Minerals Engineering 62 (2014) 4354

Contents lists available at ScienceDirect

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

CFD modeling and analysis of the multiphase ow and performance

of dense medium cyclones
Shibo Kuang a,, Z. Qi a, A.B. Yu a, A. Vince b, G.D. Barnett c, P.J. Barnett c
a

Laboratory for Simulation and Modelling of Particulate Systems, School of Materials Science and Engineering, The University of New South Wales, Sydney, NSW 2052, Australia
Elsa Consulting Group Pty Ltd., PO Box 8100, Mt Pleasant, QLD 4740, Australia
c
Minco Tech Australia Pty Ltd., P.O. Box 142, Cardiff, NSW 2285, Australia
b

a r t i c l e

i n f o

Article history:
Available online 4 November 2013
Keywords:
Dense medium cyclone
Multiphase ow
Separation
Modeling and simulation
Coal preparation

a b s t r a c t
A computational uid dynamics (CFD) model is proposed to describe the multiphase ow in a densemedium cyclone (DMC). In this model, the volume of uid (VOF) multiphase model is rst used to determine the initial shape and position of the air core, and then the so called mixture model is employed to
describe the ows of the medium, coal particles and air, where the turbulence is described by the Reynolds stress model. The validity of the proposed approach is veried by the reasonably good agreement
between the measured and calculated results in terms of separation efciency. On this base, this model is
used to quantify the effects of the ratios of spigot to vortex nder diameters (U:O) and medium to coal
(M:C) on performance. The results are shown to be generally comparable to those reported in the literature. It reveals that when vortex nder or spigot diameter is varied at the same U:O ratio, the offset and
medium split nearly remain the same, however, the coal feed rate and Ep are different under the conditions considered. It is also shown that the sh-hook phenomenon is observed when spigot diameter is
equal to or slightly larger than vortex nder diameter, and a normal operation becomes less stable with
decreasing U:O ratio. The key phenomena predicted are explained by the calculated inner ows.
2013 Elsevier Ltd. All rights reserved.

1. Introduction
Dense medium cyclones (DMCs) are the work horses of the
modern coal preparation to upgrade run-of-mine coal of size 0.5
50 mm by separating gangue from product coal. The density of
valuable coal particles is generally smaller than 1500 kg/m3 while
that of rejects or gangue particles larger than 1500 kg/m3. Therefore, a uid of density about 1500 kg/m3 is needed for effective
separation. This is usually achieved by use of a mixture of water
and ne magnetite particles, and the mixture is called as medium in practice.
Designers/controllers traditionally rely on empirical equations
for predicting DMC performance as a function of variables related
to operational, geometrical and materials conditions, and in the
past many studies have been made in this direction (Davis, 1987;
Wood, 1990; Napier-Munn, 1991; Restarick and Krnic, 1991;
Zughbi et al., 1991; Ferrara et al., 1999; Honaker et al., 2000; Sripriya et al., 2001; Barbee et al., 2005). In the empirical approach
however, different sets of experimental data lead to different
equations even for the same basic parameters. Moreover, it is
difcult to identify, through an empirical approach, small decien-

Corresponding author. Tel.: +61 2 93854443; fax: +61 2 93855856.

E-mail address: s.kuang@unsw.edu.au (S. Kuang).
0892-6875/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.mineng.2013.10.012

cies in design of DMC. In principle, these problems can be overcome by numerical simulation.
The numerical modeling of DMCs is very challenging, because
the ow in a DMC is extremely complicated with the presence of
swirling turbulence, an air core and segregation of medium and
coal particles, involving multiple phases: air, water, coal and magnetite particles of different sizes, densities and other properties. In
comparison with extensive numerical studies of other cyclone separators such as gas cyclone and hydrocyclone (see, e.g. the reviews
by Cortes and Gil (2007) and Narasimha et al. (2007a)), numerical
studies of DMCs are few to date. Nonetheless, recent efforts made
to DMCs make it possible to study DMC multiphase ow and performance by numerical models, as briey reviewed below.
The mathematic descriptions required to model DMCs generally
fall into two main aspects: one is the modeling of medium ow and
another is the modeling of coal particle ow, while allowing for
their mutual interaction. Two approaches can be used for such a
purpose: the continuum approach at a macroscopic level represented by the two-uid Model (TFM), and the discrete approach
at a microscopic level represented by the combined approach of
computational uid dynamics and discrete element method
(CFD-DEM). In the TFM approach, both uid and solid phases are
treated as interpenetrating continuum media in a computational
cell that is much larger than individual particles but still small
compared to the size of process equipment. Because of its

44

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

Nomenclature
ak
d
D
DT,ij
Ep
fdrag
g
t
p
Pij
u
xi
Re

acceleration of phase k, m/s2

particle size, m
diameter, m
turbulence diffusion term
Ecart probable
uid drag force, N
gravitational acceleration, 9.81 m/s2
time, s
pressure drop, pa
stress production term
uid velocity, m/s
Cartesian coordinate
Reynolds number

