Chemical Engineering Science 56 (2001) 6303–6311

[Link]/locate/ces

Simulation of the e ect of froth washing on !otation performance

S. J. Neethling , J. J. Cilliers ∗
Froth and Foam Research Group Chemical Engineering Department, UMIST PO Box 88, Manchester M60 1QD, UK

Abstract
The froth phase is extremely important in the operation of a !otation cell, seeing that -it is critical in determining the amount
of unwanted gangue collected to the concentrate and thus the purity of the product. This paper uses a fundamentally based model
of !owing froths to simulate the performance of a !otation cell. The study concentrates speci,cally on the e ect of wash water
addition on the overall performance, i.e. the grade and the recovery. The froth model that is used within this work includes a large
number of the e ects seen within a !otation froth and approaches their description from a fundamental point of view. Some of the
phenomena that are included are bubble coalescence, liquid drainage including the e ects of gravity, surface tension and viscous
dissipation, particle settling and particle dispersion.
The results show the advantages and disadvantages of di erent water addition strategies on the performance of !otation vessels.
Since most recent !otation work has been concerned with improving recoveries, rather than grade, the addition of water into
!otation froths has been largely limited to column cells. This work demonstrates how water addition can be optimised in terms of
both water addition point and quantity in order to produce the desired performance. ? 2001 Published by Elsevier Science Ltd.

Keywords: Flotation; Simulation; Wash water

1. Introduction The ,rst section gives a brief description of the assump-

tions inherent in the model and the mathematical for-
The froth phase of the !otation system, to a great ex- mulations and boundary conditions resulting from these
tent, determines the separation performance, as the grade assumptions. A fuller description of the mathematical
(ratio of desired to total solids recovered) of the product modelling of the liquid and gas motion can be found in
(concentrate) depends primarily on its structure and sta- previously published work (Neethling & Cilliers, 1998;
bility. The froth also contributes to the recovery (fraction Neethling, Cilliers, & Woodburn, 2000).
of valuable solids recovered from the pulp to the con-
centrate) achieved, since the amount of desired material
drop-back from the froth, together with the kinetics of
the pulp phase, determines the recovery. 2. Mathematical modelling
This paper uses a fundamentally-based model of !ow-
ing froths to simulate the performance of a !otation cell.
The study will concentrate speci,cally on the e ect of The mathematical model described below is solvable
wash water addition on the overall performance, i.e. the in both 2 and 3 dimensions, but since the computer
grade and the recovery. code that has been written to solve it takes the best
Wash water is traditionally added to column !otation part of a day to solve a two-dimensional simulation, no
cells, but it has found some, though more limited, use in three-dimensional simulations have yet been attempted.
the operation of more conventional !otation cells. This This calculation time is likely to improve dramatically
paper will examine not only the e ect of wash water rate as the code is further optimised for the non-linear set
on the performance of !otation cells, but also the e ect of equations that are being solved. In 2-D, the model
of the wash water addition point (see Fig. 1). has been implemented using cylindrical and Cartesian
co-ordinates, but the examples show later are of the
∗Corresponding author. Cartesian type. All the equations are expressed relative
E-mail address: [Link]@[Link](J. J. Cilliers). to a stationary reference frame.

0009-2509/01/$ - see front matter ? 2001 Published by Elsevier Science Ltd.

PII: S 0 0 0 9 - 2 5 0 9 ( 0 1 ) 0 0 2 4 8 - 2
6304 S. J. Neethling, J. J. Cilliers / Chemical Engineering Science 56 (2001) 6303–6311

Fig. 1. Schematic diagram of !otation froth.

