You are on page 1of 13

chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362

Contents lists available at ScienceDirect

Chemical Engineering Research and Design

journal homepage: www.elsevier.com/locate/cherd

Mixing and gas dispersion in mineral flotation cells

G.M. Evans a,∗ , E. Doroodchi a , G.L. Lane b , P.T.L. Koh b , M.P. Schwarz b
a School of Engineering, University of Newcastle, Callaghan, NSW 2308, Australia
b CSIRO Minerals, Box 312, Clayton South, Vic 3169, Australia

a b s t r a c t

Mineral flotation in mechanically agitated vessels (cells) involves complex interaction between bubbles, particles
and the liquid phase. Ideally, just enough power input from the impeller is needed to so that the frequency of
particle–bubble collision and attachment is maximised, while at the same time detachment events are minimised.
This paper firstly investigated how the slip velocity of 2–10 mm diameter bubbles, a size commonly encountered
in flotation devices, was influenced by turbulence intensity. The measurements confirmed the earlier correlation
by [Lane, G.L., 2005, Numerical modelling of gas–liquid flow in stirred tanks, Ph.D. Thesis, University of Newcastle,
Australia], which was then inputted into a computational fluid dynamic model to describe the gas dispersion in a
mechanically agitated tank. The model provided turbulence intensity values that were then coupled with both slip
velocity and critical Weber number models to generate both bubble size and gas holdup profiles for the entire vessel.
Moreover, a simple equation was introduced to allow prediction of cavity formation behind the rotating impeller
blades, which is a common occurrence in most flotation cells they normally operate at high gas loadings. This
inclusion allowed the model to predict power reduction resulting from the presence of the cavities. Finally, extension
of the computational model to include flotation hydrodynamics, such as probabilities of collision, adhesion and
stabilisation of the particles at the bubble surface, is also described. The model is able to compute net attachment
rates, and hence the particle flux entering the froth recovery phase, as a function of bubble and particle diameter,
gas flowrate and power input.
© 2008 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.

Keywords: Flotation; Computational fluid dynamics; Turbulence; Slip velocity; Gas dispersion

1. Introduction rises and forms a froth on the top of the cell. Eventually, the
water within the froth is allowed to drain back into the pulp,
The extraction of valuable minerals from ores involves the pro- leaving behind a concentrated solid product.
cessing of vast tonnages of material. For example, world pro- The objective of the flotation process is to concentrate the
duction of coal in 2006 was over 6000 Mtonne (Hetherington et valuable minerals by discarding unwanted gangue, so that
al., 2008); with Australia alone producing over 400 Mtonnes of the volume of material entering the later stages of process-
treated product (Abare, 2007). Consequently, relatively minor ing (e.g. smelting) is as small as possible. The challenge then
improvements in efficacy and energy efficiency of any mineral is to maximise recovery of valuable minerals while at the
processing operation can yield substantial cost savings. One same time discard as much gangue as possible so as to max-
such mineral processing operation is flotation, which plays a imise the grade. While optimal preparation of the ore (e.g.
key role in beneficiation of most copper, lead, zinc and nickel grinding) is important, it is critical to control the complex
ores and coal, as well as being used in some circumstances for physical–chemical phenomena involving the interaction (e.g.
others such as iron ore. Mineral flotation involves the use of Dai et al., 2000; Nguyen and Schulze, 2004) between the mul-
bubbles to separate valuable minerals from unwanted gangue tiphase fluid mechanics and the surface chemistry of both
material. Separation is achieved by rendering the surface of the bubbles and particles. Additionally, there are engineering
the valuable mineral particle hydrophobic, usually with a col- constraints on design of flotation cells, such as the desirabil-
lector, so that it preferentially attaches to a bubble which then ity of minimising energy consumption and capital cost, and


Corresponding author. Tel.: +61 2 49215897; fax: +61 2 49216920.
E-mail address: Geoffrey.Evans@newcastle.edu.au (G.M. Evans).
Received 13 July 2008; Accepted 21 July 2008
0263-8762/$ – see front matter © 2008 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.cherd.2008.07.006
chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362 1351

Nomenclature
ε rate of dissipation of energy per unit mass
b defined as b = (1 + CA )/(2 /1 + CA ) (m2 s−3 )
Bp buoyancy reduction force (N m−3 ) r ratio of turbulent time scale to particle relax-
C constant in Eq. (3) ation time
CA added mass coefficient  ratio of length scales for bubble or particle tur-
CD drag coefficient bulence interaction
CD mean drag coefficient L laminar viscosity (Pa s)
CD,0 mean drag coefficient in quiescent liquid T turbulent viscosity (Pa s)
Ctd coefficient in Eq. (9)  ratio of bubble terminal velocity to fluctuating
C␤ constant in Eq. (17) velocity of liquid
d particle or bubble diameter (m)  density (kg m−3 )
deq average equivalent spherical diameter (m) 0 reference density (kg m−3 )
D12 isotropic turbulent diffusivity coefficient p bubble or particle relaxation time (s)
(m2 s−1 ) t
12 time scale of turbulence as seen by bubble or
f oscillation frequency (s−1 ) particle (s)
fc fractional surface coverage of particles on the ϕ source or sink term in number density equation
bubble (m−3 s−1 )
F force vector (N m−3 )
FD drag force (N m−3 ) Subscripts
g gravity vector (m s−2 ) 1 liquid or continuous phase
H initial distance between grids (m) 2 gas or dispersed phase
I unit tensor a attachment
k turbulent kinetic energy per unit mass (m2 s−2 ) b bubble
KD1 fitted constant in Eq. (2) br break-up
KD2 fitted constant in Eq. (2) co coalescence
K1 constant in Eq. (21) (m) d detachment
L11 integral length scale of turbulence (m) g gas
m1 exponent in Eq. (23) i phase number
m2 exponent in Eq. (23) l liquid
M grid opening size (m) p particle
M interphase force (N m−3 ) ph phase change
n bubble or particle number density (m−3 ) UG ungassed
nD exponent in Eq. (2)
p modified pressure (N m−2 )
q overall turbulence intensity (m s−1 )
optimising incorporation into overall beneficiation circuits.
S mass source term (kg m−3 s−1 )
As a result of all these considerations flotation cells have
SL stroke length (m)
increased in size over the last decade, with single cells now
St Stokes number, defined as St =  p /TL
reaching 300 m3 (Weber et al., 2005).
t time (s)
According to a recent review by Yianatos (2007), despite
T turbulent dispersion force vector (N m−3 )
great advances in terms of process knowledge over several
TL characteristic time of the large turbulent eddies
decades, industrial flotation cells are still not sufficiently well
(s)
understood to allow optimal design and scale-up. This conclu-
u average RMS turbulent velocity in X direction
sion is not surprising given the complex interactions occurring
(m s−1 )
within a flotation cell (O’Connor, 2005). For instance, in a
u fluctuating velocity vector (m s−1 )
mechanically agitated cell the impeller provides the energy
U Reynolds-averaged velocity vector (m s−1 )
input for bubble breakup, solids suspension (where the solids
US slip velocity (m s−1 )
loading is typically around 30 wt%), and mixing so that bub-
UT terminal velocity in stagnant fluid (m s−1 )
bles and particles can contact each other. The amount of
U0 turbulence fluctuation velocity (m s−1 )
energy input is very important because too little input will
v average RMS turbulent velocity in Y direction
cause some solids to settle as well as not producing sufficient
(m s−1 )
particle–bubble collisions to achieve the desired recovery. Con-
Vp particle volume (m3 )
versely, too much energy input will cause collected particles to
x position vector (m)
detach from the bubbles resulting in a loss in recovery. There
X horizontal coordinate
have been a number of studies examining the role of energy
Y vertical coordinate
input, or more specifically turbulence intensity, on the over-
all performance of the flotation process (e.g. Schubert, 1999;
Greek symbols
Bloom and Heindel, 2002; Sherrell, 2004). In spite of these
˛ volume fraction
efforts it is still a real challenge to design a flotation cell with
ˇ ratio of turbulent fluctuating velocity and par-
the right amount of energy input to achieve the desired level
ticle terminal velocity
of gas–solid dispersion, optimum bubble size, and maximum
number of particle–bubble collisions with minimal detach-
ment of any collected particles.
1352 chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362

