Minerals Engineering 18 (2005) 762–771

This article is also available online at:

www.elsevier.com/locate/mineng

Using the SUPASIM flotation model to diagnose and

understand flotation behaviour from laboratory through to plant
Martyn P. Hay *

Eurus Mineral Consultants, P.O. Box 1249, Cramerview, Gauteng 2060, South Africa

Received 6 December 2004; accepted 11 January 2005

Available online 31 May 2005

Abstract

For a model to be successful it must address the concerns and needs of a target audience. Further, the model must demonstrate
the same patterns of behaviour as the real system, and need only be complex enough to meet this objective. If the model is too com-
plex it can become unusable to those who need it most.
With this in mind, SUPASIM has been developed for the practical metallurgist requiring only laboratory flotation rate data as
input. Some 35+ successful plant simulations have been completed since 1985. The approach of SUPASIM is to characterise an ore,
organise information in an understandable way and add structure to the complex flotation situation and hence support the metal-
lurgist to interpret the relationship between variables and optimise the flotation process.
The paper describes the basis on which the model was developed and sets out to provide an insight into interpreting laboratory
flotation kinetics and their physical meaning in a production plant and the influence floatable gangue has on float performance and
plant design. A case study illustrates how the ratio of slow floating mineral to gangue can influence circuit design and improve
recovery on a plant scale.

2005 Elsevier Ltd. All rights reserved.

Keywords: Froth flotation; Flotation kinetics; Modelling; Precious metal ores; Process optimisation

1. Introduction an average of all streams in a base metal sulphide circuit

and floatable gangue constitutes probably 92%. The
There is an enormous amount of useful data in the re- easily recoverable, fast floating portion of mineral is
sults of a laboratory scale rate test. Each variable of recovered in about the first quarter to one third of a cir-
fraction and rate, for mineral, metal and gangue has a cuit. Thereafter, ore performance, circuit design and the
physical meaning in the plant. The link between float capex that goes with it is predominantly about how not
kinetics and the plant in respect of recovery, grade, pulp to float the gangue! On average the slow floating rate of
density, pulp flowrate, residence time, number of stages mineral is 0.0300 and that of gangue is 0.0023. Once the
of cleaning and circuit configuration has been estab- stream masses are taken into consideration it is not sur-
lished by the author in the process of developing prising that plants struggle to convert rougher recovery
SUPASIM. in the last half of the rougher bank into final concentrate
The debilitating effect that floatable gangue has upon recovery—most, if not all ends up as cleaner tailings. A
plant performance may not be fully appreciated. Take kinetic balance of the slow floating fractions is as
follows:

*
Tel.: +27 11 2175412; fax: +27 11 884 5852. Values: 8% mass · 0.0300 = 0.0024
E-mail address: [email protected] Gangue: 92% mass · 0.0023 = 0.0021

0892-6875/$ - see front matter
2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mineng.2005.01.029
 M.P. Hay / Minerals Engineering 18 (2005) 762–771 763

