Minerals Engineering 66–68 (2014) 2–12

Contents lists available at ScienceDirect

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

Modern practice of laboratory flotation testing for flowsheet

development – A review
N.O. Lotter a,⇑, E. Whiteman a, D.J. Bradshaw b
a
XPS Consulting & Testwork Services, Sudbury, Ontario, Canada
b
SMI, University of Queensland, Brisbane, Australia

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

Article history: The flotation testing of sulphide ores for flowsheet development, or for the improvement of existing flow-
Received 9 January 2014 sheets in operations, has been practiced for a century or so. This practice has evolved at both laboratory
Revised 18 April 2014 and operations scales, as a result of contributions by various workers in this field. In this review, two
Accepted 24 April 2014
major contributions to improved practice are discussed, viz Process Mineralogy and representative sam­
Available online 22 May 2014
pling. A description of modern best practice is proposed, particularly in the context of circuit changes or
reagent selection and the use of mixed collectors. Process Mineralogy has contributed significantly by
Keywords:
way of powerful information that reveals process implications such as those resulting from grinding
Process Mineralogy
Sampling
strategies or flotation selectivity challenges. Only recently has the best practice of sampling been con­
Flowsheet development nected to flotation testing. High Confidence Flotation Testing, which incorporates appropriate sampling
Flotation testing models, was proposed in 1995, and used Gy’s minimum sample mass and Safety Line models. Statistical
Benchmark Surveying, a method for extracting representative suites of survey samples from an operating
plant, was added in 2005. A new addition is the small scale evaluation of floatability using the JKMSI,
which enables the testing of small samples such as of drill core, and is demonstrating good agreement
with operations data. Two generations of improved practice are reviewed. The first is when this practice
was retrofitted to serve existing concentrators that had been conventionally designed, in a reactive
approach. The second is serving new design opportunities before commissioning, where predictive value
is added to the project with a more complete understanding of the process implications drawn from the
sampling and characterisation of drill core. It is shown that when these connections are made and mod­
ern quality controls are applied to the flotation testing, much clearer conclusions are drawn, and tighter
metal balances achieved, with better metallurgical performance. This all results in a lower level of error in
the metallurgical test data, reducing project risk, offering significantly shorter project schedules, and bet­
ter startup performance for the project, and also, as the results are more precise, allowing comparison of
options with smaller recovery and grade gains.
© 2014 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1. The project requirements of laboratory flotation testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2. The milestone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3. 1st Generation: opportunities in conventionally-designed concentrators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2. Modern best practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1. Sampling and subsampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1. Alternative sampling method for drill-core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2. Statistical Benchmark Surveying . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3. Mineralogy and process implications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4. Grinding strategy, grinding media, and size distribution effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

⇑ Corresponding author. Tel.: +1 705 699 3400; fax: +1 705 699 3431.
E-mail address: Norman.Lotter@xps.ca (N.O. Lotter).

http://dx.doi.org/10.1016/j.mineng.2014.04.023
0892-6875/© 2014 Elsevier Ltd. All rights reserved.
 N.O. Lotter et al. / Minerals Engineering 66–68 (2014) 2–12 3

2.5. Flotation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.6. Flotation testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.7. Floatability assessment of small samples (drill core) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.7.1. Test methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3. 2nd Generation: greenfield opportunities and modern best practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.1. Case study 1 – prominent hill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2. Case study 2 – the kamoa project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5. Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1. Introduction
1.1. The project requirements of laboratory flotation testing
This review discusses flotation testing at the 1–2 kg batch scale.
The metallurgical engineer arranging the flotation test programme
typically wants clear, unambiguous results showing ‘‘before’’ and
‘‘after’’ test conditions that are significant, cogent, reproducible,
and have a good probability of successful scale-up. All too often,
however, much attention is paid to the flotation test treatment
details, little is paid to the replication thereof, less is paid to the
metal balances across the flotation tests, and regrettably even less,
to the representativity of the ore sample being tested. Much of this
situation is driven by limited budgets and demanding schedules,
and is exacerbated by a limited understanding of what it takes to
be able to claim that the ore being tested is a true sample – and
not a specimen. Equally, it is a challenge to justify replicate flota­
tion tests when the project manager does not see how the averag­
ing process from the Central Limit Theorem works to his advantage
(Box et al., 1978).
1.2. The milestone
Many regard the work of Henley (1983), as the turning
point when conventional mineral processing and more modern
mineralogy together engaged the overall perspective of geology,
mineralogy, and implications for mineral processing, together with
the ongoing advancement of mineral processing itself. This was
largely enabled by the development of the QEM*SEM (later, the
QEMSCAN) by Grant et al. (1976), and others such as Gottlieb;
and of the MLA by Gu (2003), and others. The quantitative informa­
tion that these instruments provide by way of mineral processing
implications has made a significant difference to the focus of the
mineral processor. However, the reliability and value of these mea­
surements by the QEMSCAN and by the MLA are dependent to a
large extent on the representativeness, or trueness, of the samples
presented for measurement, and on the skill and experience of the
mineralogist. Whilst several such as Gottlieb and Gu worked on
advancing the capabilities of the QEMSCAN and the MLA respec­
tively, others turned to investigating the importance and method­
ology of the sampling and reproducibility for both plant surveys
and flotation testing (Restarick,1976; Hartley et al., 1977; Lotter,
1995a,b; Lotter, 2005). A theory of mixed collectors for synergy
in flotation was consolidated and proposed by Bradshaw (1997)
and Lotter and Bradshaw (2010), and was validated in a plant trial
at the Eland Platinum operations in South Africa (Lotter et al.,
2011). Others such as Wightman and Evans, 2012, developed
new liberation models for grinding strategies. Together with many
other significant contributions to the best practice, when consoli­
dated and used as an integrated toolbox, flowsheet diagnosis and
flotation testing became empowered with more effective capabili­
ties. The overall outcome has been to shift the balance of technical
activities away from reactive and more towards predictive.
1.3. 1st Generation: opportunities in conventionally-designed
concentrators
Conventionally designed concentrators offer the modern Pro­
cess Mineralogist much opportunity in plant performance
improvement. This is because the fuller process implications from
the mineralogy of the orebody, including variability, were not
apparent or available at the time of design. The early generation
of Process Mineralogy thus had to take a reactive approach to these
opportunities so as to gain traction and credibility. The Mount Isa
history is a good example of this.
An excellent example of the value of the practical application of
Process Mineralogy comes from the Mt Isa operations from 1980s
onwards has been reported extensively (Johnson et al 1998; Pease
et al., 2006; Pease, 2010). From 1982, over a period of ten years as
the ores became progressively more finely grained. Fig. 1 shows
the resulting reduced Zn recovery as a result of decreasing sphaler­
ite liberation, and Fig. 1 shows typical composite minerals respon­
sible for the Zn tailings losses.
The insight to the reasons for the losses lead to the development
of the IsaMill and inclusion of fine grinding in the circuit gave a
substantial benefit to Zn recovery, as seen in Fig. 1.
The engagement of the Process Mineralogy toolbox with exist­
ing concentrators requiring performance improvements was
well-demonstrated by Martin et al. (2003), in the case of the Lac
des Iles expansion project in Ontario, Canada. The operation was
expanded from a 2400 tonnes per day (tpd) operation to a much
larger 15,000 tpd business. This required a new concentrator,
which was designed from a prefeasibility study. One major differ­
ence between the two flowsheets was the 80% passing size (d80)
size of the float feed, presumably recognising the need for a finer
grind to liberate the discrete PGM. The change in d80 size was from
150 to 75 lm. Additionally the flotation residence time was
increased from 19 to 55 min. The collector suite used was a mix­
ture of Potassium Amyl Xanthate (PAX) and di-isobutyl dithiophos­
phate. A new heavy (750 g per tonne (g/t) milled) dose of talc
depressant as Carboxy-Methyl Cellulose (CMC) was used in the
rougher float. This was another change in the practice. Methyl Iso­
butyl Carbinol (MIBC) frother completed the reagent suite. Primary
concentrates were reground in vertimills to a d80 size of 20
microns before cleaning in two separate cleaner circuits. Shortly
after commissioning in October 2001, it became apparent that,
whereas the concentrate grade was almost in agreement with
the design value of 170 g/t Pd, the recovery of Pd was short of
design. Actual Pd recoveries amounted to 67.5%, as compared to
the design requirement of 82%.
Several plant surveys ensued, supported by mineralogy as well
as size-by-size paymetal analysis of streams, each delivering clues
to flowsheet improvement. The survey methodology was not
4 N.O. Lotter et al. / Minerals Engineering 66–68 (2014) 2–12

