Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2, pp.

109125 (2007)

A REVIEW OF CFD MODELLING FOR PERFORMANCE PREDICTIONS


OF HYDROCYCLONE
M. Narasimha*, Matthew Brennan and P. N. Holtham
Julius Kruttschnitt Mineral Research Centre, The University of Queensland, Isles Road,
Indooroopilly 4068, Queensland, Australia.
*E-Mail: n.mangadoddy@jkmrc.com (Corresponding Author)
ABSTRACT: A critical assessment is presented for the existing numerical models used for the performance prediction
of hydrocyclones. As the present discussion indicates, the flow inside a hydrocyclone is quite complex and there have
been numerous numerical studies on the flows and the particle motions in hydrocyclone, with a wide range of turbulence
and multiphase models tested. Two-equation k- and RNG k- models flow velocities with empirical modifications
were led to poor results, especially the tangential components in comparison with experimental measurements. Most of
the recent studies have utilized the Reynolds stress models (RSM) with different degrees of complexity in the pressurestrain correlation. These RSM studies showed good agreements with velocity measurements. Unfortunately, the velocity
profiles were not validated in most of the RSM cases where multiphase particle tracking were applied. Finally, large
eddy simulation (LES) is the most advanced turbulence model applied in recent hydrocyclone numeric studies. Besides
the additional information on prcising the air core correctly, LES provides an additional accuracy in predicting the
velocity profiles or the grade efficiency in comparison to the RSM.
The multiphase models have been successfully applied in a hydrocyclone to model the Lagrangian motions of spherical
particles. Eulerian-Eulerian model have been used to account for the particles effect on the fluid viscosity. Simplified
Eulerian model (mixture) model predictions for solid transportation in cyclone were well predicted. Further, the
inclusion of modified slip velocity calculation in the Mixture model improves the efficiency predictions close to the
experimental data at low feed solid loadings. In future studies, the focus should be to model the three-dimensional flow
in a hydrocyclone using at least the Reynolds stress model/LES. The particle tracking should at least include the effects
of the turbulence on the particles. All these developed models will only applicable to low feed solid concentration levels.
Since most of these models neglect the particle-particle interactions, a more comprehensive numerical method of
modified Mixture model is applied for simulating solids flow in hydrocyclones for high feed solids concentration.
Explicit models for accounting hindered settling and turbulent diffusion investigated for high feed solid concentrations
in industrial cyclones are encouraging.
Keywords:

hydrocyclone, multi-phase modelling, computational fluid dynamics, classification, turbulence, viscosity

particulate material based on size from the medium


in which it is suspended.
The flow behavior in hydrocyclone is quite
complex. This complexity of fluid flow in cyclone
is basically due to the existence of different size
particles as well as the dominance of turbulent
length scales on separation. The complexity of flow
processes has led designers to rely on empirical
equations
for
predicting
the
equipment
performance. These empirical relationships are
derived from an analysis of experimental data and
include the effect of operational and geometric
variables. Different sets of experimental data lead to
different equations for the same basic parameters.
However, these models have limitationsthey can

1. INTRODUCTION
Hydrocyclones are widely used in the mining and
chemical industries, mainly due to their design and
operational simplicity, high capacity, low
maintenance and operating cost, and small physical
size. A typical hydrocyclone consists of a
cylindrical section with a central tube connected to
a conical section with a discharge tube. An inlet
tube is attached to the top section of the cylinder.
The fluid being injected tangentially into
hydrocyclone causes swirling and thus generates
centrifugal force within the device. This centrifugal
force field brings about a rapid classification of

Received: 10 Dec. 2006; Revised: 10 Feb. 2007; Accepted: 2 Mar. 2007


109

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

air/water interface was treated using a shear free


boundary condition. Turbulence anisotropy was
incorporated into the model by using a modified
mixing length turbulence model where a different
mixing length constant was used for each
component of the momentum equation. Although
the model required calibration, it was able to predict
velocities measured by these authors using Laser
Doppler Anemometry with reasonable accuracy.
k-epilson models intrinsically make the assumption
that the turbulence is isotropic because only one
scalar velocity fluctuation is modelled. Further the
Bousinessq approximation on which the eddy
viscosity relies intrinsically implies equilibrium
between stress and strain. These assumptions are
known to be unrealistic for swirling turbulent flows
and this would suggest that k-epilson models are
not suitable for modelling turbulence in
hydrocyclones and this has been shown to be the
case by Ma et al. (2000), Sevilla and Branion
(1997),
Petty
and
Parks
(2001),
Witt et al. (1999) and Delgadillo and Rajamani
(2005) and others. However Dyakowski and
Williams (1993) have suggested that the
k-epilson model can be used on small (<44 mm
radius) hydrocyclones. To address this, other
authors have used the RNG k-epison model with the
swirl correction (Fraser et al., 1997; He et al., 1999;
Suasnabar, 2000; Scheutz et al., 2003; Narasimha et
al., 2005). However Suasnabar (2000) found that
the swirl constant in the RNG model needed to be
increased to improve predictions but beyond a
certain point, further increases caused numerical
instability. As an alternative, Suasnabar (2000)
adjusted the constants in the standard k-epsilon
model but acknowledged that this approach was
limited.
Stress transport models, in particular the full
Differential Reynolds Stress model (DRSM), such
as that developed by Launder et al. (1975), solve
transport equations for each individual Reynolds
stress. This enables stress transport models to model
anisotropic turbulence and strained flows where the
Bousinessq approximation is known to be flawed.
Whilst more computationally intensive than kepsilon models, stress transport models are being
used to model turbulence in hydro-cyclones.
Boysan et al. (1982) used an algebraic stress model
but the full DRSM model has been used in more
recent work. Cullivan (2000), Suasnabar (2000),
Slack (2000), and Brennan (2003) have all used
variants of the Launder et al. (1975) model.

