Minerals Engineering 20 (2007) 6071

This article is also available online at:

www.elsevier.com/locate/mineng

CFD simulation and experimental validation studies on hydrocyclone

a,*
K. Udaya Bhaskar , Y. Rama Murthy a, M. Ravi Raju a, Sumit Tiwari a,
J.K. Srivastava b, N. Ramakrishnan a
a
Regional Research Laboratory, Hoshangabad Road, Bhopal 462 026, India
b
Ujjain Engineering College, Ujjain 456 001, India

Received 17 January 2006; accepted 25 April 2006

Available online 19 June 2006

Abstract

Hydrocyclone is a key unit operation in mineral process industry and simulation of which using CFD techniques is gaining popularity
in process design and optimization. The success of the simulation methodology depends primarily on how best the results are matching
with the experimental values and the computational time it requires for obtaining such results. In the present investigation, attempts are
made to develop a methodology for simulating the performance of hydrocyclone. Initial work included comparison of experimental and
simulated results generated using dierent turbulence models i.e., standard ke, ke RNG and RSM in terms of water throughput and
split with the help of suitably designed experiments. Among the three modeling methods, predictions using RSM model were found bet-
ter in agreement with experimental results with a marginal error between 4% and 8%. Parametric studies have indicated that a decrease in
the spigot opening increased the upward vertical velocity of water more compared to a decrease in the downward vertical velocity. An
increase in the inlet pressure has increased the axial velocities of water in both the upward and downward directions and increased the
mass ow rates through the cyclone. An increase in the inlet pressure has also increased the static pressure dierential along the radius
within the cyclone body and hence more water split into overow. Further, an increase in the inlet pressure has also increased the tan-
gential velocities and reduced the cyclone cut size. The simulated particle distribution values generated using the particle injection tech-
nique are found matching with the experimental results while achieving cut sizes between 4.9 and 14.0 lm.
 2006 Elsevier Ltd. All rights reserved.

Keywords: Computational uid dynamics; Modelling; Hydrocyclones; Partition numbers; Cyclone cut-size

1. Introduction restricted most of the research works focused on the empir-

ical modeling till recent times (Lynch and Rao, 1975; Plitt,
Hydrocyclones have been widely accepted in the area of 1976). However, the advent of high speed computational
mineral processing due to several advantages, such as ease systems, in last couple of decades made researchers focus
of operation, high throughput, less maintenance, less oor performance simulations using Computational Fluid
space requirement etc. Despite its invention in late 18th Dynamics (CFD) techniques (Pericleous and Rhodes,
century, thorough works related to understanding the prin- 1986; Hsieh and Rajamani, 1988; Monredon et al., 1992;
ciples began only in mid fties (Kelsall, 1952). Kelsall stud- Rajamani and Milin, 1992; Dyakowski and Williams,
ies on the axial, radial and tangential velocity proles 1993; Dyakowski et al., 1994; Malhotra et al., 1994; Harg-
formed the basis for subsequent research on hydrocy- reaves and Silvester, 1990; Devulapalli and Rajamani,
clones. The complex phenomenon involved coupled with 1996; Griths and Boysan, 1996; Slack and Wraith, 1997;
non-availability of high-speed computational systems Slack and Boysan, 1998; Stovin and Saul, 1998; Dyakowski
et al., 1999; Suasnabar and Fletcher, 1999; Slack et al.,
2000; Ma et al., 2000; Nowakowski et al., 2000; Grady
*
Corresponding author. Tel.: +91 755 2471956; fax: +91 755 2488323. et al., 2002; Slack et al., 2003; Nowakowski and Dyakow-
E-mail address: [email protected] (K.U. Bhaskar). ski, 2003; Grady et al., 2003; Schuetz et al., 2004; Cullivan

0892-6875/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mineng.2006.04.012
 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071 61

Nomenclature

~
g gravitational acceleration YM contribution of the uctuating dilatation in
t time compressible turbulence to the overall dissipa-
u, m velocity magnitude tion rate
m
~ overall velocity vector ak and ae the inverse eective Prandtl numbers for j and
q density e
s
~ stress tensor Sk and Se user-dened source terms
l molecular viscosity of the uid CD drag coecient
lt turbulent eddy viscosity qp density of the particle
j turbulent kinetic energy ql density of the liquid
e dissipation rate of the turbulent kinetic energy Dp diameter of the particle
Gk generation of turbulent kinetic energy due to the gx acceleration due to gravity
mean velocity gradient Fx additional force like buoyancy, etc. acting on
Gb generation of turbulent kinetic energy due to particle
buoyancy FD(up  u) drag force
C1e, C2e and C3e constants X angular velocity
rk and re the turbulent Prandtl numbers for k and e / equivalence ratio

et al., 2003, 2004; Nowakowski et al., 2004). A numerical

technique needs development of suitable methodology Over Flow
and thorough validation with the actual data prior to
applications in performance simulation and design.
Though most of these studies have validated with water Feed Inlet
Cylinder

ow characteristics, only few reported simulated results

of solids separation behavior. The present study is meant
for establishing suitable methodology including turbulence Vortex Finder
model selection for simulating both water and solid distri-
bution data on dierent geometries of cyclones. Commer-
cially available CFD software Fluent 6.1.22 was used.
The simulation results carried out on a 76 mm diameter
hydrocyclone are validated with experimental data gener-
Frustum

ated in the laboratory in terms of cyclone throughout;

water split and solids cut size.

