Flotation Plant Optimisation

A Metallurgical Guide to Identifying and

Solving Problems in Flotation Plants
Edited by C J Greet

Number 16
 Flotation Plant
Optimisation
A Metallurgical Guide to Identifying and
Solving Problems in Flotation Plants
Edited by Christopher J Greet

The Australasian Institute of Mining and Metallurgy

Spectrum Series 16

ISBN 978 1 921522 21 5

COPYRIGHT DISCLAIMER
© The Australasian Institute of Mining and Metallurgy 2010
No part of this publication may be may be reproduced, stored in a retrieval system or
transmitted in any form by any means without the written consent of the publisher.

The AusIMM is not responsible as a body for the facts

and opinions advanced in any of its publications.

NUMBER 16

Published by:
THE AUSTRALASIAN INSTITUTE OF MINING AND METALLURGY
Ground Floor, 204 Lygon Street, Carlton Victoria 3053 Australia
 HOW TO USE THIS BOOK
It is intended that Chapter 1 (The Eureka Mine – An Example of How to Identify and Solve Problems in a
Flotation Plant, by Christopher Greet) be used in parallel with subsequent chapters to gain a greater
understanding of what is involved as one goes through the process of optimising the plant.
The basic road map to follow on your quest to improve the metallurgical performance of your plant is given in
the figure below. The first step in this process is the acquisition of quality data from the plant (or mine, in the
case of a geometallurgical study (Chapter 12: Operational Geometallurgy by Dean David)). Chapter 2 (Existing
Methods for Process Analysis by Bill Johnson), provides details of how to collect, mass balance, size and
interpret metallurgical data collected from the plant. Mass Balancing Flotation Data by Rob Morrison
(Chapter 3) describes in detail the considerations that must be taken into account when mass balancing survey
data collected from a flotation plant. Having defined the recovery-by-size characteristics of the concentrator, the
next step is to establish their mineralogical character; that is, their locking or liberation characteristics. Alan
Butcher, in Chapter 4 (A Practical Guide to Some Aspects of Mineralogy that Affect Flotation), provides a
summary of the mineralogical techniques available, their application and suitability to various situations. The
combination of recovery-by-size and liberation-by-size and mineral class will define whether the losses
occurring in the tailing stream are liberated or composite for each size fraction. The same can be determined for
gangue minerals reporting to the concentrate.

Geometallurgy
(Dean David)
Data acquisition
(Bill Johnson) Mass balancing
(Rob Morrison)

Questions? Mineralogy
(What and where?) (Alan Butcher)

Cell characterisation
(Greg Harbort
Problem definition Sarah Schwarz) Electrochemistry
Machine (Ron Woods)
Liberation Pulp chemistry
Chemistry
(Stephen Grano )
Surface chemistry
Laboratory testing (Alan Buckley)
(Kym Runge)

Trials and stats

(Tim Napier -Munn)

Communication
(Joe Pease)

Schematic showing the broad road map of how to identify and quantify problems within base metal sulfide flotation plants.
Once the weaknesses within the concentrator have been identified it is possible to start looking for the root
cause. For example, are the inefficiencies of the process related the equipment being used (Chapter 5:
Characterisation Measurements in Industrial Flotation Cells by Greg Harbort and Sarah Schwarz)? If liberation
is an issue then it may be a case of adding additional size reduction capacity at an appropriate position within
the circuit. However, if fine liberated values are leaving the circuit via the tailing there may be a problem with
the chemistry. Stephen Grano’s chapter (Chapter 6: Chemical Measurements During Plant Surveys and Their
Interpretation) provides an explanation of the typical pulp chemical measurements that can be made in the plant
to identify where the chemistry of the system is not optimal for separation. Ron Woods (Chapter 7:
Electrochemical Aspects of Sulfide Mineral Flotation) discusses the significance of electrochemistry of the
system and its impact on flotation. In some instances it is necessary to use more sophisticated surface analysis
techniques to identify the species on the surfaces of minerals particles. Chapter 8: Surface Chemical
Characterisation for Identifying and Solving Problems within Base Metal Sulfide Flotation Plants, by Alan
Buckley presents the various techniques available (ie XPS, ToF-SIMS), discusses their merits and describes
how the data can be interpreted.
The above describes the steps, with increasing degrees of sophistication, one may take to identify the where,
what and how in the focusing questions. The tremendous strength of these principles lies in the fact that they
can be applied to almost any processing operation, not only the flotation of base metal sulfides, and at any scale
(ie laboratory, pilot plant and industrial scale).
Having identified where the weaknesses in your operation are, it is possible to devise laboratory experiments to
test potential solutions. These programs may range from diagnostic test conducted on plant pulps to determine
if changes made in the plant have resulted in a positive shift in metallurgy to simple laboratory flotation tests to
screen reagents. The development of these ideas is discussed by Kym Runge in Chapter 9: Laboratory Flotation
Testing – An Essential Tool for Ore Characterisation.
Invariably encouraging laboratory solutions are applied to the operating plant with varying degrees of success.
In Chapter 10 (Designing and Analysing Plant Trials), Tim Napier-Munn will discuss the steps required to
conduct a successful plant trial and how the data generated may be analysed to give a statistically meaningful
outcome.
Finally, Joe Pease (Chapter 11: Economics and Communication) will supply a philosophical note on how best
to communicate solutions developed by the technocrat to both operations personnel and management teams
such that the economic benefits can be realised.
 CONTENTS
Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
The Eureka Mine – An Example of How to Identify and Solve Problems in a Flotation Plant
C J Greet

Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Existing Methods for Process Analysis
N W Johnson

Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Mass Balancing Flotation Data
R Morrison

Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
A Practical Guide to Some Aspects of Mineralogy that Affect Flotation
A R Butcher

Chapter 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Characterisation Measurements in Industrial Flotation Cells
G J Harbort and S Schwarz

Chapter 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Chemical Measurements During Plant Surveys and Their Interpretation
S R Grano

Chapter 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Electrochemical Aspects of Sulfide Mineral Flotation
R Woods

Chapter 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Surface Chemical Characterisation for Identifying and Solving Problems Within Base Metal Sulfide Flotation
Plants
A N Buckley

Chapter 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Laboratory Flotation Testing – An Essential Tool for Ore Characterisation
K C Runge

Chapter 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
Designing and Analysing Plant Trials
T J Napier-Munn

Chapter 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Economics and Communication
J D Pease

Chapter 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Operational Geometallurgy
D David
 HOME

CHAPTER 1

The Eureka Mine – An Example of How to Identify and

Solve Problems in a Flotation Plant
Christopher Greet

MAusIMM, Manager Metallurgy – Minerals Processing Research, Magotteaux Australia Pty Ltd, 31
Cormack Road, Wingfield SA 5013. Email: christopher.greet@magotteaux.com

Chris commenced his working life as a Trainee Metallurgist at Bradken’s Adelaide steel foundry in 1978. He
subsequently worked in a number of foundries, before becoming a shift foreman at Seltrust’s Teutonic Bore
Mine in 1982. In 1985 he decided to formalise his knowledge by studying for a Bachelor of Engineering in
Metallurgical Engineering at the South Australian Institute of Technology. Upon graduating, he worked as a
Plant Metallurgist at Ok Tedi Mining Limited and Bradken Adelaide, before undertaking a PhD at the Ian
Wark Research Institute in 1992. Since leaving the Wark, Chris has held applied research positions at
Mount Isa Mines Limited, Pasminco, AMDEL, and now with Magotteaux Australia, where he leads the
technical group identifying the impact of grinding chemistry on downstream processing.

Abstract
Introduction
The Eureka Mine
Data Acquisition
Problem Definition
Solution Development and Testing
The Cycle Begins Again
Communication
Conclusions
References
Appendix 1 – The Down-the-Bank Survey
Appendix 2 – Estimated Mineral Assays from Elemental Data

ABSTRACT
As the title implies, this chapter will provide, by way of example, a
methodology for identifying and solving problems in a flotation plant. To
do this a ‘mythical’ concentrator (The Eureka Mine) will be described
and used to demonstrate to the reader how to go about the process of
identifying where the losses of valuable mineral occur, and what gangue
species are diluting the concentrate. It is intended that this chapter be
used in parallel with subsequent chapters to guide the reader through the
steps involved in the process of optimising the plant.
INTRODUCTION
Many of us have, at one time or another, flicked through the
newspaper or searched online for a job, and stumbled upon an ad
not unlike the one that appears below. Whether we are jaded with
our current role, looking for a step up to the next level, or wanting
a new challenge, we prepare our resume and send it off in the
vain hope that we may be the successful candidate.
With the interview process out of the way, the waiting and self
doubt start. And, after what seems an eternity you receive a phone
call or letter telling you that you have got the job.
Congratulations! Now what? You’re new, you have ambition,
you have drive, and you want to make your mark! But, there’s a
right way and a wrong way to do this. The first thing to remember
is that this place has a history, and the people you are going to
work with have been here much longer than you have. So,
communication and respect are keys to your success. You need to
discover the history of the concentrator, and discuss its operation
with other members of staff (operators, metallurgical technicians,
shift foremen, plant metallurgists, the chemist, mechanical and
electrical maintenance, the mine (ie geologists and mining
engineers) and supply). They will all give you their perspective,

Flotation Plant Optimisation Spectrum Series 16 1

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT
SENIOR PROJECT METALLURGIST
The key objectives of this position are to:
• identify opportunities for improvement within the
process,
• develop, test, evaluate and implement process
improvements,
• identify and evaluate new technologies that will improve
our process, and
• be actively involved in our future ores testing program.
Operating in a climate of continuous improvement, you will
be required to actively participate in the promotion of safe
work practices, have good interpersonal skills and a
professional work ethic.
Skills and experience
The successful applicant candidate will have the following
attributes:
• Bachelor of Engineering – Metallurgy or equivalent,
• five years or more process experience,
• a working knowledge of grinding and flotation
processes,
• good communication skills,
• proven problem solving skills,
• good time management skills, and
• have sound leadership qualities and management skills.
Employment conditions
Eureka Mining Limited is a wholly owned subsidiary of
Stockade Resources Limited, an Australian based mining
and exploration company with interests in Australia, New
Guinea, Zambia and Peru.
Reporting to the Metallurgy Manager, the successful
candidate will work closely with production to maintain and
improve plant performance. This is a residential position,
with an attractive renumeration package commensurate with
your qualifications and experience.
To apply
Please submit your application including a cover letter and
current copy of your resume via email to ...
and you need to respect their point of view. You will also work
out who among these people hold the knowledge, the history; the
real story of your concentrator.
While you are establishing relationships you need to determine
what (if any) data exists that will help you to develop a technical
perspective of how your concentrator performs.
The focusing questions in any process improvement strategy
are:
• Where and how do the losses of valuable mineral occur?
• What gangue minerals are diluting the concentrate and how
did they get there?
The intention of this book is to provide you with a sequence of
logical steps to follow so that you can collect the necessary data
to be able to define the problem(s) within your operation. Once
the problems are defined, you can then prioritise them and
develop experimental strategies that may lead to solutions that
can be implemented in the plant.
2 Spectrum Series 16
To assist you in this journey the performance of the Eureka
Mine will be scrutinised and used as an example. The Eureka
Mine treats a complex polymetallic sulfide ore supplied from an
underground mine, producing three saleable concentrates. This
chapter provides a description of the Eureka Concentrator, and its
metallurgical performance since commissioning. Then provides,
by example, the methodology used to collect plant data, analyse
and interpret it to determine where the metallurgical problems lie.
THE EUREKA MINE
Location and history
Extensive geological surveying of the region north of Laylor
River by Stockade Resources Limited resulted in the discovery of
the Eureka deposit in 1990. Throughout 1991 and 1992, drilling
continued to delineate the deposit. Preproduction geological
studies indicated probable reserves of approximately 25 million
tonnes of greater than 15 per cent zinc plus lead, with
economically significant copper, silver and gold grades.
A one million tonnes per annum processing plant was
commissioned by Stockade in 1995 to treat ore with an average
head grade of 0.4 per cent copper; three per cent lead; 12 per cent
zinc; 130 g/t silver; and 2.0 g/t gold.
Geology
The Eureka volcanogenic massive sulfide deposit occurs within
the Laylor-Eureka Volcanic sequence of the Mount Rush
Volcanics. The deposit was formed when hot mineralised
solutions were spewed out on to the ocean floor and were rapidly
quenched by the surrounding seawater. Hence, the sulfide
minerals that were precipitated from solution formed very fine
crystals and intricate mineral textures. Subsequent geological
changes to the orebody were few; therefore many of the original
fine grain textures remained intact.
Deposit mineralogy
To fully appreciate the complexity of flotation at Eureka it is
necessary to have a rudimentary understanding of the mineralogy
of the orebody. Eureka is unusually sulfide rich, and contains a
relatively simple mineral suite: 58 per cent pyrite, 20 per cent
sphalerite, four per cent galena, two per cent arsenopyrite, one per
cent chalcopyrite, with minor amounts of tetrahedrite. The
remaining 15 per cent of the ore consists of: quartz, barite,
calcite, chlorite, sericite and siderite.
Macroscopically the mineral textures are diverse, however, the
orebody can be divided into two distinct metal zones. The
demarcation between the two zones is set, arbitrarily, at 100 g/t of
silver, and represents a continuous horizon across the deposit.
Above this level is the hanging wall enrichment zone characterised
by higher lead, zinc, silver, gold, and arsenic grades.
Macroscopically the sulfides within the enrichment zone tend to be
banded and very fine grained. The footwall-depleted zone occurs
below the 100 g/t silver horizon. Pyrite and chalcopyrite are the
dominant minerals within this part of the orebody, with reduced
lead, zinc, silver, gold, and arsenic grades. The footwall-depleted
zone is highly recrystallised, therefore the grain structure is
comparatively coarse when compared with those observed in the
hanging wall enrichment zone.
It is important to note that the Eureka orebody is relatively free
of non-sulfide gangue mineralisation. So, pyrite is the dominant
gangue mineral, and is associated with all other minerals within
the deposit. Therefore, the properties of pyrite will influence
greatly the behaviour of all other minerals during processing.
Flotation Plant Optimisation
 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

The pyrite textures vary markedly across the deposit from coarse
recrystallised grains in the footwall-depleted zone to compact
microcrystalline masses, spongy and colliform clots, such as
melnokovite (an amorphous pyrite of colloidal origin). Ultra-fine
intergrowths of pyrite with other sulfides are common, particularly
with galena and arsenopyrite (Figure 1). The association of
auriferous arsenopyrite with pyrite is also of significance.
Process description
Laboratory testing of the Eureka ore clearly demonstrated that it
was possible to produce saleable copper, lead and zinc
concentrates. The flow sheet developed in the laboratory was
tested at pilot scale to prove that the process route selected was
robust, and to produce sufficient quantities of concentrate for
smelter testing. The final Eureka process flow sheet is presented
in Figure 2.
As the ore is mined the mine geologists classify it into three
broad ore types based on texture (ie enrichment zone ore
(banded), or footwall depleted zone ore (coarse grained)), and
estimated pyrite content. Each ore type is crushed in batches to
nominally 100 per cent passing 100 mm, in the underground jaw
crusher before being trucked to the surface in 50 tonne dump
trucks. Upon delivery to the run-of-mine (ROM) pad each ore
type is stockpiled separately. The ore is fed onto a conveyor belt
that leads to an open stockpile in specific ratios of each ore type.
Apron feeders, underneath the open stockpile, feed the blended
ore onto the primary mill feed conveyor at nominally 120 t/h.
The primary mill is a low aspect ratio semi-autogenous
grinding (SAG) mill in open circuit. The SAG mill product
discharges into a common sump shared with the ball mill. The
pulp is pumped from the mill discharge sump to cyclones in
closed circuit with a secondary ball mill. The cyclone underflow
feeds the secondary ball mill, and the cyclone overflow reports to
flotation feed. The cyclone configuration is designed to produce a
P80 of 75 microns.
The secondary cyclone overflow feeds a sequential copper/
lead/zinc flotation circuit. Each flotation section consists of a
rougher/scavenger, with the rougher concentrate reporting to the
cleaner circuit. The copper cleaning is achieved without regrinding,
and with only one stage of cleaning. The lead rougher concentrate
feeds the first of three stages of cleaning. The lead scavenger
concentrate and lead first cleaner tailing are reground and recycled
back to the head of the lead rougher. The lead scavenger tailing
reports to the zinc circuit feed. The zinc rougher concentrate
reports to two stages of cleaning. The zinc scavenger concentrate
and the zinc first cleaner tailing are reground and recycle back to
the zinc rougher feed. The concentrates produced from the copper,
lead and zinc flotation circuits are pumped to thickeners. The
thickened concentrate is filtered. The filter cake is stockpiled
before loading into rail cars for shipment to the smelter.
The flotation tailing is dewatered, and used as paste backfill in
the underground workings.
Metallurgical performance
The Eureka concentrator was commissioned in December 1995,
reaching name plate throughput by July 1996. A further 18
months were required to achieve the design concentrate grades
and recoveries. Typical metallurgical performance since 1998 is
summarised in Table 1.

FIG 1 - Photomicrographs of various galena ore textures: (A) galena replacement in pyrite matrix, like melnokovite (magnification × 10);
(B) galena blebs in pyrite matrix (magnification × 20); (C) galena in crystal voids around pyrite (magnification × 5); and (D) galena
replacement in melnokovite colloform (magnification × 40). (Note: The blue/grey areas are galena and the golden areas are pyrite.)

Flotation Plant Optimisation Spectrum Series 16 3

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

Copper Circuit Lead Circuit Zinc Circuit

Cu Ro Pb Ro Pb Scav Zn Ro Zn Scav
Feed
Zn Scavenger Tailing

Grinding
st
Zn 1 Cl
P80 = 75 microns

Copper Cleaner Concentrate

Cu Cl
Zn 2nd Cl
Pb 1st Cl

Zinc 2nd Cleaner Concentrate

Pb 2nd Cl Zinc Regrinding

P80 = 38 microns
Lead Regrinding
P80 = 38 microns

Pb 3rd Cl

Lead 3rd Cleaner Concentrate

FIG 2 - The Eureka Concentrator flow sheet.

TABLE 1
Typical metallurgical performance of the Eureka Concentrator since 1998.

Stream Wt % Grade (%) Recovery (%)

Ag (ppm) Cu Pb Zn Ag Cu Pb Zn
Flotation feed 100.0 130 0.4 3.1 12.4 100.0 100.0 100.0 100.0
Cu concentrate 0.9 3190 25.4 5.2 6.2 21.9 53.2 1.5 0.4
Pb concentrate 3.6 1356 1.2 61.5 10.8 37.6 10.1 70.4 3.1
Zn concentrate 20.0 102 0.3 2.0 53.4 15.9 16.4 13.0 86.9
Final tailing 75.3 42 0.1 0.6 1.6 24.6 20.3 15.1 9.6

An examination of these data suggest that with the exception of

DATA ACQUISITION
The first step towards a better understanding of how your plant is
performing is being able to measure the plant’s performance. In
today’s modern concentrator the process can be monitored using
a multitude of sensors, however the data collected from inventory
samples on a shift, daily, weekly and monthly basis, coupled with
well executed metallurgical surveys can be invaluable in defining
where valuable mineral losses occur and what gangue minerals
are diluting the concentrate.
In the first instance, as the new metallurgist you should
acquaint yourself with the existing plant data. That is, review the
shift mass balance data, interrogate the monthly composite data,
and examine any plant surveys that have been conducted in the
past. This analysis should provide some indication of where the
metallurgical weaknesses lie in the concentrator. However, do not
be surprised if the only data that is up-to-date and readily
available are the shift mass balances. Therefore, it should be
considered good practice to organise for a plant survey to be
completed reasonably early in the piece so that you can quantify
the metallurgical performance of each section of the plant.
The shift data for the copper, lead and zinc circuits of the
Eureka Concentrator appear in Figure 3 for the first half of 1998.
the lead and zinc concentrate grades the plant performance is
somewhat unstable. The copper concentrate grade and recovery
tend to fluctuate wildly, lead recoveries are highly varied, with
greater than ten per cent zinc grade in the lead concentrate, and
zinc recoveries are more often than not in the low to middle 80s.
Your gut feeling when you look more closely at the copper,
lead and zinc recoveries should tell you that they appear lower
than you would expect, therefore it should be possible to improve
the plant performance. But you don’t know what the limiting
factors are so completing a comprehensive plant survey is in
order.
The next item to decide is what level of detail does the survey
have to go? This will obviously depend on what work has been
completed previously. In this example it will be assumed that the
available data is scattered and incomplete. Therefore, the
objective of the survey will be to collect as much data as possible
to provide you with sufficient information to describe the pulp
chemistry and metallurgical performance of the plant. Ideally
both sets of data can be collected in tandem, and complement
each other. To add further value to your survey the collection of
gas hold-up, superficial gas velocity and bubble size data will
provide information about the hydrodynamics of the flotation
cells.

4 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

(a) 90.0 30.0 product should be conducted. Ideally, a more detailed

down-the-bank survey of the rougher/scavenger sections would
75.0 27.0
also be included. An example of a down-the-bank survey is
Cu recovery, %

provided in Appendix 1. Further, to gain greater appreciation of

Cu grade, %
60.0 24.0
how the circuit operates, more than one plant survey should be
45.0 21.0
completed over a number of days. In fact, a good strategy to
30.0 18.0 follow would be one detailed survey and several less detailed
block surveys be completed to give a more balanced view of the
15.0 15.0
metallurgical performance of the plant.
0.0 12.0 Organisation and communication are the key to successfully
completing surveys in a concentrator. In the first instance, decide
8

23 8

21 8
12 8

26 8
98

98

98

98
/98
29 8

12 8

26 8
/9

/9

/9
9

9
1/

1/

2/

3/

3/

4/

5/

6/
2/
01

04

05

06
/0

/0

/0

/0

/0
/0

/0

/0

/0
on the sampling points and the type of sample to be taken. A list
1/

9/

7/

4/
15

18
Time, days of samples for the Eureka concentrator is given in Table 2. It is
Cu recovery Cu grade
important to note that all tailing samples are taken in duplicate to
(b) 90.0 90.0 ensure sampling consistency, and are used as internal checks
(note: the recovery calculation is dependent on the tailing assay).

Pb and Zn grade, %
75.0 75.0
Pb recovery, %

60.0 60.0

45.0 45.0 TABLE 2

The sampling points and sample type for the metallurgical survey
30.0 30.0 of the Eureka flotation circuit.
15.0 15.0
Sample Process stream Sample type
0.0 0.0 number
1 Flotation feed (ball mill cyclone Half moon cutter
98

98

98
8

8
98

98

98

98

98

98
/9

/9

/9

/9
1/

2/

3/

4/

5/

6/
1/

2/

3/
01

04

05

06
/0

/0

/0

/0

/0

/0

/0

/0

/0

overflow)
1/

9/

7/

4/
12
15

29

12

26

26

23

21

18

Time, days 2 Copper rougher concentrate Lip sample

Pb recovery Pb grade Zn grade in Pb con 3 Copper rougher tailing A Dip sample
(c) 100.0 60.0
4 Copper rougher tailing B Dip sample
90.0 50.0
Zn and Fe grade, %

5 Copper cleaner concentrate Lip sample or OSA

Zn recovery, %

80.0 40.0
6 Copper cleaner tailing A Dip sample
70.0 30.0 7 Copper cleaner tailing B Dip sample
60.0 20.0 8 Lead rougher concentrate Lip sample
9 Lead rougher tailing A Dip sample
50.0 10.0
10 Lead rougher tailing B Dip sample
40.0 0.0
11 Lead scavenger concentrate Lip sample
98

98

98
8

8
98

98

98

98

98

98
/9

/9

/9

/9
1/

2/

3/

4/

5/

6/
1/

2/

3/
01

04

05

06

12 Lead scavenger tailing A Dip sample

/0

/0

/0

/0

/0

/0

/0

/0

/0
1/

9/

7/

4/
12
15

29

12

26

26

23

21

18

Time, days 13 Lead scavenger tailing B Dip sample

Zn recovery Zn grade Fe grade in Zn con 14 Lead first cleaner concentrate Lip sample
15 Lead first cleaner tailing A Dip sample
FIG 3 - Time series data for: (A) copper; (B) lead; and (C) zinc
16 Lead first cleaner tailing B Dip sample
circuits.
17 Lead second cleaner concentrate Lip sample
18 Lead second cleaner tailing A Dip sample
The metallurgical survey (Chapter 2) 19 Lead second cleaner tailing B Dip sample
20 Lead third cleaner concentrate Lip sample or OSA
Why don’t metallurgists do surveys? 21 Lead third cleaner tailing A Dip sample
The author has visited many plants around the world, and it is 22 Lead third cleaner tailing B Dip sample
apparent that there is a wide spectrum of knowledge and 23 Zinc rougher concentrate Lip sample
experience. However, plants that conduct frequent, well focused 24 Zinc rougher tailing A Dip sample
surveys and have a clear understanding of their metallurgical
25 Zinc rougher tailing B Dip sample
performance, its strengths and weaknesses, are few and far
between. The reasons for this are many and varied, but generally 26 Zinc scavenger concentrate Lip sample
boil down to not knowing how to conduct a survey, and the fear 27 Zinc scavenger tailing A Dip sample or OSA
of mass balancing! 28 Zinc scavenger tailing B Dip sample or OSA
29 Zinc first cleaner concentrate Lip sample
What do I have to do? 30 Zinc first cleaner tailing A Dip sample
The objectives of the survey(s) are to provide information about 31 Zinc first cleaner tailing B Dip sample
rougher/scavenger flotation performance in each of the flotation 32 Zinc second cleaner concentrate Lip sample or OSA
circuits, and to examine how the concentrates up grade during
cleaning. In the first instance block surveys of each of the 33 Zinc second cleaner tailing A Dip sample
flotation sections (roughers, scavengers and cleaners) for each 34 Zinc second cleaner tailing B Dip sample

Flotation Plant Optimisation Spectrum Series 16 5

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT
There are other process streams that you may consider collecting.
For example, the copper rougher feed, lead rougher feed, and zinc
rougher feed, just to name a few, that are redundant samples in
the mass balance, but do provide valuable data for checking
assays, and ensuring that the mass balanced data makes sense.
With the sample list decided upon it is necessary to brief the
team conducting the survey, prepare the equipment, and ensure
that the plant is running in a fashion that will allow it to be
surveyed. During this briefing, delegate tasks to each person
taking part in the survey so that:
• sufficient sample buckets with lids are cleaned, weighed and
labelled;
• the sampling equipment is checked, clean and ready for use;
• each person taking part in the survey knows what is expected
of them; and
• set the ground rules for the survey (ie number of rounds over
what time interval).
Further, it is wise at this point to inform operations of your
intentions to conduct a plant survey(s). Give then the reasons for
conducting the surveys and check that the plant will be available
at the time you anticipate the survey(s) will be completed. You
also need to inform the chemical laboratory that you intend to
submit a large number of samples from a plant survey, and you
would like to have them assayed for a range of elements.
The day before conducting the survey it is wise to visit the plant
and inspect the flotation circuit to make sure that all those
participating in the survey know where the sampling points are that
they are responsible for. It is also a time when flotation cell lips can
be checked for build-up and cleaned in readiness for the survey. Of
equal importance during the plant inspection is to ensure that the
work area is free of hazards, and is safe to work in.
On the day of the survey, attend the morning production
meeting to establish how the plant has been operating overnight
and inform operations of your intentions to survey the plant. Once
you have the all clear commence setting up for the survey, set out
the buckets in the correct positions, station the sampling
equipment accordingly, check the flotation cell lips and clean
them again. Check again that your team is familiar with the task
ahead of them.
Now that you are ready, go to the control room and check on
the status of the concentrator. Ensure that the feed tonnage is
steady, the feed grades are steady, and the circuit has not
experienced a major disturbance in the last few hours. Once you
are satisfied that the plant is running smoothly you can start the
survey.
It is suggested that a survey of this type be conducted over a
number of hours, whereby multiple rounds of samples are collected
to form a composite of the sampling period. In this particular case,
the survey was conducted over a three hour period, during which
time four ‘cuts’ from each sampling point were collected randomly.
Once the metallurgical survey is completed, the samples are
gathered together and taken to the laboratory.
What data do I need?
In the laboratory the samples are weighed (to determine the wet
weight), filtered, dried, weighed, prepped and submitted for
assay. A summary of this data is given in Table 3. The wet and
dry weights are used to determine the per cent solids of each
sample, and calculate a water balance.
In terms of assays, apart from those pertaining to the valuable
minerals you are separating from the gangue (in Eureka’s case –
copper, lead, zinc, silver and gold), it is also wise to assay for
other elements. For example, iron and sulfur, to aid in performing
mineral conversions so that information about the sulfide gangue
can be extracted from the elemental assays, and deleterious trace
6 Spectrum Series 16
elements like arsenic, antimony, bismuth, mercury and/or
cadmium. Knowing how these penalty elements deport to the
concentrate can lead to a way of minimising their recovery, which
translates to a potential decrease in smelter penalties.
Other data that can be useful when analysing the plant survey
mass balance are:
• the throughput at the time of the survey,
• reagent additions and other plant operating parameters (ie air
flow rates and pulp levels),
• OSA readings, and
• information about the ore being treated.
In the context of a one off survey some of these pieces of
information may not be of great value. However, when the
analysis is extended to include other surveys on other ore blends,
circuit configurations, reagent suites, these data provide a vital
link in the comparison.
How do I analyse the data collected (Chapter 3)?
Once the assays have returned from the chemical laboratory the
fun starts! You will be confronted with a list of numbers not
unlike that presented in Table 4. Initially, this may be a little
daunting, but once you have organised the data into a logical
format, and brought order to the chaos you will be in a position to
mass balance this survey.
At the beginning of the mass balancing process it is necessary to
make sure that you have received all the assays and that they are in
good order. That is, do the tailing assays match? Do the
concentrate assays follow a logical trend (for example, the rougher
concentrate grade is higher than the scavenger concentrate, the
cleaner concentrate grades increase as you pass from the first to the
third cleaner)? Once you have satisfied yourself that the assays are
good you can start the mass balancing process.
Remember, if a water balance of the flotation circuit is required
this can be achieved using the per cent solid values for each
process stream. That is, you treat the per cent solids as an assay,
and include it in your mass balance calculations.
There are numerous mass balancing packages available, but the
basic steps involved in completing the mass balance are the same
for all. Unfortunately, it is not a case of plugging the numbers in
and pressing ‘GO’, as invariably this leads to process streams
having zero flow!
The basic steps to following when mass balancing a survey are
given below. A more detailed explanation is given in Chapter 3:
Mass Balancing Flotation Data by Rob Morrison.
Step 1
Determine what the plant throughput was when the survey was
completed. If the feed tonnage is not available, assign the fresh
flotation feed a value of 100 per cent. It is further assumed that
you have a high level of confidence in this value.
Step 2
Balance the ‘outer’ circuit. That is, complete a mass balance of
the feed, final concentrate and tailing for the copper, lead and
zinc circuits (Figure 4). This balance will provide estimates of the
tonnes of copper, lead and zinc concentrates produced. These
values can be used in subsequent balances to estimate tonnages of
the internal process streams within rougher, scavenger and
cleaner circuits.
Assign each process stream a number, and identify each of the
nodes. So, the process streams are:
1. flotation feed,
Flotation Plant Optimisation
 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

TABLE 3
Raw data from the metallurgical survey of the Eureka Concentrator.

Sample Process stream Sample weight (g) % Solids

number Gross wet Bucket Cutter Net wet Dry
1 Flotation feed 2083.5 202.6 1880.9 909.8 48.4
2 Copper rougher concentrate 1961.2 229.2 347.8 1384.2 708.2 51.2
3 Copper rougher tailing A 2509.5 466.3 2043.2 949.7 46.5
4 Copper rougher tailing B 2480.9 466.0 2014.9 936.1 46.5
5 Copper cleaner concentrate 1386.7 225.0 324.8 836.9 536.3 64.1
6 Copper cleaner tailing A 1622.0 202.6 1419.4 144.8 10.2
7 Copper cleaner tailing B 1668.4 202.6 1465.8 141.8 9.7
8 Lead rougher concentrate 4842.2 225.3 341.0 4275.9 2695.1 63.0
9 Lead rougher tailing A 2395.6 464.3 1931.3 812.3 42.1
10 Lead rougher tailing B 2374.1 477.7 1896.4 772.0 40.7
11 Lead scavenger concentrate 1317.1 206.9 340.9 769.3 482.4 62.7
12 Lead scavenger tailing A 2416.1 465.0 1951.1 823.7 42.2
13 Lead scavenger tailing B 2357.1 446.3 1910.8 782.0 40.9
14 Lead first cleaner concentrate 10638.1 190.4 341.1 10106.6 6660.8 65.9
15 Lead first cleaner tailing A 2646.2 234.5 2411.7 1347.1 55.9
16 Lead first cleaner tailing B 2640.5 224.8 2415.7 1343.7 55.6
17 Lead second cleaner concentrate 5832.8 214.4 325.3 5293.1 3547.9 67.0
18 Lead second cleaner tailing A 2814.2 233.6 2580.6 1608.0 62.3
19 Lead second cleaner tailing B 2828.9 228.2 2600.7 1424.7 54.8
20 Lead third cleaner concentrate 2721.8 200.6 341.0 2180.2 1395.7 64.0
21 Lead third cleaner tailing A 2905.2 238.1 2667.1 1635.1 61.3
22 Lead third cleaner tailing B 2881.6 225.8 2655.8 1650.7 62.2
23 Zinc rougher concentrate 2779.3 226.4 341.2 2211.7 1287.8 58.2
24 Zinc rougher tailing A 2090.7 224.9 1865.8 681.5 36.5
25 Zinc rougher tailing B 1992.0 234.7 1757.3 638.8 36.4
26 Zinc scavenger concentrate 1565.6 200.4 324.9 1040.3 500.5 48.1
27 Zinc scavenger tailing A 1983.0 234.9 1748.1 629.1 36.0
28 Zinc scavenger tailing B 1753.0 225.5 1527.5 536.5 35.1
29 Zinc first cleaner concentrate 3945.3 202.4 341.2 3401.7 1947.7 57.3
30 Zinc first cleaner tailing A 2316.8 225.6 2091.2 1004.9 48.1
31 Zinc first cleaner tailing B 2321.4 225.4 2096.0 1005.5 48.0
32 Zinc second cleaner concentrate 4209.1 211.9 341.2 3656.0 2020.2 55.3
33 Zinc second cleaner tailing A 2466.7 225.5 2241.2 1250.5 55.8
34 Zinc second cleaner tailing B 2474.3 234.0 2240.3 1248.2 55.7

3. Cu scav tailing 5. Pb scav tailing

1. Flotation feed Cu Circuit Pb Circuit Zn Circuit 7. Zn Scan Tail

(1) (2) (3) (Final Tail)

2. Cu Clnr Con 4. Pb 3rd Clnr Con 6. Zn 2nd Clnr Con

(Final Cu Con) (Final Lead Con) (Final Zn Con)

FIG 4 - The ‘outer’ circuit.

2. copper cleaner concentrate, 5. lead scavenger tailing,

3. copper rougher tailing, 6. zinc second cleaner concentrate, and
4. lead third cleaner concentrate, 7. zinc scavenger tailing.

Flotation Plant Optimisation Spectrum Series 16 7

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

The nodes are:
1. flotation feed = Cu cleaner concentrate + Cu rougher tailing
(or, 1 = 2 + 3);
2. Cu rougher tailing = Pb third cleaner concentrate + Pb
scavenger tailing (or, 3 = 4 + 5); and
3. Pb scavenger tailing = Zn second cleaner concentrate + Zn
scavenger tailing (or, 5 = 6 + 7).
In this case, at the time of surveying the plant, throughput was
120 t/h.
An examination of the tailing assays reveals that the assays are
reasonably similar and in the expected range (indicating that the
sampling was of a good standard). Therefore, taking an average
of the two tailing sample assays is acceptable. However, if one
assay was considerably higher than the other (for example, the
copper cleaner tailing A, in Table 4), one suspects that this dip
sample has been contaminated with froth during collection, and
this assay would be rejected from the data set. It is likely that
having this sample reassayed would only confirm that the sample
was contaminated during collection. However, care must be taken
when rejecting assays because it may not always the case. This
sample has a similar assay to that obtained for the copper rougher
concentrate, so another possibility may be that samples have been
wrongly labelled.
Having sorted the data, identified the process streams, nodes
and feed tonnage, the numbers can be plugged into your mass
balancing program. In this case MATBAL, using a Monte Carlo
simulation, was employed. The resultant mass balance is
provided in Table 5, which includes a sigma value. If sigma is
less than five per cent, then the data are considered to be good.
Step 3
The next step in the mass balance is to balance the internal, or
‘inner’ circuits within the copper, lead and zinc flotation circuits.

TABLE 4
Raw assay data from the metallurgical survey of the Eureka Concentrator.

Sample Process stream % Solids Raw assays

number Ag (ppm) Cu (%) Pb (%) Zn (%) Fe (%)
1 Flotation Feed 48.4 135 0.43 4.02 15.80 14.30
2 Copper rougher concentrate 51.2 588 16.50 15.20 12.30 14.20
3 Copper rougher tailing A 46.5 92 0.19 4.04 15.40 13.40
4 Copper rougher tailing B 46.5 98 0.21 4.04 15.60 13.60
5 Copper cleaner concentrate 64.1 229 22.40 4.07 7.80 16.10
6 Copper cleaner tailing A 10.2 524 15.45 16.30 11.30 13.80
7 Copper cleaner tailing B 9.7 424 5.50 26.40 18.50 9.00
8 Lead rougher concentrate 63.0 184 0.32 21.42 29.97 8.97
9 Lead rougher tailing A 42.1 43 0.16 1.22 17.90 13.60
10 Lead rougher tailing B 40.7 34 0.16 1.25 16.60 13.50
11 Lead scavenger concentrate 62.7 179 0.36 8.16 39.84 11.57
12 Lead scavenger tailing A 42.2 38 0.14 0.67 15.20 13.90
13 Lead scavenger tailing B 40.9 18 0.15 0.50 14.60 13.70
14 Lead first cleaner concentrate 65.9 600 0.30 44.68 14.10 4.10
15 Lead first cleaner tailing A 55.9 184 0.28 18.50 30.65 8.80
16 Lead first cleaner tailing B 55.6 176 0.28 20.50 30.75 9.00
17 Lead second cleaner concentrate 67.0 960 0.51 50.90 10.10 2.50
18 Lead second cleaner tailing A 62.3 814 0.23 47.30 13.40 3.85
19 Lead second cleaner tailing B 54.8 800 0.27 47.40 13.60 3.95
20 Lead third cleaner concentrate 64.0 1060 1.08 60.00 9.40 1.70
21 Lead third cleaner tailing A 61.3 890 0.32 47.00 11.26 2.40
22 Lead third cleaner tailing B 62.2 900 0.40 49.00 11.34 2.80
23 Zinc rougher concentrate 58.2 53 0.33 1.63 45.10 8.40
24 Zinc rougher tailing A 36.5 16 0.13 0.45 1.72 15.35
25 Zinc rougher tailing B 36.4 15 0.13 0.55 1.74 15.65
26 Zinc scavenger concentrate 48.1 72 0.47 2.23 15.20 14.30
27 Zinc scavenger tailing A 36.0 12 0.09 0.32 0.85 15.32
28 Zinc scavenger tailing B 35.1 13 0.11 0.36 0.87 15.48
29 Zinc first cleaner concentrate 57.3 43 0.33 1.30 55.70 6.00
30 Zinc first cleaner tailing A 48.1 63 0.38 2.02 19.40 12.10
31 Zinc first cleaner tailing B 48.0 60 0.42 1.94 19.60 12.30
32 Zinc second cleaner concentrate 55.3 39 0.29 1.17 58.40 5.30
33 Zinc second cleaner tailing A 55.8 62 0.48 1.60 42.70 8.50
34 Zinc second cleaner tailing B 55.7 64 0.44 1.64 44.70 8.50

8 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

TABLE 5
Adjusted assays for the ‘outer’ circuit mass balance.

No Stream t/h Adjusted assays (%)

Ag (ppm) Cu Pb Zn Fe
1 Flotation feed 120.00 0.09 135 0.43 3.99 15.04 13.17
2 Cu cleaner concentrate 1.21 3.31 229 22.40 4.07 7.80 16.11
3 Cu rougher tailing 118.79 0.10 95 0.21 3.99 15.12 13.14
4 Pb 3rd cleaner concentrate 6.76 2.19 1060 1.07 61.01 9.42 1.70
5 Pb scavenger tailing 112.03 0.16 28 0.15 0.55 15.46 13.83
6 Zn 2nd cleaner concentrate 27.70 2.14 39 0.29 1.18 59.92 5.35
7 Zn scavenger tailing 84.33 0.73 12 0.11 0.34 0.86 16.62

1. Flotation feed Cu rougher (1) 3. Cu rougher tailing

2. Cu rougher concentrate
5. Cu cleaner tailing

Cu cleaner (2)

4. Cu cleaner concentrate
(Final Cu Con)

FIG 5 - The copper circuit ‘inner’ mass balance.

These internal balances are broken down into ‘bite size’ pieces to In Balance 1 (the lead second and third cleaners), by using the
simplify the balancing process, and generate estimates of third cleaner concentrate estimated tonnage generated in the
tonnages that can be used in subsequent ‘inner’ circuit balances. ‘outer’ mass balance in Step 2 it is possible to estimate the
The copper circuit (Figure 5) is comparatively easy to balance tonnage for the lead first cleaner concentrate and the lead second
using the tonnage estimates for the flotation feed and copper cleaner tailing. Using these two values, it is then possible to mass
cleaner concentrate generated in the ‘outer’ circuit mass balance. balance the lead first cleaner. This will give tonnage estimates for
In this case the process streams are: the lead rougher concentrate and the lead first cleaner tailing.
1. flotation feed (120 00 t/h), Balance 3 uses the tonnage estimate for the lead scavenger
tailing determined in the ‘outer’ balance in Step 2. The mass
2. copper rougher concentrate,
balance of the lead scavengers yields an estimate of the tonnage
3. copper rougher tailing, for the lead scavenger concentrate.
4. copper cleaner concentrate (1.21 t/h), and Once these three ‘inner’ mass balances are complete it is
possible to fix certain tonnages (ie copper rougher tailing, lead
5. copper cleaner tailing. scavenger tailing, lead first, second and third cleaner concentrates),
The nodes are: and complete a mass balance of the lead circuit. The process
1. flotation feed + Cu cleaner tailing = Cu rougher concentrate streams used for this balance are:
+ Cu rougher tailing (or, 1 + 5 = 2 + 3); and 1. copper rougher tailing,
2. Cu rougher concentrate = Cu cleaner concentrate + Cu 2. lead rougher concentrate,
cleaner tailing (or, 2 = 4 + 5).
3. lead rougher tailing,
The resultant mass balance assays are provided in Table 6.
A similar approach is adopted for the lead circuit (Figure 6) 4. lead scavenger concentrate,
where the circuit is balanced in four parts: 5. lead scavenger tailing,
1. the lead second and third cleaners, 6. lead first cleaner concentrate,
2. the lead first cleaner, 7. lead first cleaner tailing,
3. the lead scavenger, and 8. lead second cleaner concentrate,
4. the lead rougher/cleaner circuit. 9. lead second cleaner tailing,
While this arrangement may appear to be counter intuitive, the
10. lead third cleaner concentrate, and
reasons for moving backwards through the lead circuit are driven
by the need to estimate tonnage figures for the recycle streams. 11. lead third cleaner tailing.

Flotation Plant Optimisation Spectrum Series 16 9

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

TABLE 6
Adjusted assays for the copper circuit ‘inner’ mass balance.

No Stream t/h Adjusted assays (%)

Ag (ppm) Cu Pb Zn Fe
1 Flotation feed 120.00 0.10 135 0.44 3.99 15.04 13.17
2 Cu cleaner con 2.39 1.23 588 14.78 15.20 12.75 13.10
3 Cu rougher tailing 118.79 0.10 95 0.21 3.99 15.12 13.14
4 Cu cleaner con 1.21 0.09 229 23.63 4.14 7.71 16.83
5 Cu cleaner tailing 1.18 2.52 424 5.59 26.69 18.00 9.22

Pb rougher feed Balance 3

1. Cu rougher tailing Pb rougher 3. Pb rougher tailing

Pb scavenger
5. Pb scavenger tailing
(1) (2)

2. Pb rougher concentrate
4. Pb scavenger concentrate
Balance 2
Pb 1st cleaner feed Pb 1st cleaner 7. Pb 1st cleaner tailing
(3)

6. Pb 1st cleaner concentrate

Pb 2nd cleaner feed Pb 2nd cleaner

9. Pb 2nd cleaner tailing
(4)

8. Pb 2nd cleaner concentrate

Balance 1
Pb 3rd cleaner feed Pb 3rd cleaner 11. Pb 3rd cleaner tailing
(5)

10. Pb 3rd cleaner concentrate

(Pb final con)

FIG 6 - The ‘inner’ mass balances for the lead circuit.

And, the nodes are: balancing (Balance 2) step involves the zinc scavenger circuit.
When these two internal mass balances are complete, it is
1. Cu rougher tailing + Pb scavenger concentrate + Pb first
possible to mass balance the zinc circuit while fixing certain
cleaner tailing = Pb rougher concentrate + Pb rougher flows. The process streams used for this balance are:
tailing (or, 1 + 4 + 7 = 3 + 3);
1. lead scavenger tailing,
2. Pb rougher tailing = Pb scavenger concentrate + Pb
scavenger tailing (or, 3 = 4 + 5); 2. zinc rougher concentrate,
3. Pb rougher concentrate + Pb second cleaner tailing = Pb 3. zinc rougher tailing,
first cleaner concentrate + Pb first cleaner tailing (or, 2 + 9 4. zinc scavenger concentrate,
= 6 + 7);
5. zinc scavenger tailing,
4. Pb first cleaner concentrate + Pb third cleaner tailing = Pb
second cleaner concentrate + Pb second cleaner tailing (or, 6. zinc first cleaner concentrate,
6 + 11 = 8 + 9); and 7. zinc first cleaner tailing,
5. Pb second cleaner concentrate = Pb third cleaner 8. zinc second cleaner concentrate, and
concentrate + Pb third cleaner tailing (or, 8 = 10 + 11).
9. zinc second cleaner tailing.
The mass balanced assays for the lead circuit are given in
Table 7. And, the nodes are:
The same approach is employed when mass balancing the inner 1. Pb scavenger tailing + Zn scavenger concentrate + Zn first
circuits for zinc flotation (Figure 7). The first balance (Balance 1) cleaner tailing = Zn rougher concentrate + Zn rougher
examines the zinc first and second cleaner, while the second mass tailing (or, 1 + 4 + 7 = 3 + 3);

10 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

TABLE 7
Adjusted assays for the ‘inner’ lead circuit mass balance.

No Stream t/h Adjusted assays (%)

Ag Cu Pb Zn Fe
1 Cu rougher tailing 118.79 0.05 87 0.20 4.00 14.86 13.17
2 Pb rougher con 134.72 1.66 208 0.32 21.53 29.91 8.81
3 Pb rougher tailing 122.16 0.09 40 0.16 1.23 17.25 13.68
4 Pb scavenger con 10.14 1.34 179 0.36 8.20 39.79 11.55
5 Pb scavenger tailing 112.02 0.09 27 0.15 0.59 15.20 13.86
6 Pb 1st cleaner con 110.85 0.10 631 0.30 44.67 13.63 3.91
7 Pb 1st cleaner tailing 127.95 1.17 162 0.28 19.48 30.99 9.18
8 Pb 2nd cleaner con 31.98 0.09 940 0.51 50.72 10.54 2.45
9 Pb 2nd cleaner tailing 104.08 0.11 602 0.25 43.65 13.91 4.06
10 Pb 3rd cleaner con 6.75 0.09 1077 1.08 60.39 9.35 1.71
11 Pb 3rd cleaner tailing 25.23 0.11 903 0.36 48.13 10.86 2.65

Zn rougher feed Balance 2

Zn rougher Zn scavenger
1. Pb rougher tailing 3. Zn rougher tailing 5. Zn scavenger tailing
(1) (2)

2. Zn rougher concentrate
4. Zn scavenger concentrate

Zn 1st cleaner feed Zn 1st cleaner 7. Zn 1st cleaner tailing

(3)

6. Zn 1st cleaner concentrate

Zn 2nd cleaner feed Zn 2nd cleaner 9. Zn 2nd cleaner tailing

(4)

Balance 1

8. Zn 2nd cleaner concentrate

(Zn final con)

FIG 7 - The ‘inner’ mass balances for the zinc circuit.

2. Zn rougher tailing = Zn scavenger concentrate + Zn copper rougher is actually the fresh flotation feed plus the copper
scavenger tailing (or, 3 = 4 + 5); cleaner tailing. The sum of these recovery values should be the
same as the copper rougher concentrate plus the tailing. That is,
3. Zn rougher concentrate + Zn second cleaner tailing = Zn for copper:
first cleaner concentrate + Zn first cleaner tailing (or, 2 + 9
= 6 + 7); and Rflotation feed + RCu cleaner tailing = RCu rougher con + RCu rougher tailing

4. Zn first cleaner concentrate = Zn second cleaner concentrate
+ Zn second cleaner tailing (or, 6 = 8 + 9).
The adjusted assays for the zinc circuit are provided in Table 8.
Step 4
With the inner mass balances complete it is now possible to mass
balance the whole circuit. The resultant mass balance is given in
Table 9. When reviewing your mass balance it is always a good
idea to check that the internal workings of the circuit are
balanced. For example, in the copper circuit, the feed to the
100.00 + 12.54 = 66.15 + 46.39
112.54 = 112.54
The balance holds, so we can have confidence that the mass
balancing calculations have been completed correctly.
What does it mean? (Chapter 2)
With the mass balance completed it is time to analyse the data and
determine what it means. From this analysis it is possible to
establish what and where the weaknesses are in the circuit, and

Flotation Plant Optimisation Spectrum Series 16 11

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

TABLE 8
Adjusted assays for the ‘inner’ zinc circuit mass balance.

No Stream t/h Adjusted assays (%)

Ag Cu Pb Zn Fe
1 Pb scavenger tailing 112.02 0.07 28 0.15 0.57 15.15 13.20
2 Zn rougher con 43.74 0.07 53 0.33 1.52 44.32 7.98
3 Zn rougher tailing 89.83 0.18 16 0.13 0.47 1.74 15.69
4 Zn scavenger con 5.51 2.56 72 0.47 2.26 15.17 14.29
5 Zn scavenger tailing 84.32 0.09 12 0.11 0.36 0.86 15.78
6 Zn 1st cleaner con 37.44 0.09 43 0.33 1.32 54.80 6.15
7 Zn 1st cleaner tailing 16.04 0.10 63 0.40 2.04 19.55 12.53
8 Zn 2nd cleaner con 27.70 0.09 39 0.29 1.22 58.66 5.34
9 Zn 2nd cleaner tailing 9.74 0.47 64 0.46 1.61 43.84 8.42

TABLE 9
Mass balance of copper, lead and zinc circuits of the Eureka Concentrator.

