Minerals Engineering 64 (2014) 5162

Contents lists available at ScienceDirect

Minerals Engineering
journal homepage: www.elsevier.com/locate/mineng

A phenomenological model for an industrial ash otation cell

Bianca Newcombe
JKMRC The University of Queensland, 40 Isles Rd, Indooroopilly, QLD 4068, Australia

a r t i c l e i n f o a b s t r a c t

Article history: The measured slurry properties of an operating industrial ash otation cell treating a refractory gold ore
Received 5 January 2014 have been tested using a number of different modelling methods, including: axial dispersion; classica-
Accepted 24 March 2014 tion/partitioning; sedimentation dispersion; and rate equations. Limited success was achieved with the
Available online 4 May 2014
conventional approach to describing otation vessels and instead a novel approach of interpreting the
data from within the cell was developed. This method uses the residence time from within the quies-
Keywords: cent/settling zone of the cell (the region between the mixing zone surrounding the impeller and the froth
Froth otation
zone). In this situation particle residence time in the settling zone increases with increasing height in the
Modelling
Simulation
cell, and axial prole data can be used to determine recovery by size at varying heights relative to the
Sulphide ores mixing zone. Valuable mineral (pyrite) recovery is observed to decrease with increasing residence time
Gold ores in the settling zone, predominantly as a function of the internal cell geometry (inner cone). Plotting
Flotation machines the log rst order kinetic rate constant for each size class ki versus the residence time within the settling
zone of the cell, ss, a near linear relationship becomes evident which is described by the relationship:
ki ai ebi ss ; where a and b are empirically tted parameters for each size class i.
The residence time within each zone has been assumed to be constant for all size classes considered for
this initial part of the model development work, as a very detailed residence time distribution study
would be required to determine the variation in residence time for each size class.
Numerous mass balancing approaches have been tested, some of which attempted to incorporate the
froth phase, however due to the complexities of trying to model the froth the most accurate t was
obtained when the froth zone was excluded from mass balancing calculations. In this case, b was found
to be essentially constant across all size classes considered, at an average value of 1.41 (with a standard
deviation of 0.056). The average value of b changes with the method of mass balancing used, with the
attempts to incorporate the froth bringing in a greater level of deviation from the average. The value
of a changes with size, and follows a pattern similar to a recovery by size curve, peaking through the size
range of optimal oatability (212/+53 lm). The shape of this curve, and also the values of a for each size
class considered are similar regardless of the mass balancing method used.
Whether these two parameters (a and b) are intrinsic to the ore, the machine or the system as a whole
would take a considerable amount of detailed sampling from within the ash otation cell under
consideration and consequent analysis work which were beyond the scope of this initial study. The rela-
tionships developed here require validation in other systems, however if they are found to be robust and
universal, would allow for the unit recovery to be calculated using standard hydrodynamic equations.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
The ash otation environment is different from other types of
mineral otation devices (columns, tank cells, etc.) in that the cell
resides within the comminution circuit and receives very coarse
and thick slurry as its feed. This feed material is typically the
cyclone underow, which may be up to 80% solids and contain
particles that are better described as small rocks. In order to deal
E-mail address: biancanewcombe@yahoo.com.au
with this type of slurry the ash otation cell has some unique
features including a tangential feed inlet, conical bottom with a
centrally located tailings outlet, a secondary tailings outlet located
at a higher point and an inner cone to assist in froth transportation
to the concentrate launder. These features are depicted in Figs. 1
and 2 and allow two additional sub-processes to occur as com-
pared to a conventional otation device: the direct and intentional
bypass of very coarse material/small rocks directly to the tailings
outlet; and the classication/segregation of particles within the
otation zone of the cell. The effectiveness of these sub-processes
is what makes ash otation possible and understanding how they