2. Model description

2.1. Geometry
Spigot
The hydrocyclone geometry used for simulation and for
experimental studies is presented in Fig. 1. The cylindrical Under Flow
body is 76 mm diameter and 85 mm in length with a closed Fig. 1. Hydrocyclone geometry.
end at the top surface and bottom face open. A frustum
with a larger diameter of 76 mm and smaller diameter of
10 mm maintained at a cone angle of 10 is connected to 2.2. Meshing scheme
the main cylindrical body with the face having larger diam-
eter. A cylindrical vortex nder with an inner diameter of Hydrocyclones truly cannot be modeled in a 2-D plane
25 mm and outer diameter of 40 mm protrudes into the due to non-axisymmetric nature at the feed inlet opening.
main cylindrical body extending over a length of 60 mm Earlier reports also indicate that the results using a 3-D
inside and 30 mm above the top closed surface. A rectangu- model are better matching with the experimental data com-
lar 20 10 mm tangential feed inlet opening is connected to pared to the results with axisymmetric geometry
the cylindrical surface at a height of 15 mm below the top [www.psl.bc.ca/downloads/presentations/cyclone/cyclone.
surface. Studies were carried out on this geometry by html]. Thus 3-D computational model was selected for
changing the bottom diameter of the frustum at 10, 15, the study. Five dierent mesh densities between 50,000
20 and 25 mm at a constant cone angle of 10. and 200,000 computational elements were attempted to
62 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071

optimize the mesh density for a reasonable solution. A mental steps where in general a preset value of convergence
mesh density of 150,000 computational cells was opti- criteria 1 106 was achieved.
mized. Enormous computation time had become a con-
straint though the results at higher mesh densities were 4. Experimental
found slightly better. An unstructured grid based on tetra-
hedral cells with T-grid meshing scheme pattern was used The experimental setup consisted a slurry tank of 200
while discretizing the geometry. liters capacity mounted on a stable platform. A centrifugal
pump with 3-phase, 5.5 kW motor was connected to the
2.3. Boundary and initial conditions slurry tank at the bottom. Feed slurry consisting of clay
material at 10% solids density was pumped into the cyclone
The rectangular cyclone feed inlet face was dened as through the pipeline connected to the pump. The other end
pressure inlet. The overow and underow outlets faces of the pipeline was connected to the inlet opening of a
were designated as pressure outlets. Radial pressure distri- 76 mm diameter hydrocyclone. The pressure drop inside
bution from the cyclone axis to the edges is opted at both the cyclone was maintained at required level with the help
the pressure outlets. Backow direction was specied as of by-pass arrangement actuated through a control valve
normal to the boundary zones and backow turbulence on the pipeline. The pressure drop in the cyclone was mea-
intensity was assigned a value of 10%. The primary water sured with the help of a diaphragm type pressure gauge t-
phase with a density of 998.2 kg/m3and a viscosity value ted near the feed inlet. The hydrocyclone was positioned
of 1.003 106 kg/m s was allowed to enter the cyclone upright above the slurry tank.
at inlet pressure values of 55 kPa and 83 kPa. The outlet The experimental program was designed to achieve a
at the underow was varied at 10 mm, 15 mm, 20 mm wide range of water splits into the overow and underow
and 25 mm at each test run and water ow behavior was products suitably selecting the spigot opening and feed
simulated. Inert solid particles with density of 2650 kg/m3 inlet pressures. Hydrocyclone main body was xed to the
and of dierent sizes varying between 1 and 25 lm were test-rig. Required opening spigot as per the experimental
injected from the feed inlet boundary zone along the sur- design was tted to the hydrocyclone bottom. Initially, dis-
face. The particles entering any of the pressure outlet zones tribution studies were carried out by pumping water into
were assigned to escape the vessel. the cyclone at dierent spigot openings and feed pressures.
Later, solids consistency of 10% by weight was maintained
3. Simulation in the slurry tank by mixing measured amount of clay and
water. Tests were carried out on the solid slurry by pump-
The simulations carried out on hydrocyclone were ing into the cyclone body at desired feed inlet pressure and
assumed to be operating without air core. Cartesian coor- at dierent spigot openings. Samples of overow and
dinate system was used for numerical simulations. Flow underow products were collected simultaneously for a
simulation was carried out using a 3-D double precision, specic time in dierent containers. The underow and
steady state, and segregated solver. In this method, the gov- overow products collected were ltered, dried and
erning Navier Stokes equations (Annexure) are solved weighed. Particle size distribution of representative samples
sequentially using iterative methods till the dened values of the dried products was analyzed using in Malvern laser
of convergence are met. Initially, the properties of the particle size analyzer. Distribution points based on report
water were used along with the pressure and face mass of each size fraction in the feed to the underow product
uxes for calculating the momentum equations and further were generated.
update the velocity eld. PRESTO (Pressure staggered
option), a pressure interpolation scheme which was 5. Results and discussion
reported to be useful for predicting highly swirling ow
characteristics prevailing inside the cyclone body (Fluent The simulated results of cyclone throughput (mass ow
Europe Ltd., 2002) was adopted. For turbulence calcula- rate through overow + mass ow rate through under-
tions ke, ke RNG, Reynolds stress model (RSM) (models ow), and on the water-split (percent report of total water
presented in annexure) were independently used to evaluate entering the cyclone) into overow product were used for
the comparative simulation results. To obtain the pressure validating the predictions at dierent test runs with the
eld inside the system, SIMPLE (Semi-Implicit Pressure experimental values. The simulation studies were carried
Linked Equations) algorithm scheme, which uses a combi- out using standard ke, ke RNG and RSM turbulence
nation of continuity and momentum equations to derive an models, at three spigot openings, i.e. 10 mm, 15 mm,
equation for pressure was used. Interpolation of eld vari- 20 mm and 25 mm. All these simulations were carried out
ables from cell centers to faces of the control volumes was at a constant feed inlet pressure of 83 kPa. Initial discus-
opted with higher-order Quadratic Upwind Interpolation sions cover comparison of results obtained using the three
(QUICK) spatial discretisation scheme as it was reported turbulence models specically validating with the experi-
to be useful for swirling ows (Fluent Europe Ltd., mental data. Having identied the superior predictions
2002). Simulations were carried out for about 10,000 incre- with RSM model, ke and ke (RNG), further simulation
 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071 63

