105–114.
from two-phase flow simulations directly on digital images of porous rocks. Yang, R.Y., Jayasundara, C.T., Yu, A.B., Curry, D., 2006. DEM simulation of the flow of
Transp. Porous Media 94 (2), 487–504. grinding media in IsaMill. Miner. Eng. 19 (10), 984–994.
Robinson, M., Cleary, P.W., 2011. The influence of cam geometry and operating Zhang, Y., Reuterfors, E.P., McLaury, B.S., Shirazi, S.A., Rybicki, E.F., 2007.
conditions on chaotic mixing of viscous fluids in a twin cam mixer. Am. Inst. Comparison of computed and measured particle velocities and erosion in water
Chem. Eng. J. (AIChE J.) 57, 581–598. and air flows. Wear 263, 330–338.
Robinson, M., Cleary, P.W., 2012. Flow and mixing performance in helical ribbon Zhu, Y., Fox, P.J., 2002. Simulation of pore-scale dispersion in periodic porous media
mixers. Chem. Eng. Sci. 84, 382–398. using smoothed particle hydrodynamics. J. Comput. Phys. 182 (2), 622–645.