90

85
% Liberated Sphalerite

80 Zn Rec to Zn Con

75
%Liberated Sphalerite
or %Recovery 70

65

60 GF Flowsheet
P5 99/00
55 Mine to Mill
Review
P7 96/97
Pb Regrind
50 Mill P7 94/95

45
Fine Grinding
from P4 92/93
40

3
P8 /92

P5 /96
P6 /82

P3 /85

P2 /86

P5 /95
8

P9 /91

P7 /93

P6 /94

P5 /97

8
P1 /87

2
P5 /83

P4 /84

N 3
-0
P1 /8

P1 8/8

P1 9/9

P5 7/9
P1 7/8

P5 1/0
P5 /9

P5 9/0

P5 0/0

/0
91

92

93

94

95

96
90
81

82

83

84

85

86

87

98

02

ov
9
8

0
3

0
2

1
P7

Period / Year

Fig. 1. Plot of zinc recovery at Mt Isa over time, overlain by a plot of zinc liberation. (from Pease et al., 2006 and Pease, 2010.)

described; however the mineralogy was performed by QEMSCAN. In the case of the Raglan base metal project in Northern Québec,
These were implemented across a schedule and progressively the concentrator was designed from conventional mineral process­
advanced the grade and recovery of the saleable concentrate. The ing and successfully commissioned into its designed performance
heavy talc depressant dose in the rougher float was lightened so in early 1998 (Langlois and Holmes, 1999). The designed nickel
as to allow some talc to float. This stabilised the froth. Additional recovery of 87.0% was well-met by the commissioned recovery of
cleaner capacity was found by recommissioning the cleaner circuit 86.7%, whilst the designed concentrate grade of 16% Ni was exactly
from the older, smaller concentrator. A key discovery was the matched at 16% by the new operations. However, the development
bimodal size distribution of the palladium mineral host grain sizes. of a new Process Mineralogy resource by Falconbridge in their Met­
The two modes of this distribution are approximately 20 and 5 lm. allurgical Technology Group enabled the survey and diagnosis of
It was found that mostly, it was the coarser size distribution of dis­ the newly commissioned plant in mid-1998 using a prototype of
crete PGM, as kotulskite and palladoarsenide, that were being Statistical Benchmark Surveying, subsequent QEMSCAN measure­
recovered. The appropriate regrinding of all primary concentrates ment of sized samples, and supporting laboratory scale flotation
in the vertimills had a major effect on both concentrate grade testwork. This work identified three flowsheet improvement
and recovery. The regrind product size was a d80 of 20 lm. The opportunities (Lotter et al., 2002):
two cleaner circuits were simplified to a single circuit. All of these
changes improved the saleable concentrate to a grade of 240 g/t Pd
at a recovery of 74%.
In a separate case for the INCO Clarabelle mill in Sudbury,
Ontario, the mill was expanded in 1991 from a daily ore treat­
ment rate of 32,000–40,000 short tons per day (stpd) by the
addition of a Semi-Autogenous (SAG) mill to the grinding circuit
and associated additional flotation equipment. A key part of this
circuit relied on successful magnetic separation of the monoclinic
pyrrhotite gangue at the rougher float feed. In 1998, a new ore-
body was brought into production from the North Mine. The pyr­
rhotite occurring in this new orebody did not respond to
magnetic separation, and was believed to be pseudohexagonal
(Becker et al., 2010). This key mineralogical information was dis­
covered after the fact of commissioning this new orebody into
production. The tonnage of these new arisings only amounted
to 5% of total ore milled. However, this change caused a signifi­
cant loss in selectivity in the flotation process, which in turn
caused a reduction of nickel recovery from 78% to 70% (Kerr
et al., 2003). Laboratory and pilot scale testwork at INCO’s
research facility in Sheridan Park, Mississauga, showed that by
reallocating a regrind mill ahead of the circuit, and introducing
TETA and sulphite to depress the non-magnetic pyrrhotite, as
well as adding an extra cleaner flotation step, would restore this
lost performance (Xu et al., 2000). These changes were success­
fully implemented at the operation in 1999. Fig. 2. Flow of activities in high-confidence flotation testing (after Lotter, 1995a,b).
 N.O. Lotter et al. / Minerals Engineering 66–68 (2014) 2–12
1. Rerouting of the Recleaner Column Tailings from the
Scavenger Bank Feed to the Cleaner Column Feed, so as to
allow the predominantly liberated and fine to ultrafine
pentlandite to be recovered to Final Concentrate.
2. Commissioning of a regrind mill to treat the Cleaner
Column Tailing before its presentation to the Scavenger
Flotation Bank. This was to improve the state of base metal
sulphide liberation.
3. Introduction of a mainstream rougher flotation dose of a
CMC depressant to control gangue flotation.
These changes were implemented between September 1998
and September 2000, and yielded recovery gains of 2.1% Ni, 1.5%
Cu, 1.9% Pd, and 4.1% Pt, as well as a concentrate grade gain from
16% to 18% Ni.
In the then Phelps-Dodge Candelaria operation, which was
regarded as the best-performing copper concentrator in that
company with a 95.3% copper recovery to saleable concentrate, a
Process Mineralogy team surveyed the plant and diagnosed the
remaining performance opportunities (Baum et al., 2004; Kendrick
et al., 2003). The thinking was to the effect that, if performance gains
could be found at this operation by using Process Mineralogy, then
larger gains could be expected at other copper and gold concentra­
tors in that company. The project identified the form of tailings
losses and increased recoveries by 0.2% Cu and 10% Au. One of the
changes included reducing the controlled pH of the rougher circuit
from 10.5 to 9.0, as identified by laboratory scale flotation testwork.
This change in pH was key to improving the gold recovery. The min­
eralogy identified a concentration of copper losses in the topsize
fractions of the rougher tailings. The locked particles accounted for
58% of the copper losses to rougher tailings, representing a recovery
target of 2.5%. The ore hardness was a limiting factor in addressing
the locked chalcopyrite opportunity.
2. Modern best practice
2.1. Sampling and subsampling
The difference between a sample and a specimen has a major
impact on the reliability of the project test results. The failure of
mining or metallurgical undertakings can nearly always be traced
back to the confusion between ‘‘specimens’’, on the basis of which
no sane financial decision should ever be made; and ‘‘samples’’,
known to be representative of the object to be valued (orebody,
shipments of concentrates, etc.) within the limits of a certain con­
fidence interval that can be estimated and relied upon (Gy, 1979).
Although his sampling theories and models are commonly
regarded as excessively complex and difficult to work with, Gy’s
sampling models are practical and do work very well. Moreover,
this subject is not commonly taught in the undergraduate schools,
leaving the mineral processor to work away at the sampling refer­
ence book on his own. However, unless some solid, quantifiable
proof can be shown as a chain of evidence from the lot to be sam­
pled to the replicate flotation test lot, even the most careful flota­
tion testwork has no claim of validity. It is thus a firm
recommendation that a sampling equation be obtained for the case
in hand using a version of the fifty-piece experiment, accompanied
by observance of the Safety Line, both models proposed by Gy
(1979). The outline of these two models has been reviewed and
published elsewhere (Lotter and Oliveira, 2013; Lotter et al.,
2013a). The basic understanding of the term ‘‘representative
sample’’ is best expressed as follows.
A representative sample differs from the lot or flowing stream
from which it was taken only in that its mass is smaller than that
of the lot or flowing stream. The sample’s composition is exactly
5
the same as that of the lot or flowing stream. There are several
rules that must be followed to assure the unbiased extraction of
increments to that sample, as well as the satisfaction of the mini­
mum mass rule for the stipulated fundamental variance. If any one
of these rules is not complied with, the sample automatically
becomes disqualified and is called a specimen. These rules are
explained in detail elsewhere (Gy, 1979; Lotter and Oliveira, 2013).
When the drill core sampling is performed on the basis of geo­
met units, an improved visibility of the individual characteristics
and the process implications develops. Geometallurgical units
(Lotter et al., 2003; Fragomeni et al., 2005) can be defined as an
ore type or group of ore types that possess a unique set of textural
and compositional properties from which it can be predicted they
will have similar metallurgical performance. Sampling of an ore-
body based on Geometallurgical units will define metallurgical
variability and allow process engineers to design more robust flow-
sheet options. This variability can be muted when samples from
different Geometallurgical units are blended and tested as one
sample. Composites are created by ensuring grade and grade distri­
butions from a specific area defining the Geometallurgical unit
within a resource are maintained. The method used to divide an
orebody into Geometallurgical units is based on a review of geolog­
ical data including host rock, alteration, grain sizes, texture, struc­
tural geology, grade, sulphide mineralogy and metal ratios with
focus on characteristics which are known to affect metallurgical
performance. The foregoing list is, however, not complete and also
uses hardness testing and the grade/recovery curve as characteriz­
ing parameters (Fragomeni et al., 2005 for example). Statistical
analysis is often used to help define preliminary units. In addition,
it is recommended that a variability program based on smaller
samples from throughout a Geometallurgical unit is completed
prior to finalising the divisions between Geometallurgical units.
This approach will quantify the range in performance that can be