works were carried out using RSM turbulence model. Sim- Turbulence models
ulated general ow patterns in the cyclone in terms of 100
Experimental
velocity and static pressures are discussed. Further, the dis-
90 RSM
cussions cover comparison of simulated and experimental RNG

Report to overflow (%)

solid particles distribution curves. Finally the eects of 80 K-e
variables on axial and tangential velocities at 55 kPa feed
inlet pressure and on static pressures are discussed while 70
corroborating the experimental results.
60

5.1. Validation results

50

The results obtained on the water throughput and water 40

split under similar experimental and simulated conditions
using dierent turbulence models is compared. The results 30
5 10 15 20 25 30
are discussed in the following sections. Spigot opening (mm)

Fig. 3. Experimental and simulated values of water split using dierent

5.1.1. Water throughput
turbulence models.
The results on the water throughput obtained during the
experimental work at dierent spigot openings are com-
pared with the simulated values obtained using dierent the entire range of spigot openings. It can also be observed
turbulence models in Fig. 2. The gure indicates that all from the gure that among the results obtained with RSM
the three models predicted an increase in throughput with model, the prediction at widest spigot opening (25 mm) is
increase in the spigot, which can also be observed from slightly higher than the experimental values where as at
the experimental results. The results obtained with the narrowest spigot opening the simulation prediction is lower
RSM model are closer to the experimental results followed than the experimental results. However, at intermediate
by the results obtained using ke RNG turbulence model in spigot openings the data points indicated good match with
turn followed by the results using standard ke model. The the experimental data. The dotted line enclosed the region,
simulated results match with the experimental results with which corresponds to water split range of 4587% in the
marginal error about 1520% when ke model was used overow product. In this region the simulated data agrees
1015% with ke RNG and 48% when RSM turbulence well with the experimental results with a reasonable accu-
model was used. racy. The general range of water splits in a normal operat-
ing hydrocyclone will be within this region of good
5.1.2. Water split prediction limits.
Water split in hydrocyclone is the percentage of the Having observed the better simulation results generated
water reporting to over ow. The experimental and simu- using RSM turbulence model for predicting the hydrocy-
lated results of water split using dierent turbulence models clone performance, the general ow patterns in terms of
are presented in Fig. 3. Among the methods of simulations, velocities and static pressures at dierent positions in verti-
the method using RSM model has better predictions over cal and horizontal planes, the simulation results obtained
using this model are discussed.

1.7
Experimental
5.2. General ow behavior
RSM
1.6
RNG The general ow behavior in terms of velocity and pres-
Water throughput (kg/s)

K-e sure distributions at dierent planes inside the system are

1.5
discussed with a comparison of the published ow proles
1.4 of hydrocyclones (Williams et al., 1995; Dyakowski et al.,
1999; Grady et al., 2003; Slack et al., 2003; Nowakowski
1.3 and Dyakowski, 2003; Nowakowski et al., 2004; Cullivan
et al., 2003, 2004).
1.2

1.1
5.2.1. Axial velocity
The vertical velocity contours obtained in radial planes
1.0 at dierent axial heights are presented in Fig. 4A and in
5 10 15 20 25 30 one axial plane corresponding to central radial position
Spigot opening (mm) are presented in Fig. 4B. It can be observed from the g-
Fig. 2. Experimental and simulated data on water throughput using ures that two kinds of vertical ows one traveling upward
dierent turbulence models. indicated by values of positive axial velocity and the other
64 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071