only be used within the extremes of the


experimental data from which the model parameters
were determined. In view of this shortcoming,
mathematical models based on fluid mechanics are
highly desirable. Alternatively they can be
modelled
more
fundamentally
by
CFD
(Computational Fluid Dynamics). In this paper, a
comprehensive review of the CFD modelling for
performance predictions in hydrocyclones is
described.
Previous literature reviews on the numerical
modelling of hydrocyclones have been published by
Narasimha et al. (2006a), Pericleous (1987),
Chakraborti &Miller (1992), and Nowakowski et al.
(2004). Since these reviews were published,
different methods had been developed and tested
independently, and an updated review is necessary
to develop a comprehensive CFD model for
predicting
the
performance
of
industrial
hydrocyclones for a wide range of design and
operating variables. Three main phenomena are of
interest to the modelling of particle classification in
a hydrocyclone: (i) turbulent flow in the
hydrocyclone, (ii) the influence of the pulp
rheology on the flow field, and (iii) multiphase flow
with different size particles (Narasimha et al.,
2006b; Hsieh and Rajamani, 1991; and Sevilla and
Branion, 1995). Extensive work had been done in
each of these areas, but combining them cohesively
to develop a comprehensive model for predicting
performance of hydrocyclone for industrial
applications would be an enormous task. This
review outlines the important development in each
of these fields.
2. CFD MODELLING OF CYCLONESTURBULENCE MODELLING
Industrial hydrocyclones typically operate at
velocities where the flow is turbulent. However the
strong swirl and the flow reversal and flow
separation near the underflow introduce anisotropy
and strain into the turbulence. Most hydrocyclones
in mineral processing applications develop an air
core and the free surface between the air and the
water introduces further turbulence anisotropy.
These
characteristics
make
modelling
hydrocyclones using CFD difficult and the addition
of solids adds even more complexity.
Hsieh(1988) and Devullapalli (1994) modelled
hydrocyclones using a 2D axi-symmetric grid
where the air core was not resolved and the
110

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

approaches such as the Mixture and VOF models


and the Lagrangian approach.
The full Eulerian multiphase flow approach, where
a set of continuity, momentum and turbulence
equations for each phase has been used for systems
with very high dispersed phase concentrations,
where solid/solid interactions carry a significant
amount of the stress. The disadvantage of the full
Eulerian multiphase modelling approach has been
its
high
computational
cost.
Further
implementations in commercial CFD codes have
until recently been limited to using the
k-epsilon/RSM models for turbulence.
The Lagrangian approach where the paths of
individual particles are tracked based on the
velocity predicted by a CFD simulation of the fluid
is suited to systems where the dispersed phases are
dilute and where the particles interact mostly with
the fluid without significantly changing the fluid
transport properties. In particular the Lagrangian
approach is well suited to systems where small
numbers of large particles are encountered.

However even here the predictions are not what


they could be and there is debate about appropriate
modelling options. Whilst Slack (2000) found that
the DRSM model gave good predictions of
velocities in gas cyclones, both Brennan (2006) and
Delgadillo and Rajamani (2005) found that the
DRSM, where the air core was being resolved with
the VOF model, under predicted tangential
velocities in simulations of the 75 mm
hydrocyclone (Hsieh, 1988). Cullivan (2003) has
suggested that a DRSM simulation of a
hydrocyclone needs to use the Quadratic Pressure
Strain correlation of Spezial et al. (1991) as a
minimum. However our experience is that velocity
predictions from the Spezial model (1991) and the
simpler linear pressure strain model of Launder et
al. (1975) are much the same once the air core is
established. Further we have also found that the
constants in the linear pressure strain correlation
need to be adjusted to match velocity predictions.
This implies that even the Launder et al. (1975)
DRSM has limitations for this problem.
Recent advances in computational power have
begun to make Large Eddy Simulation (LES)
practical for engineering problems and the fact that
LES resolves the large turbulent structures without
modelling suggests that it should be appropriate for
modelling cyclone separators. LES is intrinsically a
dynamic simulation and requires a 3D grid. Slack et
al. (2000) has modelled gas cyclones using LES and
found good predictions of the velocities but the
technique needed a finer grid than the DRSM
simulation of the same geometry. De Souza and
Silveria (2002) modelled the 76 mm hydro-cyclone
of Debair (1983) but this was without an air core.
However both Delagdillo and Rajamani (2005) and
Brennan (2006) found that Large Eddy Simulation
(LES) gave the best overall velocity predictions
although this was at extra computational expense in
part due to finer grid requirements and shorter time
steps.

3.1

Lagrangian models for cyclones

By balancing the forces that act on a particle in


motion in a carrier fluid, a particle can be tracked
along its trajectory. Additionally corrections of the
particle trajectory due to interaction with its
surrounding environment can be included. The
influence of particles on the fluid can be included
by considering a source term in the governing
equations of the fluid. Also turbulence dispersion of
the particles can be included (Soo, 1990 and Crowe
et al., 1996).
Following the path of a solid particle, the general
equation of motion based on the effects treated by
Basset, Boussinesq, and Oseen is given (Soo, 1990)
by:
du p 4 3
4 3
Dp p
D p p FD (u f u p )
=
dt
3
3
4 3 P 1 4 3
d
Dp

Dp f
(u f u p )
+
3
r 2 3
dt
tp
(d / d )(u f u p )
+ 6 D p2 f f d
+ Fe
t p
t p0

3. MULTIPHASE CFD MODELLINGPERFORMANCE PREDICTIONS


The flow in a hydroyclone is a multiphase flow
consisting of solid particles which are dispersed
throughout water. In addition there is the air core.
Multiphase flows can be solved by a number of
CFD techniques. These include the full Eulerian
multiphase
approach,
simplified
Eulerian

(1)
where u f and u p are velocities of the fluid and the
solid particle, f and p are the densities of the
111

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

fluid and solid particles, Fe is the external force

u z ,slip

due to potential field, and FD is the time constant


for momentum transfer due to drag force defined
by:
18 CD Re p

(u u p )
FD = D p2

p 24 f

(2)

(3)

The application of this equation is also limited to


particle diameters much less than the local
turbulence length scale (Crowe et al., 1996).
In hydrocyclones, modelling, the complete equation
(1) has not been considered. Most of the previous
numerical studies included only the drag,
centrifugal and Coriolis forces in their calculation
of the particle trajectory. While the centrifugal and
Coriolis forces are determined from the velocities
alone, the drag force requires additional modelling.
The particles are often assumed to be travelling in a
Stokesian flow, where the particle Reynolds
number is small. Other simulations have accounted
for the history force by taking the life-time of the
eddy through which the particle is travelling into
account.
Heieh (1988), Hsieh and Rajamani (1990), He et al.
(1990), Rajamani and Milin (1992), and Monredon
et al. (1992) only considered the drag and
centrifugal forces. Particle dispersion due to
turbulence of the fluid was neglected by both
groups of researchers. In all these studies, an
expression was derived for the radial direction by
balancing the centrifugal force with the radial
component of the drag force. Similarly, for the axial
direction, the drag force was balanced with the
gravity force. Additionally, for tangential direction
He et al. (1999) assumed that the particles and the
fluid have the same tangential velocity and
expressed the particle slip axial and radial velocities
as:

ur , slip

4 m u2 d p

= g

3 l rCD

0.5

(5)

