Minerals Engineering 16 (2003) 1103–1109

This article is also available online at:

www.elsevier.com/locate/mineng

SUPASIM: a flotation plant design and analysis methodology

a,* b
M.P. Hay , C.M. Rule
a
Eurus Mineral Consultants, P.O. Box 1249, Camerview, Gauteng 2060, South Africa
b
Anglo Platinum Management Services Pty Ltd, P.O. Box 62179, Marshall town, 2107, South Africa
Received 13 May 2003; accepted 23 June 2003

Abstract
A methodology was developed in the mid-1980s to predict plant performance from standard laboratory flotation tests. The
technique is based on a simple, empirical kinetics model, and is tailored for use by the practical metallurgist.
To date the performance of more than 20 flotation plants has been predicted, encompassing copper, lead, zinc, nickel, phosphate,
pyrite, graphite, cassiterite, platinum and various slags. The method has successfully been used to predict and design four flotation
plants using data generated from bench-scale laboratory flotation testwork. All have since been commissioned, and have performed
as predicted.
The technique has now received broad acceptance within the South African mining industry, for example in the last year alone it
has been used to drive flowsheet development for seven new plants. It is also being used to assist in the optimisation of existing
operating plants.
The methodology is described in this paper. Also, verification of the methodology is presented using a number of case studies.

2003 Elsevier Ltd. All rights reserved.

Keywords: Froth flotation; Process optimisation; Modelling; Precious metal ores; Sulphide ores

1. Introduction In this work Kelsall’s unmodified equation incorpo-

rating two rate terms describing slow and fast floating
This paper presents a methodology for characterising components is used as the basis of the method. The
an ore’s flotation response and linking this to its metal- flotation behaviour of an ore is described by a set of
lurgical performance in a commercial plant. The meth- numbers, which can be compared in the evaluation of
odology incorporates a flotation simulation program ore types, mineralogy and operating conditions. In so
‘‘SUPASIM’’ which translates flotation performance doing a precise relationship is developed between met-
from bench-scale tests to full-scale plant. To date, the allurgical performance in terms of grade and recovery
successful simulation of 15 commercial flotation circuits and flotation kinetics.
treating platinum, base metal, cassiterite, graphite and SUPASIM was initially developed in the mid-1980s
phosphate ores as well as platinum and copper furnace specifically to design and simulate the operation of the
slag has validated ‘‘SUPASIM’’. Two of these plants op- Northam Platinum concentrator (Hay et al., 1994). It
erate rougher concentrate regrind circuits and two have was not possible to pilot the process and the challenge
treated more than two different ore types. ‘‘SUPASIM’’ was to proceed from 1 kg bench-scale tests to full-scale
successfully simulated these circuits including the flota- plant design and performance. At that time minimal
tion performance of the different ores. Five pilot plant published literature on the processing of Merensky ore
campaigns have also been analysed and simulated. Since was available and the existing platinum producers kept
1988 four concentrators treating platinum Merensky ore, the technology a closely guarded secret. Northam was
copper–zinc ore, copper furnace slag and platinum fur- commissioned in 1992. Predicted and operating perfor-
nace slag have been designed using only laboratory data. mance correlates very well.

2. The development of SUPASIM

*
Corresponding author. Tel.: +27-11-217-5412; fax: +27-11-884-
5852. From the outset, SUPASIM was developed with the
E-mail address: [email protected] (M.P. Hay). operator and design metallurgist in mind. The basis of
0892-6875/$ - see front matter
2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mineng.2003.06.009
1104 M.P. Hay, C.M. Rule / Minerals Engineering 16 (2003) 1103–1109