expected from within a unit, and provides a cross check that the
Geometallurgical unit definition is robust.
2.1.1. Alternative sampling method for drill-core
In the case where insufficient drill-core is available to satisfy
Gy’s minimum sample mass models, but representativity is still
desired in the test material, an alternative has been proposed by
Oliveira and first published in Kormos et al. (2013), and then in
Lotter and Oliveira (2013). In this approach, non-random sampling
of the drill-core is performed so as to match the distribution of
paymetal grades and lithological types in the host rock. She and
her co-authors showed that it is relatively easy to deliberately con­
struct a metallurgical sample in a non-random manner that con­
tains a feature of interest in the correct proportion. The preferred
approach is to review geological and lithological data with project
geologists to determine the potential features of interest and their
associated geospatial distribution based upon the existing core log­
ging data base and ore model, then to specifically select core to
represent each potential geological type. Thus, whilst the samples
are non-random, they nonetheless achieve good representativity of
the features of interest in quantities suitable for metallurgical
testing without requiring an extensive supplementary drilling
campaign.
The actual structure of the sampling activities in this approach
depends on the state of development of the drilling programme.
Key metrics such as the proposed cutoff grade determine what
has to be excluded from any sampling. Proposed mining methods
will identify and estimate whatever dilution will be introduced
into the mined material (and the sample material). It is obvious
from these points that better sampling will be possible from a later
stage of drilling and resource modeling than from earlier stages of
the project. Modelling of the geometallurgical units is performed
using basic statistics and histograms that define the mean grade
6 N.O. Lotter et al. / Minerals Engineering 66–68 (2014) 2–12
and distribution of paymetals. The objective is to formulate a
subsample that aligns with a comparable paymetal distribution
to the parent population. The target distribution is typically
subdivided into deciles. The mass required for the metallurgical
testwork is calculated, and then redistributed across these deciles
in agreement with the parent distribution. Additionally, lithogical
data are tested in a similar manner to ensure that the variations
in host rock are representatively included in the subsample. Care
is also taken to ensure that the spatial extraction of the subsample
engages the known range of space for each geometallurgical unit.
Ultimately the subsample distribution parameters and those of
the parent distribution are compared using the v2 distribution to
test for significant agreement.
2.2. Statistical Benchmark Surveying
A structured, statistical method whereby an operating concen­
trator could be surveyed for mass and value balancing, and the
resultant samples presented for subsequent mineralogical analysis,
was developed by Lotter (2005), and Lotter and Laplante (2007a),
and is called Statistical Benchmark Surveying. This approach used
replicate two-hour survey units separated by sufficient time to
break any autocorrelation in feed grades to the flotation plant.
The entire flowsheet was sampled, including circulating loads
and intermediate products such as rougher and scavenger concen­
trates. Outlier rejection was managed with independent reference
distributions to adjudicate feed, concentrate and tailings grades as
being characteristic of the normal operation. Composites are pre­
pared from the accepted two-hour survey units, then sized and
assayed prior to mineralogical measurement by QEMSCAN. This
practice produces a benchmark data set that typify the normal
metallurgical behaviour of the normal mill feed and existing
flowsheet.
A second plant survey model developed by Lotter and Laplante
(2007b), called the Campaign Survey Model, campaigns a specifi­
cally selected ore mixture from known mining areas of interest
in a planned proportion to represent some near future ore
resource. The selected ore is mined and stockpiled ahead of the
concentrator, then milled during a week of campaign operations.
This approach was tested and validated the Raglan nickel opera­
tions, Northern Québec. This model also uses quality controls to
prove that representative samples have been extracted for the
mass and value balance, and for mineralogical study.
Whereas other survey methods have been developed and which
may be regarded as valid, statistical Benchmark Surveying has
quantitative statistical proofs that the sample suite accepted into
the final composites is representative at the 95% confidence level.
2.3. Mineralogy and process implications
Mineral processing plants treat minerals, not assays. Whereas
the assays provide information for a mass and value balance to
report grade and recovery, they cannot be relied on to give any
accurate indication of the mineralogy that is hosting these paymet­
als. The last two decades have seen a great deal of advancement in
the effectiveness and automation of mineral measurement. Plat­
forms such as the QEMSCAN and MLA are good examples. Again
predicated on the condition that representative sample material
is presented to these automated mineralogical instruments, use
of the data produced provides powerful flowsheet implications to
guide the development of the flowsheet (Lotter, 2011).
2.4. Grinding strategy, grinding media, and size distribution effects
The grinding media type and size distribution produced at the
flotation feed stage, set the platform for the rougher flotation pro­
cess. The use of a graded rod charge in the laboratory batch mill to
simulate the size distribution of a production scale ball mill oper­
ating in closed circuit with a hydrocyclone is well-known and prac­
ticed (Wills, 1992). The shape of the size distribution produced by
the grinding process, as well as its cumulative passing size, has a
critical influence on the flotation process (Trahar, 1981; McIvor
and Finch, 1991; Bazin et al, 1994). Adam et al. (1984), report that
in the case of pyrrhotite flotation, depression of pyrrhotite
occurred by the use of mild steel media, however with austenitic
stainless steel, there was less interaction and the pyrrhotite floated
more successfully. The depression mechanism proposed was the
galvanic interaction of the pyrrhotite with active metals resulted
in the formation of hydroxides on the pyrrhotite surface. Peng
and Grano (2010), showed that for chalcopyrite and galena, it
was the -10 lm size fraction that showed the most response to
iron oxidation species emanating from grinding media. With more
inert steels such as chrome-enriched stainless steel, with chrome
ranging from 12% to 20%, it is possible to more closely simulate
the likely operating Eh produced by the high-chrome grinding
media which are favoured by operations because of their lower
wear rate. This was explained by Greet et al. (2000), who proposed
that broadly speaking, sulphide minerals are more noble than
forged steel grinding media. This causes a galvanic couple between
the media and the sulphide minerals, which in turn corrodes the
grinding media. The corrosion products, mostly iron oxides and
hydroxides, precipitate on the sulphide minerals and depress flota­
tion to varying degrees. With the high-chrome grinding media, cor­
rosion is more limited because these media are more inert. The
type of steel selected for use as the grinding media is influential
on the redox potential of the milled ore. Depending on the types
of sulphides present in the ore sample, the redox potential may
either aid or depress sulphide flotation (Iwasaki, 1988).
The choice of primary and secondary grind sizes in projects
studying new orebodies has traditionally been empirical, however
recent work on the MLA and QEMSCAN platforms has introduced a
more fundamental approach because of sulphide grain size distri­
bution data becoming available for modelling in the grinding strat­
egy. Initial application using multiple polished thin sections and
Sichel’s t-estimator to calculate the mean grain size and associated
confidence limits did prove successful (Fragomeni et al., 2005).
More recently, Wightman and Evans (2012) have proposed a liber­
ation model that plots the percent particle volume occurring as the
sulphide of interest versus the particle size for sized classes of
crushed ore. The plot takes the form of a horizontal shelf in the
coarser particle sizes, but changes gradient sharply as the particle
size becomes smaller and approaches the liberation size at L0. It
will be shown later in a case study how this model is used.
With new tools such as this to hand, a far more predictive esti­
mate of the grinding strategy before performing any flotation tests.
2.5. Flotation
The technology of flotation is approximately 100 years old. It is
a complex process dominated by interactions, using pH modifiers,
dispersants, activators, collectors, depressants and frothers, as well
as energy, air and water (Bradshaw, 1997). Engineers are com­
pelled because of this complexity to investigate flotation on an
empirical basis at various scales, be they laboratory investigations,
pilot plants or production operations (Napier-Munn, 1995; Lotter
et al., 2010; Lotter et al., 2011; Bradshaw and Vos, 2013; Lotter
et al., 2013a for example). Important synergies between different
collectors, for example, produce small but economically significant
improvements in concentrate grade and recovery (Lotter and
Bradshaw, 2010), however their proof requires an accurate and
reproducible measurement system, for example in the plant scale
trial at the Eland Platinum operations, South Africa (Lotter et al.,
 N.O. Lotter et al. / Minerals Engineering 66–68 (2014) 2–12 7