2.1. Gas motion and coalescence 2.2. Liquid motion and content

In order to model the motion of solids within a !ota- The motion of the liquid in the froth is described using
tion cell, the motion of both the gas and liquid phases are an extension of the model of Verbist, Weaire, and Kraynik
required, since the motion of all three phases are interde- (1996), and is described brie!y here (Neethling et al.,
pendent. 2000).
The froth motion is modelled using Laplace’s equa- The liquid velocity, û, is estimated from a force balance
tion, and assumes the !ow to be incompressible and irro- along a Plateau border incorporating gravity, capillary
tational. Since !otation froths have a relatively low liquid and viscous dissipation forces. Since the froth is slow
content and !ow quite slowly, the assumptions inherent in !owing, inertial forces are ignored, and the three forces
Laplace’s equation produce accurate descriptions of the sum to zero. This yields a liquid velocity equation, with
froth motion (Murphy, Zimmerman, & Woodburn, 1996; the cross-sectional Plateau border area, A, as the only
Neethling & Cilliers, 1998). This yields a !ow velocity dependent variable.
vector, v̂, for each position within the system.
The boundary condition for Laplace’s equation is the  
k2 @A
gas !ux through each of the boundaries. The !ux through −√ × + vx
 A @x  g
the pulp-froth interface is the amount of air being added û = 

;
 where k1 = ;
k2 @A 150
to cell. Gas can leave the froth in the cell by escaping −k1 A − √ × + vy
from bursting bubbles at the surface, or within the froth A @y
that !ows over the weir. The permeability of the burst- √
ing surface to air (i.e. the super,cial !ux of air released 3 − =2 × 
k2 = (1)
by bursting) is a direct function of froth stability and is 300
required model parameter. The !owrate of the gas in the
bubble over!owing the weir is obtained by di erence. This must be combined with a continuity equation
The diameter of the bubbles in the moving froth is re-
quired as a function of position, because, as we will see
later, a number of froth phenomena are dependent on the  · (û × A × ) = Q(x; y) (2)
local bubble diameter. The model incorporates a coales-
cence algorithm based on two criteria: the drainage time to yield a boundary value problem that is solved numeri-
of the lamellae, and the stability of the lamellae at the cally. The length of Plateau borders per volume of froth,
equilibrium thickness. The equilibrium ,lm thickness is , determines the e ect of bubble size on the behaviour
a function of liquid content, since it is a balance between of the liquid within the system and is of the form  ˙
the pressure exerted on the ,lm by the curvature of the (bubble radius)−2 : Q(x; y) is the volume of water being
Plateau border, van der Waals forces and the electric dou- added per volume of froth per time as a function of po-
ble layer repulsion. Both criteria are a function of bub- sition and allows wash water addition.
ble size and therefore, the stability of individual lamellae The solution of Eq. (2) gives the liquid content and
varies within a single !otation froth. liquid velocity vectors at each point through the froth.
 S. J. Neethling, J. J. Cilliers / Chemical Engineering Science 56 (2001) 6303–6311 6305

2.3. Solids motion settling velocity of particle type i; DG the dimensionless

Geometric dispersion coeJcient (≈ 0:25); db the bubble
2.3.1. General diameter, A the cross-sectional Plateau border area and 
In !otation there are, based on surface chemistry, two the length of Plateau borders per volume of froth.
di erent types of solids, hydrophilic mineral that will not The value of DG of 0.25 was estimated by considering
attach to bubble lamellae and hydrophobic mineral, able a two-dimensional grid of hexagons.
to attach.
In the froth phase two classes of material exist: particles
[Link]. Plateau border dispersion. Plateau border dis-
attached to the liquid=gas interfaces, and the unattached
persion is caused by the velocity pro,le of the !uid in
material that is found within the liquid in the Plateau
the Plateau borders, which results in a particle velocity
borders. While the attached particles are assumed to be
pro,le. Plateau border dispersion is in the same direction
hydrophobic, the unattached material can be either hy-
as the liquid motion and its !ux is described by the fol-
drophilic or hydrophobic.
lowing equation.
Each class of solids can be subdivided according to
size and density, and varying degrees of hydrophobicity.  
@Csi
−Dp × dp × A ×  × |vx − ux | ×
 @x 
2.3.2. Attached material F̂ P Net = 