http://dx.doi.org/10.1016/j.mineng.2014.03.023
0892-6875/ 2014 Elsevier Ltd. All rights reserved.
52 B. Newcombe / Minerals Engineering 64 (2014) 5162
Fig. 1. Schematic of ash otation cell operation (Newcombe, 2013a).
Fig. 2. Kanowna Belle ash otation cell detail (Murphy, 2012).
occur is essential for process optimisation. To date there exists no
hydrodynamic model specic to the ash otation environment
available in the published literature; this is possibly due to the
difculty in obtaining samples from both inside and around ash
otation cells in the operating environment. This paper provides
a rst attempt at mathematically describing the operating data
that has been obtained through extensive survey campaigns of a
refractory gold concentrator (Kanowna Belle Gold Mine), and inter-
nal hydrodynamic characterisation work of the ash otation cell
operating within that plant. Current otation modelling methods
will be discussed and their use in describing a ash otation cell
will also be addressed.
Results from the hydrodynamic characterisation of the ash
otation cell at Kanowna Belle (KB) have led to the division of
the cell into various zones, as depicted in Fig. 1. A detailed review
of the internal hydrodynamics has been previously presented
(Newcombe et al., 2013a) but in summary the zones within the cell
can described as follows:
 Bypass this accounts for the short-circuiting of very coarse
material and rocks to the tailings outlet, effectively acting as a
rst stage classier.
 Mixing zone this is the area surrounding the impeller where
good mixing and suspension of solids is observed.
 Settling zone this is a more quiescent zone, but differs from
the quiescent zone of a conventional otation cell in that segre-
gation of coarse particles is still occurring, this region acts as a
dual otation/classication zone. Plant sampling involved tak-
ing samples at 30 cm increments with increasing depth into
the cell, allowing this zone to be further divided into regions
of discrete volume and measured slurry characteristics.
 Froth zone similar to the froth zone of other otation
machines, however where an inner cone is in place there is a
very limited open area of froth at the top of the cell.
The ash otation cell that is the subject of this work is situated
within the grinding circuit and receives a portion of the cyclone
underow stream as its feed. To allow the reader to understand
how this cell interacts with the surrounding equipment, a process
ow schematic is provided in Fig. 3. Cyclone underow (CUF) mate-
rial is discharged by the cyclones into a distribution box, a portion
of the cyclone underow is sent directly to the ball mill feed, with
the remainder owing via a weir and orice plate arrangement to
the ash otation cell. The weir and orice plate are used as a surge
control system for the ash cell feed, with any large surges back-
owing to the ball mill feed off-take. Dilution water is added to
the distribution box directly above the ash feed inlet to allow
the percentage solids of the ash feed material to be reduced to a
level that is more amenable to otation (approximately 65% solids),
this water addition is controlled via an on-line density gauge and
ow indicator through the process control system. The segregation
of solids within the cell allows for an axial distribution of solids:
decreasing per cent solids with increasing height in the cell, result-
ing in the region that is directly below the froth to have a much
lower per cent solids than the tailings stream of the cell. This reduc-
tion not only in per cent solids, but also in average particle size is
 B. Newcombe / Minerals Engineering 64 (2014) 5162
Fig. 3. Flash otation cell location relative to other equipment in the Kanowna Belle circuit (after Newcombe, 2013b).
what makes the otation process successful in this environment
and this will be discussed further in subsequent sections. In many
concentrators a dual outlet is used, which takes a small ow of
material from the upper region of the cell, this is often used to bal-
ance the per cent solids around the circuit, however, in the cell
under consideration, this top outlet line was not in use during
any of the survey work. A comprehensive review of the literature
regarding ash otation has been previously published by the
author (Newcombe et al., 2012a) and the reader is directed to that
paper should further information be required.
2. Current applied otation modelling practise
Numerous applied modelling methods exist both commercially
and in the literature which can be used to various extents to pre-
dict the otation performance of an ore. These models divide the
feed stream into components, with each division typically based
on oatability. The application and usefulness of these models
in the plant environment introduces uncertainties as a result of
the complexities occurring in the process at an industrial level
(Ross, 1990). Parameters for these models are often derived from
laboratory testwork, which produces data which may be an order
of magnitude different from the performance observed in the pro-
cessing plant. Where parameters have been established from the
plant itself, it is usually under a xed (often ideal) set of operating
conditions, limiting the usefulness of the model application to con-
ditions and ore types variant to the conditions under which it was
developed. All applied modelling methods reviewed as part of this
work require a combination of the following to be performed:
 Extensive plant sampling.
 Batch otation testwork.
 Metallurgical characterisation.
 Mineralogical analysis.
 Chemical analysis.
The method that is chosen for a given situation will often
depend on:
53
 The detail of the required results.
 What is available for testing (ore sample versus plant testing).
 The amount of funds available for testwork.
 The timeframe in which results are required.
Each method has its own pros and cons, as will be highlighted
by this review. However, there is no single method that stands
out as the all-encompassing tool for geo-metallurgical perfor-
mance prediction. The driving force behind on-going research into
this area is the desire to be able to predict from an ore sample of
known mineralogical composition, using a basic and inexpensive test
method, the consequent metallurgical response (overall plant concen-
trate grade and recovery) all-be-it under a given set of operating
conditions.
Whilst there are many methods that go some way into predict-
ing plant performance, none of them (none that are published in
the literature) incorporate all three requirements of:
 Ore feed mineralogy/characteristics and mineralogical
response.
 Batch/laboratory testing methods.
 Plant performance characterisation.
with the results of model predictions being comprehensively
tested against the observed/measured response of the industrial
plant.
Many articles provide a predicted response on the basis of com-
bined ore characterisation and batch testing (Vianna, 2004;
Sutherland, 1989), and many others use batch testing and plant/
pilot plant performance characterisation (Welsby et al., 2010;
Amini et al., 2009; Lamberg and Bernal, 2009; Runge, 2009;
Pietroben et al., 2004; Alexander and Morrison, 1998), whilst a
few use ore feed properties and plant characterisation (Runge
et al., 2007; Bazin et al., 1994), however non could be found that
attempt to relate the results of both batch testing with a plant per-
formance characterisation to the ore feed mineralogical properties
and provide a prediction for a new ore/process condition that has
been validated with a subsequent plant survey.
54 B. Newcombe / Minerals Engineering 64 (2014) 5162
The model that is being developed for the Kanowna Belle ash
otation cell has an advantage in that when the plant surveys were
performed a belt cut was taken of the plant feed material. By doing
this the results of any laboratory testwork, that is performed on
that sample of ore can be directly related to the measured plant
performance at the time the ore was obtained. Further, any model
that is developed can be directly compared to the known plant
response, and as this process was performed twice, any testwork
methods or models developed under one set of conditions can be
tested against the second set of conditions. Linkages to the feed
ore mineralogy can also be developed and compared with both lab-
oratory and plant scale results. A laboratory batch otation test
method has been developed and validated using the results of
two survey campaigns that can be used to produce a batch test cell
concentrate that is nearly identical to the concentrate produced by
the plant ash cell on the same ore in terms of the distribution by
size of the valuable target minerals (pyrite and gold) and the liber-
ation characteristics of pyrite (Newcombe, 2013a). This paper will
take this predictive ability to the next level by developing a math-
ematical relationship between the particles present in the mixing
zone of the cell and the measured internal slurry properties
approaching the froth interface, with the ultimate aim of predict-
ing the cells recovery by size.
There are two signicant applied methods of modelling ota-
tion circuits that have been identied and studied extensively in
the literature, these being empirically based component models
and property based models, and whilst other methods exist (such
as the shaped distribution model), it is these two that show the
most relevance to the industrial situation. Both model types are
based on the concept of ore oatability.
2.1. Floatability
The current applied otation models utilise oatability as the
basis of classication of particles. Floatability is a property of the
feed stream and is a function of both particle and stream parame-
ters such as particle size, mineralogical characteristics, surface
composition and shape as well as pulp chemical characteristics
such as Eh, pH, reagent type and dosage (Welsby et al., 2010;
Runge, 2009; Vianna, 2004). The oatability response of a otation
system cannot be directly measured in any way and must be calcu-
lated from the observed response of an ore. Savassi (2006) ana-
lysed industrial plant data and linked increased recovery with
the volumetric grade of a particle. Welsby et al. (2010) used a com-
ponent method which directly related physical properties such as
size and liberation to the otation response. Whilst Runge (2009)
stated that oatability is not dependant on operating conditions
within the cell, Vianna (2004) related oatability to cell hydrody-
namic conditions such as gas dispersion, agitation and turbulence.
The use of oatability as a means of classifying particles relies
on the premise that all particles within a particular class/compo-
nent oat at a similar rate. The majority of models are based on
the assumption that once a particle has been assigned a oatability
value, this value will not change throughout the processes within a
otation circuit (with the exception of processes that are directly
intended to cause a change in oatability such as regrinding or
reagent addition) (Runge, 2009).
Through extensive research, relationships between an ores o-
atability and other parameters of the otation process have been
developed; of these the most signicant relationships for use in
predicting performance came from Gorain et al. (1997) and
Savassi (1998).
Gorain et al. (1997) found a relationship existed between the
otation rate (k), the bubble surface area ux (Sb) and the recovery
in the froth (Rf). The ore oatability (P) could be determined from
the slope of the line relating these three factors using:
k Sb Rf P 1
This was further investigated by Massinaei et al. (2009) who
showed that in otation columns the collection zone rate constant
kc is linearly dependant on Sb or gas hold up Eg , which in turn are
signicantly affected by air addition rate (Jg) as well as slurry per
cent solids and frother dose and type. Savassi (1998) took the k
Sb relationship one step further to incorporate more of the process
variables that affect otation performance (water recovery, Rw;
residence time, s; and a factor that allowed for entrainment in
the froth, ENT):
PSb sRf 1  Rw ENT Rw
R 2
1 PSb sRf ENT Rw
These two relationships form the basis of the research
performed in AMIRA P9 project group and are the foundation equa-
tions for the JKSimFloat modelling and simulation software
package.
2.2. Empirically derived models
Empirically derived otation models use batch test data and/or
plant data to calculate or t otation rate constant values to each
of the components in the system (the number of components for
each mineral of interest will be system specic, however Runge
(2009) indicated that one or two oating and one non-oating
component would generally be sufcient).
Whilst empirical models have the advantage that they do not
require size by liberation data to classify particles, this is also an
inherent disadvantage, as the groupings of particles are not related
to any measureable physical property, and consequently cannot be
connected to other mineral processing operations such as milling
or cycloning (Vianna, 2004). Empirical models of otation circuits
have been successfully developed for many existing operations,
however where these models have been used to simulate changes
to the plant only limited information regarding the validation of
performance predictions have been published. As these models
do not incorporate physical characteristics there use is limited in
circumstances where there is a change in ore properties (particu-
larly liberation), size distribution or percentage solids of the feed
to a circuit (Vianna, 2004). Bazin et al. (1994) and McIvor and
Finch (1991) provide examples of empirical models that can incor-
porate changes in particle size, however in both cases these models
are ore specic.
Where recovery is known, the following two equations can be
used to determine the rate constants of each component for either
a batch test or an industrial cell (where perfect mixing is assumed)
(Runge, 2009).
For batch otation test data, the recovery is described as:
Xn
Rt i1
mi 1  eki t 3
where R(t) = overall recovery at time t (this is the sum of all the
individual component recoveries), mi = proportion of a mineral in
each component and ki = rate constant of each component.
Plant data for a perfectly mixed cell relates recovery using:
Xn  
ki s
Rt m
i1 i
4
1 ki s
where s = the residence time for the cell or bank of cells being
considered.
Both these equations require the recovery of the target mineral
to be known, usually on a sized basis, which in the case of an