Fig. 3. Gy’s Safety Line, showing starting position of 2 kg crushed ore at A (topsize 1.7 mm), milling to flotation size to B (212 lm topsize) and C (150 lm topsize), then
subsampling the size fractions down to 5 g each at D and E.

2011), and at the Raglan nickel operations, Québec (Lotter et al., PGE and PGM-bearing ores. This system was further advanced
2010). A plant scale test design using appropriate statistics to test and validated in a postgraduate study at the University of Cape
for significant differences in recovery and grade was reported by Town (Lotter, 1995a,b). The platform provided a means of produc­
Napier-Munn (1995), Napier-Munn (2012) and with another ing flotation test data sets at the 95% level of confidence. Further
version using reference distributions and the analysis of variance, refinements were made at Falconbridge Ltd. and Xstrata Process
by Lotter et al. (2010). Support in Canada to focus this model on base metal applications
(Lotter et al., 2002; Lotter and Fragomeni, 2010). There are two
parts to the model: the correct sampling and subsampling of the
2.6. Flotation testing
ore to the flotation test charge level; and the replicate flotation test
system using the powerful averaging effects of the Central Limit
Laboratory batch flotation, in which 1–2 kg test charges of ore
Theorem (Box et al., 1978), with an outlier rejection system written
are using a standard set of operating conditions, is the standard
and proposed for small data sets by Grubbs (1969).
technique used to assess ore floatability and characterise the kinet­
Ore samples that are crushed and milled to smaller size distri­
ics in order to work out the plant circuit capacity, for example
butions before assaying present easier platforms for the accurate
Dobby and Savassi (2005). Traditionally these tests have been used
measurement of ore grade (or, as Gy refers to this measurement,
extensively to determine flotation design, optimise grinding strat­
as ‘‘content’’). Provided that the primary sample satisfies the sam­
egies, screen reagent suites, etc. (Runge, 2010). These tests have
pling equation with a known and acceptable fundamental variance,
usually been applied to samples of feed to the concentrator, but
and that the crushing and milling are performed in keeping with
the practice has recently been extended to sampling intermediate
the provisions of the Safety Line, the subsample integrity is main­
operations process streams such as recycling streams, and rougher
tained (Gy, 1979). This high-confidence flotation system recog­
concentrates. Key ore floatability parameters such as flotation rate
nizes and manages this challenge by estimating the correct
constants can be derived from such work, assisting in the ranking
sampling equation for that size distribution and paymetal variance
of candidate test conditions. A detailed description of the equip­
from Gy’s fifty-piece experiment, and by using the Safety Line to
ment and operational parameter controls is provided in Runge’s
appropriately reduce the primary sample after crushing and blend­
publication. The reproducibility of the test results is of paramount
ing to replicate test charges with smaller size distributions and
importance. The author correctly recognises the limitation of this
standard deviations. The ‘‘assay heads’’ are thus obtained from
platform in the very shallow froth depth at which the laboratory
milled test charges at the rougher float feed size distribution
setup operates, as compared to the operational scale flotation cell,
(commonly with a topsize of 106 lm).
which has a much deeper froth.
Thereafter, the system uses sets of replicate flotation tests with
A flotation testing system called FLEET was developed and val­
application of an outlier rejection model for small data sets as pro­
idated with three cases studies being presented by Dobby and
posed by Grubbs (1969). First concentrate replicate mass data
Savassi (2005). This system was designed to predict flotation plant
performance using a standardised platform producing a set of
timed concentrates which are sized into size classes before weigh­
ing and assaying. The system uses three stages:

1. Direct measurement of the kinetics of true flotation in the pulp,

independent of hydraulic entrainment and other froth effects in
the batch cell.
2. Benchmarking of the industrial plant over a broad range of ore
types and operational conditions.
3. Model calibration to describe froth effects in the plant, using a
minimal amount of parameters.