SUPASIM is highlighted in Fig. 1 where flotation per- Kelsall’s unmodified equation was chosen over the
formance is a result of the ore’s mineralogical charac- classic first-order Klimpel and Jowett/modified Kelsall
teristics, which ultimately determines circuit design. The models because it was the only model that did not use
prescription for success has been centred on ease of recovery at infinite time. Instead, this model assumes a
obtaining input data and ease of application. These are recovery of 100% at infinite time:
accommodated as; R ¼ ð100
hÞ½1
expð
kf tÞ þ h½ð1
expð
ks tÞ ð1Þ
• A standard flotation rougher rate test that can be per- where, h ¼ slow floating fraction; R ¼ per cent recovery
formed equally as well on the plant as in the labora- at time t; kf ¼ fast floating rate constant (min
1 );
tory and ks ¼ slow floating rate constant (min
1 ); t ¼ time (min).
• A simulation program, which takes the product of a To be successful, the continuous flotation process
laboratory rate test and basic information on, plant should be modelled from batch bench-scale data to a
feed tonnage and circuit size. sufficiently accurate degree so that all the process op-
tions of interest can be investigated. A method was de-
Apart from the assay and mass data from the flota- vised to quantify the behaviour and characteristics of an
tion test, inputs into the model were specifically limited ore and to establish a link between this and its perfor-
to only those factors which are easy to measure in the mance in a production-scale flotation plant.
plant––namely feed solids tonnage, pulp density, cell
volume, circuit configuration and an estimation of air
into the float cell or bank (expressed as a % of maxi- 3. Validation
mum).
The flotation process in the plant is thus characterised 3.1. Cell aeration rate, mass pull and concentrate pulp
by seven variables; density

1. A measurement of the ore’s mineral kinetics from a Airflow into a flotation cell (or bank of cells) effects
laboratory scale rate test, mass pull and concentrate pulp density. Mineral grade
2. A measurement of the ore’s gangue kinetics from a and recovery is also affected in line with the ore’s specific
laboratory scale rate test, mineralogical characteristics. The influence of airflow
3. Plant feed tonnage, was taken into account by deriving a relationship be-
4. Pulp density of fresh feed, tween concentrate froth flow (i.e. airflow into the cell
5. Cell size, measured visually on a scale of 0–100) and a water rate
6. Circuit configuration and constant (water transfer from the pulp phase into the
7. A calibration curve describing the relationship be- froth phase). The relationship was derived from a
tween percentage air into the cell and the correspond- number of case studies of operating plant conducted to
ing water rate constant. validate the model. Froth mass pull (expressed as a

Flotation performance influence diagram (Flotation "PID")

MINERALOGY

Defines
CONTROLS
FLOTATION
KINETICS
Determines

CIRCUIT DESIGN

Effects
FLOTATION
PERFORMANCE
(Grade/Recovery) &
PROFIT

Fig. 1. The link between mineralogy, flotation performance and circuit design.
 M.P. Hay, C.M. Rule / Minerals Engineering 16 (2003) 1103–1109 1105

