See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/248400810

A compartment model for the mass transfer inside a conventional

flotation cell

Article  in  International Journal of Mineral Processing · October 2005

DOI: 10.1016/j.minpro.2005.02.003

CITATIONS READS
27 288

1 author:

Orivaldo Normando Savassi

SGS GEOSOL
16 PUBLICATIONS   269 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

COMPREHENSIVE MODELLING OF FLOTATION View project

All content following this page was uploaded by Orivaldo Normando Savassi on 29 November 2017.

The user has requested enhancement of the downloaded file.

 Int. J. Miner. Process. 77 (2005) 65 – 79
www.elsevier.com/locate/ijminpro

A compartment model for the mass transfer

inside a conventional f lotation cell
O.N. Savassi*
Departamento de Engenharia de Minas, Universidade Federal de Minas Gerais, R. Espirito Santo 35, Belo Horizonte, CEP 30160-030, Brasil
Received 17 June 2004; received in revised form 7 February 2005; accepted 21 February 2005
Available online 12 May 2005

Abstract

A model is developed by taking into account the simultaneous mechanisms of true flotation and entrainment in a
conventional flotation cell. The total volume of the cell is divided into three compartments: pulp collection zone, pulp quiescent
zone and froth region, with the mechanisms being modeled as occurring at the same time but originating at different places: true
flotation from the collection zone and entrainment from the quiescent one. A particle is referred to as suspended in water or
attached to an air bubble, depending upon its original state before crossing the pulp–froth interface (whether or not it remains in
that state all the way to the concentrate launder). The model is obtained by solving a set of equations describing the mass
conservation of solids and water between adjacent compartments. The principal mass transfer factors are identified as: the
flotation rate constant, the mean residence time in the collection zone, the froth recovery of attached particles, the degree of
entrainment through the froth and the water recovery from the feed to the concentrate. The development presented here allows
the intricate nature of the mass transfer in a flotation cell to be reduced to one single equation, overcoming the need of numerical
methods for simulation purposes. Moreover, it is shown that reliable prediction of grade and recovery can be obtained without
detailed information on the pulp hydrodynamics or on any froth sub-process either than drainage, bubble bursting and bubble
coalescence.
D 2005 Elsevier B.V. All rights reserved.

Keywords: flotation modeling; mass transfer; true flotation; entrainment; pulp hydrodynamics