The flotation testing method known as ‘‘High-Confidence Flota­

tion Testing’’ was developed at Rustenburg Platinum Mines Lim­
ited, South Africa by Lotter and Munro (1994), for the specific Fig. 4. Copper recovery by size for the five ore types showing their differing
problem of true sampling and reproducible flotation testing of response (from Bradshaw and Vos, 2013).
8 N.O. Lotter et al. / Minerals Engineering 66–68 (2014) 2–12
Fig. 5. Copper recovery vs. size relationship for the different feed size fractions for
total copper and chalcopyrite (from Bradshaw and Vos, 2013).
must satisfy the criterion of demonstrating a relative standard
deviation of less than 5%. Finally, composite formulation and repli­
cate assaying lead to metal balancing of built-up head grades to the
mean assay head grade to satisfy a criterion of less than ±3.3%
error. This practice of replicate flotation tests being used with
appropriate quality control to identify and reject outliers from
small data sets as a means to composite sample assembly success­
fully engages the powerful averaging effects of the Central Limit
Theorem, leading to reduced residual errors in the composite with
a sample mean of zero, approximating a Normal Distribution. Use
of this approach enables the comparison of flotation test conditions
with smaller differences in grade and recovery, and reducing the
scale-up risk. This system is thus half a sampling and subsampling
system, and half a flotation testing system, as shown in Fig. 2.
2.7. Floatability assessment of small samples (drill core)
A method is being developed to characterise the floatability of
small samples (5–10 g) such as drill core samples using the Julius
Kruttscnitt Mineral Separation Index (JKMSI) (Coleman et al.,
2006; Bradshaw, 2011; Bradshaw and Vos, 2013; Morgan et al,
2012). Its success has been demonstrated in characterising the var­
iable copper recovery obtained for the various size fractions for
selected ore types from copper porphyry ores. It is critical that
the representatively of the samples at this small mass is estab­
lished, and it can be shown that these samples and their subsam­
ples are located to the left of Gy’s Safety Line during their
processing, thus their integrity has not been compromised. The
use of the Safety Line is shown in Fig. 3.
2.7.1. Test methodology
The test methodology followed in this paper is similar to that
previously reported (Bradshaw and Vos, 2013). In these tests three
size fractions, 212–150 lm, 150–75 lm and 75–38 lm were
obtained after wet milling each ore for 5 min in a stainless steel
rod mill. Each JKMSI test was done on a 5 g sample. These samples
were conditioned in an ultrasonic bath before being transferred to
the JKMSI device where reagents as specified by site were added.
After reagent additions the slurry in the test-tube is pneumati­
cally shaken for a set number of cycles, after which it is stopped
and the suspension left to settle. The tails (settled solids) and con­
centrate (particles attached to froth layer) samples are collected,
filtered and dried in preparation for assays. Multiple tests are done
for a particular set of conditions and the variation in the percent of
dry masses recovered to the concentrate is assessed according to
Chauvenet’s criterion to identify outliers resulting from experi­
mental error (Chauvenet, 1863). If a particular test is identified
as an outlier that sample will not be used to create the composite
from a particular set of conditions but discarded. Additional tests
were performed for some ores to obtain the required mass for ele­
mental and mineralogical analysis. The repeat samples were com­
bined, then split into representative sub-samples and sent for
chemical analysis by ICP–MS as well as mineralogical analysis of
the concentrates and tails.
Five different ores from Laguna Seca, Escondida Minera Limit­
ada, five ores showed the greater variation in the coarse fraction
which was attributed to liberation limitations resulting from
smaller grain sizes. Fig. 4 shows the effect of ore variability, and
Fig. 5, the grind-recovery relationship, on copper recovery for
particularly the 212–150 lm fraction.
3. 2nd Generation: greenfield opportunities and modern best
practice
The modern practice of Process Mineralogy prefers to predict
probable process implications ahead of design, rather than react
to process problems after commissioning. This approach makes a
significant difference to the success of the startup, and contributes
towards the probability of a Type 1 startup in which the designed
rates and quality of production are achieved in the first year
post-commissioning (McNulty, 1998).
3.1. Case study 1 – prominent hill
In the case of the Prominent Hill project and subsequent
commissioning of operations in 2009, it was shown how the fuller
process implications may be drawn from the drill core sampling
and associated testwork in the design work (Barns et al., 2009).
The ore mined contains copper-gold breccia zones with mineralisa­
tion groupings of chalcopyrite–bornite, chalcocite–bornite, chalco­
pyrite-pyrite, and a separate gold-enriched unit called EGO. The
chalcocite–bornite grouping dominates the first five years of min­
ing and processing. Haematite breccia represents approximately
55% of the whole rock assemblage, followed by andesite and steely
haematite each at 10%.
Chalcocite is the coarsest of the copper sulphides averaging
40 lm in grain size. Bornite is finer at 24 lm. Chalcopyrite is inter­
mediate at 50% less than 33 lm, and occurs as a texture with born­
ite. Approximately 70% of the copper sulphides are associated with
haematitic gangue. Another 15% is associated with pyrite.
Notable features of the host rock mineralogy included alteration
in the haematite breccia, with associated fluorite (CaF2). The clasts/
relict clasts of the host rock appeared to be highly altered, being
variably rich in sericite, quartz, haematite, siderite, chlorite and
minor biotite, titanomagnetite, rutile, anorthite and sulfides. The
fluorite is a major process implication, because fluorine is a penalty
element in saleable copper concentrate. Grades of F in saleable
concentrate exceeding 300 ppm attract penalties at the point of
sale, for example at the OK Tedi operation, where the fluorine
grades range from 400 to 1500 ppm (Lauder et al., 2003).
With this collection of powerful mineralogical and geological
data, the laboratory testwork programme was well-informed and
focussed to the flowsheet needs for this project. A key feature
was the principle that the plant design had to be simple and easy
to operate and maintain (Colbert et al., 2009).
A primary grind of 106 lm d80 size was finally selected after a
range of finer and coarser grinds had been selected. A simple
xanthate-based collector regime was chosen, with occasional pH
modification by lime, presumably to control pyrite flotation.
Rougher concentrate was reground to a size of 25 lm d80. Three-
stage cleaner flotation released grades of 25% Cu and 5–10 g/t Au
at recoveries of 87% and 70% respectively for the chalcopyrite­
pyrite ore. The chalcocite ore produced higher copper grades in
the range of 60% Cu with 5–10 g/t Au, with similar recoveries to
those of the chalcopyrite–pyrite ore.
 N.O. Lotter et al. / Minerals Engineering 66–68 (2014) 2–12 9

The saleable concentrates produced at laboratory scale

assayed high grades of fluorine. The chalcopyrite-pyrite ore type
produced concentrate at 5700 ppm. The chalcocite ore type pro­
duced a lower fluorine grade at 1000 ppm. Both of these values
exceed the penalty limit of 300 ppm. Mineralogical study of the
concentrates found that most of the fluorine was associated with
the non-sulphide gangue, thus reduction of fluorine in saleable
concentrate amounted to non-sulphide gangue rejection (Barns
et al., 2009).
A benchmark of equivalent fluorine rejection was published by
workers in Brazil from their studies on the equivalent problem at
their operations in the Carajas region of the Para state (Andrade
et al., 1999). Their solution was regrinding to 70% mass passing
16 lm at cleaner feed, plus dosage of sodium hexametaphosphate
depressant in the cleaners. This reduced the fluorine level from
1000 to 200–300 ppm in saleable concentrate. Subsequent labora­
tory work for the Prominent Hill project, using this information,
showed a promising reduction in fluorine grades by regrinding to
20 lm and adding froth washing to the cleaner operation, e.g. with
a Jameson cell in the plant design. A saleable concentrate showing
a grade of 800 ppm fluorine (with penalties) was produced by this
test work.
Since the plant has been commissioned, the saleable concen­ Fig. 6. Distribution of Cu metal balance errors, total 34 observations, mean 0.00%.
trates graded approximately 700 ppm F, in good agreement with
the laboratory testwork. The copper data showed a recovery of
85% for a concentrate grade of 55% for the chalcocite–bornite ore. (2) show the sampling models for hypogene and supergene drill
core respectively:
Hypogene:
3.2. Case study 2 – the kamoa project
424:95
Ms ¼ ð1Þ
In the case of the Ivanhoe Kamoa project, located in the Demo­ fv
cratic Republic of the Congo, the modern toolbox has been fully
Supergene:
applied. The concentrator is still at its design stages in a pre­
feasibility study at the time of writing. The case history to date
218:10
of this project follows as a summary. Ms ¼ ð2Þ
fv
To introduce this case study, an explanation of the term ‘‘geo­
met unit’’ is necessary. In Eqs. (1) and (2), ‘‘fv’’ represents the quantitative parameter
The Kamoa copper resource is a high-grade, world-class discov­ ‘‘fundamental variance’’, and Ms represents the minimum sample
ery, located some 40 km west of Kolwezi, in the Katanga Province mass equivalent to that fundamental variance. Best practice of
of the Democratic Republic of the Congo. Drilling at Kamoa in Gy’s sampling models recommends that practitioners use a prede­
2008–2009 had discovered laterally continuous, sediment-hosted, termined value of 5–8% for fv. The higher coefficient in Eq. (1) for
high-grade stratiform copper mineralisation in a newly discovered the hypogene reflects the higher variance of copper grades in the
copper district that forms a previously unrecognised western drill-core data. For a fundamental variance of 8%, the minimum
extension of the famous Central African Copperbelt – which hosts sample mass required is thus 66.40 kg for hypogene and 34.08 kg
such world-class deposits as Kolwezi, Tenke-Fungurume, Konkola for supergene. Due to the large amount of testwork identified for
and Nchanga. This discovery showed that the western end of the flowsheet development of hypogene ore, a total of 265.62 kg of
Copperbelt in the Congo does not terminate at Kolwezi quarter HQ hypogene core was configured. This mass exceeds the
(Friedland and Broughton, 2009). The resource contains hypogene minimum sample mass, and is therefore representative at the sta­
and supergene geomet units with a transitional zone located in ted level of fundamental variance. In the case of supergene ore, a
between the (upper) supergene and the (lower) hypogene. In the primary sample mass of 193.21 kg HQ quarter core was configured.
hypogene unit, the major copper sulphides are bornite, chalcopy­ This figure also exceeds the minimum sample mass of 34.08 kg,
rite and chalcocite, whilst in the supergene unit; the sequence is and is therefore representative. The sample material presented to
led by chalcocite, then bornite, and finally chalcopyrite. Covellite the laboratory scale flotation testwork is thus representative in
is present in both units as a minor copper sulphide. Unlike more terms of the definitions of minimum sample mass according to
typical copper orebodies, the Kamoa ore carries little in the way Gy (1979).
of clay minerals. To date the only clay mineral identified and quan­ In both the hypogene and supergene geomet units, use was
tified is kaolinite, at an amount of 1-3% of the whole rock assem­ made of the Safety Line to release true replicate 2-kg flotation test
blage. Both geomet units will be mined and treated in one charges as well as the assay head samples. The assay head subsam­
concentrator, either as blends or as campaigns. This implied that ples and mineralogical subsamples were extracted at rougher float
a single flowsheet successfully treating both geomet units was a feed topsize (106 lm), culminating in 10 determinations of copper
requirement for the design. grade by sodium peroxide fusion, acid dissolve and ICP–OES finish.
The sampling of drill-core for the metallurgical testwork to This arrangement follows an equivalent strategy to that described
develop the flowsheet for the concentrator followed the guidelines in Fig. 3, and demonstrates together with use of the minimum
discussed in this paper, with minimum sample mass models sample mass model that the sample material presented for
derived from the drill-core data for hypogene and supergene ore testwork was representative. The supergene sample underwent
types respectively (Lotter, 2010; Lotter et al., 2013). Eqs. (1) and equivalent treatment.
10 N.O. Lotter et al. / Minerals Engineering 66–68 (2014) 2–12