industrial circuit, would require a plant survey. The measured
recovery, mineral masses (calculated by assay) and residence or
test times can then be used to determine the rate constant k value
 B. Newcombe / Minerals Engineering 64 (2014) 5162
for a given particle class. It should be noted at this point that many
authors use k as an indication of oatability and the two terms are
often used interchangeably. In the current work k will be used to
denote rate constant exclusively.
These equations form the basis to the many and varied mathe-
matical methods available for empirical based modelling available
in the literature. A comprehensive example of industrial circuit
modelling is provided by Runge (2009), who used both batch
otation test data and plant survey information to develop a com-
ponent model for a galena ore.
2.3. Property based models
Property based models use one or more measureable properties
of the ore to dene groups into which particles can be classied.
The most basic level of grouping is based on particle size alone,
with more complexity being introduced when size and libera-
tion/mineralogical data are used or further when size, liberation/
mineralogical data and surface chemical properties are used. Prop-
erty based models are often thought to be much more expensive
than empirically derived models due to the nature of the analysis
required to gather the data, and the error associated with these
models is speculated to be higher due to the amount of sample
processing required to obtain the necessary data for model
development (Runge, 2009). As with empirically derived models,
particles within a particular grouping are required to have the
same oatability, and otation rates for each component are deter-
mined using a measured recovery.
Extensive work by both Runge (2009) and Vianna (2004)
showed that there exists a distribution of oatabilities within the
same size/liberation class, indicating another factor must be
considered, signicantly the surface composition of the mineral.
Factors such as non-uniform collector coverage or surface
oxidation may be playing a signicant role in an individual
particles otation response.
The use of size as a basis for partitioning particles has the inher-
ent complexity of the different behavioural responses of particles
of different sizes:
 Coarse particles are more sensitive to changes in reagent addi-
tion (Gontijo et al., 2007; Bravo et al., 2005; Vianna, 2004; Bazin
and Proulx, 2001; Senior et al., 1994; Klimpel and Isherwood,
1991; Trahar, 1976).
 Larger particles have different drainage rates to smaller parti-
cles (Newcombe, 2013b; Seaman et al., 2006; Ross, 1991b).
 Each size class will respond differently to hydrodynamic condi-
tions (Newcombe et al., 2013a; Grano, 2006; Schubert, 2008;
Deglon, 2005; Kallioinen and Heiskanen, 1993; Lins and
Adamain, 1993; Ahmed and Jameson, 1985).
 Froth structure, stability and mobility are all inuenced by the
characteristics of particles (Farrokhpay, 2011; Barbian et al.,
2006; Tsatouhas et al., 2006; Barbian et al., 2003; Klimpel and
Isherwood, 1991).
The amount and liberation characteristics of valuable mineral
present in a particle (at the surface of the particle) is also known
to be of importance (Suthers et al., 2004; Vianna, 2004; Bazin
et al., 1994; Sutherland, 1989). As shown by Vianna (2004), where
target valuable minerals (galena) have an inherently high oatabil-
ity, liberated valuable particles always have a signicantly higher
oatability than composites for all size classes considered. The o-
atability of composites was further complicated by the nature and
proportion of the constituent minerals.
Runge (2009) compared both empirically derived and property
based methods to predict a response in a galena otation system.
55
The work showed that when sufcient numbers of parameters
are used the two types of models predicted the same result.
2.4. Column otation (axial dispersion) models
Review of the hydrodynamic data from within the ash ota-
tion cell under consideration in this work illustrated key differ-
ences between ash otation and conventional mechanically
driven otation cells, signicantly the non-homogeneity of the
slurry with segregation of solids resulting in an increase of both
slurry per cent solids and solids size distribution (P80) with increas-
ing depth into the cell, and also the trend of decreasing gas velocity
and hold-up with increasing depth (Newcombe et al., 2013a). Cur-
rent modelling practise for conventional otation cells cannot
account for this type of variation due to their high levels of mixing
making the slurry essentially homogeneous. In order to nd a
model type which can be used as a foundation for the description
of the quiescent settling zone within the ash otation cell, ota-
tion column models have been considered. These models have
been reviewed for their potential as a basis for modelling the set-
tling zone of the ash otation cell (which does not incorporate
the region of good mixing around the impeller), with the under-
standing that some modication to the model structure could be
required.
Tuteja et al. (1994) classied column otation models as either
kinetic or non-kinetic, with the kinetic models using empirically
tted parameters based on the observed performance, and most
commonly based on axial dispersion theory. Mills and OConnor
(1990) stated that either the mixers in series or the axial dispersion
model (plug ow) could be used to describe the behaviour of a con-
tinuously operated column. In the case of the axial dispersion
model, the dispersion coefcient (D) is used to describe the degree
of mixing and the following equation is used to describe the
change in concentration of a tracer (C) with time (t) (Mills and
OConnor, 1990):
@ 2 C @c @c
D ui 5
@z2 @t @z
where ui is the particle velocity (or liquid velocity where particle
size is small) and z is the distance along the column.
There are some key assumptions that form the basis of axial dis-
persion models including (Massinaei et al., 2009; Tuteja et al.,
1994; Dobby and Finch, 1991b; Luttrell and Yoon, 1991; Mills
and OConnor, 1990):
 First order kinetics.
 Good mixing (i.e. no short-circuiting).
 Uniform gas dispersion within the pulp/quiescent zone.
 Consistency of residence time of liquid and particles.
 Cell height /aspect ratio is also a consideration.
Determination of dispersion coefcients requires both liquid
and sized solid tracer tests, however as pointed out by Dobby
and Finch (1991a,b) for particles less than 150 lm the solids and
liquid dispersion coefcients are considered to be equal. Further
analysis of column models by Xu and Finch (1992) indicated that
the key assumptions these models are based on may only be rele-
vant for particles up to 100150 lm and further recommend using
a ratio of particle to liquid residence time. The translation of this to
the ash otation environment limits the breadth of application, as
particles over 850 lm are required to be incorporated into the
model. The residence time of an 850 lm particle is expected to
be considerably less than that of a 38 lm particle for example;
with the work of Yianatos et al. (2005, 2001) illustrating this in a
conventional mechanically driven cell, nding that even where
56 B. Newcombe / Minerals Engineering 64 (2014) 5162
mixing is good, coarse particle residence time was up to 8% lower
than that of the ne particles. Greater exibility can be achieved by
compartmentalising the column so that a mixers in series approach
can be used with each compartment assigned their own parame-
ters, although this level of complexity is not necessarily required
for a otation column (Dobby and Finch, 1991a,b).
As pointed out by Tuteja et al. (1994) and Dobby and Finch
(1991a,b) axial dispersion models do not take into account varia-
tions in radial mixing and non-uniform velocity proles on column
mixing, they have also been derived for a different application in
which particles are not being deliberately attached to bubbles
and where variation in feed properties, particularly grade, is not
an issue. Dobby and Finch (1991a,b) also highlight that the method
of parameter tting and boundary conditions will affect the result.
Jg used in these models is typically in the range of 0.92 cm/s
(Massinaei et al., 2009). Robust correlations exists where Jg lies
within a normal range, the lower limit of which is 0.5 cm/s; how-
ever from the available data taken from within the operating ash
cell, all measured Jg values were less than 0.35 cm/s (Newcombe
et al., 2013a).
Models for otation columns proposed by Alford (1992), Luttrell
and Yoon (1991) and Massinaei et al. (2009) require operating vari-
ables such as gas dispersion properties, slurry viscosity and density
to be known. The degree of mixing is taken to be constant within a
vessel, but has been shown to vary considerably with the per cent
solids (Mills and OConnor, 1990) which has implications for the
adaption of this model to ash otation cell in which the per cent
solids are very high and also change substantially with height in
the cell.
Due to the complexities of incorporating the froth zone, the
majority of models assume froth zone recovery is 100% (Tuteja
et al., 1994) and are effectively modelling the overall recovery and
rate constant. As with most kinetics based calculations, recovery
must be determined at a number of different residence times, which
can be achieved in otation columns by changing the feed ow rate.
Dobby and Finch (1991a,b) found that the residence time
distribution (RTD) of a otation column could be described by
the plug ow axial dispersion model, which requires the average
residence time and the vessel dispersion number Nd. Alford
(1992) later found that for industrial columns the axial dispersion
model with perfect mixing gave best results. The work of Alford
(1992) involved a systematic review of previous column models
and concluded that where the overall performance of a column is
assumed to be dominated by the pulp zone, recovery is a function
of rst order kinetics and residence time distribution. The follow-
ing equations were used as the foundation of his analysis, and will
be referred to throughout this paper.
Perfect mixing allows recovery to be calculated via:
ks
R 6
1 ks
where each size class is considered separately, Eq. (4) is used.
The rate constant for a particle of a given size i and mineralog-
ical classication j, can be determined using:
lw
kij C ij v g 0:75 7 1. Plant survey data, allowing the ows into and out of the ash
lL
(Alford, 1992) where Cij is a mineral specic parameter, lw and lL
are the viscosities of the water and pulp respectively; and v g is
the corrected air rate, which incorporates the column cross-
sectional area and the concentrate mass ow rate into its
calculation.
Alford (1992) went on to use the following empirical equation
to model the effects of particle size:
a time. Slurry samples were taken from each sample point using
C ij C j dp expbdp 8
where C j is a mineral specic term independent of particle size, dp
represents particle diameter and a and b are constants.
Where collection zone kinetics are to be considered separately
to the froth zone, the froth zone recovery needs to be calculated
so that it can be effectively removed from collection zone calcula-
tions. This is typically achieved by taking surveys at varying froth
depths to determine recovery, then extrapolating the recovery
factor back to a zero froth depth (Massinaei et al., 2009)
Relationships established between the collection zone kinetics
and gas dispersion properties, specically Sb indicate a linear rela-
tionship exists. In the ash otation environment gas dispersion is
very poor and there is non-uniformity of gas dispersion both radi-
ally and with axial height within the cell (Newcombe et al., 2013a).
The relationships developed by workers such as Gorain et al.
(1997) and Massinaei et al. (2009) may not be valid at very low
gas dispersion levels. The kSb relationship has been shown to
change with changes in operating parameters such as per cent
solids by Massinaei et al. (2009) indicating that the high per cent
solids and non-uniform distribution of per cent solids within the
ash otation cell will have a substantial effect on gas dispersion.
Another key issue that has been highlighted by numerous work-
ers in column otation in the scale-up of models is that of carrying
capacity, which is where the froth becomes the rate limiting factor
due to overloading, this in-turn means that the froth recovery com-
ponent of an overall kinetic rate constant may contribute signi-
cantly (Luttrell and Yoon, 1991; Dobby and Finch, 1991a,b). In
the ash otation environment the cells are typically operated
with very shallow froths to maximise recovery, so the limiting
effect of the froth should be minimised.
Flotation column models can provide a foundation in uid
mechanics that will be built on in subsequent sections of this
paper, however the models themselves are not considered appro-
priate for the ash otation environment as some of the underpin-
ning assumptions are not valid in a ash otation cell and the ow
regime is considerably different. This review of column modelling
has led to the evaluation of a modied sedimentation dispersion
model, which has a similar founding as the axial dispersion model
but takes into account variations in solids dispersion with axial
height, as is investigated and tested in Section 5.1.
3. Experimental
Data used in the development of the ash otation model has
been collected as part of a larger ash otation study, conducted
at Kanowna Belle Gold Mine (Barrick). The ore treated by this plant
is a refractory gold ore, with pyrite being the mineral of interest for
otation. Pyrite is the dominant sulphide; tomographic (3D) anal-
ysis of the ore has shown that approximately 82% of the pyrite
grains are between 38 and 425 lm (AMIRA, 2012). Detailed infor-
mation on how the surveys were conducted has been previously
published (Newcombe et al., 2012b, 2013a, 2013b), and only rele-
vant aspects of the work will be presented here.
Two key sources of information were required as inputs into the
ash otation model:
otation cell to be determined on a size by assay basis.
2. Axial prole data from within the operating cell to establish the
variation of slurry composition with a change in axial height.
3.1. Plant survey work
Samples around the ash otation cell were obtained as part of
a full plant survey of 2 h duration with a 10 min sample interval
sample cutters sized specically for each individual sample
 B. Newcombe / Minerals Engineering 64 (2014) 5162 57