This assumption can only be true if the Stokes


number of the particle is less than one (in fact the
models were developed for particles in the size
range of microns) and calculations were carried out
for dilute concentration.
Despite the simplicity of the model, all the
calculated fractionation efficiency agreed well with
experimental data. The particle trajectories from
Hsiehs simulation are shown in Fig. 1.
Applying these equations, Monredon et al. (1992)
calculated the grade efficiency curve which is
shown in Fig. 2(a).
The grade-efficiency curve is quite well predicted
but the model predicts greater fractionation than the
experiments at large particle diameters. Though the
model is able to predict the liquid-phase velocity
profiles accurately, it seems to have difficulty in
predicting separation efficiency in the coarser size
range. In such cases the only explanation would be
that the model is perhaps not accounting for certain
transport mechanisms found in the actual
hydrocyclone operation. The short-circuiting flow,
which carries the coarse particles to the overflow
stream without experiencing proper classification, is
a most likely candidate. It should be noted that the
short-circuiting flow has been implicitly accounted
for in the model formulation but not adequately.
The cause of this disagreement is attributed to the
inability of the numerical simulation to predict the
short-circuit flow. In the experiment, the large
particles are carried to the vortex finder with the
short-circuit flow and this phenomenon would
worsen the performance on the grade-efficiency
curve. An involute creates less radial velocity
components near the inlet and the short-circuit flow
can be reduced. The collection efficiency in such a
hydrocyclone could be accurately predicted in the
2DA model. The experimental results from a
cyclone using an involute inlet are shown in
Fig. 2(b), and better agreements between the
experimental and numerical curves are observed.

Finally, the drag coefficient can be expressed as a


function on Reynolds number and is written as:

CD = CD (Re)

4 m gd p

= g

l
CD

0 .5

(4)

112

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

(a)

(b)

(c)

Fig. 1 Particle trajectories of different trial in the simulations showing (a) fluid streamline, (b) effects of initial position,
and (c) effects of particle size (Hsieh and Rajamani, 1991).

(a)

(b)

Fig. 2 Comparison between the calculated and experimental grade efficiency from Monredon et al. (1992). (a) shows the
grade-efficiency curve taken in a 75 mm hydrocyclone with a single tangential inlet with 4.98 wt. % limestone and (b)
shows the grade-efficiency curve taken in a 150 mm hydrocyclone with an involute inlet tube (i.e. axisymmetric inlet flow)
with 9.01 wt. % limestone.

113

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

particles from different initial positions. The


particle with the inner-most position reached the
vortex finder without traveling through the cyclone
body. The stochastic turbulence effects on the path
of a single particle is shown in Fig. 3(c), and one
can observe the random velocity path due to the
probabilistic turbulent oscillation that was included
in the stochastic models.
Further, to calculate the particle trajectory, Ma et al.
(2000) had to determine the particle drag force. A
method similar to that used by Boysan et al. (1982)
was used, but the velocity fluctuation due to
turbulence was not included in the formulation of
the drag force equation. The grade-efficiency curve
was determined numerically. Although the
calculated particle cut-off diameter agrees with the
experimental data, the grade efficiency curve shows
that particles with small diameters respond very
differently in numerical simulation than in
experiments. Ma et al. (2000) attributed the
difference to the large effects of turbulent flow
oscillation on the smaller particles with smaller
inertia.
The inclusion of the stochastic turbulence
fluctuation eliminates the sharp cut in particle
diameter in terms of the fractionation efficiency.
Fig. 4(a) shows the experimental and numerical
efficiency curves of the high efficiency Stairmand
cyclone. Boysan et al. (1982) attributed the
discrepancies in the results to possible numerical
deficiency in boundary layer definition, reentrainment of particles via the apex, large intervals
in experimental particles size, and experimental
error in measurement of the particle diameter.
Nevertheless, the numerical collection efficiency
showed good agreement with the experiment.
The eddy life-time method was used again in
another study by Griffiths and Boysan (1996). The
same multiphase model described above was
applied to the same Stairmand high efficiency
cyclone whose length scale was increased by 50%
from the Boysan et al. (1982) study. Despite the
change in geometry, the results could be compared
by matching both the swirl number and Reynolds
number of the cyclone. The 73-mm diameter vortex
finder curve in Fig. 4(a) shows the grade-efficiency
of the Stairmand cyclone with a Sw of 3.80 and a
Re of 1.9105 calculated with a ARSM model and
a eddy life-time particle tracking method. Fig. 4(b)
shows the grade-efficiency of the Stairmand
cyclone with a swirl number of 3.80 and a Re of
2105 calculated with a RNG k- model and a

The influence of the carrier turbulence on the


particles trajectory can be included with stochastic
models (Crowe et al., 1996) or more recently with
the cloud model (Baxter and Smith, 1993). A
stochastic model was implemented for gas cyclones
by Boysan et al. (1982). The stochastic particle
tracking technique described above was used to
determine the paths of 10 particles, from 1 to 10 m
in diameter, from the same initial position, and the
paths of 3 m particles released at five different
inlet positions. These simulations were performed
in the geometry of a Stairmand high efficiency gas
cyclone. The mean results are show in Fig. 3(a) and
3(b). Fig. 3(c) shows the random path of a 2 m
particle under the influence of turbulence
fluctuation.

(a)

(b)

(c)

Fig. 3 Particle trajectories results from Boysan et al.


(1982) showing (a) different particle sizes from the same
initial point, (b) 3 m particle from four different inlet
positions and (c) particle trajectories in the stochastic
turbulent flow field.

From Fig. 3(a) one can observe that large particles


are quickly collected at the wall, while the other
smaller particles follow the streamline of the flow
and show pattern of recirculation. The 1 m particle
exits through the vortex finder, while the 2 m
particle is trapped in a loop and its position at the
end of the calculated time interval is marked by a
circle. Fig. 3(b) shows the trajectories of five 3 m
114

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

eddy life-time particle tracking method. The


simulated velocity field, from which the particle
trajectories and the grade efficiency shown in
Fig. 4(b) were calculated, was not validated against
experiments. However, it is able to produce a
satisfactory behaviour in the grade-efficiency curve.
This leads one to suspect that accurate velocity
prediction, using second moment turbulence
closure, is not necessary in calculating the gradeefficiency curve.