1. Introduction

1.1. Overview of the mass transfer inside a

conventional flotation cell

* Currently at MinnovEX Technologies, 1140 Sheppard Ave W.
# 6, Toronto, Canada M3K2A2.
E-mail address: savassio@minnovex.com.
0301-7516/$ - see front matter D 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.minpro.2005.02.003
It is well established that the pulp in a conventional
flotation cell must be thoroughly agitated in order to
promote solids suspension as well as effective bub-
66 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79
ble–particle collision (Degner, 1985). Excessive agita-
tion, on the other hand, not only wastes power but also
impairs metallurgical performance by increasing the
probability of bubble–particle detachment in the
impeller vicinity (Jameson, 1977; Jowett, 1980;
Ahmed and Jameson, 1989). At the proper range of
impeller speed, there is circulation of suspended par-
ticles through the pulp and the turbulence drops
swiftly away from the impeller blades (Fallenius,
1987; Gosman et al., 1992). The efficiency of bub-
ble–particle collision, however, is also known to
decrease with the turbulence in the pulp (Jameson,
1977; Dobby and Finch, 1987). Therefore, the effi-
ciency of bubble–particle collision in a conventional
cell is a function of spatial position, dropping from the
impeller blades towards the pulp–froth interface
(Inoue, 1984). Unfortunately, such a function cannot
be quantified without mapping the pulp hydrody-
namics for specific cell geometries and operating
conditions, which is not a trivial task.
The efficiency of bubble–particle attachment must
also be considered, as not all particles colliding
against a bubble will attach to it (Dobby and Finch,
1987; Crawford and Ralston, 1988). In this sense, as a
bubble travels through the impeller vicinity, it tends to
collect the most hydrophobic particles in its path, this
being the primary source of flotation selectivity. How-
ever, depending on the pulp chemical environment,
several physico-chemical processes may occur that
will affect the efficiency of bubble–particle attach-
ment, such as bageingQ of particle surfaces and
agglomeration between hydrophobic and hydrophilic
particles (Greene and Duke, 1962).
With increasing distance from the impeller blades,
there is a transition to a less turbulent or quiescent
zone, which is essential for the stability of the pulp–
froth interface. Cell manufacturers have recognised
that the quiescent zone, bbeing less turbulent, permits
the upward migration of mineral-laden air bubbles
with minimal opportunity for bubble–particle separa-
tion (Smith et al., 1982)Q. At the pulp–froth interface,
air bubbles ascending from the quiescent zone cause
water and suspended particles to become entrained in
the froth, irrespective of particle hydrophobicity
(Smith and Warren, 1989). Despite the fact that fine
valuable particles can also be recovered by entrain-
ment (Engelbrecht and Woodburn, 1975), this
mechanism is a drawback in terms of the selectivity
of separation, always resulting in lower concentrate
grades (Johnson, 1972; Trahar, 1981).
Particle detachment within the froth is caused pri-
marily by bubble bursting and bubble coalescence
(Feteris et al., 1987; Falutsu, 1994). The detached
particles become suspended in the voids between the
remaining air bubbles, together with the particles that
entered the froth by entrainment. A high proportion of
the particles suspended in the voids (entrained and
detached particles) are rejected to the pulp due to
water drainage. The suspended particles that survive
drainage and the ones still attached to the air bubbles
are transported to the concentrate launder by the
ascending motion of the froth (Moys, 1978; Yianatos
et al., 1988). Additional to bubble bursting, bubble
coalescence and water drainage, other sub-processes
have been postulated (without supporting experimen-
tal evidence) to take place within the froth, including
selective detachment of weak hydrophobic particles
that would be caused by friction with the water drain-
ing from between the bubbles.
The fact that increasing froth depth leads to higher
concentrate grade in a cleaner cell is sometimes
invoked as proof for the significance of selective
detachment. However, that fact can just as well be
explained by a simple sequence of events: (a) unse-
lective detachment of both weak and strong hydro-
phobic particles due to increased bubble bursting and
bubble coalescence in the froth; (b) drainage of the
detached particles back to the pulp; (c) faster re-
attachment of the strong hydrophobic particles in the
pulp; and (d) rejection of the particles that did not re-
attach to the tail of the cell, the majority of which
being weak hydrophobic. This not only explains the
increase in grade but also the drop in recovery.
Against the significance of selective detachment, it
can certainly be argued that most of the weakly
attached particles will detach in the turbulent impeller
vicinity, long before the aggregate reaches the pulp–
froth interface. In addition, a study of the forces acting
upon the aggregates in the froth shows that attachment
is about three orders of magnitude stronger than the
friction generated by water drainage (Falutsu, 1994).
Moreover, in one of the rare cases where the grade of
attached particles was measured directly, there was no
grade variation in going from the pulp to the froth or
within the froth itself (Falutsu and Dobby, 1992). This
clearly indicates that selective detachment was not
 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79
significant, otherwise an upgrading of the attached
particles should have been observed as the bubbles
ascended through the froth.
1.2. Modeling approach
The central assumption of the model is that true
flotation and entrainment are the main mass transfer
mechanisms occurring in the cell. Neither particle
ageing nor particle agglomeration will be considered
for model development, due to the practical difficulty
of measuring these phenomena in industrial cells.
Therefore, model application must be restricted to
flotation systems where particle hydrophobicity is
not a function of residence time and where the parti-
cles suspended in the pulp are reasonably dispersed
from each other.
Another key assumption is that all particles within
the same class exhibit the same behaviour. This means
that class boundaries should be defined according to
the physical properties that most strongly affect true
flotation and entrainment and also that there should be
no significant difference in hydrophobicity from one
particle to another within the same class. Neverthe-
less, the definition of class boundaries for any indus-
trial application will always depend on a compromise
between the cost of data acquisition and the accuracy
of model prediction; both of which increase in the
following order: (a) mineral classes; (b) mineral-by-
size classes (Trahar, 1981); and (c) mineral-by-size-
by-liberation classes (King, 1976).
Current attempts at applying computer fluid
dynamics, CFD, to investigate the relationship between
pulp hydrodynamics and the efficiency of bubble–par-
ticle collision are still beset by a large number of
unconfirmed assumptions. In this paper, the problem
will be simplified by assuming that there is a collection
zone around the impeller blades, wherein all particles
of the same class are collected at the same flotation rate
constant, irrespective of spatial position. It will also be
assumed that the volume between the collection zone
and the froth region constitutes a quiescent zone, where
the entrainment mechanism originates from and where
neither attachment nor detachment occurs at a signifi-
cant rate. The collection zone will be assumed to be
perfect mixed for water and for all suspended particles.
The quiescent zone, on the other hand, will be
assumed to be agitated enough for perfect mixing of
67
water and fine suspended particles, but not as much as
to cause re-circulation of bubble–particle aggregates.
In this sense, all particles exiting the collection zone
attached to an ascending air bubble should remain in
that state until they cross the pulp–froth interface. In
addition, the mass transfer through the froth will be
related to the flow of originally suspended and ori-
ginally attached particles entering that region, which
is the same as treating the entire froth as a black box.
Therefore no assumption is required on the extent of
any sub-processes that might be taking place within
the froth region.
In summary, the total volume of the cell is divided
into three compartments: pulp collection zone, pulp
quiescent zone and froth region. The main mass trans-
fer mechanisms are modeled as occurring at the same
time but originating at different places: true flotation
from the collection zone and entrainment from the
quiescent one. A particle is referred to as suspended
in water or attached to an air bubble, depending upon its
original state before crossing the pulp–froth interface
(same as treating the entire froth region as a black box).
Once the cell is assumed to be at steady state, it is not
necessary to take into account the path of individual
particles but rather the mass transfer between adjacent
compartments (that is, the recent origin of the parti-
cles). This approach allows the mass conservation of
solids and water inside the cell to be described by one
single equation, and thus overcomes the need of numer-
ical methods for simulation purposes. Dividing the
pulp into collection and quiescent zones, however,
brings about the problem of estimating the relative
volume of these compartments. This is a key issue
that will be revisited later in the paper.
2. Model development
2.1. Recovery by entrainment
Entrainment of water and suspended particles into
the froth is caused by air bubbles ascending through
the pulp–froth interface (Subrahmanyam and For-
ssberg, 1988; Laplante et al., 1989; Smith and Warren,
1989). This mechanism is independent of particle
surface properties and thus affects both the hydropho-
bic and hydrophilic particles suspended in the pulp
(Engelbrecht and Woodburn, 1975; Trahar, 1981).
68 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79

Only the fine particles (say below 60 Am) tend to be
recovered by entrainment, as coarser particles have
much higher drainage rates from the froth (Bisshop
and White, 1976). Since entrainment is caused by the
same bubbles that bring the attached particles to the
froth, it will always occur simultaneously to the true
flotation mechanism.
The motivation of this section of the model devel-
opment is to establish an expression for the recovery
of a target particle class by the entrainment mechan-
ism exclusively. True flotation will be approached
later in the paper.
Fig. 1 illustrates the transfer of water and sus-
pended particles (subscripts bwatQ and bsusQ, respec-
tively) in terms of the following flowrates: F: feed
throughput; Y: suspension due to impeller action,
which tends to be of the same magnitude of the
feed throughput (see Appendix A); IN: influx to the
froth; DR: drainage from the froth; X: re-circulation
by both impeller action and particle settling; C: con-
centrate; T: tail. It must be emphasized that the sub-
script bsusQ when applied to the pulp indicates
suspended particles, but the same subscript when
applied to the froth indicates particles that crossed
the pulp–froth interface by entrainment (whether or
not those particles remain suspended in water all the
way to the concentrate launder). As discussed above,
this approach eliminates the need of any assumption
regarding the extent of the sub-processes taking place
within the froth region, greatly simplifying model
development.
The relationship between the transfer of water and
suspended particles in a flotation cell is often quantified
by the degree of entrainment, which is defined by the
following ratio (Johnson, 1972; Trahar, 1981; Subrah-
manyam and Forssberg, 1988; Smith and Warren,
1989):
degree of entrainment
recovery of entrained particles to the concentrate
¼ :
recovery of water to the concentrate
ð1Þ
The advantage of using the degree of entrainment
for modeling purposes is that this ratio is not strongly
affected by variations in the water recovery for a broad
range of operating conditions (Johnson, 1972). The
degree of entrainment is often calculated on the basis
of the feed throughput, that is, in terms of the recov-
eries from the feed to the concentrate (Trahar, 1981):
Rsus
ENTFEED ¼ : ð2Þ
Rwat
From a phenomenological point of view, a more
rigorous approach is to calculate the degree of entrain-
ment on the basis of the influx of suspended particles to
the froth, that is, in terms of the recoveries from the
pulp to the concentrate (Johnson, 1972; Bisshop,
1974). The degree of entrainment through the froth
is thus defined as (the superscript bFRQ indicates the
froth region):
RFR
sus
ENT ¼ : ð3Þ
RFR
wat
According to the flowrates represented in Fig. 1,
the recovery of a target particle class by entrainment
and the recovery of water from the feed to the con-
centrate are given, respectively, by:
Csus
Rsus ¼ ð4Þ
Fsus