Greek letters
a
volume fraction
eij
dissipation term
rt
PrandtlSchmidt number
q
uid density, kg/m3
s
solid stress tensor, kg/m/s2

computational convenience, the TFM approach is preferred in process modeling and applied research. It was used to model the medium ow in a DMC in terms of the so called mixture model by
different investigators (Zughbi et al., 1991; Brennan, 2003;
Narasimha et al., 2006; Wang et al., 2009a), where the interface
between the air and the medium was also predicted by the same
approach. On the other hand, in the CFD-DEM approach, the motion of discrete particles is obtained by the DEM which applies
Newtons laws of motion to every particle, and the ow of continuum uid is described by the local averaged NavierStokes equations that can be solved by the traditional CFD, with the coupling
of CFD and DEM through particleuid interaction forces. Chu
et al. (2009a) developed the rst CFD-DEM model for DMCs by
combing mixture model for the medium ow and DEM for the coal
particle ow, and used it to study various aspects in the DMC process (Chu et al., 2009b, 2012a,b,c). Although theoretically rational,
the CFD-DEM approach is computationally very demanding. It is
thus applied mainly to coarse coal particles. Even in this case, each
CFD-DEM simulation of DMC lasts for a long time to obtain meaningful results, from a few weeks to a few months depending on
applications and computer facilities. This problem does not happen
to Lagrangian particle tracking (LPT) method, which only traces the
motion of one particle and can be thought as a simplied DEM
model. By combing LPT with mixture model, different investigators
studied the separation behaviors in DMCs (Zughbi et al., 1991;
Narasimha et al., 2007b; Wang et al., 2009a,b, 2011). Based on
the results generated by the CFD-LPT and CFD-DEM simulations,
Chen et al. (2012) recently developed a PC-based model for
optimizing design and operation of DMCs. However, The CFD-LPT
approach should be limited to DMC operations at a large M:C
(medium-to-coal) volume ratio, because it ignores the effect of
inter-particle interactions of coal and the reaction of coal particles
on the uid.
In order to overcome the deciencies associated with the
CFD-LPT approach and alleviate the computational loading in
DMC simulations, a mathematical model, which describes the
ows of the medium and coal particles as well as air in DMCs in
terms of continuum mixture model, is proposed in this study. First,
the applicability of the proposed model is veried by comparing
the measured and calculated separation efciency. This model is

gt
cH
/l,k
/ij

turbulent diffusivity
collisional dissipation of energy, J
energy exchange between the lth solid phase and the
kth solid phase, J
pressure strain term
granular temperature, m2/s2
uid viscosity, kg/m/s

Subscripts
c
corrected
cut
cut size
dr
drift velocity
k
phase k
l
phase l
L
liquid
m
mixture
mag
magnetite
w
water
r
radial
t
tangential

then used to study the effects of the ratios of M:C and vortex nder
to spigot diameters (U:O) on DMC ow and performance. This
study suggests that the developed CFD approach offers a convenient way to examine the ow and performance of DMCs in relation to geometrical, operational, and materials conditions.
2. Simulation method
2.1. Model strategy
Because of the complexity of the ow in a DMC, the modeling
is developed into two steps, similar to the previous CFD-LPT and
CFD-DEM modeling procedures (Chu et al., 2009a; Wang et al.,
2009a) (see Fig. 1). In step 1, only air and slurry with a certain
density are considered. The two phases are treated as uids of
homogeneous viscosity and density. The turbulent ow of gas
and liquid is modeled using the Reynolds stress model (RSM),
and the interface between the liquid and air core is modeled
using the Volume of Fluid (VOF) model (Hirt and Nichols,
1981). The primary air core position and the initial velocity distribution are here obtained and used as part of the initial conditions
in the next step.
In step 2, magnetite and coal particles are added to simulate the
ows of the medium, air, coal, and magnetite in DMCs and estimate the separation performance. The multiphase model is changed from the VOF model to the mixture model. Also, a correction
is needed to estimate the viscosity effect of size distribution of
magnetite particles, as suggested by Wang et al. (2009a). In this
step, detailed density and velocity distributions of different phases
(air, water, magnetite, and coal) are obtained.

Fig. 1. Steps used in the present modeling.

45

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

2.2. Mathematical model

Table 1
DMC geometric parameters in the present simulation.

Parameter

Symbol

Dimensiona

Diameter of the body, mm

Side length of inlet (involute), mm
Diameter of vortex nder, mm
Diameter of spigot, mm
Length of cylindrical part, mm
Length of vortex nder, mm
Length of conical part, mm

Dc
Li
Do
Du
Lc
Lv
Lp

1000
266
450 (240674)
337 (225630)
1200
700
1880

For base case, with their varying ranges in the brackets.

Therefore, the whole process involves three CFD models (RSM,

VOF, mixture model) and one viscosity correction model, as
described below. Note the VOF model is well documented elsewhere (Nowakowski et al., 2004; Wang and Yu, 2010) and not
included in this paper for brevity.

The mathematical model proposed in this study is a simplied

TFM model, also known as mixture model. In the model, both
uid (liquid and air) and solid phases (magnetite and coal particles) are treated as interpenetrating continua. Particles of different
sizes or densities represent different phases. The ow of liquid
gassolid mixture (as a single phase) is calculated from the continuity and the NavierStokes equations based on the local mean
variables over a computational cell considering slip velocities between different phases (Manninen et al., 1996), which are given
by:

@
@
q
q u m 0
@t m
@xi m

and

Table 2
Operational conditions in the present simulation.
Parameter

Units
3

Coal density distribution

Coal particle diameter
Gas density
Gas viscosity
Water density
Water viscosity
Magnetite density
Magnetite sizes (and volume fractions)
Medium feed density
M:C ratio
Orientation angle
Gauge pressure at the inlet
a

kg/m
mm
kg/m3
kg/m/s
kg/m3
kg/m/s
kg/m3
lm
kg/m3

kPa

Valuea
13002200
0.2511
1.225
1.8  105
998.2
0.001
4945
10 (30.5%), 20 (25.6%), 30 (14.6%), 40 (11.5%), 50 (9.9%) and 80 (7.9%)
1550
5.6 (450)
10 to horizontal
158

For base case, with their varying ranges in the brackets.