percentage of new feed to the plant) in all these plant The scale-up factors were found to be independent of
were different ranging from, ore type and are a function of the flotation process (i.e.
bubble–particle attachment) as it occurs in the labora-
• High: phosphate (77% roughers, 23% final cleaners); tory compared to how it occurs in the production plant.
zinc (74% roughers, 25% final cleaners); lead (60% These factors take into consideration the differences
roughers, 11% final cleaners), between batch and continuous conditions such as cir-
• Medium: Pt Merensky (27% roughers, 2% final clean- culating load, recycling of plant water and contained
ers); Pt UG2 (23% roughers, 1.1% final cleaners); reagents, the degree of liberation associated with a plant
copper (17% roughers, 8% final cleaners); zinc (16% grind and flotation cell efficiency.
roughers, 5% final cleaners), SUPASIM is in the form of a series of mass-balanc-
• Low: copper (10% roughers, 0.9% final cleaners); Pt ing equations in which the proven scale-up factors are
UG2 (8% roughers, 0.8% final cleaners); Pt UG2 re- applied to laboratory rougher rate data. In this manner,
grind (2.2% roughers, 0.4% final cleaners). mass balances are generated for a commercial plant and
plant parameters such as retention time, circuit size and
The Pt Merensky and Pt UG2 plants were also op- configuration and number of cleaning stages can be
erated for a period at maximum concentrate mass pull varied to investigate metallurgical performance.
to define what grade of concentrate results at maximum
possible recovery. This management-driven test was
utilised to obtain maximum values of mass pull, con- 4. Interpretation of laboratory data
centrate pulp density, a visual measurement of mass pull
(in %) and allowed the water rate constant to be calcu- The flotation response of an ore is characterised by
lated. conducting standard rougher and cleaner rate tests.
In the model, airflow is chosen as a percentage of Changes in recovery and grade are obtained by varying
maximum (i.e. 100%) similar to a plant operator ad- test conditions. The fast and slow flotation kinetics are
justing a control valve. The model has the facility to calculated from Eq. (1) and are not only a numerical
‘‘operate’’ float cells by dialling in an air-rate, which is description of metallurgical performance but also a
equivalent to a desired mass pull. The above plant representation of the ore’s mineralogy and its liberation
measurements allowed a relationship between percent- characteristics. The mineralogist’s description and the
age air and water rate constant to be derived. This is metallurgist’s kinetics are closely linked, as one is a de-
expressed in the model as a calibration curve. Each ore scription of the other.
type has its own calibration curve and has allowed Each kinetic variable of rate and fraction including
benchmarks to be established for particular ore types, particular ratios have a physical meaning in the plant
for example the above mass pull categories of high, and describe metallurgical performance and plant design
medium and low each have a typical calibration curve parameters. Mineralogically clean ore (e.g. a good ore)
which varies very little with ore type. containing little alteration and/or oxidation products is
characterised by high concentrate recovery and grade,
3.2. The link between laboratory kinetics and plant minimal intermediary products and low circulating load.
performance The kinetics that describes the flotation performance of
a good ore has high values of IMF, kMF and kMS (fast
From a series of tests and case studies on ores from floating mineral fraction, fast floating mineral rate and
operating flotation plants, the link between laboratory slow floating mineral rate, respectively) and low values
and plant values for rates and fractions (i.e. the scale-up of IGF, kGF and kGS (fast floating gangue fraction,
factors) was established with good accuracy. In all cases fast floating gangue rate and slow floating gangue rate,
and for all ores tested the scale-up factors were found to respectively). The opposite is true of a highly altered
be similar with the values for each factor falling within a and/or oxidised ore. Typical kinetics for a clean and
narrow range. highly altered ore is shown in Table 1.

Table 1
Flotation kinetics for Ôgood’ and Ôbad’ Merensky ore
Fast gangue Slow gangue Fast PGMs Slow PGMs Laboratory Ro. concentration after 25 min
% Mass Grade (g/t) Recovery (%)
Clean ore Fractions 0.0473 0.9527 0.9213 0.0787 10.3 47.8 98.1
Rates 0.0659 0.0028 2.7811 0.0560