Design of mechanically stirred tanks, and flotation cells expressing the effect of weak turbulence with intermediate
in particular, have traditionally been based on empirical cor- length scales on the motion of bubbles in terms of dimension-
relations, but these provide only global information such as less groups ˇ and :
power consumption or gas holdup, and are usually limited in
applicability to ‘standard’ tank configurations. Scale-up has US ˇ2
≈ 1 − 0.75 (1)
been based upon simple rules such as constant impeller tip UT 
speed or constant specific power, so that just one aspect of the
process remains constant. This leads to uncertainty regard- where US is the slip velocity. The relationship in Eq. (1)
ing other characteristics of the process at the larger scale indicates that a spherical bubble moving at high Reynolds
which are not kept constant. More recently, there have been numbers is on average slowed down by the turbulence.
a number of studies utilising computational fluid dynamics According to the authors the reduction in velocity is due to
(FCD) to model the behaviour inside flotation cells, not only (i) viscous forces which make the bubbles adapt their speed
for mineral processing operations (e.g. Tiitinen et al., 2003; to the fluid velocity fluctuations, and (ii) lift forces induced by
Koh and Schwarz, 2003, 2006, 2007, in press; Koh et al., 2000, the lateral motion which redirect them laterally to regions of
2003; Lichter et al., 2007) but also for water treatment (e.g. downflow.
Kostoglou et al., 2007) and pyrometallurgical operations (e.g. Lane et al. (2005) found that various numerical and exper-
Maniruzzaman and Makhlouf, 2002a,b; Kwon et al., 2006) as imental data (Spelt and Biesheuvel, 1997; Brucato et al., 1998;
well. The advantage with CFD is the potential to model any Poorte and Biesheuvel, 2002), collapsed onto a single curve if
tank design at any scale of operation, providing a wealth the ratio US /UT is plotted against the ratio ˇ/. This latter ratio
of detail such as internal velocities, shear rates, turbulence is equivalent to a ratio of time scales, or a Stokes number,
parameters, distributions of phases, bubble sizes, and resi- given as  p /TL , where TL is a characteristic time of the large
dence time distribution, as well as overall quantities such as turbulent eddies. It was further argued that the curve would
power consumption and mixing time (e.g. Lane, 2005; Deglon have a minimum at a Stokes number of 1 (the point of maxi-
and Meyer, 2006). mum interaction between bubbles and turbulence). Therefore,
The work by Lane (2005) in particular has highlighted that the effect of turbulence on slip velocity was assumed to follow
turbulence can significantly change the slip velocity of both the function:
the bubbles and particles, where under certain conditions US

p
nD  p

the slip velocity can be as low as half that for a quiescent = 1 − KD1 exp −KD2 (2)
UT TL TL
(non-turbulent) system. Ultimately, it is the velocities of the
bubbles and particles that have a major influence on the where nD , KD1 and KD2 are experimentally determined con-
particle–bubble collision process and eventual flotation col- stants. Eq. (2) was tested in the CFD model for gas dispersion,
lection rates and for this reason it is important to be able as described below, giving values of the parameters of KD1 = 1.4,
to relate the slip velocity to spatial and temporal variations nD = 0.7 and KD2 = 0.6.
of turbulence intensity throughout the cell. The aim of this The studies described above have focussed on how turbu-
paper, therefore, if firstly to provide some insight into the influ- lence influences the drag on bubbles with equivalent diameter
ence of turbulence on the motion of bubbles and particles, of equal or less than 1 mm. However, many multiphase sys-
particularly for bubbles in the 2–10 mm diameter range com- tems deal with intermediate size bubbles of typical diameters
monly encountered in mechanically agitated flotation cells. ranging between 1 and 10 mm. These bubbles are significantly
Secondly, a computational modelling approach is presented non-spherical and lie within the so-called ellipsoidal regime
which includes both turbulence effects on bubble and par- and undergo shape oscillation and follow a zigzag or helical
ticle motion as well as the formation of gas cavities formed path when rising in a stagnant fluid (Clift et al., 1978). These
behind the rotating impeller blades. Finally, the computational shape deformations affect the drag on the bubble, and it is not
modelling analysis is extended to the actual flotation process clear how the slip velocity will be affected when turbulence
where net attachment rates (leading to flotation) are presented is present. For this reason, experiments were conducted on
for a Denver flotation cell. bubbles, in the industrially significant size range of 2–10 mm
diameter, under carefully controlled turbulence conditions.
2. Influence of turbulence on bubble and
particle motion 2.1. Slip velocity experiments

There is a growing body of evidence which suggests that the An experimental investigation was carried out in an oscillating
free stream turbulence can significantly reduce the terminal turbulence generator in which a pair of grids was employed
velocity, UT , of bubbles and particles in a liquid by increas- to produce stationary near-isotropic turbulent flow fields. In
ing the average drag force. Various studies have indicated that such devices, the oscillating grids generate a system of wakes
the reduction of terminal velocity primarily depends on turbu- and jets that merge with each other to produce a sustained
lence structure and its intensity. In studies according to Spelt turbulence with zero mean flow (De Silva and Fernando, 1994).
and Biesheuvel (1997) and Poorte and Biesheuvel (2002), bub- The main advantage of this generator over its flow-through
ble motion in weak turbulence was characterised using two counterparts (e.g. a water tunnel) is that the bubble velocities
dimensionless groups, namely (i) ˇ = U0 /UT , where U0 is tur- can be measured more accurately in the absence of a net mean
bulence fluctuation velocity, and (ii)  = L11 / p ·UT , where L11 flow (Yang and Shy, 2003).
is the integral length scale of the turbulence and  p is the The experimental setup is shown in Fig. 1. The turbulence
bubble or particle relaxation time. The latter dimensionless generator consisted of a rectangular perspex tank with a width
group is often used to indicate how quickly the bubbles or of 300 mm containing a pair of vertically oriented grids of size
particles respond to the turbulent velocity fluctuations. Poorte 150 mm × 150 mm. The grids were made of 6-mm thick alu-
and Biesheuvel (2002) proposed a semi-analytical relationship minium sheets with a mesh size of 30 mm (i.e. the distance
chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362 1353

Fig. 1 – Experimental apparatus: (a) grids, (b) motor, (c) control system, (d) data acquisition system, and (e) camera.

between centres of two successive openings was 30 mm) and


an overall opening of 64%. The grids were connected to step-
per motors, through eccentric cams, which led the in-and-out
movement of the grids. The design details of the generator are
given in Doroodchi et al. (2008). Generally, the turbulent flow
field was altered by changing the system frequency, strokes
and/or the initial distance between the grids over a range of
5–8 Hz, 18–32 mm and 70–110 mm, respectively.
The experiments were carried out using bubbles with aver-
age equivalent spherical diameter, deq , of 2–10 mm. A novel
bubble generator, as shown in Fig. 2, was designed and fabri-
cated in-house. The bubble generator consisted of an injection
port, a collecting surface and a release arm. A micro-syringe
pump was used to inject air at a rate of 19.5 mL min−1 through
the injection port located at the lower section of the vessel
side-wall (i.e. 20 mm above the bottom of the vessel). The size
of the bubbles was controlled by adjusting the total volume
of injected air. The air was guided through a capillary tube of
Fig. 3 – Overall turbulence intensity as a function of the
1.5 mm in diameter to the centre of the vessel. The released
operational parameters (18 < SL < 32 mm, 70 < H < 110 mm,
air bubbles were then collected under a concave surface where
M = 30 mm and 5 < f < 6 Hz).
small bubbles of about 2.5 mm in size formed at the tip of the
capillary rapidly coalesced forming a larger bubble. The bub-
ble was then released through a ±90◦ rotation of the surface established for a few minutes by out-of-phase oscillation of
by pushing/pulling the release arm. the grids. Particle image velocimetry (PIV) was used to obtain
A systematic approach similar to that described in our the average RMS turbulent velocities u and v in the hori-
earlier work (Doroodchi et al., 2008) was adopted in the exper- zontal (X) and vertical (Y) directions, respectively. From these
imental work. First, the operational parameters including the measurements the experimental turbulence intensity, q, was
vibration frequency, the size of the stroke and the distance calculated in accordance with Shy et al. (1997), who also sug-
between the grids were set. A turbulent flow field was then gested that turbulence intensity is related to the grid geometry
and operating conditions by the expression:

0.5 −1.5
q = CfS1.5
L M H , (3)

where f is the frequency, SL is the stroke length, H is the ini-


tial distance between the two grids and M is the distance
between the centres of two successive openings. The exper-
imental measurements have been plotted according to Eq. (3)
in Fig. 3, and produce a linear relationship.
The motion of bubbles released at the bottom of the ves-
sel was captured using a high-speed video camera. The rising
velocity for each bubble was determined from a plot of vertical
distance versus time (i.e. the particle trajectory in y direction
as a function of time) using Phantom5 software. As there was
Fig. 2 – Bubble generation device. no net motion in the vessel, the slip velocity was equivalent
1354 chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362

Fig. 4 – Typical trajectories of a 2.4-mm bubble


(u = 33 mm s−1 ).

to the bubble rising velocity. Each run was repeated a num-


ber of times with the velocity being reported as the average
value. The motion of single bubbles in stagnant distilled water
was first investigated. It was observed that the smaller bubbles
of about 2.5 mm maintained their oblate shape with convex
interface as it rose in the quiescent water. The shape of bub-
bles with sizes greater than 5 mm in diameter however was
difficult to characterise as they underwent periodic dilations
and random wobbling motions during their rise in the stag-
nant water under the force of gravity. The rising path of all
bubbles in quiescent fluid however followed a relatively sym-
metric zigzag path in the XY plane. As shown in Figs. 4 and 5,
introduction of turbulence typically increased the wobbling
behaviour of the bubbles and their lateral movements forcing Fig. 6 – (a) Slip velocity ratio vs. stokes number: data from
them to travel longer paths. These typical measurements were literature. (b) Slip velocity ratio vs. Stokes number:
carried out at u = 33 mm s−1 . validation of Lane et al. (2005) correlation using the
Fig. 6 shows the slip velocity ratio, US /UT , as a function experimental data obtained in the present study.
of Stokes number, St . The data from literature (Poorte and
Biesheuvel, 2002; Lane et al., 2005; Spelt and Biesheuvel, 1997) 2–10 mm diameter bubbles (i.e. ellipsoidal bubbles) are pre-
for bubble diameters of less than 1 mm are presented in Fig. 6a sented in Fig. 6b. First, it was found that US /UT of the ellipsoidal
while the experimental results obtained in this study for bubbles on average was reduced by the turbulence as St num-
ber was increased over the range of 0.02–0.08. This trend is
similar to that observed for bubbles with d ≤ 1. Secondly, the
new experimental data for the larger bubbles is well matched
by the prediction of Lane et al. (2005), given by Eq. (2), while
that of Spelt and Biesheuvel (1997) tended to overestimate the
influence of turbulence on the slip velocity. More generally, it
can be seen that there is considerable spread in the data, pos-
sibly indicating that the influence of turbulence on the slip
velocity cannot be entirely captured in terms of Stokes num-
ber alone. Furthermore, the results presented here are limited
to low values of St and ˇ, and quite clearly more work is needed
for intermediate size bubbles at higher turbulence intensities.
Meanwhile, the correlation by Lane et al. (2005) provides a rea-
sonable measure of how turbulence influences the slip veloc-
ity of bubbles and particles in the 1–2-mm diameter size range.

3. Gas dispersion modelling in


mechanically stirred tanks

Fig. 5 – Typical trajectories of 8 mm bubble at Simulations of mechanically stirred tanks have been reported
(u = 33 mm s−1 ). in the literature since the late 1970s (Daskopoulos and Harris,
chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362 1355

1996). The capabilities of CFD models have improved over the viscosity concept, according to
years, due to continual improvements in computers and mod-  
elling procedures. For single-phase flows especially, extensive 2
−˛i i ui ui = ˛i T (∇Ui + ∇UT
i )− kI , (6)
development and validation against experimental data has 3 i i
been reported (e.g. Luo et al., 1994; Lee et al., 1996; Wechsler
where ki is the turbulent kinetic energy per unit mass, I is
et al., 1999; Aubin et al., 2004). For multiphase flows how-
the unit tensor, and T is the turbulent viscosity. The turbu-
ever, especially gas dispersion, the number of studies has
lent viscosity in each phase is obtained through the k–ε model,
been fewer (e.g. Brady et al., 2006). Compared to single-phase
which incorporates additional equations for k and ε, the rate
flow, there are many additional considerations in the mod-
of dissipation of turbulent energy per unit mass. The effect
elling method, but simulations have tended to be more limited
of bubbles on the turbulence was modelled by an additional
in their accuracy and predictive capabilities. For example, in
contribution to the turbulent viscosity due to bubble slip (Sato
many cases (e.g. Gosman et al., 1992; Bakker, 1992; Morud
and Sekoguchi, 1975), although this contribution turns out to
and Hjertager, 1996; Djebbar et al., 1996; Jenne and Reuss,
be relatively small.
1997), the impeller was not simulated directly, but instead
The interphase force term, Mi , is very important in order to
a ‘black box’ approach was adopted, where the effect of the
describe the gas distribution and holdup in the tank. The total
impeller was modelled using a momentum source or empir-
force on bubbles includes terms for drag, added mass, lift force
ically determined boundary conditions. Such an approach
and the history force (Lathouwers, 1999). Of these, only the
severely limits the predictive capabilities of the model. In
drag force was included. The added mass and lift forces were
some later studies, such as Ranade and Deshpande (1999),
not included because they led to numerical instabilities. In
and Deen et al. (2002), the impeller was included as part
any case, it was estimated that the added mass and lift forces
of the simulation domain, but other limitations remained
were only significant in the impeller region, but the form of
evident. For example, those authors assumed a single con-
the equations is uncertain here due to high turbulence and
stant bubble size throughout the vessel. Also, there has been
high velocity gradients. The history force was also neglected,
no general agreement on the specification of various terms
as usual for turbulent flows (Lathouwers, 1999).
in the two-phase equations, such as the interphase forces,
and validation of modelling results has often been quite
3.2. Modelling the drag force on bubbles
limited.
Further development of modelling methods for gas dis-
The drag force term in the averaged momentum conservation
persion in mechanically stirred tanks was undertaken by one
equation may be given as
of the authors (Lane et al., 2002, 2005; Lane, 2005). Here, the
impeller was represented explicitly, and the CFD method was
3 CD
developed further to take into account the interaction of the F2 = −F1 = − ˛2 1 |U2 − U1 |(U2 − U1 ), (7)
4 d
gas with the impeller, since ventilated gas cavities may form,
with a subsequent large effect on power draw and pumping where U1 and U2 are the Reynolds-averaged velocities of the
capacity of the impeller. Further investigation was undertaken gas and liquid, d is the local Sauter diameter, and CD is the
relating to the appropriate specification of the forces acting mean drag coefficient which is related to the slip velocity by
between gas and liquid, especially the drag force and the tur-
bulent dispersion force. Prediction of the bubble sizes in the CD
 U −2
S
= , (8)
tank was also considered. CD,0 UT