Fig. 2. Schematic and grid representation of the DMC considered.

46

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

100

Partition number (%)

Partition number (%)

(a) 100
75

50

25

Mixture model (7mm)

CFD-DEM
LPT (7mm)
Experiment
Mixture model (all sizes)

0
1.2

1.4

1.6

1.8

2.0

75
M:C Ratio = 4
M:C Ratio = 5.6
M:C Ratio = 10
M:C Ratio = 15
M:C Ratio = 25
M:C Ratio = 50

50

25

0
1.2

2.2

1.4

1.6

1.8

2.0

2.2

Paritcle density (RD)

Particle density (RD)

Fig. 4. Effect of M:C ratio on partition curve when particle size is 2 mm.

Partition number (%)

(b) 100
75

3
1

50

Mixture model (2.5mm)

CFD-DEM
LPT (2.5mm)
Experiment
Mixture model (all sizes)

25

0
1.2

1.4

1.6

1.8

2.0

2.2

Particle density (RD)

Partition number (%)

(c) 100

Mixture model (0.5mm)

CFD-DEM
LPT (0.5mm)
Experiment
Mixture model (all sizes)

75

Fig. 5. Denitions of the sections used in this work.

50

25

turbulence problems as encountered in DMCs, is adopted for such

a purpose:

0
1.2

1.4

1.6

1.8

2.0



@ 
@ 
q u0 u0
q uk u0i u0j DT;ij Pij /ij eij
@t m i j
@xk m

2.2

Particle density (RD)

Fig. 3. Comparison of the simulated and measured partition curves for different
sized particles: (a) 411 mm, (b) 1.44 mm, and (c) 0.251.4 mm.

In Eqs. (1)(3), the mass-averaged velocity umi, mixture density

qm and mixture viscosity lm of a mixture are respectively dened

based on all phases involved:

n
@
@
@p
@ X
qm umi
qm umi umj 

p
@t
@xj
@xi @xi k3 k
 

@
@umi @umj
lm

@xj
@xj
@xi

@ 
qm u0mi u0mj

@xj
!
n
@ X
qk udr;ki udr;kj g qm

@xj k1

n
X

ak qk uki

umi

qm

k1

qm
n
X

ak qk

k1

where g is the gravitational acceleration, pk is the solid pressure,

udr;ki is the drift velocity, and qm u0mi u0mj is the Reynolds stress term
which includes turbulence closure and must be modeled to close
Eq. (2). RSM, which is necessary to describe the anisotropic

lm

n
X

ak lk

k1

where n is the number of phases, and k represents different

phases where k = 1 for water (the primary phase), and 2 for air,
3n for the kth type of coal or magnetite particles (the secondary
phases).

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

47

Fig. 6. Distributions of coal particles of different densities: (a) inside the base DMC, and (b) on the wall when M:C ratio = 15.

The volume fraction of phase ak is obtained according to the

continuity equation for phase k:

@
@
@
ak qk
ak qk umi 
ak qk udr;ki
@t
@xi
@xi

The drift velocity is determined by the algebraic slip mixture

model assuming that the phases should be reached over a short
spatial length (Manninen et al., 1996):

udr;ki


 X

2
n 
qk  qm dk
g rak ra1
ak qk u1k;i

ak;i  t

18l1 fdrag
rt ak
a1
qm
k1

where rt is the PrandtlSchmidt number set to 0.75, and gt is the

turbulent diffusivity. ak,i is the acceleration of phase k. fdrag is the
drag force on particles or air bubbles, which is determined according to the well known Ergun (1952) and WenYu correlation (1966)
for particles and Schiller and Neumann correlation for air bubbles
(1933).

When Eq. (7) is applied to air phase to predict the air core, the
concept of air bubble is introduced to calculate the interaction
force (i.e. drag force) between air and liquid. In this study, the bubbles are assumed as rigid spheres having a constant diameter. The
bubble diameter is set to 105 m after some tests, so that the mixture model gives almost the same air core as obtained by the VOF
under the same condition. Note that the VOF model is described by
the same governing equations as used in the mixture model except
that its continuity equation for evaluating air volume fraction excludes the item of drift velocity, and do not need bubble size in
the model. This is because unlike the treatment of interpenetrating
phases in the mixture model, the phases in the VOF are treated to
segregate totally with a clear interface between gas and liquid.
Therefore, the corresponding air volume fraction in a cell is discontinuous, which, however, is still within the range of 0 and 1 and the
number is then used to identify gasliquid interface in cyclones, as
done in the mixture model.
The solid properties such as viscosity and solid pressure are described by the kinetic theory based on the algebraic model of temperature model (Syamlal et al., 1993):

48

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

pk I sk : ruk  cH /l;k 0

where pk I ss : ruk is the generation of energy by the solid

stress tensor, cH is the collisional dissipation of energy, and /l;k is
the energy exchange between the lth uid or solid phase and the
kth solid phase.
The above discussed mixture model is in principle similar to
that recently proposed for hydrocyclones (Kuang et al., 2012),
whose main differences from others (Zughbi et al., 1991;
Narasimha et al., 2006; Wang et al., 2009a) lie in the modeling of
solid properties and particleuid interaction. Our numerical tests
however indicated that this model cannot properly predict the
measured performance when directly applied to DMCs. On the
other hand, the previous modeling of DMCs showed that the
medium viscosity is an important parameter describing the behavior of the medium ow and the separation of particles in DMCs
(Narasimha et al., 2006; Chu et al., 2009a; Wang et al., 2009a,b).
In those studies, the effect of coal on the rheology of the medium
was ignored, and hence the mixture viscosity was approximated
by the medium viscosity given by some established correlations.
Such a treatment is also adopted in this study. Note that medium
viscosity is dependent of many variables, such as magnetite
particle size distribution, particle shape, medium density and
medium contamination, and is complicated to determine. For
simplicity, it is assumed to be controlled by the solid volume
fraction of magnetite, and calculated by the modied correlation
of Ishii and Mishima (1984) by Wang et al. (2009a) according to
the experimental study:

lm 3:8lw 1 

amag 
0:62

4. Results and discussion

4.1. Model applicability
As described in Section 2, the proposed modeling involves two
steps, because the ow in a DMC is complex and difcult to measure. The stepwise approach offers a way to use the existing data to
verify the proposed model.
The VOF model for step 1 is the same as previously used in the
modeling of the gasliquid ow in a hydrocyclone. The validity of
the model has been veried by the good agreement results between the measured and calculated results, as reported elsewhere
(Wang and Yu, 2006; Kuang et al., 2012). The present model can
also reproduce the key phenomena of the medium-only ow in a
DMC, as done in the study of Wang et al. (2009a). The resulting results are not included in this paper to avoid repeating. In the following, we focus on the applicability of the new mixture model
in step 2 in describing DMC performance.
Fig. 3 compares the measured and calculated partition curves,
obtained according to the mass portions of particle recovery to
the underow. In this gure, the particle density is expressed as
the relative density (RD), dened as the density ratio of particle
to water. This applies to all the following results. In addition, the
calculated results from this study are generated by two methods
using the present mixture model. In Method 1, the particles considered are classied into three groups according to particle size,
as done in the experimental measurement (Rong, 2007), and the
size distribution in each group is represented by a mean size, while

10

where lw is the water viscosity, and amag is the volume fraction of

magnetite particles. Then, Eq. (10) is used to replace Eq. (6) to determine the mixture viscosity. This is implemented via a user dened
function (UDF) in ANSYS Fluent.

0.030

1.8

0.025

Ep
D50

D50

1.7

0.020
1.6
0.015

0.010

15

30

1.5

45

M:C Ratio
Fig. 7. DMC performance as a function of M:C ratio.

100

Mass flow rate (kg/s)

Tables 1 and 2 list the geometrical and operational conditions

considered in this study. Those in the base case are selected
according to the experimental work of Rong (2007). In this way,
the validity of the proposed model can be examined by comparing
the measured and calculated results. In addition, two variables are
considered: M:C and U:O ratios. The effect of each variable is studied, while others are kept the same as those in the base case, as
listed in Tables 1 and 2.
Fig. 2 shows the geometry and mesh representation of the DMC
in the base case. The mesh consisting of 80,318 hexahedral grids is
the same as used in the studies of Chu et al. (2009b) and Wang
(2009). In the vicinity of the walls and vortex nder, the grid is
more detailed than the remainder of the cyclone. Our tests indicated that the solution is independent of the mesh size used. This
also applies to other DMCs simulated.
A pressure-inlet boundary condition is used at the cyclone inlet, and the pressure-outlet condition at both the outlets, corresponding to a normal operation in DMC practice. The gauge
pressure at the two outlets (vortex nder and spigot) is zero, corresponding to the ambient atmospheric pressure and the gauge
pressure at the inlet is 158 kPa.
All the simulations are conducted using the ANSYS Fluent CFD
software package (version 14) at the National Computational Infrastructure in Australia. 32 CPUs are assigned to each simulation. It
lasts for about 14 days for each run of simulations to ensure that
the ow simulated can achieve a steady state at which the macroscopic ow characteristics do not change much with time. Unless
otherwise noted, all the results shown are time-averaged.

Ep

3. Simulation conditions

75

50

25

15

30

45

M:C ratio
Fig. 8. Coal feed rate as a function of M:C ratio.

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

49

Fig. 9. Fluid and solid ow patterns in the base DMC when: (a) M:C ratio = 4, and (b) M:C ratio = 15.

the density distribution by 12 mean densities. Three independent

simulations are then conducted at the same M:C ratio and different
density distributions, and ignore the interplay between particle
sizes. In Method 2, only one simulation is carried out, considering
both size and density distributions of particles, thus the interplay
between particle sizes, similar to the experiment (Rong, 2007);
and the size distribution is represented by 3 mean sizes and each
of the density distributions by 4 mean densities in the simulation.
For comparison, the results obtained by both CFD-DEM and CFDLPT simulations in the studies of Wang (2009) and Chu et al.
(2009b) are also included in Fig. 3. As seen from this gure, the
mixture, CFD-LPT, CFD-DEM models all give reasonably good predictions of the measured separation efciencies for the three
groups of particle sizes, and reproduce the breakaway size phenomenon where the separation efciency sharply decreases and
slows down with decreasing particle size. Notably, the simulations
based on the mixture model with and without consideration of the
interplay between particle sizes give consistent results. This result,
to some degree, supports the CFD-LPT approach, which ignores the
interplay between particle densities, between particle sizes, and
between particle size and density, however, can reasonably describe DMC performance under a certain condition.
Overall, the results predicted by the mixture model appear better agreement with the measured results compared to the CFD-LPT
model, because the mixture model considers the reaction of particles on the uid and the particleparticle interaction, which are not
considered by the CFD-LPT model. On the other hand, compared to
the CFD-DEM model, the mixture model gives a better prediction