Highly altered ore Fractions 0.1151 0.8849 0.5422 0.4578 23.3 16.5 76.9
Rates 0.2902 0.0057 0.7215 0.0273
 1106
M.P. Hay, C.M. Rule / Minerals Engineering 16 (2003) 1103–1109
Table 2
Comparison of actual and simulated flotation plant performance
Plant or material Simulations or plant data as Head Head Head Final concentrate
specified PGMs copper nickel % New Grade Recovery Grade Recovery Grade Recovery
(g/t) (%) (%) Feed Cu (%) Cu (%) PGMs + Au PGMs + Au Ni (%)
(g/t) (%)
Great Dyke Platinum, Simulation based on average of 4.80 0.162 0.263 6.05 2.27 84.8 64 80.5 3.25 74.6
Zimbabwe all lab borecore tests. 1993
Great Dyke Platinum, Average Plant Data, November 2.97 0.114 0.199 4.62 1.76 71.4 51 79.3 2.39 55.6
Zimbabwe 96–January 99
Great Dyke Platinum, Simulation based on average of 4.35 0.160 0.260 5.11 2.43 77.8 64 75.4 3.51 69.1
Zimbabwe lab borecore tests––oxidised/nor-
mal ore mix 1993
Great Dyke Platinum, Average Plant Data––50% oxi- 3.03 0.145 0.244 4.64 1.25 40.0 47 72.3 1.90 36.1
Zimbabwe dised/normal ore mix. November
96-January 99
Great Dyke Platinum, Simulation based on average of lab 5.20 0.160 0.260 6.99 2.10 91.8 64 85.7 2.98 80.1
Zimbabwe borecore tests–normal ore. 1993
Great Dyke Platinum, Average Plant Data––normal ore 3.38 0.099 0.200 4.21 1.98 83.4 68 84.7 3.07 64.6
Zimbabwe November 96–January 99
Merensky, Rustenburg, Simulation based on mill feed 3.59 0.085 0.143 1.91 3.67 79.7 160 82.7 5.34 69.5
South Africa sample, 1988
Merensky, Rustenburg, Plant Data, 1988 3.59 0.085 0.143 1.91 3.69 82.5 154 81.9 6.15 82.3
South Africa
Nortbam Platinum Mine, Simulation based on lab borecore 7.60 0.073 0.163 3.46 1.68 79.7 198 89.9 2.54 54.0
South Africa composite tests. 1989
Northam Platinum Mine, Average plant data, July–August 5.85 0.063 0.211 3.63 1.56 84.6 143 88.8 3.14 54.1
South Africa 1998
 Final concentrate
– Head Head % New Grade Recovery Grade Zn Recovery
copper zinc feed Cu (%) Cu (%) (%) Zn (%)
(%) (%)
Palabora Copper Mine, Simulation based on mill feed – 0.51 – 1.25 35.30 83.1 – – – –
South Africa sample. Open Pit ore 2001
Palabora Copper Mine, Plant Data Open Pit ore 2001 – 0.51 – 1.24 35.20 85.0 – – – –
South Africa
Palabora Copper Mine, Simulation based on mill feed – 0.80 – 2.52 18.00 56.9 – – – –
South Africa sample. Furnace Slag 2001
Palabora Copper Mine, Plant Data Furnace Slag 2001 – 0.80 – 2.14 18.60 49.8 – – – –
South Africa
Maranda Copper Zinc Mine, Simulation: Copper float based – 1.30 17.00 4.70 24.30 88.10 9.40 2.61 – –

M.P. Hay, C.M. Rule / Minerals Engineering 16 (2003) 1103–1109

South Africa on borecore sample. 1990
Maranda Copper Zinc Mine, Simulation: Copper float based – 1.30 17.00 4.70 24.30 88.10 9.40 2.61 – –
South Africa on borecore sample. 1990
Maranda Copper Zinc Mine, Plant Data: Copper float March– – 2.70 16.30 7.66 28.00 79.50 7.70 3.62 – –
South Africa December 1992
Maranda Copper Zinc Mine, Plant Data: Copper float March- – 2.70 16.30 7.66 28.00 79.50 7.70 3.62 – –
South Africa December 1992
Maranda Copper Zinc Mine, Simulation: Zinc float based on – 0.45 17.40 28.60# 1.39 88.32# 57.30 94.21# – –
South Africa borecore sample. 1990
Maranda Copper Zinc Mine, Simulation: Zinc float based on – 0.59 14.95 24.21# 0.87 35.74# 56.95 92.24#
South Africa mill feed sample. March 1992
Maranda Copper Zinc Mine, Plant Data: Zinc float March–De- – 0.60 17.10 27.40# 1.30 59.37# 57.90 92.78# – –
South Africa cember 1992
# stage recovery (Zinc float follows after Copper float).