analysed for per cent solids (using wet and dry sample weights),
solids SG, head assays for Au, total S, Fe, Cu, MgO and SiO2 and
distribution of the assayed elements by size.

3.3. Sizing

The procedure for sizing each of the samples involved wet

Fig. 4. Plan view of froth camera housing location/sample access portal (Newcombe
et al., 2013a).
screening at 38 lm; drying the screen products; dry screening
p
the +38 lm material from 38 lm upwards following the 2 series
(the maximum size of screen used was stream dependent, for
example the cyclone underow stream had a top screen size of
16 mm, whilst the cyclone overow had a top screen size of
850 lm). The 38 lm material from the dry screening was then
combined with that obtained from the wet screening.
4. Key results

location. Stream samples were analysed for per cent solids (using
wet and dry sample weights), solids SG, head assays for Au, total
S, Fe, Cu, MgO and SiO2 and distribution of the assayed elements
by size.
3.2. Axial prole data
The inner cone of the ash cell does not allow for multiple
access sites for axial proling of the operating cell below. In the
case of KB the froth camera viewing portal was utilised as an access
hole for sampling of the cell. The data used in this model develop-
ment therefore represents a single radial location. The location of
the sample point is shown in Fig. 4, and is located near the cell wall
on the opposite side of the cell to where the feed is added.
Slurry samples have been taken using a spring loaded sample
dipper with increasing depth into the cell (at increments of
30 cm) at the location shown in Fig. 4. The procedure for using this
apparatus was as follows:
The relevant results obtained for both the survey work and axial
proling will be presented here. Full sets of survey results are
available in Newcombe et al. (2012b, 2013b), and hydrodynamic
characterisation results, which includes the axial prole data are
presented in Newcombe et al. (2013a). The relevant mass balanced
survey results for the November 2011 survey are presented in
Table 1, this survey is associated with axial prole data set 2. The
rst axial prole was taken as a stand-alone dataset to determine
the magnitude of the segregation that was occurring in the operat-
ing cell. It is important to note that during all survey and prole
work, the middle discharge (dual outlet) of the cell was not in
operation, meaning that there is only a single tailings ow from
this cell.
Tables 2 and 3 are presented to illustrate the magnitude of the
changes in slurry composition within the settling zone. In both
Table 3
Total solids P80 with increasing depth (Newcombe et al., 2013a).