Cwat
Rwat ¼ : ð5Þ
froth C sus Fwat
C wat
DR sus DR wat Also according to the flowrates in Fig. 1, the froth
recovery of entrained particles and the froth recovery
IN wat IN sus of water (both from the pulp to the concentrate) are
X sus X wat quiescent given by, respectively:
Fsus Ywat Ysus Tsus Csus
RFR
sus ¼ ð6Þ
Fwat
collection
Twat IN sus
Cwat
Fig. 1. Transfer of water and suspended particles in a conventional RFR
wat ¼ : ð7Þ
cell. IN wat
 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79 69

Using the symbol x to represent pulp concentra- ENTFEED, as compared to a hydrophilic class of
tion in terms of mass of solids per unit mass of water, same size and density. In this sense, the use of the
gives: degree of entrainment through the froth is more con-
venient for modeling purposes, as this factor is not
Csus
xcon
sus ¼ ð8Þ directly affected by the true flotation mechanism.
Cwat
Combining Eqs. (2) and (13) to eliminate
ENTFEED gives:
Fsus
xfeed
sus ¼ ð9Þ
Fwat xQZ
sus
Rsus ¼ d ENT d Rwat : ð15Þ
xfeed
sus
Tsus
xtail
sus ¼ : ð10Þ In order to use Eq. (15) for simulation purposes, it
Twat
is still necessary to know the effect of the operating
According to the bubble swarm theory proposed by conditions upon the concentration of suspended par-
Smith and Warren (1989), the entrainment mechanism ticles beneath the pulp–froth interface, a measurement
causes water and suspended particles to enter the froth that is rarely available. Nevertheless, the regime in the
in the same proportion as they exist in the quiescent pulp of a conventional flotation cell can safely be
zone (superscript bQZQ), that is: assumed to be perfect mixing for water and fine
IN sus suspended particles (Johnson, 1972). According to
¼ xQZ
sus : ð11Þ that assumption, the concentration of fine suspended
IN wat
particles should be the same at every point of the pulp
Combining Eqs. (2), (4), (5), (8) and (9) gives: region, including the quiescent zone and the tailing
stream outlet, that is:
xcon
sus
ENTFEED ¼ : ð12Þ
xfeed
sus xQZ tail
sus ¼ xsus : ð16Þ
Combining Eqs. (3), (6-8) and (11) gives: Combining Eqs. (15) and (16) allows the recovery
xcon by entrainment to be calculated by:
sus
ENT ¼ : ð13Þ
xQZ
sus xtail
sus
Rsus ¼ d ENT d Rwat : ð17Þ
Eqs. (12) and (13) demonstrate that it is not neces- xfeed
sus
sary to determine the influx of suspended particles to Note that Eqs. (16) and (17) are strictly valid for
the froth in order to estimate the degree of entrain- fine particles only, in view of the difficulty of keeping
ment. Rather it is only necessary to measure the coarse particles suspended away from the impeller
concentration of originally suspended particles in blades. Nevertheless, applying those equations to the
the feed, pulp and concentrate. Moreover, the relation- whole particle range is not likely to introduce any
ship between the two forms of expressing the degree appreciable error in the model, since neither the recov-
of entrainment is given by combining Eqs. (12) and ery by entrainment nor the degree of entrainment is
(13) as in: relevant to the overall recovery of coarse particles. In
xQZ this sense, both sides of Eq. (17) tend to zero for
sus
ENTFEED ¼ d ENT : ð14Þ coarse particles, irrespective of any concentration
xfeed
sus aspect.
According to Johnson (1972), particles of similar
size and density exhibit similar drainage rates and thus 2.2. Recovery by true flotation
similar degree of entrainment through the froth, ENT.
However, Eq. (14) demonstrates that if a hydrophobic In a conventional flotation cell, bubble–particle
particle class has a lower concentration in the pulp due collision occurs most effectively in the turbulent
to the recovery by true flotation, those particles will eddies generated by the impeller blades, where sus-
also have a lower degree of entrainment from the feed, pended particles of a broad size range can be accel-
70 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79