1107
1108 M.P. Hay, C.M. Rule / Minerals Engineering 16 (2003) 1103–1109

The various kinetic variables influence the mass bal- tonnage and plant circuit volume and configuration. Pa-
ance in a plant as follows; labora open pit ore and furnace slag were individually
treated through the same plant flotation section (Sec-
• Concentrate recovery and grade are determined pri- tion 6).
marily by IMF (fast floating mineral fraction). A por- Fig. 2 shows the linear relationship between actual
tion of IMS (slow floating mineral fraction) can be plant recovery and simulated plant recovery for a vari-
recovered at concentrate grade when kMS/kGS is ety of ores. The correlation coefficient (r2 ) is 0.942 and
above a certain threshold value. includes less accurate simulations predicting the flota-
• Concentrate grade, mass pull and pulp density are de- tion performance of impurity metals such as copper in
termined by IGF and kGF (fast floating gangue frac- sphalerite flotation, zinc in chalcopyrite flotation and
tion and rate respectively). zinc in galena flotation. The numbered markers refer to
• Cleaner tailings mass is determined primarily by IGF primary metals, i.e. the recovery of platinum group
and kGS (slow floating gangue rate) and to a lesser metals (PGMs) in a PGM float, copper in a copper float,
extent kGF. lead in a lead float etc.
• The slow floating gangue rate, kGS, strongly effects
grade, mass pull and pulp density of every stream 5.1. The number of variables required to simulate flotation
in the plant especially cleaner tailings. It, however,
only weakly influences recovery except when kMS/ A number of flotation plant have been simulated that
kGS is above a certain threshold value. incorporated a variety of circuit configurations encom-
• The number of cleaning stages required is influenced passing rougher and cleaner scavenging banks, two or
by the ratio IMF/IGF. three stages of cleaning, regrinding of rougher concen-
• Retention time in the roughers and whether the use trate, open or closed circuit cleaning and pulling final
of scavenging banks is beneficial is influenced by concentrate from the first cell(s) of a roughing or first
the ratio kMS/kGS. stage cleaning bank. The plants also represented a range
of capacity from 6000 to 150 000 tons per month, in-
cluding individual cell volumes from 0.6 to 38 m3 . The
5. Simulation of flotation plant cells included columns, agitated square and tank cells, of
the forced air and self-induced type. Cell make included
Table 2 compares actual and predicted flotation Wemco, Outokumpu, Denver, Dorr-Oliver, Svedala and
performance for five plants. All simulations are based on in some cases in-house or ‘‘on-mine’’ hybrids. Some cells
testwork of borecore or mill feed samples; plant feed were high powered, fitted with oversized mechanisms

EURUS
MINERAL
ACTUAL vs SIMULATED PLANT RECOVERY CONSULTANTS
PLATINUM & BASE METAL SULPHIDE ORES
100
LEGEND 10
95 y = 0.9694x + 2.2387 8
1 - Foskorite (SA) 2 9
90 R = 0.9421 7
2 - Canadian Cu ore 6
85 3 - Maranda Cu (SA) 4
80 4 - Palabora Cu ore (SA) 3 5
5 - Palabora Cu ore (SA) 2
75 6 - Gt Dyke Pt ore (Zimbabwe) Cu ore (SA) 1
7 - Northam Merensky ore Pt (SA)
SIMULATED PLANT RECOVERY

70 Gt Dyke Pt Oxidised (Zimbabwe)

8 - Maranda Zn (SA)
65 9 - Black Mountain Pb (SA) O'Okiep Cu Slag (SA)
60 10 - Black Mountain Cu (SA) Palabora Cu
Slag (SA)
55 Pb in Cu circuit (SA)
50
Merensky Pt Oxidised (SA)
45
Zn in Pb circuit (SA)
40
35 Zn in Cu circuit (SA)

30
25
Pb in Zn circuit (SA)
20
Zn in Cu circuit (SA)
15
10
5 Cu in Zn circuit (SA)
0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
ACTUAL PLANT RECOVERY

Fig. 2. Comparison of actual and simulated flotation recovery.