traveling downward indicated by negative axial velocity body and cyclone wall than friction between individual
exist. It can be observed from Fig. 4A that concentric lay- water layers, which are away from the wall.
ers of constant axial velocities exist at all the heights in the The axial velocity proles at dierent radial distances
cyclone body. Further, it can be observed from Fig. 4B that and at dierent axial heights in the cyclone body are inte-
with increasing radial distance from the cyclone axis, the grated and presented in Fig. 5. It can be observed from
positive values of axial velocity decreases till it reaches zero the gure that a peak positive axial velocity of 2.32 m/s is
at some radial distance away from the cyclone axis. Nega- immediately below the bottom of the vortex nder and a
tive axial velocity begins beyond the radial distance of zero minimum value of positive axial velocity 0.25 m/s is at a
vertical velocity, and increases with increase in radial dis- vertical distance of 275 mm from the top of the hydrocy-
tance. However, as the radial distance approaches cyclone clone. The gure also indicates that at an axial height of
wall, the values of negative vertical velocity again decrease. 325 mm, there is no positive vertical velocity at all the
This can be due to higher friction between layers of water radial distances indicating no further classication of water
in this region. Further the gure indicates a negative axial
velocity about 0.23 m/s in magnitude near the walls of
cyclone body at an axial distance of 25 mm from the top.
At a vertical distance of 325 mm from the top, a maximum
value of 0.75 m/s can be observed.
The observations on the simulated results indicate that
water split takes places to a maximum extent early in the
cylindrical section below the vortex nder opening. At pro-
gressively lower heights, smaller splits are allowed into the
overow. Near the spigot opening the water split will be
insignicant.

5.2.2. Tangential velocity

The centrifugal force eld necessary for classication
inside the system will be generated by the tangential-veloc-
ity component of the continuous phase. The simulated

Radial distance in mm

Cyclone height in mm

15 mm
-40 -30 -20 -10 0 10 20 30 40
Base line

25 mm
50 mm
Fig. 4A. Contours of axial velocity of water (m/s) in axial plane.
75 mm

125 mm

175 mm

225 mm

2.5 275 mm
2.0
Each division = 2.75m/s 1.5
1.0
0.5
0.0 325 mm
-0.5
-1.0
-1.5
-2.0
-2.5

Fig. 5. Axial velocity distribution of water at dierent vertical positions

Fig. 4B. Contours of axial velocity of water (m/s) in radial planes. inside the cyclone.
 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071 65

Table 1
Radial distance in mm Design details of hydrocyclone used for simulation
Dimensions (mm) Cyclone 1 Cyclone 2 Cyclone 3 Cyclone 4
CD 76 76 76 76
Cyclone height in mm
CyL 85 85 85 85
15 mm VFD 25 25 25 25
-40 -30 -20 -10 0 10 20 30 40 VFL 90 90 90 90
Base line
FI (l w) 20 10 20 10 20 10 20 10
25 mm CA* 10 10 10 10
SPD 10 15 20 25
50 mm
CD: cyclone diameter; CyL: cylindrical length; VFD: vortex nder
75 mm diameter; VFL: vortex nder length; FI: feed inlet dimensions
(length width); CA*: cone angle in degrees; SPD: Spigot diameter.

125 mm
The partition curve represents the report of any size mate-
rial in the feed to the underow product (this number is
175 mm also referred as distribution point or partition number).
Particle distribution behaviour inside the cyclone is sim-
ulated using Lagrangian particle tracking approach upon
225 mm the Eulerian continuous phase predictions. Particle-track-
8.0
7.0
6.0
ing approach, which is reported to be useful technique
5.0
4.0 275 mm
3.0
2.0 Each division = 4.35m/s 100
1.0
0.0 325 mm
Percentage report in UF

75
Fig. 6. Tangential velocity distribution of water at dierent vertical
Sim. 20mm spi.
positions inside the cyclone.
50 Exp. 20mm spi.
Sim. 15mm spi.
Exp. 15mm spi.
results of tangential velocity at dierent vertical heights are
plotted in Fig. 6. It can be observed from the gure that ini- 25 Sim. 10mm spi.
Exp. 10mm spi.
tially with increasing radial distance from the axis the tan-
gential velocity increases. The tangential velocity values
0
after achieving a maximum value at some radial position 0 5 10 15 20 25
decrease with a further increase in the radius approaching Particle size (microns)
towards the walls. The observations made are similar to
Fig. 7. Experimental and simulated partition numbers at dierent spigot
the reports of earlier workers in the literature (Slack and
openings and at 55 K Pascal feed pressure.
Wraith, 1997). The proles of tangential velocity remain
similar at dierent axial heights in the cyclone body. Max-
imum values of tangential velocities are observed in the
100
cylindrical portion. It can also be noted that the maximum
values of tangential velocity decrease with decrease in axial
Percentage report in UF

height. For example, at an axial height of 125 mm, the 75

maximum tangential velocity observed is 5.70 m/s and at
an axial height of 275 mm the maximum value is 4.99 m/ Sim. 20mm spi.
s. The observation infers that at lower cyclone heights Exp. 20mm spi
50
towards the spigot opening relatively lower centrifugal Sim.15mm spi.
elds of force are generated. This phenomenon in hydrocy- Exp. 15mm spi
clone re-orients the entrapped ne size fractions for pro- 25 Sim. 10mm spi.
gressive report into the vertical axial ow while treating Exp. 10mm spi
solid materials (Cullivan et al., 2004).
0
5.3. Particle separation behavior 0 5 10 15 20 25
Particle size (microns)
The performance of hydrocyclone is represented in Fig. 8. Experimental and simulated partition numbers at dierent spigot
terms of distribution or partition curves (Wills, 1997). openings and at 83 K Pascal feed pressure.
66 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071