3.1. Equations for two-phase flow where Eq. (2) has been used to account for the influence of
turbulence. The drag coefficient for bubbles rising in quiescent
The simulation method was based on an Eulerian–Eulerian water, CD,0 , was calculated from the correlation of Ishii and
model, where equations for the conservation of mass Zuber (1979).
and momentum of each phase are derived by condi-
tional ensemble averaging (Lathouwers, 1999), and may be 3.3. Turbulent dispersion
given as
Averaging of the two-phase equations has been shown to
lead to additional correlation terms in the momentum con-
∂(˛i i )
+ ∇ · (˛i i Ui ) = Si , (4) servation equation, which are usually modelled as a turbulent
∂t
dispersion force. Typically, the turbulent dispersion force, T,
takes the form (Kurul and Podowski, 1990; Lahey et al., 1993;
Lo, 2000):
∂(˛i i Ui )
+ ∇ · (˛i i Ui Ui ) = −˛i ∇p + ˛i (i − 0 )g + ∇ · (˛i L,i (∇Ui
∂t
T2 = −Ctd 1 k1 ∇˛2 . (9)
T
+(∇Ui ) ) − ˛i i ui ui ) + Mi + Si Ui , (5)
Different values of the coefficient, Ctd , have been applied to fit
where i is the phase number (i = 1 for liquid, i = 2 for gas), ˛i is results to experimental data, e.g. Kurul and Podowski (1990)
the phase volume fraction, i is the phase density, 0 is a ref- recommend a value of 0.1, whereas Lo (2000) used this equa-
erence density, t is time, Ui is the Reynolds-averaged velocity tion with Ctd equal to 1.0. However, it is more likely that Ctd
vector, Si is a mass source term (e.g. at inlets or spargers), p is is not a constant but depends on the detailed characteristics
the modified pressure, g is the gravity vector, L is the lami- of the bubbles and the turbulent flow. Thus, a more detailed
nar viscosity, and Mi is the interphase force. The term in ui ui is model for the turbulent dispersion force was implemented,
the turbulent Reynolds stress, which is defined using the eddy following an approach similar to Simonin and coworkers
1356 chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362

(Simonin, 1990; Bel F’dhila and Simonin, 1992; Viollet and 3.4. Bubble size distribution
Simonin, 1994). The turbulent dispersion force is derived by
averaging of a Lagrangian equation for bubble motion, which Prediction of the bubble size distribution is important since, in
leads to additional force terms representing the correlation combination with the gas volume fraction, a prediction of the
between fluctuations in gas volume fraction and fluctuations interfacial surface area can be obtained, which is important for
in the instantaneous drag force, added mass force, and liquid gas–liquid mass transfer. There are various levels of complex-
pressure gradient. Thus, the turbulent dispersion force was ity at which the bubble size distribution can be approached.
modelled as the sum of three terms: In some studies (e.g. Ranade and Deshpande, 1999; Deen et
al., 2002) a fixed average bubble diameter was specified. The
D12 bubble diameter was then determined either from experimen-
T2,1 = −˛2 1 FD ∇˛2 , (10)
˛1 ˛2 tal measurements or from published empirical correlations
  such as that of Calderbank (1958). A much more detailed mod-
2 b2 − b elling approach is the population balance model. This involves
T2,2 = − 1 k1 CA ∇˛2 , (11)
3 1 + r the solution of an equation describing the probability density
function of the bubble diameter or bubble volume, account-
2
b +  
r ing for the formation or destruction of bubbles by break-up
T2,3 = 1 k1 ∇˛2 . (12)
3 1 + r and coalescence. This is a very detailed and computationally
demanding approach. As a compromise, a simplified model
In these equations, the diffusivity coefficient, D12 , is given by was adopted, where the local average bubble size is calculated
from the bubble number density equation (Bakker, 1992). This
2
b +  
D12 =
r t
k1 12 . (13) equation can be derived by integrating the population balance
3 1 + r over all bubble sizes (Kocamustafaogullari and Ishii, 1995). The
bubble number density, n, is defined as
FD is given by

˛
3 CD n= . (19)
FD = UT (14) ( /6)d3eq
4 d

where CD is the drag coefficient for rise in a stagnant liq- The equation for conservation of bubble number density is
uid, calculated from the correlation of Ishii and Zuber (1979), written as
and UT is the corresponding terminal velocity. Also, in these
expressions:
∂n(x, t)
+ ∇ · (nU2 ) = ϕbr − ϕco + ϕph , (20)
∂t
1 + CA
b= (15)
(2 /1 ) + CA
where the terms on the right hand size represent source or
where the added mass coefficient is taken as CA = 0.5, and sink terms due to break-up, coalescence, and phase change.
The rate of break-up was derived based on a consideration of
t the collision rate between bubbles and turbulent eddies, and
12
r = . (16)
p the efficiency as related to the energy available for break-up.
The rate of coalescence was derived based on the collision
The ratio r compares the characteristic time scales of the rate between bubbles and the collision efficiency, which was
turbulence and of the particle. The characteristic time scale of related to the drainage time of the film between two bubbles.
the turbulence is calculated from Further details may be found in Lane et al. (2005) and Lane
(2005).
TL
t
12 =  (17)
1 + C␤  2
3.5. Gas cavity formation

where TL is the large eddy time scale of the turbulence, and


In gas-sparged tanks, gas cavities form in the low-pressure
region on the trailing side of impeller blades. This can have
UT
=  (18) a strong effect on the power consumption and the discharge
(2/3)k1
flow rate of the impeller. According to various studies (e.g.
Warmoeskerken and Smith, 1985) there are several regimes
t
12 is the turbulent eddy time scale as seen by the parti- for the interaction of gas with the Rushton turbine. At low gas
cle, taking into account the so-called ‘cross-trajectory effect’ flow rates, the vortex cavity regime is observed, while at higher
according to Csanady (1963) where, because of the slip velocity gas flow rates, ventilated gas cavities form which extend to
of the bubbles, the bubbles cut through the turbulent eddies, the blade surface, referred to as the ‘clinging’ and ‘large’ cavity
and the time scale of the eddies as seen by the particles is regimes. An attempt was made to account for cavity formation
reduced. through some modifications to the modelling equations. This
The inclusion of the turbulent dispersion force was found mainly involved a modification to the drag force expression,
to be important, since otherwise the gas accumulated at unre- Eq. (7), for high gas volume fractions.
alistically high volume fractions in the centres of the ring It was considered that as the gas accumulates near an
vortices of the bulk flow pattern. Further details may be found impeller blade, rapid coalescence will occur. Thus, it was sup-
in Lane et al. (2005) and Lane (2005). posed that the bubble size could be expressed as a function of
chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362 1357

Table 1 – Operating conditions and some key CFD predictions


Case number Case 1 Case 2 Case 3 Case 4 Case 5

Impeller type Rushton Rushton Rushton Rushton Lightnin A315


Impeller speed (rpm) 180 250 180 285 600
Gas flow rate (m3 s−1 ) 0.00164 0.00164 0.00687 0.00687 5.57 × 10−4
Gas flow number 0.015 0.011 0.062 0.039 0.010
Froude number 0.31 0.59 0.31 0.77 1.81
Impeller cavity regime Vortex Vortex Large Large Vortex
Gas holdup (%) experimental 3.0 3.7 7.8 9.7 4.6
Gas holdup (%) simulation 2.1 3.0 7.5 9.4 4.7
PG /PUG experimental 0.82 0.84 0.52 0.50 –
PG /PUG simulation 0.70 0.79 0.65 0.54 1.25