of separation efciency for the ne particles (Fig. 3c) but less for
the coarse particles (Fig. 3a). This may be explained as follows.
The CFD-DEM approach is theoretically more rational compared
to the mixture model (Zhou et al., 2010). However, to simulate ne
particles with the current computer capacity, the former has to
introduce a simplication to alleviate the computational loading:
an assembly of ne particles, whose properties are assumed to
be the same, is represented by a parcel particle of much larger
diameter than those of real particles. Nonetheless, the results
shown in Fig. 3 suggest that the proposed model can be used to
predict the separation behavior in DMCs, at least quantitatively.
4.2. Effect of M:C ratio
M:C ratio has been known as one of the key factors in DMC
operation. Its effect on DMC ow and performance is examined
by the present model. Here, relatively ne particles of different
densities are focused, because little attention was paid to such particles in the previous numerical studies of DMC. In addition, considering that the number of phases involved in the mixture
model cannot exceed 20 in the ANSYS Fluent, although such a limitation is theoretically not necessary, only one size is considered.
This allows us to use adequate mean densities in simulations to
represent the density distribution involved to generate partition
curves that are smooth enough for assessing DMC performance.
Fig. 4 shows the partition curves at different M:C ratios and reveals
that the separation efciency increases with the increase of M:C
ratio, in particular, when RD = 1.51.8 and M:C ratio is relatively

50

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

Partition number (%)

(a)

100
U:O ratio = 0.5
U:O ratio = 0.6
U:O ratio = 0.7
U:O ratio = 0.8
U:O ratio = 0.9
U:O ratio = 1.0
U:O ratio = 1.1
U:O ratio = 1.2
U:O ratio = 1.3
U:O ratio = 1.4

75

50

25

0
1.2

1.4

1.6

1.8

2.0

2.2

Particle density (RD)

Partition number (%)

(b) 100
U:O ratio = 0.5
U:O ratio = 0.6
U:O ratio = 0.7
U:O ratio = 0.8
U:O ratio = 0.9
U:O ratio = 1.0
U:O ratio = 1.1
U:O ratio = 1.2
U:O ratio = 1.3
U:O ratio = 1.4

75

50

25

1.2

1.4

1.6

1.8

2.0

2.2

Particle density (RD)

Fig. 10. Effect of U:O ratio on partition curve when: (a) Du = 337 mm, and (b)
Do = 450 mm.

small. A similar result was also obtained in the CFD-DEM study of

DMC by Chu et al. (2009a) for particles with diameter 5 mm (2 mm
in this study). This again suggests that the CFD-DEM and mixture
models can give consistent results, at least qualitatively. The latter
is however computationally much more efcient, particularly for

ne particles. Note that the CFD-LPT model cannot be used to study

the effect of M:C ratio due to its inherent deciency.
In order to better understand the particle behaviors at different
M:C ratios, we examine representative distributions of particles of
different densities on the vertical 22 plane dened in Fig. 5, and
the results are given in Fig. 6. In this gure (and Figs. 9, 11 and
13), the white region corresponds to the area occupied by the air
core, which is not shown for clarity because the properties of air
and liquid phases are fairly different. Three typical particle behaviors generally observed in a DMC can be identied from the coal
volume fraction distributions (Fig. 6a): (a) the light particles (e.g.
RD = 1.3 and 1.525) ow down along the outside wall of the vortex
nder to join the overow stream within the vortex nder, and
thus by-pass the separation process; (b) the particles around the
cut density (e.g. RD = 1.575) is separated around the air core from
the vortex nder tip to the spigot; and (c) the heavy particles (e.g.
RD = 1.675 and 2.179) aggregate on the wall and spiral down to the
underow. Interestingly, the helical path of particles on the wall
can be observed for both light and heavy particles under the present condition (see Fig. 6b), which should account for the groove
wear pattern, as experimentally observed by Wood (1990). Note
that the above discussed particles behaviors can be observed at
all the M:C ratios considered. However, the densities corresponding to different particle behaviors vary with M:C ratio. In addition,
it should be pointed that the ndings from Fig. 6a can also be obtained according to the particle distributions on the 11 vertical
plane. It is this case for the analysis of ow properties in the
following.
Fig. 7 shows the effects of M:C ratio on cut density D50 and Ecart
probable Ep. Here, D50 is the density of the particles reporting to the
underow at a portion of 50%, and Ep is calculated by (D75D25)/2,
where D75 and D25 are respectively the densities of the particles
reporting to the underow at portions of 75% and 25%. It is shown
that both D50 and Ep decrease sharply and then slow down. These
results are in line with the experimental observations of Wood
(1990) and Sherritt et al. (2010). In addition, a similar trend of
D50 and Ep is also observed for the coal feed rate at the inlet (Fig. 8).
Fig. 9 shows the representative distributions of pressure drops
(dened as the pressure relative to the ambient atmospheric

Fig. 11. Distributions of axial velocities in the DMC when U:O ratio = 1 (Du = 337 mm and Do = 337 mm), corresponding to Fig. 5.

51

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

(a) 0.05

(b) 0.20
Du = 337

0.04

off-set (RD)

Ep (RD)

Do = 450

0.03

0.02

0.10

Du = 337

0.01

0.00

0.15

Do = 450

0.6

0.7

0.8

0.05

0.9

0.6

0.7

(c)

0.9

(d) 120
90

100

Mass flow rate (kg/s)

Du = 337

Medium split (%)

0.8

U:O ratio

U:O ratio

Do = 450

70

50

30

80
60
40
20

Du = 337

10
0.4

Do = 450

0.6

0.8

1.0

1.2

1.4

U:O ratio

0
0.6

0.8

1.0

1.2

1.4

U:O ratio

Fig. 12. Effect of U:O ratio on DMC performance: (a) Ep, (b) off-set, (c) medium split, and (d) coal feed rate.