;
@Csi 
This class is relatively simple to model, as it follows −Dp × dp × A ×  × |vy − uy | ×
the bubbles. Coalescence in the froth or bursting on the @y
surface is due to the rupture of the lamellae to which (4)
particles are attached. It is assumed that when a ,lm
ruptures all the particles that were attached to it become where F̂ P Net is the net !ux due to Plateau border disper-
detached. This is an initial modelling assumption that will sion, dp the characteristic Plateau border diameter (2×
radius of curvature) and DP the dimensionless Plateau
need to be improved by means of experimental studies.
border dispersion coeJcient (≈ 0:3).
2.3.3. Unattached material The Plateau border dispersion coeJcient (Dp ≈ 0:3) is
estimated from the equivalent axial dispersion coeJcient
The unattached material, while predominantly follow-
(1=Peclet Number) for !ow in a cylindrical tube.
ing the liquid in the Plateau borders, also moves relative
to that liquid. As noted previously, attached particles are
added to the unattached class as they detach from burst- [Link]. Hindered settling of particles. The rate at
ing or coalescing bubbles. which the particles settle under gravity is a function of
Three di erent ways in which the particles move rel- their size, density and concentration.
ative to the net !uid motion have been identi,ed; geo- Fine particles will predominantly follow the liquid,
metric dispersion, Plateau border dispersion and hindered while coarse particles will tend to settle out of the froth.
particle settling. These mechanisms disperse the particles This leads to a strati,cation of particle sizes within the
throughout the froth. froth and hence particle size selective transport of parti-
cles from the pulp, through the froth, and across the weir.
[Link]. Geometric dispersion. Geometric dispersion The settling velocity of each particle size class can be
is caused by the geometric layout of the Plateau borders estimated from the following hindered settling formulae
through which the particles move. The Plateau borders (Coulson & Richardson, 1993):
form a three-dimensional network of interconnected
channels, with four Plateau borders meeting at a vertex. g × ( s − ) × d2i
vTerm; i =
When a particle enters a vertex from a Plateau border, 18
(5)
there are three other Plateau borders by which it could vTerm; i × (4:65 )
leave. Geometric dispersion provides components of vSettling; i =
3
particle velocity that are perpendicular to the direction
of net liquid motion. The net !ux is described by means where s is the solid density, the liquid density, di
of the following equation. the particle size,  the voidage (volume fraction of liquid
  within the system), vTerm the terminal settling velocity
@Csi
 −D G d b A |vy − vSettling;i − u y | in the absence of other particles and vSettling the apparent
@x 
F̂ G Net = 

;
 (3) settling velocity of particles within the froth.
@Csi
−DG db A |vx − ux | The !ux of particles due to the !ux of liquid and their
@y
settling is given by
where F̂ G Net is the net !ux due to Geometric dispersion, 
Csi the mass of particle type i per volume of liquid, v the vx × A ×  × Csi
F̂ Bulk = : (6)
liquid velocity, u the gas velocity, vSettling; i the hindered (vy − vSettling; i ) × A ×  × Csi
6306 S. J. Neethling, J. J. Cilliers / Chemical Engineering Science 56 (2001) 6303–6311

2.3.4. Continuity equations for, at most, 3% of the volumetric feed to the cell. The
A continuity equation is required for the !ow of solids solids kinetics were taken from an industrial operation
within the Plateau borders. !oating a 1% chalcopyrite ore. Three size classes were
simulated: −38 m; +38–106 m and +106 m.
 · (F̂ Bulk + F̂ G Net + F̂ P Net )
The feed concentrations were as follows:
= Rate of addition per volume: (7)
The rate of addition of solids to the Plateau borders is the −38 m +38–106 m +106 m
same as the rate at which the particles become detached Desired material 4:2kg=m3 3:5kg=m3 1:7 kg=m3
from the lamellae. If it is assumed that all bubbles are Gangue 224 kg=m3 288 kg=m3 128 kg=m3
equally loaded, then the rate of particle addition to the Rate constant 0:85 min−1 0:75 min−1 0:34 min−1
Plateau borders is proportional to rate of loss of bubble
surface area due to coalescence and bursting. The system was simulated using cartesian co-ordinates.
Eq. (7) must be solved for each class of solids, i. The The simulated cell was 0:5 m long, 2 m wide and the froth
i equations are only indirectly interdependent in that the depth was 0:275 m for all the simulations given in this
settling velocity depends on the total concentration of paper. The permeability of the surface in every simulation
solids within the Plateau borders. was such that 80% of the gas entering the froth leaves
The addition of wash water does not appear directly due to bubble rupture at the surface. This is typical of
in the solids motion equations, as water addition does industrial rougher !otation cells.
not directly a ect solids mass continuity. However, it It is recognised that water addition has, potentially,
can dramatically a ect the liquid motion, which a ects a signi,cant e ect on froth stability. However, without
the solids motion through enhanced liquid transport and a suitable model or experimental data, this e ect was
dispersion. neglected.