gas fraction by a simple power law according to to several criteria, including sufficient reduction of the mass
residuals, an accurate balance between rates of gas entering
K1 and leaving the tank, and a steady gas holdup.
d= m . (21)
(1 − ˛2 ) 1 The modelling method was compared extensively against
experimental measurements of local bubble size and local gas
Also, the drag coefficient decreases with increasing gas vol-
fraction according to Barigou and Greaves (1992) and Bakker
ume fraction (the bubble “swarm” effect). According to Ishii
(1992). These comparisons are detailed in Lane et al. (2005)
and Zuber (1979), for bubbles at ˛2 > 0.3, the drag coefficient
and Lane (2005). Typical distributions of the gas fraction and
can be given by
bubble diameter are illustrated in Figs. 7 and 8. Overall, reason-
able agreement was obtained. Predictions of overall gas holdup
8 m
CD = (1 − ˛2 ) 2 , (22) were in good agreement with measurements, as summarised
3
in Table 1. For the tank with the Rushton turbine, the power
where Ishii and Zuber gave a value of m2 = 2, while Gidaspow draw under gassed conditions was found to be consistent with
(1994) recommended m2 = 4. Combining Eqs. (21) and (22), an empirical correlation of Bakker et al. (1994).
expression for the ratio CD /d in the drag force equation is With the tank stirred by a Rushton turbine, the accumu-
lation of the gas behind the impeller blades was found to
CD 8 m +m correspond fairly well with the observations of Barigou and
= (1 − ˛2 ) 1 2 , (23)
d 3K1 Greaves (1996). The gas accumulation was relatively low for
Cases 1 and 2, where the vortex cavity regime had been
where m1 , m2 and K1 are fitted parameters. Adjustment of the
reported. In Cases 3 and 4, where the ‘large’ cavity regime was
fitted parameters gave fair agreement with experimental data
reported, more substantial regions with high gas fraction were
over a range of conditions.
obtained. However, the resolution of the gas cavities was lim-
In addition to modifying the drag force equation, it was
ited by the number of finite volume cells near the impeller. To
found to be important to model turbulence in the gas phase
obtain better resolution of these gas cavities, the grid density
as well as the liquid, since in regions of segregated flow with
was increased. Results are shown in Figs. 9 and 10 for a sim-
100% gas, the gas phase may become turbulent.
ulation with ∼440,000 cells covering a 60◦ periodic section of

3.6. Implementation of the modelling equations

The CFD method was applied to model gas dispersion (in


air–water mixtures) in baffled tanks stirred either by a Rush-
ton turbine or a Lightnin A315 impeller. A series of different
cases were considered where the tank geometry and operat-
ing conditions were chosen to match with those for which
experimental measurements were available in the literature
(Barigou and Greaves, 1992, 1996; Bakker, 1992). Five different
cases have been considered, as summarised in Table 1. Simula-
tions were carried out using the commercial code CFX4 (CFX4
Solver Manual, 2002), with some additional user-supplied sub-
routines. The basis of the code is a finite-volume method with
all variables defined at the centres of cell volumes. Impeller
motion was accounted for using either the multiple frames
of reference (MFR) method or the sliding mesh method. The
MFR method is a pseudo-steady state method, where a solu-
tion to the equations is found based on one fixed position of
the impeller, coupling different frames of reference for the
impeller region and the outer part of the tank. The sliding
mesh method is a transient method, where the impeller region
is rotated in small angular steps and the fluid flow field is
recalculated until periodically steady results are obtained. Sat- Fig. 7 – Gas volume fraction (Case 2) in vertical plane half
isfactory completion of the simulations was judged according way between baffles.
1358 chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362

Fig. 10 – Gas volume fraction isosurface (98%) for Case 4


using higher density grid (large cavity formation shown on
trailing side of impeller blade).

Fig. 8 – Local bubble diameter in vertical plane half way


between baffles (Case 2) (bubble diameter reported in mm).

the tank. A well-defined region with 100% gas was obtained,


and the cavity shape has a resemblance to that of a ‘clinging’
cavity (Smith, 1985).
The modelling method contains a number of different
fitted parameters, some of which may be sensitive to the
operating conditions, impeller type, or the fluids involved. In
particular, the fitted parameters for the coalescence model
will be very sensitive to the actual fluids used and the level
of surfactants. Therefore, there is still a need to test the CFD
method over a wider range of conditions. Further work is
also necessary to include a more accurate expression for the
effect of turbulence on the drag coefficient, based on the lat-
est available data. Simulations with the Rushton turbine led
to a significant under-prediction of the turbulence levels. This Fig. 11 – Surface mesh of CSIRO Denver cell including the
difficulty has also been found by other authors (e.g. Ng et al., shroud and impeller.

1998). It is likely that this is due to insufficient grid resolu-


tion in the region of the impeller. Improvements in the CFD
method should be possible due to increasing computer capac-
ity. By employing much larger finite volume grids, it should be
possible to obtain better predictions of the turbulence. Also,
higher density grids in the region of the impeller will assist in
better prediction of gas cavities. It would be interesting to see if
the modelling method could predict features such as the ‘3–3’
cavity regime on a Rushton turbine, where large and clinging
cavities occur on alternating impeller blades.

4. Flotation modelling

The computational gas dispersion modelling described above


can be extended to predict the recovery performance of actual
flotation cells. Fig. 11 shows the surface mesh of a Denver
flotation cell which is widely used in the mineral processing
industry.
The flotation process is modelled as three sub-processes
involving (1) collision, (2) attachment and (3) detachment. A
Fig. 9 – Gas volume fraction in horizontal plane at 1/4 turbulent collision model has been used by Koh et al. (2000)
impeller height (Case 4) (higher density grid superimposed to estimate the rate of bubble–particle encounters, employ-
and logarithmic scale is used for the colour map). ing the local turbulent velocity and the size and number
chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362 1359

Table 2 – Equations for bubble–particle interactions (taken from Koh and Schwarz, 2006; refer to original paper for
nomenclature)
dnp1
Net attachment rate dt
= −k1 np1 nb (1 − ˇ) + k2 nb ˇ

Attachment rate constant k1 = Z1 Pc Pa Ps


Detachment rate constant k2 = Z2 (1 − Ps )
np2 db
2
Bubble loading ˇ= Smax nb where Smax = fc S and S=2 dp
dp +db
2 1/2
Collision frequency across eddies Z1 = 5.0 2 (Up2 + Ub2 )
15f U2
Critical diameter of particle or bubble d2i > d2crit = i ε
f

 dp +d
3
1/2
8 ε
Collision frequency within eddies Z1 = 15
b
2

0.4 ε4/9 d7/9  −


2/3
Turbulent fluctuating velocity of particle or bubble Ui = i i
f
f
1/3

C1 ε1/3
Detachment frequency Z2 =
(dp +db )2/3
4

d2p
Probability of collision Pc = 1.5 + 0.72
15 Reb d2
b
db Ub
Bubble Reynolds number Reb =
  −(45+8Re0.72 )U 
2 b tind
Probability of adhesion Pa = sin 2 arctan exp b
15 db (db /dp +1)
75 0.6
Induction time tind = dp
1