The kinetic differential of slow value to gangue (the slow make sense of and manage future events. But attempts
floating ratio—SFR) is 13.04, but when the masses in to deal with nonlinear dynamic systems, such as flota-
the float plant in which they are active are brought into tion, using ordinary processes of description and debate
the equation the differential drops to only 1.14! lead to inconsistencies and mental models are often
A marked improvement in recovery can be realised if logically incomplete (Forrester, 1987). Very often,
SFR can be beneficially manipulated. An operating strat- actual behaviour differs from expected behaviour.
egy to do just this was successfully developed by Mintek The development of the SUPASIM flotation model is
in the early 1990s and called the ‘‘Mintek two-concen- an attempt to organise, clarify and unify flotation data
trate’’ process. The kinetic mechanism of this process is and to help to communicate how structure (kinetics)
briefly explored via a case study of an operating plant. creates behaviour (flotation response). Its aim is to
augment the individualÕs mental model and bring about
a better understanding of the flotation process. The
2. Terminology overall objective of SUPASIM is to demonstrate the
same general patterns of behaviour as the real system
2.1. Acronyms and to be only complex enough to meet this need. There
are a considerable number of variables that influence
Various acronyms relating to the two component flotation behaviour and a model which accommodates
Kelsall equation and flotation kinetics are used in the these is too complex and becomes unusable to those
report. who need it most. By applying the Pareto (80/20) princi-
ple 14 variables were identified as being the minimum
IPF fast floating fraction of PGMs number required to account for +90% of flotation
kPF fast floating rate of PGMs behaviour,
kPS slow floating rate of PGMs
FFR fast floating ratio (kPF/kGF), the fast floating • three kinetic values of metals (and any others
flotation rate of PGMs relative to gangue assayed): IMF, kMF and kMS;
SFR slow floating ratio (kPS/kGS), the slow floating • three kinetic values of floatable gangue: IGF, kGF and
flotation rate of PGMs relative to gangue kGS;
• an empirical relationship developed between cell air
In all cases, I = fraction; k = rate, F = fast and rate with its associated water rate constant (two
S = slow, P = PGMs; G is substituted for gangue and variables);
M is substituted for metal or mineral. • six scale-up factors which accommodate the differ-
ence between laboratory and pilot or production
plant scale. One for each kinetic variable of metal
3. The basis on which SUPASIM has been developed and gangue.

Since the mid 1980s the approach used to develop The starting point is the laboratory rate test. Kinetics
SUPASIM has been grounded in systems thinking as measured from this test account for an oreÕs flotation
applied in the field of System Dynamics (Website). Rich- behaviour and this is carried through to predicting
mond (1994) defines systems thinking as Ôthe art and behaviour on a larger scale by applying scale-up factors.
science of making reliable inferences about behaviour To date, predictions of production plant recovery from
by developing an increasingly deep understanding of laboratory data agree with a correlation coefficient (r2)
underlying structureÕ. of 0.9421 (Hay and Martin, 2004). The full range of
The average flotation circuit is a complex system and an oreÕs kinetics and possible behaviours (responses) is
complex systems defy intuitive solutions. The behaviour measured by testing an array of metallurgical condi-
of an ore and its mass balance in a typical circuit is most tions. Each different condition brings about a new
likely unsolvable by inspection particularly when a vari- response and set of kinetics which is either better or
ety of operating conditions are considered. Determining worse than the chosen benchmark.
the likely response of a system can be approximated by
benchmarking against a similar operation although the 3.1. Theoretical foundation
accuracy is dependant upon the relationship between
the two operations being linear or at least close to it. Facts do not cease to exist because they are ignored—
This, together with experience creates an individualÕs Aldous Huxley.
mental model which according to personal construct If error is corrected whenever it is recognised as such,
theory is a means whereby individuals make sense of the path of error is the path of truth—Hans Reichenbach.
their world by comparing and contrasting events, and, Model development has been by a process that is
as a result, detect patterns and themes that help them iterative, involving a certain amount of trial and error,
764 M.P. Hay / Minerals Engineering 18 (2005) 762–771

and often requiring significant time and effort to come to of sulphide minerals has taken place (Fig. 2) and an
fruition. Development is continuous and the approach ore that has been subjected to considerable alteration
has largely adhered to guidelines described by Homer processes (Fig. 3).
(1996), which are, On milling, the minerals in the altered ore have a lower
degree of liberation and a greater proportion of chatted
• one of evaluation and revision and by an insistence
upon empirical evidence to support hypothesis and
formulations,
• where all discrepancies between model and evidence
are investigated and their causes isolated to determine
whether the model can not only reproduce history,
but also do so for the right reasons,
• recognising model shortcomings and following
through with solid improvements,
• continuously update the model to incorporate new
data and capture changes in the relationships between
variables. Each update reflects a deeper understand-
ing of the issues and helps to clarify the precise mean-
ing of elements in the model,
• greatest success is gained by not focussing on fixing
the problem, but by understanding what is causing it. Fig. 1. Unaltered ore: high degree of liberation and recovery.