Most of above studies have aimed to simulate only


the flow of water and solid in a hydrocyclone to
understand the flow physics; very few attempts
have been made to predict the performance of
cyclone. Narasimha et al. (2005) made an attempt to
develop a comprehensive CFD model to predict
cut-size of the separation along with water splits in
a laboratory 102 mm hydrocyclone using
Lagrangian stochastic modified k- turbulence
model superimposed on continuous water phase at
low feed concentrations. The model predictions are
all over predicted compared to experimental values.
The apparent over-prediction of the computed
collection efficiencies for the larger particles may
be due to the occurrence of turbulent bursts in the
wall boundary layer which may cause parcels of
particles to shoot in the radial direction towards the
axis of the cyclone, and to re-entrainment of
particles from the vortex finder. This phenomenon
is associated with underestimation of swirling
patterns with the k- turbulence models as
discussed in the review of turbulence models. The
discrepancies for fine sized particles were due to
used
model
unaccountability
of
air-core
estimations.
Maxey postulated that particles would accumulate
in regions with high strain rate or low vorticity
(Maxey, 1987). Dyakowski and Williams (1996)
observed in their simulations that the strain rates are
large near the inlet and the vortex finder due to the
generation of swirl components, the changing of
flow direction from horizontal to vertical, and the
presence of vortex motion near the vortex finder.
These zones of large strain rate are in qualitative
agreement with the high particle concentration in
the same regions calculated by Pericleous and
Rhodes (1986), Rajamani and Milin (1992), and
observed by Abdullah et al. (1993).
These areas of high concentration are observed by
Cullivan et al. (2004). Fig. 5 shows that the large
particles have a high concentration along the wall
and near the apex. For the smaller particle, there are
two circular zones of high concentration, and they
coincide with the zones of recirculation in the same
simulation.

(a)

(b)

Fig. 4 Particle fractionation grade-efficiency curve


calculated with the eddy life-time particle tracking
method for the Stairmand high efficiency cyclone. (a) is
the cyclone with a Re of 1.9105. The curves for the
vortex finder diameters of 40, 56 and 73 mm correspond
to Swg of 2.1, 2.9 and 3.8. The solid line represents the
experimental data from Mothes et al. (1981) and the
circles are the results with a ARSM turbulence model
from Boysan et al. (1982). (b) is the cyclone with a Re of
2.0105 and a swirl number 3.8. with a RNG k-
turbulence model from Griffiths and Boysan (1996).

115

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

Fig. 5 Contour of particle concentration in the


hydrocyclone. Figure on the top shows the contour for
particles with a radius of 0.25 m and figure on the
bottom shows the contour for particles with a radius of
2.5 m (Cullivan et al., 2004). (No absolute scale of the
contours was given.).

Fig. 6 The standard deviation of the differences


between the deterministic and stochastic Lagrangian
particle path. Contours from the 0.25 m particles are on
the top and contours from the 2.5 m particles are on the
bottom (Cullivan et al., 2004). (No absolute scale of the
contours was given.).

Most studies simulated the particle motion as a


deterministic behaviour that is a balance between
drag and centrifugal forces. In contrast, Averous
and Fuentes (1997) suggest that the radial
turbulence fluctuation could transport the particles
rapidly from the wall to the core of the
hydrocyclone. Cullivan et al. (2004) concluded that
there exists a preferential particle radial direction
which is attenuated by the turbulent fluctuation
components. They plotted the standard deviation of
the difference between the deterministic path and
the stochastic path of particles in a cross section of
the hydrocyclone, and the results are shown in
Fig. 6. The effects of including the stochastic
turbulent component are particular largely visible
near the wall and the cylindrical section.

The Lagrangian approach has however been


extended to modelling cyclones at large particle
concentrations by Rajamani and Milin (1992). The
effect of solid concentration on fluid viscosity could
be directly coupled by magnitude of the solid
concentration. Rajamani and Milin (1992) and
Sevilla and Branion (1997) used the Thomas
expression for viscosity to define the modified fluid
viscosity as:

m
= 1 + 2.5v + 10.5v2 + 0.00273 exp(16.6v )
0

(6)

where m is the modified viscosity, 0 is the


liquid viscosity and v is the concentration of the
solids. The concentration, v , is proportional to the
116

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

As shown in Fig. 8, overall, the predictions showed


a much sharper efficiency than the experimental
one. However, some discrepancies are observed in
the fine and coarse particle size ranges. The
simulated efficiency curves show slightly lower
water split ration to the underflow; this was mainly
due to overprediction of the air-core diameters in
the model calculations, which has a strong influence
on the flow splits. In case of high feed
concentration (>30 wt %), some difference were
observed in the particle size range of 530 microns,
an inadequate estimation of hindered settling while
calculating the Re- CD correlations. Another
important issue while tracking the particle PDF
through the flow domain is the interaction of the
cloud with the walls. Also, turbulent diffusion
which was not accounted for this modelling work
may contribute to fine particle migration.

particle residence time in a mesh element and the


size of the particle. Increasing particle residence
time or increasing particle diameter would both
increase the concentration scalar, v . The method
was used to predict limestone partition curves for
feeds with up to 35% by weight limestone with
good accuracy. The technique also predicted
limestone concentrations, but these were not
compared to experimental data.
The calculation is initially started with the density
and viscosity of water. After one iteration, the
seeded particles trajectory through the domain is
calculated and the concentration distribution
determined. The viscosity is adjusted using
equation (6) before the Navier-Stokes equations are
solved again. The particle trajectories are now
calculated again and this process is iterated until a
converged solution is obtained. An improved
collection efficiency curve is obtained and shown in
Fig 7.

(a)

Fig. 7 The grade-efficiency curve obtained by


iteratively accounting for the non-Newtonian effects of
thick limestone (Rajamani and Milin, 1992).

Further, a probability density function (PDF) model


was used by Devulapalli (1996) to track particles in
a Lagrangian reference frame at high solids loading.
The technique involved tracking particles clouds
rather than the individual particles. The distribution
of the particles with the cloud is represented by a
PDF. The interaction between the fluid and the
particle phase were accounted for by calculating the
density and viscosity for the mixture. These
modified physical properties were used to calculate
the mean position of the particles. The solution
procedure was iterative and the particle submodel
was interfaced with the fluid-phase calculation.