Table 2 sample runs were carried out and the report of particles
Experimental and simulated cyclone cut size values at dierent test runs into the overow and underow outlet streams were aver-
Test FP SPD Throughput Split Cut size d50 aged. The partition numbers were obtained for each size
run (kPa) (mm) (kg/s) of (kg/s) (lm) and geometry in consideration (Table 1).
Exp. Sim. The partition curves generated at 10 mm, 15 mm, and
1 55 10 0.932 94.8 11.4 14.0 20 mm spigot openings at a feed pressure of 55 K Pascal
2 55 15 0.947 79.2 10.8 12.0 using dierent partition numbers, are presented in Fig. 7.
3 55 20 1.0333 55.9 5.8 6.3 Also presented in the gure are the experimental values
4 83 10 1.161 95.7 9.5 10.0
5 83 15 1.174 81.5 8.3 8.9
obtained under similar conditions. It can be observed that
6 83 20 1.306 58.7 4.5 4.9 the simulated data points are reasonably matching with the
FP: feed pressure; SPD: Spigot diameter; Exp.: experimental; Sim.:
experimental values at all the test conditions. Similar obser-
simulated. vations can also be made at other set of test conditions car-
ried out at 83 K Pascal (Fig. 8). Further, the cut size of the
cyclone (d50), (the size of the particle having a partition
for simulation purposes when particle concentration is
value of 50%) of the simulated and experimental conditions
below 10% by weight (Stovin and Saul, 1998), is used for
are presented in Table 2. The data indicates that the pre-
achieving the distribution behavior of dierent sizes. Parti-
dicted values are closely matching with the experimental
cles are injected through the feed inlet into the cyclone
results.
body after achieving the convergence. Sample group of
1000 particles of a dened sizes within a selected range were
injected into the cyclone body along the face of the inlet 5.4. Parametric studies
opening. The density of the material is maintained constant
at 2650 kg/m3which corresponds to the density of clay The eect of important parameters like spigot opening
material used in the experimental studies. Each time 10 and feed inlet pressure on the simulated results of axial

2.5 2.5
12psi, 10spd 12psi, 10spd
Axial Height 12psi, 15spd
Axial Height 12psi, 15spd
2 2
75mm 12psi, 20spd 125mm 12psi, 20spd

1.5 1.5

1 1

0.5 0.5

0 0
-40 -30 -20 -10 0 10 20 30 40 -40 -30 -20 -10 0 10 20 30 40
-0.5 -0.5

-1 -1

-1.5 -1.5

2.5 2.5
12psi, 10spd 12psi, 10spd
Axial Height 12psi, 15spd
Axial Height
12psi, 15spd
175mm 2 12psi, 20spd 225mm 2 12psi, 20spd

1.5 1.5

1 1

0.5 0.5

0 0
-40 -30 -20 -10 0 10 20 30 40 -40 -30 -20 -10 0 10 20 30 40

-0.5 -0.5

-1 -1

Fig. 9. Eect of spigot openings on the axial velocity distribution of water at dierent radial points (feed pressure 83 K Pascal).
 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071 67

and tangential velocities, static pressures are discussed as 0.27 m/s from 0.63 m/s to 0.90 m/s. Similar observa-
follows. tion can be made at all the axial heights. Due to increase
in constriction for water ow at smaller spigot opening,
5.4.1. Eect of spigot opening the kinetic energy within the water body is dissipated for
The eect of change in spigot opening from 10 to increase in velocities in the axially upward direction and
20 mm on the axial velocities at 83 kPa feed inlet feed further increase the water split to overow. This can be
pressure, is presented in Fig. 9. It can be observed from well observed with the experimental values presented in
the gure that the maximum values of positive axial veloc- Fig. 4. This increase in mass ow through the vortex nder
ities indicating an upward vertical ow around the cyclone increases the chances of relatively coarser particles report
axis is higher at lower spigot opening at all the axial posi- into overow, which is well validated, by an increase in
tions. On the other hand, the negative axial velocities indi- the cut size of the cyclone at lower spigot opening (Table
cating a downward ow near the cyclone walls are smaller 2).
at smaller spigot opening. It may also be observed that the
magnitude of increase in upward vertical velocity is more 5.4.2. Eect of feed inlet pressure
compared to the magnitude of decrease in downward The simulated data on the axial velocities at dierent
velocity. For instance, at 125 mm axial height, a decrease radial distances and axial heights along the central cyclone
in the spigot opening from 20 mm to 10 mm has increased plane (results obtained at 10 spigot opening) are presented
the positive axial velocity by 0.47 m/s from a value of in Fig. 10. It can be observed from the gure that an
1.37 m/s to 1.84 m/s. A similar decrease in the spigot increase in feed inlet pressure from 55 kPa to 83 kPa
opening has decreased the negative axial velocity by increases the rising axial velocity near the central region

3 3
8 psi, 10spd 8psi, 10spd
10psi, 10spd 10psi, 10spd
Axial Height 2.5 12psi, 10spd
Axial Height 2.5 12psi, 10spd
75mm 125mm
2 2