erated towards the air bubbles (Degner, 1985; Fall- pulp indicates particles attached to air bubbles, but the
enius, 1987). However, even in the turbulent impeller same subscript when applied to the froth indicates
vicinity, particles of too low inertia (below 10 Am, particles originally attached to bubbles before enter-
say) cannot be easily stricken out of the pulp stream- ing the froth (whether or not the particles remain in
lines, and thus tend to slip around the bubbles with that state all the way to the concentrate launder).
very low chances of collision (Dobby and Finch, According to Fig. 2 the recovery of a particle class
1987). This is the primary reason for the drop in by true flotation is given by:
true flotation recovery typically observed in the ultra-
Catt
fine range. If a particle collides successfully, it will Ratt ¼ : ð18Þ
Fsus
slide on the bubble surface due to the hydrodynamic
drag of the streamlines. Then, if the particle is suffi- The froth recovery of attached particles, which is
ciently hydrophobic, a three-phase contact line (solid, an indicator of the mass pull in the cell, is defined as:
gas and liquid) will form along its contour, resulting in
Catt
attachment (Schulze, 1984). The attached particle will RFR
att ¼ : ð19Þ
IN att
be pushed by the pulp streamlines until it is hidden
behind the bubble, allowing several other particles to The recovery in the collection zone (CZ) is calcu-
attach in the same manner. However, if attachment lated on the basis of the total solids influx to that zone,
does not occur during the sliding time, the particle is that is, the particles in the feed stream as well as those
simply swept off the bubble. The bubble–particle being re-circulated away from the quiescent zone:
aggregates that are created in the impeller vicinity
IN att
ascend to the froth from where only a fraction is RCZ
att ¼ : ð20Þ
Fsus þ Xsus
effectively transported to the concentrate launder
due to bubble bursting and bubble coalescence The mean residence time of pulp in the collection
(Feteris et al., 1987; Falutsu and Dobby, 1992). zone is defined as:
The motivation of this section of the model devel- volume of the collection zone
opment is to establish an expression for the recovery sCZ ¼ :
total volumetric flowrate entering that zone
by true flotation, accounting for both the kinetics of
ð21Þ
collection in the pulp and the efficiency of transport
through the froth. This must be done in a manner that Given the density of each particle class suspended
allows model calibration from data that is readily in the pulp, q, and the volume of the collection zone,
available in the plant. VCZ , the mean residence time in that zone can be
The transfer of a target particle class by the true calculated by:
flotation mechanism is illustrated in Fig. 2. The flow-
rate DB represents the drop-back of attached particles V CZ
sCZ ¼   : ð22Þ
from the froth due to detachment followed by drai- P Fsus þ Xsus
þ Fwat þ Xwat
nage. Note that the subscript battQ when applied to the q
If the aeration in the impeller vicinity is sufficiently
froth C att high to prevent competition for attachment on the
bubbles, the kinetics of true flotation can be described
DBatt by a first-order differential equation of the type:

dxCZ
quiescent
IN att  ¼ k CZ d xCZ : ð23Þ
X sus dt
Fsus Tsus The flotation rate constant, kCZ , in Eq. (23) is a
collection function of the probabilities of bubble–particle colli-
sion, attachment and detachment as well as the bubble
Fig. 2. Transfer of attached particles in a conventional cell. superficial area flux in the collection zone (Dobby and
 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79
Finch, 1987). Assuming perfect mixing and integrat-
ing gives the recovery of attached particles as:
k CZ d sCZ
RCZ
att ¼ : ð24Þ
1 þ k CZ d sCZ
Unfortunately, the lack of detailed information
about the pulp hydrodynamics in most industrial
applications precludes the use of Eqs. (20), (22) and
(24) for simulation purposes, as the terms X sus and
VCZ cannot be estimated without a map of the velocity
vector for every particle class suspended in the pulp as
a function of the impeller speed. Therefore, further
simplifications are necessary to develop a model that
is dependent only on data normally available in the
plant. To that end, let us define an apparent recovery
of attached particles in the collection zone in terms of
the net flowrate of suspended particles between quies-
cent and collection zones, as in:
IN att
4RCZ
att ¼ : ð25Þ
Fsus þ Xsus  Ysus
Let us also define an apparent mean residence time
in the collection zone, based both on the feed through-
put and an arbitrary value for the volume of that zone,
as in:
4V CZ
4sCZ ¼   : ð26Þ
P Fsus
þ Fwat
q origin or hydrophobicity, any individual particle sus-
Finally, let us define an apparent flotation rate
constant in the collection zone, according to:
4 CZ 4k CZ d 4sCZ
Ratt ¼ : ð27Þ
1 þ 4k CZ d 4sCZ
Obviously, the simplifications introduced above
constitute a mathematical artefact, which is needed
to overcome the lack of detailed information about
the pulp hydrodynamics. In this sense, it is essential
to investigate the impact of these simplifications
upon the reliability of model prediction. This is the
motivation for the calculations presented in Appen-
dix A, which demonstrate that a model based upon
Eqs. (25)–(27) has the same predictive power as one
based upon Eqs. (20), (22) and (24) over a broad
71
range of variation in the operating conditions. More-
over, any arbitrary value can be used to represent the
volume of the collection zone for calibration purposes,
as long as that value is kept fixed during simulation.
2.3. Overall recovery
There is a strong interconnection between true
flotation and entrainment in a conventional flotation
cell: true flotation lowers the concentration of hydro-
phobic particles suspended in the pulp and thus
reduces the chances of those particles becoming
entrained in the froth, while water drainage causes
entrained particles to drop from the froth for another
chance of attachment in the collection zone, depend-
ing upon particle hydrophobicity. The motivation of
this section is thus to develop an expression for the
overall recovery of a particle class accounting for the
simultaneous contribution of true flotation and
entrainment mechanisms.
Fig. 3 illustrates the mass transfer in a conventional
cell by both true flotation and entrainment (the sub-
script bovrQ indicates the overall outcome of these two
mechanisms). Note that any individual particle sus-
pended in the quiescent zone have three possible
recent origins: rejection from the froth due to water
drainage only, DR sus; rejection from the froth due to
detachment followed by water drainage, DB att; and
suspension from the bottom of the cell due to impeller
action, Y sus. Nevertheless, irrespective of their recent
pended in the quiescent zone can be either transferred
to the froth by entrainment, IN sus, or to the collection
zone for a chance of collision against an air bubble,
X sus.
froth C ovr
DR sus DR wat DBatt
C wat
IN wat IN sus
IN att
X sus X wat quiescent
Fsus Ywat Ysus Tsus
Fwat
collection
Twat
Fig. 3. Mass transfer in a conventional cell by both true flotation and
entrainment.
72 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79