(b)

Fig. 8 The grade-efficiency curve obtained by


iteratively accounting for the feed concentration effects
of thick limestone with (a) 27.7% and (b)47.2 % by
weight in a 250 mm hydrocyclone (Devulapalli, 1996).

117

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

different granular diameters using a fully coupled


RSM. The results show, as in Fig. 11, that the
10 m particles remained evenly suspended in the
solution. The ratio of amount of particles leaving
through the vortex finder and the apex is
proportional to the flow split ratio in the cyclone. In
contrast, most of the larger 30 m particles are
collected along the wall and leave via the apex at
high concentration.

A similar model (Rajamani and Milin, 1992) has


been re-investigated using LES turbulence model
by Delgadillo and Rajamani (2005) and found that
the prediction of particle classification follows very
close to the experimental values as shown in Fig. 9.
But the predictions for the high feed solid
concentrations are over-estimated compare to the
experimental classification data in a bigger cyclone
(see Fig. 10).

(a)

Fig. 9 Classification function for a 10.5% (by weight)


slurry in a 75 mm-hydrocyclone (Delgadilo and Rajmani,
2005).

(b)

Fig. 11 The predicted volume fractions for 10 and 30


micron particle sizes in a hydrocyclone using an
Eulerian-Eulerian methodology (Cokljat, 2003).

Fig. 10 Classification function for a 38.4% (by weight)


slurry in a 250 mm-hydrocyclone (Delgadilo and
Rajmani, 2005).

3.2

Also, Suasnabar (2000) used the full Eulerian


approach with granular flow modelling for the
particulate phases to model a DSM pattern dense
medium cyclone. The technique has also been used
more recently by Nowakowski et al. (2000, 2003
and 2004).

Eulerian-Eulerian models for cyclones

The full Eulerian-Eulerian interaction between


different phases has been modelled by Cokljat et al.
(2003). He simulated the flow with one liquid
phase, one air phase and four phases of particle with
118

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

3.3

directions is defined identically to equations (4) and


(5).
The simulation was run with a water phase, an air
phase and three solid phases with diameters of 5, 10
and 15 m. With increasing particle diameter, the
build-up of particles towards the underflow
increases. The particle equilibrium line, where the
particle drag and centrifugal forces are equal, also
moves towards the vortex finder as the particle
diameter increases. The closer the equilibrium line
is to the vortex finder, the more likely it is for the
particle to be removed from the hydrocyclone
through the vortex finder.
The collection efficiency from the simulations is
shown in Fig. 12 and the curve is corrected to
account for the fact that small particles with small
Stokes numbers could not be fractionated since they
follow the fluid motion exactly. The d 50 of the
simulated device is 4.5 m. The results are
compared against an empirical model proposed by
Pericleous and Rhodes (1986) and showed good
agreement. This model implicitly having the
axi-symmetric and low lever turbulence (prandtl
mixing length model) assumptions, which are
incapable of predicting highly swirling flow
patterns completely as discussed in turbulence
section.

Mixture models for cyclones

The Volume of Fluid model (VOF) (Hirt and


Nichols, 1981) and the Mixture model (Manninen et
al., 1996) are simplified Eulerian multiphase
approaches where the equations of motion are
solved for the mixture and additional transport
equations for the volume fractions of additional
phases are solved. The VOF model and the Mixture
model solve significantly less transport equations
than the full Eulerian approach and thus
numerically more efficient. The VOF and Mixture
models are implemented in commercial CFD codes
such as Fluent with the option of being used for
turbulent flows with the turbulence model enabled
for the mixture.
The basic model equations of the Mixture model are
defined as:

m mu m,i
+
= 0,
t
xi
u m , i
t

+ m g j +

u m , i u m , i
x j

xi

k k

+
x i x j

Dk ,i uDk , j

(7)
u m , i u m , j
+
m
x i
x j

(8)

k =1

where m and m are the mixture density and


viscosity and are the sums of all the density and
viscosity contributions from each phase. u Dk is the
drift velocity which is the difference between each
individual phases velocity and the mass-averaged
velocity. k is the volume fraction of each phase.
Pericleous and Rhodes (1986) coupled the particle
and fluid equations through modifying the mixture
density and the effective viscosity. The mixture
density, m , is calculated by accounting for the
effect from all phases, and is expressed as:
n

1 ci
i =1

c
i =1

(9)

where l is the liquid density, i is the particle or

Fig. 12 The grade-efficiency curve from Pericleous and


Rhodes (1986).

air density, and ci is the fraction of the i-th phase.


The turbulence viscosity is implicitly affected by
the distribution of the different phases through the
variation of fluid density. Subsequently, the slip
velocity of the particle in the radial and axial

Grady et al. (2003) used the algebraic Mixture


model which modelled each phases continuity and
momentum equations. As shown in Fig. 13, the
model successfully predicted the fractionation
119

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

350 mm DSM pattern dense medium cyclone using


a single medium size, the basic Schiller and
Naumann (1935) drag law and no viscosity
corrections. Brennan (2003) found that medium
segregation was over-predicted when compared to
Subramanians GRT data (2002). The simulations
also predicted that the highest concentration of
medium was at the wall and also that a film of pure
water was predicted to form just below the air core.
Neither of these effects was observed in
Subramanians GRT measurements (2002).
Subsequently the modelling was extended where
medium was simulated with a size distribution, and
wall lift forces based on Saffmans expression
(1965) were included, the Ishii and Mishimi (1984)
slurry viscosity model was used, and the CFD used
Large Eddy Simulation (Narasimha et al., 2006b).
In subsequent work (Narasimha et al., 2006c), a
Lagrangian approach was superimposed on the
Mixture model simulations (using medium) to
construct the partition curves for coal particles with
reasonable accuracy. In parallel to this work, the
same multiphase CFD approach was used to model
the classification efficiency of the 75 mm Hsieh
(1988) classifying hydrocyclone (Brennan et al.,
2006).
As shown in Fig. 14, the cyclone efficiency curves
on all simulations are being predicted with quite
good accuracy, except that the simulations using
Bounded central differencing on the momentum
equation predict more short-circuiting of the 25 and
35 m size fractions to the overflow. The finer grid
seems to improve the predictions at the smaller size
range, but has a small effect on the short-circuiting.
However, short-circuiting has been reduced by
using the 3rd Order Muscl scheme on the
momentum equation on the coarse grid.
Currently, authors are investigating the effect of
additional forces like Bagnold dispersive forces
additional to the saffmans lift forces using the
Mixture model superimposed with advanced
turbulence models such as LES for high feed solid
concentrations in hydrocyclones.

efficiency of particles above 20 m but overpredicted the fractionation of smaller particles.