No Stream t/h Cu (%) Pb (%) Zn (%) Fe (%)

Grade Recovery Grade Recovery Grade Recovery Grade Recovery
1 Flotation feed 120.00 0.05 0.43 100.00 4.00 100.00 14.92 100 13.06 100.00
2 Cu rougher con 2.39 0.41 14.29 66.15 15.10 7.50 12.76 1.70 13.10 1.99
3 Cu rougher tailing 118.79 0.05 0.20 46.39 4.00 98.97 14.99 99.48 13.02 98.69
4 Cu cleaner con 1.21 0.11 22.74 53.61 4.07 1.03 7.71 0.52 16.84 1.31
5 Cu cleaner tailing 1.18 0.84 5.52 12.54 26.54 6.47 17.99 1.18 9.22 0.69
6 Pb rougher con 134.72 3.12 0.32 85.07 21.44 611.75 29.86 228.56 8.84 77.31
7 Pb rougher tailing 122.16 0.23 0.17 39.52 1.24 31.39 17.40 119.05 13.52 105.65
8 Pb scav con 10.14 3.00 0.36 7.25 8.17 17.72 39.71 23.11 11.58 7.70
9 Pb scav tailing 112.02 0.10 0.14 32.27 0.58 13.67 15.33 95.94 13.70 97.95
10 Pb 1st cleaner con 110.85 0.10 0.30 64.56 44.68 1030.87 13.63 84.37 3.91 27.69
11 Pb 1st cleaner tailing 127.95 2.27 0.28 70.95 19.41 526.44 30.92 225.02 9.21 76.58
12 Pb 2nd cleaner con 31.98 0.10 0.51 31.72 50.75 337.91 10.54 18.84 2.45 5.00
13 Pb 2nd cleaner tailing 104.08 0.11 0.25 50.44 43.64 945.57 13.91 80.84 4.06 26.95
14 Pb 3rd cleaner con 6.75 0.09 1.08 14.12 60.64 85.30 9.36 3.53 1.71 0.74
15 Pb 3rd cleaner tailing 25.23 0.13 0.36 17.60 48.10 252.61 10.86 15.30 2.65 4.26
16 Zn rougher con 43.74 0.39 0.33 27.91 1.48 13.51 44.77 109.42 8.00 22.33
17 Zn rougher tailing 89.93 0.24 0.13 21.96 0.50 9.38 1.74 8.75 16.30 93.53
18 Zn scav con 5.51 3.59 0.47 5.12 2.23 2.58 15.14 4.71 14.26 5.07
19 Zn scav tailing 84.32 0.07 0.10 16.85 0.39 6.79 0.86 4.04 16.44 88.46
20 Zn 1st cleaner con 37.44 0.09 0.33 24.19 1.30 10.16 55.33 115.74 6.17 14.74
21 Zn 1st cleaner tailing 16.04 1.07 0.40 12.49 1.99 6.64 19.52 17.52 12.51 12.83
22 Zn 2nd cleaner con 27.70 0.09 0.29 15.42 1.19 6.88 59.41 91.91 5.38 9.50
23 Zn 2nd cleaner tailing 9.74 0.38 0.46 8.76 1.62 3.28 43.76 23.83 8.41 5.23

determine what additional tests are required to give more definition
to the data. As the mass balance was completed using elemental
assays it is now possible to use this data to calculate what is
happening on a mineral basis. That is, the mass balanced
elemental assays can be converted to minerals by making certain
assumptions about the elemental composition of the various
minerals of interest within your system. Appendix 2 provides an
example of these element to mineral calculations. The conversion
to minerals allows you to examine the flotation behaviour of the
iron sulfide and non-sulfide gangue species. Combining the
calculated non-sulfide gangue mass balanced data with the water
recovery data it is possible to gain an appreciation of how these
minerals are being recovered (ie entrainment). Chapter 2: Existing
Methods for Process Analysis by Bill Johnson provides a detailed
description of how to analyse and interpret your plant survey data.
The copper grade/recovery curve for the copper circuit appears
in Figure 8. This data, in conjunction with Table 9, indicates that
there is a small circulating load (about 12 per cent) of copper
returning to the rougher feed via the copper cleaner tailing. This
is not a major concern. However, the poor selectivity for
chalcopyrite against galena and sphalerite during copper roughing
(ie the high lead and zinc grades) is an issue, and requires further
investigation.
Figure 9 contains the lead grade/recovery curve for the Eureka
circuit. It is immediately obvious that there is a large circulating
load of galena within the lead circuit centred around the lead first

12 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

25.0

Cu cleaner concentrate

20.0

Cu grade, %
15.0
Cu rougher concentrate

10.0
Cu rougher feed

5.0

Flotation feed
0.0
0.0 20.0 40.0 60.0 80.0 100.0 120.0
Cu recovery, %

FIG 8 - The copper grade/recovery curve for the copper circuit within the Eureka Concentrator.

70.0

Pb 3rd cleaner concentrate

60.0
Pb 2nd cleaner concentrate
50.0 Pb 2nd cleaner feed
Pb grade, %

st
Pb 1 cleaner concentrate
40.0

Pb 1st cleaner feed

30.0

Pb rougher concentrate
20.0

Cu rougher tailing
10.0 Pb rougher feed

Flotation feed
0.0
0.0 200.0 400.0 600.0 800.0 1000.0 1200.0 1400.0 1600.0 1800.0
Pb recovery, %

FIG 9 - The lead grade/recovery curve for the lead circuit within the Eureka Concentrator.

70.0
nd

60.0
st

50.0
Zn grade, %

Zn rougher concentrate st
40.0

30.0

20.0
Zn rougher feed
Pb rougher tailing
10.0

Cu rougher tailing Flotation feed

0.0
40.0 60.0 80.0 100.0 120.0 140.0 160.0
Zn recovery, %

FIG 10 - The zinc grade/recovery curve for the zinc circuit within the Eureka Concentrator.

cleaner bank. A closer examination of the lead circuit mass
balanced data in Table 9 reveals that accompanying the very large
circulating loads of galena is a large circulating load of sphalerite.
This would suggest that there may be liberation issues within this
part of the circuit. Again, this warrants further investigation.
The zinc grade/recovery curve (Figure 10) also reveals that there
are circulating loads of zinc around the zinc first cleaner. While
these loads are not as severe as those noted in the lead circuit,
undoubtedly they are having an impact on the zinc metallurgy.
Consulting Table 9, it was noted that there may be an issue with
iron, particularly around the rougher/scavenger/first cleaner circuit.
Further investigation is again needed to define the problem.
It is apparent that there are opportunities for improvement in all
three circuits. Equally, it is obvious that making improvements in

Flotation Plant Optimisation Spectrum Series 16 13

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT
the copper circuit will have ramifications on both the lead and
zinc circuits. And this makes for a good rule of thumb: start at the
front of the circuit and work downstream. The test programs need
to be prioritised so that resources can be focused on the project
that has the greatest potential metallurgical (and financial) benefit
to the plant. The survey data indicates that the lead circuit is very
unstable with high circulating loads, and could be impeding both
the lead and zinc circuit performance. It is recommended that
further work be completed to determine the recovery-by-size and
liberation characteristics through the lead circuit to see where
improvements may be made.
Sample identification and storage
At the risk of stating the obvious, having a workable sample
identification and storage system will make life easier. A lack of
attention to detail in this area will lead to mistakes in the assay
laboratory with poor labelling, and can jeopardise the ability to
locate samples for check assaying or for further testing (for
example, sizing or mineralogical examination). Therefore, it is
necessary to set up a system of identification that is both simple
and effective. For example, all surveys are labelled with an ‘S’
and numbered consecutively, with each sample in that survey
given a number. So, if the survey described in this chapter is the
tenth survey completed in the Eureka Concentrator its
identification code would be S10/1 to S10/34, where the sample
numbers 1 to 34 correspond to the sample numbers listed in
Table 2. The next survey would be S11, and so on. Flotation tests
could be identified according to the operator’s initials, for
example, CG351/1 to CG351/5, which would be read as flotation
test 351 completed by CG with five test products.
Once a numbering system has been put in place it makes it
easy to develop a storage system. In developing your storage
system you must set some rules regarding the length of time you
are going to keep a sample, and then allow the laboratory
technicians time, periodically, to maintain the storage facility. For
example, it is wise to keep flotation test produces for nominally
three months and plant surveys for up to 12 months. It is also
wise have a clean up every three months to dispose of samples
that have passed their storage date.
The pulp chemistry survey (Chapter 6)
Concurrent to the metallurgical survey, pulp chemical data was
collected from the plant to determine the pulp chemical conditions.
What do I have to do?
The pulp chemical survey involves collecting Eh, pH, dissolved
oxygen, and temperature data from the following process streams:
• SAG mill discharge,
• cyclone underflow,
• ball mill discharge,
• cyclone overflow,
• copper circuit feed,
• lead circuit feed,
• zinc circuit feed, and
• final tailings.
To complete a pulp chemical survey, a sample of slurry is ‘cut’
from the process stream of interest, and poured into a small
beaker. The sample is then stirred gently with the probes for
nominally two minutes until equilibrium readings are obtained.
The Eh, pH, dissolved oxygen, and temperature data are logged
14 Spectrum Series 16
on a data logger. The logged data can then be downloaded from
the data logger to a laptop computer where it is able to be
manipulated.
All of the probes should be checked to ensure they are clean
and in good working order, and it is imperative that they are
calibrated before use. The pH probe should be calibrated using
pH buffer solutions seven and ten, if this is the range expected in
the plant. Alternative buffer solutions should be used if the pulp
pH is outside this range. The Eh probe should be calibrated, for
example, using Zobell solution (1:1 solution of Part A and B =
231 mV at 24°C), the dissolved oxygen probe calibrated in a
0.2 g/L solution of sodium sulfite for the zero calibration, and air.
The temperature probe can be calibrated with the aid of a
thermometer, using iced and heated water to give a two point
calibration.
To complete the data set, EDTA extractions should be
completed on the same process streams as the pulp chemistry. For
the Eureka survey, each stream was ‘cut’ and the sample poured
into a small wide mouth, screw top sample bottle. The samples
were taken back to the laboratory, and the wet weight recorded.
The EDTA extraction procedure used was:
• stir the pulp sample with a 25 ml syringe to produce
homogeneous slurry;
• syringe a 25 ml aliquot of slurry from the sample bottle and
weigh;
• inject the contents of the syringe into a 400 ml beaker
containing 250 ml of three per cent (by weight) EDTA
solution, pH modified to 7.5 with sodium hydroxide;
• thoroughly mix the slurry and EDTA solution using a
magnetic stirrer, for five minutes;
• filter the slurry using a 0.2 micron millipore filter; and
• submit the filtered EDTA solution for assay.
The remainder of the pulp sample was pressure filtered, and the
solids dried. The dry solids were weighed and submitted for
assay.
What data do I need?
The pulp chemical data (pH, Eh, dissolved oxygen and
temperature) are stored on a data logger. These data now need to
be downloaded into Excel, massaged and put into a logical format
for analysis and interpretation.
The samples for EDTA extraction (solids and liquids) are
prepared and submitted for assay. Usually these samples are
assayed for the base metals of interest; in the case of the Eureka
Concentrator, the solids and liquids were assayed for copper,
lead, zinc and iron.
As with the metallurgical surveys, other data that can be useful
when analysing the EDTA survey are:
• the throughput at the time of the survey,
• reagent additions and other plant operating parameters (ie air
flow rates and pulp levels),
• OSA readings, and
• information about the ore being treated.
In the context of a one off survey some of these pieces of
information may not be of great value. However, when the analysis
is extended to include other surveys on other ore blends, circuit
configurations and reagent suites, these data provide a vital link in
the comparison.
It can be beneficial to collect process water samples routinely
to determine the free ions in solution (ie base metal ions (copper,
lead, zinc, iron), calcium, magnesium, sulfate and chloride).
Flotation Plant Optimisation
 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

How do I analyse the data? examining the pulp chemistry of a regrind/cleaner circuit, it is
normal to collect a sample of the fresh concentrate that feeds the
With the exception of the Eh data the other pulp chemical
regrinding section. It is also good policy when completing surface
measurements can be read directly from the measured data. The
analysis comparing pairs of streams (ie rougher concentrate and
Eh should be adjusted so that it is referenced against the standard
hydrogen electrode (SHE). Chapter 6: Chemical Measurements rougher tailing), to collect the pulp chemistry of the two process
During Plant Surveys and Their Interpretation by Stephen Grano streams of interest.
provides an explanation of how to manipulate and analyse the
data you have collected. What does it mean?
One method of analysing the EDTA data is given by Rumball The pH, Eh, dissolved oxygen, and temperature data should be
and Richmond (1996), who developed a simple relationship for
plotted in such a way to represent the pulp flow profile through
calculating the percentage of oxidised mineral present in a pulp
the circuit to assist with the analysis and interpretation of the
from EDTA extraction data (Equation 1):
data. The profiles through the Eureka circuit are displayed in
Figures 11 to 14, respectively.
Mass of M in EDTA solution
% EDTA extractable M = × 100 (1) The pH profile (Figure 11) indicated that grinding occurs at
Mass of M in solids
natural pH (in the range 8.0 to 8.5). The slight reduction in pH in
the ball mill discharge maybe attributed to pyrite oxidation.
where: Sodium metabisulfite (SMBS) was added to the copper rougher
M is the metal ion under investigation feed, to depress galena during copper flotation. The addition of
SMBS resulted in the pH being reduced to approximately 6.0.
The data is generally presented graphically with the pulp
The pH was then increased to 8.0 for galena flotation and 11.5
chemical parameter on the y-axis and the circuit position on the
x-axis. The process streams are set to mimic the normal flow of sphalerite flotation thought the addition of lime.
the slurry through the circuit. In this case: Figure 12 contains the Eh profile through the circuit. The Eh
shifted to slightly less oxidising pulp potentials as the pulp
1. SAG mill discharge, flowed from the SAG mill through the ball mill, and became
2. cyclone underflow, more oxidising during flotation where air is used as the flotation
gas.
3. ball mill discharge,
The dissolved oxygen profile (Figure 6) through the circuit
4. cyclone overflow, indicated that the oxygen content of the pulp was negligible
5. copper rougher feed, during both grinding and flotation, registering zero on the meter.
It was not until the zinc flotation stage that the dissolved oxygen
6. lead rougher feed, content of the pulp was increased to 4.5 ppm in the final tailing.
7. zinc rougher feed, and These data suggest that the ore is very reactive and continues to
consume oxygen throughout the process.
8. final tailing.
The pulp temperature ranged 33 to 35°C across most of the
When investigating the pulp chemistry within your circuit often circuit (Figure 14). The spike in the pulp temperature in the ball
the urge is to sample every stream in the concentrator. While this mill discharge can be attributed to part of the addition grinding
may be a good idea initially to get an idea of the variations energy being dissipated as heat.
between process streams, it soon becomes apparent that some of The Eh-pH data for the grinding and flotation circuits of the
this data is redundant. For example, concentrate streams usually Eureka Concentrator have been plotted in Figure 15 to determine
have high dissolved oxygen contents and very oxidising pulp where the reactions are occurring. From the Nernst Equation 2
potentials because the froth in the concentrate contains significant there is a dependence of redox potential on pH:
amounts of air. Therefore, when conducting a pulp chemistry
survey of the primary grinding/rougher flotation circuit
0.059 ⎛ α Reactants ⎞
measuring the pulp chemistry of the rougher concentrate can E = E °+ log 10 ⎜ ⎟ (2)
produce numbers that skew the analysis. However, if you are n ⎝ α Products ⎠

12.0

10.0

8.0
pH

6.0

4.0

2.0

0.0
w

g
w

d
ge

d
ge

lin
ee

ee
ee
f lo
flo
ar

r
ha

ai
rf
rf

rf
er
r
ch

de

rt
he
ov

he

he
sc
s

he
un

di
di

ug
ug

ug
ne

ug
ne

ill
ill

ro
ro

ro
clo
m
m

ro
clo

Cu

Zn
Pb
Cy
G

ll

Zn
Ba
Cy
SA

Circuit position

FIG 11 - The pH profile through the grinding and flotation circuits of the Eureka Concentrator.

Flotation Plant Optimisation Spectrum Series 16 15

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

200.0

160.0

Eh, mV (SHE)
120.0

80.0

40.0

0.0

g
w

d
ge

d
ge

lin
ee

ee
ee
flo

f lo
ar

r
ha

ai
rf
rf

rf
r

er
ch

de

rt
he
he

he
sc

ov
s

he
un

di
di

ug
ug

ug
ne

ug
ne

ill
ill

ro
ro

ro
clo
m
m

ro
clo

Cu

Zn
Pb
G

Cy
ll

Zn
Ba
Cy
SA

Circuit position
FIG 12 - The Eh profile through the grinding and flotation circuits of the Eureka Concentrator.

6.0

5.0

4.0
DO, ppm

3.0

2.0

1.0

0.0
w

g
w

d
ge

d
ge

lin
ee

ee
ee
f lo
flo
ar

r
ha

ai
rf
rf

rf
er
r
ch

de

rt
he
ov

he

he
sc
s

he
un

di
di

ug
ug

ug
ne

ug
ne

ill
ill

ro
ro

ro
clo
m
m

ro
clo

Cu

Zn
Pb
Cy
G

ll

Zn
Ba
Cy
SA

Circuit position
FIG 13 - Dissolved oxygen profile through the grinding and flotation circuits of the Eureka Concentrator.

45.0
40.0
35.0
Temperature, C
o

30.0
25.0
20.0
15.0
10.0
5.0
0.0
g
w

d
ge

d
ge

lin
ee

ee
ee
flo

f lo
ar

r
ha

ai
rf
rf

rf
r

er
ch

de

rt
he
he

he
sc

ov
s

he
un

di
di

ug
ug

ug
ne

ug
ne

ill
ill

ro
ro

ro
clo
m
m

ro
clo

Cu

Zn
Pb
G

Cy
ll

Zn
Ba
Cy
SA

Circuit position
FIG 14 - Temperature profile through the grinding and flotation circuits of the Eureka Concentrator.

Applying the Nernst equation to water results in a Pourbaix E 0 2 = +1.23 + 0.015 log 10 po2 − 0.059 pH (3)
diagram that describes three domains, separated by lines of
equilibria. The upper most of these is the water-oxygen line This can be simplified further (Johnson, 1988; Natarajan and
(Equation 3), above which water decomposes and oxygen is Iwasaki, 1973) for an oxygenated aqueous solution with no well
evolved, and below which water is stable: defined redox couples to (Equation 4):

16 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

E 0 2 = +0.9 − 0.059 pH
In broad terms, if the changes in Eh and pH result in a line
parallel to the water-oxygen line this means that water equilibria
is being maintained. That is, any change in Eh is directly
proportional to a change in pH with a similar relationship to that
expressed in Equation 4. If the changes in Eh and pH result in a
line that is perpendicular to the water-oxygen line then the
evidence suggests that oxidative reactions are occurring.
Figure 15 shows that the Eh-pH lines between points 1 to 4 are
perpendicular to the water-oxygen line. This indicates that
oxidative reactions are occurring in the grinding circuit. It is
likely that these oxidative reactions are corrosion of the grinding
media and oxidation of the sulfide minerals. As the pH is
decreased across the ball mill it is suggested that pyrite oxidation
(an acidic reaction) may be one of the dominant reactions. The
changes in the Eh-pH curve between Points 4 and 5, 5 and 6, and
6 and 7 can be attributed to the addition of reagents that alter the
pH of the system. During copper flotation it is the addition of
SMBS, while in the lead and zinc circuits it is the addition of
lime.
The EDTA extractable copper, lead, zinc and iron data are
presented in Figure 16. The EDTA extractable copper profile
(Figure 9) suggested that the percentage of oxidised chalcopyrite
250
(4) remained negligible throughout the grinding circuit. The per cent
EDTA extractable copper increased during the copper flotation
step to approximately 2.5 per cent and this increases to 17 per
cent in the zinc rougher feed, which can be attributed to the
addition of copper sulfate for sphalerite activation.
The first observation to be made is that the percentage of EDTA
extractable lead is at least an order of magnitude greater than the
values reported for zinc and iron. The percentage of EDTA
extractable lead is higher as galena is a reactive mineral when in
contact with other sulfide minerals, particularly pyrite. The EDTA
extractable lead profile shows that the percentage of oxidised
galena remained approximately constant through the grinding
circuit, with values of around 1.0 per cent. The per cent EDTA
extractable lead increased in the cyclone overflow, and remains
reasonably constant through the copper circuit. After the lead
flotation stage the lead scavenger tailing per cent EDTA extractable
lead was 17 per cent, and then increased to almost 40 per cent after
zinc flotation. These data suggest that the lead species that remain
in the lead scavenger tailing are more heavily oxidised.
The EDTA extractable zinc profile exhibits a very similar trend
to that observed for lead. That is, through the grinding circuit the
percentage EDTA extractable zinc remained approximately
constant, with values of around 0.1 per cent. This indicated that
sphalerite oxidation was largely unchanged through this circuit.

200 6
8

150
Eh, mV (SHE)

4
5
100 1
1. SAG mill discharge;
2
2. Cyclone underflow;
50 3. Ball mill discharge;
4. Cyclone overflow; 7
5. Cu rougher feed; 3
0 6. Pb rougher feed;
7. Zn rougher feed; and
8. Final tailing
-50
4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0
pH

FIG 15 - The Eh-pH curve for the grinding and flotation circuits of the Eureka Concentrator.

42.0 1.2
EDTA extractable Zn and Fe, %
EDTA extractable Cu and Pb, %

35.0 1.0

28.0 0.8

21.0 0.6

14.0 0.4

7.0 0.2

0.0 0.0
w
w

d
e

ed

g
ge

ee
rg

ee

lin
flo

flo

fe
ar
ha

ai
f

rf
r

er
h
de

er

lt
sc

he

he
sc

ov

na
un
di

di

ug

ug

ug
ne

Fi
ne
ill

il l

ro

ro

ro
lo
m

m
lo

yc

Pb

Zn
G

ll
yc

C
Ba

C
SA

Circuit position
Cu Pb Zn Fe

FIG 16 - The EDTA extractable copper, lead, zinc and iron profiles through the grinding and flotation circuits of the Eureka Concentrator.

Flotation Plant Optimisation Spectrum Series 16 17

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT
There was an increase in the per cent EDTA extractable zinc
during the copper and lead flotation circuits (ie from 0.13 to
0.37 per cent) and then a significant increase after the zinc
flotation circuit (to 1.0 per cent). This increase indicated that the
zinc species that remained in the zinc scavenger tailing were
more heavily oxidised.
The EDTA extractable iron profile gives the best indication of
the impact of grinding media corrosion on the system. The EDTA
extractable iron through primary grinding ranged from 0.05 to
0.16 per cent, and then increased to 0.30 per cent in the cyclone
overflow. That is, the pulp entering the flotation circuit contains
elevated levels of EDTA extractable iron, presumably after
contact with the forged grinding media. Interestingly, the levels of
EDTA extractable iron decrease through the flotation circuit. This
is probably due to the continuing oxidation of the iron species to
higher level iron oxides, which are not soluble in EDTA.
Summary of data acquisition
The metallurgical survey suggests that there are a number of
opportunities for improving the metallurgical performance of the
Eureka Concentrator. It is suspected that these may be liberation
related, and further data is required to identify exactly what the
liberation characteristics are. The data strongly suggests that the
first priority should be improving the lead flotation circuit.
The pulp chemistry suggests that the ore is reasonably reactive,
particularly during grinding, and consideration should be given to
improve this aspect of plant performance. However, this should
not be rectified until the liberation issues have been resolved.
PROBLEM DEFINITION
From the metallurgical survey the circulating load observed in the
lead rougher/lead first cleaner circuit were observed to be
extremely high (Table 10), for both lead and zinc. In fact, the
copper scavenger tailing contributed only 15.4 and 28.6 per cent
of the lead and zinc units, respectively, to the lead rougher feed.
The recycle from the lead first cleaner tailing contributed by far
the most lead and zinc to the lead rougher feed (81.9 and 64.7 per
cent, respectively).
Recovery by size (Chapter 2)
The mass balanced data gives you information about the grades
and recoveries in the circuit. The data also provides information
about the circulating loads. However, to gain a greater
appreciation for where problems occur and their magnitude, it is
advisable that samples from selected process streams be sized,
and each size fraction assayed. By mass balancing the size
fractions, the performance of each size fraction can be assessed.
An examination of Table 9 and Figure 9 indicates that the lead
rougher/first cleaner section of the plant has a large circulating
load of galena and sphalerite. It is likely that this large circulating
load produces instability within the circuit, and is probably due to
composite particles. However, at this stage you are only guessing.
TABLE 10
Mass balanced data for the lead rougher feed, with respect to flotation feed. Note: Both the lead scavenger concentrate and the lead first
cleaner tailing are recycled back to the head of the lead rougher.
Stream t/h Grade (%)
Cu Pb Zn
Cu scav tailing 118.8 0.2 4.0 15.0
Pb 1st cleaner tailing 128.0 0.3 19.4 30.9
Pb scav concentrate 10.1 0.4 8.2 39.7
Pb rougher feed 256.9 0.3 11.8 23.9
18 Spectrum Series 16
You need to gather solid data that tells you which size fractions
are being rejected from the circuit, and what their mineralogical
characteristic is. Therefore, the first step in this analysis is to
examine the recovery-by-size around the lead rougher feed.
What do I have to do?
When you conducted the plant survey you may have had every
intention to have the samples sized so that you can complete a
recovery-by-size analysis. However, before having your
metallurgical technicians launch into a large body of work sizing
every sample collected during the survey it is best to mass
balance the plant survey. Sizing samples prior to completing the
mass balance may waste time, resources and money if the survey
does not mass balance easily (ie poor sampling). Once you have
established that the mass balance is good, you need to decide if
completing a recovery-by-size analysis is warranted, and if it is,
what samples you would like to have sized. It is not always
necessary to size every sample in the survey. The sample
selection will depend on what the survey reveals about plant
performance. For example, sizing the concentrate and tailing
from a section will show which size fractions of the valuable
species are misbehaving (ie not being recovered efficiently), and
which size fractions of the gangue species are being recovered.
Sometimes completing recovery-by-size analysis on a down-
the-bank survey can add value because it allows you to determine
where size fractions are recovered in the circuit, and can be used
to determine the kinetics of various species on a size basis.
With these decisions made you need to retrieve the samples
from storage, ensure that you have sufficient sample to complete
a sizing (generally 200 grams is sufficient), and determine what
sizes you require. If you need to obtain size fractions below 38
microns you will need to use a cyclosizer, which will give you six
more size fractions. However, if cyclosizer is used in concert with
a precyclone and a centrifuge the number of size fractions
obtained can be extended to seven.
What data do I need?
With the sizing completed you will require the mass collected in
each size fraction. If the sizing has been restricted to a simple
sieve analysis, then you will need to determine how much mass
you have in each size fraction to ensure that you have sufficient
for assay (usually 5 grams is more than enough). If you do not
have enough sample in each size fraction for an assay you need to
either complete another sizing on this sample, or you combine
size fractions to yield enough mass. However, if you do combine
size fractions you must apply this across all of the samples within
this data set. Not to do so will make it impossible to complete the
analysis. That is, if you were required to combine the +300, +212
and +150 micron fractions in the feed sample, then you will need
to do the same for all other samples in the suite of samples for
this survey.
When it comes to sizing the subsieve fractions using a
precyclone/cyclosizer/centrifuge combination, you will need to
know the water temperature, the elutriation time, the water flow
Recovery (%)
Fe Cu Pb Zn Fe
13.0 46.4 99.0 99.5 98.7
9.2 71.0 526.4 225 76.6
11.6 7.2 17.7 23.1 7.7
11.1 124.6 643.1 347.6 183
Flotation Plant Optimisation
 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

rate and the specific gravity of the mineral being separated. With
these variables known using the cyclosizer manual (Warman
International Limited, 1991) the ‘cut’ size for each cyclone can be
calculated. That is, for quartz the cyclone ‘cut’ sizes for the
cyclosizer are:
• C1 = 44 microns,
• C2 = 33 microns,
• C3 = 23 microns,
• C4 = 15 microns, and
• C5 = 11 microns.
The precyclone generally ‘cuts’ at about 7 microns. To convert
these ‘cut’ sizes to values that are more representative of minerals
such as chalcopyrite, galena, sphalerite and pyrite, then the ‘cut’
size for quartz must be multiplied by the overall correction factor:
f = fT × fSG × fFM × fET
where:
fT is the temperature correction factor
fSG is the mineral specific gravity correction factor
fFM is the flow meter reading correction factor
fET is the elutriation time correction factor
It is important to note that you should correct for each mineral
in the system as each mineral has a different specific gravity. The
cyclone ‘cut’ sizes for each of the cyclosizer cones are given in
Table 11.
Further, it must be realised that sizing the -38 micron material
using the precyclone, cyclosizer and/or the centrifuge represents a
change in the way the particles are sized. When using sieves the
sizing is achieved by passing the particles over a nest of sieves.
Particles that are larger than the screen aperture are retained on
the screen and those that are smaller fall through. However, when
sizing using the precyclone, cyclosizer and/or centrifuge sizing is
completed hydraulically based on size and mineral specific
gravity. When the two sizing methods are used together, the
transition point from one technique to the other can lead to some
unusual shaped curves which are purely due to the change in the
sizing method. To overcome this it is usual to combine the -53
micron to +C2 size fractions. This smooths the curves and
effectively removed the transition between sizing methods.
If you are considering completing a mineralogical analysis it is
at this point that you would ‘split’ out a small representative
portion from the samples of interest prior to pulverising the
samples for assay. Once you have reviewed the weights in each
size fraction for each of the samples you have had sized, and
decided which size fractions to combine, and ensured that there is
enough sample for assay in each size fraction, the samples can be
prepared and submitted of assay.
How do I analyse the data collected?
When the assays return it is important to check that the assays are
in good order. This can be done by calculating the head assay
from the size fraction assays and comparing it with the actual
head assay. If they are in good agreement, then you can assume
that both the sizing and the assays were completed properly.
To construct a graph similar to that presented in Figure 17 the
(5) size and assay data for the copper rougher tailing, the lead first
cleaner tailing, the lead scavenger concentrate and the lead
rougher feed are required. To simplify the analysis only the lead
assay data will be considered in this example. The starting point
for determining the recovery-by-size behaviour in this part of the
circuit is the mass balanced data from the plant survey given in
Table 9, and for this section of the plant Table 10. Using the mass
balanced tonnages for each of the process streams given in
Table 10 the size distribution data determined from the sizing
process (ie the weight per cent data) for each of these sample
points were applied to yield the tonnes of ore in each size fraction
(Table 12). The weight per cent and the assay for each size
fraction are used to calculate the head grade:
n
( mi × ai )
Head grade = ∑ (6)
i=1 ∑ mi
where:
m is the weight per cent is size fraction i
a is the assay for that size fraction
This calculated head grade is then compared with the assay of
the head sample to determine how well the sizing and assay
procedures have been completed. Unfortunately, in this instance
the mass and assay of the -5.6 micron fraction was calculated by
difference. That is:

TABLE 11
Cyclosizer cyclone ‘cut’ sizes for quartz, chalcopyrite, galena, sphalerite and pyrite.

Correction factor Mineral

Quartz Chalcopyrite Galena Sphalerite Pyrite
fT 0.9815 0.9815 0.9815 0.9815
fSG 0.7181 0.5038 0.7416 0.6423
fFM 0.9519 0.9519 0.9519 0.9519
fET 1.0063 1.0063 1.0063 1.0063
f 0.6751 0.4737 0.6972 0.6038
Cyclone ‘cut’ size
C1 44 30 21 31 27
C2 33 22 16 23 20
C3 23 16 11 16 14
C4 15 10 7 11 9
C5 11 7 5 8 7

Flotation Plant Optimisation Spectrum Series 16 19

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

100.0

90.0

80.0

Pb distribution, % 70.0

60.0

50.0

40.0

30.0

20.0 Pb rougher feed

10.0 Pb 1st cleaner tail

0.0 Pb scav con

Total
-5.6
+5.6 Cu scav tail
+7.6
+11.6
+16.7
Particle size, microns +22.2

FIG 17 - Lead distribution by size in the process streams contributing to the lead rougher feed.

TABLE 12
The lead recovery-by-size data for the lead rougher feed calculated from the copper rougher tailing, lead first cleaner tailing and the
lead scavenger concentrate.
Stream/size Wt % Wt (t/h) Pb parameter (%)
Grade Distribution Recovery to Pb Ro feed
Cu scavenger tailing
+22.2 μm 48.79 57.96 4.06 49.49 7.73
+16.7 μm 6.66 7.91 3.00 5.00 0.78
+11.6 μm 10.18 12.09 1.88 4.78 0.75
+7.6 μm 5.32 6.32 2.54 3.38 0.53
+5.6 μm 3.77 4.48 2.24 2.11 0.33
-5.6 μm 25.28 30.03 5.58 35.24 5.51
Head 100.00 118.79 4.00 100.00 15.62
Pb First cleaner tailing
+22.2 μm 82.44 105.48 22.12 93.96 76.72
+16.7 μm 6.13 7.84 8.84 2.79 2.28
+11.6 μm 3.19 4.08 8.64 1.42 1.16
+7.6 μm 1.17 1.50 5.10 0.31 0.25
+5.6 μm 0.81 1.04 4.10 0.17 0.14
-5.6 μm 6.26 8.01 4.18 1.35 1.10
Head 100.00 127.95 19.41 100.00 81.65
Pb scavenger con
+22.2 μm 70.32 7.13 9.66 83.14 2.26
+16.7 μm 10.48 1.06 8.20 10.52 0.29
+11.6 μm 6.69 0.68 2.62 2.15 0.06
+7.6 μm 2.15 0.22 3.20 0.84 0.02
+5.6 μm 1.37 0.14 3.68 0.62 0.02
-5.6 μm 8.99 0.91 2.49 2.74 0.07
Head 100.00 10.14 8.17 100.00 2.72
Pb rougher feed
+22.2 μm 66.40 170.57 15.46 86.72 86.72
+16.7 μm 6.55 16.82 6.05 2.25 2.25
+11.6 μm 6.56 16.85 3.55 1.97 1.97
+7.6 μm 3.13 8.03 3.03 0.80 0.80
+5.6 μm 2.20 5.65 2.62 0.49 0.49
-5.6 μm 15.16 38.95 5.22 6.68 6.68
Head 100.00 256.88 11.84 100.00 100.00

20 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

n−1
A similar approach is adopted for the more traditional
m−5 .6 μm = mH − ∑ m1 (7)
i−1
recovery-by-size curves presented in Figure 18 and 20.

and What does it mean?

Figure 17 contains the distribution by size, for galena, for the
⎛ ⎛n−1 ⎞⎞ three process streams contributing to the lead rougher feed. The
⎜ mH × aH − ⎜ ∑ m1 × ai ⎟ ⎟
⎝ ⎝ i−1 ⎠⎠ vast majority of the galena in the lead rougher feed is in the
a−5 .6 μm = (8) coarse (+20 micron) lead recycled back from the lead first cleaner
m−5 .6 μm
tailing (80 per cent of the total lead in the lead rougher feed).
Thus, the flotation characteristics of the lead rougher/scavenger
where: unit are dominated by the behaviour of this material.
mH and aH are the weight per cent and assay for the head The recoveries versus size data for the lead final concentrate
are provided in Figure 18. These data show clearly that the lead
By calculating the weight per cent and assay for the finest size
final concentrate consists predominantly of fine (-20 micron)
fraction all the errors associated with the sizing, assaying and
galena, with greater than 90 per cent lead recovery for the
sampling are accumulated to this size fraction. In an ideal world
intermediate size fractions (+5/-20 microns). Either side of this
as sample of this material would be collected.
size range the recoveries decrease. The recoveries of gangue
Having worked out the tonnes of ore and the assay for each minerals (sphalerite, iron sulfides and non-sulfide gangue) are
size fraction the distribution-by-size within each process stream, low across all size fractions.
and the recovery-by-size against a particular process stream can
be calculated. The distribution within a process stream is given The galena distribution, by size, of the lead scavenger tailing,
by: with respect to flotation feed, is provided in Figure 19. Clearly the
losses of galena from the lead circuit are bimodal, with significant
⎛ ( m × ai ) ⎞ losses in the fine (-5 microns) fraction (5.6 per cent) and coarse
Distributiona = ⎜ i ⎟ × 100 (9) (+20 microns) fractions (19.9 per cent). Of greatest concern
⎝ ( mH × aH )⎠ however are the losses in the coarse fractions. It is presumed that
these coarse particles occur as galena deficient/sphalerite rich
The distribution of lead within each process stream is provided composites.
in Table 12. In this case the recovery-by-size is calculated for the Figure 20 shows the recovery by size data for the zinc final
lead rougher feed because the copper scavenger tailing, the lead concentrate. These data show the peculiar behaviour of the coarse
first cleaner tailing and the lead scavenger concentrate all (+20 microns) galena, which exhibits recoveries as high as 60 per
combine to form the lead rougher feed. So, the recovery for each cent. It is presumed that this behaviour is related to the
size fraction, i, is given by:
composition of these galena particles (ie galena deficient/
sphalerite rich composites).
⎛ n ⎞
⎜ ∑ ( m p × a p )⎟
Recoveryi , a = ⎜ P= 1 ⎟ × 100 (10) Mineralogy (Chapter 4)
⎜ ( mH × aH ) ⎟
⎜ ⎟ Once the recovery-by-size data has been analysed and a theory
⎝ ⎠
formed, select streams should be examined mineralogically. As
this is a relatively expensive process, it is wise to complete the
where:
work already discussed, prior to obtaining mineralogical analysis.
P are the process streams of interest The recovery-by-size analysis indicates that a significant amount

90.0

80.0

70.0

60.0
Recovery, %

50.0

40.0

30.0

20.0

10.0

0.0
1.0 10.0 100.0 1000.0
Geometric mean particle size, microns
Lead Zinc Copper Iron Silica

FIG 18 - Recovery-by-size data for the lead final concentrate, with respect to flotation feed for the Eureka Concentrator.

Flotation Plant Optimisation Spectrum Series 16 21

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

6.0

Distribution (with respect

5.0

to flotation feed), %
4.0

3.0

2.0

1.0

0.0
2.3

6.0

7.9

11.7

17.6

28.0

41.0

48.8

63.0

89.2

126.1

178.3
Geometric mean particle size, microns

FIG 19 - The galena distribution by size, with respect to flotation feed, in the lead scavenger tailing for the Eureka Concentrator.

100.0
90.0

80.0

70.0
Recovery, %

60.0

50.0

40.0

30.0

20.0

10.0

0.0
1.0 10.0 100.0 1000.0
Geometric mean particle size, microns
Lead Zinc Copper Iron Silica

FIG 20 - Recovery-by-size data for the zinc final concentrate, with respect to flotation feed for the Eureka Concentrator.

of material is recycled back to the lead rougher feed from the lead
first cleaner tailing, and that the vast majority of this material
occurs in the +20 micron size fraction. It is highly likely that
these particles are galena/sphalerite composites. However, in
order to prove this it is necessary to complete a mineralogical
examination.
What do I have to do?
To complete a mineralogical analysis of samples from you plant
you need to provide the company completing the examination for
you with a representative sample of the process stream that has
not been pulverised. For liberation analysis, the sample provided
will be of a number of closely sized size fractions. If you have
completed a recovery-by-size analysis you should have prepared
the samples for mineralogical examination at the same time.
With the samples available for analysis it is then necessary to
choose the method by which the mineralogical data will be
collected. These methods range from optical microscopy through
to automated methods such as QEMSCAN and MLA. Chapter 4:
A Practical Guide to Some Aspects of Mineralogy that Effect
Flotation, by Alan Butcher, describes the methods available.
What data do I need?
When shipping samples off to a laboratory for mineralogical
analysis it is wise to provide the mineralogist with a scope of
work, and assays of the samples that will be examined. The scope
of work will inform the mineralogist of the questions you would
like answered. For example, at Eureka you are interested in the
locking and liberation characteristics of galena and sphalerite
around the lead rougher/first cleaner circuit. It is then clear to the
mineralogist that the examination should provide information
about the degree of liberation of these minerals, and how they are
locked with other minerals. Supplying the assays for the samples
you have supplied for examination allows the mineralogist to
reconcile his calculated head grade based on the minerals present
back to the assays. It is a check that the mineralogist uses to
ensure the quality of the data that is provided.
How do I analyse the data collected?
The mineralogical report provided will depend on the technique
employed by the mineralogist to examine your samples.
Essentially, the data provided should provide nominally the same
information. That is, the percentage of liberated mineral within

22 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

the sample, the percentage of binary particles present and what
the two minerals are, and the percentage of more complex
composite particles measured. Table 13 provides a simplified list
of the liberation data for galena in the lead rougher feed of the
Eureka Concentrator.
What this data means is that of the lead in the +28 micron size
fraction 66 per cent occurs as liberated particles of galena and
28 per cent are present as binary composites with sphalerite.
Minor amounts of galena are associated with pyrite and occur in
ternary particles.
What does it mean?
Mineralogical examination of strategic process streams feeding
into the lead rougher feed showed that about one third of the
galena in the lead rougher feed occurs as coarse (+20 micron)
galena/sphalerite composite particles (Figure 21), and the
majority of this material emanates from the lead first cleaner
tailing. Thus, two thirds of the galena in the lead rougher feed is
liberated, which infers that the maximum lead recovery to the
final lead concentrate, at grade, would be approximately 70 per
cent. Therefore, galena liberation was considered to be the
limiting factor impinging of improved metallurgical performance.
Problem definition summary
The metallurgical survey suggested that the lead circuit was
potentially unstable with large circulating loads of galena and
sphalerite around the lead rougher/first cleaner circuit. The use of
recovery-by-size and liberation analysis has clearly indicated that
the recovery of +20 micron galena within the lead circuit is
limited. The liberation analysis clearly shows that the galena in
the +20 micron fraction is locked with sphalerite.
Thus, the problem is: poor liberation of galena in the +20 micron
size fractions.
SOLUTION DEVELOPMENT AND TESTING
It is apparent that to improve liberation one needs to grind finer.
The first step in proving that improved regrinding is required in
the lead circuit is to complete a series of laboratory tests examining
the impact of regrinding the main feed source to the lead rougher,
the lead first cleaner tailing.
Laboratory investigation (Chapter 9)
What do I have to do?
As the lead first cleaner tailing contributed the highest flow of
material to the lead rougher feed, and this process stream contains
a significant percentage of galena/sphalerite composite particles
greater than 20 microns, it was decided that this process stream
would be the best one to complete laboratory tests examining the
impact of regrinding on lead rougher flotation. It is your job to
design a series of laboratory flotation tests that would identify
what changes would occur as the lead first cleaner tailing was
reground to progressively finer sizes.

TABLE 13
Galena liberation by size and mineral class for the lead rougher feed of the Eureka Concentrator.

Size (μm) Liberation class (%) Total (%)

Liberated Binary with … Ternary
Sphalerite Pyrite Gangue
+89 26 54 5 4 11 100
+28 66 28 3 0 2 100
+10 69 25 5 1 1 100
+6 93 6 0 0 0 100
-6 94 5 0 0 0 100
Head 66 28 3 0 2 100

100
90
80
Distribution

70
60
50
40
30 Liberated
Ga-Sp
20
Ga-Py
10
Ga-Gn Liberation class
0
Head Ternary
2.3 6.0 10.1 28.2
Size, microns 89.2

FIG 21 - Size by liberation class data for the lead rougher feed. Ga = galena; Sp = sphalerite; Py = pyrite; and Gn = gangue.

Flotation Plant Optimisation Spectrum Series 16 23

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

With the laboratory test program designed, sufficient pulp was
collected from the lead first cleaner tailing. In addition to the lead
first cleaner tailing pulp, a bulk sample of process water was also
collected. The pulp was filtered and ‘cut’ into 1000 gram lots.
The test program called for one test to be completed without
regrinding, and then tests completed on ore ground for five, ten,
15 and 20 minutes. The tests were completed in a 2.5 litre
flotation cell. Timed concentrates were collected after 0.5, 1.5,
2.0, 2.0, 2.0, 2.0 and 2.0 minutes for a total flotation time of
12 minutes. The seven concentrate and tailing samples generated
for each test were prepped and submitted for assay.
What data do I need?
The weights for each of the concentrates and the tailing samples
as well as the assays are required to complete a mass balance for
each of the tests. Additional information such as the pH, reagent
additions and any observations made during the test are also
helpful.
How do I analyse the data collected?
The weights and assays are to be put into a spreadsheet that
allows you to calculate the cumulative grades and recoveries for
each of the tests. These data can then be used to construct lead
grade/recovery curves (Figure 22). From the mass balanced data
it is also possible to determine the kinetics (ie flotation rate
constant and maximum recovery) using Equation 11, and the
selectivity for galena against the gangue minerals.
R = Rmax × (1 − e − kt ) (11)
To assist with interpretation the kinetic data and the selectivity
data can be tabulated. For example, Table 14 shows the lead
concentrate grade, as well as the zinc and iron sulfide recoveries,
at 80 per cent lead recovery. Selectivity curves can also be drawn.
What does it mean?
Laboratory lead rougher flotation tests conducted on strategic
process streams suggested that the introduction of regrinding
significantly improved the position of the lead grade/recovery
curve. Figure 22 shows that as the degree of regrinding increased
the position of the lead grade/recovery curve improved. For
example, at 80 per cent lead recovery the lead grade was
increased from 30.8 per cent for the test conducted without
regrinding, to 42.2 per cent after 20 minutes regrinding. The
improvement in lead grade was due to better selectivity for galena
against sphalerite and iron sulfides (Table 14).
In the laboratory, regrinding improved lead metallurgy, and
with the improved selectivity for galena against sphalerite it is
expected that zinc metallurgy would also improve.
Plant trial (Chapter 10)
As the existing lead regrind mill was inadequate for this duty, a
larger mill would be required if improvements in metallurgical
performance were to be realised. In light of the very first
comment made at the beginning of this chapter about utilising the

60.0

55.0

50.0
Pb grade, %

45.0

40.0

35.0

30.0

25.0
10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Pb recovery, %

No regrinding 5 minutes regrinding 10 minutes regrinding

15 minutes regrinding 20 minutes regrinding

FIG 22 - Lead grade/recovery curves for lead rougher flotation tests conducted on lead first cleaner tailing, demonstrating the effect
of regrinding.

TABLE 14
Lead grade and diluent recoveries, at 80 per cent lead recovery, for lead rougher flotation tests conducted on lead first cleaner tailing
demonstrating the impact of regrinding.

Test description Pb grade (%) Diluent recovery (%)

Zn IS
No regrinding 30.8 71.2 51.6
5 minutes regrinding 34.3 61.9 37.2
10 minutes regrinding 37.6 50.8 30.4
15 minutes regrinding 42.0 42.6 24.3
20 minutes regrinding 42.2 42.2 25.8

24 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

resources around you, and especially people, this problem was • the zinc grade in the lead concentrate was reduced by 1.8 ±
discussed with the operations team and a solution was quickly 0.6 per cent, with greater than 99 per cent confidence.
found. As capital was not readily available an experienced
These changes to the lead circuit resulted in a decrease in the
flotation operator suggested swapping the lead and zinc
amount of sphalerite reporting to the lead concentrate, and shifted
regrinding mills, as the zinc regrinding mill was larger and more
this material into the zinc circuit. Hence, zinc metallurgy was also
able to cope with the higher tonnages experienced in the lead improved, with:
circuit.
Redirecting the lead scavenger concentrate and lead first • the zinc recovery increased by 1.5 ± 0.4 per cent, with
cleaner tailing process streams to the zinc regrind mill was greater than 99 per cent confidence; and
accomplished through a simple change in the pipe work. The • the zinc concentrate grade increased by 2.1 ± 0.5 per cent,
existing pipe work to the old lead regrind mill was left intact so with greater than 99 per cent confidence.
that a plant trial could be conducted. A randomised block on/off
trial was designed, following the guidelines prescribed by Tim Other statistical methods (for example, comparison of
Napier-Munn in Chapter 10: Designing and Analysing Plant regression lines) produced similar values.
Trials, that would use the shift mass balanced data to statistically
analyse the metallurgical performance of the Eureka Concentrator THE CYCLE BEGINS AGAIN
with and without effective lead regrinding.
Despite the application of effective regrinding in the lead circuit,
When the new lead regrind mill was turned on the impact on
and the massive improvements in metallurgical performance,
lead and zinc metallurgy was immediate and dramatic. Because
there is still room for improvement. There is not time to rest on
of the magnitude of the improvements the on/off trial was
your laurels, as the Senior Project Metallurgist it is your job to
abandoned. However changes of this magnitude are not always
identify problems and develop cost-effective solutions to
apparent, and strong statistical data analysis is often required as
implement in the plant. You’ve only solved part of the problem,
proof that there is a statistically confident change in the plant
and opportunities for further improvement are available. So, sorry
which was due to your implemented solution. In some cases,
but the Senior Project Metallurgist’s work is never finished! The
trials can run for a number of months but persistence pays off
key areas where efforts can be made to improve metallurgy are:
when you achieve the result you were aiming for.
The lead recovery to final concentrate from June 1998 through • the optimisation of both the lead and zinc regrinding
to February 2000 is presented in Figure 23. It is apparent that the circuits, to improve liberation and subsequently concentrate
lead recovery has increased on average from 65 per cent up to grades and recoveries;
81.5 per cent. The most significant improvement to lead recovery • the optimisation of chemistry for flotation to improve
was achieved through the application of an effective regrinding selectivity of valuable minerals against gangue minerals, and
stage. consequently increased concentrate grades; and
Despite not undertaking the randomised block trial as first • to continue minimising variations in the process.
thought, it was possible to complete a statistical analysis using
plant data from before and after the change. Using the Student
t-test to analyse the data showed that installing effective lead
More surveys
regrinding produced the following results: To move forward it is necessary to continue monitoring the
• the lead recovery was improved by 16.5 ± 2.5 per cent, with process through:
greater than 99 per cent confidence; • analysis of shift mass balances,
• the lead concentrate grade improved by 4.0 ± 0.8 per cent, • analysis of monthly composite samples (by size and
with greater than 99 per cent confidence; and liberation class),
90.0

85.0 New Pb Feed distributor

80.0
MF4 used as
lead regrind mill
Average % Rec.

75.0

70.0 Complete Pb froth level control

+ new 6" Pb regrind cyclones
+ Dedicated float operators
65.0

60.0 Initial Pb froth level control

Level on Pb conditioning tank
dropped
55.0

50.0
13-May-98

26-May-99
5-May-99
18-Nov-98

10-Nov-99
18-Feb-98

10-Feb-99

24-Mar-99
16-Sep-98

29-Sep-99
3-Mar-99

8-Sep-99

2-Feb-00
28-Jan-98

20-Jan-99

12-Jan-00
7-Jan-98

15-Jul-98

28-Jul-99
26-Aug-98

18-Aug-99
30-Dec-98

22-Dec-99
24-Jun-98

5-Aug-98

9-Dec-98

16-Jun-99

1-Dec-99
22-Apr-98

14-Apr-99

7-Jul-99
11-Mar-98

1-Apr-98

3-Jun-98

28-Oct-98

20-Oct-99
7-Oct-98

W/E

Average Average CL UCL LCL

FIG 23 - Lead recovery versus time data from June 1998 to February 2000. Some milestones are highlighted.

Flotation Plant Optimisation Spectrum Series 16 25

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT
• routine metallurgical and pulp chemical surveys,
• flotation cell characterisation (Chapter 5), and
• the introduction of a future ores testing program
(Geometallurgy, Chapter 12).
After all, to measure is to know!
Chemistry
With the improvements in the lead circuit through better
regrinding perhaps it is time to study the chemistry of the
system.
Pulp chemistry
The pulp chemistry indicated that the lead rougher circuit
operated in oxidising conditions, with Eh values between 150 and
200 mV (SHE). Figure 15 shows the Eh-pH curve for the circuit.
The curve is essentially parallel to the water-oxygen line for the
lead rougher circuit, and suggested that any changes in Eh or pH
were due to the maintenance of water equilibria. These conditions
points to a system operating in oxidising conditions.
Solution chemistry
The EDTA extractable lead increased from approximately five
per cent in the copper rougher tailing up to 17 per cent in the lead
scavenger tailing. The EDTA extractable zinc and iron values
were an order of magnitude lower than those reported for lead.
These data suggest that the galena in the lead circuit feed is
moderately oxidised, while the galena remaining in the lead
scavenger tailing is heavily oxidised. The main point to come
from this analysis is that there would appear to be a significant
quantity of galena oxidation products present within the pulp,
which under the right circumstances could lead to the inadvertent
activation of sphalerite.
Surface chemistry
To confirm your theory developed from the pulp chemistry
observations it is wise to complete surface analysis on selected
samples. Alan Buckely, in Chapter 8: Surface Chemical
Characterisation for Identifying and Solving Problems Within
Base Metal Sulfide Flotation Plants, provides details of the types
of surface analysis and how they may be used in your quest for
more information.
What do I have to do?
With the decision made to complete surface analysis to validate
your hypothesis you need to provide the institution conducting
the surface analysis with a scope of work. It is imperative that
you provide the surface scientist with a focused question that
defines what you hope to get out of the surface analysis. In your
case, at the Eureka Mine, you are very interested in knowing the
surface chemistry of liberated sphalerite that reports to the lead
rougher concentrate, and how this compares with free sphalerite
in the lead rougher tailing. This essentially means that the
surface scientist will examine liberated sphalerite in lead
rougher concentrate and tailing process streams to identify
differences that may explain why some of the liberated
sphalerite is being recovered into the rougher concentrate. The
surface scientist will suggest the surface analysis technique that
they feel is best for your application. ToF-SIMS is frequently
used because of its high surface sensitivity and its ability to
analyse individual particles.
26 Spectrum Series 16
What data do I need?
When collecting the samples for surface analysis it is important to
use the right sample preparation protocols. The surface scientist
will provide you with these. Generally, once the samples have
been collected they need to be purged with nitrogen gas to
remove oxygen from the pulp to prevent further oxidation. The
sample is then sealed in the sample vial, and snap frozen in liquid
nitrogen. The sample pairs are then shipped to the surface
analysis facility frozen in a cryogenic container.
Ideally, when you collect the samples for surface analysis you
should complete a block survey of that part of the circuit of
interest. At Eureka, a block survey of the lead rougher circuit
would be the order of the day. Conducting a survey gives you a
reference point in terms of metallurgical performance that can be
used for comparison purposes. It is also wise to complete a pulp
chemical and EDTA extractable metal ion analysis on the sample
process streams that the surface analysis samples were collected
from. This information should be transmitted to the surface
scientist as it gives them some idea of the chemistry of the system
and helps explain the observations they will make.
How do I analyse the data collected?
The surface scientist will provide you with an analysis of the
findings. You are not expected to know everything! However, it is
wise to read the surface analysis report and ask questions. If you
do not understand the analysis of the surface chemical data ask
the surface scientist to explain. This will allow you to question
the surface scientist to determine how they reached the
conclusions and help you to understand these techniques, and the
valuable information they can add to an investigation. It is very
important to link the surface chemical observations to the pulp
chemistry and metallurgical performance of the plant. After all
the data generated will be used to confirm your theory as to why
something is happening (at Eureka identification of the activating
species on liberated sphalerite), and from this you will propose
solutions for testing in the laboratory to solve the problem.
What does it mean?
Surface analysis was completed on samples of lead rougher
concentrate and tailing samples using ToF-SIMS to determine the
dominant species on the surfaces of the liberated sphalerite
particles. The surface analysis investigation indicated that the
liberated sphalerite particles contained within the lead rougher
concentrate had statistically more collector, copper, silver and
lead species on their surfaces than those reporting to the tailing
(Figures 24 and 25).
The conclusions reached from the chemical analysis were that:
• The flotation pulp in the lead rougher circuit operated in
oxidising conditions.
• The level of EDTA extractable lead within the lead circuit
was high, indicating the presence of a large quantity of
galena oxidation species capable of activating sphalerite.
• The surface analysis confirmed the suspicion that the
liberated sphalerite was activated by lead ions. Copper and
silver were also observed on the surfaces of the sphalerite
particles, but in lower concentrations.
These data would suggest that some of the sphalerite is being
recovered into the lead concentrate because it is activated by lead
ions. You should propose that laboratory test work be completed
examining the effect of various reagents to depress activated
sphalerite. Should these tests prove positive, and the solution is
economically sound, then it should be tested in the plant using
statistically rigorous trial methodology.
Flotation Plant Optimisation
 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

(+) SIMS sphalerite

Normalised Intensity
0.3
0.25 Pb cleaner con
0.2 Pb cleaner tail
0.15
0.1
0.05
0

Si
g
Ag

Pb
Fe

Zn

H
a

H
M

O
O

Pb
Fe
(+) Fragment

(-) SIMS sphalerite

Normalised Intensity

0.5
Pb cleaner con
0.4
Pb cleaner tail
0.3
0.2
0.1
0
C CH O OH S SO3 SO4 IEX
(-) Fragment

FIG 24 - Positive and negative mass spectra for sphalerite particles in the lead cleaner concentrate and tailing process streams – confidence
intervals calculated for 95 per cent.