1. Lower it into the cell to the required depth (indicated by marks Sample depth (cm)
along the shaft) with the sample chamber sealed. 0 30 60 90 120 135 150
2. Open both ends to allow slurry to ow through for 10 s.
Overall solids P80 #1 (lm) 150 180 206 315 340 345
3. Release the springs to seal the slurry sample inside. Overall solids P80 #2 (lm) 295 294 367 413 526 522
4. Rinse off the outside of the sample pot using water (as the sam-
ple pot would have to be raised through the layer of froth to be
removed from the cell, the outside could be contaminated with
high grade material). Table 4
Mass balanced sulphur grades by size with depth prole data #2.
5. Release sample into a bucket.
Size class (lm) Zone within cell
This process was repeated 1014 times at each required depth 9060 6030 300
to produce a single sample of sufcient mass for each depth to
+850 0.54 0.53 0.53
allow for all subsequent analysis. Samples for each depth were 850/600 0.52 0.53 0.53
600/+425 0.42 0.42 0.42
425/+300 0.51 0.49 0.49
Table 1 300/+212 0.57 0.61 0.60
Flash otation survey results. 212/+150 0.75 0.78 0.77
150/+106 1.28 1.36 1.34
Stream Tonnes % Solids Au (g/t) S (%)
106/+75 2.03 2.24 2.17
Cyclone underow(ash feed) 413 83.4 5.6 1.12 75/+53 2.52 2.86 2.76
Flash concentrate 9.3 37.5 43.5 8.69 53/+38 1.98 2.23 2.16
Flash tail 404 65.1 4.7 0.94 38 0.74 0.80 0.79

Table 2
Slurry per cent solids with increasing depth (Newcombe et al., 2013a).

Sample depth (cm)

0 30 60 90 120 135 150
Slurry density (% Solids) #1 36.5% 41.5% 45.3% 55.6% 57.1% 57.5%
Slurry density (% Solids) #2 36.0% 53.8% 63.5% 66.4% 66.3% 67.5%
58 B. Newcombe / Minerals Engineering 64 (2014) 5162

 
Table 5 w U pi  U sL Lx
Mass balanced throughput (tph) for each increment prole data #2. C si C 1 C 2 exp  L 10
Esi
Size class (lm) Zone within cell
The solids concentration Cs for each size fraction i is therefore a
9060 6030 300
function of:
+850 0.08 0.03 0.02
850/600 0.48 0.13 0.08  The average liquid fraction of the slurry (w  L ), which intrinsically
600/+425 1.67 0.49 0.30
425/+300 5.25 1.69 1.05
takes per cent solids into consideration;
300/+212 9.35 3.50 2.15  C1 and C2 are constants.
212/+150 8.30 3.50 2.12  Esi is the dispersion coefcient for solids (m2/s).
150/+106 4.28 1.88 1.12  L(m) is the height of the slurry column.
106/+75 2.14 0.98 0.57
 x(m) is the normalised distance from the column base (height in
75/+53 1.30 0.59 0.35
53/+38 0.56 0.27 0.15 cell).
38 4.78 2.23 1.31  The settling velocity of solid particles is given by Up while UsL is
the linear liquid velocity (m/s).