3.2. Target audience

Modelling flotation may be seen as addressing the

impacts that cannot be easily thought through, whereas
direct consequences can be thought through. This may
mean that the effects cannot be fully understood and
therefore quantified. The purpose of SUPASIM is to
support metallurgists working on multifaceted problems
by adding structure and helping explication. This sug-
gests that an easy to use model would aid in the han-
dling of such problems, so allowing the metallurgist to
explore complex issues and negotiate an understanding
of them.
The ultimate success of a model is one that addresses
the concerns of a target audience and changes the way
people think about a system (flotation). The target audi- Fig. 2. Silicates altered more than the sulphides. Mineral recovery
ence is therefore any metallurgist/company with access good, but gangue component very floatable.
to standard laboratory equipment. Thus at the outset
model requirements were defined which would not
require large investments of time and money to acquire
the necessary inputs.

4. Structure and behaviour

In the context of this paper structure is defined as the

oreÕs physical make-up including the associations of its
values with sulphide and oxide minerals, i.e. its mineral-
ogy. Structure determines flotation behaviour which can
be numerically described by a set of kinetics estimated
from KelsallÕs two-component equation. As an illustra-
tion consider a mineralogically clean and essentially Fig. 3. Altered ore: overgrowth of sulphides by secondary silicate
unaltered ore (Fig. 1), an ore where medium alteration minerals. Low to medium degree of liberation and recovery.
 M.P. Hay / Minerals Engineering 18 (2005) 762–771 765

mineral-silicate particles compared to the unaltered ore PGMs) is apparent. The better flotation response is cor-
giving a slower flotation response and a lower recovery. related with larger values of fast floating PGMs (IPF and
The recovery–time curves of all three together with their kPF) indicating a higher association with sulphides and
flotation kinetics are shown in Fig. 4 and mass recovery good liberation. The correlation is less obvious for mass
curves in Fig. 5. recovery. The unaltered and partially altered ores have a
The relationship between mineral flotation response similar profile despite very different fast floating gangue
and the associated kinetics, IPF, kPF and kPS (in this case kinetics. If Fig. 5 was the only means of analysing the

TYPICAL RECOVERY-TIME CURVES AND ASSOCIATED FLOTATION KINETICS

Unaltered Silicate altered more than sulphide Altered

100

95 Unaltered

90
LABORATORY ROUGHER CONCENTRATE

85

80 Silicate altered more

than sulphide
75
RECOVERY %

70

65
Altered
60

55
Mineral Gangue
50 IPF k PF k PS IGF kGF kGS
45 Unaltered 0.900 1.600 0.115 0.049 0.358 0.0063
40 Sulphides altered
0.766 0.642 0.097 0.152 0.032 0.0064
less than silicates
35
Altered 0.510 1.130 0.055 0.249 0.353 0.0043
30
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
LABORATORY ROU GHER RATE TIME (min)

Fig. 4. Typical mineral flotation response and kinetics of a ‘‘clean’’ and altered ore.

TYPICAL MASS-TIME CURVES AND ASSOCIATED FLOTATION KINETICS

Unaltered Silicate altered more than sulphide Altered

34
32
LABORATORY ROUGHER CONCENTRATE MASS %

30
28
26
24
22
Altered
20
18
16
14
12
Mineral Gangue
10
IPF kPF kPS IGF kGF kGS
8 Unaltered
Unaltered 0.900 1.600 0.115 0.049 0.358 0.0063
6
Sulphides altered
4 Silicate altered more 0.766 0.642 0.097 0.152 0.032 0.0064
less than silicates
2 than sulphide
Altered 0.510 1.130 0.055 0.249 0.353 0.0043
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
LABORATORY ROUGHER RATE TIME (min)