(A) (B)

FIG 25 - ToF-SIMS images of sphalerite particles within the lead rougher concentrate: (A) the zinc ion image; and (B) the lead ion image.

COMMUNICATION
The technical aspects of concentrator performance are very
important in achieving optimum metallurgy. However, the best
technical expertise is of little consequence if people and
maintenance issues are not adequately addressed.
Process variations are probably the hardest to monitor and
control, and can result in operations personnel treating symptoms
rather than causes (ie ‘fire fighting’). In most instances process
variations, such as ore hardness, mineralogical changes with ore
type, and head grade variation, are very difficult to measure
online, but can be partially overcome with an effective ore
characterisation program (geometallurgy) and good commun-
ication between the mine and the mill.
Behavioural variations are often easily identified, but are
difficult to correct because of the human factor. One way of
improving the behavioural variations within a concentrator is to
document and standardise each task. That is, if each operator
performs the same task the same way, by following as standard
work procedure, then behavioural variations can be minimised.
Unfortunately, this is easier said than done.
Another key element in altering behaviour to maximise
metallurgical performance is the development of a proactive,
interventionist metallurgical team, such that there are definite
metallurgical targets and procedures prescribed to achieve these
goals. A change in the perception of the metallurgist is required
such that he should be viewed as a member of the team with
specialist knowledge that is on-call when problems arise.
CONCLUSIONS
At the Eureka Concentrator the use of classical metallurgical
techniques (recovery- and liberation-by size analysis) has
provided a high degree of certainty in determining the source of
the metallurgical problem. Initially galena/sphalerite liberation
was identified as a significant impediment to achieving better

Flotation Plant Optimisation Spectrum Series 16 27

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT
lead and zinc metallurgy. With the application of greater
regrinding capacity the focus shifted from liberation to chemistry.
The use of pulp and solution chemistry provided an indication
that inadvertent activation of sphalerite by lead, copper and silver
ions was probably the reason for the high recovery of liberated
sphalerite into the final lead concentrate. The judicious
application of appropriate surface chemical techniques provided
confirming evidence. The Eureka metallurgists have now
identified the reasons for the losses of sphalerite from the circuit,
and can now develop solutions to this problem for testing in the
laboratory, prior to trialling in the plant.
REFERENCES
Johnson, N W, 1988. Application of electrochemical concepts to four
sulphide flotation separations, in Proceedings Electrochemistry in
Mineral and Metal Processing II, pp 131-149.
Natarajan, K A and Iwasaki, I, 1973. Practical implications of Eh
measurements in sulphide flotation circuits, in AIME Transactions,
256:323-328.
Rumball, J A and Richmond, G D, 1996. Measurement of oxidation in a
base metal flotation circuit by selective leaching with EDTA,
International Journal of Mineral Processing, 48:1-20.
Warman International Limited, 1991. Cyclosizer Instruction Manual –
Particle Size Analysis in the Sub-Sieve Range, Bulleting WCS/2
(Warman International Limited: Sydney).
APPENDIX 1 – THE DOWN-THE-BANK SURVEY
Introduction
Often it is of value to complete metallurgical surveys were
samples are collected down a bank of flotation cells. This type of
survey is referred to as a down-the-bank survey, and differs from
a block survey, as it provides more detailed information about the
flotation behaviour of the various species contained within that
bank of flotation cells. In conducting a block survey you collect
samples of the feed, concentrate and tailing from the section of
the circuit you are interested in. This information provides you
with data about the overall performance of that block of cells.
The down-the-bank survey provides detail of the internal
workings of that block of flotation cells.
How do I do a down-the-bank survey?
Chapter 2: Existing Methods for Process Analysis, by Bill
Johnson provides a detailed description of the two common
methods employed. An example, using one of those methods is
given below. The first step in completing a down-the-bank survey
is to identify the samples that you wish to collect. Figure A1 is a
schematic of the bank of 12 flotation cells you want to survey.
You have decided to divide the flotation cells into groups such
that you will collect five concentrate samples, in addition to the
combined concentrate sample that the bank produces. The sample
list is given in Table A1.
1. T bank feed
Cell 1 Cell 2 to 4 Cell 5 to 7
3. Con 1 4. Con 2 5. Con 3
2. T bank combined concentrate
FIG A1 - A bank of 12 flotation cells, and the cell grouping for a down-the-bank survey.
28 Spectrum Series 16
TABLE A1
The sample list for a down-the-bank survey conducted on the
bank of 12 flotation cells pictured in Figure A1.
Sample Sample name Sample type
number
1 T bank feed Dip sample
2 T bank combined concentrate Lip or OSA sample
3 Con 1 – T bank cell 1 Timed lip sample
4 Con 2 – T bank cells 2 to 4 Timed lip sample
5 Con 3 – T bank cells 5 to 7 Timed lip sample
6 Con 4 – T bank cells 8 to 10 Timed lip sample
7 Con 5 – T bank cells 11 and 12 Timed lip sample
8 T bank tailing A Dip sample
9 T bank tailing B Dip sample
With the sampling points identified you need to communicate
your intentions to various people involved and prepare your
equipment. The same actions discussed above the Section: The
Metallurgical Survey, What do I have to do? apply to conducting
a down-the-bank survey. Communication and organisation are the
key to success.
In terms of sampling equipment you will need:
• dip and lip samplers that are clean and good working order;
• sufficient clean buckets, with lids, that have been tared;
• a stop watch for timing the lip sample collection; and
• a note book to record the times and any other observations
that may be of use later.
Conducting a down-the-bank survey is usually a job for two
people. One person collects the sample, and the other measures
and records the time taken to collect each of the timed lip
samples. Prior to conducting the survey it is wise to inspect the
flotation bank to be samples, clean the cell lips to ensure that the
froth flows freely, make the work area safe and free of tripping
hazards, and establish how the lip sampling is going to occur,
then perform a dry run to practice how the timed samples will be
collected. One method of sampling that works effectively is for
the sampler to yell ‘GO’ and ‘STOP’ as he starts and finishes the
sample collection. On these commands the time keeper starts and
stops the stop watch. It is imperative that the two people
collecting the timed lip samples agree on the methodology, and
work in concert. So, with all the preparations complete, the plant
operating correctly it is time to conduct the survey.
As with other surveys you would have decided before hand
how many ‘cuts’ from each process stream will be taken over a
specified sampling time. The two people detailed to conduct the
down-the-bank survey will then collect the nine samples over the
prescribed sampling period. Once the sampling is finished, the
samples are gathered together and taken to the laboratory.
8. T bank tailing
Cell 8 to 10 Cell 11 and 12
6. Con 4 7. Con 5
Flotation Plant Optimisation
 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

TABLE A2
Summary of the down-the-bank survey data collected.

No Sample name Weights (g) % solids

Gross Tare Wet Dry
Bucket Lip
1 T bank feed 1585.3 350.0 1235.3 471.9 38.2
2 T bank combined con 1891.1 350.0 245.0 1296.1 331.8 25.6
3 Con 1 – Cell 1 1789.7 350.0 245.0 1194.7 630.8 52.8
4 Con 2 – Cells 2 to 4 2217.0 350.0 245.0 1622.0 361.7 22.3
5 Con 3 – Cells 5 to 7 2049.4 350.0 245.0 1454.4 315.6 21.7
6 Con 4 – Cells 8 to 10 2142.3 350.0 245.0 1547.3 379.1 24.5
7 Con 5 – Cells 11 and 12 1795.6 350.0 245.0 1200.6 187.3 15.6
8 T bank tailing A 1428.8 350.0 1078.8 372.2 34.5
9 T bank tailing B 1610.0 350.0 1260.0 459.9 36.5

What data do I need? Other data that can be useful when analysing the plant survey
mass balance are:
In the laboratory the samples are weighed (to determine the wet
weight), filtered, dried, weighed, prepped and submitted for • the throughput at the time of the survey,
assay. A summary of this data is given in Table A2. The wet and • reagent additions and other plant operating parameters (ie
dry weights are used to determine the per cent solids of each airflow rates and pulp levels),
sample, and calculate a water balance. In terms of assays, apart
from those pertaining to the valuable minerals you are separating
• OSA readings, and
from the gangue. In example Eureka’s assayed for: silver, copper, • information about the ore being treated.
lead, zinc, iron and silica. In the context of a one off survey some of these pieces of
You will also need the times recorded for the timed lip information may not be of great value. However, when the
samples, the lip lengths of the flotation cells, and the lip sample analysis is extended to include other surveys on other ore blends,
cutter length. With these data it will be possible to calculate the circuit configurations, reagent suites, these data provide a vital
flow rate of concentrate from each of the groups of flotation cells. link in the comparison.
The data are supplied in Table A3.
How do I analyse the data collected?
TABLE A3 The assays will be returned to you in a form similar that shown in
Lip sample details. Table A4. Before proceeding with mass balancing the survey it is
necessary to check these numbers are in good order. In this
No Sample name Sample Length (mm) example, the tailing assays are in good agreement, and the silver,
time (s) Cell lip Cutter lip copper and lead assays all trend in the right way (ie from higher
3 Con 1 – Cell 1 31.6 1250 × 2 105 assays in Con 1 to progressively lower numbers by Con 5).
4 Con 2 – Cells 2 to 4 28.9 1250 × 6 105
Further, the combined concentrate assay falls within the extremes
of Con 1 and Con 5 assays.
5 Con 3 – Cells 5 to 7 28.4 1250 × 6 105
Mass balancing this survey involves a number of steps. The
6 Con 4 – Cells 8 to 10 30.8 1250 × 6 105 first step is to calculated the mass flow rate from the timed lip
7 Con 5 – Cells 11 and 12 39.7 1250 × 4 105 samples. The mass flow rate is given by:

TABLE A4
Elemental assays for the down-the-bank survey.

No Sample name Assay (%)

Ag (ppm) Cu Pb Zn Fe SiO2
1 T bank feed 98 0.05 2.84 10.7 15.5 21.9
2 T bank combined con 444 0.24 14.20 11.1 23.0 9.2
3 Con 1 – Cell 1 644 0.47 21.40 10.7 21.2 7.6
4 Con 2 – Cells 2 to 4 490 0.28 15.70 11.3 22.9 8.5
5 Con 3 – Cells 5 to 7 406 0.21 12.60 11.3 23.8 9.1
6 Con 4 – Cells 8 to 10 346 0.18 10.60 11.6 24.2 10.0
7 Con 5 – Cells 11 and 12 306 0.18 9.15 11.7 24.3 10.9
8 T bank tailing A 70 0.02 2.04 10.5 14.8 22.6
9 T bank tailing B 72 0.02 2.13 10.6 14.9 22.6

Flotation Plant Optimisation Spectrum Series 16 29

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

Dry weight of sample, g 3600 concentrate produced is 6.96 tonnes per hour, compared to 11.76
t/ h = × ×
Sample time, s 10 6 (A1)
determined from the timed lip tonnages. The difference is
Cell lip length, mm reconciled by applying the tonnage distribution from Table A5 to
number of cell lips × the mass balanced tonnage given in Table A6. Thus, using the
Cutter lip length, mm ‘outer’ circuit mass balanced tonnage new values are calculated
for each of the concentrates. These new values are presented in
Thus, by substituting the dry weights for each of the timed Table A7, and are used in the second mass balance incorporating
concentrate samples from Table A2, and the sample times, the down-the-bank concentrates.
number of cell lips, cell lip length and cutter lip length from
Table A3 in Equation (A1), you can calculate the tonnes per hour
TABLE A7
recovered into each concentrate. For example, for Con 1 – cell 1
The ‘new’ tonnes per hour for the five concentrates collected in the
Equation (A1) becomes: down-the-bank survey calculated from the tonnage distribution
630.8, g 3600 1250, mm and the ‘outer’ mass balanced tonnage recovered into the
t/ hCon 1 = × ×2× combined concentrate.
31.6, s 10 6 105, mm
No Sample name New tonnes per hour
t/ hCon 1 = 1.71 t/ h 3 Con 1 – Cell 1 1.01

So, from the timed lip sample collected from Cell 1, the tonnes 4 Con 2 – Cells 2 to 4 1.90
per hour recovered are 1.71. The same calculation was completed 5 Con 3 – Cells 5 to 7 1.69
for each of the other concentrate. The values obtained are given in 6 Con 4 – Cells 8 to 10 1.87
Table A5. The tonnage distribution over the five concentrate 7 Con 5 – Cells 11 and 12 0.48
samples was calculated, and also appear in Table A5.
Total 6.96

TABLE A5
The tonnes per hour and distribution of tonnes for the five The down-the-bank mass balance has eight process streams:
concentrates collected in the down-the-bank survey.
1. T bank feed,
No Sample name Tonnage data 2. T bank combined concentrate,
Tonnes per hour Distribution
3. Con 1 – Cell 1,
3 Con 1 – Cell 1 1.71 14.55
4 Con 2 – Cells 2 to 4 3.22 27.36
4. Con 2 – Cells 2 to 4,
5 Con 3 – Cells 5 to 7 2.86 24.30 5. Con 3 – Cells 5 to 7,
6 Con 4 – Cells 8 to 10 3.17 26.91 6. Con 4 – Cells 8 to 10,
7 Con 5 – Cells 11 and 12 0.81 6.88 7. Con 5 – Cells 11 and 12, and
Total 11.76 100.00
8. T bank tailing.
There are two nodes:
The next step is to mass balance the ‘outer’ circuit. The process 1. T bank feed = Con 1 + Con 2 + Con 3 + Con 4 + Con 5 + T
streams are: bank tailing (ie 1 = 3 + 4 + 5 + 6 + 7 + 8); and
1. T bank feed, 2. Con 1 + Con 2 + Con 3 + Con 4 + Con 5 = T bank
2. T bank combined concentrate, and combined concentrate (ie 3 + 4 + 5 + 6 + 7 = 2).

3. T bank tailing. The mass balance is accomplished by fixing the tonnage values
for the T bank feed, and the five down-the-bank concentrates,
There is only one node in this mass balance: then using a mass balancing software fitting the data. The
down-the-bank mass balance is given in Table A8.
T bank feed = T bank combined concentrate
+ T bank tailing (1 = 2 + 8). The mass balanced down-the-bank data can now be used to
construct a lead grade/recovery curve for this bank of flotation
In completing this mass balance it was assumed that the feed cells, as shown in Figure A2. This data can also be used to
tonnage was 100 per cent. The mass balanced data are provided generate kinetic data examining the flotation rate constant and
in Table A6. The mass balance has calculated that the tonnes of maximum recovery of the various species.

TABLE A6
The mass balanced data for the ‘outer’ balance.

No Sample time Wt (%) Adjusted assay (%)

Ag (ppm) Cu Pb Zn Fe SiO2
1 T bank feed 100.00 98 0.04 2.91 10.67 15.48 21.78
2 T bank combined con 6.96 444 0.24 14.20 11.10 23.00 9.23
3 T bank tailing 93.04 72 0.03 2.07 10.63 14.92 22.72

30 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

TABLE A8
The down-the-bank mass balance.

No Sample name Wt Weight (%) Recovery (%)

(%) Ag Cu Pb Zn Fe SiO2 Ag Cu Pb Zn Fe SiO2
1 T bank feed 100.00 98 0.04 2.9 10.6 15.5 21.8 100.0 100.0 100.0 100.0 100.0 100.0
2 T bank combined con 7.00 440 0.26 14.0 11.3 23.3 9.1 31.5 42.0 33.6 7.4 10.5 2.9
3 Con 1 – Cell 1 1.02 644 0.47 21.4 10.7 21.2 7.6 6.7 11.2 7.5 1.0 1.4 0.4
4 Con 2 – Cells 2 to 4 1.91 490 0.28 15.7 11.3 22.9 8.5 9.6 12.5 10.3 2.0 2.8 0.7
5 Con 3 – Cells 5 to 7 1.70 406 0.21 12.6 11.3 23.8 9.1 7.0 8.3 7.4 1.8 2.6 0.7
6 Con 4 – Cells 8 to 10 1.88 346 0.18 10.6 11.6 24.2 10.0 6.6 7.9 6.9 2.0 2.9 0.9
7 Con 5 – Cells 11 and 12 0.49 306 0.18 9.2 11.7 24.3 10.9 1.6 2.1 1.5 0.6 0.8 0.2
8 T bank tailing 93.00 72 0.03 2.1 10.6 14.9 22.7 68.5 58.0 66.4 92.6 89.5 97.1

25.0 Mineral conversions

Each conversion is based on the atomic mass of the elemental
20.0
components of the mineral.
Pb grade, %

15.0
Galena
10.0 Galena contains lead (207.2 amu) and sulfur (32.06 amu),
therefore the atomic mass of galena (PbS) is:
5.0
PbS = Pb + S
0.0 PbS = 207.2 + 32.06 amu
0.0 10.0 20.0 30.0 40.0 50.0 60.0
Pb recovery, % PbS = 239.26 amu
OUTER balance data Down-the-bank balanced data
Thus, ‘pure’ galena contains 86.6 per cent lead and 13.4 per
cent sulfur. The conversion factor (fGa) to convert the lead assay
FIG A2 - The lead grade/recovery curve constructed from the to galena is given by:
mass balanced down-the-bank survey data.
PbS amu
fGa =
Pb amu
239.26
APPENDIX 2 – ESTIMATED MINERAL ASSAYS fGa =
207.2
FROM ELEMENTAL DATA
fGa = 1155
.
Introduction So, an assay of ten per cent lead is equivalent to 11.6 per cent
As iron occurs in a variety of minerals present in the Eureka galena.
orebodies’ interpretation of iron deportment from elemental assay The sulfur in galena conversion factor (fSGa) is given by:
data from flotation tests and plant surveys is a complex issue.
% sulfur in PbS
Therefore, by making a few simple assumptions regarding the f SGa =
composition of the dominant sulfide minerals present in the % lead in PbS
orebody, it is possible to estimate mineral assays from the 13.4
elemental assay data. f SGa =
86.6
Assumptions f SGa = 0155
.
The first set of assumptions relate to the minerals themselves.
That is, the dominant sulfide minerals are: galena; sphalerite, Sphalerite
chalcopyrite and pyrite. The lead occurs as galena; the zinc as ‘Pure’ sphalerite (contains no iron) is made up of zinc
sphalerite; the copper as chalcopyrite; and the iron occurs in (65.38 amu) and sulfur (32.06 amu), therefore the atomic mass of
sphalerite, chalcopyrite and pyrite. ‘pure’ sphalerite (ZnS) is:
The mineral conversions are based on the assumption that each
mineral is ‘pure’, for example galena is PbS, pyrite is FeS2, etc. ZnS = Zn + S
Eureka sphalerites are known to contain, in solid solution, ZnS = 65.38 + 32.06 amu
moderate iron levels. In this exercise 3.0 per cent was chosen.
This value is an estimate of the average value. ZnS = 97.44 amu
A calculated assay of the non-sulfide gangue (NSG) is made by Thus, ‘pure’ sphalerite contains 67.1 per cent zinc and 32.9 per
assuming that everything that is not galena, sphalerite, cent sulfur. Unfortunately, Eureka sphalerite contains an average
chalcopyrite, or pyrite is non-sulfide gangue. of 3.0 per cent iron in solid solution. Further, it is assumed that

Flotation Plant Optimisation Spectrum Series 16 31

CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

the iron replaced zinc in the sphalerite lattice. Thus, the 183.52
fCh =
composition of Eureka sphalerite is 64.1 per cent zinc, 3.0 per 63.55
cent iron and 32.9 per cent sulfur. Hence, the conversion factor
(fSp) to convert the zinc assay to sphalerite is given by: fCh = 2.888

100.00 So, an assay of ten per cent copper is equivalent to 28.9 per
f Sp = cent chalcopyrite. To determine the amount of iron present in
% Zn in sphalerite
chalcopyrite it is simply a matter of multiplying the copper assay
100.00 by the ratio of the per cent iron in the chalcopyrite and the per
f Sp = cent copper in the chalcopyrite. So, the iron in chalcopyrite
641.
conversion factor (fFeCh) is given by:
f Sp = 1.560
% iron in chalcopyrite
fFeCh =
So, an assay of ten per cent zinc is equivalent to 15.6 per cent % copper in chalcopyrite
sphalerite. To determine the amount of iron present in sphalerite
it is simply a matter of multiplying the zinc assay by the ratio of 30.43
fFeCh =
the per cent iron in the sphalerite and the per cent zinc in the 34.63
sphalerite. So, the iron in the sphalerite conversion factor (fFeSp) is
given by: fFeCh = 0.879

% iron in sphalerite Thus, for a sample containing ten per cent copper the per cent
fFeSp = iron associated with the chalcopyrite is 8.8 per cent (ie 10 ×
% zinc in sphalerite
0.879). To check this, convert the copper assay to chalcopyrite (ie
3.0 10 × 2.888), and multiply this by the amount of iron in solid
fFeSp = solution in chalcopyrite (ie 30.43 per cent), and the per cent iron
641
.
associated with chalcopyrite is 8.8 per cent.
fFeSp = 0.047 Similarly, the sulfur in chalcopyrite conversion factor (fSCh) is
given by:
Thus, for a sample containing ten per cent zinc, the per cent
iron associated with the sphalerite is 0.47 per cent (ie 10 × 0.047). % sulfur in chalcopyrite
f SCh =
To check this, convert the zinc assay to per cent sphalerite % copper in chalcopyrite
(ie 10 × 1.560), and multiply by the amount of iron in solid
solution in sphalerite (ie 3.0 per cent), and the per cent iron 34.94
f SCh =
associated with sphalerite is 0.47 per cent. 34.63
Similarly, the sulfur in the sphalerite conversion factor (fSSp) is
f SCh = 1.009
given by:
% sulfur in sphalerite Thus, for a sample containing ten per cent copper the per cent
f SSp = sulfur associated with the chalcopyrite is 10.1 per cent (ie 10 ×
% zinc in sphalerite 1.009). To check this, convert the copper assay to chalcopyrite (ie
32.9 10 × 2.888), and multiply this by the amount of sulfur in solid
f SSp = solution in chalcopyrite (ie 34.94 per cent), and the per cent
641
.
sulfur associated with chalcopyrite is 10.1 per cent.
f SSp = 0.513
Pyrite
Thus, for a sample containing ten per cent zinc, the per cent
sulfur associated with the sphalerite is 5.13 per cent (ie 10 × Assuming that pyrite is the dominant iron sulfide mineral present
0.513). To check this, convert the zinc assay to per cent sphalerite (ie no pyrrhotite) it is relatively easy to calculate the per cent
(ie 10 × 1.560), and multiply by the amount of sulfur in sphalerite pyrite in a sample. Pyrite contains iron (55.85 amu) and sulfur
(ie 32.9 per cent), and the per cent sulfur associated with (32.06 amu), therefore the atomic mass of pyrite (FeS2) is:
sphalerite is 5.13 per cent.
FeS2 = Fe + S
Chalcopyrite FeS2 = 55.85 + (2 x 32.06) amu
Chalcopyrite is made up of copper (63.55 amu), iron ( 55.85 amu) FeS2 = 119.97 amu
and sulfur (32.06 amu), therefore the atomic mass of chalcopyrite
Thus, ‘pure’ pyrite contains 46.6 per cent iron and 53.4 per cent
(CuFeS2) is:
sulfur. Depending on the assay data the pyrite content of the ore
CuFeS2 = Cu + Fe + S can be calculated using either the iron or the sulfur assay.
Generally speaking, the estimation of the pyrite concentration is
CuFeS2 = 63.55 + 55.85 + (2 x 32.06) amu more accurate using the sulfur assay because it is assumed that
CuFeS2 = 183.52 amu the vast majority of the sulfur in the ore is associated with the
sulfide minerals. Iron, on the other hand, can be present in the
Thus, stoichiometric chalcopyrite contains 34.63 per cent sulfides as well as the non-sulfide gangue, for example as iron
copper, 30.43 per cent iron and 34.94 per cent sulfur. Hence, the oxides such as haematite or magnetite, and feldspars to name but
conversion factor (fCh) to convert the copper assay to chalcopyrite a few. Thus, the pyrite conversion factor (fFePy) based on the iron
is given by: assay is given by:
CuFeS 2 amu FeS 2 amu
fCh = fFePy =
Cu amu Fe amu

32 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 1 – THE EUREKA MINE – AN EXAMPLE OF HOW TO IDENTIFY AND SOLVE PROBLEMS IN A FLOTATION PLANT

119.97 However, if the pyrite conversion factor (fSPy) based on the

fFePy =
5585
. sulfur assay then:
fFePy = 2148
. f SPy = 1871
. × (% S − (% sulfur in galena +

So, based on the conversion factor developed from the iron
assay, an assay of ten per cent iron is equivalent to 21.5 per cent
pyrite.
Using sulfur, the pyrite conversion factor (fSPy) is given by:
FeS 2 amu
f SPy =
S 2 amu
119.97
f SPy = (% S − ( 0155
6412
.
f SPy = 1871
. Thus, if an ore sample contained ten per cent sulfur, ten per
So, based on the conversion factor developed from the sulfur
assay, an assay of ten per cent iron is equivalent to 18.7 per cent
pyrite.
Unfortunately, iron is also associated with sphalerite and
chalcopyrite in the ore, so these contributions must be subtracted
from the iron assay before determining the pyrite content. Thus,
the pyrite conversion factor (fFePy) based on the iron assay
becomes:
fFePy = 2148
. ×
(% Fe − (% iron in sphalerite + % iron in calcopyrite))
From the discussion above the conversion factor for iron in
sphalerite from the zinc assay was determined to be 0.0.47, and a
similar calculation revealed that the conversion factor for iron in
chalcopyrite is 0.879. So, the per cent pyrite is given by:
fFePy = 2148
. ×
(% Fe − ( 0.047 × % Zn + 0.879 × %Cu))
Thus, if an ore sample contained ten per cent iron, ten per cent
zinc, and 0.1 per cent copper, the per cent pyrite is 20.28 per cent.
% sulfur in sphalerite + % sulfur in chalcopyrite ))
From the discussion above the conversion factor for sulfur in
galena from the lead assay is 0.155; sulfur in sphalerite from the
zinc assay was determined to be 0.513, and a similar calculation
revealed that the conversion factor for sulfur in chalcopyrite is
1.009. So, the per cent pyrite is given by:
f SPy = 1871
. ×
. × % Pb + 0.513 × % Zn + 1.009 × %Cu))
cent lead, ten per cent zinc, and 0.1 per cent copper, the per cent
pyrite is 6.0 per cent.
Non-sulfide gangue
If it is assumed that galena, sphalerite, chalcopyrite, and pyrite
are the only significant sulfide minerals present in the orebody it
is possible to estimate the per centage of non-sulfide gangue
present. This is achieved by subtracting the per cent galena,
sphalerite, chalcopyrite, and pyrrhotite from 100 per cent. That is:
% NSG = 100 – (%Ga + %Sp + %Ch + %IS)
Thus, if a sample assayed ten per cent lead, zinc and iron, and
0.1 per cent copper, the per cent non-sulfide gangue is 57.8 per
cent.
Final comment
These element to mineral conversions have been completed based
on a lead/zinc ore, however a similar approach can be applied to
other ore types provided the stoichiometry of the minerals in the
system is known, and the assumptions made in the calculations
are clearly explained.

Flotation Plant Optimisation Spectrum Series 16 33

 HOME

CHAPTER 2

Existing Methods for Process Analysis

Bill Johnson

FAusIMM(CP), Senior Principal Consulting Engineer, Mineralurgy, Suite 2, Level 1, 42 Morrow Street,
Taringa Qld 4068. Email: johnson10@bigpond.com

Bill Johnson obtained a PhD in mineral processing from the University of Queensland in 1972. After working
for ASARCO in Arizona until 1976, he lectured at the University of Melbourne. He joined the CSIRO Division
of Mineral Engineering (1978 - 1982) where research on the Lead-Zinc Concentrator at Mount Isa Mines
Limited was his main project. In 1982, he moved to Mount Isa where he continued research on the
Lead-Zinc Concentrator and other plants and ores. He was the Minerals Processing Research Manager
(1989 - 1997). Development of a circuit for McArthur River ore recommenced in 1989 under his direction,
one key outcome being the IsaMill technology. He was Professor of Minerals Engineering at the University
of Queensland (1998 - 2005) and is presently Senior Principal Consulting Engineer at Mineralurgy.

Introduction
Solid Balance and Water Balance Calculation by Surveying
Calculation of Mineral Recoveries
Calculation of Recovery-Size Data
Interpretation of Recovery-Size Curves
Addition of Liberation Data to Recovery-Size Data
The Role of Surface and Solution Analyses for Organic and Inorganic Species
Inclusion of Froth Recovery in Surveys
Process Analysis with Down-the-Bank Flotation Data
Use of Summary Graphs with Limits Imposed by Liberation
General Outcomes of the Analysis – Magnitudes of Significant Process Weaknesses and
Solutions to the Weaknesses
References
Appendix
References

ABSTRACT
This introductory chapter describes options for surveying flotation banks
and calculation of their solid and water balances, including mineral
recoveries. The calculation and interpretation of mineral recovery - size
data and the more advanced mineral recovery - size - liberation data are
then covered. The related role of surface analysis for elucidating
unexplained process weaknesses in the mineral recovery - size -
liberation data is described. The chapter is aimed at recognition of the
location and magnitude of process weaknesses in industrial plants and
identification of the mechanism involved in each, allowing technical
solutions for some or all weaknesses to be proposed and evaluated. Later
chapters expand on topics in this chapter.
INTRODUCTION
The first step in analysis of the flotation process is to obtain a mass
or solid balance, which may be accompanied by a water balance,
from surveying of the relevant portions of the plant. The key
benefit of obtaining the corresponding water balance is that water
flow rates, and hence pulp flow rates, can be calculated and hence
residence times can be calculated, ie kinetic data can be obtained.
Decisions have to be made in the design of an experiment or
survey on the level of sampling detail which is appropriate for
each portion of the plant. In general, the following options for
sampling detail exist and different levels of detail may be selected
for various parts of the plant:

Flotation Plant Optimisation Spectrum Series 16 35

CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
• block level, eg grouping of all cleaner banks or rougher
banks;
• bank level, eg obtaining of information on feed, combined
concentrate and tailing assays; or
• down-the-bank data (where data on each concentrate or
grouping of concentrates are obtained).
Of course, the whole plant may be treated as a block and this
treatment is applied commonly for obtaining shift or daily
composite samples from automatic inventory samplers from
which shift or daily metallurgical performance is often calculated
routinely. Such data can also be complementary to surveys on the
banks or blocks in the circuit. In normal metallurgical accounting,
daily samples from inventory samplers are often used to prepare
monthly composite samples for the feed to and products from a
concentrator.
Samples from inventory samplers on all the inputs to and
products from a plant can be processed at various levels of detail,
to establish causes of misplaced valuable mineral and causes of
dilution of concentrates by unwanted minerals. In broad terms,
these are at the recovery-size level for each of the minerals, and at
the next level which can be summarised as the recovery-
size-liberation level. Analysis of such data allows definition of
the weaknesses (and strengths) of the plant.
While such data from inventory samples are valuable, the
overall plant data do not provide clues to the location of the
weaknesses in the plant, eg rougher section, cleaner section or
retreatment section. Hence, surveys of the whole plant or of
individual sections at appropriate times are needed to identify the
origin of the weakness and to check progress during later
corrective steps. Judgement is required in the frequency and
design of such surveys.
Surveys of the sections in a pilot plant circuit can be conducted
by similar methods as for concentrators with some modification
of the scale of sampling equipment. Such surveys can serve two
purposes. One is to record the behaviour of an ore source or ore
type in a new orebody for later comparison with the behaviour in
a full-scale plant. The availability of detailed data within the pilot
plant circuit and at a sized level for minerals (at least) will greatly
improve the ability to perform troubleshooting in the later
start-up of a new concentrator. A second purpose can be to
perform a cycle of recognition of process weaknesses with later
correction in a pilot plant scale development of a new flow sheet.
As far as possible, such process development ought to be
conducted at the laboratory scale where the development costs
are much lower. Nevertheless, the methodology can be applied to
pilot plants along the lines discussed earlier for full-scale plants.
Samples which are produced from open circuit laboratory flow
sheets or from closed circuit laboratory cycle tests can also be
subjected to analysis at the mineral recovery-size-liberation level.
The samples are easier to obtain in this laboratory situation but
care is required to ensure enough sample exists for the envisaged
procedure. For closed circuit cycle tests, the circuit must be at
equilibrium for those cycles from which the samples are taken.
Paired surveys are sometimes employed to provide insights.
These surveys should be conducted with the same feed to plant or
pilot plant scale processes. For plants in particular, the two
surveys would normally be conducted within one day. For one of
the surveys, a major change in conditions would be implemented
and the effect of the change would be evaluated by comparison
with the ‘reference’ data under normal conditions. Such paired
tests can be valuable when surface analysis is included in the
design for the experiment as convenient ‘reference’ surface
analysis data are provided from the survey under normal
conditions.
36 Spectrum Series 16
In past analysis of recovery-size data for minerals, there has
been the convenient assumption that the solid in the pulp was
fully dispersed. In other words, it has been assumed that any
mineral reporting in a given size fraction in the recovery-size data
existed in that fraction in the real pulp. While this is likely to be a
sound assumption for most pulps, it is expected that there are
pulps for which the lack of dispersion is sufficient to transfer
minerals from one size fraction to another during the sizing
procedure due to shearing apart of aggregated particles. In a plant
pulp, a liberated valuable particle which adheres to other gangue
minerals by some mechanism will usually be less recoverable.
Further, if the liberated valuable particle is recovered, the grade
of the concentrate will be detrimentally affected due to the
adhering gangue mineral.
For convenience, hand-held samplers will be considered in
three categories: conventional hand-held or manual samplers for
sampling streams at weirs or leaving launders or pipes, dip
samplers for sampling the pulp in the pulp zone of a flotation cell,
and lip sampling for providing a sample of the concentrate from
one or more flotation machines where this usually also involves
obtaining a value for the flow rate of solid by a timing procedure.
Manual samplers may be used in conjunction with well formed
sampling points in automatic inventory samplers in special
circumstances.
There have been major changes in the type and scale of cells
employed in the last two decades of the 20th century. The
introduction of flotation cells with washed froth technology has
changed the nature of sampling problems. Further, the installation
of very large unit cells has increased sampling difficulties in
many new plants due to lack of access to the large machines and
decisions during the plant layout phase of design which tend to
increase sampling difficulties.
SOLID BALANCE AND WATER BALANCE
CALCULATION BY SURVEYING
Solid balance
A solid balance is calculated after carefully planned data
collection in a plant. A water balance can also be calculated from
the same survey provided the relevant per cent solid data are
collected. The complete procedure is relatively long and involves
many steps, requiring careful planning, a plant at steady state
during the survey, careful checking of the results and the
execution of the necessary metallurgical calculations. The steps
are summarised in Figure 1.
In contrast, a solid balance is easily obtained from a laboratory
batch testing procedure from the readily obtained weights of dry
solid in the various concentrates and tailings, along with their
assay values.
There are a limited number of methods for collection of plant
data from a bank that enables the calculation of a solid balance
and related metallurgical information.
To provide clarity on the key aspect of obtaining a solid
balance, it is valuable to state what the generation of a solid
balance does not involve. It does not involve obtaining estimates
of the flow rates of solid in the relevant streams by independent
and quite different means, and assembling these values from
disparate sources. In contrast, it does involve using principally or
solely the assay values for the relevant streams and a known
reliable tonnage (eg from a weightometer) as the scaling factor.
A survey of a total plant (many banks) or part of a plant (one
bank or a limited number of banks) is obtained by matching the
limited number of methods to the type of equipment in each part of
the plant and the availability of sampling points, and from
consideration of the purpose of the survey, ie the type of
metallurgical data required from the survey. The limited number of
Flotation Plant Optimisation

1. Establish objectives, ie type of survey, level of detail, etc
2. Plan Survey
• Determine number of people/skill levels
• Prepare all sampling equipment that match tasks
• Review plans with sampling staff
• Check condition of sampling points
• Practise at sampling points
3. Ensure circuit at steady state and fixed ore source to continue
• Steady inputs
• Steady circulating loads
• Steady assays from on-line systems
4. Execute survey (1 to 3 hours)
• Continue to monitor that circuit is at steady state (see 3)
5. Filter/dry products, perform sample preparation and extract
samples for analysis
6. Construct initial mass balance and review the quality of the
survey data
7. Move to next steps, if warranted, for more detailed data, eg
sizing of samples and assaying of size fractions, obtaining
liberation data on size fractions
FIG 1 - Summary of steps in planning and execution of a survey.
methods will be described in following sections. The methods
represent the building blocks from which surveys are designed and
conducted.
Water balance
The water flow rate in a stream can be calculated from the value
for the solid flow rate which has been determined as part of the
solid balance. The following relationship holds:
Flow rate of water =
Flow rate of solid (100 − per cent solids by weight)
Per cent solids by weight
The per cent solid value (by weight) for flotation concentrates
requires some discussion for the concentrate stream. The value of
interest for analysis of flotation data is the per cent solid value for
the concentrate as it traverses the cell lip. The value for the per
cent solid in the concentrate should not reflect the water which
has been added to the cell launder, the pump box or the gland seal
water for the concentrate pump. Taking the concentrate sample at
the cell lip provides the required per cent solid value.
Similarly, in laboratory batch tests, the water in the concentrate
(for process analysis purposes) should not include water which is
added to wash the cell lip or the scraper during the collection of a
given concentrate. This can be arranged by adding such additional
water from a wash bottle for each concentrate and weighing the
wash bottle before and after each concentrate is gathered.
Description of some basic designs for data
acquisition
Three traditional designs for acquisition of metallurgical data are
depicted in Figure 2. These designs are presented as an
Flotation Plant Optimisation
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
introduction, noting that knowledge of these designs can assist in
determining the design of a survey for acquiring metallurgical data
in more complicated situations. These designs can be summarised
as listed:
• two product (samples of feed, one concentrate and single
tailing);
• three product (samples of feed, two concentrates and single
tailing); or
• four product (samples of feed, three concentrates and single
tailing).
Conceptually, the five product and ‘higher’ cases could be used
but these cases are almost without exception impractical. In fact,
the three and four product cases can only be utilised under
appropriate circumstances which are now discussed.
To apply the two product design in Figure 2, the following
form of the two product equation is used to calculate the
fractional recovery of solid (ie the solid split):
α ( f − t)
Solid recovery ( fractional ) = = (2)
F ( c − t)
The concentration of at least one mineral (eg chalcopyrite or
the copper assay) in the feed needs to be significantly different
from the value for the tailing, ie there must be a clear difference
between feed and tailing assays in the numerator, otherwise f - t
will be a very low value (and possibly negative), and very
susceptible to considerable error from sampling and assaying.
Therefore, the calculated fractional solid recovery will be prone
to unacceptable errors. One contributor to the overall sampling
error, the fundamental error, is very dependent on the sample
mass and this issue is described in the appendix.
This separation of the mineral in question would then normally
result in a significant difference for the denominator (c - t). The
assay values (feed, concentrate and tailing) for the key element
(eg copper) can then be used as shown in Figure 2 to calculate the
solid recovery (α/F) or the absolute value for α if F is known. In
addition to a large separation occurring for the element used in
the calculation, it is desirable that the selected element can be
assayed accurately and that sampling errors are not unusually
high. Low-grade samples containing some free gold can
introduce large sampling errors.
To apply the three product case, the concentration of at least
(1)
one mineral in the feed would need to be significantly elevated in
the first concentrate while the concentration of a second mineral
would need to be significantly elevated in the second concentrate.
A typical example is the separation of galena (PbS) and sphalerite
(ZnS) into two separate concentrates and the two equations in the
three product case in Figure 2 are solved by use of lead and zinc
assays to obtain the solid balance.
The four product case can be thought of as an extension of the
three product case. However, three different minerals require
concentrating into their respective concentrates. An example is
the concentration of chalcopyrite, galena and sphalerite into their
respective concentrates and the use of copper, lead and zinc
assays to solve the three equations in the four product case, to
obtain the solid balance.
The three and four product cases can usually only be applied to
the feed and products of a concentrator (ie inventory sampler
products and shift or daily performance calculations) of a
concentrator with multiple single mineral concentrates, as the
special conditions for their successful application exist for the
products from such concentrators and do not usually exist along
individual banks. Therefore, the three and four product cases
and ‘higher’ cases can usually be expected to play no role or
only a supporting role in designing a survey for the banks in a
Spectrum Series 16 37
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

Definitions
Let f, c and t be the assays for a particular element in the feed, concentrate and tailings
Let F = t/h of solid in feed (known)
Let α, β and γ = t/h of solid in concentrate(s)

Solid and Element Balances for Three Cases

TWO PRODUCTS THREE PRODUCTS FOUR PRODUCTS
(one equation with one (two equations with two (three equations with three
unknown α) unknowns α and β) unknowns α, β and γ)
F F–α F F – α– β F F – α– β – γ

α α β α β γ
Element 1 (Major separation occurring)

fF = cα + t(F – α) fF = c1 α + c2 β + t(F – α – β) fF = c1 α + c2 β + c3 γ + t(F – α – β – γ)

Element 2 (Major separation occurring also for the second element)

Not needed to solve for α. f F = c1 α + c 2 β + t (F – α – β)

l 1 1 1
f1 F = c1 1 α + c21 β + c3 1 γ + t 1(F – α – β – γ)

Element 3 (Major separation occurring also for the third element)

Not needed to solve for α. Not needed to solve for α and β. f11 F = c1 11α + c2 11β + c311 γ + t 11(F – α – β – γ)

Solving for single unknown, α. Solve for α and β using two Solve for α, β and γ using three
fF = cα + tF – tα simultaneous equations. simultaneous equations.
α f -t
∴ =
F c- t
f -t
or ∴α = F
c- t

FIG 2 - Three traditional designs for acquisition of metallurgical data, including the commonly used two product formula.

concentrator. In contrast, the two product case has often an
important role in data acquisition from individual banks within a
flotation plant.
Description of some designs for data acquisition
from individual banks
A traditional approach to data acquisition for a bank of small
flotation machines is depicted in Figure 3. Provision is made in
Figure 3 for intermixing or exchange of pulp between adjacent
pulp zones. This situation existed in the layout of many
commercial flotation machines which were bolted together to
form banks and which were relatively small in volume, eg less
than 10 to 20 m3 per impeller. Often, these banks contained ten
to 20 flotation machines in series in order to provide the
required residence time for relatively large tonnages of solid and
pulp, given the relatively small flotation machines.
For some banks in a plant, the objective of the survey may be
to obtain the overall performance for the bank (and not down-the-
bank data). However, because no sampling point exists for the
combined concentrate from the bank, a more detailed design for
data collection (obtaining down-the-bank data) may be required,
making use of access to the concentrate discharge lips for
example (to compensate for the absence of a sampling point for
the combined bank concentrate).
The approach depicted in Figure 3 represents repeated
application of the two product equation to the data for each
machine to obtain the solid balance. It must be recognised that
the method is based on some simplifying assumptions:
• the pulp zone of each machine is perfectly mixed (to allow
ready and reliable sampling via a dip sampler), and
• the effect of the intermixing flow rates (M) is disregarded in
the mass balance for each machine (see Figure 3).
There are some other properties of this type of sampling:
• A relatively large number of samples are required (compared
to some other options) as a pulp sample from each cell is
required.
• Experimental error in sampling, preparation and assaying of
the series of pulp samples can result in irregularities in the
data particularly towards the end of the bank where the
additional recovery in each cell is low. For example, the
recorded assay for the valuable element in the pulp region of

38 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

Individual concentrates from cells

f X Feed sampling point

c1, c2 … cn • Sampling points for concentrate from each cell
p1, p2 … p n ° Sampling points for pulp region in each cell
t Sampling point for bank tailing after last cell

Note: For the last cell in the bank, the tailing sample can be taken from the tailing stream
°
( ) or from the pulp in the last cell ( ) (if the mixing state is known for the last cell).

M = exchange of pulp due to intermixing between adjacent cells (zero if cells

are isolated) in units of M t/h of solid

Assays for calculating of solid balance for each cell using two product equation (ie
ignoring the solid flows (M) from intermixing):

Feed Assay Concentrate Assay Tailing Assay

Cell 1 f c1 p1
Cell 2 p1 c2 p2

Cell n pn-1 cn pn

For the first cell in the bank:

Let C = t/h of solid in concentrate
and F = t/h of solid in feed
and T = t/h of solid in tailing

Stating the two product equation and inserting the symbols for cell 1:
C f -t
Solid Recovery * = =
F c-t
C f - p1
=
F c1 - p 1
f - p1
C =F×
c1 - p1
T = F-C
*expressed as a fraction
This calculation is reapplied for cell 2, cell 3, etc.

FIG 3 - Bank survey design involving repeated use of the two product equation.

cell n can exceed its assay in cell n-1. It should be noted that
data smoothing techniques available since approximately
1970 may handle this issue but, for manual calculations, this
property remains a difficulty.
Use of the approach in Figure 3 declined during the period
from 1970 to approximately 1990 because a more effective
method emerged in terms of the quantity of samples required and
its theoretical basis. This more effective method will be discussed
shortly in conjunction with Figure 5. However, since approx-
imately 1990, a change in the type and volume of flotation
machines being installed (ie the use of large unit cells) has
resulted in increased use of a closely related variant of the
method in Figure 3. This variant is illustrated in Figure 4.
The design of the unit cells precludes the possibility of any
intermixing of pulp zones and removes any objections on this
basis. However, for taking dip samples from the pulp zone of
each cell, the degree of mixing in the pulp zone must be
considered. However, it is also possible for valid sampling points
to exist in the pipework between the cells, eliminating the use of
dip samples which is time consuming and presents difficulties if
the pulp zone is not perfectly mixed. Banks of unit cells contain
typically six to ten cells and the moderate number of cells makes
the work load of the method of sampling manageable.
A competing method for sampling a bank emerged from
approximately 1970 (Figure 5). It required samples of the feed,
tailing and the various concentrates for which some grouping was
possible. For each concentrate or concentrate grouping, the flow
rate of solid discharging from the various cell lips had to be
determined also (Restarick, 1976).
From Figure 5, no assumptions on pulp intermixing between
machines or the level of mixing within machines were required,
the only possible exception being the collection of the tailing
sample from the pulp zone (for which an alternative point,
sampling the combined tailing stream after exiting the last cell,
exists in some cases). The elimination of all or virtually all the
pulp sampling reduced the quantity of work. However, a more
advanced method of concentrate sampling, preferably requiring
two people, was required to obtain values for the solid flow rate

Flotation Plant Optimisation Spectrum Series 16 39

CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

Individual concentrates from cells

F X Feed sampling point

c1, c2 … cn • Sampling points for concentrate from each cell
t1, t2 …t n ° Sampling points for tailing from each cell

Assays for calculating of solid balance for each cell using two product equation:

Feed Assay Concentrate Assay Tailing Assay

Cell 1 f c1 t1
Cell 2 t1 c2 t2

Cell n tn-1 cn tn

For the first cell in the bank:

Let C = t/h of solid in concentrate
and F = t/h of solid in feed
and T = t/h of solid in tailing

Stating the two product equation and inserting the symbols for cell 1:
C f -t
Solid Recovery * = =
F c-t
C f -t1
=
F c1 - t 1
f - t1
C =F×
c1 - t 1
T = F-C
*expressed as a fraction
This calculation is reapplied for cell 2, cell 3, etc.

FIG 4 - Bank survey design involving repeated use of the two product equation for a series of tank cells.

for each machine or grouping of machines. Details of this aspect
of the sampling are provided in the appendix.
In Figure 5, the two product equation is demonstrated to be a
special case of a general equation for removal of many
concentrates (designated as n). Figure 5 presents the general
equation and the method for initial examination of the data from
the survey. Concentrate assays and flow rates by lip sampling
must be obtained for each concentrate and a feed and tailing
assay for the bank must also be obtained.
It is important to discuss the reliability which can be attributed
to the measured values for the solid flow rate (C1, C2, C3, … Cn)
in each concentrate (see Figure 5) and hence the appropriate
means for processing the data from this type of sampling.
Reliability of measured concentrate flow rates
A method of bank surveying has been described (Figure 5)
requiring both the assay and concentrate solid flow rate for each
flotation machine or grouping of machines along the bank. In
using this method, it is important to appreciate the reliability of
the method, in particular the required reliability for the measured
flow rates of solid in each concentrate.
As a minimum, the measured solid flow rate values for each of
the concentrates along a bank must be correct on a relative basis.
It is therefore advisable that the same lip sampling personnel
(usually two people) conduct all the tonnage measurements along
a given bank. It is also advisable that the personnel receive prior
training on the use of correct methods and that the training
includes a convenient small sampling exercise where the
correctness of the measured flow rates on an absolute basis could
also be established for the personnel.
By way of guidance, traditional sources of absolute solid
tonnages, eg a weightometer on the grinding circuit feed should
always be used as the ‘scaling tonnage’ in processing of plant
surveys (Figure 5) and surveys should be designed to ensure that
such sources of absolute solid tonnages are available. In some
cases, extra metallurgical samples may have to be taken to link the
plant section being surveyed in detail to the source of an absolute
solid tonnage, eg weightometer, thereby removing reliance on the
correctness of concentrate tonnages on an absolute basis.
For various levels of reliability of the measured solid flow rates
in the concentrate, the metallurgical performance information
which can be reliably calculated is summarised in Figure 6.
In summary, the measured solid flow rates in the concentrate
from lip sampling should only be taken as correct on an absolute
basis in unusual circumstances where a large amount of training
and checking has been completed by the lip sampling personnel.
Further, an unusual set of circumstances needs to exist where
traditional scaling tonnages, eg weightometer tonnages do not
exist. In other unusual circumstances, the combination of a
magnetic flow meter and a density gauge (both of which are
calibrated properly and checked regularly) may be used to
provide a scaling factor for the tonnes per hour of solid for a key
stream in a survey of some peripheral part of a circuit.