1.5 1.5

1 1

0.5 0.5

0 0
-40 -30 -20 -10 0 10 20 30 40 -40 -30 -20 -10 0 10 20 30 40
-0.5 -0.5

-1 -1

3 3
8psi, 10spd 8psi, 10spd
Axial Height 10psi, 10spd Axial Height 10psi, 10spd
2.5 2.5
175mm 12psi, 10spd 225mm 12psi, 10spd

2 2

1.5 1.5

1 1

0.5 0.5

0 0
-40 -30 -20 -10 0 10 20 30 40 -40 -30 -20 -10 0 10 20 30 40
-0.5 -0.5

-1 -1

Fig. 10. Eect of feed inlet pressure on the axial velocity distribution of water at dierent radial points (10 mm spigot opening).
68 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071

around the axis and also increases the maximum values of and minimum near the radial position near to the axis of
negative axial velocity at radial distances near the walls. An the cyclone. The observation indicates that higher-pressure
increase in upward and downward axial ow rates at dierential between radial positions near the walls and
higher feed inlet pressures increases the total mass ow of towards the cyclone center is generated at higher feed inlet
water through the cyclone. This is in agreement with the pressures. This increase in dierential feed pressure
experimental results (Table 2). increases the inward radial ow of water eventually report-
Further, it can be observed that the increase in the mag- ing to the overow product.
nitude of upward vertical velocity is more compared to the The eect of feed inlet pressure on the tangential veloc-
increase in downward velocity magnitude. For instance, at ity distribution at dierent radial positions and at dierent
175 mm axial height, an increase in the feed pressure from axial positions is presented in Fig. 12. The gure indicates
55 kPa to 83 kPa increased the positive axial velocity by that an increase in feed pressure increases the tangential
0.22 m/s from a value of 1.03 m/s to 1.25 m/s. A similar velocity at all the radial positions being maximum near
increase in the feed inlet pressure has increased the negative the walls to minimum near the cyclone center. At any point
axial velocity only by 0.08 m/s from 0.53 m/s to 0.61 m/ on the radial position, an increase in tangential velocity will
s. The observation indicates an increased ow in the verti- increase the centrifugal forces on suspended solid particles
cal direction and increased split to overow. This observa- and hence relatively ner particles also get centrifuged
tion is found similar to the experimental split values towards the cyclone walls to report into the underow.
presented in Table 2. The remnant ner fractions at shorter radial distances
The eect of feed inlet pressure on the static pressure from the cyclone axis will be aected by the drag to report
values along the radial distances at dierent axial heights into the overow. Thus at higher feed pressures, a decrease
are presented in Fig. 11. It can be observed from the gure in solids cut size is expected. The results obtained through
that an increase in the feed inlet pressure increases the sta- experiments are matching with the expected increase in the
tic pressure values dominantly near the walls of the cyclone cut size at higher feed pressures (Table 2).

60000 60000
Axial Height Axial Height
50000 75mm 50000 125mm

40000
40000

30000
30000

20000
20000
8psi
8psi
10000 10psi
10psi
12psi 10000 12psi
0
-40 -30 -20 -10 0 10 20 30 40
0
-10000 -40 -30 -20 -10 0 10 20 30 40

60000 60000

Axial Height Axial Height

50000 175mm 50000 225mm

40000 40000

30000 30000

20000 20000
8psi 8psi
10psi 10psi
10000 12psi 10000 12psi

0 0
-40 -30 -20 -10 0 10 20 30 40 -40 -30 -20 -10 0 10 20 30 40

Fig. 11. Eect of feed inlet pressure on the pressure distribution at dierent radial points (10 mm spigot opening).
 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071 69

6 6

5
5

4
4

3
3

2
Axial Height Axial Height 1
8psi
8psi 10psi
75mm 10psi
125mm
1 12psi
12psi 0
-40 -30 -20 -10 0 10 20 30 40
0
-1
-40 -30 -20 -10 0 10 20 30 40

6 6

5 5

4 4

3 3

2 2

8psi 1 8psi
1
Axial Height 10psi Axial Height 10psi
12psi
175mm 12psi 225mm
0 0
-40 -30 -20 -10 0 10 20 30 40 -40 -30 -20 -10 0 10 20 30 40

Fig. 12. Eect of feed inlet pressure on the tangential velocity distribution of water at dierent radial points (10 mm spigot opening).

6. Summary for ow predictions. Under steady state conditions, the

equations for mass and momentum in a general form are
A CFD simulation and supporting validation study on as follows:
76 mm hydrocyclone have demonstrated the applicability
r  qm 0; 1
in the analysis of the water throughput, splits and cyclone
cut sizes for various test conditions. Among the turbulence r  q~m rp r  ~
m~ s q~
g; 2
models studied, the simulation results adopting RSM
where p is a static pressure, q~ g is the gravitational body
model is found to have better predictions with experimen-
force ~
s is the stress tensor given by
tal results. The solid particle distribution simulation  
through discrete phase modeling technique was found 2
s leffective rm  r:q~
~ m2 ; 3
matching with the experimental results at 10% solids by 3
weight with reasonable accuracy. Parametric results have where leective = l + lt.
indicated that the feed inlet pressure has major inuence
on mass ow through the cyclone and spigot opening on The standard ke model
percent split into the overow. The study has indicated that
the methodology adopted can be useful for predicting the The standard ke model is a semi-empirical model based
performance of hydro- cyclones. on model transport equations for the turbulent kinetic
energy (k) and its dissipation rate (e), and are given by
Appendix   
o o o li ok
qk qkui l Gk  qe; 4
The model used for ow simulation solves the conserva- ot oxi oxj rk oxj
  