Fig. 5. Typical mass flotation response and kinetics of a ‘‘clean’’ and altered ore.
766 M.P. Hay / Minerals Engineering 18 (2005) 762–771

character of the floatable gangue then it would be as- tently confirmed irrespective of size, power input, geom-
sumed that they behave in a similar manner. However etry and type of commercial float cell and points to the
Figs. 1 and 2 show that the structure of the gangue is fact that the inherent floatability of the fast floating frac-
very different and in their respective plant operations tion exceeds any effect that the mechanical and hydrau-
unaltered ore is treated with 150 g/t depressant and the lic environment may exert on it. It appears that the
partially altered ore is treated with 450–600 g/t. Final requirements of a commercial cell is to provide adequate
concentrate mass pull in the plant is 3.46% in both cases. suspension, mixing and bubble size/density and the
The level of depressant addition is influenced but not kinetics of the mineral will do the rest! Since the mid
primarily driven by IGF (fast floating fraction of 1980s, case studies of laboratory to plant simulations
gangue). have been conducted involving cells ranging from 30 l
to 150 m3 without a single anomaly to this observation
being encountered. There is little data supporting the
5. Laboratory kinetics and plant performance case for a maximum, or preferred maximum cell size
based on an assumed limit of flotation efficiency.
Clearly there is a link between structure and behav- Note that this observation applies only to the fast
iour. But how is behaviour measured in the laboratory floating fraction of mineral or metal.
tied-in with plant performance and can it indicate an The laboratory test accentuates the flotation of fast
appropriate design and circuit configuration? floating gangue and this is balanced by an under estima-
tion of the rate of slow floating gangue. In a plant the
5.1. Laboratory vs. plant behaviour effect of slow floating gangue has considerably more im-
pact than in the laboratory. The importance of the slow
The discrepancy in flotation performance between floating ratio (SFR) in optimising float performance and
laboratory and plant is well known and has been ex- determining circuit design is illustrated below.
plained in a variety of ways, mostly in terms of physical
properties and a difference in efficiency. Despite this, the
batch test measures flotation response in a way that is 5.2. Residence time
translatable to larger scales once the methodology to
do so has been empirically derived and validated. The Minimum residence time in the roughers should be
intensity of the laboratory cell imparts certain peculiar- sufficient to capture the fast floating fraction. Figs. 6
ities to how mineral and gangue float relative to one and 7 compare recovery–time profiles for a UG2 and
another. However, an overestimation in one fraction is Merensky ore tested in the laboratory with the plant
balanced by compensation in another. profile and the simulated profile using laboratory gener-
If a mineral or floatable gangue particle behaves in a ated kinetics. Each graph shows the time at which fast
certain way in a laboratory cell how does it behave in the floating PGMs are recovered in the laboratory float,
plant? The magnitude of the scale-up factor between actual plant and simulated plant. On the plant profile,
laboratory and plant is an ideal measure of this differ- the diagonally shaded arrow marks final concentrate
ence in behaviour and takes into account all the various recovery achieved in the plant. These two ores were cho-
dissimilarities in mechanical and physio-chemical sen because their PGM fast floating fractions and rates
efficiency as well as batch vs. continuous operation and their fast and slow floating rates of gangue are very
(Table 1). similar (Table 2).
The first thing Table 1 illustrates is that the fast float- For comparison the UG2 ore profile has been plotted
ing and easily recoverable fraction of metal or mineral is in Fig. 7. Considering the similarity in their kinetics the
recovered in the plant to almost the same extent as in the profiles are more different than would be expected and
energy intensive laboratory cell, although the rate at highlight the marked effect of fast floating gangue (IGF)
which this occurs is much reduced. This has been consis- at the beginning of the float and slow floating PGMs
(kPS) at the end. The nominal rougher residence time suf-
ficient to capture all PGMs that are recovered to final con-
centrate is 23 min for UG2 ore and 35 min for Merensky
Table 1 ore. Based on case studies conducted to date on PGM and
Magnitude of scale-up factors BMS ores, in a plant environment the fast floating frac-
Metal and/or mineral Fast fraction (IMF) !
1:1 tion is typically recovered in a nominal residence time of
Fast rate (kMF) !
1:1 25 min, and therefore, this is considered to be the mini-
Slow rate (kMS) ! <1.0 mum residence time in a primary stage of flotation.
Floatable gangue Fast fraction (IGF) ! 1.0 Minimum rougher residence time in secondary or
Fast rate (kGF) ! 1.0 cleaner tailings regrind circuits is greater and is deter-
Slow rate (kGS) ! 1.0 mined in the same way according to their kinetics.
 M.P. Hay / Minerals Engineering 18 (2005) 762–771 767