40 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

Individual concentrates from cells

f x Feed sampling point

c1, c2 … cn  Sampling points for concentrate from each cell allowing estimate of solid flow rate
(C1, C2 … Cn) where the values are correct on a relative basis (as minimum
standard) and provision of solid assay (c1, c2, … cn)
t  Tailing sample point (after end of bank or possibly pulp region in last machine)
General case:
Let Ci = t/h of solid in the ith concentrate
and F = t/h of solid in the feed to bank
and T = t/h of solid in the bank tailing

It can be shown that:

n

¦ C (c i i - t) n
F i 1
and T F - ¦ Ci
f -t i 1
The calculated value for F is compared with its known value from the weightometer. The observed lip
tonnages for each concentrate (C1, C2 … Cn) are then rescaled upwards or downwards to ensure the
value for F based on the rescaled values for C1, C2, … Cn equals the true value from the weightometer.

For the general case above applied to a bank with seven concentrates:

Individual concentrates from cells

For n=7 in the above equation:
7

¦ C (c i i - t) 7
F i 1
and T F - ¦ Ci
f -t i 1
Rewriting the two product equation where F is calculated from the concentrate flow rate
C and the feed (f) and the product assays (c and t):
C(c - t)
F and T F - C
f -t
It can be observed that the two product equation for F is a special case of the above
equations for F with 7 and n cells in the bank.

FIG 5 - Bank survey requiring assays for the feed, tailing and concentrates and the relative flow rates for each concentrate
(no knowledge of the degree of intermixing (M) is needed for legitimate use of this method – see also Figure 3).

CALCULATION OF MINERAL RECOVERIES
The traditional technologies for obtaining the composition of ore
samples have provided elemental assays. While assays for
elements are very useful for providing information for down-
stream processes such as smelting, elemental assays would not
normally be the first choice for a concentrator metallurgist in most
circumstances in a perfect world.
It is now possible to obtain the assays for minerals directly.
However, there are various limitations to the currently available
methods.
The most basic information on the behaviour of a mineral is its
recovery, ie its distribution between the concentrate and tailing
streams for a separation. Hence, in this section, the objective is to
discuss the calculation of recovery values for minerals from survey
data. The necessary prerequisite is to obtain mineral assays by
suitable means.
Obtaining mineral assays
After laboratory preparation of dry solid from each sampling
point in the survey, the next step is to submit representative
portions of each sample for assaying. Metallurgists would prefer
to receive assays of both minerals and elements, each for different
purposes.
Traditionally, the industrial assaying methods (atomic
absorption spectroscopy (AA), X-ray fluorescence (XRF),
inductively coupled plasma emission spectroscopy (ICP) and fire
assaying) have provided the concentration of valuable metallic
elements such as Cu, Ni and Au and the concentration of other
metallic elements such as Fe, often contained in both gangue and
valuable minerals. There has been an increase in the ability to
provide assays for entities often associated with gangue minerals,
such as silica (SiO2) and calcium oxide (CaO). Elements such as
barium may be associated solely with the mineral barite (BaSO4)
in some ores.
These methods do not provide mineral assays but provide the
total concentration of elements such as copper, iron and silica
resulting from the various minerals in which they exist. The iron
assay represents this element in gangue sulfide minerals such as
pyrite and pyrrhotite, in non-sulfide gangue minerals such as
dolomite and various silicate minerals and may include iron in a
valuable copper-bearing mineral such as chalcopyrite (CuFeS2) or

Flotation Plant Optimisation Spectrum Series 16 41

CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

Case 1
Lip tonnages correct on absolute basis
ie compatible with reference tonnage and correct relative to each other

Residence times measured correctly

Grade

Recovery

Case 2
Lip tonnages correct relative to each other but all 10% too high in comparison to
reference tonnage (lip tonnages not rescaled as described in Figure 2.5)
Calculated residence times lower by
10% than correct value
Grade

Position of grade-recovery curves for

cases 1 and 2 is the same.

Recovery

Case 3
Lip tonnages correct relative to each other but all 15% too low in comparison to reference
tonnage (lip tonnages not rescaled as described in Figure 2.5)

Calculated residence times higher by

15% than correct value
Grade

Position of grade-recovery curves for

cases 1, 2 and 3 is the same.

Recovery

Case 4
Lip tonnages not correct relative to each other

Position of grade recovery curve becomes subject to additional errors and is no longer the
same as for cases 1, 2 and 3

FIG 6 - Illustration of the reliability of calculated metallurgical performance for a valuable mineral/element using the method with
measured lip tonnages (see also Figure 5).

bornite (Cu5FeS4). Equally, a silica assay usually does not reflect
the occurrence of one non-sulfide gangue mineral.
In the supergene or oxidised portion of a copper orebody, a
copper assay is likely to reflect the existence of various co-
dominant copper-bearing minerals such as chalcocite (Cu2S),
covellite (CuS), native copper (Cu), secondary chalcopyrite
(CuFeS2), bornite (Cu5FeS4) and others. Clearly, in these
examples, mineral assays cannot be calculated from the elemental
assays alone.
In the fortunate case where all the copper exists as
chalcopyrite, which is often the case in the deeper primary zone
in a copper orebody, the stoichiometry of the mineral can be used
to convert the copper assay to a chalcopyrite assay. In the primary
zones of orebodies, it is reasonably common for the valuable
element to reside in just one mineral. However, for species such
as iron or silica, this situation is almost non-existent.
Mineral assays can be obtained by three methods:
1. the use of a quantitative analysis method for minerals such
as quantitative X-ray diffraction;
2. the use of point counting or related methods involving an
optical microscope, image analyser or automated electron
microscope for mounted samples, often size fractions and
sometimes on the original sample containing all sizes; and
3. the use of approximate methods for conversion of elemental
assays to mineral assays where the approximations are
specific to a given ore or site.
Methods 1 and 2 provide mineral assays but some
understanding of their limitations is required. Quantitative X-ray
diffraction is suitable for assaying total samples or individual size
fractions from samples. Point counting methods as practised via
an optical microscope are usually applied to size fractions and the
finest size fraction from a given sample cannot be measured by
this method. In this situation, assumptions have been needed to
sum the data across the size fractions to obtain the mineral head
grade. In general, these mineral assay methods are less accurate
than traditional assaying methods for elements and they have
increased technical difficulty and/or become much more costly if
minerals are being assayed in the low concentration region of less
than one per cent by weight. In practice, methods 1 and 2 need to
be used in conjunction with elemental assays.
Further, in many cases, the use of all the methods in combination
is warranted, allowing a method which draws on the strengths of
the data from various methods for obtaining mineral assays.
Suitable programs are needed for execution of the necessary
calculations to produce a combined set of mineral assays.

42 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

It has been observed that operating sites with both traditional
assaying facilities for elements and recently acquired assaying
facilities for minerals generally do not merge the two sources of
assay data well to maximise the benefit to the site.
Method 3 relies on the elemental assays which are very reliable
but requires assumptions for conversion to mineral assays, which
are of various levels of reliability. As an example, consider a
simple ore containing lead (only as galena), pyrite, a small
amount of copper (only as chalcopyrite) and various non-sulfide
gangue minerals including talc.
Clearly, the lead and copper assays can be converted to galena
and chalcopyrite assays, preferably by using the measured
stoichiometry for galena and chalcopyrite in that ore. If sulfur only
existed in the galena, chalcopyrite and pyrite, the sulfur assay
could be corrected for sulfur in the galena and chalcopyrite, and
the corrected sulfur assay converted to a pyrite assay, preferably
using the measured stoichiometry for pyrite in that ore. By
subtracting the mineral assays for galena, pyrite and chalcopyrite
from 100 per cent, a useful estimate of the non-sulfide gangue
assay would be obtained. If a substantial and variable proportion of
the magnesium in the ore existed in non-talc minerals, conversion
of magnesium assays to talc assays may provide unreliable
information on the talc. In such circumstances, a combination of
elemental assays which are converted to mineral assays and
directly measured mineral assays for the talc (eg by X-ray
diffraction) could be devised for that particular ore.
Obtaining mineral recoveries with introduction of
the recalculated feed
After calculation of the solid balance, obtaining the recovery (the
distribution of a mineral, or element, between the concentrate and
tailing streams for a separator) becomes a simple calculation. It is
recommended that the recovery is calculated with respect to the
recalculated feed, which is illustrated in Figure 7 for a separator
producing two products and for which the raw assays are used (ie
no statistical adjustment of the assays to provide ‘smoothed’ or
completely internally consistent assays has occurred).
In Figure 7, the solid balance was calculated using the method
shown for the feed, concentrate and tailing assays of mineral A,
the only mineral for which there was a large difference between
the feed and tailing assays. As a result, a mineral balance must
also exist for mineral A (see the same flow rates for mineral A in
the actual feed and recalculated feed (from summation of flows of
mineral A in the products) in Figure 7).
However, for minerals B and C, it is very unlikely that an exact
mineral balance will exist using the observed assays for minerals B
and C and the solid balance calculated from mineral A. In
Figure 7, the different flow rates of minerals B and C are shown for
the actual feed and for the recalculated feed (defined as the sum of
the flow rates in the concentrate and tailing). The recovery values
for minerals B and C can be expressed with respect to the
recalculated or the actual feed. It is recommended that the recovery
value is calculated with respect to the recalculated feed because:

The following raw mineral assays were obtained for a separator producing two products.

Feed Concentrate Tailing Feed Tailing

Mineral A 10.2 81.3 1.2
Mineral B 19.7 11.6 21.6 F F –D
Mineral C 70.1 7.1 77.2 t/h solid t/h solid
Total 100 100 100
Concentrate
D t/h solid

The major separation is occurring for mineral A and its assays will be used to calculate
the solid split.

Using the two product equation and noting the feed flow rate is 1000 t/h of dry solid:

D f t
F ct
10.2  1.2
81.3  1.2
0.112
?D 0.112 u 1000
112 tph

Summary of Flow Rates and Recoveries

Feed Recalculated Feed # Concentrate Tailing

(t/h) (t/h) (t/h) (t/h)
Mineral A 102 102 91 11
Mineral B 197 205 13 192
Mineral C 701 693 8 685
Solid 1000 1000 112 888

Recovery* Recovery**
Mineral A 89.2% 89.2%
Mineral B 6.6% 6.3%
Mineral C 1.1% 1.2%
Solid 11.2% 11.2%

# sum of mineral flow rates in concentrate and tailing

* with respect to actual feed
** with respect to recalculated feed

FIG 7 - Illustration of calculation of mineral recoveries with respect to the recalculated feed. A similar method can be used for
calculations based on element assays.

Flotation Plant Optimisation Spectrum Series 16 43

CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
• the resulting recovery value is less affected by sampling and
assay errors existing in the data, particularly for minerals
with higher flow rates in the concentrate than the tailing;
• the values are bounded between zero per cent and 100 per
cent; and
• a consistent approach is provided.
There are several practical advantages in obtaining the recovery
of minerals, aside from the previously stated observation that,
because minerals are being separated, logic dictates that mineral
data are the most relevant.
When metallurgical data are analysed on the basis of elements,
some or all of the solid in the non-sulfide gangue category may
be ignored. Further, knowledge of an ore’s behaviour via minerals
allows ready calculation of the effects from changing ore head
grade, assuming all other properties remain fixed. Because this
approach requires a knowledge of the stoichiometry of each
mineral and because the concentrations of minor and trace
elements in the lattice of each mineral should also be found in
detailed analyses, more complete information on the real limits to
a separation can be obtained, based on the properties of the
minerals. For example, most sphalerite contains some iron in
solid solution and the mineral marmatite reflects naturally
occurring zinc sulfide with very high levels of iron.
CALCULATION OF RECOVERY-SIZE DATA
The calculation of recovery values for each mineral for a bank in
the previous section specifies the behaviour of each mineral but,
if some values are abnormal or if there is an economic imperative
to improve the recovery value in some size fractions, there is no
direct clue to the mechanism by which the gangue mineral was
recovered or the valuable mineral was not recovered. The first
step towards determining the clues is to obtain recovery values
for all the minerals in all size fractions and to summarise the
information for all the minerals (and also water) in a graph of
mineral recovery (y-axis) plotted with the mid-point of each size
fraction using a logarithmic scale (x-axis).
The key steps in generating this information are now listed:
1. Deciding if there is sufficient quality in the mass balance
calculated from the original survey to warrant generation of
more detailed data.
2a. Selection of the sizing method(s).
2b. The sizing of the samples.
3. Specification of any grouping of size fractions before
assaying to lower costs, ensuring consistent grouping of size
fractions for each group of samples involved in a particular
bank or plant section. (This extends to a consistent approach
to the size fractions combined at or near the boundaries of
sizing methods based on different principles, eg screening
(sizing method based on physical dimensions) and
cyclosizing (sizing method based on hydraulic size), for
example.)
4. Specification of the assays (elemental and/or mineral) to be
obtained and submission of the size fractions to the assay
laboratory after pulverising the relatively coarse fractions to
provide a low fundamental error (see the appendix) in
sampling a very small mass for chemical analysis. For each
stream, a ‘head’ sample (unsized) should be submitted for
checking consistency of the assays on the size fractions.
A number of checking steps are required during the procedure,
particularly with the sizing step (2b) and the internal consistency
of the assays. These are now discussed.
After each sizing is completed, the size distribution should be
graphed to detect any irregularities in the shape of the curve
44 Spectrum Series 16
which may indicate experimental errors. Further, the position of
the curve, ie its coarseness or fineness should be compared with
the typical range of values for that stream. If the sizing is outside
the expected range, further checking is needed.
For the various assays, the first check is to calculate the head
grade for the sample by using the size distribution and the assay
for each size fraction. The consistency between this value and the
head grade for the sample (assayed at the same time as the size
fractions) needs to be checked. Further, these two values for the
head grade of the sample should be compared with the value
obtained for the sample when the survey was conducted and
assayed initially.
After checking the integrity of the data, a system is needed for
processing the data to provide useful information on the recovery
values for each mineral in each size fraction. The concept of the
recalculated feed remains relevant for the sized data.
Initially, consider sized data for the simplest situation, ie two
output streams and one input. The following two options exist for
processing the data when there is no size reduction inside the
section considered:
1a. use the solid balance established with the original survey
data to calculate the flow rate for each mineral in each size
fraction in the recalculated feed (by summing the flow rates
of each mineral in the concentrate and tailing), and compare
each value with the observed value for the feed stream;
1b. calculate the recovery value for each mineral in each size
fraction with respect to its flow rate in the recalculated feed;
2a. use a data smoothing program to adjust the assays
statistically and provide balanced data for each mineral in
each size fraction; and
2b. calculate the recovery value for each mineral in each size
fraction, noting that the recovery values with respect to the
actual feed and recalculated feed will be the same as
the data were adjusted statistically to provide internally
consistent assays in 2a.
When there is size reduction inside the section considered, it is
only possible to employ methods 1a and 1b for the recovery
calculation for each mineral in each size fraction, there being no
basis for data smoothing. It is also relatively common that no
sized data exist for the feed stream. This can arise because a
sampling point did not exist or the available sample was not
selected for sizing. In this case, the recoveries can only be
calculated by use of the recalculated feed.
The next step which allows interpretation of the recovery-size
information more readily is to summarise the information in a
graph of mineral recovery (y-axis) and the mid-point of the size
fraction (x-axis) using a logarithmic scale on the x-axis. All the
minerals should be plotted on the one graph for a given
processing stage and the values for the mid-points on the x-axis
should reflect the specific gravity of the minerals when the sizing
device operates on the basis of hydraulic equivalents such as for a
cyclosizer, its predecessor (infrasizer) or for sizings by beaker
decantation. When the sizing is obtained by sieving, the
mid-points for all minerals will be the same.
To appreciate fully recovery-size data for minerals, the flow
rates of each mineral in each size fraction in the recalculated/
actual feed also need to be reviewed. The values for mineral
recovery are key indicators of metallurgical performance but the
practical and economic significance of the recovery values
depends on the quantity of each mineral in each size fraction on
which the recovery value acts. While it is less common to include
this information in the recovery-size curves, this can be done by
way of a histogram along the base of the recovery-size curve.
Alternatively, tables or other means may be devised to allow
convenient recognition of this important information.
Flotation Plant Optimisation

INTERPRETATION OF RECOVERY-SIZE CURVES
The hydrophobic mineral
The position and shape of the recovery-size (log scale) curves
have metallurgical significance for liberated valuable and gangue
minerals from the sequence of subprocesses in the pulp zone,
which are particle/bubble collision and adhesion, followed by
successful transportation through the pulp zone. Detachment of a
particle from a bubble can be caused by turbulence during the
transportation step. The next necessary step is successful
transportation through the froth zone. Survey methods now exist
for examination of the recovery of the hydrophobic and other
minerals across the froth zone as discussed in a following section.
Particle/bubble collision depends on physical properties of the
particles (eg particle diameter and density) and the system (eg
level of turbulence and bubble size). The chemical surface
properties of the particles (particularly their hydrophobicity) and
bubbles are dominant in the particle/bubble adhesion step and
remain important for a successful transportation step, ie the
avoidance of detachment. Appropriate settings for both the
physical and chemical properties of the particles and the system
are important to minimise detachment. Hence, the position and
shape of the resulting recovery-size (log scale) curves are related
to the physical and chemical properties of the system through
their effect on the subprocesses in the pulp zone for liberated
particles.
For technical and economic reasons, complete liberation of all
minerals cannot be achieved. Hence, preceding comments in this
section which referred only to liberated particles have to be
tempered with the superimposed effects from incomplete
liberation which affect all size fractions, but which typically
affect the less liberated coarse size fractions to the greatest extent
for a given ore. It must be noted that, by obtaining liberation data
as a next step, the effects of this complication can be understood
for a given ore. This avoids having full reliance on deductions
from recovery-size graphs, particularly for an ore whose
liberation characteristics are not well known.
It is also assumed that a particle which reports to a given size
fraction in the recovery-size curves existed in that size fraction in
the flotation process. The existence of a fully dispersed pulp is
therefore assumed.
It is useful to summarise the general form of some basic
relationships (Pyke, Fornasiero and Ralston, 2003) between the
efficiency of each subprocess in the pulp zone and particle
diameter (Figure 8). Figure 8 arises from ongoing investigations
at the Ian Wark Research Centre. The authors used the following
equation to describe the collection or capture efficiency (Ecoll) for
a particle and bubble in terms of the efficiencies of the
subprocesses in the pulp zone:
Ecoll = EcEaEs
where:
Ec = collision efficiency
Ea = adhesion efficiency
Es = stability efficiency
For each of the efficiencies in the subprocesses (Ec, Ea and Es),
Pyke, Fornasiero and Ralston (2003) provided equations relating
them to properties of the flotation system.
For the particle/bubble collision step, there is a direct
relationship between collision efficiency and size (see Figure 8).
This arises because small particles approaching a bubble tend to
be swept along the stream-lines around bubbles while larger
particles with higher momentum have an increased ability to
cross the stream-lines and complete a collision. This inability to
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
FIG 8 - Calculated relationships between collision efficiency Ec,
adhesion or attachment efficiency Ea and stability efficiency Es
and particle diameter dp, and the resulting calculated relationship
between the first-order rate constant k and particle diameter dp for
the pulp zone, where cases a, b and c are for advancing contact
angles of 50°, 65° and 80° respectively (Pyke, Fornasiero and
Ralston, 2003).
cross stream-lines means that the probability is low for small
particles to approach a bubble sufficiently closely for a collision
to have occurred, ie sufficiently closely for the adhesion process
to commence. For a particle of a given diameter, its momentum
and its ability to cross the stream-lines is increased if it has a
higher density.
The second subprocess (known as adhesion or attachment)
commences after a particle and bubble approach each other very
closely. Pyke, Fornasiero and Ralston (2003) described the
approach of the particle and the bubble:
Should they approach quite closely, within the
range of attractive surface forces, the intervening
liquid film between the bubble and particle will
drain, leading to a critical thickness at which
rupture occurs. Movement of the three-phase
contact line (the boundary between the solid
particle surface, receding liquid phase, and
advancing gas phase) then occurs, until a stable
wetting perimeter is established.
There is an inverse relationship between the adhesion
efficiency and particle diameter (Figure 8), noting that this
subprocess is referred to as attachment in the reference (Pyke,
Fornasiero and Ralston, 2003). Small particles slide more slowly
over the surface of a bubble as they are ‘protected’ due to their
low diameter by existing in the more slowly moving boundary
region of the water phase near the surface of the bubble. The
lower sliding velocity for smaller particles allows a greater time
for the adhesion subprocess to be successful, ie for the actual
contact time between the particle and the bubble to exceed the
needed contact time for adhesion, known as the induction time.
Similarly, for the transportation step in the pulp zone, there is
also an inverse relationship between the stability efficiency and
particle diameter (Figure 8). Small particles are subjected to
lower forces of detachment and have a higher probability of
successful transportation as a particle/bubble aggregate to the
base of the froth phase.
45
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
In Figure 8, Pyke, Fornasiero and Ralston (2003) provided the
summation of the relationships for the three described
subprocesses by calculating the relationship between the first-
order rate constant (k) and particle diameter for the hydrophobic
mineral for a set of conditions. This calculation involved the use
of typical parameters for the induction time.
The first-order rate constant (k) for a mineral is closely related
to its recovery (see section – Process Analysis with Down-the-
Bank Flotation Data) and summarises the propensity for the
flotation process for a mineral to proceed in the pulp region for a
set of conditions. The rate constant for the pulp region is linked to
the efficiencies in the three subprocesses by the following
equation (Pyke, Fornasiero and Ralston, 2003):
k = z Nb Ec Ea Es
where:
z = frequency of particle bubble collision
Nb = number of bubbles per unit volume
It can be seen that the relationship between the rate constant
and particle diameter in the example has a maximum as a result
of the interaction of the direct relationship between efficiency and
particle diameter for the collision subprocess and the two indirect
relationships discussed for the adhesion and transportation
subprocesses. In some cases, a small plateau region may result
from the interactions of the three relationships. The importance of
a high particle hydrophobicity in the adhesion subprocess and in
avoiding detachment in the transportation subprocess is
highlighted in Figure 8 where calculated outcomes are provided
for contact angles of 50, 65 and 80 degrees.
For convenience of presentation and interpretation, industrial
recovery-size data are graphed with a logarithmic scale for the
particle size on the x-axis. This step spreads the values for
particle diameter conveniently, as for plotting sizing distribution
data.
The hydrophilic minerals
The entrainment mechanism is a non-selective physical
mechanism for transfer of minerals from the pulp zone to the
concentrate launder. The entrainment mechanism does act on all
minerals – both hydrophilic and hydrophobic – because it is a
non-selective physical mechanism. It will be discussed
principally in terms of the hydrophilic minerals because the
mechanism provides a much higher proportion of the total
recovery of these minerals. For the finest size fractions, the
entrainment mechanism can be the sole recovery mechanism for
the non-sulfide gangue minerals in some ores.
The pattern of behaviour of hydrophilic particles of various
sizes in the froth phase is better described through knowledge of
the entrainment mechanism. Throughout a perfectly mixed pulp
zone, hydrophilic particles of all sizes exist at a uniform
concentration in each unit of water in the pulp zone. These
particles are subject to hydraulic classification in the froth region
because the water in the froth region must contain a
representative sample of the hydrophilic particles when the water
enters the froth. It must be noted that water from the pulp zone is
the only source of water essential for a stable froth zone in a
conventional flotation machine.
TABLE 1
Values for entrainment efficiency factors (ENTi) for siliceous non-sulfide gangue (Lynch et al, 1981).
-11 -16 +11 -23 +16
Entrainment efficiency 0.83 0.44 0.24
46 Spectrum Series 16
Knowledge of the water recovery assists in interpretation of the
relationship between the recovery of unwanted minerals (usually
gangue sulfide minerals and non-sulfide gangue minerals) and
particle diameter (log scale). With the value for water recovery for
a bank, it can be seen if the entrainment mechanism explains all
the recovery of the gangue minerals or only a portion. This is
possible because it is observed that the recovery of liberated,
hydrophilic gangue in the 0 - 10 μm region is 0.8 of the value for
the water recovery. If the water recovery were ten per cent, a
normal flotation system would exhibit a recovery of eight per cent
in the 0 - 10 μm fraction for a hydrophilic gangue mineral (specific
gravity of 2.7) from the entrainment mechanism.
It can be visualised that, for a perfectly mixed pulp zone of a
conventional flotation machine, the recovery of ten per cent of the
water would result in recovery of ten per cent of the 0 - 10 μm
liberated non-sulfide gangue by this non-selective physical
mechanism if there was no drainage of this mineral in the froth
region. In this situation, the entrainment efficiency value would
be 1.0. In real flotation systems, the observed entrainment
efficiency factor is 0.8 for the 0 - 10 μm fraction for siliceous
non-sulfide gangue.
The entrainment efficiency value for a size fraction can be
defined formally:
Entrainment Efficiency Value for size i (ENTi )
(Mass of free gangue per unit mass water)Con
=
(Mass of free gangue per unit mass of water)Top of pulp
The entrainment efficiency value for size i has been called the
classification function (Johnson, 1972) or the classification vector
(Lynch et al, 1981). It is commonly described by the term ENTi
in the present literature and the topic was reviewed recently
(Johnson, 2005). Typical values for ENTi for a range of size
fractions of siliceous non-sulfide gangue are provided in Table 1.
On a sized basis, the simplest mathematical depiction of
entrainment is a series of entrainment efficiency values
approaching unity for the fine sizes and which approach zero for
sizes exceeding 50 μm. If a gangue mineral has an elevated
specific gravity, its value can be determined to establish if lower
values for ENTi are required for that system.
The following relationship can be demonstrated from the
definition of the entrainment efficiency factor (Lynch et al, 1981)
for the first-order rate constant (ki) of a mineral in size fraction i
being recovered by entrainment:
ki = ENTi kw
where:
kw = the first-order rate constant for water recovered in the
concentrate
Because the first-order rate constant is closely related to the
recovery of a mineral or water (see section – Process Analysis
with Down-the-Bank Flotation Data) from a cell or bank, the
previously described comparisons of the water recovery value and
the recovery of gangue minerals in the relatively fine size
fractions are validated as part of interpretation of mineral
recovery-size (log scale) data.
Size fraction (μm)
-33 +23 -44 +33 -75 +44 +75
0.11 0.04 0.03 0.0
Flotation Plant Optimisation

Review
To understand the behaviour of a particular ore, there is often a
benefit from observing its flotation behaviour in the absence of
collector but in the presence of a normal frother addition (and
possibly depressant addition). The benefit can be increased if the
recovery-size behaviour is obtained for the recovered minerals.
Minerals which display normal flotation through the particle/
bubble collision sequence in the absence of collector can be
readily recognised (recovery values in some size fractions
considerably larger than water recovery) and the size fractions in
which this occurs can be readily identified (Johnson and Jowett,
1982). Minerals which display this behaviour can be valuable
sulfides, non-valuable sulfides and non-sulfide gangue (eg talc).
Hydrocarbons present during ore formation and which remain
associated with a portion of one mineral (eg pyrite with a rimming
of hydrocarbons) can cause this behaviour (Croxford et al, 1961).
As a result of the various mechanisms for flotation and the
steps involved in each mechanism, recovery-size curves for the
valuable mineral being recovered by flotation adopt a general
form with many variations, of which a few are demonstrated in
Figure 9. It is given that the curves were observed after high
residence times, ie their position and shape were changing only
very slowly with additional residence time. For the valuable
mineral in all graphs, two scenarios (Johnson, 2006) are shown
for the fine fractions. In scenario 1, the diminished recoveries are
typical of those arising from deficiencies in the collision
subprocess. In scenario 2, the more greatly diminished recoveries
result from an additional effect beyond the collision deficiency
likely to result from an imbalance in the ratio of adsorbed
hydrophobic/hydrophilic species.
For the minerals for which recovery is not being sought, an
extremely wide range of positions and shapes exist (Figure 9).
For the hypothetical fully liberated feed, the following are
demonstrated:
• entrained liberated non-sulfide gangue (cases A, C and E);
• entrained and hydrophobic (collector), liberated sulfide
gangue (case C); and
• entrained and naturally hydrophobic liberated sulfide gangue
(case E).
For the realistic cases (B, D and F) with an imperfectly
liberated feed, cases are shown with increased recoveries of
sulfide and non-sulfide gangue due to their recovery in
composites, along with lower recovery of coarse valuable mineral
due to the lower hydrophobicity of composite particles containing
some valuable mineral.
ADDITION OF LIBERATION DATA TO
RECOVERY-SIZE DATA
The addition of liberation data to recovery-size data eliminates
the need for deductions and inferences about the state of
liberation of a mineral in given size fraction of a given product.
For a new ore or for an ore which is unfamiliar to a metallurgist,
the collection of liberation data in initial phases of the work can
TABLE 2
Example of some point count data for sphalerite in one size fraction.
Liberation data (categories)
% Liberated % in binary composite with listed minerals
Galena
% in each category 60 16
† Includes quaternaries.
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
increase confidence in the interpretation of recovery-size data for
the ore in question such that less liberation data may be required
in further stages of experimental work on the same ore.
Liberation data have been collected traditionally on size
fractions and there are lower and upper limits (mounting size
limitations) to the size fractions which can be examined. The
lower limits result from the physics of the method employed
(Jones, 1987). For techniques based on optical microscopy or
generation of X-rays by electron microscopy, different reasons
exist for the lower limit but the value is typically in the region of
5 μm, sometimes slightly lower. The limit can be much lower if a
different method with different limitations is useable for a given
ore, eg one based on backscattered electrons. Size fractions
submitted for liberation analysis must not have been pulverised.
For a given size fraction in a separation with one concentrate
and one tailing, the liberation data can be used in two
complementary ways. For the concentrate, the liberation state of
the unwanted minerals can be recognised, providing strong clues
to their mechanism of recovery. Equally, the liberation state of the
valuable mineral is determined. If significant losses of valuable
minerals in composites are observed, corrective steps involving
grinding or regrinding may be assessed.
Liberation data allow the flow rate of each mineral in each size
fraction of each product (obtained from the recovery-size (log
scale) level of analysis) to be distributed between a number of
categories (typically from five to ten). The performance of the
various categories becomes the basis of the next level of analysis.
Liberation data for a mineral are supplied in one of two basic
forms. The traditional point counting method with use of an
optical microscope provides liberation data in the general format
as shown in Table 2 for a size fraction.
Automated scanning electron microscopes can also provide the
data in the point counting format illustrated in the preceding
table. The data from these devices are also commonly supplied in
another format based of the percentage of the mineral of interest
in each category. An example is shown in Table 3.
In the type of data indicated in the preceding table, some
grouping of the original data has been performed. Typically, the
data are provided with increments of ten per cent from category
to category.
The reader is reminded that several calculation steps, as
described in preceding sections of this chapter, are the necessary
precursors to incorporation of liberation data. These steps are
listed:
1. Calculation of the mass or solid balance for the circuit. The
water balance should also be calculated.
2. Use of size distributions and assays for the size fractions of
the concentrate and tailing (as the minimum relevant
streams), along with the solid flows from item 1, to
calculate the flow rates of minerals in each size fraction of
the concentrate and tailing.
3. Calculation of the recovery of each mineral in each size
fraction with respect to the recalculated feed.
% in ternary composites†
Iron sulfide Non-sulfide gangue
4 11 9
47
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

Fully Liberated Feed Typical Feed

(All minerals 100% Liberated) (Liberation of valuables >80%)
(Liberation of other minerals 70 to 90%)
Case A Case B
100 100

90 90

80 80

70 70
Recovery (%)

Recovery (%)
60 60

50 50

40 40

30 30

20 20

10
Water Recovery (10%) Water Recovery (10%)
10

0
0
1 10 100
1 10 100
log Size (um) log Size (um)

Case C Case D
100
100

90
90

80
80

70
70
Recovery (%)

60
60
(%)
Recovery
50
50

40
40

30
30

20
Water Recovery (10%) 20
Water Recovery (10%)
10
10

0
0
1 10 100
1 10 100
log Size (um)
log Size (um)

Case E Case F
100
100

90
90

80
80

70
70
Recovery (%)

60
Recovery (%)

60

50 50

40 40

30 30

20 20

Water Recovery (10%) Water Recovery (10%)

10 10

0 0
1 01 001 1 10 100

log Size (um) log Size (um)

Valuable Mineral (Scenario 1) – ‘collision effect’ for fine valuables

Valuable Mineral (Scenario 2) – additional deleterious effect beyond ‘collision effect’ for fine valuables

Non-sulfide Gangue (Case 1)

Non-sulfide Gangue (Case 2) – lowered contribution from recovery in composites with valuable mineral
compared to case 1

Sulfide Gangue

FIG 9 - Examples of some mineral recovery – particle diameter (log scale) curves for a feed with perfect liberation (cases A, C, E) and for a
feed with acceptable liberation (cases B, D, F) for which recovery of unwanted minerals in composites becomes a possible mechanism.

Steps 2 and 3 represent the calculations required for graphing
the mineral recovery – particle size (log scale) curves as
discussed in earlier sections. The next level of analysis (liberation
level) is now discussed. An example in the literature with the
same methodology can also be reviewed (Johnson, 1987).
To provide an example of the type of calculations required for
the liberation level of analysis, some sphalerite liberation data for
a single selected size fraction are presented for a system with
such data for the concentrate and tailing only. In Table 4, the
calculations for one size fraction of one mineral are illustrated.
The flow rate of sphalerite in the concentrate and tailing were
6188.2 and 1155.9 kg/h respectively. These flow rates (column
A) were multiplied by each liberation value (each divided by
100) in the five columns labelled as B to distribute the mineral

48 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

TABLE 3
Example of second form of liberation data for sphalerite in a size fraction.

Liberation data (per cent of mineral of interest defined in each category)

100 90 - 100 70 - 90 50 - 70 30 - 50 20 - 30 10 - 20 0.1 - 10
Percentage in category 61.2 30.4 4.1 2.1 1.1 0.4 0.4 0.3

TABLE 4
Example of processing of point count form of some sphalerite liberation data for one size fraction.
Product Mineral Liberation data (B) Mineral flow in categories (kg/h) (C)
flow Liberated % in binaries with Tern Liberated Binaries with Tern
(kg/h) (A) category category
Galena Iron Non-sulfide Galena Iron Non-sulfide
sulfide gangue sulfide gangue
Concentrate 6188.2 75.6 5.1 12.2 2.4 4.6 4678.3 315.6 755.0 148.5 284.7
(97.0) (90.4) (73.0) (54.8) (33.2)
Tailing 1155.9 12.7 2.9 24.2 10.6 49.6 146.8 33.5 279.7 122.5 573.3
(3.0) (9.6) (27.0) (45.2) (66.8)
Recalculated 7344.1 65.7 4.8 14.1 3.7 11.7 4825.1 349.1 1034.7 271.0 858.0
feed (100.0) (100.0) (100.0) (100.0) (100.0)

Notes: Tern = category containing ternary and quaternary composites.

Recalculated feed = recalculated feed from summation of flows in concentrate and tailing as shown in columns labelled as (A) and (C).
Liberation values for each category in the row ‘Recalculated feed’ and in columns labelled as (B) calculated using the flow rates for the
recalculated feed in columns labelled as (C) and the total flow rate of sphalerite (7344.1 kg/h) in column (A).

flow rates amongst the various liberation categories (columns
labelled as C).
The recalculated feed is obtained in the table by summation of
the flow rates in each liberation category. In a further step, the
distribution of sphalerite between the concentrate and tailing in
each liberation category is calculated (see values in brackets). In
other words, the recovery of mineral in each liberation category
has been calculated with respect to the recalculated feed.
Although the calculations are straightforward, the presentation
of the calculated quantities is more difficult for the listed reasons:
• there is interest in information on the feed, concentrate and
tailing streams;
• there are many size fractions and minerals; and
• there are many liberation categories of relevance.
For compactness in some examples of presentation methods,
data are used for a bank where only four size fractions encompass
all the solid in the concentrate and tailing. The first presentation
uses tables only (see Table 5). This table provides the flow rates
and recoveries for all the minerals in the various liberation
categories and size fractions in the concentrate. This table is
particularly valuable for recognition of the major sources of
gangue dilution in the concentrate.
Therefore, the dominant flow rates for the gangue minerals
(>200 kg/h) in the concentrate have been highlighted. In reading
the flow rates for minerals in a typical binary in Table 5, the
reader is reminded of the following example for the sphalerite/
iron sulfide binary (+38 μm fraction): 281.8 kg/h of sphalerite and
117.8 kg/h of iron sulfide. The table also provides the recovery
values for all the categories in the various size fractions.
The recovery values are particularly valuable for examination of
metallurgical behaviour of the various liberation categories of the
valuable mineral in the concentrate. The recovery values represent
the metallurgical performance for the valuable mineral in the
category. However, it must be noted that the significance of a high
or low recovery value for the valuable mineral in a category
depends also on its flow rate in the feed. For example, a category
with a high flow rate in the feed and a metallurgical recovery of
95 per cent may result in a much higher flow rate of valuable
mineral to the tailing than another category with a low flow rate in
the feed and a low metallurgical recovery of 20 per cent.
Some recovery values have not been entered in Table 5. In
collecting data from a standard number of particles, the number of
observations in some liberation categories may be too small to
provide reliable recovery values. For example, galena was present
in the tailing in small amounts only and no observations existed for
some liberation categories in the tailing, implicating a recovery of
100 per cent. A second phase of data collection recording only
information on a selected mineral (in this case for galena) in some
liberation categories is required in such circumstances.
The graph of recovery-particle diameter (log scale) for the
valuable mineral can be updated with the additional liberation
categories for the valuable mineral as shown in Figure 10 where
the curve for sphalerite contained six data points. With the
grouping of size fractions, the curves for the various liberation
categories contained four size fractions in this example. It can be
noted that the curve for liberated sphalerite displayed higher
recoveries than for the overall sphalerite curve. Further, the points
available for sphalerite-galena binaries were in a similar region to
the liberated sphalerite because both the galena and sphalerite were
hydrophobic in these binary particles. For the other liberation
categories containing sphalerite in composites with less
hydrophobic unwanted minerals, lower recoveries were observed
in general. Similar patterns of behaviour are often observed in data
sets of this type.
Graphical presentations can be extended to three-dimensional
graphs (with axes of size fraction/liberation category/measure of
quantity) of the types listed:
1. distribution of a selected mineral in the feed, concentrate or
tailing;
2. flow rate of a selected mineral in the feed, concentrate or
tailing; and

Flotation Plant Optimisation Spectrum Series 16 49

CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

TABLE 5
Listing of flow rates (kg/h) and recoveries for sphalerite, galena, iron sulfide and non-sulfide gangue in the various liberation categories
and size fractions for a zinc rougher. All gangue mineral flow rates >200 kg/h have been highlighted to indicate the dominant gangue
mineral flow rates.
Flow rate of mineral in all liberation categories in concentrate
Flow in binary with
Mineral Liberated Galena Sphalerite Iron sulfide NSG Ternary
Galena 0.6 - 61.1 0.0 108.3 117.8
Sphalerite 1199.3 858.5 - 281.8 180.2 756.9 +38 μm
Iron sulfide 43.4 0.0 117.8 - 0.0 256.4
NSG 84.5 65.2 297.6 71.3 - 362.8
Flow in binary with
Mineral Liberated Galena Sphalerite Iron sulfide NSG Ternary
Galena 3.9 - 61.8 10.8 25.5 127.8
Sphalerite 4678.3 315.6 - 755.0 148.5 284.7 -38 μm +11 μm
Iron sulfide 266.5 13.1 339.1 - 13.9 139.1
NSG 214.3 0.0 267.2 80.7 - 641.6
Flow in binary with
Mineral Liberated Galena Sphalerite Iron sulfide NSG Ternary
Galena 7.0 - 11.0 1.3 4.4 4.7
Sphalerite 1336.7 29.4 - 29.4 63.3 13.3 -11 μm +8 μm
Iron sulfide 116.9 3.5 46.4 - 18.9 0.0
NSG 143.9 12.4 87.5 43.7 - 0.0
Flow in binary with
Mineral Liberated Galena Sphalerite Iron sulfide NSG Ternary
Galena 56.5 - 89.3 10.3 35.6 37.9
Sphalerite 5978.3 131.7 - 131.7 283.1 59.3 -8 μm
Iron sulfide 575.3 17.4 228.3 - 93.2 0.0
NSG 1206.4 103.8 733.5 366.7 - 0.0
Recovery of mineral in all liberation categories in concentrate
Recovery in binary with
Mineral Liberated Galena Sphalerite Iron sulfide NSG Ternary
Galena # - # 0.0 12.9 6.9
Sphalerite 80.5 92.3 - 39.1 9.9 19.9 +38 μm
Iron sulfide 1.6 0.0 13.5 - 0.0 5.0
NSG 0.2 2.8 12.7 0.3 - 2.6
Recovery in binary with
Mineral Liberated Galena Sphalerite Iron sulfide NSG Ternary
Galena # - 79.3 20.0 12.9 29.4
Sphalerite 97.0 90.4 - 73.0 54.8 33.2 -38 μm +11 μm
Iron sulfide 3.3 10.5 31.0 - 1.0 11.1
NSG 0.9 0.0 33.1 1.9 - 29.6
Recovery in binary with
Mineral Liberated Galena Sphalerite Iron sulfide NSG Ternary
Galena # - # # 9.1 24.2
Sphalerite 92.4 # - 66.5 63.3 44.1 -11 μm +8 μm
Iron sulfide 7.2 # 37.3 - 12.3 ##
NSG 2.5 28.1 58.0 8.2 - ##
Recovery in binary with
Mineral Liberated Galena Sphalerite Iron sulfide NSG Ternary
Galena # - # # 5.3 15.1
Sphalerite 74.6 # - 32.5 29.4 16.0 -8 μm
Iron sulfide 5.0 # 29.0 - 8.8 ##
NSG 2.0 24.3 53.2 6.9 - ##

Notes: NSG = non-sulfide gangue; = gangue flow >200 kg/h; # = insufficient observations of this category in tailing for reliable value; ## =
insufficient observations of this category in concentrate for reliable value.

50 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

100

90

80

70

60
Recovery (%)

Total Sphalerite
50
Liberated Sphalerite
Binary with Galena
40 Binary with Iron Sulfide
Binary with NSG
Ternaries
30

20

10

0
1 10 100
Particle Diameter (μm - log scale)

FIG 10 - Relationship between total sphalerite recovery and particle diameter (μm – log scale) for data from the recovery-size level of
analysis and from the recovery-size-liberation level of analysis illustrating the recovery of the following sphalerite liberation categories in a
sphalerite rougher – liberated, binaries with galena, binaries with iron sulfide, binaries with non-sulfide gangue (NSG) and ternaries.

3. recovery values for a selected mineral (often the valuable TABLE 6

mineral) in the concentrate or tailing. Calculation method for the total liberation of a mineral in a
The types of graphs in item 1 are demonstrated in Figure 11 given stream.
because their structure is difficult to explain without an actual
Input information:
demonstration. However, the graph types in items 2 and 3 are not
demonstrated because their structure is straightforward. The Column 1 Flow of mineral in each size fraction (from size distribution
graph type in item 2 is suitable for demonstrating how each and assays)
gangue mineral is diluting a concentrate and also for Column 2 Liberation value for mineral in size fraction
demonstrating the flow rate of the valuable mineral in each
category and size fraction in the tailing. It can be noted that items Column 1 Column 2 Column 3
2 and 3 effectively represent a means of graphical presentation of
the type of data presented in tabular form in Table 5. Size Mineral flow Liberation Flow of
value liberated
Liberation data can also be used to identify the extent of mineral
liberation achieved at each grinding or regrinding stage at the -105 μm +53 μm 5 20 1
commencement or within a processing circuit. This information
can be obtained by including samples of the combined feed and -53 μm +CS2 8 25 2
product for a grinding or regrinding circuit. For each size -CS2 +CS3 10 40 4
fraction, measurement of the liberation values for the minerals in -CS3 +CS4 10 60 6
the feed and product allows the liberation value for minerals in -CS4 +CS5 10 75 7.5
each stream to be calculated. For each mineral, the increase in
liberation across the grinding circuit is calculated by difference. -CS5 +CS6 20 80 16
The method for calculating the liberation value within a given -CS6 +CS7 37 80* 29.6
stream is illustrated in Table 6. Total 100 → A 66.1 → B
The taking of a few extra samples in a plant survey may allow
more reliable quantification of the changes in liberation at size The liberation value is calculated as:
reduction steps. For example, to improve the quality of the data, it
is preferable that a single combined feed sample and a single B
× 100
product sample be taken for an overall regrinding system, which A
often will contain a mill for size reduction and a classification 66.1
device. It is advisable that a survey plan is reviewed for the ie × 100 = 66.1%.
100
directness by which liberation data may ultimately be obtained.
In other words, the technical viability of a ‘liberation survey’ This is simply the weighted average of the liberation values in Column 2.
within the larger plant or circuit survey needs to be addressed Note: in this example, observations could not be made on the CS7
separately in the planning steps. Equivalent samples can be taken fraction. The liberation value for this size fraction (denoted by *) was
in pilot plant work and in laboratory batch or cycle tests involving assumed to equal the value for the C6 fraction because the liberation
regrinding. values had essentially reached a plateau region.

Flotation Plant Optimisation Spectrum Series 16 51

CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

Recalculated Feed

100

90
Distribution of Sphalerite in Recalculated
80

70
Zinc Rougher Feed

60

50

40

30
Total

ory
20
Ternaries

g
ata
10 Sphalerite in Binaries with NSG

nC
Sphalerite in Binaries with Iron Sulfide

ti o
0
Sphalerite in Binaries with Galena

era
Total
-8um Liberated Sphalerite

Lib
+8um
+11um
+38um
Size Fraction

Concentrate

100
Distribution of Sphalerite in Zinc Rougher
Concentrate (with respect to Recalculated

90

80
Zinc Rougher Feed)

70

60

50

40

30

Total
ory
20
Ternaries ata
g
10 Sphalerite in Binaries with NSG
nC

Sphalerite in Binaries with Iron Sulfide

0
io

Sphalerite in Binaries with Galena

rat

Total
-8um Liberated Sphalerite
e

+8um
Lib

+11um
Size Fraction +38um

Tailing

100
Distribution of Sphalerite in Zinc Rougher
Tailing (with respect to Recalculated

90

80
Zinc Rougher Feed)

70

60

50

40

30
Total
20
Ternaries
ry
go

10 Sphalerite in Binaries with NSG

ata

Sphalerite in Binaries with Iron Sulfide

nC

0
Sphalerite in Binaries with Galena
t io

Total
-8um Liberated Sphalerite
era

+8um
+11um
Lib

Size Fraction +38um

FIG 11 - Examples of usage of three-dimensional graphs with axes of size fraction/liberation category/percentage of sphalerite (with
respect to the recalculated feed). The diagram displays data for the zinc rougher concentrate, the zinc rougher tailing and also the
recalculated feed to the zinc rougher.