According to the flowrates represented in Fig. 3, ble recovery by entrainment (Shaning Yu, 1985; Finch
the overall recovery of a particle class is given by: and Dobby, 1990). Finally, the overall recovery of a
particle class in a conventional cell, including the
Covr simultaneous contribution by true flotation and
Rovr ¼ : ð28Þ
Fsus entrainment, is given by combining Eqs. (27), (35)–
(37) as in:
The overall mass balance of solids and water
around the cell is expressed by:
4k CZ d4sCZ d RFR
att d ð1  Rwat Þ þ ENT d Rwat
Rovr ¼ :
Fsus ¼ Covr þ Tsus ð29Þ ð1 þ 4k CZ d 4sCZ d RFRatt Þd ð1  Rwat Þ þ ENT d Rwat
ð38Þ
Fwat ¼ Cwat þ Twat : ð30Þ
Eq. (38) expresses the compartment model in
The mass balance of solids entering the froth by terms of the flotation rate constant, the mean residence
entrainment or true flotation is expressed by: time in the collection zone, the froth recovery of
IN sus ¼ Csus þ DRsus ð31Þ attached particles, the degree of entrainment through
the froth and the water recovery from the feed to the
IN att ¼ Catt þ DBatt : ð32Þ concentrate. It is important to note that the terms
accounting for the true flotation mechanism always
The mass balance of solids around the quiescent appear together in a product. This greatly simplifies
zone is expressed by: model calibration, as discussed further in the next
Ysus þ DRsus þ DBsus ¼ Xsus þ IN sus : ð33Þ section.

The solids flowrate in the concentrate consists of

particles that were originally suspended in the pulp or 3. Model calibration
originally attached to an air bubble before entering the
froth, that is: The basic dataset for calibration of the compart-
ment model is the overall recovery–R ovr–of the main
Covr ¼ Csus þ Catt : ð34Þ mineral species in the system. The use of metal spe-
Combining Eqs. (4), (18), (28) and (34), gives: cies must be avoided when the same metal is part of
different minerals (for instance, iron in a copper ore
Rovr ¼ Rsus þ Ratt : ð35Þ can be part of chalcopyrite, pyrite and hematite, each
The recovery by entrainment has an additional mineral having a different flotation rate). The dataset
relationship with the overall recovery in the cell, can also be sub-divided into more particle classes,
which is given by combining Eqs. (5), (9), (10), allowing the effect of size and liberation to be
(17), (28)–(30), as in: included in the model. As discussed earlier in the
text, however, the decision on whether or not obtain
1  Rovr
Rsus ¼ d ENT d Rwat : ð36Þ such detailed information will always depend on a
1  Rwat compromise between model accuracy and experimen-
The recovery by true flotation can be expressed in tal costs.
terms of the collection zone recovery and the froth For more reliable results, it is strongly recom-
recovery of attached particles, by combining Eqs. (4), mended to mass balance the experimental data prior
(18), (19), (25), (31)–(33), as in: to model calibration. Including the solids percentage
of the streams around the cell in the mass balance
4RCZ FR
att d Ratt d ð1  Rsus Þ allows the water recovery–R wat–to be promptly esti-
Ratt ¼ : ð37Þ mated. If the ore contains a hydrophilic mineral spe-
4RCZ FR
att d Ratt þ 1  4Ratt
CZ
cies (for instance, non-sulphide gangue in a sulphide
Note that Eq. (37) is similar to the one proposed in ore), that species can be used as a tracer for the
the literature for flotation columns assuming negligi- estimation of the degree of entrainment—ENT.
 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79 73