Fig. 6. Laboratory, plant and simulated plant PGM recovery–time curves. UG2 ore.

Fig. 7. Laboratory, plant and simulated plant PGM recovery–time curves. Merensky ore.

Table 2
Kinetics of UG2 ore and Merensky ore
Mineral Gangue FFR SFR
IPF kPF kPS IGF kGF kGS IPF/IGF kPS/kGS
UG2 ore No 1 0.7609 1.3571 0.0767 0.0881 0.5452 0.0028 8.64 27.39
Merensky ore 0.7693 1.5019 0.0216 0.0527 0.5483 0.0024 14.60 9.00
768 M.P. Hay / Minerals Engineering 18 (2005) 762–771

5.3. Recovery and slow floating ratio (SFR) of Taggart but ‘‘largely ignored in the push for circuit
simplification and larger float cells’’. Another example
Unless unusual circumstances are in effect, which is of particle–particle interference is the effect of floatable
outside the scope of this paper, a plant should always re- gangue upon the behaviour of recoverable values con-
cover its fast floating fraction of mineral/metal. Beyond tained within minerals and metal alloys. Similarly for
this, recovery is sourced from the slow floating fraction coarse and fine particles operating strategy and circuit
and is referred to as incremental recovery (i.e. total design needs to be tailored to provide the maximum
recovery less fast floating percentage). For this to hap- opportunity for values to be recovered selectively from
pen the differential rate of flotation between slow float- gangue.
ing mineral/metal and slow floating gangue must be The effect of SFR on plant recovery is one of the key
great enough to allow the values to be upgraded to a aspects of flotation. Manipulating this can have a
point at which they are recovered into final concentrate. marked impact on performance as is shown in the case
This differential is called the slow floating ratio (SFR) study below.
and is the ratio of the slow floating rate of values
(PGMs) divided by gangue or kPS/kGS. If SFR is less 5.5. Manipulating the slow floating ratio: the Mintek
than a certain value then that increment of recovery in ‘‘two concentrate’’ process
the roughers greater than IPF drops out in the cleaners
and becomes a circulating load in the cleaner tailings. One of the means of modifying SFR is by the addi-
SFR has a mineralogical significance and is indicative tion of depressant. If mineral–gangue liberation is good
of the degree of alteration, mineral–gangue chatting and then it will be possible to selectively change gangue
liberation. kinetics relative to mineral. The fast floating fraction
Figs. 6 and 7 are annotated with the difference of mineral is then less encumbered by floatable gangue
between plant recovery and the fast floating percentage. and can lead to an improvement in both recovery and
The UG2 plant does not recover its entire fast floating grade. Thus without altering the value of fast floating
fraction and the Merensky plant has an incremental fraction (usually achieved by finer milling) performance
recovery of +2.7%. An inspection of the shape of the improves as a result of the environment being tailored to
plant recovery–time profiles and the SFR values in its particular needs. A typical example of the selective
Table 2 shows that the value of SFR drives incremental action of depressant is shown in Table 3. Gangue float-
recovery. A rougher residence time of 40–45 min would ability is progressively reduced with increasing depres-
be adequate in the Merensky plant and a minimum time sant addition but so too are both fast and slow
of 30 min is preferable in the UG2 plant. floating rates of PGMs. In this case SFR remains constant
which usually indicates that slow floating mineral and
5.4. Improving recovery: metallurgical conditions and gangue are in the form of binary particles and further
circuit configuration grinding is required to improve the degree of liberation
between the two. IPF increases as a result of being less
In a paper laying out the requirements for improving encumbered by floatable gangue. The effect of overdosing
recovery of fine particles, Pease et al. (2004) comments can be seen at 450 g/t where kPS is significantly reduced.
that ‘‘sub 20 lm particles do perform poorly because Fig. 8 shows the mass balance of the plant treating
they are mixed with coarser particles with much different UG2 prior to optimisation.
needs’’ and ‘‘that they will (float) in the right condi- The circuit was changed to produce two concentrates
tions. . .and you must design your circuit to suit the and the addition of depressant was both increased and
needs of fine particles’’. proportioned between the two cleaner banks. The results
Attempting to simultaneously float coarse and fine were spectacular as shown in Fig. 9. Recovery improved
particles results in poor performance and is an example 4.2% and grade increased from 405 to 660 g/t.
of particle–particle interference in the flotation process. This is the application of the Mintek ‘‘two-concen-
As Pease points out this has been known since the days trate’’ process developed in the early 1990s specifically