52 Spectrum Series 16 Flotation Plant Optimisation


Historically, metallurgists have often sought to determine the
benefits of regrinding by examination of the separation results
(often using the flotation process) with various levels of
regrinding. Particularly for regrinding to fine sizings, it is much
safer to establish initially if the regrinding step is causing a
significant increase in the liberation level of the valuable mineral
and the gangue minerals. If a significant increase in liberation is
measured, the necessary conviction is provided to seek conversion
of the increase in liberation into an increase in performance of
the following separation. Such increases in performance may not
emerge after initial cursory separation tests, even if a significant
improvement in liberation is resulting from the regrinding.
Optimisation of the chemical conditions during regrinding and
the following flotation separation may require many months of
tests for some systems.
Liberation data can now be provided to customers with
stereological corrections included. Such corrections allow for the
well-known observation (Jones, 1987) that data from two-
dimensional mounts may overestimate the levels of liberation of
each mineral because a two-dimensional intersection of a particle
may, by chance, occur in one mineral only, even if there is more
than one mineral in the particle. This is a phenomenon which is
often quoted in texts. The existence of this technical issue along
with the cost of obtaining liberation data may be two reasons for
the limited use generally made by industrial metallurgists of the
type of data in the past.
It can be argued that the two-dimensional data, ie uncorrected
data remain useful for process analysis. In terms of technical
purity, it is preferable to employ corrected liberation data.
However, uncorrected two-dimensional data may be used with
caution and with other checks to improve understanding of the
process for the listed reasons.
Firstly, in calculation of the recovery values for liberation
based species in Table 5, there may be correction factors (a and b)
needed to the flow rates in the numerator and denominator of the
following equation:
% Recovery = (a × flow rate in conc × 100) /
(a × flow rate in conc + b × flow rate in tailing)
In percentage terms, the magnitude of the correction factor
(slightly less than or equal to 1) for the tailing flow rate (b) may
differ from the correction factor for the concentrate flow rate (a),
ie b does not equal a. However, because the direction of the
correction must be the same for both the numerator and
denominator, there is at least partial cancellation of the
stereological effect. When the flow rate for the concentrate in the
equation is very large compared to the flow rate for the tailing,
the cancellation of the stereological effect is more complete.
Secondly, if a metallurgist is suspicious of the flow rates in
Table 5 provided by uncorrected liberation data, it is possible
with care and patience to section a given set of particles in a
briquette at various levels to obtain, for practical purposes,
three-dimensional liberation data. Note that this step is a
possibility for particularly unusual circumstances or when very
high confidence is required in the liberation data.
Thirdly, the liberation level (two-dimensional basis) for the
valuable mineral(s) in a process feed is used to judge if the sizing
of the feed is at a value where an acceptable separation could be
expected. On a two-dimensional basis, the guidelines shown in
Table 7 can be used in process engineering where the majority of
the liberation is occurring in a grinding circuit at the start of the
circuit, ie very little or no regrinding exists inside the circuit.
Of course, obtaining a high level of liberation does not guarantee
an efficient separation as the appropriate settings for the process
variables must also be obtained. For an efficient froth flotation
separation, both a high level of liberation and suitable settings for
the physical and chemical variables in the process are needed.
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
In porphyry copper circuits, liberation levels for the copper
minerals may sometimes be less than 50 per cent in the rougher
feed. However, the regrinding steps in the circuit must raise the
liberation level of the copper minerals such that the guidelines in
Table 7 are achieved for the recalculated feed (summation of the
final concentrate(s) and tailing(s)). This applies to other circuits
with major regrinding and liberation steps in the flotation circuit.
TABLE 7
The maximum potential quality of separation possible at various
liberation levels (two-dimensional data) for the feed or
recalculated feed.
Liberation level (%) Potential quality of separation
(2D data)
>80% High, very efficient separation possible
70 to 80% Sound, moderate separation possible
<70% Poor, inadequate separation only possible
With liberation data that has been converted from 2D to 3D
values, assuming that extensive experience in the future indicates
substantial lowering of the two-dimensional values occurs upon
correction, a new set of boundaries based on experience with 3D
liberation data may be established.
The use of liberation data in conjunction with recovery-size
data can be employed in existing plants for surveys, monthly or
yearly composite samples on feed and products, or for samples
from supporting laboratory or pilot plant development work.
Plotting of the level of liberation of the valuable mineral(s) in the
recalculated feed of a concentrator versus time (monthly or
quarterly) has been demonstrated to be very useful in
understanding declines and monitoring improvements in plant
performance (Young et al, 1997). Equally, for a new deposit, the
technique can be applied to laboratory test work and pilot plant
surveys as a part of flow sheet development and optimisation.
THE ROLE OF SURFACE AND SOLUTION
ANALYSES FOR ORGANIC AND INORGANIC
SPECIES
The use of recovery-size data in conjunction with liberation data
will sometimes reveal the recovery of a liberated gangue mineral in
quantities which indicate it was recovered due to hydrophobicity
imparted by unintentionally adsorbed collector, possibly assisted
by activating metal ion adsorption. It cannot be explained solely by
entrainment as the recovery value is much greater than the
yardstick provided by the water recovery and the entrainment
efficiency factors, and there is no evidence of recovery of the
mineral (beyond entrainment levels) in tests with frother but
without collector.
Further, for the valuable mineral, there may be cases of
unexpectedly low recovery of the liberated valuable mineral,
typically in the finest size fractions. In principle, further clues to
the causes of these two types of process weakness can be sought
by examination of surfaces. For the surface of the valuable mineral,
an unacceptable ratio between collector based hydrophobic and
detrimental hydrophilic species (derived from the system) may be
found by analysis for organic and inorganic species.
Such surface analysis provides an additional fourth level in the
system for process analysis. However, the surfaces which are
analysed must reflect their condition in the real process. Therefore,
additional samples which may be used for surface analysis must be
taken at the time of the survey. Alternatively, for an initial survey,
the samples can be processed to obtain recovery-size and liberation
data and weaknesses recognised. A decision can then be made if
surface analysis is warranted and a new survey taken for both
recovery-size/liberation data with separate samples for surface
analysis.
53
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
The minerals in the samples for surface analysis must retain their
properties from the process pulp. Therefore, samples that have
been dried and possibly sized as part of a normal survey procedure
are of no use. Samples which have not been specially taken and
handled will provide misleading surface analyses because of
desorption of key species or the introduction of artefacts on the
surface. The correct handling of samples for surface analysis will
depend on the requirements of the surface analysis facility but may
include protection of the sample at low temperature (low reaction
rates) and under nitrogen (absence of oxidising agent).
A method has been developed which uses classical chemical
methods (Rumball and Richmond, 1996) to assess the aqueous
phase and then to reassess the aqueous phase after a complexing
agent, ethylenediaminetetra-acetic acid (EDTA), has been used to
return cations existing as hydrophilic debris (for example metal
hydroxides) and possibly other forms on the surfaces of minerals
to the aqueous phase. The increase in concentration of each
cation is related to the quantity existing on the surfaces of
minerals. Hence, general information is provided as the method
cannot indicate the minerals on which the debris was deposited in
greatest surface concentration.
Surface analysis of particular minerals by instrumental
methods with a strong focusing ability is required to provide such
information. However, the EDTA method will indicate the
relative increase in the quantity of precipitated species from the
start of the circuit to the exit points. The increase in the extent of
oxidation in the circuit can be established by a relatively simple
method.
INCLUSION OF FROTH RECOVERY IN SURVEYS
An industrial flotation machine contains two distinct regions –
the pulp and froth zones. Traditional flotation surveys seek to
obtain the combined performance of the two regions, ie an overall
mass balance. Until recently, there has been no practical way of
establishing a separate mass balance for each of the two zones as
part of a plant survey. With collection of data from new types of
measuring/sampling devices, the establishment of a separate mass
balance for the two zones is now possible.
The froth recovery Rf of a mineral in a size fraction is defined
as the ‘fraction of particles attached to bubbles entering the froth
phase that are collected in the concentrate launder’ (Vianna,
2004) and this term is included in the modelling framework for
the first-order rate constant (k) for the overall process (combined
pulp and froth regions) as used at the Julius Kruttschnitt Mineral
Research Centre:
k = P Sb Rf
P = mineral floatability in pulp region
Sb = bubble surface area flux
Rf = froth recovery
The key additional step requires a novel device for capturing
and measuring the upward flow rate of minerals in particles
attached to bubbles at the top of the pulp zone, just before the
bubbles enter the froth region. The quantity of work and therefore
time involved in obtaining the extra data during a survey will
determine its usage in surveys. It is likely that this additional
measurement will only be made initially at selected cells.
However, as techniques improve, the method may become
sufficiently routine for its use at all cells.
The obtaining of the froth recovery across the froth region is
relevant for both conventional flotation machines (including tank
cells) and cells with water addition to the froth region.
Novel measuring devices (Seaman, 2006) for determining the
transfer or flow rate of attached particles at the top of the pulp
54 Spectrum Series 16
region at steady state during the survey have the advantage that
there is no disruption to the plant. Some other methods require
alteration of the froth height after the survey and collection of
data for the cell in question at each froth height. Methods that
require such disruption to the plant and are also time consuming
are much less practical.
Froth stability and therefore froth recovery (Rf) are believed to
depend on factors such as frother type and concentration and
particle properties such as size, number, hydrophobicity and
shape. The role of particle hydrophobicity is of particular interest
because two recent studies have shown there is a maximum in the
relationship between froth recovery and hydrophobicity (Ata,
Ahmed and Jameson, 2003; Schwarz and Grano, 2005). These
findings corroborate broadly with earlier more fundamental
findings reported by Dippenaar (1982). While increases in
hydrophobicity assist the subprocesses of adhesion and stable
transportation in the pulp zone, such increases may only assist in
the froth region up to an intermediate hydrophobicity, indicated
by a contact angle in the region 60 to 70 degrees.
PROCESS ANALYSIS WITH DOWN-THE-BANK
FLOTATION DATA
Down-the-bank industrial flotation data can be presented and
analysed in two forms:
1. cumulative metallurgical performance along the bank via
cumulative grade/cumulative recovery curves for the
valuable mineral or element and one method for presenting
the associated selectivity curves for the sulfide and non-
sulfide gangue minerals (see example in Figure 12); and
2. kinetic data (fraction remaining (log scale) versus nominal
residence time) for the valuable and other minerals (see
examples in Figure 14).
Equivalent data can also be obtained from batch laboratory
flotation tests.
Another example of cumulative metallurgical performance via
cumulative grade/cumulative recovery curves for the valuable
mineral is given in Figure 13 for a three stage closed circuit
cleaning system. Two types of graphs are shown. Information on
the magnitude of each circulating load of valuable mineral (and
its assay relative to the stream it joins) is incorporated in the more
complete version located at the base of Figure 13.
The behaviour of each mineral may be expressed in kinetic
terms where the standard graph (Figure 14) is the fraction or
percentage remaining in the bank (log scale) on the y-axis and
nominal residence time (average residence time per cell
multiplied by the number of cells to that point in the bank) on the
x-axis for plant data. For batch laboratory tests, the cumulative
flotation time is employed on the x-axis. Such relationships for
the batch and plant cases allow recognition of the kinetic
components which describe the behaviour of each mineral in the
bank. The number of kinetic components, the first-order rate
constant of each and the proportion of the mineral with that rate
constant value can be determined. Such graphs can be on an
overall, unsized basis or they can be used for minerals or
elements in size fractions.
It must be emphasised that the rate constant components can
only be determined from data with progressive values for fraction
remaining and residence time. If the relationship between mineral
recovery and particle diameter (log scale) has been determined
along with the residence time at the end of a bank or a batch
flotation stage, without progressive data on reaching the overall
performance for the stage, a single rate constant value can be
calculated for that mineral. The rate constant calculated by this
method will represent a weighted average of more than one rate
Flotation Plant Optimisation

FIG 12 - Analysis of down-the-bank flotation data for a single
rougher or cleaner bank using the grade recovery curve for the
valuables and selectivity curves for each gangue mineral.
constant component in many cases. While the single rate constant
value can be useful for some purposes, its limitations must also
be recognised.
Some examples are provided in Figure 14. In the simplest case
in Figure 14, the graph of fraction remaining (log scale) versus
time for a single rate constant is depicted, providing a straight
line relationship. This case is followed by an example where the
kinetic behaviour with two, quite different values (high and low)
for the rate constant is depicted. In this case, two components
existed for the mineral. It can be noted that a closely related case
with two components exists when the low rate constant is a
non-recoverable component with a rate constant of zero min-1. A
third case is also depicted in Figure 14 with three components, ie
three quite different values (high, low and medium) for the rate
constant.
By comparing the batch and continuous cell examples in
Figure 14, it can be seen that a higher value for the fraction
remaining (ie lower recovery) exists at a given residence time for a
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
given set of rate constants for the plant cells. By comparing cases 2
and 3 and noting the apparent similarity of the fraction remaining
(log scale) and residence time relationships, it can be surmised that
there may be difficulty in accepting the existence of the third
component, ie a fit deemed acceptable may still be obtained using
the equation in the form with two rate constant components.
Valuable liberated mineral with a medium rate constant will be
recovered in a rougher section with adequate installed residence
time. The diminished rate constant component of the valuables
(medium or even lower rate constant) may reflect solely the
deficiencies of fine particles in the collision subprocess, or it may
reflect an additional issue arising from an imbalance in
hydrophilic and hydrophobic species on the surface of the
component. Such liberated valuable mineral would be expected to
report to the final concentrate on the basis of its liberation state.
However, the medium rate constant valuable mineral may have
more difficulty in being sufficiently competitive during
conventional cleaning for the majority of it to report to a high-
grade final concentrate.
The design of laboratory batch rougher or cleaner tests with
collection of a series of concentrates also allows collection of
traditional cumulative concentrate grade/cumulative concentrate
recovery curves and kinetic data as just described for a bank in a
plant. The usefulness of such data depends on the times selected
for collection of each concentrate in a laboratory batch test.
For both laboratory and plant processes, it is advisable to have
many data points at the commencement of a stage to allow
checking for anomalous behaviour because the process was
allowed to commence before the feed was fully prepared. It is
also often useful to survey the last cell individually to obtain
directly the concentrate grade for the final cell. For a laboratory
batch test, a separate concentrate for a short time at the end of the
test is the equivalent. Experience with surveying improves a
metallurgist’s judgement on the appropriate level of detail in
down-the-bank data collection.
Down-the-bank data can be collected for a series of bank
surveys or laboratory tests with changed conditions for each test.
An example is given in Figure 15 (Grano, Ralston and Johnson,
1988). The following points can be made in establishing the
changes that occurred in the performance of each mineral for
each survey or laboratory test:
• A graph of cumulative grade/cumulative recovery for the
valuable element or mineral provides a summary of the
performance of the bank and indicates the trade-off between
grade and recovery in each portion of the bank. The graph
does not allow analysis of changes in the behaviour of the
unwanted minerals when a series of tests is being compared.
• A convenient method for analysis of changes in behaviour of
the unwanted minerals is by the use of selectivity curves.
There is more than one method for plotting selectivity
curves. A common feature of all methods is that the recovery
of each mineral (or an equivalent measure) is plotted against
the recovery of the valuable mineral in a series of graphs.
• From the use of selectivity curves of the type illustrated in
Figures 12 and 15, changes in behaviour of the other
minerals relative to the valuable mineral can be easily
recognised. These changes in conjunction with changes in
the flotation rate constant of the valuable mineral itself can
be used to understand the origins of changes in position of a
series of grade-recovery curves for the valuable mineral.
• The type of selectivity curves shown in Figures 12 and 15 are
legitimate for convenient comparison of the recovery of each
gangue mineral at a given recovery for the valuable mineral,
the primary purpose of the described analysis. (However, for
55
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

1 2 3 4 5 6 7 8
Rougher First Cleaner First Cleaner
Concentrate Tailing
A
B
8 7 6 5 4 3 2 1 C

Second Cleaner

D
1 2 3 4 5 6 7 8
Major Water Addition
(for dilution cleaning) Third Cleaner

Third Cleaner Concentrate

Grade Recovery Curve of Valuable Element in Cleaning System

45
Cumulative Grade of Valuable Element (%)

1
40
1-3
1-2 1-5
35
1-4 1-7
Combined Third Cleaner Concentrate
1 1-6 1-8
30
1-3
1-2 1-5
25
1 1-4 1-6
1-7 Combined Second Cleaner Concentrate
20
1-2
1-3 1-4 1-5 1-8
15
1-6 Combined First Cleaner Concentrate
1-7
10
1-8
5
Rougher Concentrate
0
0.00 20.00 40.00 60.00 80.00 100.00 120.00
Cumulative Recovery of Valuable Element (%)

More Complete Grade Recovery Curves for Valuable Element in Cleaning System
(providing additional information on magnitude and grade of circulating loads of
valuables)
45
Cumulative Grade of Valuable Element (%)

1
40
1-3
1-5
1-2
35
1-4 1-6
1-7 Combined Third Cleaner Concentrate
1
30
1-8
1-3 1-4
1-2 1-5 Combined Second Cleaner Concentrate
25
1-6 1-7
1
20
1-8 Second Cleaner Feed (C+D)
1-2 1-3 1-6
1-4 1-5
15
1-7 Combined First Cleaner Concentrate (C)
1-8
10

First Cleaner Feed (A+B)

5

0
Rougher Concentrate (A)
0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 90.00 100.00 110.00 120.00

Cumulative Recovery of Valuable Element (%)

FIG 13 - Examination of down-the-bank flotation data for three stages of closed circuit dilution cleaning using grade recovery
curves for the valuable element using a straightforward method and a more advanced method.

considering a single test in isolation, this type of selectivity zinc sulfate. The reagent also caused a slight decrease in
curve is not legitimate for evaluating changes in selectivity selectivity of iron sulfide and non-sulfide gangue relative to the
between the valuable mineral and gangue mineral under galena. From Figure 15 alone, no discussion is possible on
examination from the commencement to the end of the test. changes in selectivity between the galena and each unwanted
For this purpose on a single test, the logarithm of the recovery mineral during each test.
value for both the valuable mineral and the gangue mineral in
question must be plotted. A straight line relationship indicates USE OF SUMMARY GRAPHS WITH LIMITS
no change in selectivity along the bank or during the batch IMPOSED BY LIBERATION
test.)
The set of curves in Figure 15 shows that the selectivity of The sections in this chapter have described analysis methods for
sphalerite relative to galena was improved clearly by the use of the separation which has taken place in a laboratory batch

56 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

Laboratory Batch Flotation Series of Isolated and Perfectly Mixed Plant Cells
Equation applied Equation applied
1 1 1
k f t FR (1  Is  Im )  Is  Im
FR (1  Is  Im )e  I s e  k s t  Im e  k m t (1  k f O ) N
(1  k s O ) N
(1  k m O ) N

Where:
FR = fraction remaining
kf, ks, km = first order rate constant for high, low and medium rate constant components (min-1)
Is , Im = fraction of component with low or medium rate constant
t = time for batch operation and O = nominal residence time for each of the N continuous
cells in series for continuous operation
Case 1
Time (min) Nominal Residence Time (min)
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
1 1
Fraction Remaining

Fraction Remaining
0.1 Case 1 0.1 Case 1

0.01 0.01

Cases 2 and 3
Time (min) Nominal Residence Time (min)
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
1 1
Fraction Remaining

Fraction Remaining

Case 2 Case 2
0.1 0.1
Case 3 Case 3

0.01 0.01

kf
(min-1)
ks
(min-1)
km
(min-1)
Is Im Comment
Case 1 0.2 0 0 0 0 High rate constant for all valuables (one component)
Case 2 0.2 0.01 0 0.3 0 High rate constant for 70% of all valuables (two components)
Case 3 0.2 0.01 0.08 0.2 0.1 High rate constant for 70% of all valuables (three components) and 10% with a medium rate
constant

FIG 14 - Demonstration of various relationships between fraction remaining for the valuable mineral (log scale) and (i) time (min) for a batch
process and (ii) nominal residence time (min) for a series of perfectly mixed plant cells and with various rate constant components.

process, a pilot plant or a full-scale plant. Strengths and queue, and composite particles containing the valuable mineral
weaknesses of the separation can be recognised and targets for follow and are ranked in terms of decreasing quality of the
process improvement can be identified. In addition, a different composite particles. This approach is shown schematically in
perspective on the obtained separation can be reached by Figure 16 for a size fraction. The theoretical performance curve
comparing the observed performance with the best achievable assumes that no liberated gangue minerals reach the concentrate
performance based on the level of liberation of the valuable and that gangue mineral can only reach the concentrate in
mineral(s) in the relevant feed. Only those particles containing composite particles with the valuable mineral. The theoretical
some valuable mineral are considered and, in effect, a queue is performance curve in Figure 16 is expressed as a grade-recovery
produced in which the liberated particles form the start of the curve for the valuable mineral in the process and is obtained by

Flotation Plant Optimisation Spectrum Series 16 57

CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

Zinc Sulfate 0.0 kg/t, pH 8.5 Zinc Sulfate 2.0 kg/t, pH 8.5
FIG 15 - Example of comparison of results of laboratory batch tests using the grade recovery curve for galena and selectivity curves
for the unwanted minerals (sphalerite, iron sulfide and non-sulfide gangue) (Grano, Ralston and Johnson, 1986).

summing the liberation information from all size fractions in the
feed. Comparisons of the observed and theoretical performance
for each size fraction can also be obtained.
The physical meaning of key points on the theoretical grade
recovery curve is provided in Figure 16.
For the feed information, liberation data for a new orebody or
deposit, or the feed to a plant or a bank in a plant could be used.
In some cases, the liberation data for a recalculated feed must be
used to calculate the relevant theoretical performance curve. This
situation would arise for a number of reasons:

58 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS

Consider the listed particles as separator feed:

Note: Darkened mineral is the valuable mineral (V).

Liberated (group 1 – 40% of V) High quality composites (group 2 – 30% of V)

Medium quality composites (group 3 – 20% of V) Low quality composites (group 4 – 10% of V)

Particles containing no valuable mineral (group 5)

Construction using additional liberation information

• Point B – % of valuable mineral that is liberated in group 1 (40%) and its grade
• Point C – % of valuable mineral in groups 1+2 (70%) and its grade
• Point D – % of valuable mineral in groups 1+2+3 (90%) and its grade
• Point E – % of valuable mineral in groups 1+2+3+4 (100%) and its grade
Step 1 Step 2

100 MAXIMUM 100

MAXIMUM
mineral/element in concentrate (%)

B C POSSIBLE B C POSSIBLE
CUMULATIVE GRADE of valuable

CUMULATIVE GRADE of valuable

90
CONCENTRATE 90
CONCENTRATE
D D
mineral in concentrate (%)

80 GRADE (100%) 80
GRADE (100%)
70 70
THEORETICAL GRADE-RECOVERY
60
E 60
CURVE FOR VALUABLE MINERAL E
50
THEORETICAL GRADE-RECOVERY 50

40 CURVE FOR VALUABLE MINERAL 40

30 30

20 20
THEORETICAL GRADE-RECOVERY
A FEED GRADE OF CURVE FOR VALUABLE ELEMENT A FEED GRADE OF
10 VALUABLE 10 VALUABLE
MINERAL (10%) 0
MINERAL (10%)
0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100

CUMULATIVE RECOVERY of valuable mineral in concentrate (%) CUMULATIVE RECOVERY of valuable mineral/element in
concentrate (%)

group 1 group 1
group 1+2 group 1+2
group 1+2+3 group 1+2+3
group 1+2+3+4 group 1+2+3+4

FIG 16 - Illustration of the concepts in construction of a theoretical grade recovery curve for separation of a valuable mineral (dark phase)
based on the group of particles in a size fraction of the feed to the separation (step 1). The theoretical grade recovery curve for the
valuable element in the valuable mineral is included (step 2) based on the stoichiometry of the valuable mineral.

1. for a bank, liberation data on the concentrate(s) and tailing position for both the actual and theoretical performance curves.
only were obtained for a bank, possibly to reduce costs; The non-recovery of liberated valuable mineral can be another
contributor, by ‘pushing’ the observed grade recovery curve to
2. for a total plant or section of a plant, grinding or regrinding the left.
existed within the section and hence a ‘feed’ sample would
not represent the total achieved liberation, unlike the GENERAL OUTCOMES OF THE ANALYSIS –
summed concentrate(s) and tailing (ie the recalculated feed) MAGNITUDES OF SIGNIFICANT PROCESS
which would represent the total achieved liberation; and WEAKNESSES AND SOLUTIONS TO THE
3. for a laboratory batch separation, only the concentrate(s) WEAKNESSES
and tailing exist for that particular sample tested. The type of process analysis sequence described in the
As a further step, bands can be indicated between the actual preceding sections provides information on the magnitude of
and theoretical performance curves to indicate the various sources weaknesses in the process in the fine, intermediate and coarse
of inefficiency from unwanted minerals. Examples are the size fractions. An example of a weakness is a community of
contribution of each liberated sulfide gangue mineral, each particles, typically in either the fine, intermediate or coarse
liberated non-sulfide gangue mineral and composite particles fractions, which has reported to the wrong product and which
containing no valuable mineral (all mentioned categories contain has a similar liberation state, eg fine liberated valuable mineral
no valuable mineral and were not included in construction of the reporting to the final tailing. Such information cannot be
theoretical grade recovery curve) to creation of the difference in obtained from non-sized data.

Flotation Plant Optimisation Spectrum Series 16 59

CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
In general terms, the weaknesses in a process exist in one or
more of the following categories:
1. valuable mineral reporting to the wrong product (tailing or
incorrect concentrate), and
2. gangue mineral(s) diluting a concentrate.
Examples of weaknesses are losses of valuable mineral to the
tailing in low quality composites, due to insufficient liberation of
the ore in the grinding and regrinding stages, or recovery of a
liberated sulfide gangue mineral in a concentrate due to its
accidental activation (a process chemistry issue which may
require the use of surface analysis for full diagnosis). The
misreporting of liberated valuable mineral to the wrong
concentrate or the tailing is another example.
Further, the magnitude of the significant weaknesses can be
determined from such data, leading to a ranking of the
importance of the weaknesses.
From the diagnoses of the mechanisms causing the
weaknesses, solutions to the weaknesses can be developed. Test
work is usually required at this stage. From a combination of the
magnitude of the weaknesses, the capital and/or operating cost of
the corrective action in the plant, along with the probability of
success of the solution, the most prospective corrective actions
can be selected and implemented in the plant.
Ongoing collection of further sized/liberation data can be used
to monitor the change in process performance resulting from
implementation of process improvements.
REFERENCES
Ata, S, Ahmed, N and Jameson, G, 2003. A study of bubble coalescence
in flotation froths, Int J Miner Process, 72:255-266.
Croxford, N J W, Draper, N and Harraway, D H, 1961. Some aspects of
the carbonaceous fraction of Mount Isa lead concentrate, The
AusIMM Proceedings, 197:149-161.
Dippenaar, A, 1982. The destabilization of froth by solids, 1. The
mechanism of film rupture, Int J Miner Process, 9:1-14.
Grano, S R, Ralston, J and Johnson, N W, 1988. Characterization and
treatment of heavy medium slimes in the Mount Isa Mines lead-zinc
concentrator, Minerals Engineering, 1(2):137-150.
Johnson, N W, 1972. The flotation behaviour of some chalcopyrite ores,
PhD thesis, The University of Queensland.
Johnson, N W, 1987. Application of kinetics and liberation data to
analysis of an industrial flotation process, CIM Bulletin,
80(899):113-117.
Johnson, N W, 2005. A review of the entrainment mechanism and its
modelling in industrial flotation processes, in Proceedings Centenary
of Flotation Symposium, pp 487-496 (The Australasian Institute of
Mining and Metallurgy: Melbourne).
Johnson, N W, 2006. Liberated 0 - 10 μm particles from sulphide ores,
their production and separation – Recent developments and future
needs, Minerals Engineering, 19:666-674.
Johnson, N W and Jowett, A, 1982. Comparison of lead primary rougher
behaviour of several Mount Isa lead zinc ores, in Proceedings Mill
Operators’ Conference, pp 271-285 (The Australasian Institute of
Mining and Metallurgy: Melbourne).
Jones, M P, 1987. Applied Mineralogy – A Quantitative Approach
(Graham and Trotman: London).
Lynch, A J, Johnson, N W, Manlapig, E V and Thorne, C G, 1981.
Mineral and Coal Flotation Circuits – Their Simulation and Control
(Elsevier: Amsterdam).
Pyke, B, Fornasiero, D and Ralston, J, 2003. Bubble particle
heterocoagulation under turbulent conditions, J Colloid and
Interfacial Science, 265:141-151.
Restarick, C J, 1976. Pulp sampling techniques for steady state
assessment of mineral concentrators, in Proceedings Sampling
Symposium, pp 161-168 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
60 Spectrum Series 16
Rumball, J A and Richmond, G D, 1996. Measurement of oxidation in a
base metal flotation circuit by selective leaching with EDTA, Int J
Miner Process, 48:1-20.
Schwarz, S and Grano, S, 2005. Effect of particle hydrophobicity on
particle and water transport across a flotation froth, Colloids and
Surfaces, A: Physiochem Eng Aspects, 256:157-164.
Seaman, D, 2006. Selective transport of attached particles across the froth
phase, PhD thesis, The University of Queensland.
Vianna, S M, 2004. The effect of particle size, collector coverage and
liberation on the floatability of galena particles in an ore, PhD thesis,
The University of Queensland.
Young, M F, Pease, J D, Johnson, N W and Munro, P D, 1997.
Developments in milling practice at the lead/zinc concentrator of
Mount Isa Mines Limited from 1990, in Proceedings Sixth Mill
Operators’ Conference, pp 3-12 (The Australasian Institute of
Mining and Metallurgy: Melbourne).
APPENDIX
Calculation of minimum sample mass
The purpose of this appendix is to describe a means for
calculation of a minimum mass of sample required from a much
larger population to allow a specified, acceptable statistical error
for the properties of the sample relative to the properties of the
population. Errors are introduced at all steps in the chain of
events associated with obtaining the sample, starting with its
initial sampling from the process stream, its later handling/size
reduction/subsampling and the following chemical or other
analysis of the subsample. Various types of error can be
recognised at each of the steps.
To assist this discussion, the concepts of a correct sample and a
representative sample will now be described. François-Bongarçon
and Gy (2002a) recently discussed sample correctness:
A sample is said to be ‘correct’ when any
fragment in the lot to be sampled has the same
probability of being selected in the sample as any
other one.
For a sample to be representative (François-Bongarçon and Gy,
2002a), it must satisfy the preceding statement of correctness (to
ensure no bias exists in the sample) and, additionally, it must have
an acceptably small variance (to ensure the sample is sufficiently
reproducible).
One type of error (fundamental sampling error) connected with
the heterogeneity of the particles in the population being sampled
is now described because of its importance in calculation of the
minimum mass of sample required. François-Bongarçon and Gy
(2002a) described the heterogeneity of the particles (the shape,
density and liberation state of each particle being examples) in
the population as the ‘natural, constitutive heterogeneity’ of the
population of particles.
To calculate the minimum mass ms for a sample from a flowing
stream with steady properties at a specified fundamental sampling
relative variance s2FSE, the following equation developed by Gy
(1982) must be used:
ms = Cd3a2/ s2FSE
where:
C = the sampling constant of the material being sampled for a
particular particle size and component of interest (depends
on all the relevant properties of the set of particles submitted
to sampling)
d = nominal top size of the material (cm)
a = fractional concentration of the material of interest
Flotation Plant Optimisation

For a specified fundamental variance, it can be observed that
particle size has a dominating effect on the required minimum
sample size, while the properties of the set of particles (expressed
via the sampling constant C) and the fractional concentration of
the mineral of interest (a) play a role.
The sampling constant C is given by:
C = clfg
where:
c = mineralogical composition factor (defined in a following
equation)
l = liberation factor (d1/d)b where d1 is the nominal top size at
which complete liberation occurs and parameter b is
determined for the system by experiment (François-
Bongarçon and Gy, 2002b)
f = particle shape factor (often taken as 0.5)
g = size range factor (0.25 to 1.0 as shown in Table A1)
The size range factor g can be estimated from the ratio of the
nominal top size d to the lower size limit d' (size corresponding to
approximately five per cent undersize) as listed in Table A1.
TABLE A1
Guidance for selection of size range factor (g).
Description of size Value for d/d' Resulting value for g
range
Broad >4 0.25
Medium breadth 2 to 4 0.50
Narrow <2 0.75
Uniform 1 1.00
The mineralogical composition factor c is calculated from:
c = (1-a)[(1-a)ρ1 + aρ2]/a
where:
ρ1 = density of the particles of the mineral of interest (g/ml)
ρ2 = density of the other mineral(s) (g/ml)
The liberation factor (l) which contributes to the sampling
constant C was defined earlier as (d1/d)b. The ratio d1/d is 1.0 (its
maximum value) when the mineral of interest is fully liberated
and d becomes equal to d1. The liberation value (l) remains equal
to 1.0 irrespective of the value for parameter b when the mineral
of interest is fully liberated. It can be noted that a larger sample
mass is required to ensure a specified s2FSE when the population is
fully liberated, due to its increased heterogeneity.
The setting of d1/d to 1.0 when the value for d1 is not known
for the mineral of interest is a convenient conservative
assumption under some circumstances. The calculated sample
mass becomes larger than the true value and may be quite
misleading for coarse and, in reality, poorly liberated solid
(d>>d1) which is being sampled.
Consider a chalcopyrite ore which has the following properties
(fully liberated (l = 1) with a top size of 200 μm and containing
two per cent chalcopyrite (a = 0.02)). The following calculations
can be performed using the previously described method (Gy,
1982), where the density of the chalcopyrite is 4.2 g/ml and the
density of the combined gangue is 2.6 g/ml, and assuming the
fundamental error for chalcopyrite does not exceed 0.06 per cent
chalcopyrite (sFSE = 0.0006).
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
c = (1-a)[(1-a)ρ1 + aρ2]/a
= (1 - 0.02)[(1 – 0.02) × 4.2 + 0.02 × 2.6]/0.02
= 204.2
C = clfg
= 204.2 × 1 × 0.5 × 0.25
= 25.5
ms = Cd3a2/ s2FSE
= 25.5 × (0.02)3 × (0.02)2/(0.0006)2
= 0.23 g
The calculation can be repeated for coarser streams with the
same particle properties. For streams with top sizes (d) of 0.2, 2.0
and 20 cm associated with grinding and crushing circuits, the
calculated values for ms increase to 23 g, 2269 g and 226 889 g
respectively (assuming d1 is 0.02 cm and letting parameter b
equal 1.0). For the coarser solid in grinding circuits and
particularly in crushing circuits, care is therefore required such
that sufficient sample mass is obtained to ensure a given
fundamental error is not exceeded. For each stream in crushing
and grinding circuits, the appropriate calculations for the sample
size must be performed using the real properties of the particles
in the population being sampled.
To comply with sample correctness as discussed earlier, one
aspect is that the width of the parallel opening of the sampler
must be commensurate with the top size of the particles in the
stream. Large particles must not have a lower probability than
smaller particles entering the sample because of interference with
their motion. Standards exist for the key dimensions of the
samplers in such cases. Further, the samplers must be emptied in
a way that ensures all particles irrespective of their size or density
are removed from the sampling device or system and continue to
be part of the sample.
The equation for ms can be expressed with the fundamental
error as the unknown. Holmes (1992) noted that:
… a fundamental characteristic of s2FSE is that it
diminishes very quickly when d is reduced and not
so quickly when ms is increased. However, it can
never be eliminated no matter what crushing and
homogenisation procedures are used, although for
fine flotation concentrates it is negligible when the
sample mass exceeds about 100 g.
In collection of samples from a flotation circuit with several
rounds of sampling and using conventional types of samplers, the
resulting sample mass is usually unavoidably in the region of
500 g (or larger). Hence, as dictated by the use of conventional
sampling devices in a flotation circuit, the sample mass tends to
be in a region where the value for the fundamental error is at
acceptable low values.
To obtain an assay for a sample or a size fraction, a very small
mass (typically 0.25 g to 1 g) has to be sampled from a much
larger mass of dry solid in the laboratory. The fundamental error
for this step can also be calculated using the equation from Gy
(1982) as described earlier. If the fundamental error is
unacceptably high, the sample or size fraction can be pulverised
to lower the top size, and greatly lower the mass of sample
providing an acceptable fundamental error. Of course, if a
property is to be measured which is affected by size reduction (eg
liberation state), this approach cannot be taken.
For a particular operation, it is clearly more effective to obtain
a plant feed sample from the flotation circuit feed, rather than the
coarser grinding circuit feed or the even coarser crushing circuit
feed. For a given sampling error, the minimum mass of the
61
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
required sample is much smaller for the flotation circuit feed and
cost savings result from handling of the smaller sample, aside
from the convenience. In collecting a sample, a large number of
frequent correctly executed sampling events with collection of the
same small sample weight in each sampling event is superior to a
small number of infrequent correctly executed sampling events
which produces the same final sample weight.
Some additional sources of information on sampling principles
can be accessed (Weiss, 1985; Birnbaum, 1992).
Lip sampling
There are various types of data which can be collected in the lip
sampling procedure:
1. a reliable sizing and assay of the solid being recovered at
the lip,
2. a reliable value for the per cent solid of the pulp being
recovered at the lip, and
3. an estimate of the solid flow rate passing over the lip.
To satisfy requirements 1 and 2, the sampler should be moved
at right angles to the lip across the full width of the cell lip at
constant speed to intercept the discharging froth. The speed
should be selected to ensure that overflowing of the sampler is
not possible and to ensure that uninhibited entry of the
discharging froth into the sampler is possible during the
sampling. Such lip samplers have typically the following
dimensions: width 9 cm, height 30 cm and length 20 cm. The
method was described by Restarick (1976).
To satisfy requirement 3, the time for which the sampler was
intercepting the stream has to be recorded. Therefore, the lip
sampling procedure is performed using a crew of two people with
one recording the sampling times. The following points can be
noted:
• The requirements for the starting and finishing points in the
movement of the lip sampler are illustrated in Figure A1.
These starting and finishing points arise because of the
requirement for determination of the solid flow rate per unit
width of the sampler as expressed in Equation A1, ie the full
width of the sampler must be collecting sample during all
the timing period. (This requirement differs subtly from the
correct sampling motion of starting and finishing the
sampler’s motion outside the flowing concentrate.)
Lip Width
Cell
Cell Discharge Lip
Starting Finishing
Position Position
Cutter Width
Total weight ( g ) 3600 Lip width (mm)
TPH Solid = × × (A1)
Total time (sec) 10 6 Cutter width (mm)
FIG A1 - Illustration of the lip sampling technique as described in
the appendix for one cell. The lip sampler is positioned under the
lip from which concentrate is discharging in this plan view.
62 Spectrum Series 16
• The time taken for the sampler to intercept the concentrate
stream has to be approximately the same for each traverse to
obtain reliable estimates of the solid flow rate.
• For a group of cells being sampled together, the time taken
for the sampler to intercept the concentrate stream for each
cell also has to be approximately the same for each traverse
of each cell. Initially, all the cells in the grouping have to be
considered in selection of the required time for the group of
cells.
For obtaining the solid flow rate, the details of the correct lip
sampling procedure for a single cell are illustrated in Figure A1
and the equation for conversion of the measurements of the total
mass of collected solid and the total sampling time (for all
traverses on the various rounds of sampling) is provided. The
total mass of collected solid divided by the total sampling time is
effectively the observed flow rate of solid per unit width of the
sampler. Movement of the sampler across the cell width at fixed
velocity allows this value to be averaged across the entire
concentrate flow. In the equation, the ratio lip width/cutter width
then allows scaling up of the observed flow rate of solid per unit
width of the sampler to the full width of the cell lip.
To this point, all the discussion of lip sampling to obtain the
solid flow rate has been based on movement of the lip sampler
across the cell lip at right-angles at fixed speed as indicated in
Figure A1. This movement of the sampler addresses any
variations in concentrate assay and flow rate across the full width
of the lip. For a cell in good mechanical condition, such
variations are usually minimal.
However, given the large size and shape of some cells, the
existence of floor grating over the top of the cells, the design and
maintenance of the cell lips or the existence of water pipes/sprays
or other obstacles in the launder, it is not possible to move the lip
sampler across the entire lip width or a portion of the lip. In this
situation, an estimate of the observed flow rate of solid per unit
width of the sampler can be obtained by placing the sampler in
one or more fixed positions where a portion of the cell lip is
accessible. The observed value can then be scaled up to the full
length of the cell lip.
Large tank cells can be one example of the described situation
for several reasons:
• the existence of relatively inaccessible external and internal
launders and sometimes cross launders, and
• the existence of floor grating over the top of cells with
possibly some small trapdoors for access.
For sampling a single cell, the discharge of concentrate into the
various types of launders may have to be treated as separate but
parallel sampling steps for that cell. It is also worth noting that,
depending on the overall sampling scheme, the objective from
sampling a single unit cell may be a sample of the combined
concentrate for analysis without the need for an estimate of the
solid flow rate. While such an objective appears less onerous,
timing of the concentrate collection with the sampler in various
fixed positions may still be required.
For example, taking a cell with one internal launder and one
external launder, if the concentrate discharging to the internal
launder happened to be at a higher flow rate per unit length of the
lip than for the external launder, its assay may differ from the
concentrate at an external launder for the same cell. To obtain the
assay of the combined concentrate for the entire cell, two
approaches could be taken:
1. The same lip sampling time for the concentrate discharging
into the internal and external launders could be used as the
basis for a concentrate sample representing the entire
concentrate.
Flotation Plant Optimisation

2. The assays for the solid at the internal and external launders
along with the flow rate estimate of each could be used to
calculate the assay for the combined concentrate. (Clearly, it
is preferable that the plant designer provides a sampling point
to enable sampling of the entire concentrate from all launders
at a single point, to obtain the overall assay of the stream.)
It can be noted that timing and recording of the lip sampling
times would assist in correct execution of approach 1 even though
obtaining a value for the lip tonnage is not the objective. Further,
timing and recording of the lip sampling times at the internal and
external launders is an integral part of execution of approach 2.
REFERENCES
Birnbaum, P M, 1992. An evaluation of sampling errors in a mineral
concentrator, in Proceedings Sampling Practices in the Mineral
Industry, pp 49-58 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 2 – EXISTING METHODS FOR PROCESS ANALYSIS
François-Bongarçon, D and Gy, P, 2002a. Critical aspects of sampling in
mills and plants: A guide to understanding sampling audits, J S Afr
Inst Min Metal, Nov/Dec, pp 481-484.
François-Bongarçon, D and Gy, P, 2002b. The most common error in
applying ‘Gy’s Formula’ in the theory of mineral sampling, and the
history of the liberation factor, J S Afr Inst Min Metal, Nov/Dec,
pp 475-479.
Gy, P M, 1982. Sampling of Particulate Materials – Theory and Practice,
second edition (Elsevier: Amsterdam).
Holmes, R J, 1992. Sampling of mineral process streams, in Proceedings
Sampling Practices in the Mineral Industry, pp 33-37 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Restarick, C J, 1976. Pulp sampling techniques for steady state
assessment of mineral concentrators, in Proceedings Sampling
Symposium, pp 161-168 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
Weiss, N L, 1985. SME Mineral Processing Handbook, (Society of
Mining Engineers, AIME: New York).
63
 HOME

CHAPTER 3

Mass Balancing Flotation Data

Rob Morrison

MAusIMM, Chief Technologist, Julius Kruttschnitt Mineral Research Centre, 40 Isles Road,
Indooroopilly Qld 4068. Email: R.Morrison@uq.edu.au

Rob is currently Chief Technologist at the JKMRC with responsiblity for technical oversight of the research
activities of the centre and for promoting the transfer of technology from research to commercial application.
His experience includes operations and process development during four years at Bougainville Copper and
plant design, construction and commissioning during six years with Fluor Daniel Australia. Rob managed
JKTech for nine years. He led Program 2 of the Centre for Sustainable Resource Processing for its first two
years. Rob was also the leader of the AMIRA Metal Accounting project P754, which developed a Code of
Practice and a text book. Technical interests include energy efficient mineral processing, simulation, mass
balancing and measurement techniques. Rob has published more than 80 technical papers, contributed to
several text books and holds several patents.

Abstract
Introduction
Accuracy Considerations
The Simplest Case
The Method of ‘Mass Flow Errors’
An Analytical Solution
Estimating the Accuracy of the Flow Split
The Monte Carlo Approach
Generating Numerically ‘Exact’ Data for Further Analysis
A First Pass at a More General Solution
Elements Versus Minerals
Practical Application – Single Level Balancing
Selection of Measurement Points
Analysis of More Complex Circuits
Multi-Dimensional Balances (Size by Assay)
Size by Assay Example
Recap
Conclusions
Acknowledgements
References
Further Reading
Useful Websites

ABSTRACT In general, these measured data will not be consistent around any
separation unit or junction within the circuit. A fair assessment of circuit
A flotation circuit survey will typically produce a substantial volume of performance requires a set of numerically consistent data. Otherwise key
data – usually as metal assays and flow rates. performance indicators (KPIs), such as recovery in each section, will

Flotation Plant Optimisation Spectrum Series 16 65

CHAPTER 3 – MASS BALANCING FLOTATION DATA
depend of which measured values are chosen as the basis of the
calculation.
Generating a set of self consistent data provides a basis for a single set
of KPIs. The set of self consistent data should in some way be a ‘best
estimate’ of the state of the circuit during the survey.
This chapter discusses strategies for selection of flow measurement and
sampling points. Various ways of estimating measurement accuracy are
considered. Lastly, some calculation methods are described. These are
supported by some simple examples.
INTRODUCTION
The basis of mass balancing is that all measurements are subject
to statistical variation. If we could make each measurement
(sample, assay or flow rate) many times, the results would have a
spread of values. In practice, we can typically only afford to take
a few replicate samples at each measurement point. If we have
some knowledge of the degree of variation expected in each
measurement, we can analyse all of the data together to try to find
a best estimate of the mass balance. Data which is self consistent
with this mass balance can be used to assess and compare
performance and as the basis of mathematical models of the
process.
The mass balancing process can also incorporate redundant
data to produce not only mass split factors but a further estimate
of how well that mass split is defined. The ‘traditional’ approach,
using a two product formula based on a single assay, will also
produce an estimate of the mass split. However, it provides no
estimate of the accuracy of that mass split and therefore provides
at best a poor basis for decision-making.
Note that these techniques are not at all new. The first general
purpose system is due to Weigel (1972) who developed the
original MATBAL code. This chapter focuses on how to attack
the simpler cases using spreadsheet technology.
The other key factor is that the flow sheet being balanced must
be operating in reasonable ‘balance’ during the measurement
period. That is, operating at as close as possible to steady state for
both flow rates and separation processes. Otherwise the
fundamental assumption that ‘what goes in is equal to what goes
out’ is not justified.
Objectives
• We wish to produce sets of self consistent data to suit a range
of measurement strategies to characterise the performance of
flotation circuit or a part of it, and
• we wish to generate estimates of the self consistency of the
data and of the accuracy of the flow rate measurements.
For the general case we can express these objectives in
mathematical terms as three types of data:
• x i is a measurement of some sort – assay, flow rate, size
fraction, liberated mineral fraction, whatever. These values
are generally not self consistent.
• x ∗ is the true value of the measured quantity. We can
i
estimate this value at various levels of accuracy but can
never know it exactly. These values are self consistent.
• x$ i is an adjusted value of xi which satisfies all of the
constraints – that is – is self consistent and is in some way a
best estimate of the true value.
Hence we can define an adjustment Δ i of each measurement as:
Δ i = ( x i − x$ i )
The simplest way to find a ‘best’ set of adjusted data is to find
strategy which minimises the sum of squares (SSQ) of the
66 Spectrum Series 16
adjustments. However, this strategy does not take any account
of how well (or badly) a measurement may be defined. The
estimated or measured standard deviation σi provides us with an
estimate of how well each piece of data is defined. Hence, a
better criterion for minimisation is a sum of squares where each
adjustment is divided (or weighted) by our estimate of the
standard deviation of its associated measurement.
SSQ = ∑ (Δ i )2
i
The weighted sum of squares (WSSQ) is the sum of squares of
the adjustments where ‘weighted’ means that the adjustment is
scaled in terms of the expected variation of that data.
2
⎛Δ ⎞
WSSQ = ∑ ⎜ i ⎟
i ⎝ σi⎠
It is also necessary that the adjusted values satisfy all flow
sheet and summation constraints. That is, what goes in is equal to
what comes out and the various kinds of assays and subassays
add up to one or to the assay in the next level of measurements.
ACCURACY CONSIDERATIONS
We can estimate the standard deviation (‘sd’ or error for short) σi
from experience or from repeated sampling and assaying or
repeated measurements against a known standard. In practise this
means taking five to ten replicate samples at several key points
within a circuit and subjecting all of them to the same sample
preparation and assaying process which you intend to use in the
actual test work. The formula for sampling variance or standard
error then provides an estimate of standard deviation at each
measurement point:
σ 2x = ∑ (x i - x)2 / ( n − 1)
i
x = ∑ x i / n – the mean of the n measurements.
i
Strictly speaking, the standard deviation refers to the complete
distribution (which we can never measure completely) while the
standard error is the same property (the square root of the
variance) of the sampling distribution – which we can measure.
However, the two terms are often used interchangeably.
There are some suggestions in the examples for sd estimation
based on experience, but it is usually worth doing some
repeatability testing to obtain actual estimates.
In general this does not mean doing a full sampling tree (AS
28841 precision and bias of mineral sands measurement) to split
the errors into their components. If you cannot reduce process
and assaying variability to less than a few per cent (relative),
there is very little point in carrying out detailed test work as the
results will mean very little.
If in general the measurement variations are small, then the
required adjustments will also be small and be drawn from the
same population of differences. Therefore, we expect the average
value of ( x i − x i ) / σi to be about equal to one and, as a
consequence, the WSSQ to be roughly similar to the number of
measurements.
However, each time we apply a flow sheet or a summation
constraint we lose one degree of freedom. This reduces the
expected value of the WSSQ by one.
Hence we now have a quite general way of looking at data
quality. If the WSSQ is of the same order as the number of
measurements, our data is likely to be suitable for further
analysis.
A note of caution: if you have taken multiple data sets, a few of
these may balance well by pure chance and still be nonsense.
Flotation Plant Optimisation
 CHAPTER 3 – MASS BALANCING FLOTATION DATA

Balances which are self consistent by chance are rarely sensible
in terms of other criteria such as the size by recovery response.
The WSSQ can also be used to provide an estimate of the
standard deviation of each calculated result x$ i. If a small change
in x i (or of a flow rate estimate) causes a large change in the sum
of squares, then x$ i is well defined. If it causes no change, then it
is not defined at all.
This property is particularly useful for checking whether
calculated flow rates are well defined.
The formal name for this rule is the ‘propagation of variance’.
An alternative approach is to apply some artificial (random)
variation to each measured value and solve the problem many
times. This is called a Monte Carlo method. The variation in
results provides an estimate of the variation which might be
expected from a single experiment and analysis if repeated many
times.
This overall approach is usually called the ‘minimisation of
weighted squared errors’ or simply ‘least squares’. If for some
reason the measured values are drawn from a highly asymmetric
probability distribution, outlying values may cause biases in the
calculations. An alternative approach to finding x$ i is to use what
are called ‘maximum likelihood methods’. For measurement and
analysis of flotation data, the probability distributions are
sufficiently symmetrical for least squares methods to be quite
adequate.
THE SIMPLEST CASE
The simplest case is also the most common and warrants quite
extensive consideration.
Consider a single separator (or node) which has a feed stream
of flow rate A, a product stream of flow rate B and a reject stream
of flow rate C.
For the simplest possible solution we would like to solve for a
single variable and the best way to do this is to consider the flow
split as the ratio of B/A flowing into product stream B – which
we will call ‘beta’ or β.
B* = β * A * and C * = (1 − β *) A ∗
0 = ai∗ − β * bi∗ − (1 − β ∗ )ci∗
β* = ( ai∗ − ci∗ ) / ( bi∗ − ci∗ )
This equation for β* offers the useful insight that – even with
perfect data – this method is not going to work for splitters where
we can expect:
ai = bi = ci
to within measurement variation.
For this case β* is undefined as zero/zero, which is not zero but
undefined. Beware of flow split estimates which are only based
on experimental noise.
If the flow split is defined to any reasonable degree, then the
splitter is functioning as some kind of separator – which is not
usually desirable.
This general approach also works for measured data and
provides a useful way to do an initial evaluation of a set of data.
We can set up the preliminary balance around our separator in
MS Excel following the general structure shown in Table 1.
A spreadsheet containing some real data from a section of a
flotation circuit is shown in Table 2.
TABLE 1
Example of a preliminary balance set up around the separator
in MS Excel.

A B Feed Product Reject (ai - ci) (bi - ci) Beta

assay assay assay (ai - ci)/
(bi -ci )
A B C
Flows
C
Assayi
FIG 1 - The simplest case is a single node which represents Assay 1 ' ' '
any process with one input and two outputs.
Assay 2 ' ' '
Assay 3 ' ' '
This single node case might represent a complete flotation
circuit or a single flotation cell. Each stream is sampled and …
measured for i assays. Assay n
For stream A we have assays a1, a2 .. …an. For stream B we
have assays b1, b2.. …bn and so on.
From the measured flow rate of copper cleaner concentrate, the
Note that these assays may be any self consistent, additive
mass split to concentrate (beta) should be about one per cent.
property – size or specific gravity fractions can also be used. Per
Note that 890 kg/h is not too difficult to measure as the flow is
cent solids is not suitable as an assay but can be used if converted
to per cent water. This is because per cent solids is based on the only about 15 kg/min.
total stream flow of ore and water while the other assays are The estimates of mass split due to the change in each assay
percentages of the solid phase only. vary from 30 per cent to -66.7 per cent, but perfect data should
We also have some knowledge of (by repeated measurements generate identical values which would be the same as the perfect
or experience) the standard deviations for flow rates (σA, σB, σC) mass split.
and assays σa1, σb1, σc1 and so on. The calculated mass split based on the copper assays is almost
We could tackle the problem directly but there are advantages exactly one per cent. The split based on the silver assay is 1.5 per
in considering flow splits first and then looking at assay data cent. In each of these cases, the process strongly concentrates the
adjustment. mineral containing that element.
For the ‘true’ data (designated by a *): Our estimate of beta is the ratio of two differences. If these
differences are small compared with the accuracy of
A ∗ = B ∗ + C ∗ and A ∗ ai∗ = B ∗ bi∗ + C ∗ ci∗ for all of i.
measurement, then the ratio will be poorly defined – as it is here.
where: If the differences are large compared with the accuracy of
A, B and C are the total solids flow rate in each stream sampling and assay, then the ratio should be well defined. This is
almost certainly not the case for the lead assays, where both
ai, bi and ci refer to a series of assays in each of those streams concentrate and tailing assays exceed the feed value.

Flotation Plant Optimisation Spectrum Series 16 67

CHAPTER 3 – MASS BALANCING FLOTATION DATA

TABLE 2
Spreadsheet containing some real data from a section of a flotation circuit.

Stream Float feed (A) Cu clnr conc (B) Cu rghr tail (C)
Measured t/h 87.90 0.89
Relative flow 1 Beta (1-Beta) Beta for each
assay
ai bi ci (ai - ci) (bi - ci) (ai - ci)/(bi - ci)
Measured Pb (%) 4.02 4.07 4.04 -0.02 0.03 -0.667
Measured Zn (%) 15.80 7.80 15.50 0.30 -7.70 -0.039
Measured Cu (%) 0.43 22.40 0.20 0.23 22.20 0.010
Measured Fe (%) 14.30 16.10 13.50 0.80 2.60 0.308
Measured Au (g/t) 1.05 47.50 0.94 0.11 46.56 0.002
Measured Ag (g/t) 142 3023 97 45.00 2926.00 0.015

It is also clear that zinc is being rejected from the concentrate
and that the zinc grade of the tailings is apparently reduced as a
result. A more reasonable possibility is that the zinc assay of the
tails is higher than that in the feed, but our measurement accuracy
is insufficient to demonstrate this effect.
The absolute value of an assay is also important. One gram per
tonne is 0.0001 per cent or one per cent is 10 000 ppm. Hence the
silver assays are really 0.0142 per cent, 0.3032 per cent and
0.0097 per cent.
The gold assays are two orders of magnitude smaller again and
the mass split of 0.2 per cent is a reflection of good sampling and
assaying at those concentrations.
Error models
There are some useful generic ways to ‘estimate’ likely
measurement errors based on assay magnitude. The most popular
method is to assume a constant relative (or per cent) error. This
approach implies that the best defined values would be the
tailings assays for gold and silver and is plainly nonsense.
The next commonly used approach is to assume that the
expected error is constant or (usually) one. This simplifies the
arithmetic and is a more sensible estimate. The larger assays are
now assumed to have the smallest relative error.
base metals in the per cent ranges would be at ppm values. The
reason for this is that the assaying technique itself uses some
preconcentration before measurement.
THE METHOD OF ‘MASS FLOW ERRORS’
If we switch to measured values in the component balance
equation, we can expect a measured ‘mass flow error’ for each
assay (or size fraction) measured:
Δ i = ai − βbi − (1 − β )ci
Note that only one β value is required for all of the n assays
and that Δ i should be zero for perfect data.
Hence, a process which makes a large difference to the stream
assays will have a better defined flow split than one which only
makes a small difference, such as a final cleaner or scavenger
bank.
Similarly if a splitter is working well, it is essential to measure
or estimate its flow split as mass balancing will not be helpful.
Transposing and simplifying we get for each component:
Δ i = ( ai − ci ) − β ( bi − ci )
If we square both sides and sum up all of the components:
SSQ = ∑ Δ2i = [( ai − ci ) − β ( bi − ci )]
2

The JKMRC approach is to measure the error distribution i

through repeated sampling or to use a well-proven heuristic – that
is – a rule which should behave in a sensible manner. This rule is
often attributed to Bill Whiten – who disclaims ownership.
We assume that for assays (or size fractions) in the per cent
range that the minimum sd is 0.1 per cent (absolute) and that for
assays greater than nine per cent the absolute sd is one per cent.
In between those two limits, the sd is 0.1 plus 0.1 times the
assay value.
Hence, the relative error at a measured value of 0.1 per cent is
200 per cent. The relative error at an assay value of ten per cent is
also ten per cent.
At a measured value of 0.5 per cent, the absolute error is 0.1
plus 0.05 per cent or 30 per cent relative.
Some people refer to this as a ‘two term’ error model where 0.1
is the fixed error and 0.1 is the fractional error between zero and
nine per cent.
Another term is a ‘one over x’ model because the relative error
curve has much the same shape as that function.
If the analysis method used is very different, we can expect the
error models to change. For example, the fire assays typically
used for precious metals should be intrinsically more accurate at
ppm values than standard techniques such as XRF and AAs for
We might reasonably expect the value of β, which minimises
SSQ, to be a reasonable estimate of β* which we can call β.
As for the first case, this can use a quite generic spreadsheet
format as shown in Table 3. Set up this spreadsheet as shown in
Table 4. Then work through it with estimates of β of 0, 0.005,
0.01, 0.02, 0.1, 1.0, as we know from the measured flow rate that
the mass split is about one per cent.
The gold assays have been divided by 100 and the silver assays
by 1000 to bring them closer to per cent values. (As an exercise,
check out the minima with the unadjusted values.)
If your set-up is correct, the SSQ should go from 0.783 at 0.0
to 550 at 1.0 with a minimum value of 0.742 at 0.01.
It is also a useful exercise to tabulate β and plot it against SSQ
as shown in Figure 2. This graph shows a well-defined minima at
β is approximately 0.01. It is well defined because making a
small change (±0.005) in β makes a ±10 per cent change in the
SSQ. However, it is necessary to plot the log of the SSQ to be
able to see the minima.
Go through the β sequence again and watch how the SSQ or
(Δ i )2 for each assay varies.
They do not all have a minima at the same value of β.