Probability of stabilisation Ps = 1 − exp As 1 − Bo∗

d2 2/3 ((d /2)+(d /2))−1/3 ]+1.5d ((4/d )−d  g)sin2 ( −( /2))


p [ p g+1.9p ε p b p b b f
Bond number Bo∗ = |6 sin( −( /2)) sin( +( /2))|

concentrations of bubbles and particles obtained from CFD velocities in highly turbulent flow. In regions where velocities
modelling. The probability of collision accounts for the for the particle and bubbles are not independent, the collision
streamline effect of fine particles moving around a larger bub- equation by Saffman and Turner (1956) is applicable for fine
ble. The local attachment and detachment rates are calculated particles and bubbles which are confined within eddies in low
from the collision frequency and the probabilities of colli- turbulent regions.
sion, adhesion and stabilisation (Koh and Schwarz, 2006). The The collision efficiency accounts for the tendency of par-
effects of the turbulence intensity on the local rates of attach- ticles to follow the fluid streamlines around the bubble and
ment and detachment and the local bubble loading can be avoid actual contact. This is especially true for particles that
quantified. are much smaller than the bubble, and the equation proposed
In modelling flotation kinetics, the local turbulent energy by Yoon and Luttrell (1989) is appropriate for intermediate
dissipation rates are obtained by CFD modelling of the cell. bubble Reynolds number. The probability of adhesion is deter-
This is a significant improvement over other models where an mined by the sliding time of the particles on the bubble
average dissipation rate or an assumed Gaussian distribution surface and the induction time for rupture of the disjoining
for turbulent velocities was used for the whole cell (e.g., Bloom film between the particle and bubble. If the sliding time is
and Heindel, 1997, 2003). CFD modelling provides a realistic longer than the induction time, adhesion is likely. Yoon and
approach to flotation models without additional assumptions Luttrell (1989) derived an expression for the adhesion prob-
on turbulent energy dissipation rates. ability dependent on the particle and bubble sizes, bubble
For simulating flotation kinetics, the transfer of particles Reynolds number and induction time.
between the pulp and bubbles is achieved by applying source The bubble–particle detachment frequency is dependent
terms for particle number concentration in the population bal- on the relative velocity between the particle–bubble aggregate
ance equation as follows: and the surrounding fluid. The probability of stabilisation or
destabilisation of the bubble–particle aggregate, as proposed
∂(˛i ni ) by Schulze (1993), is based on the Bond number which is the
+ ∇ · (˛i ni Ui ) = −ϕa + ϕd , (24)
∂t ratio of detachment to attachment forces.
While attachment and detachment processes are occurring
where ϕa and ϕd are sources or sinks specifying attachment
continuously throughout the flotation cell, the particle–bubble
and detachment rates respectively. The transient simulation
aggregates are being transported at the same time by con-
uses variable time steps such that the mass error is less than
vection and buoyancy to the froth interface for recovery. The
0.1% for each time step. The equations used in the model for
gravitational force acting on particles attached to bubbles
bubble–particle collision frequency, and the probabilities of
affects the rise velocity of the aggregate. The effect is signif-
collision, adhesion and stabilisation (Koh and Schwarz, 2006)
icant especially with larger particles at higher loadings. The
are shown in Table 2.
body force acting on the attached particles is incorporated in
In turbulent flows where inertial effects are the pri-
the model by adding a source term, Bp , to the gas phase as
mary cause of collisions, the equation for the number of
follows:
particle–bubble collisions per unit time and volume is based
on work of Abrahamson (1975), and applied by Schubert and  2
db
Bischofberger (1979) to collisions in the flotation cell. The con- Bp = −4fc (p − l )Vp gnb , (25)
dp
dition for using the model is independent of bubble or particle
1360 chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362

where fc is the fractional surface coverage, nb the local num-


ber concentration of bubbles, p the density of particles, and
Vp the volume of individual particles, and db and dp are the
diameters of bubble and particles, respectively. Essentially, the
source term reduces the buoyancy effect of the gas phase and
is based on the number of particles attached on individual
bubbles at a given time and location in the cell. A similar effect
in turbulent flow is expected where the bubble rise velocity is
reduced by the attached particles. The bubble–particle aggre-
gates can sink if the particles are sufficiently large and the
bubble loadings are significantly high.

4.1. Flotation simulation

The CSIRO Denver cell has an eight-bladed impeller driven


from below and enclosed in a conical shroud. The meshing
for the cell is shown in Fig. 11. Air is injected at a rate of Fig. 13 – Predicted net attachment rates (m−3 s−1 ) for 60 ␮m
8 L min−1 through a nozzle located within the shroud and diameter particles (flotation time of 228 s).
directed downwards at the rotating impeller. The flotation cell the limiting factor in the recovery rate at the froth interface.
has a volume of 3.78 L and an impeller diameter of 72 mm. This explains why the relationship between flotation rate and
Results of two-phase simulations for the CSIRO Denver cell bubble surface area flux is generally used as a criterion for
have been obtained. A typical plot of the liquid velocity vec- designing flotation cells. In fact, transport rates to the froth
tors, as shown in Fig. 12, is similar to those obtained previously layer contribute quite significantly to the overall flotation per-
(Koh and Schwarz, 2003). The predictions indicate a complex formance (Koh and Schwarz, 2006).
flow field within the flotation cell and large fluid velocities From the simulations with flotation, predicted flotation
observed in the impeller region. Gas holdup distribution and rate constants can be obtained by analysing the total num-
bubble–particle collision rates in the cell have been reported ber of particles (both free and attached particles) remaining
previously (Koh et al., 2000). in the cell with time, and this involves summing all the parti-
The distribution of net attachment rates in the CSIRO Den- cles in all finite volumes in the cell at each time step (Koh and
ver cell operating at 1200 rpm is shown in Fig. 13. The plot Schwarz, 2006). By assuming first order kinetics, flotation rate
shows that detachment rates (with negative values) are also constants can be determined from the slopes of the logarithm
large near the impeller tip, but the maximum attachment of the particles remaining in the cell as a function of time.
rates are outside the impeller zone. The two zones are sepa- The effect of bubble loading (the amount of particles car-
rated by the stator shroud. The negative regions are inevitable ried by the bubble surface) on the flotation rate has been
because of the need to use a high impeller speed to generate investigated by CFD simulations (Koh and Schwarz, in press).
fine bubbles for the necessary bubble surface area flux to oper- Simulations have been performed with assumed maximum
ate. Flotation cell design should ideally minimise the negative bubble loading of a half mono-layer coverage (fc = 0.5) and with
regions while maximizing the attachment rates in the cell. a maximum coverage of 0.2. Flotation rates of quartz parti-
The predicted concentration of particles remaining in the cles in a Rushton-turbine stirred tank reported in the literature
cell indicates that the particle recovery rate to the pulp–froth (Ahmed and Jameson, 1985) have been used for comparison.
interface is much slower than the net attachment rates after The CFD predicted flotation rates for the assumed maximum
accounting for detachments. The bubbles are loaded with of 0.2 fraction of a mono-layer match experimental data better
particles quite quickly, and the bubble surface area flux is than the case for 0.5 fraction of a mono-layer. The predicted
bubble loadings for both cases are compared against mea-
sured bubble loadings in a micro-flotation cell (Bradshaw and
O’Connor, 1996). An alternative method is to assume an aver-
age loading based on total numbers of particles and bubbles in
the flotation cell, and this assumption was used in the latest
model by Bloom and Heindel (2003). However, from the latest
results by Koh and Schwarz (in press), the case for maximum
loading of 0.2 fraction of bubble coverage matched experi-
mental values better than the 0.5 cases. The bubble loading
actually obtained in the flotation cell is expected to be depen-
dent on the particle surface properties, collector and frother
chemistry, as well as cell geometry and hydrodynamics.
The predicted flotation rates also showed that fine and
large particles do not float as well as intermediate size par-
ticles (Koh and Schwarz, 2007). This is consistent with the
recovery generally observed in practice.