pressure), medium densities, medium tangential and axial velocities, and coal volume fraction on the vertical 22 plane at small
and large M:C ratios, at which the performance varies signicantly.
As seen from this gure, when M:C ratio is decreased, the coal volume fraction or amount of coal particles in the cylindrical region
increases signicantly, similar to those predicted by the CFDDEM approach (Chu et al., 2009a). However, the medium axial
velocities do not change much, particularly for the locus of zero axial velocity (the black line in Fig. 9), which divides the ow into
two parts: the downward (negative) and upward (positive) ows.
On the other hand, a smaller M:C ratio leads to increased medium
densities, and at the same time, decreased tangential velocities in
the entire DMC. Because of the decreased tangential velocities at
a smaller M:C ratio, the pressure drops generally decrease, as reected in Fig. 9. It is known that in a DMC, the separation of particles is essentially determined by the radial accelerations due to the
centrifugal force that throws a particle towards the wall (=v 2t =r)
and the pressure gradient force that moves a particle towards
the DMC center (=Dpr/q). This suggests that smaller tangential
velocities and radial pressure drops deteriorate separation of particles of different densities. It is this case for the operation at a small
M:C ratio. On the other hand, when M:C ratio is very large, the variation of M:C ratio leads to negligible variations of medium density,
tangential velocity and pressure drop in the DMC. Note that these
results are not included into this paper for brevity. Accordingly, the
DMC performance does not change much.
4.3. Effect of U:O ratio
Diameters of vortex nder and spigot are interrelated in terms
of U:O ratio, however, they are often treated as two individual

variables in optimizing cyclone design and operation (He and Laskowski, 1995; Chen et al., 2012). In this section, the relationships
among spigot diameter, vortex nder diameter and U:O ratio are
studied with respect to DMC ow and performance by the proposed model. For such a purpose, within a wide range of U:O ratios,
we change one of vortex nder and spigot diameters and x another as used in the base case to achieve the same U:O ratio, when
M:C ratio is 5.6 and particle size is 2. 5 mm.
Fig. 10 shows the predicted partition curves and reveals that the
separation efciency increases with increasing U:O ratio. Here, the
variation of U:O ratio is achieved by changing either spigot diameter or vortex nder diameter. Moreover, it is observed that when
U:O ratio is larger than 1, the separation efciencies of relatively
light particles, which are supposed to mainly report to the overow, are fairly large. This is because many of such particles ow
with the medium and report to the underow due to the relatively
large spigot diameter. Thus, a large U:O ratio cannot ensure a normal operation, and is not expected in DMC practice. Notably, when
U:O ratio is equal to 1 and 1.1 regardless a xed vortex nder or
spigot diameter, the separation efciency increases with decreasing particle density beyond a critical particle density. Thus, the partition curve presents a sh-hook shape. It should be pointed out
that sh-hook phenomenon is widely observed in hydrocyclones
(Majumder et al., 2007; Wang and Yu, 2010), which is however a
behavior in relation to particles of different sizes rather than different densities. Recently, Wang et al. (2010) also observed a similar
sh-hook phenomenon in a DMC to that in hydrocyclones.
To explain the sh-hook phenomenon obtained in this study,
we examine the details of ows in the DMCs at different ratios
and the key results are given in Fig. 11. As seen from this gure, under the condition where a sh-hook phenomenon occurs, strongly

52

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

Fig. 13. Fluid and solid ow patterns in the DMC when: (a) U:O ratio = 0.6 (Du = 337 mm and Do = 561 mm), and (b) U:O ratio = 0.9 (Du = 337 mm and Do = 374 mm).

and medium densities in the DMC are generally increased, particularly in the upper region. This increased pressure drop is not expected for high throughput but increases the possibility for light
(coal) particles to move to the DMC center. Conversely, the tangential velocities in the upper region are generally smaller at a larger
U:O ratio, leading to deteriorated DMC separation performance.
Moreover, in the DMC with a larger U:O ratio, a proportion of the
feed medium passes directly across the cyclone roof and down
the outside wall of the vortex nder to join the overow stream
within the vortex nder, which represents a short circuit ow
(point A in Fig. 13b). Furthermore the recirculating currents known

1.6

Normalised mass flowrate

mixed downward (point A) and upward (point B) ows are developed around the air core in the cylindrical region of DMC, although
an upward ow is supposed to be mainly developed in this region
for a normal operation. The downward ow brings light particles to
move downwards and report to the underow, leading to their increased separation efciencies and thus sh-hook phenomenon.
This mechanism is different from that identied for the sh-hook
phenomenon in relation to particle size, which is found to be
attributed to the signicant turbulent dispersion effect of ne particles (Wang and Yu, 2010).
Fig. 12 compares the DMC performance when U:O ratio is varied at a xed vortex nder or spigot diameter. It is observed that
with increasing U:O ratio, Ep increases (Fig. 12a), and at the same
time, the off-set (=D50qm, where qm is medium density) and medium split decrease (Fig. 12b and c). This result is in line with the
experimental observations of DMC (He and Laskowski, 1995).
Moreover, the last two performance parameters are largely the
same at both xed spigot and vortex nder diameters, whereas
Ep at a xed vortex nder diameter is generally larger. On the other
hand, when U:O ratio is increased, the coal feed rate increases at a
xed vortex nder diameter but decreases at a xed spigot diameter (Fig. 12d). Note that the operations where the separation efciency of the lightest particles is too large, are not included into
Fig. 12a and b, because their Ep and/or off-set cannot be determined by denition.
Fig. 13 shows the representative distributions of ow properties
on the vertical 22 plane at small and large U:O ratios, including
pressure drop, medium density, medium tangential and axial
velocities, and coal volume fraction. It can be seen from this gure
that at a lager U:O ratio, the pressure drops, coal volume fraction,

1.3

1.0

0.7
U:O ratio = 0.9
U:O ratio = 0.6

0.4
30

31

32

33

34

35

Time (s)
Fig. 14. Temporal variation of mass owrate at the underow when Du = 337 mm.