Rewriting Eq. (36) for the hydrophilic tracer, with In systems where selective detachment is not sig-
R ovr = R sus = R hydrophilic, gives: nificant (which, as discussed earlier in the text, should
be the most common case), hydrophobic minerals
1  Rwat Rhydrophilic with similar size distribution and density will have
ENT ¼ d : ð39Þ FR
Rwat 1  Rhydrophilic similar froth recovery of attached particles—R att .
Therefore, if the ore contains a fast floating mineral
Ideally, the hydrophilic tracer should be fairly species whose overall recovery is dominated by true
liberated and constitute a significant fraction of the flotation, that species can be used as a tracer for the
ore (more than 10% in mass, say). Any mineral that estimation of the froth recovery of all other minerals
has similar size distribution and density of the hydro- in the system. Combining Eqs. (24) and (37) and
philic tracer will have similar degree of entrainment. rewriting for the hydrophobic tracer, with R ovr =
Nevertheless, even when those properties are differ- R att = R hydrophobic, gives:
ent, assigning the same degree of entrainment to
hydrophobic and hydrophilic minerals will not Rhydrophobic
RFR
att ¼ CZ
  ð40Þ
cause significant error in the model predictions, pro- 4khydrophobic d4sCZ d 1  Rhydrophobic
vided that the fine hydrophobic particles have a
reasonably high true flotation recovery. This is Hydrophobic and hydrophilic tracers will co-exist
demonstrated by Eq. (17), which shows that a mineral in flotation systems where the chemical environment
whose concentration in the tailing is much lower than is reasonably selective. The case where neither tracer
that in the feed will have insignificant recovery by is present is not suitable for separation by flotation. If
entrainment, irrespective of the water recovery or the only the hydrophobic tracer is present (for example,
degree of entrainment through the froth. There are silica in reverse flotation of an iron ore), the degree of
cases, however, where the ore contains no hydrophilic entrainment can still be calculated from the overall
species, which precludes the use of Eq. (39) for model recovery of one of the slow floating species in the
calibration. These cases will be discussed later in this system (iron oxide, in the same example above), by
section. combining Eqs. (38) and (40) as in:
Estimation of the flotation rate constant by batch
testwork is a common practice in the industry. To 1  Rwat
ENT ¼
ensure that the experimental results represent the Rwat
!
kinetics of true flotation in the pulp–k CZ –, it is essen- Rslow 4k CZ Rhydrophobic
tial to use fast scraping and shallow froth depth. A d  CZ slow d
1 Rslow 4khydrophobic 1  Rhydrophobic
rigorous methodology for extracting kinetic para-
meters from batch testwork is described elsewhere ð41Þ
(MinnovEX Technologies, 2004). The methodology
applies a phenomenological approach to decouple the Where separation is very difficult (for instance,
recovery by true flotation in the batch cell from that galena-spharelite ores), the use of depressants tends
by entrainment as well as to account for the differ- to slow down the flotation rate of all minerals and thus
ences in the pulp hydrodynamics mixing regime in precludes the use of Eq. (40) for calibration purposes.
relation to the industrial cells. In these cases, the froth recovery can be estimated
As demonstrated in Appendix A, using an arbitrary from the recovery of the valuable mineral, R valuable, by
volume to represent the collection zone has no effect combining Eqs. (38) and (39) as in:
upon the predictive power of the compartment model. 1
RFR
att ¼
Therefore, for the sake of simplicity, the same volume 4k CZ
d4sCZ
must be chosen for all particles in the cell, irrespective  valuable 
Rvaluable Rhydrophilic
of size, liberation or hydrophobicity. In this sense the d  : ð42Þ
1  Rvaluable 1  Rhydrophilic
residence time in the collection zone–sCZ –becomes a
function of the total volumetric flow of the feed, Note that the calibration of the froth recovery as per
exclusively. Eqs. (40) or (42) is based upon apparent values for
74 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79
both the flotation rate constant and the mean residence
time in the collection zone. In this sense, the froth
recovery is being calibrated as a reconciliation factor
between those apparent values and the recovery of
the hydrophobic species measured experimentally.
Obviously such reconciliation factor (which will be
referred to as RF for the rest of the text) is not neces-
sarily the froth recovery that would be measured
directly in the industrial cell. It must be emphasized,
however, that the experimental measurement is not
always feasible due to the high error propagation aris-
ing from the fact that the particles attached to the
bubbles constitute a minute fraction of the total amount
of solids in the pulp (Shaning Yu, 1985). Nevertheless,
since the terms accounting for the true flotation
mechanism in Eq. (38) always appear together in a
product, the question is not so much on obtaining the
absolute value of each individual term, but rather to
estimate the relative variation with the operating con-
ditions. As demonstrated in Appendix A, the reconci-
liation factor RF and the froth recovery of attached
particles may differ in absolute value but will have
similar variation with the operating conditions, pro-
vided that: (a) a fixed value is chosen to represent the
collection zone volume; and (b) the flotation rate
constant in the pulp of the batch cell has the same
relative variation with the operating conditions as the
rate in the industrial cell.
4. Industrial application
The compartment model is the calculation platform
for FLEET (Flotation Economic Evaluation Tool),
developed by MinnovEX with support by seven
large mining firms. To date, the simulator has been
applied with success at several flotation plants,
encompassing copper, gold, nickel, iron, lead, zinc
and coal operations. This section includes examples
of industrial application of FLEET to confirm the
validity of the compartment model.
A total of 10 surveys were conducted in a gold plant
over a period of fixed ore type characteristics, as indi-
cated by the MFT testwork (MinnovEX Technologies,
2004). As usual in industrial gold operations, the
flotation circuit configuration was highly complex,
aiming at maximum recovery. The circuit consisted
of conventional cells, exclusively. Cell dimensions at
some stages were close to 30 times smaller than the
roughers. Gold was balanced as such, since it was not
part of any of other mineral species. Copper, iron and
sulphur assays were converted to chalcopyrite and
pyrite assays, while the non-sulphide gangue was
calculated as the remainder. After mass balancing
each one of the 10 surveys individually, the weighted
averaged assays were used to represent the base case
of plant metallurgical performance. The compartment
model was then calibrated to each stage of the circuit
using Eqs. (39) and (40), with non-sulphide gangue
and gold used as the hydrophilic and hydrophobic
tracers, respectively. As shown in Fig. 4, a strong cor-
relation between base case and calibration results was
obtained for all species. Note that each point in the
plot represents a different assay in a different stream.
Therefore the model is able to describe the perfor-
mance the entire plant with high level of accuracy.
Large copper plants are known for wide variation in
feed characteristics, making it impossible to propose a
base case to represent the average metallurgical perfor-
mance. Fig. 5 shows calibration results for one such
case, including eight plant surveys completed with
MFT testwork on the rougher feed. The feed con-
tained several copper bearing minerals, whose relative
proportion was measured by point counting. The
lumped copper sulphide species and the non-sulphide
gangue were used as the hydrophobic and hydrophilic
tracers, respectively. Fixed values for ENT and for the
R wat/RF ratio were then calibrated to each stage of
1000
100
assay model
10
Gold
1 Pyrite
Chalco
NSG
0.1
0.1 1 10 100 1000
assay base case
Fig. 4. Calibration results for a complex gold plant.
 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79 75

60
have since been investigated by simulation include:
feed throughput and P80, circuit configuration and
50 mass pull at different stages. Changes in the pulp kine-
tics due to ore type or reagent dosages can be predicted
40 by MFT testwork conducted on drill core samples,
which is the basis for production planning projects.
assay model

Benchmark of an iron ore plant was conducted in

30
20
10
0 cuit and then required to predict the performance of
0 10 20 30 40
assay individual survey
Fig. 5. Calibration results for a large copper plant.
conventional cells, with RF varying between surveys.
This approach allows the intervention of the control
room upon the mass pull in the conventional cells to
be included in the model (details of the calibration of
column and regrind models are beyond the scope of
this paper). The results in Fig. 5, more than 250 assays
all up, demonstrate that FLEET is able to simulate the
plant performance even in this extreme case of varia-
tion in feed characteristics. Some of the aspects that
the midst of plant lay-out modification, resulting in
the rare case shown in Fig. 6 where parallel lines of
industrial cells were operated under different circuit
CuSulf configuration (at that time line 1 was yet to be chan-
Pyrite ged). This offered a unique opportunity for model
NSG validation, with FLEET being calibrated for one cir-
50 60 the other. To that end, three plant surveys were con-
ducted for a fixed ore type, including MFT testwork
on the feed to the circuit. The ore was modelled as
consisting of two species only, silica and iron oxide,
which were used as the hydrophobic tracer and the
slow flotation species in Eqs. (40) and (41), respec-
tively. Calibration of line 1 provided the value of ENT
and the Rwat/RF ratio in each stage of the circuit. The
performance of the other parallel lines could then be
simulated and compared against the results obtained
experimentally. As shown by the results in Fig. 6,
FLEET is able to predict the impact of circuit recon-
figuration with a high level of accuracy.