these Tables, a sample depth of zero represents the region just Eq. (10) has been selected for testing against the ash otation
below the froth interface with increasing depth into the cell until cell data as the concentration of each size class is known at each
the mixing zone is reached (the impeller region) at a depth of height of the cell. This form of the model also incorporates changes
approximately 90 cm. The depth at which good mixing commences in per cent solids and size distribution and includes variations in
is dependent on the operating conditions and is taken to be the cell height via:
depth at which slurry properties plateau (i.e. become similar to
the properties of the next sample down). Table 2 clearly illustrates z
x 11
the solids segregation that is occurring within the settling zone of L
the cell. A drop of 21% and 31.5% solids was observed in operating
where z is the distance from the base of the column, which is con-
conditions 1 and 2 respectively. A large change in the distribution
sidered to be the mixing zone interface within the ash cell.
of solids with size is also observed, as illustrated by the P80 results
Two sets of axial prole data exist from within the operating
of Table 3. Table 4 presents the mass balanced assay results of pro-
ash otation cell at Kanowna Belle. The rst prole was taken
le data set 2 (associated with the survey results of Table 1), while
as a stand-alone data set (June 2011), while the second was
Table 5 details the mass balanced throughput (tph) of each zone.
performed as part of plant survey campaign (November 2011). As
a otation model must be able to describe the behaviour of the tar-
5. Models tested against ash otation conditions get mineral for recovery, in this case pyrite, the measured concen-
trations of sulphur at each height z, in the cell were calculated from
5.1. Modied sedimentation dispersion model the prole data. As pyrite is the only sulphide mineral of signi-
cance in this system, sulphur can be used as a proxy for pyrite.
The sedimentation dispersion model is a phenomenological  L can also be calculated at each height, z from the per cent solids
w
model widely used in chemical processing industries to describe data for the known height of the column, L, which in this case is
the solids behaviour within slurry bubble columns. As with ota- taken as being 120 cm. A least squares tting method was then
tion column models there is no mixing device within the vessel used to minimise the difference between the measured sulphur
and solids/liquid ow is counter current to the gas ow. These concentration and that predicted using Eq. (10) for the empirically
models are not used to describe the situation where particles are tted parameters Esi and Upi for each size class; and C1, C2 and UsL,
physically attached to gas bubbles, as is the case in a ash otation which are taken as being universal system constants for the pur-
cell. These models have been selected for evaluation to determine poses of this exercise. A parity graph is presented in Fig. 5 of the
if this type of model may be appropriate, with some modication, measured versus predicted sulphur concentration for each size
to describe the settling zone of the ash otation process. class in the cell at each height. Linear regression has been used
One signicant difference between the axial dispersion and sed- to indicate the closeness of the predicted and measured values
imentation dispersion models is that the later has been developed by the proximity of the correlation coefcient and R2 to a value
for the situation where solids concentration decreases with axial
height within the column and the effect of gas velocity on axial sol-
ids concentration is not considered to be signicant (Fan, 1989).
There are two key parameters which form the basis of these mod-
els: the axial solids dispersion coefcient (E); and the solids set-
tling velocity, which ultimately reside in an equation of the form
(Fan, 1989):
dC s U Sl
Ezs  V p Cs  C f V Sl C fs 9
dz 1  eg s
The model in its original form is not readily compatible with the
type of data that is able to be gathered in an operating environ-
ment and does not take into consideration a distribution of particle
sizes being present (Smith et al., 1986), it also assumes that gas
hold-up is constant with axial height. Smith et al. (1986) developed
the following relationship for poly-dispersed particles in a slurry
bubble column, and it is this form of the model that was tested
with the operating data from the settling zone of the ash otation
cell: Fig. 5. Parameter tting evaluation for prole data set 1.
 B. Newcombe / Minerals Engineering 64 (2014) 5162
of 1. As can be seen in Fig. 5, the correlation coefcients for each
height of the cell considered are reasonable, given that this is
raw experimental data, but the R2 values are considerably less than
1. To further evaluate this model the second set of prole data has
been analysed in two ways. Firstly by tting its own unique set of
parameters in exactly the same way as has been described for the
prole 1 data set, as shown in Fig. 6; and secondly by applying the
parameters from data set 1 to data set 2 and evaluating the differ-
ence, as shown in Fig. 7.
When Fig. 6 is considered it is evident that the goodness of t as
indicated by the correlation coefcients and R2, is approximately
the same as that achieved with data set 1. What does become
apparent with data set 2, is the curved nature of the of the relation-
ship, indicating there is more at work within a ash cell than can be
accounted for by the sedimentation dispersion model as given in
Eq. (10). Interestingly, when the parameters determined for data
set 1 are applied to the raw data of prole 2, the R2 values (shown
in Fig. 7) are similar to those shown in Fig. 6, but the slope of the
lines changes signicantly. This change in slope is a function of
the different feed grades of material into the cell at the times the
proles were taken (roughly 5 months apart). Again the non-linear-
ity is evident even when the parameters for data set 1 are applied.
The analysis presented here has indicated that the classication
and segregation of particles, as well as the underlying principles of
otation itself cannot be accounted for by a modied sedimenta-
tion dispersion model alone and another method of modelling this
data must be considered.
Fig. 6. Parameter tting evaluation for prole data set 2.
Fig. 7. Evaluation of prole 1 parameters with prole data set 2.
59
5.2. Kinetic/rate based approach
Rate equations are a commonly used way of describing the
observed recovery performance of minerals as they undergo the
otation process, and are considered to be phenomenological in
their nature (Polat and Chander, 2000). In the laboratory, kinetics
are easily described from timed concentrates and their resulting
recoveries. The method of Kelsall (Kelebek and Nanthakumar,
2007) has previously been applied to laboratory data generated
from ore taken as a belt cut during a plant survey campaign and
the ore under consideration here has been shown to have second
order kinetics (Newcombe, 2013a; Newcombe et al., 2012b). The
ash otation cell itself has been found to only recover very fast
oating material; analysis of the ash otation cell concentrate
from 2 separate surveys showed that 85% and 93% of the pyrite
particles it contained were considered to be greater than 90% liber-
ated (across all size classes), with the 150/+10 lm size range con-
taining 95% and 96% of pyrite in the >90% liberated category
respectively. Although the ore system as a whole displays higher
order kinetics, the ash cell will initially be treated as though it
is treating a rst order kinetic ore (Newcombe et al., 2013b). This
assumption that the particles are fast oating is supported by the
work of Welsby et al. (2010) who found that over 71% of the mate-
rial determined to be fast oating was greater than 90% liberated
(in a galena system).
There are two primary ways of describing the kinetics of a ves-
sel and this relies on the ow regime it utilises, either plug ow or
perfect mixing. Where these are found to be inadequate, many
authors have used mixers in series, which treats a single otation
cell as though it is a combination of varying numbers of perfectly
mixed and/or plug ow vessels (Yianatos et al., 2005, 2001;
Grano et al., 2007). It must be noted that detailed hydrodynamic
studies would be required to determine an exact number of vessels
for a ash otation cell.
In a conventional otation circuit kinetic calculations are facil-
itated by taking down the bank or block surveys of the plant. The
data gathered in this manner allows recoveries to be determined
with increasing residence time. In the ash otation situation
there is only a single cell and so a method needed to be estab-
lished that would allow for recoveries to be determined with
increasing residence time within the cell itself. As column ota-
tion cells are often single cell installations, investigation of how
this is performed with that type of unit was considered, with
the nding that changing the feed ow rate is used to change
the cells residence time, with recovery being calculated for a
number of feed ow (residence time) conditions. Changing the
feed ow rate to a ash otation cell would involve changing
the way the grinding circuit performs and hence have a signi-
cant impact on a number of key process variables (such as parti-
cle size and slurry per cent solids, which in turn could affect air
dispersion, etc.). To reduce the amount of variability associated
with parameters other than residence time another method was
required. The use of prole data from within the cell, which uti-
lises slurry samples taken with increasing depth at a single axial
location, have been used to provide this information. The sample
interval in this case was 30 cm, so the cell itself has been dived
into 30 cm zones to a depth of 120 cm. The mixing zone of the
cell has been taken as the region below 90 cm cell depth, with
the region above being treated as the settling zone. 90 cm has
been used as the cut-off height for the mixing zone in this
instance as the data indicates a reasonable level of mixing below
this depth (Newcombe et al., 2013a).
Testing using rate equations encompassed a range of scenarios
including assumptions of either perfect mixing or plug ow, kinet-
ics within each 30 cm zone of the cell, or kinetics over a cumulative
volume.
60 B. Newcombe / Minerals Engineering 64 (2014) 5162

Mass balancing on a sized basis has been performed treating
each 30 cm increment as its own vessel, allowing recovery to be
calculated not only for each zone, but also for cumulative zones.
Calculating recovery for incrementally increasing volumes (and
hence residence time within the cell) in this manner allows for rate
constants to be determined for each size class. Mass balancing has
been performed using three scenarios:
1. Incorporating drop back from the froth at a grade equivalent to
the concentrate.
2. Incorporating drop back from the froth at a grade equivalent to
the 0 cm sample (pulpfroth interface).
3. Ignoring the froth.
Each of these mass balances has been assessed assuming either
perfect mixing or plug ow.
To allow for recovery to be calculated with increasing residence
time the recovery from the mixing zone to a cumulatively increas-
ing height within the cell has been used. In this case there are 3
recoveries which can be used:
1. Recovery to 60 cm relative to the mixing zone (90 cm).
2. Recovery to 30 cm relative to the mixing zone (90 cm).
3. Recovery to 0 cm relative to the mixing zone (90 cm) (0 cm is
just below the froth interface).
Residence time increases as the froth interface is approached
from the mixing zone, as detailed in Table 6. The division of the cell
is depicted graphically in Fig. 8, and Table 6 contains the volumes
of each of the zones, which takes into account the internal cell
structure.
Table 6 summarises the rate constants that have been calcu-
lated for the mass balance in which the froth zone is ignored. As
the ow regime within the cell remains undened both plug ow
and perfect mixing have been considered separately. The data pre-
sented in Table 6 has one key difference to the data that would be
measured with increasing residence time down a bank of cells: the
recovery decreases as residence time increases. There are several
factors which are contributing to this, including: cell geometry
(the inner cone reduces the available volume as the froth is
approached); decreasing per cent solids with increasing cell

Table 6
Sized recovery and kinetic constants for pyrite with increasing residence time within the settling zone.