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

as eddy ows in the region between the outer wall of vortex nder
and the wall of body (point B in Fig. 13b) are also developed. All
these features of tangential and axial velocities explain why a larger U:O ratio leads to an increased Ep.
Fig. 13 also shows that the upward axial velocities in the spigot
region are stronger at a smaller U:O ratio and thus the rejects are
more difcult to move out of the spigot. Consequently, such particles may accumulate in the spigot region, and fall down with the
help of gravity till a certain amount is reached. This leads to significant uctuation of mass ow at the underow, which can be reected by the results shown in Fig. 14. It can be seen from this
gure that the mass owrate uctuates more signicantly with
decreasing U:O ratio, even when the operations considered are regarded as normal according to Fig. 10. This suggests that the operation at a smaller U:O ratio may be less stable. In addition, it is
noted that mixed downward and upward ows (see the axial
velocity distribution in Fig. 13) around the air core in the cylindrical region are observed at U:O ratio = 0.9. However, in this case, the
upward ow much dominates and sh-hook phenomenon is thus
not observed.

5. Conclusions
A CFD model has been developed to simulate the gasliquid
solid ow and performance of DMCs, where the mixture model,
coupled with the viscosity correction model, is used to determine
the interface between air and uid, and the motion of magnetite
and coal particles. The applicability of the proposed model has
been veried by the reasonably good agreement between the
calculated and measured results in terms of separation efciency.
The calculated results indicate that the current model can describe the effect of M:C ratio on DMC performance as observed in
the experiments: both cut density and Ep decrease sharply and
then slow down with increasing M:C ratio. The deteriorated separation performance at a smaller M:C ratio is attributed to the decreased pressure drops and tangential velocities.
The relationships among spigot diameter, vortex diameter and
U:O ratio have been quantied by the proposed model. The numerical results reveal that the deteriorated separation performance at
a larger U:O ratio is due to the presence of short circuit and eddy
ows. It is also shown that when vortex nder or spigot diameter
is varied at a given U:O ratio, the offset and medium split nearly
remain the same, however, the coal feed rate and Ep are different
under the conditions considered. A sh-hook phenomenon is observed when U:O ratio is equal to or slightly larger than 1 due to
the development of strongly mixed downward and upward ows
around the air core in the cylindrical region.
It should be pointed out that although the present study demonstrates the applicability of the proposed CFD model to DMCs,
there is a limited number (up to 20) of phases that can be simulated in the ANSYS Fluent for numerical stability, which brings difculty to the simulation of operations involving a wide distribution
of particle size/density or a strong interplay among particle properties. This problem can be overcome by combing the CFD-DEM
model for coarse coal and the mixture model for ne coal. Further
studies are necessary in the future to decide the criteria for the
combination and identify the resulting benets with regard to prediction accuracy and computational efciency.

Acknowledgements
The authors are grateful to Australia Research Council (ARC)
and Minco Tech Australia Pty Ltd for the nancial support of this
work, to the National Computational Infrastructure for the use of

53

its high performance computational facilities, and to Mr. Jiang

Chen in SIMPAS for helpful discussion.