5. Conclusions
Fig. 12 – Vector plot of liquid velocities at the mid-plane of
flotation cell (operating at 1200 rpm and typical velocity is The motion of ellipsoidal bubbles, with equivalent diameters
about 0.1 m s−1 ). of 2–10 mm, in stationary near isotropic turbulent flow fields
chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362 1361

was investigated experimentally. The turbulence was gener- Barigou, M. and Greaves, M., 1996, Gas holdup and interfacial
ated using a pair of oscillating grids. A thorough assessment of area distributions in a mechanically agitated gas-liquid
the flow field using PIV techniques demonstrated that the flow contactor. Chem. Eng. Res. Des., 74(3): 397–405.
Bel F’dhila, R. and Simonin, O., 1992, Eulerian prediction of
in the central region between the grids was near isotropic with
turbulent bubbly flow downstream of a sudden pipe
an integral length scale of 8 mm. Generally, the investigations expansion, In Proceedings of the Sixth Workshop on Two-Phase
were performed for Stokes values less than 0.1. The bubble tra- Flow Prediction , pp. 264–273.
jectories as well as their mean rise velocity were obtained for Bloom, F. and Heindel, T.J., 1997, Mathematical modeling of the
horizontal turbulence intensities ranging from 0 to 50 mm s−1 . flotation deinking process. Math Comp Model, 25: 13–58.
It was observed that the bubbles often tend to travel longer Bloom, F. and Heindel, T.J., 2002, On the structure of collision and
detachment frequencies in flotation models. Chem Eng Sci,
paths due to their interaction with the turbulent eddies. The
57(13): 2467–2473.
results also indicated that, at low values of St and ˇ, US /UT of
Bloom, F. and Heindel, T.J., 2003, Modeling flotation separation in
the ellipsoidal bubbles on average reduces by the turbulence as a semi-batch process. Chem Eng Sci, 58: 353–365.
St number increases. Furthermore, it was demonstrated that Bradshaw, D.J. and O’Connor, C.T., 1996, Measurement of the
Lane et al. (2005) drag coefficient correlation can be used to sub-process of bubble loading in flotation. Miner Eng, 9:
provide a reasonable measure of how turbulence influences 443–448.
the slip velocity of intermediate size bubbles at low turbulence Brady, M.R., Telionis, D.P., Vlachos, P.P. and Yoon, R.-H., 2006,
Evaluation of multiphase flotation models in grid turbulence
intensities.
via particle image velocimetry. Int J Miner Process, 80: 133–143.
A CFD modelling method was developed for gas disper-
Brucato, A., Grisafi, F. and Montante, G., 1998, Particle drag
sion in agitated vessels. The details of the impeller geometry coefficients in turbulent fluids. Chem Eng Sci, 53(18):
were included in the model, and the interaction of the gas 3295–3314.
with the impeller was considered, including cavity formation. Calderbank, P.H., 1958, The interfacial area in gas–liquid
When specifying the drag force, the effect of turbulence was contacting with mechanical agitation. Trans IChemE, 36:
included by a modification to the drag coefficient based on a 443–463.
CFX4 Solver Manual, 2002 (AEA Technology, Oxford, UK).
local Stokes number. A detailed model for turbulent disper-
Clift, R., Grace, J.R. and Weber, M.E., (1978). Bubbles, Drops and
sion was also introduced, which was found to give a more Particles. (Academic Press).
realistic gas distribution. A bubble number density equation Csanady, G.T., 1963, Turbulent diffusion of heavy particles in the
was introduced which allowed prediction of local mean bub- atmosphere. J Atmos Sci, 20: 201–208.
ble diameter, taking bubble break-up and coalescence into Dai, Z., Fornasiero, D. and Ralston, J., 2000, Particle–bubble
account. The method was implemented in simulations of collision models: a review. Adv Colloid Interf Sci, 85(2–3):
231–256.
tanks stirred by a Rushton turbine and by a Lightnin A315
Daskopoulos, P. and Harris, C.K., 1996, Three-dimensional CFD
impeller. Simulation results were compared in detail with
simulations of turbulent flow in baffled stirred tanks: an
experimental data in the literature and fairly good agreement assessment of the current position, IChemE Symp Ser No. 140,
was found in most respects. 1–8.
The computational gas dispersion modelling has been Deglon, D.A. and Meyer, C.J., 2006, CFD modelling of stirred tanks:
extended to predict the recovery performance of the Den- numerical considerations. Miner Eng, 19: 1059–1068.
ver flotation cell. A population balance approach has been Deen, N.G., Solberg, T. and Hjertager, B.H., 2002, Flow generated
by an aerated Rushton impeller: two-phase PIV experiments
applied to account for fundamental processes involving
and numerical simulations. Can J Chem Eng, 80: 638–652.
bubble–particle collision, attachment and detachment. The
De Silva, I.P.D. and Fernando, H.J.S., 1994, Oscillating grids as a
predicted flotation kinetics has been found to agree with mea- source of nearly isotropic turbulence. Phys Fluids, 6(7):
sured data reported in the literature. The combined model 2455–2464.
with hydrodynamic and bubble–particle interaction has been Djebbar, R., Roustan, M. and Line, A., 1996, Numerical
shown to be useful for practical application in the investiga- computation of turbulent gas–liquid dispersion in
tion of flotation cells. The design and operating conditions can mechanically agitated vessels, Trans IChemE, Part A, 74:
492–498.
be identified for optimum cell operation.
Doroodchi, E., Evans, G.M., Schwarz, M.P., Lane, G.L., Shah, N. and
Nguyen, A., 2008, Influence of turbulence intensity of particle
References drag coefficients. Chem Eng J, 135: 129–134.
Gidaspow, D., (1994). Multiphase flow and fluidization. (Academic
Press).
Abare, 2007, Australian mineral statistics: September quarter
Gosman, A.D., Lekakou, C., Politis, S., Issa, R.I. and Looney, M.K.,
2007 (Australian Bureau of Agricultural and Resource
1992, Multidimensional modeling of turbulent two-phase
Economics, Canberra, Australia), ISSN 1447-1159
flows in stirred vessels. AIChE J, 38(12): 1946–1956.
Abrahamson, J., 1975, Collision rates of small particles in a
Hetherington, L.E., Brown, T.J., Benham, A.J., Bide, T., Lusty, P.A.J.,
vigorously turbulent fluid. Chem Eng Sci, 30: 1371–1379.
Hards, V.L., Hannis, S.D. and Idoine, N.E., (2008). World Mineral
Ahmed, N. and Jameson, G.J., 1985, The effect of bubble size on
Production 2002–2006. (British Geological Survey, London,
the rate of flotation of fine particles. Int J Miner Process, 14:
England). ISBN: 978-0-85272-615-0
195–215.
Ishii, M. and Zuber, N., 1979, Drag coefficient and relative velocity
Aubin, J., Fletcher, D.F. and Xuereb, C., 2004, Modeling turbulent
in bubbly, droplet or particulate flows. AIChE J, 25(5): 843–855.
flow in stirred tanks with CFD: the influence of the modeling
Jenne, M. and Reuss, M., 1997, Fluid dynamic modelling and
approach, turbulence model and numerical scheme. Exp
simulation of gas–liquid flow in baffled stirred tank reactors.
Thermal Fluid Sci, 28: 431–445.
Récents Progrès en Genie des Procédès, 11(52): 201–208.
Bakker, A., 1992, Hydrodynamics of stirred gas–liquid dispersions,
Kocamustafaogullari, G. and Ishii, M., 1995, Foundation of the
Ph.D. Thesis, Technical University of Delft, The Netherlands.
interfacial area transport equation and its closure relations.
Bakker, A., Smith, J.M. and Myers, K.J., December 1994, How to
Int J Heat Mass Transfer, 38(3): 481–493.
disperse gases in liquids. Chem Eng, 98–104.
Koh, P.T.L., Manickam, M. and Schwarz, M.P., 2000, CFD
Barigou, M. and Greaves, M., 1992, Bubble-size distributions in a
simulation of bubble–particle collisions in mineral flotation
mechanically agitated gas–liquid contactor. Chem Eng Sci,
cells. Miner Eng, 13: 1455–1463.
47(8): 2009–2025.
1362 chemical engineering research and design 8 6 ( 2 0 0 8 ) 1350–1362