 M.P. Hay, C.M. Rule / Minerals Engineering 16 (2003) 1103–1109 1109

and motors providing inputs in excess of 1 kW/m3 . The time, high-energy input, and cell configuration make
range of air usage in N m3 /m3 cell volume, bubble size, a beneficial difference and where they do not.
superficial gas velocity (SB and JB ) and particle dynamics
were not measured but it is assumed that the variation 7. Conclusions
across these plants was very large.
In all cases just seven, easily measured variables were A kinetics-based methodology and computer pro-
found to be adequate to successfully simulate an oper- gram (SUPASIM) has been developed which predicts
ating flotation plant. The flotation response of an ore, the flotation performance of operating plant with data
captured by its kinetics as measured by a laboratory obtained from batch scale tests. Validation has been
flotation rate test was found to be sufficient to define accomplished by the successful simulation of 15 com-
flotation performance at a laboratory, pilot and plant mercial plants treating a variety of sulphide and non-
scale. The scale-up factors derived from case studies sulphide ores and slags. Actual and predicted plant
were shown to accommodate the many differences be- recovery is described by a linear relationship with a
tween the laboratory and plant flotation cell such as cell correlation coefficient (r2 ) of 0.942.
efficiency, cell geometry and size, power input, bubble, The link between laboratory and plant values for rates
froth and particle dynamics, froth structure, superficial and fractions (i.e. the scale-up factors) was established
gas velocity and others of academic interest. This shows with good accuracy. In all cases and for all ores tested the
that although it is known that these particular variables scale-up factors were found to be similar with the values
and factors affect the process of flotation, their effect on for each factor falling within a narrow range. The scale-
a practical level is small. If this were not the case, up factors were found to be independent of ore type and
SUPASIM would be unable to simulate float plant. are a function of the flotation process (i.e. bubble–
Further, it appears that the power of flotation kinetics particle attachment) and take into consideration the
inherent in the ore tends to overpower the effect of the differences between batch and continuous conditions.
mechanical and hydraulic environment. It is projected Each kinetic variable of rate and fraction was found to
that this is not the case when dealing with difficult to have a physical meaning in the plant and describes
float, low grade values where cell size and power input metallurgical performance and plant design parameters.
would probably make a large difference such as the In all cases just seven, easily measured variables were
flotation of finely milled secondary cleaner tails. found to be adequate to successfully simulate an oper-
ating flotation plant. Case studies showed that the many
differences between the laboratory and plant flotation
6. Future development cell such as cell efficiency, cell geometry and size, power
input, bubble, froth and particle dynamics, froth struc-
SUPASIM has accurately simulated single stage mill- ture, superficial gas velocity and others of academic in-
float (MF1) plants. A series of case studies has already terest were adequately accommodated by these seven
been concluded where MF2 plants (i.e. a coarse grind variables. Further, it appears that, in all but very slow
mill-float followed by a fine grind mill-float) have been floating values, the power of flotation kinetics inherent
successfully simulated from laboratory batch tests. in the ore tends to overpower the effect of the mechan-
Analysis is being concluded and will be reported in a ical and hydraulic environment.
future paper.
Work is currently being conducted to simulate the
flotation of finely milled secondary cleaner tailings (d80 Acknowledgements
of 10 lm) produced within an MF2 circuit.
Future work will, The authors would like to thank the management of
Anglo Platinum Management Services, Northam Plati-
• Determine which ores are conducive to being treated num Ltd, O’Okiep Copper Company (Pty) Ltd, Meto-
by the ‘‘two-concentrate’’ process developed by the rex Ltd and Palabora Mining Company Ltd for their
platinum industry in the 1990s. Platinum UG2 ores permission to publish plant data. Special thanks is
operated by this process have shown a significant in- conveyed to those companies who wished to remain
crease in recovery at original or better concentrate anonymous but allowed plant data to be quoted under a
grade. This process involves dividing and extending general geographic region.
a single bank of final cleaner cells into two banks
each producing a final concentrate. Each bank is
Reference
set-up to maximise the recovery of fast-floating and
slower-floating values respectively. Hay, M.P., Snodgrass, R.A., DuPreez, P.J., 1994. Process development
• Aim at redesigning the current MF2 circuit. This will and design of the Northam Merensky concentrator. In: XVth
involve determining where in the circuit residence CMMI Congress, Johannesburg, SAIMM, vol. 2, pp. 341–357.