tion equations for mass (or continuity) and momentum. o o o l ok e e2
The turbulence in the system is solved through transport qe qeui l t C 1e Gk  C 2eq ;
ot oxi oxj re oxj k k
equations. NavierStokes equations for incompressible
ows along with appropriate turbulence model are adopted 5
70 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071

ouj F ij rotation production 2qXk u0j u0m eikm u0i u0m ejkm ; 12
Gk qu0i u0j : 6
oxi o h 0 0 0 i
DTi0 ij turbulent diffusion  qui uj uk pdkj u0i dik u0j ;
oxk
In these equations, Gk represents the generation of turbu-
lent kinetic energy due to the mean velocity gradients, 13
!
C1e, C2e and C3e are constants. rk and re are the turbulent 0 0
oui ouj
Prandtl numbers for k and e, respectively. The eddy or /ij pressure strain ; 14
oxj oxi
turbulent viscosity, gt dened in Eqs. (2) and (3) can be
computed by combining k and e as follows: ou0 ou0j
eij dissipation 2l i : 15
oxk oxk
k2
lt qC g ; 7 More details on the underlying equations in this method
e
are provided in the literature (Fluent Europe Ltd., 2002)
where Cg is a constant.
The model constants C1e, C2e, Cg, rk and re were Equations of motion for particles
assumed to have the following values: C1e = 1.44,
C2e = 1.92, Cg = 0.09, rk = 1.0, re = 1.3. In hydrocyclone systems operating at dilute concentra-
tion of solids (typically below a value of 10% solids by
The RNG ke model weight), discrete phase modeling technique can be adopted
for identifying the particle positions inside the system. The
The renormalization group (RNG) ke model is similar particles introduced into the system are simulated in a
in form to the standard ke model but includes an addi- Lagrangian frame of reference using the denitions for
tional terms for dissipation rate e development that signif- material parameters like particle size, specic gravity, and
icantly improve the accuracy, especially for rapidly strained initial position etc. assuming solids spherical particles.
ows. The eect of swirl on turbulence is included in the The trajectory of the discrete phase particle is obtained
RNG model, enhancing accuracy for swirling ows. by integrating the force balance on the particle. This force
The RNG ke model has a similar form to the standard balance equates the particle inertia with the other forces
ke model: acting on the particle, and can be written (for the x direc-
o o o ok tion in Cartesian coordinates) as
qk qkui ak leff Gk
ot oxi oxj oxj
dup =dt F D u  up gx qp  ql =qp F x ; 16
Gb  qe  Y M S K 8
where FD(u  up) is the drag force per unit particle mass
and
and
o o o oe e
qk qeui ae leff C 1e GK C 3e Gb F D 18l=qp  D2p  C D Re=24: 17
ot oxi oxj oxj k
e2 Here u is the uid phase velocity, up is the particle veloc-
 C 2e q  Re Se: 9 ity,l is the molecular viscosity of the uid, ql is the uid
k
density, qp is the density of the particle and Dp is the par-
In these equations, Gk represents the generation of turbu- ticle diameter, Re is the relative Reynolds number which
lence kinetic energy due to the mean velocity gradients. is dened as
Gb is the generation of turbulence kinetic energy due to
buoyancy, YM represents the contribution of the uctuat- Re q  Dp up  u=l: 18
ing dilatation in compressible turbulence to the overall dis-
sipation rate. The quantities ak and ae are the inverse References
eective Prandtl numbers for k and e, respectively. Sk
and Se are user-dened source terms. Cullivan, J.C., Williams, R.A., Cross, C.R., 2003. Understanding the
hydrocyclone separator through computational uid dynamics. Trans-
actions of the IchemE, Part A, Chemical Engineering, Research and
Reynolds stress model (RSM) Design 81 (A4), 455466.
Cullivan, J.C., Williams, R.A., Dyakowski, T., Cross, C.R., 2004. New
For steady-state, the Reynolds-Stress Model (RSM) understanding of a hydrocyclone ow eld and separation mechanism
adopted, uses the following transport equations for the from computational uid dynamics. Minerals Engineering 17, 651
660.
Reynolds stresses:
Devulapalli, B., Rajamani, R.K., 1996. A comprehensive CFD model for
o o particle- size classication in industrial cyclones. In: Hydrocyclones96.
qu0i u0j qu0k u0i u0j P ij F ij DTij /ij  eij ; 10 Mechanical Engineering Publications, Ltd., London, UK, pp. 83
ot oxk
  104.
ouj oui Dyakowski, T., Williams, R.A., 1993. Measurement of particle velocity
P ij stress production q u0i u0k u0j u0k ; 11 distribution in a vertical channel. Powder Technology 77 (2), 135
oxk oxk
142.
 K.U. Bhaskar et al. / Minerals Engineering 20 (2007) 6071 71