LINE 1
MODEL CALIBRATION _ Fe CON

SiO2 % Fe Rec
FEED survey
exp cal exp cal
Fe CON
A 3.0 3.0 89 90

B 4.9 4.7 90 86

C 3.5 3.7 88 93
SiO2 TAIL

LINES 2 to 4
MODEL PREDICTION _ Fe CON

SiO2 % Fe Rec
FEED survey
exp prd exp prd

A 2.1 2.0 88 89

B 2.4 4
2.4 85 82

C 2.0 2.3 87 91
SiO2 TAIL Fe CON

Fig. 6. Reconfiguration of an iron ore flotation circuit.

76 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79

5. Conclusions sary for the development have no significant effect on

A compartment model, accounting for the simulta-
neous contribution of true flotation and entrainment to
the performance of a conventional cell, has been
developed. The principal factors representing the
mass transfer inside the cell have been identified as:
the flotation rate constant, the mean residence time in
the collection zone, the froth recovery of attached
particles, the degree of entrainment through the froth
and the water recovery from the feed to the concen-
trate. It is demonstrated that the simplifications neces-
model predictive power, as compared to using a more
detailed description of the pulp hydrodynamics.
Model calibration is based upon batch testwork
along with the assays and solids percentages of dif-
ferent streams in the plant. No special measurements
are required. The compartment model allows the
metallurgical performance of conventional cells to
be predicted over a broad range of operating condi-
tions. Industrial application includes circuit optimiza-
tion and production planning projects based on drill
core samples.

Appendix A

A number of simplifications regarding the pulp hydrodynamics were necessary for the development of the
compartment model. The motivation of the calculations presented below is to investigate the impact of those
simplifications upon the reliability of model predictions. The mass transfer equation that corresponds to Eq. (38)
without the simplifications is given by combining Eqs. (4), (18–20), (24), (31-33), (35) and (36) as in:
k CZ d sCZ d RFR
att d ð1 þ Ysus =Fovr Þd ð1  Rwat Þ þ ENT d Rwat
Rovr ¼ ð43Þ
ð1 þ k CZ d sCZ d RFR
att Þd ð1  Rwat Þ þ ENT d Rwat

Note that Eq. (43) is based on the bactualQ flotation rate constant and mean residence time in the collection zone,
whereas Eq. (38) is based on apparent values for those variables. In order to estimate the flowrate of the solids
being suspended from the bottom of the cell due to the impeller action, Y sus, it will be assumed that the cross-
section area of the cell in the impeller vicinity is equally divided between suspended particles moving up and
downwards. This leads to the following equation:

1
Ysus ¼ d average velocity of the pulp exiting the collection zoned cell cross  section area
2
d pulp densityd mass of particle class per unit mass of pulp ð44Þ
The average velocity of the pulp exiting the collection zone is a function of impeller speed and cell geometry
mainly. Other terms in Eq. (44) depend upon particle concentration, as in:
P CZ
1þ x
pulp density ¼ P  CZsus  ð45Þ
1þ xsus =q

xCZ
sus
mass of particle class per unit mass of pulp ¼ P ð46Þ
1þ xCZ
sus
Assuming perfect mixing in the collection zone gives the concentration of a target particle class as in:

Fsus d ð1  Rovr Þ
xCZ
sus ¼ ð47Þ
Fwat d ð1  Rwat Þ
Note that, as a consequence of Eqs. (45) and (46) including a sum throughout the particle classes, the estimation
of the overall recovery by Eq. (43) is numerical in nature, depending on initial guesses for the concentration of
 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79 77

each class suspended in the collection zone, which is then updated by Eq. (47). This poses a major convergence
problem when using those equations for simulation purposes, a problem that increases in complexity with the
number of cells and circulating streams in the flotation circuit (loops within loops).
Eqs. (24), (43)–(47) will be referred to as bmodel 1Q, while bmodel 2Q will refer to the compartment model in
conjunction with the simplified residence time equation, that is, Eqs. (38) and (26), respectively. The impact of the
simplifications introduced in model 2 can thus be quantified by comparing its predictions against those obtained by
model 1. The calculations reported here consist of the following steps (note that, for the sake of simplicity, only one
particle class has been included in this exercise):

(A) Using model 1 to generate artificial data at a given operating condition: feed throughput = 500 tph; solids
percentage in the feed = 30%, water recovery = 15%; degree of entrainment = 0.5; froth recovery = 50%, cell
cross-section area = 20 m2; bactualQ collection zone volume = 25 m3; bactualQ flotation rate constant = 0.4
min1; and average velocity of the pulp exiting the collection zone = 3.8 cm/s (the settling velocity for a 500
Am particle with 4 g/cm3 density given by Stokes’ law). Overall recovery estimated by model 1 for this set of
parameters is 26.4%, while Y sus is 464 tph, same magnitude of the feed flowrate.
(B) Calibration of model 2: An arbitrary value of either 25 or 50 m3 is chosen to represent the volume of the
collection zone, which is kept fixed within each set of simulations. Given the apparent values for both the

60
RF = 93.5 RF = 46.7

40 VCZ = 25 m3 VCZ = 50 m3

R ovr RF = 58.4 RF = 29.2

20
RF = 23.4 RF = 11.7

model 1 model 2 calibration point

0
100 300 500 700 900 300 500 700 900
feed throughput (tph)
60
model 1 model 2 calibration point
RF = 93.5 RF = 46.7
40 VCZ = 25 m3 VCZ = 50 m3
R ovr RF = 58.4 RF = 29.2
20
RF = 23.4 RF = 11.7

0
0.0 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
flotation rate constant (min-1)
Fig. 7. Variation of the overall recovery in a flotation cell with the feed throughput, pulp kinetics and mass pull, as predicted by two different
models.
78 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79

flotation rate constant and the mean residence time, the reconciliation factor RF is calibrated to make model 2
match the overall recovery estimated by model 1 under the same conditions.
(C) Changing the operating conditions: When investigating the effect of the feed throughput, the flotation rate
constant is kept fixed at 0.4 min1, the same value used to generate the artificial data. When investigating
the effect of the flotation rate constant, the feed throughput is kept fixed at 500 tph. Within each set of
simulations, the froth recovery in model 1 is varied to either 20 or 80%, while the RF value calibrated in step
2 is changed in the exact same proportion (so that there is only one calibration point per each set). The overall
recovery in the new operating condition is then estimated by both models and compared.