Fig. 13 Separation purity function comparison Grady


et al. (2003).

Davidson (1994) used what was effectively the


Mixture model (Manninen et al., 1994) to resolve
particle concentrations in simulations of Kelsalls
cyclone (1952). Bagnold (1954) and turbulent
diffusion forces were incorporated into the solid
phase slip velocity calculation. Tomographic
measurements of medium segregation were not
available at the time of Davidsons work (1994) but
Davidsons predictions (1994) compare well with
the GRT density predictions of Galvin and Smithan
(1994) and Subramanian (2002) with the drop in
medium concentration at the wall being predicted
qualitatively correctly. Suasnabar (2000) used both
the full Eulerian granular approach (Ding and
Gidaspow, 1990) and the Mixture model (Manninen
et al., 1996) to model the distribution of medium in
200 mm and 350 mm DSM pattern dense medium
cyclones. Suasnabar (2000) found that both
Eulerian techniques predicted medium segregation
in a qualitatively sensible manner but the Eulerian
granular flow model predicted more correctly the
observed drop in medium concentration at the wall
in the bottom of the apex. Suasnabar (2000)
suggested that this was because the full Eulerian
granular flow model simulated Bagnold (1954)
forces on the medium through the gradient in solids
pressure in the dispersed phase momentum
equation.
The Mixture model has been used to model
dispersed phases in cyclone separators at the
JKMRC. Initially, the Mixture model with the
DRSM turbulence model was used by Brennan
(2003) to simulate medium and the air core in a
120

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

To some extent, short-circuiting of solids is also


prevailing along the vortex finder wall. A unique
radial segregation of solid particles can be seen in
section, which is resulted from the entrapment of
some relatively coarse particles in the forced vortex
flow. That is why a relatively higher solids
concentration appeared in the middle of the cyclone
apart from the cyclone wall.

Cyclone Efficiency
1
0.9

Fraction to underflow

0.8
0.7
0.6
0.5

Hsieh Series VII

0.4

Fine OF BCD

0.3

Fine UF BCD

0.2
0.1
0
0

10

15

20

25

30

35

40

Particle size - m

Fig. 14 Predicted cyclone efficiency curve (Brennan et


al., 2006). Fine grid multiphase LES. Feed mass flow
rate of water = 1.117 kg.s-1. Feed Limestone mass flow
rate = 0.0574 kg.s-1. BCDBounded central differencing
on Momentum. 3rd O 3rd Order Muscl on Momentum.
UFfraction to underflow from underflow flow rate.
OFfraction to underflow by difference from overflow
flow rate.

3.4 Simulation of high feed solids flow in


hydrocyclones
Fig. 15 Steady state contours of solids concentration in
the hydrocyclone.

The simulations were conducted on 3D body fitted


grids of Devulappallis (1997) 250 mm cyclone
geometry, which were generated in GambitTM,
were unsteady LES simulations. The modified
Mixture model (Narasimha et al., 2006c) was run
using eight phase transport equations in the
multiphase simulations, where the primary phase
was water and the dispersed phases were air and
five limestone phases of density 2700 kg m-3 with
sizes 4.25 m, 13.8 m, 27 m, 55.4 m,
110.34 m. The volume fraction of each limestone
size in the feed boundary condition was set so that
the cumulative limestone size distribution and the
total limestone volume fraction matched those used
in the feed in Devulapallis experimental runs. In
this case the feed solids concentration is about
27.2% by weight as used by Devulappallis (1997).
The flow field in hydrocyclones has been
established and well understood, but little attention
has been brought up on distribution of solids
concentration inside the cyclone. The typical
concentration contours are shown in Fig. 15. From
Fig. 15, it is observed that more or less a high solids
concentration exists along the wall of the cyclone,
indicating coarse particles presence at this zone. As
expected, most of the coarse particles will be
separated immediately after entering into the free
vortex flow domain due to high centrifugal forces.

(a)

(b)

(c)

(d)

(e)

Fig. 16 At steady state(a) 4.25 micron, (b)13.8


micron, (c)27 micron, (d) 55.4 micron and (e) 110.34
micron.

Overall, the classification of different size particles


based on the above observed information (see
Fig. 16) can be explained as follows. The fine
particles are more equally distributed in the cyclone
volume because of the predominated turbulent
diffusion. On the other hand, coarser particles are
much influenced by the centrifugal forces and
segregate at the cyclone wall. Interestingly the
121

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

and these studies have been tested against a wide


range of turbulence and multiphase models. Twoequation k- and RNG k- models flow velocities
with empirical modifications were led poor results,
especially the tangential components in comparison
with experimental measurements. Despite the
varying degree of empiricism and accuracy in
predicting the velocity profiles, all these
simulations were able to predict reasonable
classification-efficiency curves. It appears that
accurate classification-efficiency predictions do not
hinge on accurate predictions of the velocity
profiles.
Most of the recent studies have utilized the
Reynolds stress models (RSM) with different
degrees of complexity in the pressure-strain
correlation. Most of the RSM studies showed good
agreements
with
velocity
measurements.
Unfortunately, the velocity profiles were not
validated in most of the RSM cases where
multiphase particle tracking was applied. Finally,
large eddy simulation (LES) is the most advanced
turbulence model applied in recent hydrocyclone
numeric studies. Besides the additional information
on the prcising air core correctly, LES provides an
additional accuracy in predicting the velocity
profiles or the grade efficiency in comparison to the
RSM.
Presently, multiphase models have been
successfully applied in a hydrocyclone to model the
Lagrangian motions of spherical particles. EulerianEulerian model have been used to account for the
particles effect on the fluid viscosity. Simplified
Eulerian model (Mixture model) predictions for
solid transportation in cyclone were well predicted.
Further, the inclusion of modified slip velocity
calculation in the Mixture model improves the
efficiency predictions close to the experimental data
at low feed solid loadings.
In future studies, the focus should be to model the
three-dimensional flow in a hydrocyclone with at
least the Reynolds stress model/LES. The particle
tracking should at least include the effects of the
turbulence on the particles. All these developed
models will only be applicable to low feed solid
concentration levels, where mostly these models
neglect the particle-particle interactions. A more
comprehensive numerical method of modified
Mixture model should be applied for simulating
solids flow in hydrocyclones for high feed solids
concentration. Explicit models for accounting
hindered settling and turbulent diffusion

intermediate particles (here 27 microns range),


which are close to the cut-size, are equally
distributed to overflow and underflow. Generally
these particles will eventually take a long time to
report to products.
Series-II, Feed solids=27.2 w t%
100
90

Recovery to Underflow (%)

80
70
60
50

Expt

40

CFD

30
20
10
0
0

10

20

30

40

50

60

70

80

90

100

110

120

Particle size (m icrons)

Fig. 17 Comparison of predicted and measured


classification functions in a 250 mm Krebs
hydrocyclone.