Depth volume 9060 cm 9030 cm 900 cm (Froth interface)

residence time
3 3
2.075 m 3.791 m 5.03 m3
1.34 min 2.46 min 3.26 min
Size class Pyrite recovery k (min1) Pyrite recovery k (min1) Pyrite recovery k (min1)
Perfect mixing Plug ow Perfect mixing Plug ow Perfect mixing Plug ow
+850 0.010 0.0075 0.0075 0.003 0.0013 0.0013 0.002 0.0006 0.0006
850/+600 0.084 0.0681 0.0652 0.024 0.0099 0.0098 0.015 0.0046 0.0045
600/+425 0.150 0.1311 0.1207 0.044 0.0186 0.0182 0.027 0.0086 0.0084
425/+300 0.314 0.3400 0.2800 0.097 0.0435 0.0414 0.060 0.0197 0.0191
300/+212 0.423 0.5464 0.4097 0.170 0.0831 0.0756 0.103 0.0352 0.0333
212/+150 0.464 0.6444 0.4642 0.204 0.1042 0.0928 0.122 0.0426 0.0399
150/+106 0.503 0.7543 0.5208 0.235 0.1254 0.1093 0.138 0.0490 0.0454
106/+75 0.530 0.8388 0.5616 0.268 0.1487 0.1268 0.152 0.0549 0.0505
75/+53 0.524 0.8193 0.5524 0.272 0.1518 0.1290 0.153 0.0554 0.0510
53/+38 0.444 0.5938 0.4365 0.240 0.1289 0.1120 0.133 0.0473 0.0440
38 0.425 0.5502 0.4119 0.214 0.1106 0.0978 0.123 0.0431 0.0403

Fig. 8. Division of settling zone of cell into 30 cm compartments.