2.4. The e<ect of particles on =uid drainage 3.2. The e<ect of water rate on performance

The particles a ect the !uid by changing the density, These simulations examine the e ect of wash water
viscosity and the volume entering the Plateau borders. addition rate on cell performance, when added uniformly
across the upper surface of the froth. The water rates ex-
• The density of the !uid is a ected by the solid concen- plored cover both net positive and negative bias. Neu-
tration: tral bias, when there is zero net water !ux across the

n  pulp-froth interface, is at a wash water addition rate of

f = Csi + × ; (8) 108 l=h.
0 The overall performance trends (Figs. 2 and 3) are as
would be expected; an increase in water addition results
where f is the !uid density and the liquid density.
in an increase in grade, but at the expense of recovery.
• The viscosity of the !uid is a ected by the solid con-
It should be noted that the purest product theoretically
centration [5]:
obtainable is 31.5%, taking into account the copper con-
f ≈  × (1 + 2:5 × (1 − ) + 14:1 × (1 − )2 tent of chalcopyrite and imperfect liberation. Thus, at the
highest water rate simulated, the product is very pure.
+ 0:00273 × e16:6(1−) ); (9) It can be seen that the curves do not show a discon-
where f is the !uid viscosity and  the liquid tinuity at neutral bias. What is slightly unexpected from
viscosity.

3. Simulation results

3.1. System details

For all simulations, the pulp phase was considered to

have a volume of 1:5 m3 and to be well mixed. The !ota-
tion kinetics in the pulp, and the pulp feed rate, were
taken to be the same for each simulation.
The feed rate to the cell was set at 150 l=min of a
45 wt% solids slurry, while the air rate was kept at
900 l=min for all simulations. The e ect of water addition
on the pulp residence time was neglected, as it accounts Fig. 2. Grade as a function of water rate.
 S. J. Neethling, J. J. Cilliers / Chemical Engineering Science 56 (2001) 6303–6311 6307

Fig. 3. Recovery as a function of water rate.

Fig. 5. Plateau border area (m2 ) for uniform wash water addition of
108 l=h.

Fig. 4. Grade-Recovery curve for variations in wash water rate.

these curves and the grade=recovery curve (Fig. 4) is that

the position of the point of zero net bias is not as obvious
Fig. 6. Grade (%Cu) for uniform wash water addition of 108 l=h.
from the grade and recovery curves as it would be, for
instance, in the results from a column cell !oating very
,ne particles.
The reasons for this are twofold. Firstly, the critical
water rate is not that at neutral bias, but the rate at which
the upward water rate equals the particle settling rate. At
wash water rates exceeding this, particles can only move
upwards by dispersion. For very ,ne particles, these rates
are approximately equal, but as particle size increases,
this di erence becomes more signi,cant. Since these sim-
ulations have a range of particle sizes, there is also a range
of water rates over which the dominant particle transport
mechanism changes from net liquid !ux to dispersion.
Secondly, in column !otation cells with deep froths,
there is no horizontal gas or liquid motion through most
of the froth height and capillary e ects will cause the
liquid !ow to become virtually uniform over the width of Fig. 7. Gangue concentration (kg=m3 ) for uniform wash water addition
the cell. The comparatively shallow froths simulated here of 108 l=h.
have a signi,cant horizontal component, which allows
local regions of upward water !ow even when the net Figs. 5 –8 are plots of some of the data produced for
water bias is downwards. This non-uniform water !ow the uniform wash water simulation at neutral bias. The
also obscures the location of the zero net water bias on gangue and desired material concentrations shown here
the grade-recovery curve. (and elsewhere in this paper) are for the intermediate
6308 S. J. Neethling, J. J. Cilliers / Chemical Engineering Science 56 (2001) 6303–6311

Fig. 9. Grade as a function of water position.