Table 3
Effect of gangue depressant upon flotation kinetics
Additional of depressant IPF kPF kPS IGF kGF kGS IPF/IGF kPS/kGS
Plant Ro feed 90 g/t 0.7849 1.3752 0.0225 0.0235 0.4755 0.0007 33.4 31.1
Plant Ro feed 135 g/t 0.8114 0.9960 0.0113 0.0220 0.2196 0.0005 36.9 22.7
Plant Ro feed 180 g/t 0.8341 0.7420 0.0142 0.0192 0.1973 0.0005 43.5 28.9
Plant Ro feed 270 g/t 0.8578 0.6631 0.0110 0.0146 0.2850 0.0004 58.9 26.9
Plant Ro feed 450 g/t 0.8512 0.6489 0.0042 0.0140 0.4155 0.0002 61.0 19.0
 M.P. Hay / Minerals Engineering 18 (2005) 762–771 769

for the South African Platinum industry. The process

Roughers (34 min nom residence) utilises differences in mineral/gangue floatability by
addition of depressant in combination with a change
Tailings in circuit configuration as illustrated.
At this point it is necessary to distinguish between a
2 stages Cl Scav. MF2 circuit and a MF1 circuit producing two concen-
3 stages of trates. Both are often described as a two concentrate pro-
cleaning cess. The MF2 circuit produces two concentrates by
virtue of it being two mill/float plants in series. Fig. 10
shows the flowsheets of both circuits. The second stage
of the MF2 circuit fulfils a regrind function to capture
Final Concentrate PGMs and base metal sulphides locked in silicates
0.9% mass 405 g/t 78.7% recovery whose liberation size is finer than the primary grind.
Fig. 8. UG2 plant prior to optimisation of SFR. SUPASIM is used to illustrate the mechanism under-
lying the two-concentrate process. The mass balance in
Fig. 8 is reproduced as ‘‘original’’ in Table 4. Cases 1–4
Roughers (34 min nom residence) are simulations of the two-concentrate circuit in Fig. 9
with progressive variation of the flotation kinetics as de-
Tailings tailed in Table 5. The kinetics were varied in a way that
RT:feed matched the general effect seen in Table 3 such that cases
RT : Ro Conc 2 stages of cleaning 1–4 mimic successive increases in depressant. A signifi-
3 stages of cant increase in grade is seen without changing kinetics
cleaning (equivalent to the reagent suite remaining unchanged)
RT : Cl T and moving to a two-concentrate circuit (‘‘original’’ to
case 1). Case 2 increases the proportion of PGM recovery
from the secondary concentrate by reducing kPF and
Primary Concentrate Secondary Concentrate grade is increased by reducing kGS. A combined recovery
0.32% mass 0.32% mass
similar to plant actual is simulated in case 4 and shows
1121 g/t 212 g/t
that the same magnitude of recovery and grade increase
69.4% recovery 13.5% recovery
can be obtained and pinpoints the specific changes in
kinetics for this to happen.
Combined Concentrate
Although combined recovery was close to actual, the
0.64% mass 660 g/t 82.9% recovery
simulated contribution from primary and secondary
Fig. 9. UG2 plant after optimisation of SFR. concentrates did not agree (Table 4). This is because