 B. Newcombe / Minerals Engineering 64 (2014) 5162
Fig. 9. k (min1) versus residence time within the settling zone for each size class (for the mass balancing scenario which excludes the froth, assumes perfect mixing).
Fig. 10. Variation of a with size (for the mass balancing scenario which excludes
the froth and assumes perfect mixing).
height; and increasing gas hold-up as the froth is approached.
There are also many underlying sub-processes at work within the
ash otation environment which contribute to this observation,
fall-back, drainage, detachment are all occurring as the valuable
particles attempt to rise attached to a bubble through a thick slurry
trying to reach the froth.
A plot of rate constant versus residence time for each size class
is presented in Fig. 9 and shows that the trend of the curve is the
same regardless of the size fraction being considered. Similar
results are achieved when the data from the different mass balanc-
ing methods are utilised (i.e. when the froth phase is incorporated).
All scenarios tested yielded equations of the same form:
ki ai ebi ss
For each size class i: k is the rst order rate constant; a and b are
empirically tted parameters; and ss is the residence time within
the settling zone.
For the data shown in Fig. 9, b was found to be essentially con-
stant, at an average value of 1.41 (with a standard deviation of
0.056). The average value of b changes with the method of mass
balancing used, with the incorporation of the froth also bringing
in a greater level of deviation from the average. The ash otation
cell at Kanowna Belle is operated with a shallow froth (typically
between 100 and 160 mm) to maximise recovery, so froth zone
recovery should be high; and as no specic measurements were
61
made of the froth phase at the time of sampling, for the purposes
of this exercise the froth phase will be ignored.
If plug ow is considered, then the values for both a and b
change, but the nature of the relationship remains the same as
per Eq. (12). In this case the average value for b is 1.31, with a
standard deviation of 0.078.
The value of a changes with size, and follows a pattern similar
to a recovery by size curve, peaking through the size range of opti-
mal oatability (212/+53 lm). The shape of this curve, and also
the values of a for each size class considered are similar regardless
of the mass balancing method used. Fig. 10 shows the variation of a
with size.
Whether a and b are parameters intrinsic to the ore, the
machine or the system as a whole would take a considerable
amount of detailed sampling from within the ash otation cell
under consideration which was beyond the scope of this initial
study. The relationships developed here require validation in other
systems, however if they are found to be robust and universal,
would allow for the unit recovery to be calculated using standard
hydrodynamic equations. The mixing regime within the settling
zone of the cell will need to be dened so standard hydrodynamic
equations can be appropriately applied. A detailed size based resi-
dence time study is recommended for this purpose.
6. Conclusions
Current otation modelling methods have been evaluated and
(where appropriate) tested using data gathered from within an
operating ash otation cell. Of the models considered, none were
found to adequately describe the behaviour of the ash otation
cell and a novel method of interpreting the data was developed.
A mathematical relationship has been established to describe the
12
measured behaviour of the valuable particles (pyrite) within the
settling zone of an operating industrial ash otation cell. Data
from a plant survey that incorporated axial proling of the operat-
ing ash cell has been used. This relationship links the rst order
kinetic rate constant k (for pyrite) for each size class i, to the resi-
dence time within the settling zone, ss. As the time in the settling
zone increases (i.e. as the froth interface is approached) the recov-
ery is observed to decrease primarily due to effects of the inner
cone. The relationship is given as ki ai ebi ss , where a and b are
empirically tted parameters for each size class i.
The available data has also shown that b is essentially constant
for all size classes considered. The correlation developed here
62 B. Newcombe / Minerals Engineering 64 (2014) 5162
requires validation in other ore/machine systems for further
renement and potential use within simulation and modelling
software platforms.
Acknowledgements
The author wishes to acknowledge the Corporate Strategic
Technology Solutions Group (operations support) of Barrick Gold
Corporation as the nancial sponsor of the Flash Flotation Project.
Special recognition is given to Barun Gorain as the Barrick repre-
sentative for this study, as part of the AMIRA P9 project. Thanks
are also given to Danny Hillier (Barrick Perth Ofce) for his assis-
tance throughout the duration of this work, and the site team at
Kanowna Belle for allowing the author to conduct experiments at
the plant scale.
The author also gratefully recognises the contribution of
Weiguo Xie (JKMRC) for his assistance with the internal mass
balancing of the ash otation cell. Encouragement from others
at the JKMRC (Dee Bradshaw, Elaine Wightman, Rob Morrison,
Malcolm Powell, Wayne Stange, Francois Vos and Emmy Manlapig)
is also recognised with thanks.
References
Ahmed, N., Jameson, G.J., 1985. The effect of bubble size on the rate of otation of
ne particles. Int. J. Miner. Process. 14, 195215.
Alexander, D.J., Morrison, R.D., 1998. Rapid estimation of oatability components in
industrial plants. Miner. Eng. 11, 133143.
Alford, R.A., 1992. Modelling of single otation column stages and column circuits.
Int. J. Miner. Process. 36, 155174.
Amini, E., Alexander, D.J., Wightman, E., 2009. Using laboratory scale otation
testing to predict pilot scale otation performance. In: Proceedings 41st
Canadian Mineral Processors, Ottawa.
AMIRA, 2012. P9O Final Technical Report, pp. 125128.
Barbian, N., Ventura-Medina, E., Cilliers, J.J., 2003. Dynamic froth stability in froth
otation. Miner. Eng. 16, 11111116.
Barbian, N., Hadler, K., Cilliers, J.J., 2006. The froth stability column: measuring froth
stability at an industrial scale. Miner. Eng. 19, 713718.
Bazin, C., Proulx, M., 2001. Distribution of reagents down a otation bank to
improve the recovery of coarse particles. Int. J. Miner. Process. 61, 112.
Bazin, C., Grant R., Cooper, M., Tessier, R., 1994. Prediction of metallurgical
performance as a function of neness of grind. In: Proceedings 26th Canadian
Mineral Processors, Ottawa.
Bravo, A., Torem, M., Monte, M., Dutra, A., Tondo, L., 2005. The inuence of particle
size and collector on the otation of a very low grade auriferous ore. Miner. Eng.
18, 459461.
Deglon, D.A., 2005. The effect of agitation on the otation of platinum ores. Miner.
Eng. 18, 839844.
Dobby, G., Finch, J., 1991a. Column otation a selected review, Part I. Int. J. Miner.
Process. 33, 343354.
Dobby, G., Finch, J., 1991b. Column otation a selected review, Part II. Miner. Eng.
4, 911923.
Fan, L.S., 1989. Gasliquidsolid uidization engineering. Butterworths Ser. Chem.
Eng (Chapter 4).
Farrokhpay, S., 2011. The signicance of froth stability in mineral otation a
review. Adv. Colloid Interface Sci. 166, 17.
Gontijo, C.F., Fornasiero, D., Ralston, J., 2007. The limits of ne and coarse particle
otation. Canad. J. Chem. Eng. 85, 739747.
Gorain, B.K., Franzidis, J.P., Manlapig, E.V., 1997. Studies on impeller type, impeller
speed and air ow rate in an industrial scale otation cell. Part IV Effect of
bubble surface area ux on otation performance. Miner. Eng. 10, 639654.
Grano, S., 2006. Effect of impeller rotational speed on the size dependent otation
rate of galena in full scale plants. Miner. Eng. 19, 13071318.
Grano, S.R., Akroyd, T., Mular, M.A., 2007. A model study of copper recovery
optimisation at Freeport Indonesia. In: Proceedings 9th Mill Operators
Conference, Freemantle 1921st March, pp. 2537 (AusIMM).
Kallioinen, J., Heiskanen, K., 1993. Effective otation of a difcult nickel-ore based
on intelligent mineral technology. Miner. Eng. 6, 917928.
Kelebek, S., Nanthakumar, B., 2007. Characterisation of stockpile oxidation of
pentlandite and pyrrhotite through kinetic analysis of their otation. Int. J.
Miner. Process. 84, 6980.
Klimpel, R.R., Isherwood, S., 1991. Some industrial implications of changing frother
chemical structure. Int. J. Miner. Process. 33, 369381.
Lamberg, P., Bernal, L., 2009. Modelling and simulation of SkimAir ash otation as
a part of grinding-otation circuit case study in Esperanza, Chile. 41st Annual
Meeting Can. Min. Proc., Canada, January 2009, pp. 2938.
Lins, F.F., Adamain, R., 1993. The inuence of some physical variables on gold
otation. Miner. Eng. 6, 267277.
Luttrell, G.H., Yoon, R.H., 1991. A otation column simulator based on
hydrodynamic principles. Int. J. Miner. Process. 33, 355368.
Massinaei, M., Kolahdoozan, M., Noaparast, M., Oliazadeh, M., Yianatos, J.,
Shamsadini, R., Yarahmadi, M., 2009. Hydrodynamic and kinetic
characterisation of industrial columns in rougher circuit. Miner. Eng. 22, 357
365.
McIvor, R.E., Finch, J.A., 1991. A guide to interfacing of plant grinding and otation
operations. Miner. Eng. 4, 923.
Mills, P., OConnor, C.T., 1990. The modelling of liquid and solids mixing in a
otation column. Miner. Eng. 3, 567576.
Murphy, B., 2012. Outotec, Private Communication.
Newcombe, B., 2013a. Predicting plant scale ash otation performance
Validation of laboratory methodology and applications for use. Miner. Eng.
57, 5767.
Newcombe, B., 2013b. A practical guide for operation, characterisation and
performance prediction of industrial ash otation cells. In: Proceedings
Flotation 13, Cape Town. MEI.
Newcombe, B., Bradshaw, D., Wightman, E., 2012a. Flash otation and the plight of
the coarse particle. Miner. Eng. 34, 110.
Newcombe, B., Bradshaw, D., Wightman, E., 2012b. Development of a laboratory
method to predict plant ash performance. Miner. Eng. 39, 228238.
Newcombe, B., Bradshaw, D., Wightman, E., 2013a. Hydrodynamics of an operating
ash otation cell. Miner. Eng. 41, 8696.
Newcombe, B., Wightman, E., Bradshaw, D., 2013b. The role of a ash otation
circuit in an industrial refractory gold concentrator. Miner. Eng. 53, 5773.
Pietroben, M., Grano, S., Greet, C., 2004. Matching laboratory and plant performance
a case study of the Elura lead circuit, Pasminco Australia Limited. Miner. Eng.
17, 811824.
Polat, M., Chander, S., 2000. First-order otation kinetics models and methods for
estimation of true distribution of otation rate constants. Int. J. Miner. Process.
58, 145166.
Ross, V.E., 1990. A study of the froth phase in large-scale pyrite otation cells. Int. J.
Miner. Process. 30, 143157.
Ross, V.E., 1991b. An investigation of sub-processes in equilibrium froths II: The
effect of operating conditions. Int. J. Miner. Process. 31, 5171.
Runge, K., 2009. Modelling of ore oatability in industrial otation circuits. PhD
Thesis, University of Queensland.
Runge, K., McMaster, J., Ijsselstijn, M., Vien, A., 2007. Establishing the relationship
between grind size and otation recovery using modelling techniques. In:
Proceedings Flotation 07, Cape Town.
Savassi, O.N., 1998. Direct estimation of the degree of entrainment and the froth
recovery of attached particles in industrial otation cells. PhD Thesis, University
of Queensland.
Savassi, O.N., 2006. Estimating the recovery of size-liberation classes in industrial
otation cells: a simple technique for minimising the propagation of error. Int. J.
Miner. Process. 78, 8592.
Schubert, H., 2008. On the optimisation of hydrodynamics in ne particle otation.
Miner. Eng. 21, 930936.
Seaman, D.R., Manlapig, E.V., Franzidis, J.P., 2006. Selective transport of attached
particles across the pulpfroth interface. Miner. Eng. 19, 842851.
Senior, G.D., Shannon, L.K., Trahar, W.J., 1994. The otation of pentlandite from
pyrrhotite with particular reference to the effects of particle size. Int. J. Miner.
Process. 42, 169190.
Smith, D.N., Ruether. J.A., Shah, Y.T., Badgujar, M.N., 1986. Modied sedimentation
dispersion model for solids in a three-phase slurry column. A.I.Ch.E.J. 32, 426
436.
Sutherland, D.N., 1989. Batch otation behaviour of composite particles. Miner. Eng.
2, 351367.
Suthers, S.P., Clout, J., Donskoi, E., 2004. Prediction of plant performance using feed
characterisation an emerging tool for plant design and optimisation. In:
Proceedings Metallurgical Plant Design and Operating, Strategies, pp. 203217
(AusIMM).
Trahar, W.J., 1976. The selective otation of galena from sphalerite with
special reference to the effects of particle size. Int. J. Miner. Process. 3, 151
166.
Tsatouhas, G., Grano, S.R., Vera, M., 2006. Case studies on the performance and
characterisation of the froth phase in industrial otation circuits. Miner. Eng.
19, 774783.
Tuteja, R.K., Spottiswood, D.J., Misra, V.N., 1994. Mathematical models of the
column otation process a review. Miner. Eng. 7, 14591472.
Vianna, S., 2004. The effect of particle size, collector coverage and liberation on the
oatability of galena particles in an ore. PhD Thesis, University of Queensland.
Welsby, S., Vianna, S., Franzidis, J.-P., 2010. Assigning physical signicance to
oatability components. Int. J. Miner. Process. 97, 5967.
Xu, M., Finch, J.A., 1992. Solids mixing in the collection zone of otation columns.
Miner. Eng. 5, 10291039.
Yianatos, J.B., Bergh, L.G., Condori, P., Aguilera, J., 2001. Hydrodynamic and
metallurgical characterisation of industrial otation banks for control
purposes. Miner. Eng. 14, 10331045.
Yianatos, J.B., Bergh, L.G., Diaz, F., Rodriguez, J., 2005. Mixing characteristics of
industrial otation equipment. Chem. Eng. Sci. 60, 22732282.