68 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 3 – MASS BALANCING FLOTATION DATA

TABLE 3
Generic structure of a spreadsheet balance around a single node using the method of mass flow errors.

Estimated beta Initial estimate

Feed assay Product assay Reject assay (ai - ci) *(bi - ci) 2
i i
Measured A B C
Flows
Assayi
Assay 1 ' ' '
Assay 2 ' ' '
Assay 3 ' ' '
Assay 4
…
Assay n
Total SSQ =

value of the flow split. In this case, there is little room for debate
100 about the value of the flow split.
In general, it will be more useful to derive an estimate of the
standard deviation of the flow split error based on our measured
standard deviations. This estimate can be used to ‘weight’ the
10
Log of SSQ

contribution of each flow split error to the sum of squares which

is to be minimised.
SSQ
Δi
1 σΔi = [ a i − βbi − (1 − β )ci ] σΔi
0 0.05 0.1 0.15 0.2 0.25 0.3
The rule for ‘propagation of variance’ is that the variance of a
function is the sum of the product of the variance of each input
0.1 parameter and the square of its partial derivative – see Deming
Estimated Beta (1938) or almost any statistics reference.
Dropping the i for the moment:
FIG 2 - Log of the sum of squared errors at a range of estimates 2 2 2
⎡ ∂Δ ⎤ ⎡ ∂Δ ⎤ ⎡ ∂Δ ⎤
of the mass split – beta. σ 2Δ= ⎢ ⎥ ( σ 2a ) + ⎢ ⎥ ( σ 2b ) + ⎢ ⎥ ( σ 2c )
⎣ ∂a ⎦ ⎣ ∂b ⎦ ⎣ ∂c ⎦
= σ 2a + β 2 σ b2 + (1 − β )2 σ c2
TABLE 4
An example of the method of mass flow errors. If we add some additional columns to our spreadsheet, we can
Est beta = 0.0036 A B C derive a weighted sum of squares WTDSSQ of the mass flow
errors.
Stream Float Cu clnr Cu rghr
feed conc tail
Measured t/h 87.90 0.89 TABLE 5
Additional columns required for Table 3 to include a
Relative 1.000 0.004 0.996 Delta Delta
weighted sum of squares.
flow squared
ai Bi ci ai - B* a b c (1 - )2 Δ2i
bi-(1-B)
*ci
Measured Pb (%) 4.02 4.07 4.04 -0.020 0.000
WSSQ =
Measured Zn (%) 15.80 7.80 15.50 0.328 0.108
Measured Cu (%) 0.43 22.40 0.20 0.149 0.022 Note: WSSQ = weighted sum of squares.
Measured Fe (%) 14.30 16.10 13.50 0.790 0.625
Measured Au (g/t) 0.0105 0.475 0.0094 -0.012 0.000 As before, run through in increments of 0.001 for β from 0.00
Measured Ag (g/t) 0.142 3.023 0.097 0.020 0.000 to 0.02 and watch how the minima change. For the first case,
assume the sd of each assay is one. For the second case assume
SSQ = 0.759
they are each five per cent of the assay value. For the third case,
assume ten per cent. If your spreadsheet is right, the unit
How should we decide which value of β to use? Obviously, the weighting case will look quite like the original but be driven by
value which has the most accurately measured data should the larger assay values. For the second and third cases, the
dominate. smallest assays will dominate. Neither seems very sensible and
Let us first consider the case where each of the minima is quite we will consider a better balanced approach in the section on
similar. Clearly the weighting will have very little effect on the error models.

Flotation Plant Optimisation Spectrum Series 16 69

CHAPTER 3 – MASS BALANCING FLOTATION DATA

Arriving at a single numerical solution The accuracy with which the flow split is defined depends on
how rapidly the SSQ increases as we move away (in either
For the examples in MS Excel, the built-in minimiser called direction) from the best fit value of beta. Once again from high
Solver provides a very simple solution. school calculus, we recall that the curvature of a function is its
Designate SSQ (or WTDSSQ) as the target for minimisation second derivative:
and beta as the variable which can be adjusted. Run Solver.
∂SSQ
The result should be instantaneous. = ∑ (( ai − ci )( b − ci )) − ∑ β ( bi − ci )2
∂β i i

AN ANALYTICAL SOLUTION ∂ 2 SSQ

= − ∑( bi − ci )2
In the prehistoric days before PCs and Solver, a simple analytical ∂β 2 i
solution was very useful.
Hence:
Recalling some high school calculus, if we wish to find the
minimum (or maximum) of a function, we take the first derivative σ 2β = σ 2Δ ∑( bi − ci )2
with respect to the variable of interest and set the result to zero. i

SSQ = ∑ [( ai − ci ) − β ( bi − ci )]
2
In the Excel case, Solver does not provide this value for
i parameters but we can easily change beta by ±1 per cent and note
the change on the sum of squares to test for a well-defined value.
dSSQ
= 2 ∑ i [( ai − ci ) − β ( bi − ci )]( bi − ci )
dβ
THE MONTE CARLO APPROACH
0=∑ [( a i − ci )( bi − ci ) − β ( bi − ci ) 2
] An alternative approach is to add some random variation to the
i input data, run Solver (or better our analytical formula) many

β=
∑ ( a − c )( b − c )
i i i i
times and consider the results as a mean and standard deviation
for beta. There are some specialist add-ons for MS Excel for this
∑ (b − c ) i i
2
purpose. However, for a simple case like this, a Monte Carlo
investigation is not too difficult.
This is an interesting result.
You can set up the process as a ‘Macro’ (or a Visual Basic
When ai = bi = ci, it is also undefined. The assays with the Application program) or set the spreadsheet to manual calculation
largest difference will dominate each sum. Hence the target of the and do a little manual accumulation. The latter is probably more
separation process (for example, copper in a copper circuit) useful for understanding the process. You can automate it with a
provides the best defined balance. macro as an exercise.
As for the earlier cases, this approach is easy to set up in a
We will use the same original data block and try the formula
spreadsheet. We already have columns of (ai - bi) and (bi - ci) in
solution method.
the first example. Hence we only need to add a column each for
their product and the second term squared. The Excel function RAND() draws a value from a calculated
random distribution between zero and one. We can scale the
Divide the sums of these columns for an analytical estimate of
beta as shown in Table 6. range of values between ‘min’ and ‘max’ by multiplying by
((max-min) + min). This generates a uniform distribution of
random values between min and max.
ESTIMATING THE ACCURACY OF THE However, what we actually need is a random set of variations
FLOW SPLIT drawn from a probability distribution similar to the distribution
We can also estimate the variance of the residual errors at the best we were sampling by making measurements or taking samples.
fit value of beta. This is a little trickier than it sounds at first.
If we consider the cumulative (or integral) form of any
σ 2Δ = ∑ Δ2i / ( n − 1) probability distribution, by definition, it will start at zero and end
i

TABLE 6
Spreadsheet balance including the analytical solution for the ‘best fit’ mass split.

Data A B C
Flotation feed Cu cleaner conc Cu rougher tail
t/h 87.90 0.89 Analytical Estimate
1 0.010 0.990
ai bi ci (ai - ci) (bi - ci) (ai - ci)*(bi - ci) (bi - ci)2
Pb (%) 4.02 4.07 4.04 -0.02 0.03 -0.001 0.001
Zn (%) 15.80 7.80 15.50 0.30 -7.70 -2.310 59.290
Cu (%) 0.43 22.40 0.20 0.23 22.20 5.106 492.840
Fe (%) 14.30 16.10 13.50 0.80 2.60 2.080 6.760
Au (g/t)/100 0.001 0.475 0.009 -0.01 0.47 -0.004 0.217
Ag (g/t)/1000 0.142 3.023 0.097 0.05 2.93 0.132 8.561
Sum 5.003 567.669
Beta 0.008813553

70 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 3 – MASS BALANCING FLOTATION DATA

at one. Hence we need the value of our cumulative distribution
which corresponds to each random value between zero and one.
This is the inverse of the probability distribution.
This will work well most of the time but it is worth
remembering that the normal distribution extends from negative
to positive infinity. Most chemists will not accept this range of
variation and will simply not report numbers which are
statistically possible but not plausible. If we reduce the range of
possible random values to say 0.025 to 0.975, then the inverse
values will lie between plus and minus two standard deviations of
the mean value.
To test out this process, set up four blocks of similar structure
as shown in Table 7. The first one is the measured data block.
The second one is the block of estimated standard deviations.
The third one is our scaled random number. The last one is the
block of perturbed data which we can use to rerun our balance
(and other calculations) to generate an estimated range of
variation.
Each time the sheet is executed (press F9 to force a
recalculation), this process will generate an equally likely
‘plausible’ set of synthetic data, which our formula will then
solve for beta.
After each run, jot down resulting beta. If you edit the
spreadsheet, it will generate another value unless you switch to
manual recalculation (Tools/Options/Manual Recalculation).
In addition to the scaled RAND() function, you will need to
use the INVNORMAL (inverse x, mean of x and sd) function to
generate a deviation to add to the measured value. An example is
shown in Table 7.
Do as many runs as you like – but no less than ten. Now use the
mean and sd functions (MEAN(Range) and STDEV(Range))
with your row of β values as the argument. If the mean is very
different from your original β, $ there is a problem in the
spreadsheet. The sd of the synthetic betas provides an estimate of
the sd of the calculated beta. Some mass balance programs only
provide this way of estimating how well the balance is defined for
the balanced flow rates.
GENERATING NUMERICALLY ‘EXACT’ DATA
FOR FURTHER ANALYSIS
Given that we now have estimates of the flow split and its
accuracy, we can consider various ways to generate an ‘exact’
data set.

TABLE 7
Spreadsheet example of a Monte Carlo analysis.

Measured Data A B C
Stream Flotation feed Cu cleaner conc Cu rougher tail
Measured t/h 87.90 0.89
Relative flow 1 0.010 0.990
ai bi ci
Measured Pb (%) 4.02 4.07 4.04
Measured Zn (%) 15.80 7.80 15.50
Measured Cu (%) 0.43 22.40 0.20
Measured Fe (%) 14.30 16.10 13.50
Measured Au (g/t)/100 0.001 0.475 0.009
Measured Ag (g/t)/1000 0.142 3.023 0.097

Standard deviation Model A B C

Stream Flotation feed Cu cleaner conc Cu rougher tail
Standard deviations t/h 4.40 0.04 5%
ai bi ci
Standard deviations Pb (%) 0.50 0.51 0.50 0.1 +10% Or 1 if > 9
Standard deviations Zn (%) 1.00 0.88 1.00 0.1 +10% Or 1 if > 9
Standard deviations Cu (%) 0.14 1.00 0.12 0.1 +10% Or 1 if > 9
Standard deviations Fe (%) 1.00 1.00 1.00 0.1 +10% Or 1 if > 9
Standard deviations Au (g/t)/100 0.050 0.050 0.050 0.05
Random values RAND – min 0.025 RAND – max 0.975
Est beta = 0.008713 A B C
Stream Flotation feed Cu cleaner conc Cu rougher tail
Measured t/h 0.25 0.28
ai bi Ci
Synthetic Pb (%) 0.47 0.30 0.84
Synthetic Zn (%) 0.44 0.94 0.82
Synthetic Cu (%) 0.79 0.84 0.97
Synthetic Fe (%) 0.07 0.54 0.53
Synthetic Au (g/t)/100 0.70 0.12 0.80
Synthetic Ag (g/t)/1000 0.95 0.06 0.41

Flotation Plant Optimisation Spectrum Series 16 71

CHAPTER 3 – MASS BALANCING FLOTATION DATA

TABLE 7 cont …
Synthetic Data A B C
Stream Flotation feed Cu cleaner conc Cu rougher tail
Measured t/h 80.67 0.90 Analytical Estimate
ai bi ci (ai - ci) (bi - ci) (ai - ci)*(bi - ci) (bi - ci)2
Synthetic Pb (%) 3.51 3.54 3.61 -0.10 -0.07 0.008 0.006
Synthetic Zn (%) 15.13 9.12 16.04 -0.91 -6.92 6.277 47.819
Synthetic Cu (%) 0.37 22.19 0.17 0.20 22.02 4.403 484.766
Synthetic Fe (%) 13.13 16.24 12.57 0.56 3.67 2.060 13.460
Synthetic Au (g/t)/100 0.02 0.52 0.00 0.02 0.52 0.010 0.267
Synthetic Ag (g/t)/1000 0.18 2.98 0.11 0.07 2.87 0.207 8.243
Sum 12.964 554.560
Beta 0.023377

Reconstitution The mathematically preferred approach is to minimise the

If the flow split is well defined, adding the reject and concentrate
together in the flow split ratio has much to recommend it. If the
weighted sum estimate of the feed assays is within ±1 standard
deviations of the feed measurement, this realisation will be quite
adequate for further analysis.
In practice, we can insert a single column to the right of the
feed column and add β times the concentrate plus (1 - β) times
the tailing assay.
Note that the summation constraints are satisfied.
required adjustment in some way – and the sum of squares is a
good general approach. Let Δa, Δb and Δc be the minimum
adjustments and omit the i – as we can consider each assay as a
separate case. Our adjusted data must balance if:
0 = ( a i − Δ a ) − β ( b − Δ b ) − (1 − β)( c − Δ c )
And be subject to the constraint that:
Δ = Δ a − βΔ b − (1 − β )Δ c
SSQ = (Δ a )2 + (Δ b )2 + (Δ c )2

TABLE 8 We can use a Lagrange multiplier λ to impose the constraint by

setting the constraint equation to zero and adding it to the SSQ.
A self consistent data set generated by reconstituting products
at the best fit mass split. This is called a ‘modified’ sum of squares. We then minimise
each adjustment and λ as well.
Data A B C
SSQ = (Δ a )2 + (Δ b )2 + (Δ c )2 − 2λ ( −Δ + Δ a − βΔb − (1 − β) Δ c )
Flotation Reconstituted Cu cleaner Cu rougher
feed feed conc tail Then:
t/h 87.90 0.89 ∂SSQ
= 2Δa − 2 λ = 0
Beta 1 0.0088 0.991 ∂Δ a
ai bi ci
∂SSQ
Pb (%) 4.02 4.040 4.07 4.04 = 2βΔb + 2 λβ = 0
∂Δ b
Zn (%) 15.80 15.432 7.80 15.50
∂SSQ
Cu (%) 0.43 0.396 22.40 0.20 = 2Δc + 2 λ(1 − β) = 0
∂Δ c
Fe (%) 14.30 13.523 16.10 13.50
Au (g/t)/100 0.001 0.014 0.475 0.009 ∂SSQ
= 2( −Δ + Δa − βΔb - (1 - β) Δc) = 0
Ag (g/t)/1000 0.142 0.123 3.023 0.097 ∂λ
We can drop all of the ‘2’s:
This strategy has the advantage that two thirds of our numbers
are exactly the ones that we measured. Δa = λ
Δb = −βλ
Data adjustment
Δc = −(1 − β )λ
Persons of a statistical bent will favour least squares adjustment
of all of the assays. This does lead us back towards the general and substitute into the last equation:
case, so it is useful as an exercise.
β = λ + β 2 λ + (1 − β)2 λ
We can start with the unit weighted case. Recall that:
Or:
Δ i = ai − βbi − (1 − β )ci
λ = Δ / (1 + β 2 + (1 + β)2 )
If we want an exact solution, we need some strategy to
apportion Δi across the measured assays. Strictly speaking, we As we already know the value of β and Δ for each assay, we
$ However, as we are going
should use the best fit value of beta – β. can easily calculate a set of adjustments.
to do a combined minimisation a little later using β$ it is probably For this case, the adjustment is proportional to the flow of that
confusing. component (assay) in each stream.
Some possible simple solutions are to divide delta by three or Table 9 shows the adjusted data and a check that it actually
to proportion delta according to the flow in each stream. does add up.

72 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 3 – MASS BALANCING FLOTATION DATA

TABLE 9
A self consistent set of adjusted data generated by minimising the sum of squares of the required adjustments.

Beta 0.008813553 A B C 1.982528

Stream Flotation feed Cu cleaner conc Cu rougher tail
Measured t/h 87.90 0.89
Relative flow 1 0.010 0.990 Check Beta
aI bi ci Delta Lambda Delta Check
Adjusted Pb (%) 4.03 4.07 4.03 -0.0203 -0.01 0 0.008814
Adjusted Zn (%) 15.61 7.80 15.68 0.3679 0.19 0 0.008814
Adjusted Cu (%) 0.41 22.40 0.22 0.0343 0.02 0 0.008814
Adjusted Fe (%) 13.91 16.10 13.89 0.7771 0.39 0 0.008814
Adjusted Au (g/t)/100 0.01 0.47 0.003 -0.0125 -0.01 0 0.008814
Adjusted Ag (g/t)/1000 0.13 3.02 0.11 0.0192 0.01 0 0.008814

Examining the adjusted values for lead and zinc shows that in As before, we know β and the sd for each assay.
this case it would be very difficult to distinguish the likely change Each set of assays has its own Lagrange multiplier – as shown
in the tailings assays as a result of the concentrate assays – that is in Table 10. The adjustments which are shown are based on the sd
– the difference imposed by the process. By way of contrast, estimates provided in the Monte Carlo example.
copper, iron and silver are all changed significantly.
Assuming there is a reasonable quantity of gangue, the total
If we wish to scale each adjustment by its own standard
assay summation constraint can be satisfied by difference for
deviation, the solution is quite similar (subject to the same
each stream. That is, work out the assays of interest and calculate
constraints as before):
the gangue by subtraction from 100 per cent.
WSSQ = (Δa / σa )2 + (Δb / σb )2 + (Δc / σc )2 It is interesting to note that even though the Lagrange
∂WSSQ multipliers have changed considerably; the adjusted data is
= 2Δa / σa − 2 λ = 0 or Δa + λσa essentially identical with the unweighted case.
∂Δ a
If there were real differences in the estimated flow splits (not
∂WSSQ 2Δb simply measurement noise), the assay with the smallest relative
= + 2 λβ = 0 or Δb + λβσb
∂Δ b σb sds would dominate the flow split.
∂WSSQ 2Δc A note on Lagrange multipliers: these are named after their
= + 2(1 − β )λ = 0 inventor. Possibly a more instructive name would be ‘unknown’
∂Δ c σc multipliers. We are seeking a solution where each of the first
∂WSSQ derivatives is equal to zero and the required constraints are also
= 2( −Δ − Δa − βΔb − (1 − β )Δc ) satisfied. Hence the multiplier is always one result from a set of
∂λ
simultaneous equations. The technique is very unusual in that it is
Δa = λσa easier to use than to understand intuitively. See Boas (1966) for a
clear and detailed explanation.
Δb = −βλσb
Δc = (1 − β )λσc A FIRST PASS AT A MORE GENERAL SOLUTION
Δ = λσ a + β λσ b + (1 − β) λσ 2
2 2
We can also see from the above analysis that a single flow split
λ = Δ /( σ a + β 2 σ b + (1 − β)2 σ c ) generates a full set of weighted assay adjustments.

TABLE 10
A self consistent set of adjusted data generated by minimising the sum of squares of the adjustments weighted by the SD
estimates from the Monte Carlo example.

Beta 0.010778355 A B C
Stream Flotation feed Cu cleaner conc Cu rougher tail
Measured t/h 87.90 0.89
Relative flow 1 0.010 0.990 Check Beta
ai bi ci Delta Lambda Delta Check
Weighted Pb (%) 4.03 4.07 4.03 -0.0203 -0.02 0 0.010778
adjustment
Weighted adj Zn (%) 15.61 7.80 15.69 0.3830 0.19 0 0.010778
Weighted adj Cu (%) 0.44 22.40 0.20 -0.0093 -0.04 0 0.010778
Weighted adj Fe (%) 13.91 16.10 13.89 0.7720 0.39 0 0.010778
Weighted adj Au (g/t)/100 0.01 0.47 0.003 -0.0134 -0.14 0 0.010778
Weighted adj Ag (g/t)/1000 0.14 3.02 0.10 0.0135 0.14 0 0.010778

Flotation Plant Optimisation Spectrum Series 16 73

CHAPTER 3 – MASS BALANCING FLOTATION DATA

The analytical solution is more complex and requires two steps
to get to an answer.
However, if we make the weighted sum of squares for all
assays the minimisation objective for Solver and let Solver adjust
the value of the flow split, we can achieve a simple version of the
general solution.
If you also have measured flow rates and estimates of their sds,
they can easily be added to this solution.
The weighted sum of squares of assay adjustments plus flow
rate adjustments now becomes the objective to minimise.
For this data set, the results are shown below. The flow split
estimate changes from 0.0088 to 0.0108 and the weighted SSQ
changes from 0.5138 to 0.5124. Whether either change represents
a ‘genuine’ improvement is not a very useful debate.
For reasonably self consistent data, the results of each of these
methods will be very similar. For less well-defined data, the
results will be dominated by the assays with the largest
differences (usually what the circuit is trying to separate) and the
smallest standard deviations.
ELEMENTS VERSUS MINERALS
The flotation process will usually be dependent on mineral
composition not assay composition.
If an element appears only in one mineral, it will not matter if
you use elements or minerals. For example, lead is often only
present in galena.
Many other cases are not so clear cut. For example, iron can be
present in chalcopyrite, pyrite, phyrrhotite and iron silicates – or
in varying proportions in sphalerite.
If we have consistent assay ratios, we can convert our assays
into equivalent minerals and use the mineral assays for balancing.
This will often give results which are better defined and easier
to understand – and are more suitable for modelling.
In this case, we might consider converting the copper to
chalcopyrite and the remaining iron to pyrite or phyrrhotite.
However, this task is left as an exercise.
In some cases, such as several species of copper minerals or a
variable Zn:Fe ratio in sphalerite, quantitative mineralogy will
become essential – each elemental assay will generate a flow split
estimate but they will not define mineral behaviour.
PRACTICAL APPLICATION – SINGLE LEVEL
BALANCING
Now that the various approaches to balances have been covered,
we can consider some common cases before starting on
multilevel balances.
Typical strategies for data collection are:
1. the complete circuit;
2. block level, eg cleaners or scavengers;
3. bank level – combined feed, concentrate(s) and reject; and
4. down-the-bank data.
Case 1 – The complete circuit can usually be considered as a
single two product separator. There will often be metal
accounting samplers available. Flow rates will often be known –
at least for the circuit feed rate – from a (hopefully) well-
calibrated weightometer.
Because the complete circuit maximises the differences in
assays, the balance should be well defined. If it is not, it will be
essential to refine your sampling and assaying techniques before
attempting more detailed test work.
Case 2 – Block level will often be covered by the two product
case. However, it is worth reviewing the feed, rougher, scavenger
case as it is also very common.

TABLE 11
A mass balance that combines the search for the best fit mass split with the minimised weighted assay adjustments.

Adjusted Beta 0.0107783 A B C

Weighted Stream Flotation feed Cu cleaner conc Cu rougher tail
Data Measured t/h 87.90 0.89
ai Bi ci Delta Lambda Delta
Pb (%) Wtd adj Pb (%) 4.03 4.07 4.03 -0.0203 -0.02 0
Zn (%) Wtd adj Zn (%) 15.61 7.80 15.69 0.3830 0.19 0
Cu (%) Wtd adj Cu (%) 0.44 22.40 0.20 -0.0093 -0.04 0
Fe (%) Wtd adj Fe (%) 13.91 16.10 13.89 0.7720 0.39 0
Au (g/t) Wtd adj Au (g/t)/100 0.01 0.47 0.003 -0.0134 -0.14 0
Ag (g/t) Wtd adj Ag (g/t)/1000 0.14 3.02 0.10 0.0135 0.14 0

Adjustments A B C
Stream Flotation feed Cu cleaner conc Cu rougher tail
Measured t/h 87.90 0.89 Weighted
ai bi ci SSQ
Pb (%) Wtd adj Pb (%) -0.02 0.000 0.02 0.0008
Zn (%) Wtd adj Zn (%) 0.19 -0.002 -0.19 0.0741
Cu (%) Wtd adj Cu (%) -0.04 0.000 0.04 0.0025
Fe (%) Wtd adj Fe (%) 0.39 -0.004 -0.39 0.3012
Au (g/t) Wtd adj Au (g/t)/100 -0.14 0.001 0.13 0.0361
Ag (g/t) Wtd adj Ag (g/t)/1000 0.14 -0.001 -0.13 0.0366
WTDSSQ 0.4514

74 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 3 – MASS BALANCING FLOTATION DATA

This case comes in two versions – one with four samples and That is, minimise the modified sum of squares:
one with five.
SSQ + 2 λ 1 (Δ1 − Δa − βΔb − (1 − β )Δe)
Each case does have an analytical solution but they can also be
set up as mass flow errors in each component for a straight- 2 λ 2 (Δ 2 − (1 − β ) Δe − (1 − β − γ )Δd − γΔc )
forward solution in MS Excel.
∂SSQ / ∂Δa = 2Δa − 2λ 1 = 0
For the first case we only have one node. We retain the idea
that beta is the flow split into stream B and add gamma for the / ∂Δb = 2Δb − 2λ 1β = 0
flow split into stream C, as shown in Figure 3: / ∂Δc = 2Δc + λ 2 2 γ = 0
Δ i = ai − βbi − γci − (1 − β − γ)di / ∂Δd = 2Δd − 2λ 1 (1 − β − γ) = 0
SSQ = ∑ Δ i
2
/ ∂Δe = 2Δe − 2λ 2 (1 − β) = 0
i
We can express λ 2 in terms of λ 1 :
A E D
λ 2 = (Δe − λ 1 (1 − β )) / (1 − β )
λ 2 = Δe / (1 − β ) − λ 1
B C and calculate both Lagrange multipliers. This gives a direct
solution for each adjustment. As before we can extend this to a
FIG 3 - A two node, rougher scavenger circuit. direct solution of the minimum sum of squared adjustments or of
weighted squared adjustments.
As stream E is not measured for this case, we can use Solver to The following set of data in Table 12 can be used as an
find ‘best fit’ values of β and γ. exercise.
Similarly, we can use Solver to find the minimum squared data
adjustments subject to the condition that the adjusted data add up TABLE 12
to zero at the specified values of β and γ. Measured assay data for the circuit shown in Figure 3.
As before, the adjustments can be weighted by their estimated
Stream ID Pb (%) Zn (%) Cu (%) Fe (%)
standard deviations. However, tackling the complete problem in
Solver will rapidly exceed the maximum number of constraints Meas Meas Meas Meas
assay assay assay assay
available with the add-in version. Various expanded versions are
available for purchase. Pb rougher feed A 11.89 24.02 0.25 11.14
Note: Pb rougher conc B 20.20 30.10 0.31 9.10

1. You will need at least two independent assays to solve this Pb rougher tail E 1.25 16.60 0.16 13.50
balance at all as there are two unknowns. Pb scavenger conc C 8.20 39.80 0.37 11.70

2. If stream C is a middlings stream with similar composition Pb scavenger tail D 0.51 14.60 0.14 13.70
to the feed stream, the flow rate in stream C is not defined.
Almost any flow measurement estimate (including a Case 3 – The ‘simplest case’ considered earlier is applicable to
‘calibrated eyeball’) will be more useful than the result of bank level sampling and analysis for combined feed, concentrate
the balance. and reject. A more detailed case is considered in the next case.
For the second configuration (in which E is measured) there are Case 4 – Down-the-bank data – For a more detailed analysis,
two separations or nodes. Hence we need two error equations: we may wish to sample the froth from each cell. We can either
take samples of the product of each cell and a pulp sample
Δ1i = ai − βbi − (1 − β)ei between each cell or take timed lip samples of each concentrate
Δ2 i = (1 − β)ei − γci − (1 − β − γ)di and a feed and a tailings sample. Hopefully a combined
concentrate will also be available.
[
SSQ = ∑ (Δ1i )2 + (Δ2 i )2 ] This problem provides a different kind of constraint. We have
i an estimate of the ratios of the flow rate from each cell but we do
This case is also quite straightforward to set up for Solver – not know the total flow rate.
with or without weighting the errors. We may or may not know the total assay of the concentrate.
As before, if your data is accurate and self consistent, there is An analytical solution for this multiconcentrate plus each cell
much to recommend simple reconstitution to generate a self tail was developed by the author and published in Chapter 7 of
consistent set of data for further analysis. Lynch (1977). It is outlined in Figure 4.
However, it is also fairly straightforward to do a least squares The data from that example are shown in Table 13.
adjustment one assay at a time.
A1 A2 A3 A4 A5
We will work through the second case as it has two constraints.
For each value of i, we wish to minimise:
SSQ = Δa 2 + Δb 2 + Δc 2 + Δd 2 + Δe 2
B1 B2 B3 B4
subject to:
Δ1 = Δa − βΔb − (1 − β) Δe
B5
and:
FIG 4 - A typical ‘Down-the-bank’ equivalent circuit for detailed
Δ 2 = (1 − β) Δe − γΔc − (1 − β − γ) Δd sampling (after Lynch, 1977, Figure 7.8).

Flotation Plant Optimisation Spectrum Series 16 75

CHAPTER 3 – MASS BALANCING FLOTATION DATA

TABLE 13
Measured assay data for the circuit shown in Figure 4 (after Lynch, 1977, Table 7-IV) .

A1 A2 A3 A4 A5 B1 B2 B3 B4 B5
Copper 3.03 1.0 1.15 0.75 0.52 21.2 22.5 17.1 9.7 15.5
Iron 8.2 7.1 6.4 6.0 5.7 23.8 25.9 23.4 20.2 21.4
Sulfur 6.5 5.3 4.4 3.8 3.3 26.5 29.2 25.9 20.8 23.0
Insoluble 46.6 49.0 51.0 51.8 52.8 14.4 11.4 18.4 27.6 21.6

The flow ratios to A2, A3 can be designated as alpha 2, alpha 3
and so on.
The solution is a somewhat complicated matrix inversion:
⎡m11 m12 0 0 ⎤ ⎡α 2 ⎤ ⎡x 1 ⎤
⎢m21 m22 m23 0 ⎥ ⎢α 3 ⎥ ⎢ 0 ⎥
⎢ ⎥⋅ ⎢ ⎥ = ⎢ ⎥
⎢ 0 m 32 m 33 m 34 ⎥ ⎢α 4 ⎥ ⎢ 0 ⎥
⎢⎣ 0 0 m43 m44 ⎥⎦ ⎢⎣α 5 ⎥⎦ ⎢⎣x 4 ⎥⎦
Each of the ‘m’ and ‘x’ terms is a sum of differences of assays.
When a circuit reaches this level of complexity, it is
recommended that the user switch to a more general approach or
to a commercial package.
The flow ratio case is better suited to a numerical approach.
If we have measured flow rates, fi, f2 →fm we can assume that
they will be related to the total functional flow rate by some ratio ρ.
Hence our node equation becomes:
Δ1 = a − βb − (1 − β )c
Δ2 = βb − ρ( f1 conc1 + f2 conc2 + f3 conc3 KK )
or if we cannot measure b, we simply substitute:
⎛ ⎞
ρ ⎜ ∑ f j conc j ⎟ * for βb
⎝ j ⎠
In either case, we are solving the original problem for the flow
weighted average of the concentrate assays and this will usually
be a good first step to see if the lip sample data are reasonably
self-consistent. If they are, weighted addition and reconciliation
should be adequate for further analysis. Alternatively, a similar
constrained least squares adjustment can be used – either
weighted or unweighted.
At this point, it is also worth commenting that several flow
sheet mass balancers (using these principles) are available
commercially (JKSimMet, JKMBal, JKSimFloat, MATBAL,
BilMat). Some web sites which provide more detail are
referenced at the end of this chapter. For more complex flow
sheets, an ‘off the shelf’ solution has much to recommend it – lest
all the time available for data analysis be spent trying to
manufacture a balance. A useful compromise is a spreadsheet
offered by Luttrell (2004).
It is also possible to use a general purpose, constrained
minimisation routine to solve the general problem, but the sheer
numbers of adjustments and constraints make this an ineffective
approach.
Dividing the problem into flow rates estimation followed by
assay adjustment makes it much more tractable. However, we do
need samples and assays from all streams to be able to use this
approach.
The most important step towards a general solution is to
convert the measurement flow sheet into a connection matrix. A
connection matrix usually counts streams in columns and nodes
in rows. Where a stream flows into a node, the matrix value is set
to +1. Where a stream flows out of a node, the value is -1. All
other values are set to zero. Table 14 is for the multi-concentrate
example shown in Figure 4. ‘B’ streams are set in order at six to
ten and the zeros are omitted for clarity.
TABLE 14
Connection matrix for the circuit shown in Figure 4. Note that ‘B’
streams are numbered 6 to 10 and zeros are omitted for clarity.
Nodes Streams
1 2 3 4 5 6 7 8 9 10
1 1 -1 -1
2 1 -1 -1

SELECTION OF MEASUREMENT POINTS 3 1 -1 -1

Now that we have considered a range of flow sheets, we can
deduce some useful rules.
If we wish to use assays to define flow splits, we need to use
either very accurate sampling and assaying or survey around a
separation process (or a series of separations), which will make a
large difference to the assays. Hence the flow split around a
rougher section or a complete circuit will usually be well defined
by feed, product and tailing assays.
Conversely, the flow split across cleaner and scavenger sections
(and splitters!) will often be poorly defined by those assays. For
all of these cases, flow measurements are essential.
There is usually a good case for sampling and flow
measurement across a complete circuit – even though the study
objective may only be measurement of a single section.
ANALYSIS OF MORE COMPLEX CIRCUITS
The methods which have been described so far can also be
applied to more complex circuits.
4 1 -1 -1
5 1 1 1 1 -1
The advantage of this approach is that we can use matrix
arithmetic to express flow component errors and constraints in
quite general (and very compact) fashion.
If we have a vector of calculated flow rates where f1 is the flow
in stream 1 and so on, then any self consistent set of flow rates
must satisfy:
[C ].[ f ] = [ 0 ]
where:
C is the connection matrix for the flow sheet
Similarly, each assay times its stream flow rate times the
relevant stream/node entry in the connection matrix will generate
a vector of mass flow errors. Hence we can call [C i ] a connection
matrix where each row is multiplied by assay type i, then:
[C i ]. f = Δ i for each assay i

76 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 3 – MASS BALANCING FLOTATION DATA

TABLE 15
Product of the Connection Matrix (Table 14) and the copper assay matrix.

Nodes Streams
1 2 3 4 5 6 7 8 9 10
1 3.03 -1 0 0 0 -21.2 0 0 0 0
2 0 1 -1.15 0 0 0 -22.5 0 0 0
3 0 0 1.15 -0.75 0 0 0 -17.1 0 0
4 0 0 0 0.75 -0.52 0 0 0 -9.7 0
5 0 0 0 0 0 21.2 22.5 17.1 9.7 -15.5

As MS Excel has a built-in function for matrix multiplication,

TABLE 18
it is straightforward to set up a spreadsheet to define constraints
Best fit flow rate and constraint residuals around each node of
and the SSQ of mass flow errors for each assay type.
the balance. The residuals are effectively zero.
The product of the Connection Matrix (Table 14) and the
copper assay matrix is shown in Table 15. Flows Estimates Matrix product
The best fit solution for all flows (based on f1 = 100) is shown 1 100 -3.7E-14
with the copper SSQ. 2 93.13985 6.35E-14
3 90.13671 -1.5E-14
TABLE 16 4 88.13768 -2.1E-14
Best fit flow rate solution for all flows (based on f1 =100) is 5 84.63477 1.78E-15
shown with the copper SSQ.
6 6.860148
Flows Estimates Matrix product SQD 7 3.003143
1 100 64.4250163 4150.58 8 1.999028
2 93.13985 -78.0880874 6097.74 9 3.502911
3 90.13671 3.37058247 11.3608 10 15.36523
4 88.13768 -11.8850601 141.254
5 84.63477 43.0064029 1849.5
6 6.860148 MULTI-DIMENSIONAL BALANCES
7 3.003143 SSQ (SIZE BY ASSAY)
8 1.999028 12250.5 For detailed performance analysis, size by assay (or size by
9 3.502911 mineral) measurements will often be of interest as particle
composition (and degree of liberation) and size underlie flotation
10 15.36523
performance.
A major problem is that the coarse end of the size distribution
The best fit flow rates can be based on any one or any is often in short supply. This makes accurate assaying or
combination of assays by including its SSQ in the target for mineralogical measurement difficult. To overcome this problem,
Solver to minimise. Weighting and sensitivity estimates can be you can take multiple composite assays or simply wet screen
some additional sample cuts to obtain a substantial mass of
added in much the same way as for the single node case.
‘coarse end’ sample. The definition of ‘coarse’ will vary with
Once the best fit flow rates have been established, we can use a different circuits. Larger than the 90 per cent passing size is
similar strategy to find each set of data adjustments. usually a reasonable definition.
In this case, the adjustments become the parameters for Solver, The first test for size by assay data is to see if it is self
the SSQ of the adjustments is the target for minimisation and our consistent within the sample itself. To do this, simply add the
connection matrix multiplied by the flow rate of the selected assays for each size fraction weighted by the mass fraction for
component must be constrained to zero. each size fraction and see if the total assay resembles the
The results for copper are shown in Table 17 – followed by the measured total assay for that stream. If it does not, your sampling
constraints in Table 18 – which are near enough to zero. and analysis methods need to be examined in detail.

TABLE 17
Comparison of adjusted and measured copper assay data.

Stream 1 2 3 4 5 6 7 8 9 10
Adjusted -0.091 0.664 -0.166 -0.125 -0.250 -0.949 -0.437 -0.287 -0.498 2.139
Adjusted squared 0.008 0.440 0.028 0.016 0.062 0.900 0.191 0.082 0.248 4.576
Adjusted Cu 2.939 1.664 0.984 0.625 0.270 20.251 22.063 16.813 9.202 17.639
Measured 3.030 1.000 1.150 0.750 0.520 21.200 22.500 17.100 9.700 15.500
Total SSQ 6.552

Flotation Plant Optimisation Spectrum Series 16 77

CHAPTER 3 – MASS BALANCING FLOTATION DATA

Assuming that the weighted sum of the fractional assays does

agree reasonably with the measured total assays, we can now do a 100.00
beta estimate for each total assay and each size fraction using the
formula developed in the section ‘Accuracy Considerations’ or by Pb
using Solver as in the section ‘The Simplest Case’. Zn
10.00
Cu
As detailed in the section ‘Estimating the Accuracy of the Flow
Fe
Split’, we can reconstitute product and reject in the mass split
ratio to generate a feed stream which is exactly self consistent
1.00
with the products. This strategy may require a little more 0.10 1.00 10.00 100.00
explanation. After the best fit mass split (beta) has been
estimated, convert the product into component flow rates by
multiplying the concentrate size fraction by each assay (and the 0.10
gangue fraction as 100 – sum of all assays). Do the same for the
reject using (1 – the flow split) and add each component to
FIG 5 - Parity graph comparing measured total assays for each
generate each size fraction of the reconstituted feed.
metal in each stream with those based on reconstitution of the
Next divide the assay component flows in each size fraction by assays for each size fraction of each stream.
the fraction total to convert them back to assays.
As in our earlier example, if the reconstituted results are
reasonably similar to the measured, the reconstituted results will should be close to the mass split in the size fraction, which is the
usually be suitable for recovery calculations and modelling. A concentrate size fraction multiplied by the overall mass split
more quantitative way to check is too see if the reconstituted size divided by the feed size fraction.
fractions and assays are within the expected variation of the There are a range of mathematically complex ways to approach
measured feed data. As in the simpler one-dimensional case, two the general case (Hodouin and Everel, 1980; Gay, 1999).
thirds of the measured numbers are included in the data set for
analysis.
SIZE BY ASSAY EXAMPLE
For this many data points, a plot of measured (x) against
reconstituted (y) is a quick visual check for agreement. The data As noted in the previous section, a good place to start is to check
is from the example shown in Table 19. the reconstituted size fraction assays against the measured total
There are many more sophisticated approaches but they will not assay for each stream. Table 19 shows measured sizes, total
help very much with poor data – other than to spread the problems assays and assays reconstituted from size fraction assays based on
across the data set and make them more difficult to identify. samples taken from a zinc circuit.
Note that for diagnostic purposes, you can do a balance for Table 19 also shows mass split calculations for each total
each within each size fraction to see if the mass split within each component and for each group of totals based on the formula
size fraction is self consistent. The mass split calculated this way developed earlier. The total assays and the reconstituted assays

TABLE 19
Measured sizes, total assays and assays reconstituted from size fraction assays based on samples taken from zinc circuit.

Measured data A B C
Sizing Wt % Wt % Wt %
ai - ci bi - ci Beta (ai - ci) * (bi - c-)2 Beta bar
(bi - ci)
+75 20.85 11.87 18.48 2.36 -6.61 -0.3571 -15.623 43.746
+53 12.46 15.73 8.58 3.88 7.16 0.5424 27.770 51.197
+38 11.53 14.22 7.07 4.46 7.15 0.6242 31.910 51.124
-38 55.17 58.18 65.87 -10.71 -7.69 1.3920 82.343 59.155
Totals 126.400 205.222 0.616
Pb 0.59 1.19 0.39 0.20 0.80 0.2438 0.156 0.640
Zn 15.34 59.41 0.86 14.48 58.55 0.2472 847.511 3428.103
Cu 0.15 0.29 0.10 0.05 0.19 0.2368 0.009 0.036
Fe 13.70 5.38 16.44 -2.74 -11.06 0.2477 30.304 122.324
Rem 70.24 33.73 82.21 -11.98 -48.48 0.2470 580.548 2350.310
Recon totals 1458.528 5901.413 0.247
Pb 0.71 1.49 0.41 0.30 1.08 0.2801 0.326 1.163
Zn 15.43 57.50 1.10 14.32 56.40 0.2540 807.833 3180.405
Cu 0.14 0.34 0.12 0.02 0.22 0.0861 0.004 0.047
Fe 13.97 5.14 15.81 -1.84 -10.67 0.1729 19.668 113.755
Remainder 69.75 35.53 82.55 -12.80 -47.02 0.2722 601.971 2211.343
1429.802 5506.712 0.260
Overall 3014.731 11613.34 0.260

78 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 3 – MASS BALANCING FLOTATION DATA

are quite consistent and indicate a mass split to concentrate of
about 25 per cent. However, the splits based on size fractions are
not consistent with each other or the assay estimates. It would be
useful to investigate whether all of the size fractions are likely in
error or a particular one is at fault. We can investigate by applying
the 25 per cent split to the concentrate size distribution and the
75 per cent split to the tails.
This generates mass flows of 2.97 and 13.86 respectively. The
mass split within each size fraction can be estimated from
the assays within that size fraction. Table 20 estimates the
+75 micron mass split as 0.144, which gives flow rates of 3.00
and 17.85 respectively.
This simple analysis suggests that the feed and concentrate are
consistent with the assay split but the tailings are not. Hence
investigating the collection and sample preparation for the tailings
stream would be a good idea before doing more test work.
Size by assay balance
Each size fraction can be thought of as a separate process
(provided we use the same set of sieves) and we can apply the
strategies developed earlier to investigate self consistency and to
find a minimum SSQ of assay adjustments. Table 21 shows an
example for the +75 micron samples.
There is quite a bit of scatter but the split for zinc is very well
defined – as might be expected. The next size fraction (Table 21)
contains a little more mass, a lot more particles and is much more
self consistent.
The minimum data adjustment strategy is equally applicable.
Tables 20 and 21 show how to set this up for one size fraction.
Note the check sums in both directions. They do take up some
space but it is very important to ensure that the calculated data are
complete and self consistent with respect to size and assays.
Note that the ‘remainder’ assays are calculated by subtracting
the measured assays from 100 per cent. This removes a
substantial number of constraints from the solution.
The weighting factor is (1 + β 2 + (1 − β )2 ) as derived earlier.
The fully weighted case can be used in exactly the same way at
the expense of more complexity. For example, a beta value of
0.15963 gives a weighting factor of 1.73169.
The corresponding table of adjustments is shown in Table 23.
We can now generate a complete set of data adjustments within
each size fraction (weighted or not). To link the size fractions
together and tie back to the total assays it is important to realise
that the mass split within each size fraction has to be identical
for the size fraction based calculation and the assay based
calculations. Hence, we can multiply each feed size fraction by
that mass to generate a set of concentrate flows and by one minus
that split to generate a set of tailings size fraction flows and then
convert the flow back to size distributions. This strategy makes
another set of constraints implicit.
The last constraint can be imposed by calculating one of the
feed size fractions (usually the fines) by difference.
We can use the other feed size fractions and each of the size
fraction split factors as a set of parameters to minimise the total
SSQ for all of the sets of data. One way to set this up is shown in
Table 24. The parameters to minimise are identified in bold font
as are the adjusted total assays.
This approach is somewhat tedious to set up but does provide a
very detailed view of the data. For example, the problem with the
tailings size distribution identified earlier is now very clear. That
set of sizing data contributes more than half the SSQ. We can

TABLE 20
Assay mass balance within the +75 micron size fraction.

Mass splits in size fractions

Measured data A B C
Sizing Wt % Wt % Wt% Component
+75 ai - ci bi - ci Beta ai - ci * bi - ci (bi - c-)2
Pb 0.60 1.30 0.40 0.20 0.90 0.222 0.180 0.810
Zn 8.90 55.20 1.50 7.40 53.70 0.138 397.380 2883.690
Cu 0.17 0.30 0.18 -0.02 0.12 -0.125 -0.002 0.014
Fe 7.90 5.00 7.60 0.30 -2.60 -0.115 -0.780 6.760
Rem 82.44 38.20 90.32 -7.88 -52.12 0.151 410.966 2716.494
Fraction total 807.7444 5607.768
Beta bar 0.144040

TABLE 21
Assay mass balance within the +53 micron size fraction.

Measured data A B C
Sizing Wt % Wt % Wt% Component
+53 ai - ci bi - ci Beta ai - ci * bi - ci (bi - c-)2
Pb 0.65 1.00 0.50 0.15 0.50 0.300 0.075 0.250
Zn 18.3 59.00 1.30 17.00 57.70 0.295 980.900 3329.290
Cu 0.23 0.35 0.20 0.03 0.15 0.200 0.005 0.023
Fe 17.8 4.80 25.40 -7.60 -20.60 0.369 156.560 424.360
Rem 63.02 34.85 72.60 -9.58 -37.75 0.254 361.645 1425.063
Fraction total 100 100 100 1499.184 5178.985
Beta bar 0.289474

Flotation Plant Optimisation Spectrum Series 16 79

CHAPTER 3 – MASS BALANCING FLOTATION DATA

TABLE 22
Adjusted assays for the +75 micron size fraction.

Adjusted A B C
Sizing Wt % Wt % Wt% Exp Check
+75 Delta Lambda Sum
Pb 0.57 1.31 0.43 0.06 0.03 0.000
Zn 9.58 55.09 0.93 -1.17 -0.68 0.000
Cu 0.18 0.30 0.16 -0.03 -0.02 0.000
Fe 7.49 5.07 7.95 0.72 0.41 0.000
Remainder 82.18 38.24 90.53 0.44 0.25 0.000
Fraction total 100.00 100.00 100.00

TABLE 23
Assay adjustments for the +75 micron size fraction.

SSQ A B C
Sizing Wt % Wt % Wt % Delta
+75 SSQ
Pb 0.03 -0.01 -0.03 0.00
Zn -0.68 0.11 0.57 0.79
Cu -0.02 0.00 0.02 0.00
Fe 0.41 -0.07 -0.35 0.30
Remainder 0.25 -0.04 -0.21 0.11
Fraction total SSQ 1.20

TABLE 24
A simple spreadsheet strategy for a complete size by assay balance.

Calc size A B C B C
Flows Wt % Flow Flow Wt % Wt %
Adj A Adj B Adj C
+75 fit 18.89 3.02 15.87 11.81 21.31
+53 fit 13.05 3.74 9.31 14.63 12.50
+38 fit 12.08 3.76 8.31 14.74 11.16
-38 55.99 15.01 40.98 58.82 55.02
Totals 100.00 25.53 74.47 100.00 100.00
Size Splits
Beta 75 0.15963
Beta 53 0.28631
Beta 38 0.31148
Beta-38 0.26814
Adjusted Assays
Recon Flows A B C
Pb 69.76 38.14 31.62 0.698 1.494 0.425
Zn 1553.57 1468.72 84.85 15.536 57.540 1.139
Cu 16.84 8.54 8.30 0.168 0.334 0.111
Fe 1379.62 133.11 1246.51 13.796 5.215 16.737
Rem 6980.21 904.03 6076.18 69.802 35.417 81.587
Totals 10000.0 2552.54 7447.46 100.00 100.00 100.00

omit that contribution and reminimise to see how much the However, it does suggest that taking samples optimised for
results change. As might be expected, the changes are mostly in equi-probable selection is essential for this kind of work. That
the tailings size distribution but they are not very large. Hence means cutting through the stream to be sampled – and avoiding
either set of data would be similarly useful. any kind of dip sampling.