MINTEK "TWO-CONCENTRATE" PROCESS

Roughers

Tailings

Primary Conc No 1 Primary Conc No 2

MF2 (MILL/FLOAT-MILL/FLOAT) CIRCUIT

Roughers

Tailings

Primary Conc Secondary Conc

Fig. 10. MF2 and MF1 ‘‘two-concentrate’’ circuits.

770 M.P. Hay / Minerals Engineering 18 (2005) 762–771

Table 4
Simulation of the two concentrate process: UG2 ore
Actual concentrate Simulated concentrate
Primary Secondary Combined Primary Secondary Combined
% Mass Original 0.90 0.90 0.92 0.92
Grade g/t 405 405 390 390
% PGM recovery 78.7 78.7 76.0 76.0
% Mass Case 1 0.34 0.24 0.58
Grade g/t 1.64 31 638
% PGM recovery 74.5 1.5 76.0
% Mass Case 2 0.31 0.39 0.70
Grade g/t 1113 69 529
% PGM recovery 70.5 5.6 76.1
% Mass Case 3 0.32 0.32 0.64 0.29 0.41 0.70
Grade g/t 1121 212 631 1265 73 562
% PGM recovery 69.4 13.5 82.9 74.7 6.1 80.8
% Mass Case 4 0.32 0.32 0.64 0.27 0.36 0.63
Grade g/t 1121 212 631 1378 80 632
% PGM recovery 69.4 13.5 82.9 75.8 5.9 81.7

Table 5
Flotation kinetics for the simulations in Table 4
IPF kPF kPS IGF kGF kGS FFR SFR
IPF/IGF kPS/kGS
Original 0.7609 1.3571 0.0767 0.0651 0.5452 0.0028 11.69 27.39
Case 1 0.7609 1.3571 0.0767 0.0651 0.5452 0.0028 11.69 27.39
Case 2 0.7609 0.9000 0.0767 0.0651 0.5452 0.0009 11.69 85.22
Case 3 0.7609 0.9000 0.0767 0.0294 0.5452 0.0009 25.88 85.22
Case 4 0.7609 0.9000 0.0767 0.0294 0.4000 0.0009 25.88 85.22

Table 6
Simulation of the two concentrate process: Pd rich Merensky ore
Actual concentrate Simulated concentrate
Primary Secondary Combined Primary Secondary Combined
% Mass I Original 0.84 0.84 0.85 0.85
Grade g/t 240 240 241 241
% PGM recovery 75.7 75.7 77.0 77.0
% Mass I Case 1 0.27 0.19 0.46
Grade g/t 734 11 443
% PGM recovery 76.1 0.8 76.9
% Mass Case 2 0.19 0.28 0.47
Grade g/t 1057 25 438
% PGM recovery 75.1 2.6 77.7
% Mass Case 3 0.28 0.34 0.62
Grade g/t 732 10 333
% PGM recovery 76.7 1.3 78.0

simulation of the process is based on kinetics from a rate simulated using kinetics from rate tests performed on
test conducted on rougher feed (RT: feed, Fig. 9). the rougher concentrate and cleaner tailings streams
Depressant is added to both rougher concentrate and (RT: Ro Conc and RT: Cl T respectively). For the pur-
cleaner tailings streams to produce the required change poses of this paper a simpler approach was taken to sim-
in mineral/gangue floatability. The actual 69.4/13.5 ulate the two concentrate circuit by changing the
recovery split from the two cleaner banks can be kinetics of the rougher feed. This allows an easier assess-
 M.P. Hay / Minerals Engineering 18 (2005) 762–771 771