Dyakowski, T., Hornung, G., Williams, R.A., 1994. Simulation of non- multi-continuum model. Chemical Engineering Journal 80 (13), 275
Newtonian ow in a hydrocyclone. Chemical Engineering Research 282.
and Design 72 (A4), 513520. Nowakowski, A.F., Dyakowski, T., 2003. Investigation of swirling ow
Dyakowski, T., Nowakowski, A.F., Kraipech, W., Williams, R.A., 1999. structure in hydrocyclones. Transactions of the Institution of Chemical
A three dimensional simulation of hydrocyclone behaviour. In: Second Engineers, Chemical Engineering Research and Design 81 (A), 862
International Conference on CFD in the Minerals and Process 873.
Industries. CSIRO, Melbourne, Australia, pp. 205210. Nowakowski, A.F., Cullivan, J.C., Williams, R.A., Dyakowski, T., 2004.
Fluent Europe Ltd., 2002. D30-1_TAR_P32_AC Cyclone separator. Application of CFD to modelling of the ow in hydrocyclones. Is this
Thematic area 3. Available from: <http//www.qnet-cfd.net>. a realizable option or still a research challenge? Minerals Engineering
Grady, S.A., Wesson, G.D., Abdullah, M.M., Kalu, E.E., 2002. Predic- 17, 661669.
tion of ow eld in 10-mm hydrocyclone using computational uid Pericleous, K.A., Rhodes, N., 1986. The hydrocyclone classier a
dynamics. Fluid and Particle Separations Journal 14, 111. numerical approach. International Journal of Mineral Processing 17
Grady, S.A., Wesson, G.D., Abdullah, M.M., Kalu, E.E., 2003. Predic- (12), 2343.
tion of 10 mm hydrocyclone separation eciency using computational Plitt, L.R., 1976. A mathematical model of hydrocyclone classier. CIM
uid dynamics, Research article, Dept. of Chemical and Civil Bull, 114.
Engineering, FAMU-FSU College of Engineering, Tallahassee, and Rajamani, R.K., Milin, L., 1992. Fluid ow model of the hydrocyclone for
USA. concentrated slurry classication. Hydrocyclones: Analysis and Appli-
Griths, W.D., Boysan, F., 1996. Computational uid dynamics (CFD) cation: 4th Intentional Conference, vol. 12, pp. 95108.
and empirical modelling of the performance of a number of cyclone Schuetz, S., Mayor, G., Bierdel, M., Piesche, M., 2004. Investigations on
samples. Journal of Aerosol Science 27 (2), 281304. the ow and separation behavior of hydrocyclones using computa-
Hargreaves, J.H., Silvester, R.S., 1990. CFD analysis of deoiling hydro- tional uid dynamics. International Journal of Mineral Processing 73
cyclone performance. Transactions Institute Chemical Engineering 68, (24), 229237.
365385. Slack, M.D., Wraith, A.E., 1997. Modelling the Velocity Distribution in a
Hsieh, K.T., Rajamani, K., 1988. Phenomenological model of the Hydrocyclone. In: 4th International Colloquium on Process Simula-
hydrocyclone: Model development and verication for single-phase tion, pp. 6583.
ow. International Journal of Mineral Processing 22 (14), 223 Slack, M.D., Boysan, 1998. Advances in cyclone modelling using
237. unstructured grids. In: Scandinavian FLUENT UGM Gothenburg.
Kelsall, D.F., 1952. A study of the motion of solid particles in a hydraulic Stovin, V.R., Saul, A.J., 1998. A computational uid dynamics (CFD)
cyclone. Transactions, Institute Chemical Engineers 30, 87108. particle tracking approach to eciency prediction. Water Science and
Lynch, A.J., Rao, T.C., 1975. Modelling and scale up of hydrocyclone Technology 37 (1), 285293.
classiers. 11th International Mineral Processing Congress, Cagliari, Slack, M.D., Prasad, R., Baker, A., Boysan, F., 2000. Advances in cyclone
pp. 245269. modelling using unstructured grids. Transactions Institute Chemical
Malhotra, A., Branion, R.M.R., Hauptman, E.G., 1994. Modelling the Engineers 78, 10981104.
ow in a hydrocyclone. Canadian Journal of Chemical Engineering 72, Slack, M.D., Del Porte, S., Engelman, M.S., 2003. Designing automated
953960. computational uid dynamics modeling tools for hydrocyclone design.
Ma, L., Ingham, D.B., Wen, X., 2000. Numerical modelling of the uid Minerals Engineering 17, 705711.
and particle penetration through small sampling cyclones. Journal of Suasnabar, D.J., Fletcher, C.A.J., 1999. A CFD model for dense medium
Aerosol Science 31 (9), 10971119. cyclones. In: Second International Conference on CFD in the Minerals
Monredon, T.C., Hsieh, K.T., Rajamani, R.K., 1992. Fluid ow model of and Process Industries. CSIRO.
the hydrocyclone: an investigation of device dimensions. International Williams, R.A., Ilyas, O.M., Dyakowski, T., 1995. Air core imaging in
Journal Mineral Processing 35 (12), 6583. cyclonic coal separators using electrical resistance tomography. Coal
Nowakowski, A.F., Kraipech, W., Williams, R.A., Dyakowski, T., 2000. preparation 15, 149163.
The hydrodynamics of a hydrocyclone based on a three-dimensional Wills, B.A., 1997. Mineral Processing Technology, sixth ed., pp. 232234.