80 Spectrum Series 16 Flotation Plant Optimisation


RECAP
The suggested analysis sequence follows the order of the chapter.
Run a check around each node using the method of mass flow
errors detailed in earlier sections ‘The Simplest Case’ and ‘The
Method of Mass Flow Errors’.
The formulae derived in the sections ‘An Analytical Solution’
and ‘Estimating the Accuracy of the Flow Split’ are useful for a
quick check on the flow split and how well it is defined.
Where the data are very different in magnitude (ppm and per
cent assays), it is usually worth using weighted least squares for
both the flow split and the data adjustments. If the required data
adjustments are large (more than ten per cent of the
measurement) for more than a few of the measurements, redo the
experiment. The total weighted sum of squares should always be
of the same order as the number of measurements, node by node
or across a complete circuit. Hence it is worth examining each
total by node and for each set of assays.
If the adjustments are small, there is a lot to be said for
reconstituting concentrate and tailings to generate a set of self
consistent data – at least two thirds of your numbers will be the
ones you measured.
Once you have checked each node for self consistency, you can
continue on to the multinode balancing strategies outlined in the
section on ‘Practical Application – Single Level Balancing’.
However, as the circuit becomes more complex, it is well worth
considering a commercial product.
The single node checks will often pinpoint measurement
problems which may be concealed by doing the complete balance
first. Some commercial systems (such as JKSimMet) will let you
select each node – or a broader subset of the flow sheet – for a
check balance.
Plotting measured data against adjusted data is a way of
compressing a lot of data into a single graph. It is also an excellent
way of detecting bias. If all of the small assays are adjusted
upwards, there is likely a problem with low assay analyses.
For more complex data, such as size by assay or size by
mineral, the same guidelines apply. Consider the balance around
each node with each size fraction before progressing on to total
assay balances across the circuit and then total size by assay
balances across the circuit.
It is usually a poor strategy to attack the complete problem
from the start as least squares balancing can spread a single
wrongly measured (or transcribed) piece of data across the
complete set. Such errors are much easier to identify and weed
out in single node slices. Keeping backup splits of samples allows
for reassaying of dubious values when the rest of the set are self
consistent.
For mineral based analysis (MLA or Q*S) be aware of the
number of particles or sections which contribute to your
measurement. Low number errors follow the Poisson distribution
and even 100 sections will have a relative sd of ten per cent.
Overall, try to use a systematic approach, starting at the node
level and working across the circuit as you test each data point for
its actual contribution to the balance. While we may never be able
to know the ‘true’ value, careful measurement and analysis will
produce data which can provide a sound basis for flotation
modelling and for decision-making.
CONCLUSIONS
The techniques outlined in this chapter provide a reasonably
simple yet quite powerful approach to analysis of flotation data.
Using a spreadsheet for the process avoids capital outlay but
like all spreadsheet activity requires extreme care in checking
and including checks on constraints. While setting up these
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 3 – MASS BALANCING FLOTATION DATA
spreadsheets is a good way to learn about the analysis process,
they will not generally transfer between users.
Hence, for shared use – or anything at all complex – investing
in one of the commercial, flow sheet based products has much to
recommend it. Their web sites are listed at the end of this chapter.
Compared with even a very optimistic estimate of the true cost of
professional labour, the cost of commercially available mass
balancing systems is really very reasonable. A few days of time
saved would pay for most of them.
Unless you are quite an expert user, building your own
spreadsheets for more than the single node cases is a high-risk
endeavour. Using someone else’s spreadsheet is an excellent way
to generate a misleading analysis as the subsequent users do not
know what modifications have been made by others. If you are
obliged to use a spreadsheet, the author strongly recommends that
you also use it to generate checks for self consistency of your
calculated or adjusted data and graphs of all measured and
adjusted data.
The well documented and secured spreadsheet offered by
Luttrell (2004) may be a workable compromise in some cases
as it should not be subject to unauthorised and undocumented
‘improvements’.
ACKNOWLEDGEMENTS
Except as otherwise noted the data used for the examples in this
chapter were provided by Dr Chris Greet. The author gratefully
acknowledges permission granted by the Elsevier Scientific
Publishing Company to reproduce Figure 7.7 and Table 7-IV
from Lynch (1977).
REFERENCES
Boas, M L, 1966. Mathematical Methods in the Physical Sciences
(Wiley).
Deming, W E, 1938. Statistical Adjustment of Data (Dover).
Gay, S L, 1999. Avoiding the zero-flow solution in mass-balance
equations, Transactions of the Institutions of Mining and Metallurgy,
Mineral Processing and Extractive Metallurgy, 108:C121-126.
Hodouin, D and Everel, M D, 1980. A hierarchical procedure for
adjustment and material balancing of mineral processing data,
International Journal of Mineral Processing, 7:91-116.
Luttrell, G H, 2004. Reconciliation of excess circuit data using
spreadsheet tools, Coal Preparation, 24:35-52.
Lynch, A J, 1977. Mineral Crushing and Grinding Circuits, 340 p
(Elsevier: Amsterdam).
Wiegel, R L, 1972. Advances in mineral processing material balances,
Canadian Metallurgical Quarterly, 11:413-419.
FURTHER READING
Montgery, D C, Runger, G C and Hubele, N F, 2000. Engineering
Statistics, second edition (Wiley).
Morrison, R D, 1991. Material balance techniques, Evaluation and
Opitmization of Metallurgical Performance, pp 209-218 (AIME:
Denver).
Morrison, R D (Ed), 2008. An Introduction to Metal Balancing and Recon-
ciliation (Julius Kruttschnitt Mineral Research Centre: Brisbane).
Morrison, R D and Richardson, J M, 1991. JKMBal – The mass
balancing system, in Proceedings Second Canadian Conference on
Computer Applications for the Mineral Industry (CAMI ‘91), vol 1,
pp 275-286, Vancouver, September.
Wills, B A and Napier-Munn, T J, 2006. Mineral Processing Technology,
chapter 3 and associated spreadsheet examples (Elsevier).
USEFUL WEB SITES
For JKSimMet, JKMBal, JKSimFloat: <http://www.jktech.com.au>.
For BilMat and MATBAL: <http://www.algosys.com>.
Email: luttrell@vt.edu to request a copy of the spreadsheet mass
balancer.
81
 HOME

CHAPTER 4

A Practical Guide to Some Aspects of Mineralogy that

Affect Flotation
Alan Butcher

Strategic Marketing Manager, FEI Company, PO Box 2269, Milton Qld 4064.
Email: Alan.Butcher@fei.com

Alan R Butcher has over 29 years international experience as a field geologist, igneous petrologist, isotope
geochemist and applied mineralogist. His professional career has spanned the entire spectrum from
undergraduate and postgraduate teaching, to pure and applied research and consulting, and now
commercialisation of technology. He is best known for his enthusiastic evangelism and visionary
development of new and emerging capabilities in the field of Automated Mineralogy. Alan is a Fellow of the
Geological Society of London.

Abstract
Introduction
Background to Ore Deposits
Basic Ore Textures
The Concept of Liberation
A Review of Mineralogical Methods Available
Application of Mineralogy to Flotation
Example Case Studies
Concluding Remarks
References

ABSTRACT An understanding of the geological context, mineral assemblage, and

This chapter is designed to introduce some of the basic aspects of
mineralogy that can influence flotation behaviour, with particular emphasis
on sulfide flotation. The minerals industry has recently celebrated 100
years since the introduction of flotation, and we have had cause to reflect
on the many and varied developments in chemical reagents and flotation
cell technologies during this period, yet it is only more recently that the
role of mineralogy has been realised as another important component in the
better understanding of the flotation separation process.
Most of the information required by mineral processing engineers
involved in flotation can now be provided by geologists and mineralogists
using well-established optical, X-ray, laser and electron-microbeam
technologies. Metallurgically-relevant observations for flotation processing
applications can be made at any time during the life of a mine, from first
exploration drill core, which can flag-up problematic minerals and textures,
through start-up phases to full production, which can allow fine tuning of a
concentrator to produce acceptable grades and recovery. Finally,
mineralogy can be used to optimise recycling and/or disposal of tailings
during the mine closure phase.
textures of an ore is absolutely key to understanding its potential amenability
to the flotation process. Equally important is the measurement of sizes,
composition, locking and liberation, and flotation behaviour of particles as
they pass through a circuit from blasting, crushing, grinding, flotation to final
concentration, using the appropriate mineralogical techniques.
This chapter will review the many and varied mineralogical techniques
currently available, and will guide the reader through when and where to
use them. Examples will be drawn from case studies which illustrate how
mineralogy has allowed better understanding and improvement in the
flotation process on a range of ore types.
INTRODUCTION
An understanding of the geological context, mineral assemblage,
and texture of an ore is absolutely key to efficient and effective
unlocking and concentration of ore minerals during blasting,
crushing, grinding and flotation. Whilst this opening statement
might appear somewhat obvious to an experienced process

Flotation Plant Optimisation Spectrum Series 16 83

CHAPTER 4 – A PRACTICAL GUIDE TO SOME ASPECTS OF MINERALOGY THAT AFFECT FLOTATION
mineralogist, it is seldom appreciated by geologists, let alone by
mining and minerals engineers. Let us explore why this is so.
The reasons for this are not always clear, or even logical, but
experience has shown that it is due in part to one or more of the
following factors:
• Historical: in the past, only mining of high-grade ores or
easy-to-process ores was practised and, generally speaking,
these did not require strong geological input. In fact, in some
of the early mines, a geologist, let alone a mineralogist, may
not even have been on staff.
• Managerial: mineralogy is typically only considered and
commissioned during the exploration, feasibility and start-up
stages, as part of due diligence exercises. It is not that
common for a mine manager to agree to regular
mineralogical audits once an operation is up and running,
except of course when grade and recovery drop below
unacceptable limits. Also, mining geologists rarely make it
to mine manager level where they can influence behaviour
patterns in such matters.
• Financial: there is real perception (which is also sometimes
based on truth) that mineralogy is costly, requires expensive
equipment, and involves highly qualified personnel who
write not particularly timely and relevant reports, which are
all too often full of jargon, and serve only to alienate the
very audience for which they were intended.
Of course, the modern mining industry is quite different to that
in the past. Gone are the days where one could choose to mine
only the near surface, high-grade, easy-to-process ores. With
volatile metal prices, heightened expectations from shareholders,
and unprecedented industrial growth from China and India, the
race is on to extract as much from the Earth to keep up with
demand. This must of course take place in a world where the
awareness of human safety and the environmental footprint of
mining are coming under ever-increasing scrutiny. All of these
factors are challenging the viability of companies and mining
operations every day as we move to a world where mining
lower-grade, more-difficult-to-process resources will be the norm.
Given that our mantra as professionals involved in the Earth
Resources industry could be to ‘Take apart what Mother Nature
has put together. Safely … with due regard for the Environment
… and Economically’, we really need to make effective use of all
the armoury available to us to succeed in this combat exercise.
The science of mineralogy, and in particular that area of
mineralogy which is dedicated to the application of mineralogical
information to mineral processing and smelting in order to
improve understanding, solve problems and improve efficiency of
extraction, commonly referred to as Process Mineralogy, is one
such effective weapon in our mining armoury which we can draw
on. There is no doubt that process mineralogy, when used
judiciously, can make a major difference to a mining operation. In
the author’s opinion, this was first brought to the world’s attention
in a seminal paper by Henley (1983), although many others
followed shortly after that (Hiemestra, 1984; Petruk, 1984; Baum
et al, 1989). A text book has recently appeared by Petruk which
attempts to pull together his life’s experiences as a process
mineralogist and is a good source of case studies (Petruk, 2000).
Before we move on, let us consider a hypothetical, but typical
situation, which hopefully will keep the sceptics interested and
the novices alert.
The situation involves an ore type from the Eureka
Concentrator. It needs investigating because it is new. Routine
assaying by the geologists in the mine suggests this ore type
contains a three weight per cent copper head grade, similar to all
existing production ores. The temptation is to go ahead, mine and
crush the ore, and hope for the best. You might be lucky. But a
84 Spectrum Series 16
quick inspection of the ore by a mineralogist can immediately
establish if the copper is deporting to the usual phases (say
sulfides), at the same grain size (which affects grind size), and in
the same host gangue (which affects flotation conditions) as the
production ore. If conditions are the same, fine. If they have
changed, you could have saved yourself a lot of expensive
experimentation, not to mention embarrassment.
Mineral processing plants process minerals, and for mostly
practical reasons, it is accepted that conditions are often based on
chemical assays rather than on mineralogical information.
Hopefully, your opinion on the use of mineralogy will change
after reading through this chapter.
BACKGROUND TO ORE DEPOSITS
It is necessary at this point to introduce a number of technical
terms commonly used to describe ore deposits, the minerals
which make them up, and the textures which affect processing, all
with special reference to flotation behaviour.
Basic classification of deposit types
Believe it or not, it is helpful to know the geological context of a
mine, especially where mineralogical work is required to be
carried out. This is undoubtedly satisfying from an intellectual
viewpoint. But more importantly, geologists need to do this all
the time so that they can conjure up a mental 3D image of the
deposit on which they are working. It helps us see things to scale,
in some cases, put things back in order of formation and
ultimately allows us to rationalise and categorise the apparent
anarchy we find in nature.
Ores are no different to any other rocks – they simply have an
economic value and are defined as a mineral or aggregate of
minerals that form a rock which can be mined at a profit. They
can therefore be classified, like everything else in Earth Sciences,
into conveniently simplified categories. There are classifications
that are based on simple criteria, such as the minerals or metals
contained (Pb-Zn; Cu-Au; PGM, etc). Better still, we can describe
the structure of the deposit in terms of its shape or size (basin,
pluton, diatreme, vein, fault, etc), its relationship to the host rocks,
and the rocks which enclose or contain the deposit (eg stratabound,
stratiform, volcanic-hosted, sedimentary-hosted, etc).
Of less use, is to classify a deposit according to the geological
processes which combined to form the deposit (or genesis). Why?
Because most deposits are old, some very old, and whose
formation has not been witnessed by geologists today. We
therefore have to rely on interpretation alone, and given that there
is considerable debate among geologists as to the exact mode of
formation of most mineral deposits, this is not a good
classification in isolation.
So, in summary, it is best to stick to features which can be
agreed on, such as the physical description of the deposit. And
even though no two mineral deposits are exactly alike, most of
them fall into a clear category. Each category often conveniently
coincides with a generally accepted hypothesis as to how the
mineral deposit formed. So we typically end up with a physically
descriptive classification that includes a descriptor of how the
deposit formed. Once one understands that certain ore types
produce distinctive mineral assemblages and textures (and
geologists are trained to know this), then it is easier to focus on
issues that are of specific interest. Figures 1 and 2 summarise
some of the basic ore deposit types and structures likely to be
encountered by the professional minerals engineer. For a more
thorough review, the interested reader is referred to Guilbert et al
1986; Evans, 1993; Roberts and Shehan, 1988; Eckstrand, 1984;
Peters, 1978 and McKinstry, 1948.
Flotation Plant Optimisation
 CHAPTER 4 – A PRACTICAL GUIDE TO SOME ASPECTS OF MINERALOGY THAT AFFECT FLOTATION
FIG 1 - Summary diagram illustrating the main types of ore
deposit, classified according to mode of origin, host-rock
and commodity, with real examples.
Mineralisation
Stratiform Stratabound
Vein-hosted Fault-hosted
• Exploration methods
• Mining method
• Sampling Strategies
Disseminated Replacement
FIG 2 - Cartoons to illustrate how the geological structure of an
orebody controls how you explore for it, how you mine it and
how you sample it. Here six typical types are presented. Black
is mineral or ore of interest. No scale (after Butcher and Trudu,
1999).
Take, for example, the deposit on which Eureka Mine is
positioned. We know at the most basic level that it is a
polymetallic sulfide ore, and is mined underground. Armed with
just this information, an experienced geologist and process
mineralogist can mentally map this to be most likely a
sediment-hosted, volcanogenic massive sulfide deposit, with an
economic mineral assemblage of sphalerite (Zn), galena (Pb),
chalcopyrite (Cu).
With more information (such as age, geographic locality,
geological model, tectonic history, historical company reports), a
mental checklist can be immediately created, and forms the basis
on which the mineralogist starts his or her campaign. It may
include the following: look for oxidation effects of sulfides;
watch for grain size reduction if metamorphosed; look for talc if
sheared; check for more than one type of sphalerite; etc. More on
this later on in the chapter.
BASIC ORE TEXTURES
Once the macro-features (km-m) of an ore deposit have been
established, it is then necessary to concentrate on the meso-
(m-cm) and micro-features (cm-mm-μm). It is at these scales we
start to gain valuable insights into which minerals make up the
ore and how they all fit together to form what is known as the
texture. This is the area of study known as petrography.
A comprehensive study of ore deposit textures is beyond the
scope of this chapter but the interested reader is referred to the
Flotation Plant Optimisation
following texts for further details: Petruk, 2000; Craig and
Vaughan, 1994, and an excellent web-based atlas (Ixer and
Duller, 1998). Of particular interest to flotation are the following
basic types of textures. Starting at the most basic level (Figure 3),
an ore texture can be evaluated as to whether it is equigranular or
inequigranular. Milling strategies for the two extreme ores will be
quite different.
Equigranular
Inequigranular
Ore mineral
FIG 3 - Cartoons to illustrate the basics of ore textures, with special
reference to flotation. At the simplest level, ores can be considered
either equigranular (all grains same size) or inequigranular (not all
same size). The mineral of interest is shown in black. Grain
boundaries are straight or gently curved. No scale.
On a more complex level (Figure 4), it is crucial to ascertain
how the economic minerals occur: if they are present as rims,
disseminated inclusions (or exsolutions), or as interstitial phases –
as this will control breakage mechanisms during blasting,
crushing and grinding (intragranular versus intergranular), and
will make it either relatively easy, or difficult (or impossible) to
liberate and float effectively. It is not uncommon for an ore to
have multiple textural deportments for the element of particular
interest (see Figure 5).
Finally, and for completeness, a further concept is now
introduced. It was mentioned above that orebodies are generally
contained within old rocks. Practically, this means that they
have been around long enough to experience post-formation
modification, which can take one of many forms (metamorphism,
deformation, oxidation and weathering, to name a few).
Sometimes these so-called secondary processes can overprint or
obliterate those that developed at the time the deposit formed
(so-called primary features). Metamorphism, for example, is
generally bad news for the metallurgist as it often involves growth
of new (and unwanted) minerals (such as mica, talc, graphite,
serpentine, etc), which have an annoying habit of floating
along with the ore minerals to contaminate the concentrate.
Deformation can lead to fracturing of the ore, recrystallisation
and grain-size reduction, which can change the comminution
behaviour of an ore in a negative way (production of slimes).
Oxidation and weathering, however, can be both good and bad for
a minerals engineer – it can either upgrade (laterites, gossans) or
downgrade (silcretes, calcretes and regoliths) the ore depending
on which process is dominant.
Figure 6 attempts to simplify what is obviously a very complex
concept. The main point being made here is that it is the job of
the process mineralogist to: firstly, identify the presence of these
Spectrum Series 16 85
CHAPTER 4 – A PRACTICAL GUIDE TO SOME ASPECTS OF MINERALOGY THAT AFFECT FLOTATION

processes; then unravel them; and finally, understand how each

one may impact on the physical and chemical processing of the
ore under investigation, and report these results in a clear, easy to
understand and practical way to the metallurgist.

THE CONCEPT OF LIBERATION

Now that we have revealed that ore deposits form in different
ways, and that resulting mineral assemblages and ore textures, to
Rims Inclusions a large extent, reflect that heritage, we need to briefly review
what we mean by liberation from a mineralogical perspective.

Interstitial Disseminated

FIG 5 - One major outcome of a textural analysis of an ore is to

Exsolved Laminated
FIG 4 - Further cartoons illustrating complex ore textures, with
special reference to those which affect liberation and flotation
performance. Mineral of economic interest shown in black. No scale.
ascertain the deportment of elements of economic interest. In this
case, the element of interest, shown in black (say copper) occurs
as: chalcopyrite intergrown with pyrite (upper left); chalcopyrite
blebs included within sphalerite (upper right); discrete native
copper (centre); chalcocite rimming pyrite (bottom left); bornite
inclusions within silicate (bottom right); and native copper within a
cross-cutting vein (bottom left to top right). Scale: width 10 -
100 microns (after Butcher and Trudu, 1999).

FIG 6 - Cartoon illustrating some of the processes which are known to affect ores after initial formation. The fate of three pristine ore textures
is considered (one equigranular and two inequigranular), which undergo progressively more aggressive modification from left to right
(oxidation, hydrothermal alteration to metamorphism). The job of the process mineralogist is to identify the presence of these processes,
unravel them and understand how each one may impact on the physical and chemical processing of the ore under investigation.

86 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 4 – A PRACTICAL GUIDE TO SOME ASPECTS OF MINERALOGY THAT AFFECT FLOTATION
This is quite important as mineralogists often get asked to provide
an estimate of liberation or to estimate optimal grind sizes in their
reports.
Liberation is best studied, estimated and quantified in mineral
agglomerates (rock particles). Particles are typically examined in
2D sections, and this is best achieved by setting them in epoxy
resin blocks and exposing their cross-sectional geometry by
combinations of cutting, grinding and polishing.
Examination of images of particles allows for the classification
of each particle in terms of whether it is ‘pure’, ‘barren’ or
‘locked’. This classification of particles is based on the volume of
the mineral of interest in each particle.
The concept can be illustrated with a cartoon in the same
idealised format as we have seen previously. Figure 7 shows how
a complexly-textured ore (as viewed, for example, in thin-
section), and one that is essentially equigranular, starts to liberate
the mineral of interest at particle diameters of around 70 microns.
We will now see below that the degree of liberation can be
quantified using a variety of optical and SEM-based image
analysis techniques.
FIG 7 - Cartoon illustrating the effect of particle size reduction and
increasing liberation from a complex, equigranular, ore texture,
where the mineral of economic interest is shown in black. Note
that even the finest particle sizes still contain locked particles.
For a complete background to the theory of liberation, please
refer to Barbery (1991), and for further background on how
automated SEM-images can be used to derive knowledge of
particle behaviour in processing, see Sutherland et al, 1988;
Sutherland, Wilkie and Johnson, 1989; Spencer and Sutherland,
2000.
A REVIEW OF MINERALOGICAL METHODS
AVAILABLE
There are a bewildering number of techniques available to
determine the mineralogical and metallurgical behaviour of
particles (of differing sizes) as they pass through a mineral
processing circuit from blasting, crushing, grinding and flotation
through to final concentration and disposal, in terms of their
particle size and shape, composition, locking and liberation
characteristics and flotation behaviour.
The following is an essential guide to the main techniques,
which briefly covers each technique and provides when-and-
where-to-use-it advice.
Hand-specimen analysis
Every self-respecting geologist and mineralogist should be able to
gain information from examination of materials in hand-
specimen, using the naked eye or aided only by a hand lens, a
Flotation Plant Optimisation Spectrum Series 16
scratch plate and a bottle of acid. However, once materials
become reduced in size to the point where they are barely visible
to the naked eye, other techniques clearly need to be used.
Optical techniques
This probably still remains the most widely used technique
worldwide, and not surprisingly, it has been around almost as
long as the science of geology. It turns out that cleverly designed
combinations of reflected and transmitted light sources, in both
polarised and unpolarised states, allows minerals to be identified
by their characteristic optical behaviour, especially when viewed
under magnification on a petrographic polarising microscope. Ore
minerals (such as sulfides, precious metals and oxides) are best
viewed in reflected light as they tend to be opaque, whereas
gangue phases (such as silicates, carbonates, phosphates) are best
examined in transmitted light as they are often translucent.
Sample presentation normally takes the form of polished sections
or petrographic thin-sections, although loose grains can also be
usefully observed with a basic stereo microscope. Apart from the
identification of minerals, the modal mineralogy of a sample can
be determined by point counting. Textures can be observed and
recorded photographically (photomicrographs) and typically are
used to augment reports and illustrate features of the sample. A
useful text on this topic is by Gribble and Hall (1992).
In each case, optical techniques obviously require a competent
person to operate the microscope and make meaningful
observations, typically a graduate in geological sciences with a
few years relevant experience. Great microscopists, however, are
a greying workforce and are in serious decline. Good
microscopists are getting harder to find. And it is becoming more
and more difficult to encourage young graduates to move into
optical mineralogy. Automation of phase identification, modal
point counting and texture analysis goes someway to redress this
situation (Clemex Technologies Inc, 2009).
A clear advantage of this method is that, in the right hands, a
quick (and often cheap) prognosis can be obtained for both
unbroken ore as well as particulates. And that might be good
enough. But like all techniques, it must be used judiciously, and
in combination with other methodologies.
X-ray beam techniques
A commonly used technique to assess a sample for the relative
abundance of minerals present is by a method that utilises the
unique diffraction properties of minerals when subjected to a
primary X-ray beam, the so called X-ray diffraction method or
XRD for short. XRD is particularly useful when trying to
establish the nature of an unknown material, or tracking the
presence of an undesirable phase in a sample. It can be used to
gain both semi-quantitative and quantitative phase data.
XRD is a very powerful analysis technique for distinguishing
certain mineral groups such as the Clay Group (montmorillonite,
illite, kaolinite), the Serpentine Group (lizardite, talc), and certain
polymorphs (minerals with exactly the same composition but
differing only in their crystal structure, such as rutile-anatase,
quartz-coesite), or distinguishing minerals with similar, but not
exactly the same compositions (haematite-geothite-magnetite) or
optical behaviour. As it is a bulk method of analysis, it cannot
provide information on particle shape, size, or internal texture.
In certain situations, it can be a relatively quick and
inexpensive method to solve a problem. It is best used, however,
in combination with other mineral techniques.
XRD, of course, is not to be confused with X-ray fluorescence
spectrometry (XRFS), which is another established X-ray beam
technique, but unlike XRD, provides whole-rock chemical data.
87
CHAPTER 4 – A PRACTICAL GUIDE TO SOME ASPECTS OF MINERALOGY THAT AFFECT FLOTATION
It is most commonly used for grade determination (say, Fe-, Pb-
and Zn-content of ores, Ti-contents of mineral sands, P-content of
iron ores, etc).
Electron beam techniques
The technology which has most revolutionised mineralogical
investigation over the past 40 years or so has been that based on
the utilisation of signals generated as a result of interaction
between a highly focused electron beam and earth materials. As
soon as scanning electron microscopes (SEM) and later electron
probe micro-analysers (EPMA) were introduced, they found
immediate appeal among geologists, mineralogists and materials
scientists. Effectively non-destructive, and with the ability to
image, analyse and quantify a range of earth materials (both
unbroken rocks and particulates), they remain at the forefront of
mineral analysis for the earth resources industry today.
The underlying principles are common to all systems. electron-
induced, secondary X-ray, energy dispersive spectrometry (EDS)
can be used to qualitatively and quantitatively determine the
composition of a phase under the electron beam. Secondary
electrons (SEI), along with backscattered electrons (BSE) and
cathodoluminescence, all of which are simultaneously generated
along with the secondary X-rays, can be used to gain insights into
other aspects such as sample topography, composition and crystal
defects. EPMA systems (and some SEM systems) have both EDS
and wavelength dispersive spectrometers fitted (WDS – for
increased sensitivity using a particular crystal to tune into a suite
of elements). The electron beam can be used in both a scanning
or point-and-shoot mode.
A number of research groups around the world have attempted
to automate SEMs and EPMAs (King and Schneider, 1993;
Schneider and Neumann, 2004; Jones, 1984 and 1987), but only
the Australian systems have succeeded in being widely adopted
by the global minerals industry (Gottlieb et al, 2000; Baum,
Lotter and Whittaker, 2004; Gu, 2003). A particular strength of
an automated SEM system is that it removes some operator
strain, tedium and repeatability issues arising from manual SEM
work.
CSIRO developed a system known as quantitative evaluation
of minerals by scanning electron microscopy (QEM*SEM) in the
early 1980s (Grant et al, 1977; Miller, Reid and Zuiderwyk,
1982), which was used extensively by industry through ownership
of their own systems as well as through a bureau service offered
by CSIRO. The latest product of QEM*SEM technology is
known as QEMSCAN®, and is a complete automated solution
from sample preparation through analysis to data reporting (Pirrie
et al, 2004).
The University of Queensland, through the JKMRC research
centre, developed an alternative technology with similar
objectives in terms of automation. Known as the Mineral
Liberation Analyser (MLA), it began in 1997 (Keith, 1998; Dou,
1998), and was offered as a bureau service from 1999 (Gu, 2003).
Both QEMSCAN® and MLA are now owned by FEI Company,
who continue to develop, market and sell the two technologies.
Other techniques
Other technologies used by mineralogists include those which
utilise proton beams, ion beams, laser beams and near infrared
beams. A proton microprobe is ideal for mapping ultra-trace
elements within minerals. An example application at Eureka
Mine would be to examine the galena by PIXE to ascertain if
silver was present in solid solution (Goodall and Scales, 2006;
Goodall, Scales and Butcher, 2005). A secondary ion mass
spectrometer (SIMS) can be used for accurately measuring
invisible gold in sulfides – gold that is not possible to see with
88 Spectrum Series 16
SEM or EPMA microbeams as it is in solid solution (Vaughan
and Corrans, 1977). Fine grinding is not an option for this texture,
but leaching could be.
Logging of cores, chips and powders for mineralogical
composition can now be undertaken using scanners which utilise
near infrared radiation which is phase specific. This is ideal for
on-site situations where rapid (near real-time) bulk mineralogical
information is required to troubleshoot processing problems, or
predict required conditions of operation as ores and products pass
along conveyors (Huntington, 2006).
APPLICATION OF MINERALOGY TO FLOTATION
Introduction
Flotation isn’t about flotation cells. It is about
minerals. Liberating the right minerals in the
right place, with the minimum of grinding power.
Preparing the surfaces for maximum selectivity
for valuable minerals against gangue. Minimising
entrainment. Responding to the needs of different
size fractions. Doing all this at minimum cost,
using the least new equipment possible. Under-
standing the economics of the operation. And
making the whole circuit work together, simple,
responsive and operable (Xstrata Technology,
2005).
This quote appeared recently within the conference
proceedings of the Centenary of Flotation Symposium held in
Australia, which celebrated, as the name suggests, 100 years
since the introduction of flotation. It encapsulates many of the
concepts introduced so far, which is why it is reproduced here in
full. Flotation, in fact, was started because in Australia at Broken
Hill the ores were too fine to be separated efficiently by gravity
(Figure 8).
We now move on to consider the practical application of all
this know-how, particularly the role of mineralogy, in the better
understanding of the flotation separation process, with respect to
the Eureka Mine.
Let us imagine therefore, that we have been set a mineralogical
task to try and understand why the metal recoveries at the Eureka
concentrator are low despite the grades being acceptable.
Methodology
First up, we need decent primary samples to examine. These
must be representative of the problem we are trying to fix. A
good starting point is to always organise for the collection of
a feed, concentrate and tail for each of the main saleable
products. Experience has shown that you need to involve all
relevant personnel in all key areas of the mines’ activity in the
sampling, and to make them aware of the common pitfalls and
problems in sampling mineral processing plants (many of these
issues are covered in this book in other chapters, and by other
publications, eg Annels (1991). Sampling a dynamic system
such as a mineral processing plant, where crushing throughputs
can be in the region of 100 kt/d, is certainly a challenging
exercise. Results of any kind always need to be viewed with this
in mind.
Next, we need to create secondary samples from the primary
samples, using appropriate methods, which normally take the
form of one or more of the following, all of which are familiar
to the metallurgist: cone and quartering, splitting, riffling,
sieving, cyclosizing and de-sliming. From these secondary
samples, we are then able to produce genuine replicate aliquots
which can be used for all mineralogy and allied testing and
Flotation Plant Optimisation
 CHAPTER 4 – A PRACTICAL GUIDE TO SOME ASPECTS OF MINERALOGY THAT AFFECT FLOTATION

10 μm
10 μm

10 μm
10 μm

FIG 8 - Optical photomicrographs, taken using a digital camera attached to a reflected polarising petrographic microscope, of processed
particles from an Australian lead–zinc operation (McArthur River), which contain textures that historically would have been too fine for
conventional flotation, and now require combinations of staged and fine-grinding methods to separate the galena and sphalerite.

analysis (assaying, etc). Typically, these aliquots then get made • grain size estimates (used to determine optimum
into a number of products depending on what happens next. liberation grain sizes and predict or prevent problems
Polished thin-sections and polished blocks are best as these can such as poor liberation, over-grinding of the valuables,
be used for optical work, SEM and EPMA work, as well as production of slimes);
automated SEM techniques.
Then, we need to generate quality petrographic information • mineral associations (important for optimising the
from the prepared products. A typical checklist that most separation process, say galena from sphalerite,
mineralogists might run through when examining samples could chalcopyrite from sphalerite);
look something like: • grain boundary relationships (curved, straight or irregular
• Complete inventory of all known mineral phases present boundaries will affect behaviour during crushing and
(Figure 9). grinding as this imparts preferential breakage mechanisms
– eg along or across boundaries);
• Detailed modal analysis for major minerals (>5 per cent),
minor phases (>1 per cent - <5 per cent) and trace phases • elemental deportment information (important for
(<1 per cent). tracking how metals behave during processing, covered
by Figure 5); and
• Textural information on both gangue and ore minerals,
which includes: • photomicrographs of typical textures.
Ore Type Example Valuable Example Problem Gangue
PGM PGM, Ni-sulfides Talc, pyroxene
Copper sulfide Chalcopyrite Pyrite, Fe -oxides, silicates
Nickel sulfide Pentlandite Pyrrhotite, talc
Pb-Zn Galena, sphalerite Pyrite, pyrrhotite, marcasite, carbonates
Gold Gold, electrum Pyrite, arsenopyrite, Fe -oxides, quartz

Ore Crusher Grinding & Classification Separation Product

Waste
Typical Process Overview

chalcopyrite – fast floating; pyrrhotite – slow floating depressants for carbonates and talc

naturally hydrophobic minerals float

FIG 9 - Summary diagram to show how different ores, classified by the main ore minerals, have quite different gangue phases.

Flotation Plant Optimisation Spectrum Series 16 89

CHAPTER 4 – A PRACTICAL GUIDE TO SOME ASPECTS OF MINERALOGY THAT AFFECT FLOTATION

All of the above can be generated manually from careful This section covers two examples where mineralogical data
traditional optical-SEM-EPMA interactive analysis. Automated collected by the author has been turned into information and
techniques have enabled this type of information to be collected knowledge, and then action has been taken based on that
and reported in a fast, robust, operator independent, time- knowledge to realise the value. The examples have been simplified
efficient and scalable manner. and modified so as to be relevant to the present text. A summary of
information typically acquired is summarised in Figure 10.
EXAMPLE CASE STUDIES
Case study 1 – finding the optimal grind size
Introduction Following measurement of crushed drill cores as part of ongoing
For mineralogy to add value to a mining operation, one really ore characterisation exercise for a new orebody at Eureka Mine, it
needs to acquire consistently meaningful mineralogical data over was found that there was a distinct bi-modal distribution to the
a significant period of time (weeks, months, years). If a grain size distribution for galena (Figure 11). This flagged the
well-performing ore is properly characterised it is much easier to possibility of whether a two-stage grinding process could be
troubleshoot problems in an under-performing ore because there implemented. When the same distribution for sphalerite was
is a benchmark against which to compare it. Without a good day, considered, however, it was found that this displayed a normal
it is difficult to diagnose what went wrong on the bad day. distribution (Figure 12).
Mineralogical Data
Roughers Liberation
Particle size
Particle composition
Cleaners Particle surface area
Problematic particles
Penalty elements
Re-grind Mill Mineral associations

Metallurgical Implications
• Reagent regime
Scavengers • Surface preparation
• Residence time
• Mechanical entrainment
• Flash flotation options
• Staged grinding
• Grinding media

FIG 10 - Idealised view of what mineralogical data can be extracted from samples taken from a mineral processing circuit, and the
implications these data may have on metallurgical behaviour and practices.

Lead-zinc Galena
deposit
-

+
2

6
0

0
/

/+

/+

/+

0
+

0
/+

/+

/+
2

+
0

4
5

5
6

G a l e n a B i - S iz e D i s t r i b u t io n

FIG 11 - An image grid of galena-containing particles (red) from crushed drill core, organised by galena grain-size estimate, as
measured by a QEMSCAN® and plotted using iExplorer™. The distribution shows a distinct bi-modal distribution.

90 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 4 – A PRACTICAL GUIDE TO SOME ASPECTS OF MINERALOGY THAT AFFECT FLOTATION

Lead-zinc Sphalerite
deposit

+
2

6
0

0
/+

/+

0
+

0
/+

/+

/
+

+
2

5
0

4
5

5
6

0
S p h a le r it e S i z e D i s t r ib u t i o n M T G A

FIG 12 - An image grid of sphalerite-containing particles (blue) from crushed drill core, organised by sphalerite grain-size estimate,
as measured by a QEMSCAN® and plotted using iExplorer™. The distribution shows a normal distribution.

Fe-Sphalerite
Action Sphalerite
Pyrite
Given that a two-stage grinding circuit was not possible at Low Fe- Concentrate High Fe-Concentrate
Eureka, an optimal grind size of 212/+106 microns was trialled CS2 CS3+4 CS5 CS2 CS3+4 CS5
during bench-scale grinding tests, and was found to achieve the 100

required grade and recovery. 90

80

Case study 2 – tracking down reason for penalty 70

elements 60
50
After a few months of excellent results and acceptable
40
concentrator performance, a crisis hit Eureka Mine. The levels of
30
iron in the final zinc concentrate reached unacceptable levels.
20
Mineralogy was used to establish the source of the iron.
10

Answer 0
Based on more than 1000 grains/fraction
Historic samples taken on a good day were compared with those
containing high iron contents. The results showed that Fe deports FIG 13 - Comparison of samples from the final zinc concentrate,
to both pyrite and Fe-rich sphalerite. Both phases were more good and bad, in terms of Fe-content. It illustrates that when iron
abundant in the final concentrate that had higher Fe-content increases, it is due to increases in both Fe-sphalerite and pyrite.
(Figure 13).
which will help reduce particle size to the required liberation size
Case study 3 – understanding locking range. Bench-scale testing could validate this theory.
relationships
A re-evaluation of the locking characteristics of particles within CONCLUDING REMARKS
the lead circuit was initiated following a managerial decision to It is the hope of the author that, after having gone through this
set new grade and recovery targets at the Eureka Concentrator. In chapter, the reader will take something away of interest, and most
order to undertake this, a feed sample was collected and analysed importantly, something of use, which can be directly applied to
on a size-by-size basis. The results revealed that the locking their everyday work.
characteristics are complex (Figure 14). Galena is locked with
sphalerite, as expected, but also with Fe-sulfides and non-sulfide No one expects or wants minerals engineers to become
gangue (NSG), which are problematic from a contamination mineralogists or geologists – far from it. What the author
perspective. encourages, however, is that metallurgists, like geologists, mining
engineers and environmentalists, should take some of the
responsibility for developing a culture of cross-pollination of
Answer professional skill areas within the modern minerals industry.
Finer grinding of the feed ore is perhaps suggested. This will Mineralogy is clearly useful when applied in the correct way.
liberate the composite particles. Installation of secondary Rocks contain secrets that traditionally only geologists and
grinding equipment might be the solution, such as stirred mills, mineralogists have been able to unlock. With the advent of more

Flotation Plant Optimisation Spectrum Series 16 91

CHAPTER 4 – A PRACTICAL GUIDE TO SOME ASPECTS OF MINERALOGY THAT AFFECT FLOTATION

Background
Fe Sulfides
Sphalerite
Galena
Other Sulfides
NSG

FIG 14 - Image grid of particles from a feed sample, sorted according to the size fraction from which they were measured, and according
to the texture they exhibit. The results show that the locking characteristics of galena are quite complex and involve: locking with sphalerite
(high- and low-grade ternary), Fe-sulfide, non-sulfide gangue (NSG). Both barren and fully liberated particles are also present. Measured
on QEMSCAN® and plotted using iExplorer™.

companies adopting automated mineralogical techniques, it is
clear that mineralogy is experiencing a revival, and it is allowing
the engineers insights that were previously just not possible.
Finally, ‘to measure is to know’ as Lord Kelvin, the famous
English scientist, once remarked. If you do not understand your
inputs and your outputs, then you cannot hope to optimise the
overall process, whatever it is.
REFERENCES
Annels, A E, 1991. Mineral Deposit Evaluation, first edition, 436 p
(Chapman and Hall).
Barbery, G, 1991. Mineral liberation: Measurement, simulation and
practical use in mineral processing, Quebec Editions GB, p 341.
Baum, W, Lotter, N O and Whittaker, P J, 2004. Process mineralogy – A
new generation of ore charactrisation and plant optimisation, SME
Annual Meeting, Denver, 23 - 25 February 2004, pp 1-5.
Baum, W, Sanhueza, J, Smith, E and Tufar, W, 1989. The use of process
mineralogy for plant optimisation at the El Indio gold-silver-copper
operation (Chile), Erzmetall, 42(9):373-378.
Butcher, A R and Trudu, A G, 1999. Ore characterisation and its
relevance to the crushing and grinding of ores, presented to AJM
Crushing and Grinding Conference, Kalgoorlie, 27 - 28 April 1999.
Clemex Technologies Inc, 2009. Clemex web site. Available from:
<http://www.clemex.com>.
Craig, J R and Vaughan, D J, 1994. Ore Microscopy and Ore
Petrography, second edition, 434 p (John Wiley and Sons, Inc).
Dou, M, 1998. The segmentation of BSE images of mineral particles
using an object based approach, in Proceedings Seventh JKMRC
Conference, Brisbane.
Eckstrand, O R, 1984. Canadian Mineral Deposit Types: A Geological
Synopsis, Geological Survey of Canada, Economic Geology Report
36, 86 p (Canadian Government Publishing Centre).
Evans, A M, 1993. Ore Geology and Industrial Minerals – An
Introduction, third edition, 389 p (Blackwell Scientific Publications
Inc).
Goodall, W R and Scales, P J, 2006. Proceedings MEI Automated
Mineralogy ’06, Brisbane.
Goodall, W R, Scales, P J and Butcher, A R, 2005. The use of
QEMSCAN® and diagnostic leaching in the characterisation of
visible gold in complex ores, Minerals Engineering, 18:877-866.
Gottlieb, P, Wilkie, G, Sutherland, D, Ho-Tun, E, Suthers, S, Perera, K,
Jenkins, B, Spencer, S, Butcher, A and Rayner, J, 2000. Using
quantitative electron microscopy for process mineralogy applications,
Journal of Mining, April, pp 24-25.
Grant, G, Hall, J S, Reid, A F and Zuiderwyk, M, 1977. Characterisation
of particulate and composite mineral grains by on-line computer
processing of SEM Images, in Proceedings 15th APCOM
Symposium, Brisbane, pp 159-170.
Gribble, C D and Hall, A J, 1992. Optical Mineralogy, Principle and
Practice, p 295 (UCL Press).
Gu, Y, 2003. Automated scanning electron microscope based mineral
liberation analysis – An introduction to JKMRC/FEI mineral
liberation analyser, Journal of Minerals and Material
Characterisation and Engineering, 2(1):33-41.
Guilbert, J M and Park Jr, C F, 1986. The Geology of Ore Deposits, 985 p
(W H Freeman and Company).
Henley, K J, 1983. Ore Dressing Mineralogy – A Review of Techniques,
Applications, and Recent Developments, special publication no 7,
pp 175-200 (Geological Society of South Africa: Marshalltown).
Hiemstra, S A, 1984. The role of mineralogy in ore beneficiation as
illustrated by case histories, in Proceedings Second International
Congress on Applied Mineralogy, Los Angeles, 22 - 25 February,
pp 9-20.
Huntington, J, 2006. The hi-logger scanner for automated mineralogy,
presented to MEI Automated Mineralogy ’06, Brisbane.

92 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 4 – A PRACTICAL GUIDE TO SOME ASPECTS OF MINERALOGY THAT AFFECT FLOTATION
Ixer, R A and Duller, P R, 1998. Virtual atlas of opaque and ore minerals
in their associations [online]. Available from: <http://www.smenet.
org/opaque-ore/>.
Jones, M P, 1984. Recent developments in the rapid collection of
quantitative mineral data, in Proceedings Second International
Congress on Applied Mineralogy, Los Angeles, 22 - 25 February,
pp 141-156.
Jones, M P, 1987. Applied Mineralogy: A Quantitative Approach, p 253
(Graham and Trotman, Springer).
Keith, J, 1998. Mineral identification using X-rays, in Proceedings
Seventh JKMRC Conference, Brisbane.
King, R P and Schneider, C L, 1993. An effective SEM-based image
analysis system for quantitative mineralogy, KONA, 11:165.
McKinstry, H E, 1948. Mining Geology, 680 p (Prentice Hall: Englewood
Cliffs).
Miller, P R, Reid, A F and Zuiderwyk, M A, 1982. QEM*SEM image
analysis in the determination of modal assays, mineral associations
and mineral liberation, presented to XIV International Mineral
Processing Congress, Toronto.
Peters, W C, 1978. Exploration and Mining Geology, 696 p (John Wiley
and Sons Inc).
Petruk, W, 1984. Image analysis measurements and data presentation for
mineral dressing applications, in Proceedings Second International
Congress on Applied Mineralogy, Los Angeles, 22 - 25 February,
pp 127-140.
Petruk, W, 2000. Applied Mineralogy in the Mining Industry, p 286
(Elsevier Science).
Pirrie, D, Butcher, A R, Power, M P, Gottlieb, P and Miller, G L, 2004.
Rapid quantitative mineral and phase analysis using automated
scanning electron microscopy (QEMSCAN®); potential applications
in forensic geoscience, Journal of the Geological Society of London,
232:123-136.
Flotation Plant Optimisation Spectrum Series 16
Roberts, R G and Sheahan, P A, 1988. Ore Deposit Models, 194 p
Geoscience Canada, Reprint Series 3 (Geological Association of
Canada).
Schneider, C L and Neumann, R, 2004. Measurement of exposed area
by image analysis particle cross-section (eds: Pecchio et al), in
Proceedings 2004 ICAM, Sao Paulo, pp 1035-1038.
Spencer, S and Sutherland, D N, 2000. Stereological correction of
mineral liberation grade distributions estimated by single sectioning
of particle images, Image Anal Sterel, 19:175-182.
Sutherland, D, Gottlieb, P, Jackson, R, Wilkie, G and Stewart, P, 1988.
Measurement in section of particles of known composition, Minerals
Engineering, 1(4):317-326.
Sutherland, D N, Wilkie, G and Johnson, C R, 1989. Simple prediction of
processing characteristics of ores using QEM*SEM, in Proceedings
Mineralogy and Petrology Symposium, pp 81-84 (The Australasian
Institute of Mining and Metallurgy: Melbourne).
Vaughan, J and Corrans, I J, 1997. Estimation of submicroscopic gold in
ores by SIMS and ultrafine milling: A comparative investigation,
(eds: G Gillen, R Lareau, J Bennett and F Stevie), in SIMS XI Annual
Workshop, Orlando, 7 - 12 September (Wiley and Sons).
Xstrata Technology, 2005. Quote from Xstrata Technology company
profile published in Proceedings Centenary of Flotation Symposium
(The Australasian Institute of Mining and Metallurgy: Melbourne).
93
 HOME

CHAPTER 5

Characterisation Measurements in Industrial Flotation

Cells
Greg Harbort and Sarah Schwarz

MAusIMM, Principal Process Engineer, GRD Minproc Ltd, Precinct 2, Level 2, 10 Browning Street,
West End Qld 4101. Email: greg.harbort@minproc.com.au

Greg has 23 years experience in operations, research and development, design and engineering. Principal
speciality areas include flotation and ultra-fine grinding technologies. Areas of experience include base
metals, gold and coal with emphasis on flotation circuit design, optimisation and simulation, flotation
machine design and analysis and test work appraisal. Previous employment has included Peko Mines as a
production metallurgist, Xstrata Technology as a development metallurgist with both the Jameson Cell and
Isa Mill and JKTech as Technology Manager. In his current role as Principal Process Engineer with GRD
Minproc he is responsible for delivering best practice process plant design services in designated studies,
projects and consulting services.

Technology Transfer Manager, JKTech Pty Ltd, Isles Road, Indooroopilly Qld 4068.
Email: s.schwarz@jktech.com.au

Sarah has over ten years of experience in mineral processing (mainly flotation), including operations,
research and development as well as consulting. Sarah has a Bachelor of Engineering in Metallurgical
Engineering, as well as a PhD in Mineral Processing, both awarded from the University of South Australia.
She joined JKTech in 2003 and has been strongly involved in the development and release of JKSimFloat
V6, as well as developing steady-state flotation models for plants worldwide. She is currently the
Technology Transfer Manager responsible for commercialising new research outcomes from the JKMRC
and other SMI Centres, as well as coordinating business development and marketing activities. In addition
Sarah continues to have a significant role in software development, training courses and flotation circuit
optimisation.

Abstract
Introduction
Gas Dispersion Measurements
Froth Recovery Measurements
Other Plant Measurements
Conclusions
References

AMIRA P9 project. These measurements have been made in a wide range

ABSTRACT
In recent years a significant volume of research has focused on flotation
machine parameters and the impact they have on metallurgical
performance. Technology has been developed as part of the Australian
Mineral Industries Research Association (AMIRA) P9 project, to conduct
characterisation measurements (such as gas dispersion and froth recovery)
in industrial flotation cells.
Gas dispersion measurements include gas hold-up, superficial gas
velocity and bubble size, and are a good indication of the hydrodynamics
within the pulp phase. Each of these measurements can be made in
industrial-sized cells using specialised equipment developed within the
of type and size of flotation cells throughout the world.
Various techniques have also been developed to indicate the efficiency
of the froth phase, known as froth recovery. These techniques range from
physically changing the froth height, through to performing mass balances
across the froth phase, and collecting samples of the material attached to
the bubbles within the pulp phase.
This chapter discusses methods for obtaining these characterisation
measurements, as well as other necessary measurements within an
industrial flotation plant environment. These measurements can be used by
plant metallurgists to analyse and improve the performance of their
flotation circuits.

Flotation Plant Optimisation Spectrum Series 16 95

CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS
INTRODUCTION
The range of flotation cell types and sizes in flotation plants has
varied over the years, from older circuits containing a large
number of small volume cells (<3 m3), through to newer circuits
containing smaller numbers of larger cells (>100 m3). The typical
and/or optimum operating conditions of flotation cells must be
taken in context of the cell type and size. The following lists the
common types of cells currently in the flotation industry:
• mechanical air-induced cells, eg Outotec, Metso, Bateman;
• self-aerated cells, eg Wemco;
• column; and
• Jameson cells.
Each of these cell types has different operating conditions
required to produce the optimum metallurgical performance.
Note that the performance of the cell is also dependent on the
characteristics of the slurry passing through, although this will be
addressed in another chapter.
For many years, plant metallurgists have been aware of the
importance of air, agitation and froth characteristics on the overall
cell performance, but have not been able to quantify the
individual variables. Techniques have been developed in the last
15 years to measure the gas dispersion properties and the
performance of the froth phase. This has allowed plant
metallurgists to benchmark their cell operating conditions and
identify potential areas of improvement.
This chapter focuses on some of the measurement techniques
available for characterising the gas dispersion properties such as
gas hold-up, superficial gas velocity and bubble size, and the
various techniques to characterise the froth performance, in terms
of froth recovery.
GAS DISPERSION MEASUREMENTS
In simple terms, gas dispersion can be defined as how well the air
entering a flotation cell is dispersed throughout the cell. Gas
dispersion conditions within a flotation cell directly influences the
flotation efficiency (Schubert and Bischofberger, 1978; Gorain,
Franzidis and Manlapig, 1995a) and can be characterised by
measurements such as bubble size, gas velocity and gas hold-up.
Bubble size
The bubble size within the pulp phase of a flotation cell
significantly influences the efficiency of particle-bubble contact
FIG 1 - Schematic of the McGill bubble size analyser and typical set-up.
96 Spectrum Series 16
and attachment. In general, smaller bubbles are more efficient for
smaller particles and larger bubbles are more suited for larger
particles (Biswal, Reddy and Bhaumik, 1994). Also, it has been
acknowledged that smaller bubbles within the pulp phase improve
the flotation kinetics (Diaz-Penafiel and Dobby, 1994). While
these points have been in general knowledge for a while, the
problem has been actually measuring the bubble size in the pulp
phase of industrial flotation cells. A large number of bubble
measurement techniques are available, consisting of either
invasive or non-invasive techniques. The most common technique
developed to measure bubble size in industrial flotation cells
currently is the McGill bubble size analyser, due to its simplicity
and robustness. This is discussed in detail, with other measurement
devices discussed briefly, highlighting their advantages and
disadvantages, in the following sections.
McGill bubble size analyser
As part of the AMIRA P9 project a bubble viewing device, to
collect and image bubbles, was developed by Jameson and Allum
(1984). This approach has been progressively refined over the years,
leading to the McGill bubble size analyser, developed at McGill
University in Canada. This is now the most commonly used device
for industrial bubble size analysis. A schematic of the McGill
bubble size analyser (or ‘bubble viewer’) is given in Figure 1.
The McGill bubble size analyser consists of a sampling tube
attached to a viewing chamber with a window inclined at 15°
from vertical. The sampling tube is immersed in the flotation cell,
with the bottom of the tube normally just below the pulp/froth
interface in the quiescent zone. The set-up is filled with water
dosed with enough frother to prevent bubble coalescence in the
tube. Bubbles from the flotation cell rise up through the sampling
tube and slide up the inclined window of the viewing chamber.
The sloped window ensures a single layer of bubbles passes
through the image area, captured by a digital video camera. The
light source illuminates the image area, resulting in each bubble
casting a shadow, which can be detected by the software
associated with the bubble viewer. Full details of the apparatus
and methodology can be found elsewhere (Chen, Gomez and
Finch, 2001; Hernandez-Aguilar, Gomez and Finch, 2002;
Hernandez-Aguilar et al, 2004). Further refinements have since
been made to the apparatus, with LED lights operating from a
battery and a digital still camera used instead of a video camera.
The probe is known as the Anglo Platinum Bubble Sizer, or
APBS and has been described in Taute and McClelland (2006).
Flotation Plant Optimisation
 CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS

Other bubble size measurement techniques Conductivity probes

Extensive reviews of alternative bubble sizing techniques have
been conducted by Rodrigues and Rubio (2003) and Boyer,
Duquenne and Wild (2002). Many of these are restricted to
laboratory use or limited to specific applications. Brief
descriptions of these methods are given below.
Conductivity probes register the variation in conductivity between
bubbles and the surrounding pulp. Immersion conductivity probes
are also used and can measure bubble size without extracting
bubbles from the flotation machine (Sanaullah, Zaidi and Hills,
2001; Wu et al, 2001).