Fig. 8. Desired concentration (kg=m3 ) for uniform wash water addition

of 108 l=h.

sized particle class (+38–106 m). The x and y axes are

the grid numbers for the ,nite di erence solution. Since
the cell is 0:5 m wide, the grid numbers represent 0:25 cm.
It may appear anomalous that the largest Plateau bor-
ders are not found where the water is being added at the
top, but rather in a band lower down (see Fig. 5). The
main reason for this is that the wash water removes vir-
tually all the gangue and a large portion of the desired
material away from the upper reaches of the froth. Not
only does this remove the volume of the solids from these
Plateau borders, but also lowers the viscosity and allows Fig. 10. Recovery as a function of water position.
the water to drain away more easily.
The reason for the band of slightly wider Plateau bor- surface, or by a single row of sprays near the weir over-
ders across the middle of the froth is due, in part, to an !ow. In these simulations these strategies will be com-
accumulation of water where the entrained water meets pared, and the e ect of moving the single addition point
the wash water. This band also corresponds to the re- will be examined.
gion in which a sizeable amount of solids is again found For these simulations, a single water addition rate of
and where, therefore, the viscosity of the slurry increases 108 l=h was used, the neutral bias point in the uniform
dramatically. The fact that this band, while being quite water addition simulations.
broad in this region, intercepts the front wall at roughly Figs. 9 and 10 show the grade and recovery, respec-
the height of the weir is due to the water bias being about tively, as the water addition position is changed. What
neutral. is immediately evident is that the highest grade is not
The reason why the band of high desired material achieved when the water is added near the weir at the
(Fig. 8) content also corresponds to the region of high front of the cell, as is done conventionally, but occurs
water content is partly due to this being the lowest point when the water is added towards the back of the froth.
at which wash water has a direct e ect, but is also because Further, the highest grade does not correspond to the low-
the desired material is hindered from settling through the est recovery, as observed when the wash water rate was
region of high gangue concentration below it. changed (Fig. 4).
These simulations con,rm that the froth model pro- Two e ects cause a decrease in grade when water is
duces plausible results when the wash water rate is var- added near the front of the cell. Firstly, the majority of the
ied. The e ect of varying the position at which water is wash water !ows directly over the weir, observed by a
added is, however, less well understood and will be ex- decrease in the concentrate solids concentration. Further,
plored in the following section. since most of the wash water !ows straight over the weir,
the net bias is compromised and, as observed at low wash
3.3. Wash water addition position water rates, the grade decreases (see Fig. 16, as compared
to Fig. 12 for Plateau border size near the weir over!ow).
In conventional !otation cells the wash water is gener- Secondly, a very large fraction of the bubbles burst at
ally either added uniformly by sprays over the entire froth the top surface rather than over!ow the weir and release
 S. J. Neethling, J. J. Cilliers / Chemical Engineering Science 56 (2001) 6303–6311 6309

Fig. 12. Plateau border area (m2 ) for wash water addition 30% in
from back wall.

Fig. 11. Grade Recovery curve for changing water position superim-
posed on curve for varying water rate.

the attached, desired material into the Plateau borders.