The separation curve with a well-known S-shape


represents a steady state and is predicted to be
established after 5.5 sec in a 250 mm hydrocyclone.
The predicted typical classification curve is
compared with Devulapallis data (1997) and
shown in Fig. 17. Fig. 17 demonstrates that the
separation of particles by size in the large-scale
hydrocyclone is closely simulated by modified
CFD-LES model at high feed solid loadings. The
predicted cut-size is very close to the experimental
data. The beauty of the CFD model is that at high
solids loadings, it considered the interactions
between the water phase and solids phase in terms
of hindered settling correction; shear forces (lift) at
the wall and also the inter-particle interactions in
terms of Bagnold forces based on pressure gradient
of granular solids. An attempt is made in this study
to simulate the fish-hook phenomena at finer size
particles classification. It appears as a raised
partition curve below 10 microns in Fig. 17. Further
investigation on fish-hook phenomena is needed to
be done numerically in the future.
4. CONCLUSIONS
The flow inside a hydrocyclone is quite complex
and there have been numerous numerical studies on
the flows and the particle motions in hydrocyclone,
122

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

investigated for high feed solid concentrations in


industrial cyclones are encouraging.
11.
REFERENCES
1. Abdullah MZ, Conway WF, Dyakowski T and
Waterfall RC (1993). Investigation of electrode
geometries for use in cyclonic separators.
Workshop on Tomographic Techniques for
Process Design and Operation, p.127.
Manchester, England.
2. Bagnold RA (1954). Experiments on a gravityfree dispersion of large solid spheres in a
Newtonian fluid under shear. Proceedings of
the Royal Society London, Series A, 225,
4963.
3. Baxter LL and Smith PJ (1993). Turbulent
dispersion of particles: the STP model. Energy
and Fuels 7:852859.
4. Boysan F, Ayers WH and Swithenbank J
(1982). A fundamental mathematical modelling
approach to cyclone design. Trans. Inst. Chem.
Eng. 60:222230.
5. Brennan MS (2006). CFD simulations of
hydrocyclones with an air core: comparison
between large eddy simulations and a second
moment closure. Chemical Engineering
Research and Design 84A:495505.
6. Brennan MS, Narasimha M and Holtham PN
(2006) Multiphase CFD modelling of
hydrocyclones-prediction of cut size. Minerals
Engineering (in press).
7. Brennan MS, Holtham PN, Rong R and Lyman
GJ (2002). Computational fluid dynamic
simulation of dense medium cyclones.
Proceedings of the 9th Australian Coal
Preparation Conference, Yeppoon, Australia,
paper B3.
8. Brennan MS, Subramanian VJ, Rong R,
Holtham PN, Lyman GJ and Napier-Munn TJ
(2003). Towards a new understanding of the
cyclone separator. Proceedings of XXII
International Mineral Processing Congress,
Cape Town, South Africa.
9. Chakraborti N and Miller JD (1992). Fluid flow
in hydrocylones: A critical review. Mineral
Processing and Extractive Metallurgy Review
11:211244.
10. Chu Liang-Yin and Chen Wen-Mei. (1993).
Research of the motion of solid particles in a

12.
13.

14.

15.

16.

17.

18.

19.
20.
21.

22.

123

hydrocyclone.
Separation
Science
&
Technology 28(10):18751886.
Cokljat D, Slack M and Vasquez SA (2003).
Reynolds-stress model for Eulerian multiphase,
Proc. 4th Int. Symp on Turbulence Heat and
Mass Transfer, Begell House Inc. 10471054.
Crowe CT; Troutt TR and Chung JN (1996).
Numerical Models for Two-Phase Turbulent
Flows. Annu. Re. Fluid Mech. 28:1143.
Cullivan JC, Williams RA, Dyakowski T and
Cross CR (2004). New understanding of a
hydrocyclone flow field and separation
mechanism from computational fluid dynamics.
Minerals Engineering 17(5):651660.
Cullivan JC, Williams RA and Cross CR
(2003). Understanding the hydrocyclone
separator
through
computational
fluid
dynamics.
Trans.
Inst.
Chem.
Eng.
81A:455465.
Dabir D (1983). Mean Velocity Measurements
in a 3 inches Hydrocyclone Using Laser
Doppler Anemometry. PhD Thesis, Department
of Chemical Engineering, Michigan State
University, USA.
Davidson MR (1994). A numerical model of
liquid-solid flow in a hydrocyclone with high
solids fraction. Numerical Methods in
Multiphase Flows ASME Fluids Engineering
Division Summer Meeting, 2939.
Delgadillo Jose A and Rajamani Raj K (2005).
A comparative study of three turbulenceclosure models for the hydrocyclone problem.
International Journal of Mineral Processing
77(4):217230.
Derksen JJ and Van der Akker HEA (2000).
Simulation of Vortex Core Precession in a
Reverse-Flow Cyclone. American Institute of
Chemical Engineers Journal 46(7):13171331.
Devullapalli B (1994). Hydrodynamic modeling
of
solid-liquid
flows
in
large-scale
hydrocyclones. PhD Thesis, University of Utah.
Ding J and Gidaspow D (1990). A bubbling
fluidization model using the kinetic theory of
granular flow. AIChE Journal 36:523538.
Dyakowski T and Williams RA (1996).
Prediction of high solids concentration Region
within a hydrocyclone. Powder Technology
87:4347.
Fraser SM, Abdel Rasek AM and Abdullah MZ
(1997). Computational and experimental
investigations in a cyclone dust separator. Proc.
Instn. Grid. Engrs. 211E:247257.