The mineralogy and flotation testwork programmes for the Table 2

Kamoa project were performed by XPS Consulting & Testwork Ser­ Size-by-size copper data: scavenger tailings ex hypogene ore.

vices (Lotter et al., 2013b). The replicate 2 kg ore test charges pro­ Size class Mean class size lm Mass % Grade % Cu Dist’n % Cu
duced at A in Fig. 3 were used for the flotation tests. Flotation tests -75 + 53 lm 64 16.89 1.04 32.19
were performed in triplicate. For the development of the Rougher CS1-2 46 7.82 0.96 13.76
and Scavenger float, First Rougher Concentrate replicate masses CS3 28 10.70 0.64 12.55
had to pass a test of displaying less than 5% relative standard devi­ CS4-5 16 14.93 0.45 12.30
CS6 5 15.47 0.30 8.50
ation (RSD) before being taken into the composite sample series for CS7 1.5 34.20 0.33 20.69
mass and value balance. Otherwise, outlier detection and deletion Total 100.00 0.55 100.00
using the method of Grubbs (1969), was used. In the case of the full
flowsheet format including the cleaner circuit, this replicate mass
test was applied to the cumulative mass of the four Rougher Rec- tion concerning copper mineral type and abundance, with
leaner Concentrates. The point of this quality control measure is associated grain size distributions, led the flowsheet development.
to ensure that an influential point on the grade-recovery curve The key features, or process implications, of this copper
has a tight confidence limit. resource were found to be:
The built-up head grades were pooled and analysed together
with the multiple assay head grades. The two data sets averaged 1. The copper sulphide mineral hosting between hypogene
3.30% Cu for assay heads and 3.29% Cu for the built-up heads, with and supergene ranged between chalcopyrite, bornite,
relative standard deviations of 2.58% and 1.74% respectively. On the chalcocite and covellite, averaging a mineral grain size
average, this amounts to a nearly perfect metal balance for copper, of liberated sulphides: 27 lm, middling particles:
with a call factor of 99.7%. The distribution of differences between 15 lm; and locked particles: 7 lm, in a rougher flotation
the robust built-up head mean grade and the reported test condi­ feed of p80 = 75 lm. The liberation of total copper sulp­
tions forming this distribution were then calculated. This resulted hides was only 62.5%. This implied that:
in a mean of 0.00% error with a standard deviation of 1.74%, and a 1.1. The collector system would have to be a mixed collec­
distribution of metal balance errors as shown in Fig. 6. In this distri­ tor suite to span the wide range of flotation require­
bution, 80.0% of all errors fall between -2 and +2%. ments of these copper sulphides and their incomplete
The mean value of 0.00% error is no surprise. It is a tenet of the liberation (Lotter and Bradshaw, 2010).
Central Limit Theorem that, when averages of observations are 1.2. A staged grinding and flotation system for the rougher
used instead of individual values, regardless of the shape of the and scavenger float (otherwise called an MF2 platform,
parent distribution, the errors in estimating the mean tend to an short for ‘‘mill-float-mill-float’’) would be preferable to
average value of zero with a reduced total error that approximates handle the issue of size distribution of copper
a Normal Distribution. This is why High Confidence Flotation Test­ sulphides.
ing uses averages of triplicate composited flotation tests. In the 1.3. Separate cleaner circuits to treat the rougher and scav­
built-up head data set, the standard deviation is 0.057% Cu. enger concentrates would be necessary, in order to han­
The progress made with these flotation tests was greatly dle the different flotation speeds of the various copper
enhanced by the detailed quantitative mineralogy data provided sulphides present. These cleaner circuits would require
by QEMSCAN measurements, supported by microprobe measure­ regrinding of their respective cleaner feeds to improve
ment to obtain the exact composition of the key minerals. Informa- liberation.

The MF2 platform was easily configured. It demonstrated suit­

Fig. 7. Showing Liberation Topsize for Regrinding Kamoa Scavenger Tailings (After
Wightman and Evans, 2012). Test Data Courtesy of Ivanplats Ltd.
Table 1
Saleable concentrates produced by the milestone flowsheet.
able grades and recoveries to rougher and scavenger concentrates
with the MF1 grind set at a p80 of 75 lm, and the MF2 grind at a
p80 of 38 lm, using a 36%:64% w/w/ 3477 (di-isobutyl dithiophos­
phate):SIBX (sodium isobutyl xanthate) mixed collector system at
a total dosage of 125 g/t milled. The cleaner circuit testwork set out
to develop an open-circuit format with discard cleaner tailings
streams, rather than to recirculate these to the circuit. This work
soon found that the rougher cleaner tailing contained large quan­
tities of slow-floating copper sulphides, in the order of 19% of mill
head copper recovery, which needed retreatment in the scavenger
cleaner circuit. Subsequently a successful flowsheet delivering
saleable concentrate at an acceptable recovery was developed,
and was called the ‘‘Milestone Flowsheet’’, which especially incor­
porated two cleaner sections. These have been reported elsewhere
(Lotter et al., 2013b).
This circuit delivered satisfactory grades and recoveries for both
hypogene and supergene geomet units. The saleable concentrates
produced are summarised in Table 1.
The project listed two follow-up items for further improvement
of the flowsheet performance:

Geomet unit Mass pull % COPPER 1. Reduction of the silica grade in the saleable concentrate to
avoid the need for lime flux addition in the smelter feed.
Grade % Cu Recovery % Cu
2. Estimation of a different grind size at the secondary mill
Hypogene 8.6 32.8 85.3
to lower the tailings losses attributable to locked copper
Supergene 6.9 45.1 83.2
sulphide in the coarser size fraction.
 N.O. Lotter et al. / Minerals Engineering 66–68 (2014) 2–12
For the estimation of an alternative grind size at the secondary
mill, use was made of the Evans and Wightman liberation model
on QEMSCAN data from size-by-size measurement of the Hypo-
gene Scavenger Tailings. The size-by-size copper assays and
distribution are shown in Table 2. The various copper sulphides
were pooled as ‘‘Total Copper Sulphides’’ and the liberation graph
prepared. This is shown in Fig. 7.
The enrichment of copper grades and distribution in the two
topsizes is obvious. The liberation diagram for the Hypogene Scav­
enger Tailings (Fig. 7) drawn after the method of Wightman and
Evans (2012), indicates a new grind setpoint of 18–20 lm.
The residual copper sulphides in the Scavenger Tailings
would start to liberate at a grind size of approximately
18–20 lm. The project is thus in a position to consider further
improvements to the flowsheet’s grinding strategy from a predic­
tive viewpoint.
4. Conclusions
From this review, the following conclusions may be drawn:
1. The benefits to laboratory flotation testwork from improved
Process Mineralogy underpinned by effective representative
sampling have been clearly demonstrated.
2. The history of conventionally designed concentrators shows
that these provide a suitable platform from which to intro­
duce better practice of circuit diagnosis. This is done
through representative sampling, High-Confidence Flotation
Testing of the mill feed, detailed, quantitative mineralogy to
develop the process implications, and implementation of
practical plans to transfer this information to well-focussed
circuit changes.
3. The recent history of modern Process Mineralogy in cases
with new mines and metallurgical processes has so far
shown that the new concentrator commissions with better
performance than would have been the case with the con­
ventional design approach.
4. The High Confidence Flotation Testing system, using the
best practice of Gy for minimum sample mass and sample
processing with the Safety Line, followed by replicate flota­
tion testing and outlier management, has been proven to
reliably scale up to production operations. This result has
in large measure been achieved through improvement of
the representativity of the sample material tested, and by
reduction of random, unnecessary errors in the metal
balances.
5. The potential value of the smaller-scale JKSMI flotation test­
ing system has been shown by the characterisation of ore
variability as well as the relationship to production opera­
tions. The earlier concerns about representativity of the
small 5 g sample have been resolved by use of the Safety
Line.
6. It is clear that improved ore body knowledge and variability
testing at several levels, especially including mineralogy, is
key to understanding a new orebody for either a greenfield
or brownfield project.
7. These enhancements to laboratory flotation testing have
resulted in lower error in the metallurgical test data, reduc­
ing project risk, offering vastly reduced time frames and
better startup performance for the project, and also, as the
results are more precise, allowing comparison of options
with smaller recovery and grade gains.
11
5. Recommendations
This review has summarised two generations of the develop­
ment of a modern, effective toolbox that, if used correctly, leads
to better flowsheet development for a project. To summarise the
basic steps that a practitioner should follow, the following recom­
mendations are made as a check-list for the reader:
1. The prospective ore resource being investigated should be
well-understood in terms of geology and geomet units.
2. Representative sampling of each geomet unit, using the
models here described, is essential in order to ensure that
whatever sample material is used in the laboratory for test-
work and characterisation is properly representative of the
ore resource. In this regard, two methods have been
described in the paper. These are the modified 50-piece
experiment by Lotter, after the original work by Gy; and
the distribution method by Oliveira.
3. Use of Gy’s Safety Line in transforming the primary ore sam­
ples to workable batches is critical to the maintenance of
sample representativity.
4. Use of High-Confidence Flotation Testing is preferred over
conventional singleton flotation tests.
5. Quantitative mineralogy, supported by compositional infor­
mation by the microprobe, provides very useful information
leading to process implications that may be used in the
flowsheet development.
Acknowledgements
The authors gratefully acknowledge the many helpful discus­
sions over the years with operational and research staff, as well
as permission by the management of XPS Consulting and Testwork
Services to publish this paper. Ivanhoe management very kindly
granted permission for the use of the case study on Kamoa.
References
Adam, K., Natarajan, K.A., Iwasaki, I., 1984. Grinding media wear and its effect on
the flotation of sulphide minerals. Int. J. Min. Proc. 12 (1–3), 39–54.
Andrade, V.L., Santos, N.A., Nardi, R.P., 1999. Technological development for
Igarape-Bahia copper–gold project. In: Hancock, B.A., Pon, M.R.L. (Eds.), Proc.
Copper 99. Mineral Processing, Environment, Health and Safety, vol. 2, pp. 113–
127.
Barns, K.E., Colbert, P.J., Munro, P.D., 2009. Designing the optimal flotation circuit –
the Prominent Hill case. In: Proc. Tenth Mill Operators Conference, Adelaide,
October 2009, pp. 173–182.
Baum, W., Lotter, N.O., Whittaker, P.J., 2004. Process Mineralogy – a New Generation
for Ore Characterisation and Plant Optimisation. SME Preprint No. 04–12,
Denver, Colorado, February 2004.
Bazin, C., Grant, R., Cooper, M., Tessier, R., 1994. A method to predict metallurgical
performances as a function of fineness of grind. Miner. Eng. 7 (10), 1243–1251.
Becker, M., Villiers, J., Bradshaw, D.J., 2010. The flotation of magnetic and non­
magnetic pyrrhotite from selected mineral deposits. Miner. Eng. 23 (2010),
1045–1052.
Box, G.E.P., Hunter, W.G., Hunter, J.S., 1978. Statistics for Experimenters. Wiley.
Bradshaw, D.J., 1997. Synergistic Effects between Thiol Collectors used in the
Flotation of Pyrite, Ph.D. (Chem. Eng.) Thesis, University of Cape Town, South
Africa, pp. 30–49.
Bradshaw, D.J., 2011. A Small Scale Test for Rapid Characterisation of Flotation
Response. SME Denver, 1–3 March, 2011.
Bradshaw, D.J., Vos, F., 2013. The development of a small scale test for rapid
characterisation of flotation response (JKMSI). In: Proc. Canadian Mineral
Processors, Ottawa, January 2013, Paper No. 5, pp. 43–57.
Chauvenet, W., 1863. A Manual of Spherical and Practical Astronomy V. II (Reprint
of 1891, fifth ed. Dover, N.Y., 1960. pp. 474–56).
Colbert, P.J., Munro, P.D., Yeowart, P.G., 2009. Prominent Hill – designed for
operators and maintainers. In: Proc. Tenth Mill Operators Conference, Adelaide,
pp. 23–31.
12 N.O. Lotter et al. / Minerals Engineering 66–68 (2014) 2–12
Coleman, R., Alexander, D., Urtubia, H., 2006. A comparison of BHP-Billiton’s Minera
Escondida flotation concentrators. In: Proceedings Canadian Mineral Processors
Conference 17–19 January 2006, Ottawa, Canada.
Dobby, G.S., Savassi, O.N., 2005. Advanced modelling technique for scale-up of batch
flotation test results to plant metallurgical performance. In: Proc. Centenary of
Flotation, Brisbane, Brisbane, June 2005, pp. 99–103.
Fragomeni, D., Boyd, L., Charland, A., Kormos, L.J., Lotter, N.O., Potts, G., 2005. The
use of end-members for grind-recovery modelling, tonnage prediction and
flowsheet development at Raglan. In: Proc. Canadian Mineral Processors,