Koh, P.T.L. and Schwarz, M.P., 2003, CFD modelling of measurements of the flow in a tank stirred by a Rushton
bubble–particle collision rates and efficiencies in mineral impeller. Trans IChemE, 76(A): 737–747.
flotation cells. Miner Eng, 16: 1055–1059. Nguyen, A.V. and Schulze, H.J., (2004). Colloidal Science of Flotation.
Koh, P.T.L., Schwarz, M.P., Zhu, Y., Bourke, P., Peaker, R. and (Marcel Dekker, NewYork).
Franzidis, J.P., 2003, Development of models of mineral O’Connor, C.T., 2005, Contributions to an improved
flotation cells, In Proceedings of the Third Int. Conf. On CFD in Min understanding of the flotation process, Doctor of Engineering
and Proc. Industries , pp. 171–176. Dissertation, University of Stellenbosch, South Africa.
Koh, P.T.L. and Schwarz, M.P., 2006, CFD modelling of Poorte, R.E.G. and Biesheuvel, A., 2002, Experiments on the
bubble–particle attachments in flotation cells. Miner Eng, 19: motion of gas bubbles in turbulence generated by an active
619–626. grid. J Fluid Mech, 461: 127–154.
Koh, P.T.L. and Schwarz, M.P., 2007, CFD model of a self-aerating Ranade, V.V. and Deshpande, V.R., 1999, Gas–liquid flow in stirred
flotation cell. Int J Miner Process, 85: 16–24. reactors: trailing vortices and gas accumulation behind
Koh, P.T.L. and Schwarz, M.P., in press, Modelling attachment impeller blades. Chem Eng Sci, 54: 2305–2315.
rates of multi-sized bubbles with particles in a flotation cell, Saffman, P.G. and Turner, T.S., 1956, On the collision of drops in
Miner. Eng. turbulent clouds. J Fluid Mech, 1: 16–30.
Kostoglou, M., Karapantsios, T.D. and Matis, K.A., 2007, CFD Sato, Y. and Sekoguchi, K., 1975, Liquid velocity distribution in
model for the design of large scale flotation tanks for water two-phase bubble flow. Int J Multiphase Flow, 2: 79–95.
and wastewater treatment. Ind Eng Chem Res, 46(20): Schubert, H. and Bischofberger, C., 1979, On the optimization of
6590–6599. hydrodynamics in flotation processes, In Proceedings of the 13th
Kurul, N. and Podowski, M.Z., 1990, Multidimensional effects in International Mineral Processing Congress, vol. 2 , pp. 1261–1285.
sub-cooled boiling, In Proceedings of the Ninth Heat Transfer Conf Schubert, H., 1999, On the turbulence-controlled microprocesses
, pp. 21–26. in flotation machines. Int J Miner Process, 56: 257–276.
Kwon, Y., Zhang, J. and Lee, H.-G., 2006, Water model and CFD Sherrell, I.M., 2004, Development of a flotation rate equation from
studies of bubble dispersion and inclusions removal in first principles under turbulent flow conditions, Ph.D. Thesis,
continuous casting mold of steel. ISIJ Int, 46(2): 257–266. Virginia Polytechnic Institute and State University, USA.
Lahey, R.T., Jr., Lopez de Bertodano, M. and Jones, O.C., Jr., 1993, Schulze, H.J., 1993, Flotation as a heterocoagulation process:
Phase distribution in complex geometry conduits. Nucl Eng possibilities of calculating the probability of flotation, in
Des, 141: 177–201. Coagulation and Flocculation, Dobias, B. (ed) (Marcel Dekker,
Lane, G.L., 2005, Numerical modelling of gas–liquid flow in stirred New York), pp. 321–353.
tanks, Ph.D. Thesis, University of Newcastle, Australia. Shy, S.S., Tang, C.Y. and Fann, S.Y., 1997, A nearly isotropic
Lane, G.L., Schwarz, M.P. and Evans, G.M., 2002, Predicting turbulence generated by a pair of vibrating grids. Exp Thermal
gas–liquid flow in a mechanically-stirred tank. Appl Math Fluid Sci, 14: 251–262.
Model, 26: 223–235. Simonin, O., 1990, Eulerian formulation for particle dispersion in
Lane, G.L., Schwarz, M.P. and Evans, G.M., 2005, Numerical turbulent two-phase flows, In Proceedings of the Fifth Workshop
modelling of gas–liquid flow in stirred tanks. Chem Eng Sci, on Two-Phase Flow Predictions , pp. 156–166.
60: 2203–2214. Smith, J.M., 1985, Dispersion of gases in liquids: the
Lathouwers, D., 1999, Modelling and simulation of turbulent hydrodynamics of gas dispersion in low viscosity liquids, in
bubbly flow, Ph.D. Thesis, Technical University of Delft. Mixing of Liquids by Mechanical Agitation, Ulbrecht, J.J. and
Lee, K.C., Ng, K. and Yianneskis, M., 1996, Sliding mesh Patterson, G.K., Patterson, G.K. (eds) (Gordon and Breach, New
predictions of the flows around Rushton impellers. IChemE York), pp. 139–201.
Symp Ser, 140: 47–58. Spelt, P.D.M. and Biesheuvel, A., 1997, On the motion of gas
Lichter, J., Potapov, A.V. and Peaker, R., 2007, The use of bubbles in homogeneous isotropic turbulence. J Fluid Mech,
computational fluid dynamics and discrete element modeling 336: 221–244.
to understand the effect of cell size and inflow rate on Tiitinen, J., Vaarno, J. and Gronstrand, S., 2003, Numerical
flotation bank retention time distribution and mechanism modelling of an outokumpu flotation device, In Proceedings of
performance, Proceedings of the 39th Annual Canadian the Third Int. Conf. On CFD in Min and Proc. Industries , pp.
Mineral Processors Operators Conference, Ottawa, Canada, 167–170.
paper 09. Viollet, P.L. and Simonin, O., 1994, Modelling dispersed two-phase
Lo, S., 2000, Recent advances in CFD for industrial multiphase flows: closure, validation and software development. Appl
flows, in Industrial CFD and the Move Towards Multiphase Mech Rev, 47(6 (Part 2)): S80–S84.
Flow Simulations, NAFEMS Awareness Seminar, November 8, Warmoeskerken, M.M.C.G. and Smith, J.M., 1985, Flooding of disc
2000, University of Warwick, UK. turbines in gas–liquid dispersions: a new description of the
Luo, J.Y., Issa, R.I. and Gosman, A.D., 1994, Prediction of phenomenon. Chem Eng Sci, 40(11): 2063–2071.
impeller-induced flows in mixing vessels using multiple Weber, A., Meadows, D., Villanueva, F., Palomo, R. and Prado, S.,
frames of reference. IChemE Symp Ser, 136: 549–556. 2005, Development of the world’s largest flotation machine, In
Maniruzzaman, M. and Makhlouf, M., 2002, Mathematical Centenary of Flotation Symposium , pp. 285–291.
modelling and computer simulation of the rotating impeller Wechsler, K., Breuer, M. and Durst, F., 1999, Steady and unsteady
particle flotation process. Part 1. Fluid flow. Metall Mater computations of turbulent flows induced by a 4/45◦ pitched
Trans, 33B: 297–303. blade impeller. J Fluids Eng, 121: 318–329.
Maniruzzaman, M. and Makhlouf, M., 2002, Mathematical Yang, T.S. and Shy, S.S., 2003, The settling velocity of heavy
modelling and computer simulation of the rotating impeller particles in an aqueous near-isotropic turbulence. Phys Fluids,
particle flotation process. Part 2. Particle agglomeration and 15(4): 868–880.
flotation. Metall Mater Trans, 33B: 305–314. Yianatos, J., 2007, Fluid flow and kinetic modelling in flotation
Morud, K.E. and Hjertager, B.H., 1996, LDA measurements and related processes: columns and mechanically agitated
CFD modelling of gas–liquid flow in a stirred vessel. Chem Eng cells—a review. Chem Eng Res Des, 85(12): 1591–1603.
Sci, 51(2): 233–249. Yoon, R.H. and Luttrell, G.H., 1989, The effect of bubble size on
Ng, K., Fentiman, N.J., Lee, K.C. and Yianneskis, M., 1998, fine particle flotation. Miner Process Extract Metall Rev, 5:
Assessment of sliding mesh CFD predictions and LDA 101–122.

You might also like