Table 7
Flotation kinetics for the simulations in Table 6
IPF kPF kPS IGF kGF kGS FFR SFR
IPF/IGF kPS/kGS
Original 0.7693 1.5019 0.0216 0.0527 0.5483 0.0026 14.59 8.25
Case 1 0.7693 1.5019 0.0216 0.0527 0.5483 0.0026 14.59 8.25
Case 2 0.7693 0.9945 0.0216 0.0176 0.4021 0.0009 43.71 23.96
Case 3 0.7693 0.9945 0.0216 0.0176 0.4021 0.0009 43.71 23.96

ment to be made regarding the effect that depressant has mineral is caused by the floatability of the gangue frac-
upon the ore and also allows the changes to be refer- tion. The Mintek ‘‘two-concentrate’’ process provides a
enced with the data in Table 3. solution by making use of the difference in relative float-
As a comparison the exercise is duplicated for the Pd- ability between mineral and gangue to significantly
rich Merensky ore (Table 6) with its kinetics being mod- improve recovery and grade. In one case a recovery
ified in the same ratio as in Table 5. A PGM recovery improvement of 5% was obtained and shows that a sim-
improvement of only 1% is evident; however it is 3% ilar level of benefit is possible when this aspect of parti-
up on actual plant performance. Much less recovery in cle–particle interference is effectively tackled. A plant
the secondary concentrate is in keeping with the oreÕs scale case study highlighted the kinetic mechanism of
lower kPS and SFR (Table 7). the process. Simulation showed that, provided liberation
is adequate, the action of depressant on gangue is selec-
tive and although PGM kinetics are not increased, PGM
6. Conclusions floatability and recovery are both improved by a signif-
icant reduction in gangue kinetics.
SUPASIM has been developed to organise, clarify
and unify flotation data and to help to communicate
how structure (kinetics) creates behaviour (flotation Acknowledgements
response). Optimisation of flotation performance and
circuit flowsheet can be accomplished using the model The initiate part of this paper quotes extensively from
as well as facilitating a better understanding of the the published papers of Barry Richmond, Jack Homer
flotation process. and Jay Forrester, all leaders in the field of system
Three mineralogically diverse ores were analysed to dynamics and systems thinking.
illustrate how the alteration process changes flotation
kinetics and flotation response. The increase in floatable
gangue kinetics can be benchmarked against depressant References
addition in the associated plant.
A comparison between two PGM ores showed that Forrester, J.W., 1987. Lessons from systems dynamics modelling.
plant performance and circuit design can be determined System Dynamics Review 3 (2), 136–149.
from laboratory generated kinetics. It is possible to use Hay, M.P., Martin, C.J., 2004. SUPASIM: A methodology to predict
plant performance from laboratory data. In: Proceedings 36th
the characteristics of the ores to, among other aspects,
Annual Meeting of the Canadian Mineral Processors, January
determine residence time and the specific circuit config- 2004, pp. 281–299.
uration most suitable for best performance. Homer, B.H., 1996. Why we iterate: scientific modelling in theory and
Poor performance in a flotation plant is often the re- practice. System Dynamics Review 12 (1), 1–19.
sult of particle–particle interference. Interference of fine Pease, J.D., Young, M.F., Curry, D., Johnston, N.W., 2004. Improv-
particle flotation by coarse particles has been seen at ing fines recovery by grinding finer. MetPlant 2004 AusIMM.
Richmond, B., 1994. Systems thinking/system dynamics: letÕs just get
Mount Isa Mines and resolved by appropriate classifica- on with it. System Dynamics Review 10 (2–3), 135–157.
tion. Lead recovery increased by 5% and zinc recovery Website: www.systemdynamics.org provides a definition of system
by 10%. Similarly, interference of fast and slow floating dynamics.