Ottawa, January 2005, Paper No. 6, pp. 77–98.
Friedland, R., Broughton, D.W., 2009. Keynote Address, 8th World Copper
Conference, Santiago, Chile, April 2009.
Grant, G., Hall, J.S., Reid, A.F., Zuiderwyk, M., 1976. Multi-compositional particle
characterisation using the SEM-microprobe. In: Proc. Scanning Electron
Microscopy 1976. Part III, Workshop for Particulate Matter Studies in SEM,
ITT Research Institute, pp. 401–408.
Greet, C.J., Kinal, J., Steinier, P., 2000. Grinding media – its effect on pulp chemistry
and flotation behaviour – fact or fiction? In: Proc. Centenary of Flotation,
Brisbane, pp. 967–972.
Grubbs, F.E., 1969. Procedures for detecting outlying observations in samples.
Techonometrics 11 (1), 1–21.
Gu, Y., 2003. Automated scanning electron microscope based mineral liberation
analysis. JOM 2 (1), 33–41.
Gy, P.M., 1979. Sampling of Particulate Materials, Theory and Practice. Publ.
Elsevier, Amsterdam, p. 1.
Hartley, D.G., Trahar, W.J., Weller, K.R., 1977. Analysis of Concentrator Operations at
Mount Lyell, Aus. IMM., Conference Series, 6, pp. 155–169.
Henley, K.J., 1983. Ore-dressing mineralogy – a review of techniques, applications
and recent developments. Spec. Publ. Geol. Soc. S. Afr. 7, 175–200.
Iwasaki, I., 1988. Flotation behaviour of pyrrhotite in the processing of nickel–
copper ores. In: Tyroler, G.P., Landolt, C.A. (Eds.), Extractive Metallurgy of
Copper–Nickel Ores, pp. 271–291.
Johnson, N.W., Gao, M., Young, M.F., Cronin, B., 1998, Application of the ISAMILL (a
horizontal stirred mill) to the lead/zinc Concentrator (Mount Isa Mines Ltd.) and
the mining cycle. AusIMM Annual Conference, Mount Isa, pp. 291–297.
Kendrick, M., Baum, W., Thompson, P., Wilkie, G.J., Gottlieb, P., 2003. The use of the
QEMSCAN and automated mineral analyser at the Candelaria Concentrator. In:
Proc. Copper, vol. III, pp. 416–430.
Kerr, A., Barrett, J., Bouchard, A., Labonté, G., Truskoski, J., 2003. The ‘‘Mill Redesign
Project ‘‘ at INCO’s Clarabelle Mill. In: Proc. Canadian Mineral Processors,
Ottawa, Paper No. 3, pp. 29–49.
Kormos, L.J., Sliwinski, J., Oliveira, J.F., Hill, G.S., 2013. Geometallurgical
characterisation and representative metallurgical sampling at Xstrata Process
Support. In: Proc. Canadian Mineral Processors, Ottawa, pp. 3–14.
Langlois, P.A., Holmes, J.A., 1999. Process development at the Raglan Concentrator.
In: Proc. Canadian Mineral Processors, Ottawa, pp. 427–451.
Lauder, D.W., Mavotoi, M., Gladhaar, J.W., 2003. Fluorine removal in OK Tedi
copper/gold Concentrates. In: Proc. 8th Mill Operators Conference, Aus. IMM.,
Melbourne, pp. 203–209.
Lotter, N.O., 1995a. A Quality Control Model for the Development of High-
confidence Flotation Test Data, (Unpublished master’s thesis). University of
Cape Town, Cape Town, South Africa.
Lotter, N.O., 1995b. A Quality Control Model for the Development of High-
confidence Flotation Test Data, SME Annual Meeting and Exhibit, Denver,
Colorado, March 1995, Preprint 95-40.
Lotter, N.O., 2005. Statistical Benchmark Surveying of Production Concentrators,
Ph.D. Thesis, McGill University, Montréal, August 2005.
Lotter, N.O., 2010, Stratified sampling of drill-core. In: Proc. Canadian Mineral
Processors, Ottawa, January 2010, Paper No. 11, pp. 163–179.
Lotter, N.O., 2011. Modern process mineralogy: an integrated multi-disciplined
approach to flowsheeting. Miner. Eng. 24 (2011), 1229–1237.
Lotter, N.O., Bradshaw, D.J., 2010. The formulation and use of mixed collectors in
sulphide flotation. Miner. Eng. 23 (2010), 945–951.
Lotter, N.O., Fragomeni, D., 2010. High-confidence flotation testing at Xstrata Process
Support. Trans. Miner. Metallurgical Process., SME, Denver 27 (1), 47–54.
Lotter, N.O., Kowal, D.L., Tuzun, M.A., Whittaker, P.J., Kormos, L.J., 2003. Sampling
and flotation testing of Sudbury basin drill core for process mineralogy
modelling. Miner. Eng. 16, 857–864.
Lotter, N.O., Laplante, A.R., 2007a. Statistical benchmark surveying of production
concentrators. Miner. Eng. 20 (2007), 793–801.
Lotter, N.O., Laplante, A.R., 2007b. The campaign survey model – a case study at the
Raglan mine Québec. Miner. Eng. 20 (2007), 480–486.
Lotter, N.O., Munro, H.C., 1994. The development of high-confidence flotation
testing at Rustenburg Platinum Mines Ltd., Min. Met. Managers Association,
Circular 1/94, pp. 29–50.
Lotter, N.O., Oliveira, J.F., 2013. Sampling – A Practical Primer, Short Course,
University of Utah, March 2013.
Lotter, N.O., Whittaker, P.J., Kormos, L.J., Stickling, J.S., Wilkie, G.J., 2002. The
Development of Process Mineralogy at Falconbridge Ltd., and Application to the
Raglan Mill, CIM Bulletin, 95, No. 1066, pp. 85–92.
Lotter, N.O., Di Feo, A., Kormos, L.J., Fragomeni, D., Comeau, G., 2010. Design and
measurement of small recovery gains: a case study at Raglan concentrator.
Miner. Eng. 23 (2010), 567–577.
Lotter, N.O., Monnapula, Oliveira, J., Fragomeni, D., Bradshaw, D.J., 2011.
Formulation and plant trial of a mixed collector suite for Eland platinum. In:
Proc. Canadian Mineral Processors, Ottawa, Paper No. 10, pp. 161–183.23, 945–
951 (2011).
Lotter, N.O., Oliveira, J.F., Hannaford, A.L., Amos, S.R., Broughton, D.W., 2013,
Flowsheet development of the hypogene geomet unit in the Ivanplats Kamoa
copper project. In: Proc. Canadian Mineral Processors, Ottawa, January 2013,
Paper No. 7, pp. 71–85.
Lotter, N.O., Oliveira, J.F., Hannaford, A.L., Amos, S.R., 2013b. Flowsheet development
for the Kamoa project – a case study. Miner. Eng. 52 (2013), 8–20.
Martin, C.J., Ohrling, T., Olsen, T., Taggart, P., 2003. From tribulation to triumph –
flotation improvements at the new Lac des Iles Concentrator. In: Proc. Canadian
Mineral Processors, Ottawa, January 2003, Paper No. 2, pp. 13–27.
McIvor, R.E., Finch, J.A., 1991. A guide to interfacing of plant grinding and flotation
operations. Miner. Eng. 4 (1), 9–23.
McNulty, T., 1998. Developing innovative technology. Miner. Eng. 50 (10), 50–55.
Morgan, S., Bradshaw, D.J., Schwarz, S., 2012. JKMSI – small scale test to characterise
the flotation response of an ore type. In: 11th Mill Operators Conference 2012.
Hobart 29–31 October 2012.
Napier-Munn, T.J., 1995. Detecting performance improvements in trials with time-
varying mineral processes – three case studies. Miner. Eng. 8 (8), 843–858.
Napier-Munn, T.J., 2012. Statistical methods to compare batch flotation grade-
recovery curves and rate constants. Miner. Eng. 34 (2012), 70–77.
Pease, J D., 2010. Economics and communication, 2010. In: Greet, C.J. (Ed.), Flotation
Plant Optimisation A Metallurgical Guide to Identifying and Solving Problems in
Flotation Plants. Spectrum Series No. 16. The Australasian Institute of Mining
and Metallurgy, pp. 191–200 (Chapter 10).
Pease, J.D., Curry, D.C., Barns, K.E., Young, M.F., Rule, C., 2006. Transforming
flowsheet design with inert grinding media – the IsaMill. In: Proceedings 38th
Annual Meeting of the Canadian Mineral Processors, CMP, Ottawa, pp. 231–249.
Peng, Y., Grano, S.R., 2010. Effect of iron contamination from grinding media on the
flotation of sulphide minerals of different particle size. Int. J. Min. Proc. 97 (1–4),
1–6.
Restarick, C.J., 1976. Pulp sampling techniques for steady state assessment of
mineral concentrators. In: Proc. Sampling Symposium Aus. IMM. pp. 161–168.
Runge, K., 2010. Laboratory flotation testing – an essential tool for ore
characterisation. In: Greet, C.J. (Ed.), Flotation Plant Optimisation A
Metallurgical Guide to Identifying and Solving Problems in Flotation Plants.
Spectrum Series No. 16. The Australasian Institute of Mining and Metallurgy, pp.
155–173 (Chapter 9).
Trahar, W.J., 1981. A rational interpretation of the role of size in flotation. Int. J. Min.
Proc. 8 (1991), 289–327.
Wightman, E., Evans, C., 2012. Representing and interpreting the liberation
spectrum in a processing context. In: Proc. Process Mineralogy ’12. MEI
Conference, Cape Town, South, Africa, November 2012. (Chapter 12).
Wills, B.A., 1992. Mineral Processing Technology, fifth ed. Pergamon, Oxford, p. 528
(Chapter 12).