References
Barbee, C.J., Luttrell, G.H., Wood, C.J., Bethell, P.J., 2005. Simulation of heavy medium
cyclone performance. Minerals & Metallurgical Processing 22 (1), 3842.
Brennan, M.S., 2003. Multiphase CFD simulations of dense medium and classifying
hydrocyclones. In: Proceedings of the 3rd International Conference on CFD in
the Minerals and Process Industries, Melbourne, Australia.
Chen, J., Chu, K.W., Zou, R.P., Yu, A.B., Vince, A., 2012. Prediction of the performance
of dense medium cyclones in coal preparation. Minerals Engineering 31, 5970.
Chu, K.W., Wang, B., Yu, A.B., Vince, A., 2009a. CFD-DEM modelling of multiphase
ow in dense medium cyclones. Powder Technology 193 (3), 235247.
Chu, K.W., Wang, B., Yu, A.B., Vince, A., Barnett, G.D., Barnett, P.J., 2009b. CFDDEM
study of the effect of particle density distribution on the multiphase ow and
performance of dense medium cyclone. Minerals Engineering 22 (11), 893909.
Chu, K.W., Kuang, S.B., Yu, A.B., Vince, A., 2012a. Particle scale modelling of the
multiphase ow in a dense medium cyclone: effect of uctuation of solids
owrate. Minerals Engineering 33, 3445.
Chu, K.W., Wang, B., Yu, A.B., Vince, A., 2012b. Computational study of the
multiphase ow in a dense medium cyclone: Effect of particle density. Chemical
Engineering Science 73, 123139.
Chu, K.W., Wang, B., Yu, A.B., Vince, A., 2012c. Particle scale modelling of the
multiphase ow in a dense medium cyclone: effect of vortex nder outlet
pressure. Minerals Engineering 31, 4658.
Cortes, C., Gil, A., 2007. Modeling the gas and particle ow inside cyclone separators.
Progress in Energy and Combustion Science 33 (5), 409452.
Davis, J.J., 1987. A study of coal washing dense medium cyclones. PhD Thesis,
University of Queensland, Brasbane, Austraila.
Ergun, S., 1952. Fluid ow through packed columns. Chemical Engineering Progress
48 (2), 8994.
Ferrara, G., Bozzato, P., Chine, B., 1999. Performance of conical and cylindrical
separatory vessels in dynamic dense-medium separation processes. Minerals &
Metallurgical Processing 16 (2), 815.
He, Y.B., Laskowski, J.S., 1995. Dense medium cyclone separation of ne particles
part 1. The effect of medium split ratio on dense medium cyclone performance.
Coal Preparation 16 (12), 125.
Hirt, C.W., Nichols, B.D., 1981. Volume of uid (VOF) method for the dynamics of
free boundaries. Journal of Computational Physics 39 (1), 201225.
Honaker, R.Q., Singh, N., Govindarajan, B., 2000. Application of dense-medium in an
enhanced gravity separator for ne coal cleaning. Minerals Engineering 13 (4),
415427.
Ishii, M., Mishima, K., 1984. Two-uid model and hydrodynamic constitutive
relations. Nuclear Engineering and Design 82, 107126.
Kuang, S.B., Chu, K.W., Yu, A.B., Vince, A., 2012. Numerical study of liquid-gas-solid
ow in classifying hydrocyclones: effect of feed solids concentration. Minerals
Engineering 31, 1731.
Majumder, A.K., Shah, H., Shukla, P., Barnwal, J.P., 2007. Effect of operating variables
on shape of sh-hook curves in cyclones. Minerals Engineering 20 (2), 204
206.
Manninen, M., Taivassalo, V., Kallio, S., 1996. On the mixture model for multiphase
ow. VTT Publications 288, Technical Research Centre of Finland.
Napier-Munn, T.J., 1991. Modeling and simulating dense medium separation
processes a progress report. Minerals Engineering 4 (34), 329346.
Narasimha, M., Brennan, M.S., Holtham, P.N., 2006. Numerical simulation of
magnetite segregation in a dense medium cyclone. Minerals Engineering 19
(10), 10341047.
Narasimha, M., Brennan, M., Holtham, P.N., 2007a. A review of CFD modelling for
performance predictions of hydrocyclone. Engineering Applications of
Computational Fluid Mechanics 1 (2), 109125.
Narasimha, M., Brennan, M.S., Holtham, P.N., Napier-Munn, T.J., 2007b. A
comprehensive CFD model of dense medium cyclone performance. Minerals
Engineering 20 (4), 414426.
Nowakowski, A.F., Cullivan, J.C., Williams, R.A., Dyakowski, T., 2004. Application of
CFD to modelling of the ow in hydrocyclones. Is this a realizable option or still
a research challenge? Minerals Engineering 17 (5), 661669.
Restarick, C.J., Krnic, Z., 1991. The effect of underow/overow ratio on dense
medium cyclone operation. Minerals Engineering 4 (34), 263270.
Rong, R., 2007, Industrial trials of novel cyclones. ACARP report, C14067.
Schiller, L., Naumann, Z., 1933. A drag coefcienty correlation. Z Ver Deutsch Ing 77,
318320.
Sherritt, G., Meyers, A., Bennetts, A., Graham, J., 2010. Delineation of large diameter
dense medium cyclone performance. ACARP Report. C17036.
Sripriya, R., Banerjee, P.K., Rao, P.V.T., Dutta, A., Rao, M.V.S., 2001. Critical evaluation
of factors affecting the operation of dense medium cyclones treating medium
coking coals. International Journal of Mineral Processing 63 (4), 191206.
Syamlal, M., Rogers, W., OBrien, T.J., 1993. MFIX documentation: theory guide.
National Technical Information Service, Springeld, VA, DOE/METC-9411004,
NTIS/DE9400087.
Wang, B., 2009. Modelling the multiphase ow in cyclones. PhD thesis, The
University of New South Wales, Sydney, Australia.
Wang, B., Yu, A.B., 2006. Numerical study of particleuid ow in hydrocyclones
with different body dimensions. Minerals Engineering 19 (10), 10221033.

54

S. Kuang et al. / Minerals Engineering 62 (2014) 4354

Wang, B., Yu, A.B., 2010. Computational investigation of the mechanisms of particle
separation and sh-hook phenomenon in hydrocyclones. AIChE Journal 56
(7), 17031715.
Wang, B., Chu, K.W., Yu, A.B., Vince, A., 2009a. Modeling the multiphase ow in a dense
medium cyclone. Industrial & Engineering Chemistry Research 48 (7), 36283639.
Wang, B., Chu, K.W., Yu, A.B., Vince, A., 2009b. Numerical studies of the effects of
medium properties in dense medium cyclone operations. Minerals Engineering
22 (11), 931943.
Wang, B., Chu, K.W., Yu, A.B., Vince, A., Barnett, G.D., Barnett, P.J., 2011. Computational
study of the multiphase ow and performance of dense medium cyclones: effect
of body dimensions. Minerals Engineering 24 (1), 1934.

Wen, C.Y., Yu, Y.H., 1966. Mechanics of uidisation. Chemical Engineering Progress
Symposium Series 62, 100111.
Wood, C.J., 1990. A Performance Model for Coal-washing Dense Medium Cyclones.
University of Queensland, Brisbane, Australia.
Zhou, Z.Y., Kuang, S.B., Chu, K.W., Yu, A.B., 2010. Discrete particle simulation of
particleuid ow: model formulations and their applicability. Journal of Fluid
Mechanics 661, 482510.
Zughbi, H.D., Schwarz, M.P., Turner, W.J., Hutton, W., 1991. Numerical and
experimental investigation of wear in heavy medium cyclones. Minerals
Engineering 4 (34), 245262.