Typical values, as in these simulations, are approximately
80%. Desired material released into the Plateau borders at
the top surface has diJculty reaching the weir if released Fig. 13. Grade (%Cu) for wash water addition 30% in from back
behind the wash water addition point. This explains the wall.
observed reduction in recovery as the wash water addition
point is moved closer to the weir. This decrease does
not occur close to the weir as most of the wash water
!ows over the weir and a ords the desired material an
opportunity to pass under it.
Fig. 11 shows the grade-recovery behaviour observed
when the water addition position is varied, superimposed
on the grade recovery curve obtained when the water
rate is changed but added uniformly. The results obtained
when the wash water addition point is changed lie on a
similar curve to the one obtained when the water rate is
changed.
Note in particular that water addition at the neutral
bias rate at a single position 30% from the rear of the
cell yields an equivalent recovery but a higher grade than Fig. 14. Gangue concentration (kg=m3 ) for wash water addition 30%
in from back wall.
when the wash water is added uniformly across the sur-
face.
The point for water addition near the weir lies virtually similar internal characteristics to the system in which the
on (though very slightly below) the grade-recovery curve water is added 60% of the distance in from the back wall.
obtained for uniform water addition. It corresponds to an The region directly below the water addition point is,
e ective uniform water rate that is much lower than was as would be expected, characterised by very large Plateau
added as a sizeable portion of the water added in this borders (see Fig. 12). What is also observed, though,
region goes straight over the weir without in!uencing the is that the Plateau borders on either side of this water
performance of the cell. addition point are not only much smaller than those near
The behaviour of the cell in which water is added 30% where the water is added, but also smaller than the Plateau
of the distance from the back wall (Figs. 12–15) shows borders elsewhere in the froth. The reason for this is
6310 S. J. Neethling, J. J. Cilliers / Chemical Engineering Science 56 (2001) 6303–6311

Fig. 15. Desired concentration (kg=m3 ) for wash water addition 30% Fig. 17. Grade (%Cu) for wash water addition near front wall.
in from back wall.

Fig. 18. Gangue concentration (kg=m3 ) for wash water addition near
Fig. 16. Plateau border area (m2 ) for wash water addition near front front wall.
wall.

that the wash water (as it is intended to do) removes

material from the Plateau borders in this vicinity. Since
the solids content of the Plateau borders and the viscosity
of the slurry within them are both strongly and directly
related, the lower solids content in this region results in
a lower viscosity. This, in turn means that the slurry in
this vicinity is able to !ow away more easily.
As the wash water moves away from the addition point,
its strong downward motion becomes di used over a
wider region. This gives an opportunity for some of the
material released behind the wash water addition point
to settle and=or disperse to a position from which it can
move towards the weir below the region of strongest wash Fig. 19. Desired concentration (kg=m3 ) for wash water addition near
front wall.
water e ect. This is characterised by the horizontal posi-
tion where the desired solids concentration increases and
then decreases with height (see Fig. 15). The concentra- is especially true of the wash water addition position,
tion of gangue shows no similar saddle point, seeing that where it was found that the simulations indicate that the
no gangue is release into the Plateau borders by due to best performance is achieved when the wash water is
coalescence and bursting (see Figs. 16 –19). added slightly further back in the cell. This is due to these
wash water addition positions being able remove a similar
fraction of the gangue, while at the same time washing
4. Conclusions out less desired material. The ine ectiveness of water
addition right at the front of the cell is because a sizeable
This paper shows that the behaviour of !otation froths fraction of the water !ows straight over the weir without
and the particles within them are not straightforward. This having any signi,cant in!uence on the cell performance.
 S. J. Neethling, J. J. Cilliers / Chemical Engineering Science 56 (2001) 6303–6311 6311

Even the uniform wash water addition, while showing Murphy, D. G., Zimmerman, W., & Woodburn, E. T. (1996).
the totally expected trends, has a slightly unexpected wa- Kinematic model of bubble motion in a !owing froth. Powder
ter distribution, with the wettest portion of the froth being Technology, 87, 3–12.
Neethling, S. J., & Cilliers, J. J. (1998). A visual kinematic model
found in the middle. of !owing foams incorporating coalescence. Powder Technology,
While the simulation study cannot eliminate the need 101, 249–256.
for experimental studies, these studies can be carried out Neethling, S. J., Cilliers, J. J., & Woodburn, E. T. (2000). The
far more quickly and inexpensively than a set of equiva- distribution of liquid in !owing foams. Chemical Engineering
lent experiments and could be used to restrict the experi- Science, 55, 4021–4028.
Verbist, G., Weaire, D., & Kraynik, A. M. (1996). The foam drainage
mental studies to a set of promising looking alternatives. equation. Journal of Physics Condensed Matter, 8, 3715–3731.

References

Coulson, J. M., & Richardson, J. F. (1993). Chemical Engineering,

Vol. 2, Oxford: Pergamon Press.