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

36. Maxey MR (1987). The gravitational settling of


aerosol particles in homogeneous turbulence
and random flow fields. Journal of Fluid
Mechanics 174:441.
37. Monredon TC, Hsieh KT and Rajamani RK
(1992). Fluid flow of the hydrocylone: An
investigation of the device dimension.
International Journal of Mineral Processing
35:6583.
38. Narasimha M, Brennan MS and Holtham P
(2006a). A brief review of empirical and
numerical flow modelling of dense medium
cyclone. Coal preparation 26(2):5589
39. Narasimha M, Brennan M and Holtham PN
(2006b). Numerical simulation of magnetite
segregation in a dense medium cyclone.
Minerals Engineering 19:10341047.
40. Narasimha M, Brennan M, Holtham PN and
Napier-Munn TJ (2006c). A comprehensive
CFD Model of dense medium cyclone
performance. Minerals Engineering (in press).
41. Narasimha M, Sripriya R and Banerjee PK
(2005). CFD modelling of hydrocycloneprediction of cut-size. Int. J. Mineral. Process
71(1):5368.
42. Nowakowski AF, Kraipech W, Williams RA
and Dyakowski T (2000). The hydrodynamics
of a hydrocyclone based on a three dimensional
multi-continuum model. Chemical Engineering
Journal 80(13):275282.
43. Nowakowski AF and Dyakowski T (2003).
Investigation of swirling flow structure in
hydrocyclones.
Chemical
Engineering
Research and Design 81(A):862873.
44. Nowakowski AF, Cullivan JC, Williams RA
and Dyakowski T (2004). Application of CFD
to modelling of the flow in hydrocyclones. Is
this a realizable option or still a research
challenge? Minerals Engineering 17:661669.
45. Pericleous KA and Rhodes N (1986). The
hydrocyclone classifiera numerical approach.
International Journal of Mineral Processing.
17:2343.
46. Petty CA and Parks SM (2001). Flow
predictions within hydrocyclones. Filtration
and Separation 38(6):2834.
47. Rajamani RK and Milin L (1992). Fluid-flow
model of the hydrocyclone for concentrated
slurry classification. Hydrocyclone, Analysis
and Application (ed. L Svarovsky and MI
Thew), London, 95101.

23. Galvin KP and Smitham JB (1994). Use of


x-rays to determine the distribution of particles
in an operating cyclone. Minerals Engineering
7:12691280.
24. Grady SA, Wesson GD, Abdullah M and Kula
EE (2003). Prediction of 10-mm hydrocyclone
separation efficiency using computational fluid
dynamics.
Filtration
&
Separation
40(9):4146.
25. Griffiths WD and Boysan F (1996).
Computational fluid dynamics and empirical
modelling of a number of cyclone samplers.
Journal of Aerosol Science 27(2):281304.
26. He P, Salcudean M and Gartshore IS (1999). A
numerical simulation of hydrocyclones. Trans.
Inst. Chem. Eng. 77A:429441.
27. Hirt CW and Nichols BD (1981). Volume of
fluid (VOF) method for the dynamics of free
boundaries. Journal of Computational Physics
39:201225.
28. Hoesksta AJ, Derksen JJ and Van den Akker
HEA (1999). An experimental and numerical
study of turbulent swirling flow in gas
cyclones. Chemical Engineering Science
20552065.
29. Hsieh KT (1988). A phenomenological model of
the hydrocyclone. Ph D Thesis, University of
Utah.
30. Hsieh KT and Rajamani RK (1991).
Mathematical model of the hydrocyclone based
on the physics of fluid flow. AIChE Journal
37:735746.
31. Ishii M and Mishima K (1984). Two-fluid
model and hydrodynamic constitutive relations.
Nuclear Engineering and Design 82:107126.
32. Kelsal DF (1952). A study of the motion of
solid particles in a hydraulic cyclone.
Transactions of the Institution of Chemical
Engineers 30:87108.
33. Launder BE, Reece GJ and Rodi W (1975).
Progress in the development of a Reynoldsstress turbulence closure. Journal of Fluid
Mechanics 68:537566.
34. Ma L, Ingham DB and Wen X (2000).
Numerical modeling of the fluid and particle
penetration through sampling cyclones. Journal
of Aerosol Science 31(9):10971119.
35. Manninen M, Taivassalo V and Kallio S
(1996). On the mixture model for multiphase
flow. Valtion Teknillinen Tutkimuskeskus,
Espoo, Finland.

124

Engineering Applications of Computational Fluid Mechanics Vol. 1, No. 2 (2007)

48. Saffman PG (1965). The lift on a small sphere


in a slow shear flow. Journal of Fluid
Mechanics 22(2):385400. [Corrigendum,
Journal of Fluid Mechanics 31:624].
49. Schiller L and Naumann Z (1935). Z. Ver.
Deutsch. Ing. 77:318.
50. Schuetz S, Mayer G, Bierdel M and Piesche M
(2004). Investigations on the flow and
separation behaviour of hydrocyclones using
computational fluid dynamics. Int. J. of Min.
Process 73:229237.
51. Sevilla EM and Branion R (1997). The fluid
dynamics of hydrocyclones. Journal of Pulp
and Paper Science 23(2):J85J93.
52. Slack MD, Prasad RO, Bakker A and Boysan F
(2000). Advances in cyclone modeling using
unstructured grids. Chemical Engineering
Research and Design 78(A):10981104.
53. Smagorinsky J (1963). General circulation
experiments with the primitive equations. I. the
basic experiment. Month. Wea. Rev. 91:99164.
54. Soo SL (1990) Multiphase Fluid Dynamics,
Science Press and Gower Technical, Beijing,
China.
55. Speziale CG, Sarkar S and Gatski TB (1991).
Modelling the pressure-strain correlation of
turbulence: An invariant dynamical systems
approach. Journal of Fluid Mechanics
227:245274.
56. Suasnabar DJ (2000). Dense Medium Cyclone
Performance, Enhancements via computational
modeling of the physical process. PhD Thesis,
University of New South Wales.
57. Subramanian VJ (2002). Measurement of
medium segregation in the dense medium
cyclone using gamma-ray tomography. PhD
Thesis, JKMRC, University of Queensland.
58. Witt PJ, Mittoni LJ, Wu J and Shephero IC
(1999). Validation of a CFD model for
predicting gas flow in a cyclone. Proceedings
of CHEMECA 99, Newcastle, Australia.

125

You might also like