Fig. 7 summarises the results of the calculations. The strong correlation between the predictions by the two
model in each set of simulations demonstrate that: (a) the simplifications introduced in the compartment model
have no significant effect upon its predictive power, as compared to using a more complex description of the
pulp hydrodynamics; (b) any arbitrary value can be used to represent the volume of the collection zone for
calibration purposes, as long as that value is kept fixed upon simulation; c) the reconciliation factor RF and
the froth recovery of attached particles may differ in absolute value but will have similar variation with the
operating conditions, provided that the same occurs with the flotation rate constant between the batch and
the industrial cell. Therefore, the compartment model allows reliable prediction of metallurgical performance
over a broad range of operating conditions, with no special measurements being required for calibration
purposes.

References froth depth in flotation. International Journal of Mineral Proces-

Ahmed, N., Jameson, G.J., 1989. In: Laskowski, J.S. (Ed.), Flota-
tion Kinetics, Mineral Processing and Extractive Metallurgy
Review, vol. 5, pp. 77 – 99.
Bisshop, J.P., 1974. A study of particle entrainment in flotation
froths. B.Sc. thesis, University of Queensland.
Bisshop, J.P., White, M.E., 1976. Study of particle entrainment in
flotation froths. Transactions of the Institution of Mining and
Metallurgy, Section C: Mineral Processing and Extractive
Metallurgy 85, 191 – 194.
Crawford, R., Ralston, J., 1988. The influence of particle size and
contact angle in mineral flotation. International Journal of
Mineral Processing 23, 1 – 24.
Degner, V.R., 1985. Hydrodynamic influences in flotation machine
design. SME-AIME Annual General Meeting, Preprint 85-41,
New York.
Dobby, G.S., Finch, J.A., 1987. Particle size dependence in flotation
derived from a fundamental model of the capture process.
International Journal of Mineral Processing 21, 241 – 260.
Engelbrecht, J.A., Woodburn, E.T., 1975. The effects of froth height,
aeration rate and gas precipitation on flotation. Journal of the
South African Institute of Mining and Metallurgy, 125 – 132.
Fallenius, K., 1987. Turbulence in flotation cells. International
Journal of Mineral Processing 21, 1 – 23.
Falutsu, M., 1994. Column flotation froth characteristics—stability
of the bubble particle system. International Journal of Mineral
Processing 40, 225 – 243.
Falutsu, M., Dobby, G.S., 1992. Froth performance in conventional
sized columns. Minerals Engineering 5, 1207 – 1223.
Feteris, S.M., Frew, J.A., Jowett, A., 1987. Modelling the effect of
sing 20, 121 – 135.
Finch, J.A., Dobby, G., 1990. Column Flotation. Pergamon Press,
England. 180 pp.
Gosman, A.D., Lekakou, C., Politis, S., 1992. Multidimensional
modeling of turbulent two-phase flows in stirred vessels. AIChE
Journal 38, 1947 – 1956.
Greene, E.W., Duke, J.B., 1962. Selective froth flotation of
ultrafine minerals or slimes. Transactions of the American
Institute of Mining, Metallurgical and Petroleum Engineers
223, 389 – 395.
Inoue, T., 1984. Flotation process modelling. In: Jones, M.H.,
Woodcock, J.T. (Eds.), Principles of Mineral Flotation. The
Australasian Inst. Min. Metall, Parkville, Victoria, pp. 255 – 271.
Jameson, G.J., 1977. Physical factors affecting recovery rates in
flotation. Minerals. Science and Engineering 9 (3), 103 – 118.
Johnson, N.W., 1972. The flotation behaviour of some chalcopyrite
ores. PhD thesis, University of Queensland.
Jowett, A., 1980. Formation and disruption of particle–bubble
aggregates in flotation in fine particles processing. In: Soma-
sundaran, P. (Ed.), Proc. Int. Symp. On Fine Particles, Las
Vegas. AIME, N.Y., pp. 720 – 754.
King, R.P., 1976. The use of simulation in the design and modifica-
tion of flotation plants. In: Fuerstenau, M.C. (Ed.), Flotation,
A.M. Gaudin Memorial Volume, AIME, vol. 2, pp. 937 – 962.
Laplante, A.R., Kaya, M., Smith, H.W., 1989. The effect of froth on
flotation kinetics—a mass transfer approach. In: Laskowsky,
J.S. (Ed.), Mineral Processing and Extrative Metallurgy Review,
vol. 5, pp. 147 – 168.
MinnovEX Technologies, 2004. MinnovEX Flotation Test: Stan-
dard Procedure and Parameter Extraction. Available at:
www.minnovex.com. 31 pp.
 O.N. Savassi / Int. J. Miner. Process. 77 (2005) 65–79
Moys, M.H., 1978. A study of a plug flow model for flotation
froth behaviour. International Journal of Mineral Processing 5,
221 – 238.
Schulze, H.J., 1984. Physico-Chemical Elementary Processes in
Flotation. Elsevier, Amsterdam. 348 pp.
Shaning Yu, 1985. Particle collection in a flotation column. Master’s
thesis, McGill University, Montreal Canada.
Smith, P.G., Warren, L.J., 1989. Entrainment of particles into flota-
tion froths. In: Laskowski, J.S. (Ed.), Mineral Processing and
Extractive Metallurgy Review, vol. 5, pp. 123 – 145.
Smith, E.L., Prevett, M.J., Lawrence, G.A., 1982. An improved
View publication stats
79
mechanism for large flotation cells. XIV International Mineral
Processing Congress, Toronto, Canada.
Subrahmanyam, T.V., Forssberg, K.S.E., 1988. Froth stability, par-
ticle entrainment and drainage in flotation—a review. Interna-
tional Journal of Mineral Processing 23, 33 – 53.
Trahar, W.J., 1981. A rational interpretation of the role of particle
size in flotation. International Journal of Mineral Processing 8,
289 – 327.
Yianatos, J.B., Finch, J.A., Laplante, A.R., 1988. Selectivity in
column flotation froths. International Journal of Mineral Proces-
sing 15, 279 – 292.