Alternative photographic techniques Analysis of bubble sizes

A number of alternative photographic techniques are in use for
bubble size measurement. The LTM-BSizer (Rodrigues and Rubio,
2003) is similar to the McGill bubble size analyser in terms of
equipment. It varies once bubbles have entered the viewing
chamber where the bubbles are decelerated and stopped prior to
being photographed. Taking pictures through the wall of a flotation
machine is a frequently used bubble sizing technique which
requires a see-through section of wall and provides a bubble size
image in the immediate vicinity of the wall. Laser velocimetry uses
a laser sheet to illuminate a slice of fluid within a flotation
machine, which can then be photographed and analysed.
Both the UCT bubble sizer and the McGill bubble viewer
produce a bubble size distribution, the average bubble diameter
and the Sauter mean bubble diameter for each measurement. The
bubble size distributions can be compared on a normalised basis,
as shown in Figure 3. This figure shows that cells 2 and 3 had
similar shaped bubble size distributions, with a significantly
different average bubble size compared to the other two cells in
the circuit. This indicates that cells 2 and 3 have a lot of similar
smaller sized bubbles compared to cells 1 and 4, which have a
large range of bubble sizes.
1.6

Normal distribution frequency (%)

1.4
Optical fibre and sensor probes
1.2
Optical needle probes are very thin, sharp ended probes, facing 1.0
the bubble flow direction in order to pierce bubbles. Infrared light 0.8
beams are transmitted along the optic fibre to tips made of quartz
0.6
or sapphire and the difference in light transmittance when an
0.4
individual bubble is pierced is registered. They can be used in
three-phase high air hold-up conditions (Magaud, Souhar and 0.2

Boisson, 2001). The most commonly used in industrial 0.0

0 1 2 3 4 5 6 7 8
environments is the UCT bubble probe, Figure 2 (Tucker et al,
Bubble diameter (mm)
1994). This analyser consists of a sampling system and an
optical/electronic detector system. The sampler enables incoming Cell 1 Cell 2 Cell 3 Cell 4
bubbles to be cleaned of any attached particles before they enter
the capillary, where they are deformed from spheres to cylinders. FIG 3 - Normalised distribution of bubble sizes obtained from four
The length of the bubbles is detected by a pair of optical sensors. cells within a flotation circuit.

The atmosphere
Three-way valve
Flush water

Burette

The atmosphere Peristaltic

The atmosphere
Pump 1

Water reservoir
Photo-detector MPU timer Laptop
Capillary amplifier detector computer

Peristaltic
Plant water
Pump 2

Cleaning column Compressed air

The atmosphere

Pinch valve

FIG 2 - Schematic of the UCT bubble size analyser.

Flotation Plant Optimisation Spectrum Series 16 97

CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS

The average bubble diameter is calculated from Equation 1 and Air flowmeter readings
the Sauter mean bubble diameter is determined from Equation 2:
For most air-induced mechanical cells, the air flow rate is
N controlled and recorded on the control system. Using the flow
∑d i rate recorded from the flowmeter, superficial gas velocity, Jg, can
db = i=1
(1) be calculated from Equation 3:
N
Q
Jg = (3)
∑d i
3 A
d32 = i=1
(2)
∑d i=1
i
2
where:
Q is the volumetric air flow rate
where: A is the cross-sectional area of the flotation cell
di is the measured bubble diameter However, this method can be problematic if the flowmeters
themselves are not correctly calibrated, if the gas is not uniformly
N is the number of bubbles measured dispersed throughout the cell, or if the cells are self-aerated (and
Generally the Sauter mean bubble diameter is quoted from hence no flow rate recorded). The following methods enable the
these measurements, as this has been shown to be a more accurate gas velocity to be measured accurately independent of the above
representation of the bubble size distribution. Currently there is conditions.
some debate about whether the ratio of the Sauter mean to the
arithmetic mean should also be quoted, to indicate the width of The JKMRC mechanical Jg probe
the bubble size distribution.
The mechanical Jg probe was developed at the Julius Kruttschnitt
Typical bubble size ranges Mineral Research Centre (JKMRC) as part of the AMIRA P9
project. A schematic of the probe is given in Figure 5.
Generally, Sauter mean bubble sizes range from 1.0 mm to 2.0 mm The JKMRC Jg probe consists of a Perspex tube with a
within mechanical cells, with larger bubbles found in larger cells pneumatically controlled valve at one end and a water inlet and an
(>100 m3). Sauter mean bubble sizes within columns are usually air release valve at the other end. The probe is lowered into the
larger, 1.5 mm to 3.0 mm as the norm. Results of Sauter mean flotation cell, generally just below the pulp/froth interface and
bubble sizes measured in over 1000 cells are given in Figure 4.
filled with water. The air release valve is closed to the atmosphere
and the pinch valve at the bottom is opened to allow gas from the
cell to flow into the probe. This gas displaces the water within the
30
probe at a rate equal to the velocity flowing within the cell, and is
25
Column, Jameson, Flash measured by the time required for the water level to drop a
Conventional cells
predetermined distance. Corrections are made for pressure, based
Frequency (%)

20 on the length of the probe and the distance the probe is immersed
in the pulp phase. Full details of the use and set-up of this probe
15 can be found in Gorain, Franzidis and Manlapig (1996) and Power,
Franzidis and Manlapig (2000).
10

5 Worked example
0 Repeat measurements have been made using a Jg probe in a single
0.5 0.75
1 1.25
location in a flotation cell. The following is a list of the times
1.5 1.75 taken for the water level to drop 20 cm: 14.0 s, 19.1 s, 17.7 s,
2 2.25
2.5 2.75
3 3.25 15.7 s, 14.8 s. The length of the probe from the bottom valve to
Bubble size (mm) the bottom line was 2.9 m and the distance from the bottom valve
to the top of the pulp phase was 1.2 m. Calculate the average
FIG 4 - Range of Sauter mean bubble size obtained in mechanical, superficial gas velocity of this cell.
column, Jameson and Flash flotation cells (from Schwarz and
Alexander, 2006). Calculations
The experimental Jg value can be obtained using the following
Superficial gas velocity
equation:
Superficial gas velocity is a measure of the aeration ability of a
L
cell and is directly related to the flotation kinetics (Ahmed and Jg =
Jameson, 1989). If the superficial gas velocity is too high, t
however, the entrainment of gangue material into the froth phase where:
may become excessive and reduce the froth stability, as well as
decrease the concentrate grade. Measuring the superficial gas L is the distance between marks
velocity at various locations across the cell cross-sectional area t is the time taken for the water level to drop between the marks
has been shown to give a good indication of how well the air
is dispersed in the cell (Gorain, Franzidis and Manlapig, 1996). This value is calculated for each measurement, for example:
Superficial gas velocity can be measured by various means,
20
ranging from the air flowmeter readings through to sophisticated Measurement 1: Jg = = 1.43 cm / s
instruments, and these will be discussed in this section. 14.0

98 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS

Water

Atmosphere

Compressed Air
Feed 3-way Valve
Mark 1

Mark 2

Perspex
Tube

Pinch
Valve
Nozzle

Tailings

FIG 5 - Schematic of JKMRC Jg probe and typical set-up.

All measurements are then averaged to give an average 80

experimental Jg value of 1.29 cm/s. The pressure difference needs
to be taken into account, as the measurement is taken within the Changing slope P1 P1
pulp phase, at a different height to the water level dropping.
PRESSURE (cm H2O)
60
The Anglo Platinum Bubble Sizer (APBS) also measures
superficial gas velocity, using the same concept as described for Difference
the JKMRC Jg probe. Measurements of time are recorded as the 40
water level drops known distances with corrections made for Correct slope
pressure and differences in tube diameters.
P2 – level changes
20 P2
McGill online Jg probe
As part of the ongoing research in the AMIRA P9 project,
researchers at McGill University have developed a probe to 0
measure superficial gas velocity online in industrial environments. 0 50 100 150
The probe consists of two hollow plastic tubes that are placed just
TIME (s)
below the pulp/froth interface (in the quiescent zone) similar to the
JKMRC Jg probe. A typical arrangement is shown in Figure 6. FIG 7 - Output from the McGill online Jg probe.

P1 P2
⎛1033(2 P − PO )⎞
⎜ ⎟
⎝ ρb ⎠ dP
Jg sensor Reference
Jg = (4)
PL dt

L
FLOTATION
MACHINE
FIG 6 - Schematic of McGill online Jg probe and typical
arrangement.
The operating principle is based on the increase in pressure as
the slurry in the first tube (shown as P1) is pushed down due to the
gas entering the probe from the flotation cell. There is a linear
relationship between pressure and time that can be related to the
superficial gas velocity. The second tube is monitored to account
for changes in cell pulp level and the pressure versus time
relationship is subtracted from the relationship obtained from the
first tube. Repeated opening and closing of the valves enables
continuous Jg measurements to be obtained. An example of the
output from this probe is given in Figure 7, with Equation 4 used
to calculate the superficial gas velocity.
where:
Po is the atmospheric pressure
PL is the pressure at the bottom of the tube which changes as
the level changes
b is the density of the aerated pulp
Typical superficial gas velocity values
Superficial gas velocity tends to range widely depending on the
operating philosophy of the plant. In some instances, values of
0.1 cm/s have been recorded, while at the other extreme, values
greater than 3 cm/s have also been measured for mechanical
flotation cells. In general, however, values within 1.0 cm/s to
2.0 cm/s are considered to be operating in a reasonable range.
Where Jg is measured at less than 1.0 cm/s, it is generally
recommended to investigate increasing the air flow rate to
improve the recovery in the pulp phase. However, where Jg is
higher than 3.0 cm/s, flooding can occur, resulting in pulp being
recovered in the concentrate launder. This indicates a physical

Flotation Plant Optimisation Spectrum Series 16 99

CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS
25
Column, Jameson, Flash
20
Frequency (%)
Conventional cells
15
10
5
0
0.25 0.5
0.75 1
1.25 1.5
1.75 2 2.25 2.5
2.75
Superficial gas velocity (cm/s)
FIG 8 - Range of superficial gas velocity values obtained from
conventional, column, Jameson and Flash flotation cells (from
Schwarz and Alexander, 2006).
limit to increasing air flow rates, although the resulting effect on
the froth phase performance should also be investigated. As a
guide, Jg measurements obtained from over 1000 flotation cells
are displayed in Figure 8.
Gas hold-up
Gas hold-up measures the volume fraction of air contained within
the pulp phase of the flotation cell. Generally, increasing the gas
content within the cell results in more bubbles per unit volume and
hence more available surface area for particle-bubble interactions
to occur, producing improved kinetics in the pulp phase (Ahmed
and Jameson, 1989). However, there is a maximum value of gas
hold-up before the cell capacity is limited, resulting in reduced
residence time. Again, there are various measurement techniques
available to measure both overall and local gas hold-up values, and
these are discussed in the following section.
Level change method
This method involves measuring the level of the pulp with gas
flowing into the cell and then measuring the pulp level when the
gas is switched off. The change in pulp level is used to calculate
the volume of gas, εg , in the flotation cell, according to Equation 5:
ΔH
εg =
H Using Equation 6, the gas hold-up from the first measurement can
where:
H is the change in pulp level after switching off the gas
H is the pulp level with gas
This method has the advantage that no specialised equipment is
necessary and it also gives the overall gas content within the
whole flotation cell. However, it can often be quite difficult to
obtain the initial measurement of the pulp level with gas present
in the cell.
The JKMRC gas hold-up probe
Probes were also developed to measure gas hold-up as part of the
AMIRA P9 project both at the JKMRC (this section) and at
McGill University (following section). The JKMRC gas hold-up
probe is shown schematically in Figure 9 and is described in
detail in Gorain, Franzidis and Manlapig (1995b) and Power,
Franzidis and Manlapig (2000). The early probes used pinch
100
Rubber tube Rubber tube
Plastic cylinder tube
Three-way valve
Air line
Compressed air
The atmosphere Pinch valve 1 Pinch valve 2
FIG 9 - Schematic of the JKMRC pneumatic gas hold-up probe.
valves operated by compressed air, while recent versions include
mechanical swing valves or spring valves.
3
This probe takes a sample of aerated slurry from just below the
pulp/froth interface. The probe consists of a vertical cylinder with
either pneumatic or mechanical valves at the top and bottom. The
probe is lowered into the flotation cell and slurry is allowed to
pass through the probe for more than ten seconds. After this time,
the top and bottom valves are rapidly closed and the probe lifted
out of the flotation cell. The volume of the slurry collected, Vs , is
measured and compared with the known volume of the probe
between the two valves, Vp . The gas hold-up is calculated from
Equation 6:
Vp − Vs
εg = (6)
Vp
This technique is simple and rapid, giving almost instantaneous
results. However, the results are for a localised region of the cell,
and measurements must be repeated across the cross-section of
the flotation cell to give the average gas hold-up across the whole
cell. It is generally recommended that at least three measurements
of gas hold-up are performed in each location within the cell to
ensure representative results.
Worked example
Repeat measurements have been made using a JKMRC gas
hold-up probe. The volume of the probe is 1070 ml. The
following is a list of slurry volumes obtained from the same
location in the flotation cell: 1000 ml, 1010 ml and 1000 ml.
Calculate the average gas hold-up from this position within the
flotation cell.
Calculation
(5)
be calculated as:
1070 − 1000
εg = = 6.5%
1070
Similarly the other measurements result in gas hold-up values
of 5.6 per cent and 6.5 per cent, respectively. Averaging these
gives an average gas hold-up value of that location within the
flotation cell of 6.2 per cent.
The McGill online gas hold-up probe
The McGill gas hold-up probe uses the differences in
conductivity obtained from slurry and gas. It consists of two
tubes, with the arrangement shown in Figure 10.
The open cell measures the conductivity of the slurry with
bubbles present, while the siphon cell measures the conductivity
of the slurry without any gas present. The gas hold-up is
calculated using Equation 7:
Spectrum Series 16 Flotation Plant Optimisation
 CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS

Open cell Syphon cell transfer coefficient between the probe and pulp/air allows the air
(slurry, no gas)
hold-up fraction to be calculated.

Light attenuation techniques

In the case of a clean aerated liquid, the light attenuation is a
function of the interfacial area of the air phase. A high intensity
laser source is directed through the aerated liquid, with received
light being detected via a photomultiplier tube. The amount of
light scatter is proportional to the air hold-up (Aslan et al, 2006).

Radiation techniques
FIG 10 - Schematic and arrangement of the McGill online gas
hold-up probe. Radiation sources have for many years been in use for purposes
of in-stream analysis. Gases, liquids and solids have different
adsorption coefficients of X-ray or γ-ray radiation or neutron
ko beam radiation. Radiation released by a radioactive source is
1−
ks passed through the phase being evaluated. The attenuation of
εg = (7) radiation can be related to the air hold-up (George et al, 2001).
k
1 + 0.5 o
ks
Typical gas hold-up values
where: In general, gas hold-up varies between six per cent and 25 per
cent for both mechanical and column flotation cells (Power,
ko is the specific conductivity of the slurry with bubbles Franzidis and Manlapig, 2000). However, in some instances
present values of less than two per cent have been measured, indicating
ks is the specific conductivity of the slurry with no bubbles insufficient air introduced to the cell. Values greater than 50 per
cent have also been recorded, a possible indication of high pulp
Again, this technique produces gas hold-up values for a given viscosity. The range of values obtained can be seen in Figure 11.
location within the flotation cell, and can be repeated across the
cell to give the average overall gas hold-up. This technique can
also be used for continuous monitoring of the gas content within 45
the cell. 40
Column, Jameson, Flash
Frequency (%)

35
Other air hold-up measurement techniques 30 Conventional cells
25
Pressure transducers 20
The use of pressure sensor measurements has been used for many 15
years to determine the pulp level in tanks. Dual sensors spaced a 10
set vertical distance apart can also be used to calculate the density 5
of aerated pulp. Should the pulp density be known, this then 0
allows the air hold-up to be calculated. The average air hold-up 6 12 18
can be calculated by sensors set into the tank wall. Air hold-up 24 30 36
42 48 54
at specific locations within the tank can be measured by 60 66 72
non-intrusive probes such as pitot tubes. Air hold-up (%)

Ultrasonic techniques
Doppler ultrasonic probes have been used for many years for
measuring liquid velocity in pipes. Their use is based on the
propagation of sound waves, generated within a piezo-ceramic
disc. Techniques can either rely on ultrasound transmittance
through an aerated pulp (Waniewski, Hunter and Brennen, 2001),
or reflectance from bubble surfaces (Kulmyrzaev, Cancelliere and
McClements, 2000).
Acoustic emission
Acoustic emission techniques operate with a similar principle to
the use of microphone recordings to determine a grinding mill
load. Pulp, with a specific air hold-up will generate a unique
acoustic signal which can be recorded through a hydrophone
(Boyd and Varley, 2004).
Heat transfer anemometers
Heat transfer probes operate on the principle that the heat
exchanged between an electrically heated probe varies between
that of the pulp and air contained within it. Knowing the heat
FIG 11 - Range of gas hold-up values obtained from conventional,
column, Jameson and Flash flotation cells (from Schwarz and
Alexander, 2006).
Bubble surface area flux
The bubble surface area flux, Sb, combines the superficial gas
velocity and bubble size to give a measure of the rate of bubble
surface area rising through a flotation cell per unit cross-sectional
area per unit time, according to Equation 8:
6 × Jg
Sb = (8)
db
There is a strong linear relationship between bubble surface area
flux and recovery across the pulp phase (Gorain, Franzidis and
Manlapig, 1997).
The bubble surface area flux can also be estimated using
impeller dimensions and speed as well as particular properties of
the pulp phase in an empirical relationship (Gorain, Franzidis and
Manlapig, 1999).

Flotation Plant Optimisation Spectrum Series 16 101

CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS

Also useful to note is that when treating the same type of ore,
this linear relationship is independent of impeller type, as shown
in Figure 12. The slope of the relationship is related to the
floatability of the ore – with higher slopes indicating faster
floating material.
recovery varies from ten to 90 per cent (Savassi et al, 1997).
Consequently, up to 50 per cent of the overall cell recovery can
be lost due to inefficiencies in the froth phase.
This section discusses three of the main techniques available to
measure froth recovery, and the advantages and disadvantages of
each.

Froth residence time method

Many authors (Engelbrecht and Woodburn, 1975; Laplante,
Toguri and Smith, 1983; Feteris, Frew and Jowett, 1987; Vera,
Franzidis and Manlapig, 1999) have calculated froth recovery (Rf)
from the ratio of the overall rate constant, k, to the collection zone
rate constant, kc, via:

k
Rf = (9)
kc

Various authors including Vera, Franzidis and Manlapig (1999)

FIG 12 - Linear relationship between flotation kinetics in the pulp
phase and bubble surface area flux (from Gorain, Franzidis and
Manlapig, 1997).
Typical bubble surface area flux values range from 30 s-1 to
70 s-1 for both mechanical and column flotation cells, as shown in
Figure 13.
indicated the value for kc by extrapolation of the overall rate
constant (k) versus froth residence time (τ f ) graph to the y-axis,
where τ f = 0 (the collection zone) (Feteris, Frew and Jowett,
1987; Gorain et al, 1998). Froth residence time can be defined in
a number of ways. The definition referred to in this chapter is
based on the concentrate flow rate. As such, the froth residence
time is defined as (Mathe, Harris and O’Connor, 2000):

(1 − ε g )Vfroth
30 FRT = (10)
Qconc
25 Column, Jameson, Flash
where:
Frequency (%)

Conventional cells
20
εg is the gas hold-up in the froth phase
15
Vfroth is the volume of the froth phase
10 Qconc is the volumetric flow rate of the concentrate
5 By varying the froth residence time, through either varying froth
height or airflow and measuring recovery, a set of flotation rate
0 constants at various froth residence times can be calculated and
10 20 30
40 50 60 plotted. A typical example of this is shown in Figure 14. From this
70 80 90 graph an estimated flotation rate constant (kc) of 0.13 min-1 is
100 110
120
-1 determined. At 20 seconds residence time the flotation rate
Bubble surface area flux (s ) constant is approximately 0.05 min-1. Therefore, at 20 seconds
froth residence time the froth recovery will be 38.5 per cent.
FIG 13 - Range of bubble surface area flux values obtained from
Flotation rate constant (1/min)

conventional, column, Jameson and Flash flotation cells (from

0.25
Schwarz and Alexander, 2006).
0.2
FROTH RECOVERY MEASUREMENTS
0.15
The main function of the froth phase in flotation is to transfer
the particles attached to bubbles from the pulp phase to the
concentrate launder. Some bubbles with particles attached to 0.1
them may undergo coalescence in this region, leading to
detachment of the most weakly attached particles (Gourram- 0.05
Badri, Conil and Morizot, 1997), which are usually gangue
mineral particles. There is then a subsequent upgrading of 0
0 20 40 60 80 100
the concentrate because the detached particles (which may
contaminate the concentrate) will drain back to the pulp phase. Froth residence time (sec)
The effect of froth recovery on the overall cell recovery is
FIG 14 - Typical variation of flotation rate constant with froth
significant. A particle may be collected in the pulp phase and
residence time.
subsequently report to the froth phase. However, if the froth
recovery is low, the particle will probably not be recovered in the
concentrate and drop back to the pulp. This produces a decrease In operations where the flotation circuit operates with flotation
in the overall flotation recovery. In general, collection zone (pulp cells of various sizes an adjustment based on the horizontal
phase) recovery varies from 60 - 99 per cent and froth phase distance the froth travels to the concentrate launder may be

102 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS

required. This is done by dividing the froth retention time by a
typical cell length as shown below; where τ fs may be thought of as
‘specific’ froth residence time which is independent of cell size:
FRT
τ fs = (11)
L
where:
τ fs = specific froth residence time (sec/m)
L = perpendicular distance from impeller to launder (m)
Mass balance technique
The froth residence time method can result in significant
disruption to operations, in order to gain the required information.
In order to overcome this, the mass balance technique was
developed (Alexander, Franzidis and Manlapig, 2003). This
technique is based on a number of key assumptions:
• a froth sample can be generated that represents the grade of
particles attached to bubbles,
• the average grade of particles attached to bubbles does not
change over the froth phase, and
• the grade of the tail sample is equivalent to the grade of
particles that are entrained in the water.
The technique is non-disruptive and involves taking samples of
feed, concentrate, tail, froth and air hold-up. These samples are
assayed and a mass balance of attached and entrained particles at
various size fractions is conducted, as discussed below.
The air hold-up sample taken directly below the pulp/froth
interface with the superficial gas velocity measurement is also
used to estimate the mass flow rate of attached particles entering
the froth phase, based on the following equation:
M attpulp = 3.6 ⋅ Qair ⋅ B pulp (12)
matt ⋅1000
B pulp = (14)
V
where:
matt = mass of attached particles in the air hold-up probe (g)
V = volume of air measured in the air hold-up probe (ml)
The mass of attached particles in the air hold-up probe is
estimated using the following mass balance equation:
mtotal ⋅ (Gtotal − Gent )
matt = (15)
(Gatt − Gent )
where:
mtotal = total mass of particles in the air hold-up probe (g)
Gent = grade of entrained particles in the air hold-up probe
Gtotal = grade of all particles in the air hold-up probe
Gatt = grade of attached particles in the air hold-up probe
In order to calculate the mass of attached particles (matt), the
grade of attached particles (Gatt) is taken as that of the grade of
the top of froth sample, while the grade of entrained particles
(Gent) is taken to be equivalent to the grade of the tailings. This is
in accordance with the assumptions made at the beginning of this
section.
The estimate of the mass of attached particles in the air hold-up
sample obtained from Equation 15 can then be used to calculate
the mass flow rate of attached particles into the froth phase
(Equation 12).
The mass flow rate of attached particles in the concentrate can
be calculated using a similar mass balance equation to Equation 15
applied to the concentrate sample:

conc
M total ⋅ (Gtotal − Gent )
where: conc
M att = (16)
pulp (Gatt − Gent )
M att = mass flow rate of attached particles in the pulp zone
(t/h)
where:
Qair = volumetric flow rate of air (m3/s)
conc
M att = mass flow rate of attached particles in the concentrate
Bpulp = bubble load in the pulp phase (grams of solids/litre of stream (t/h)
air) conc
M total = mass flow rate of all particles in the concentrate stream
The volumetric flow rate of air can be measured directly in each
(t/h)
cell or estimated using the superficial gas velocity measurements
across the cell cross-sectional area: Gent = grade of entrained particles in the concentrate

Qair = A ⋅ J g (13) Gtotal = concentrate grade

Gatt = grade of attached particles in the concentrate
where:
The froth recovery of attached particles can then be calculated
Jg = superficial gas velocity (cm/s) using the following equation:
A = open area of cell (m2) conc
M att
The bubble load is defined as the mass of attached solids per Rf = (17)
volume of air and involves obtaining a sample using the air M attpulp
hold-up probe and using the assumptions outlined above. Air
bubbles (with particles attached) and pulp (with particles where:
entrained) pass through the air hold-up sampler in the open
position. When the air hold-up probe is closed, the attached Rf = froth recovery of attached particles
particles become detached from the air bubbles and are included conc
M att = mass flow rate of attached particles in the concentrate
with the pulp containing entrained particles and water. The
stream (t/h)
volume of air and the mass and grade of the pulp sample are
noted. The bubble load can then be calculated using: M attpulp = mass flow rate of attached particles in the pulp (t/h)

Flotation Plant Optimisation Spectrum Series 16 103

CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS

Worked example
A flotation machine operates with an air rate of 396 m3/h treating
a feed of 6.5 t/h solids. The average air hold-up air volume is
83.3 ml, giving an air hold-up of 7.79 per cent and the mass of
solids in the air hold-up probe is 28.3 g. The following zinc
assays have been recorded: feed – 11.05 per cent Zn; concentrate
– 33.72 per cent Zn; tailing – 8.80 per cent Zn; air hold-up –
9.28 per cent Zn; froth – 36.1 per cent Zn. Calculate the froth
recovery.

Calculations
The overall concentrate produced can be calculated either by
concentrate measurements, or from the two product formula.
6.5 × (11.05 − 88
.)
conc
M total = = 0.59 t / h
(33.72 − 88.)

The total mass of attached particles in the air hold-up probe is:

matt = 28.3 ×
(9.28 − 88
.)
= 0.50 g
(361. − 88
.)

The bubble load is:

0.50 × 1000
B pulp = = 6.00 kg / m3 air
83.3
The mass of particles attached to bubbles entering the froth is:
M attpulp = 3.6 × 396 × 6.00 = 8555 kg / h = 8.56 t / h
The froth recovery is:
0.59
Rf = × 100 = 6.9%
8.56
Bubble load measurement techniques
As discussed above the mass of attached particles entering the
froth is a critical measurement in determining froth recovery.
Seaman, Franzidis and Manlapig (2004) developed a new
technique to measure this, requiring a bubble load collection
chamber suspended above the froth phase. The chamber has a
sampling tube inserted into the pulp which captures rising,
mineral-laden bubbles. These bubbles rise through the tube where
they enter a collection chamber, burst and release the particles.
The captured particles are weighed and volume of air from the
rising bubbles is measured to allow a bubble load to be
calculated.
Comparison of different froth recovery
techniques
It is the decision of the plant metallurgist as to which technique is
chosen to measure froth recovery. The froth residence time
method is relatively easy to perform, but requires many samples
(feed, concentrate and tails required at different froth heights with
more than three froth heights required to generate the
relationship) and is disruptive to normal production.
The mass balance technique does not disrupt normal plant
operation, and generates only five samples per cell (feed,
concentrate, tails, top of froth and air hold-up). The main
disadvantage with this method is requiring assays of the tails and
air hold-up samples to be different enough to perform the mass
balance of attached and unattached particles in the air hold-up
sample. This has shown to be problematic in certain scavenging
operations, although the technique generally works well in
roughing and cleaning duties.
FIG 15 - Arrangement of the bubble load device (from Seaman,
Franzidis and Manlapig, 2004).
The bubble load technique is also non-invasive and generates
only two samples per cell (concentrate and bubble load sample).
This technique is performed in conjunction with a general survey
to give the flow rates of the individual cell concentrate streams.
There have been some issues with the size of the sample
collected, with long sampling periods sometimes required to
obtain enough sample for sizing analysis to be performed.
The bubble load device has been shown to give similar froth
recovery measurements to the mass balance technique, but
significantly different results to the froth height or froth residence
time methods. This is mainly due to the different basis of
calculations – the froth residence time method is concerned with
processes occurring within the froth phase only, while the mass
balance and bubble load techniques consider processes occurring
at the pulp/froth interface as well. The majority of bubble-particle
detachment has been found to occur at the pulp/froth interface,
due to the sudden change in momentum of the bubble-particle
aggregates. Since the froth residence time method does not take
this section of the flotation system into account, the froth
recovery measurements are routinely higher than for the same
cell and operating conditions using the other techniques.
OTHER PLANT MEASUREMENTS
In a mechanical flotation cell, the size of the bubbles generated is
dependent on the impeller design and cell operating variables like
impeller speed and air flow rate (with chemical conditions fixed).
In general the mean bubble size decreases with increase in
impeller speed, whereas it increases with increase in air flow rate
(Gorain, Franzidis and Manlapig, 1995a, 1995b). The manner in
which the mean bubble size changes depends on the air
dispersion capability of the impeller. Previous investigation has
also shown that feed particle size (P80) affects the rheological
properties of slurry treated in the cell and therefore affects bubble
size (O’Connor, Randall and Goodall, 1990).

104 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS
Since the Sb generated in a cell is dependent on bubble size, it
is expected that Sb will also depend on impeller speed, air flow
rate, impeller design and the particle size of the feed to the cell.
In order to represent Sb mathematically in terms of these
variables, irrespective of impeller design and cell sizes, it is
useful to convert these variables into more ‘general terms’.
Impeller peripheral speed (Ns) instead of impeller speed (N)
and air flow rate per unit cell cross-sectional area (Q/A) instead
of air flow rate (Q) are suitable conversions which enable Sb to be
represented mathematically in terms of variables whose values
are independent of cell size and impeller design.
The production of Sb in flotation machines also depends on
impeller design. Impeller diameter in general increases with cell
size and also varies from one design to another. But for a
particular impeller type, the impeller aspect ratio (ratio of
impeller height to impeller diameter) is approximately the same
irrespective of cell size. So impeller aspect ratio (As) can be
considered as a suitable criterion to characterise impellers of
different design or type.
This can be expressed mathematically as:
C
⎛ Q⎞
S b = a ⋅ N sb ⋅ ⎜ ⎟ ⋅ A ds ⋅ P80e (18)
⎝ A⎠
where:
a, b, c, d and e are the parameters for the model
Standard units of the operating variables are used in the model
viz m/s for Ns, cm/s for Q/A, which in theory is equal to
superficial gas velocity, dimensionless for As and microns for P80
which is typically used to represent particle size in flotation
operations.
A database of cell characterisation values for over 2500
flotation machines has been developed at the Julius Kruttschnitt
Mineral Research Centre (JKMRC) for flotation cells of different
sizes and designs operated using a variety of impeller types, and
for industrial scale cells on operating plants. This allows use of
up to date parameter values. Initial investigations indicate that the
same general form can be used for the Jameson cell, using jet
velocity and the downcomer diameter to orifice plate diameter
ratio.
CONCLUSIONS
Measurement of gas dispersion characteristics in industrial
flotation plant cells has aided troubleshooting and identifying
problem areas such as worn impellers, non-uniform gas
dispersion, etc. Being able to measure these parameters quickly
and relatively easily has enabled plant metallurgists to make
adjustments to their cell operation and improve the plant
performance significantly.
Measurements of the froth performance has indicated that most
flotation cells operate at reasonably low froth recovery, with a
significant proportion of material that entered the froth phase
dropping back down to the pulp phase. These measurements have
also identified that the processes within the froth phase are still
relatively unknown, but quite significant in the overall flotation
cell performance.
While a great deal of flotation plant operation has developed
based on experience and instinct, it is envisaged that with the
various techniques now available to measure critical parameters
within the flotation cells, plant metallurgists will have a greater
understanding of the flotation process. Once the process is better
understood, there is a greater potential for improving and
operating the flotation cells at their optimum levels for the best
metallurgical performance possible.
Flotation Plant Optimisation Spectrum Series 16
REFERENCES
Ahmed, N and Jameson, G J, 1989. Flotation kinetics, Mineral
Processing and Extractive Metallurgy Review, 5:77-99.
Alexander, D J, Franzidis, J-P and Manlapig, E V, 2003. Froth recovery
measurement in plant scale flotation cells, Minerals Engineering,
16(11):1197-1203.
Aslan, M M, Crofcheck, C, Tao, D and Mengüç, M P, 2006. Evaluation
of micro-bubble size and gas hold-up in two-phase gas–liquid
columns via scattered light measurements, Journal of Quantitative
Spectroscopy and Radiative Transfer, 101(3):527-539.
Biswal, S K, Reddy, P S R and Bhaumik, S K, 1994. Bubble size
distribution in a flotation column, The Canadian Journal of Chemical
Engineering, 72(1):148-152.
Boyd, J W R and Varley, J, 2004. Acoustic emission measurement of low
velocity plunging jets to monitor bubble size, Chemical Engineering
Journal, 97:11-25.
Boyer, C, Duquenne, A-M and Wild, G, 2002. Measuring techniques in
gas-liquid and gas-liquid-solid reactors, Chemical Engineering
Science, 57:3185-3215.
Chen, F, Gomez, C O and Finch, J A, 2001. Technical note: Bubble size
measurement in flotation machines, Minerals Engineering,
14(4):427-432.
Diaz-Penafiel, P and Dobby, G S, 1994. Kinetic studies in flotation
columns: Bubble size effect, Minerals Engineering, 7(4):465-478.
Engelbrecht, J and Woodburn, E T, 1975. The effects of froth height,
aeration rate, and gas precipitation on flotation, Journal of the
Southern African Institute of Mining and Metallurgy, October,
pp 125-132.
Feteris, S M, Frew, J A and Jowett, A, 1987. Modelling the effect of froth
depth in flotation, International Journal of Mineral Processing,
20:121-135.
George, D L, Shollenberger, K A, Torczynski, J R, O’Hern, T J and
Ceccio, S L, 2001. Three-phase material distribution measurements
in a vertical flow using gamma-densitometry tomography and
electrical-impedance tomography, International Journal of
Multiphase Flow, 27:1903-1930.
Gorain, B K, Franzidis, J-P and Manlapig, E V, 1995a. Studies on impeller
type, impeller speed and air flow rate in an industrial scale flotation
cell – Part 1: Effect on bubble size distribution, Minerals Engineering,
8(6):615-635.
Gorain, B K, Franzidis, J-P and Manlapig, E V, 1995b. Studies on impeller
type, impeller speed and air flow rate in an industrial flotation cell –
Part 2: Effect on gas hold-up, Minerals Engineering, 8:1557-1570.
Gorain, B K, Franzidis, J-P and Manlapig, E V, 1996. Studies on impeller
type, impeller speed and air flow rate in an industrial flotation cell –
Part 3: Effect on superficial gas velocity, Minerals Engineering,
9:639-654.
Gorain, B K, Franzidis, J-P and Manlapig, E V, 1997. Studies on impeller
type, impeller speed and air flow rate in an industrial flotation cell –
Part 4: Effect of bubble surface area flux on flotation performance,
Minerals Engineering, 10:367-379.
Gorain, B K, Franzidis, J-P and Manlapig, E V, 1999. The empirical
prediction of bubble surface area flux in mechanical flotation
cells from cell design and operating data, Minerals Engineering,
12(3):309-322.
Gorain, B K, Napier-Munn, T J, Franzidis, J-P and Manlapig, E V, 1998.
Studies on impeller type, impeller speed and air flow rate in an
industrial scale flotation cell. Part 5: Validation of k-Sb relationship
and effect of froth depth, Minerals Engineering, 11(7):615-626.
Gourram-Badri, F, Conil, P and Morizot, G, 1997. Measurements of
selectivity due to coalescence between two mineralized bubbles and
characterisation of MIBC action on froth flotation, International
Journal of Mineral Processing, 51:197-208.
Hernandez-Aguilar, J R, Coleman, R G, Gomez, C O and Finch, J A, 2004.
A comparison between capillary and imaging techniques for sizing
bubbles in flotation systems, Minerals Engineering, 17(1):53-61.
Hernandez-Aguilar, J R, Gomez, C O and Finch, J A, 2002. A technique
for the direct measurement of bubble size distributions in industrial
flotation cells, in Proceedings 34th Annual Meeting of the Canadian
Mineral Processors of CIM, pp 389-402 (Canadian Institute of
Mining, Metallurgy and Petroleum: Montreal).
105
CHAPTER 5 – CHARACTERISATION MEASUREMENTS IN INDUSTRIAL FLOTATION CELLS
Jameson, G J and Allum, P, 1984. A survey of bubble sizes in industrial
flotation cells, report to AMIRA Ltd (Australian Mineral Industries
Research Association Limited).
Kulmyrzaev, A, Cancelliere, C and McClements, D J, 2000.
Characterisation of aerated foods using ultrasonic reflectance
spectroscopy, Journal of Food Engineering, pp 235-241.
Laplante, A R, Toguri, J M and Smith, H W, 1983. The effect of air flow
rate on the kinetics of flotation. Part 2: The transfer of material from
the froth over the cell lip, International Journal of Mineral
Processing, 11:221-234.
Magaud, F, Souhar, M and Boisson, N, 2001. Experimental study of
bubble column hydrodynamics, Chemical Engineering Science,
56:4597-4607.
Mathe, Z T, Harris, M C and O’Connor, C T, 2000. A review of methods
to model the froth phase in non-steady state flotation systems,
Minerals Engineering, 13(2):127-140.
O’Connor, C T, Randall, E W and Goodall, C M, 1990. Measurement of
the effects of physical and chemical variables on bubble size, Int J
Min Proc, 28:139-149.
Power, A, Franzidis, J-P and Manlapig, E V, 2000. The characterisation
of hydrodynamic conditions in industrial flotation cells, in
Proceedings Seventh Mill Operators’ Conference, pp 243-255 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Rodrigues, R T and Rubio, J, 2003. New basis for measuring the size
distribution of bubbles, Minerals Engineering, 16:757-765.
Sanaullah, K, Zaidi, S H and Hills, J H, 2001. A study of bubbly flow
using resistivity probes in a novel configuration, Chemical
Engineering Journal, 83:43-53.
Savassi, O N, Alexander, D J, Johnson, N W, Franzidis, J-P and
Manlapig, E V, 1997. Measurement of froth recovery of attached
particles in industrial flotation cells, in Proceedings Sixth Mill
Operators’ Conference, pp 149-155 (The Australasian Institute of
Mining and Metallurgy: Melbourne).
106 Spectrum Series 16
Schubert, H and Bischofberger, C, 1978. On the hydrodynamics of
flotation machines, International Journal of Mineral Processing,
5:131-142.
Schwarz, S and Alexander, D A, 2006. Gas dispersion measurements in
industrial flotation cells, Minerals Engineering, 19:554-560.
Seaman, D, Franzidis, J-P and Manlapig, E V, 2004. Bubble load
measurement in the pulp zone of industrial flotation machines – A
new device for determining the froth recovery of attached particles,
Mineral Processing and Extractive Metallurgy Review, 5:123-145.
Taute, J J and McClelland, A J, 2006. Introduction to the Anglo Platinum
Bubble Sizer, presented to the South African Institute of Mineral
Processing Conference, August.
Tucker, J P, Deglon, D A, Franzidis, J-P, Harris, M G and O’Connor, C T,
1994. An evolution of a direct method of bubble size distribution
measurement in a laboratory batch flotation cell, Minerals
Engineering, 7(5/6):667-680.
Vera, M A, Franzidis, J-P and Manlapig, E V, 1999. Simultaneous
determination of collection zone rate constant and froth zone recovery
in a mechanical flotation environment, Minerals Engineering,
12(10):1163-1176.
Waniewski, T A, Hunter, C and Brennen, C E, 2001. Bubble
measurements downstream of hydraulic jumps, International Journal
of Multiphase Flow, 27:1271-1284.
Wu, Q, Welter, K, McCreary, D and Reyes, J N, 2001. Theoretical studies
on the design criteria of double-sensor probe for the measurement of
bubble velocity, Flow Measurement and Instrumentation, 12:43-51.
Flotation Plant Optimisation
 HOME

CHAPTER 6

Chemical Measurements During Plant Surveys and Their

Interpretation
Stephen Grano

MAusIMM, Professor, Ian Wark Research Institute, University of South Australia, Mawson Lakes
Campus, Mawson Lakes SA 5095. Email: stephen.grano@unisa.edu.au

Stephen Grano received his BEng (Metallurgical) in 1981 from the University of Queensland and his PhD in
1997 from the University of South Australia. Early in his career, Stephen Grano gained hands-on experience
at Mount Isa Mines Ltd, working in the copper and lead/zinc concentrators and heavy medium plant. His
research interests include: evaluation and diagnosis of mineral flotation plant performance; technology
transfer; grinding and pulp chemistry; fine and coarse, composite particle flotation; process development for
the separation of sulfide minerals; adsorption and modification of sulfide mineral surfaces; non-sulfide
mineral flotation; and flotation froths. A particular focus is the successful transfer of scientific research
outcomes into engineering practice through structured research and development programs within
industry-university partnerships.

Abstract
Introduction
Chemical Measurements During a Plant Survey
Analysis and Interpretation of Survey Chemical Data
Case Study – Zinc Reverse Process
Conclusions
References

ABSTRACT CHEMICAL MEASUREMENTS DURING A

This chapter addresses simple chemical measurements and their
interpretation in the optimisation of base metal sulfide flotation plants. The
measurements include pH, Eh, dissolved oxygen, dissolved oxygen
demand, extractable metals, and solution species. For each measurement,
the basic theory and practical examples of application are discussed.
Simple hydrolysis and Eh/pH diagrams are described which aid
interpretation. The key sampling points and methodology for chemical
measurements during a plant survey are outlined. The critical importance
and method of correlating the laboratory conditions and performance to the
plant is discussed. A practical case study which highlights the utility of
these chemical measurements for process troubleshooting is discussed.
INTRODUCTION
This chapter is divided into the following three sections:
1. chemical measurements during a plant survey,
2. analysis and interpretation of these chemical measurements,
and
3. a practical case study which highlights the utility of these
chemical measurements for process troubleshooting.
PLANT SURVEY
In this section, basic chemical measurements taken during a typical
plant survey will be outlined. The chemical measurements include:
• pH,
• Eh,
• dissolved oxygen,
• dissolved oxygen demand,
• ethylene diamine tetra-acetic acid (EDTA) extractable metal
ions, and
• solution ionic species.
For each measurement type, a brief theoretical background will
firstly be discussed. This will then be followed by a brief
discussion on the importance of the measurement, how the
measurement may be made, and some practical guidelines.
The combined measurement of Eh and pH will, to a significant
extent, define the thermodynamic stability of bulk species
(Garrels and Christ, 1965). This is because most equilibriums are

Flotation Plant Optimisation Spectrum Series 16 107

CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION

principally controlled by pH (eg hydrolysis) and Eh (eg
oxidation/reduction). The other important variable when
considering the predominant equilibrium species is concentration
(eg the total metal ion or collector concentration in the system),
though this parameter may be very difficult to define and
measure. However, the strict application of thermodynamic
equilibria is often not appropriate for reactions that have not
come to equilibrium, or at a mineral surface where the interfacial
pH and concentrations may not be at their bulk solution values
(Woods, 1984). However, putting aside these considerations, it is
still very valuable to measure both pH and Eh during plant
surveys and to use these measurements to ‘predict’ species
predominance (Natarajan and Iwasaki, 1973a). Several examples
are presented which shows the utility of such an approach. It is
always important to measure both pH and Eh at key sample
points, such as the ball mill discharge, flotation feed and process
water in a plant survey.
pH
This is the most common and easiest to measure of the chemical
measurements. pH is relatively easy to measure as it does not rely
on particle contact with the electrode to ensure an accurate
measurement, unlike the case of Eh, which does. The definition
of pH is given below (Garrels and Christ, 1965):
pH = -log [H+]
This definition is a formal one as it involves the activity of a
single ion, a quantity that cannot be measured directly. As a
result, the measured pH is defined in terms of an operational
procedure. For all practical purposes, the measurement of pH is
accomplished using an electromotive force (EMF) cell that
consists of a glass electrode dipping into the solution, together
with a reference electrode to complete the circuit. The reference
electrode is usually a mercury-mercurous chloride (calomel)
electrode in saturated KCl solution, connected to the solution to
be measured by a salt bridge of saturated KCl solution. The glass
electrode consists of a glass bulb containing an acid solution, and
an inner electrode of fixed potential, usually Ag-AgCl, to conduct
electrons reversibly, into and out of solution. When the bulb is
immersed in a solution, an electrical potential is developed
between the inner and outer solutions, which is proportional to
log [H+] in the external solution. The reference electrode has a
constant potential, whilst that of the glass electrode will vary with
[H+]. The overall cell may be represented as: glass electrode, sol.
x | KCl(sat.), AgCl (s); Ag (l).
The half-cell (electrode) potential of the glass electrode is
given as:
2.303 RT
E x' = C H + log [H + ]
pH value, pHs, the EMF measured in the test solution, at the same
temperature, is given by:
2.303 RT
E x = CH − pHx − E ref
F
2.303 RT
E s = CH − pHs − E ref
F
Subtracting these potentials and rearranging gives:
F(Ex − Es)
pHx = pHs −
2.303RT
This equation is the basis for the operational definition of pH.
The pH is dependent upon temperature. Thus, it is always
important to measure the temperature of the pulp as well as pH.
Temperature will also affect the oxidation of minerals,
decomposition reactions, solubility of gases in the slurry, and
precipitation from solution. For example, the temperature of a
ball mill discharge stream can be up to 70°C in some climates,
which can greatly accelerate interaction of thiol collectors with
sulfur-oxy depressants (Sheldon and Johnson, 1988) and may also
affect flotation (O’Connor, Dunne and Botelho De Sousa, 1984).
Buffer solutions are available for calibrating the pH electrode,
which, in effect, fix the values of Es and pHs. The pH meter must
be standardised with buffer solutions at the same temperature as
that of the test solution to be measured. Most instruments have a
temperature compensator that can be adjusted to the temperature
of the test solution. This adjustment will have the effect of putting
the correct value of the slope (ie 2.303 RT / F) into the pH meter.
For a well-calibrated pH electrode, the significance of the
measurement can be ±0.01 pH units. However, for slurry
measurements it is usual to determine the pH at a significance
±0.2, or at best ±0.1 pH units. In general, electrode response to
homogenous solutions is rapid, but between readings the glass
electrode should be rinsed several times with distilled water to
remove adsorbed ions. For pH measurement in slurries in
industrial flotation cells the electrode should be placed into the
pulp through the froth so that the tip of the electrode is deep
enough to be covered with the pulp. Furthermore, using
appropriate sampling equipment, transfer the slurry to an empty
bucket or vessel and while stirring, record the pH and compare
the two values.
The importance of pH in controlling the hydrolysis of metal
ions is demonstrated in Figure 1. If a system contains 10-5 mol
dm-3 of copper at pH 4, the predominant copper species will be
Cu2+ (aq) and CuCO3 (aq) in solution. Increasing the pH to 7
would give rise to the onset of Cu(OH)2 (s) precipitation. By pH 8
2+ 2+ 4-
Cu2(OH)2 Cu Cu(CO3) 3
-log Concentration/mol dm -3

F 0

ie: -1
2.303 RT -2 2-
E x' = C H − pH Cu(OH) 2 (s) Cu(CO3) 2
F -3
where: -4 +
F is Faradays constant (colombs per mole of electrons) Cu(OH)
-5
The overall cell potential is given by the difference of the
two-half cell potentials: -6
CuCO3 (aq) 2-
2.303 RT -7 Cu(OH)4
E x = E x '− E ref = CH − pH - E ref
F 2 4 6 8 10 12
If a given experimental arrangement as represented by the pH
above cell is used firstly to determine the EMF developed by a
solution whose pH (pHx) is to be measured, and then the test FIG 1 - Solubility diagram for the Cu (II)-H2O-CO2 system.
solution is replaced by a standard reference solution of assigned [CO2]tot = 10-3 mol dm-3.

108 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION
- 8.5, virtually all the copper in the system will be precipitated as
Cu(OH)2 (s) with a small amount of CuCO3 (aq) and Cu2+ (aq) in
solution. The extent of metal ion precipitation in a particular
system can be measured by extraction, while the metal ions in
solution can also be measured by separating the solid phase from
the aqueous phase. These methods are discussed further below. It
is very important to know what species are in solution at specific
points in the circuit. For example, a thickener overflow, which
contributes to the process water, containing 5 ppm of copper at
pH 7, is sufficient to inadvertently activate sphalerite during
galena flotation.
Increasing the pH further will dissolve Cu(OH)2 (s) as
Cu(CO3)22-. It is important to note that Cu(OH)2 (s) will still be
present in the system to pH values greater than ten, though
equilibrium considerations suggests it will be at a lower
concentration. Because the Cu(OH)2 (s) is in equilibrium with
negatively charged species in solution, the zeta potential of
the Cu(OH)2 (s) will be negative at pH values greater than
approximately 8 - 8.5 (pHiep). To dissolve copper hydroxide it is
necessary to decrease the pH to less than ~6, though this will
depend on the total concentration of copper in solution. Copper in
solution may be derived from the oxidation of copper minerals,
which is dependent upon the pulp Eh, while hydrolysis of the
copper is pH dependent. The adsorption of copper and iron onto
copper sulfide minerals controls their flotation properties to a
very large extent. An excellent case study of the effect of pH on
iron hydrolysis and the zeta potential of pyrite is given by
Fornasiero, Eijt and Ralston (1992), while the ensuing effect of
pH on collector adsorption is discussed by Montalti, Fornasiero
and Ralston (1991). Kristall et al (1994) has described the
influence of pH on the selective dissolution of zinc hydroxide on
sphalerite while leaving iron (III) hydroxide on pyrite intact in a
copper/zinc flotation plant.
Eh
The oxidation potential, Eh, is defined as the potential of a
half-cell, referred to the standard hydrogen half-cell, the EMF of
the standard half-cell being taken as zero at all temperatures, by
definition (Garrels and Christ, 1965). Oxidation potential is
measured with an electrode pair consisting of an inert electrode
and a reference electrode. The same reference electrode is used
for most Eh measurements as for pH measurements – the
saturated calomel. The inert electrode used most commonly is
bright platinum, although gold electrodes are also used. The role
of the calomel electrode is to supply a known EMF and to make
electrical connection with the system to be measured. The inert
electrode acts as an electron acceptor or donor to the ions in the
measured solution. When connected to the calomel electrode, the
platinum electrode can accept electrons from dissolved ionic
species, or it can give them up, depending upon whether the
potential of the half-cell involving the dissolved species is greater
or less than that of the calomel reference electrode.
The half-cell reaction for every oxidation-reduction system can
be written as:
Oxidised State + ne- = Reduced State
An example is:
Fe3+ + e- = Fe2+
The half-cell potential is given by:
RT Oxidised state
Eh = E o + ln
nF Reduced state
The reference electrode simply acts as a half-cell of constant
potential (at a constant temperature) and the overall cell EMF is
given by:
Flotation Plant Optimisation Spectrum Series 16
EMFcell = Eh Red/Ox – Eh ref
As an example, it is possible to calculate, using the Nernst
equation, the predicted EMF of the cell:
Pt (inert) | Fe3+ (aq) 0.01 M || Hg2Cl2 | Hg
| Fe2+ (aq) 0.001 M || KCl aq (sat) |
The two half-cell reactions and potentials are:
2Fe2+ (aq) = 2Fe3+ (aq) + 2e- Eh Fe2+, Fe3+
- -
2Hg + 2Cl (aq) = Hg2Cl2 + 2e Eh ref
Subtracting these two equations gives:
2Fe2+ (aq) + Hg2Cl2 = 2Fe3+ (aq) + 2Hg + 2Cl- (aq)
EMF cell = Eh Fe2+, Fe3+ - Eh ref
The value of Eh ref is constant at a given temperature and
pressure, and for the saturated calomel cell (Hg2Cl2) at 25°C is
0.2444 v. The half-cell potential for the Fe2+, Fe3+ couple is given
by:
Eh Fe2 + , Fe3 + = E °Fe2+ , Fe3 + 0.0592 1n [Fe 3+ ]2
2 [Fe 2 + ]2
Eo Fe2+, Fe3+ = 0.771 v, gives:
0.01
Eh Fe2 + , Fe3 + = 0.771 + 0.0592 ln
0.001
Eh Fe2+, Fe3+ = 0.830 v
Therefore:
EMF cell = 0.830 - 0.244 = 0.586 v
The measured potential of an inert electrode will be close to the
‘reversible’ potential calculated by the Nernst equation, provided:
1. All the species involved in the oxidation/reduction are in
equilibrium.
2. The indicator electrode is truly an inert electrode and does
not itself become oxidised or reduced.
3. There is only one couple, which dominates the potential
response. Often there are two or more redox couples, such
as the reduction of water, which may influence the potential.
In this case, the developed potential is referred to as a
‘mixed’ potential (Rand and Woods, 1984; Natarajan and
Iwasaki, 1974).
In Eh measurements most of the difficulties result from
contamination of the platinum electrode (Natarajan and Iwasaki,
1973b). It is useful to have two indicator or inert electrodes set up
with the same reference electrode to check for differences due to
contamination. Solutions containing sulfide ion often ‘poison’ Eh
electrodes. Eh measurements under ideal conditions are probably
only reproducible to ±5 mV, while for measurements in slurries it
is only reproducible to ±10 mV. In slurries, the Eh will also depend
upon contact of the inert electrode with conducting particles of
sulfide minerals (Rand and Woods, 1984). It is important to
maintain a reproducible technique to measure the Eh.
The degree to which the oxidised species predominates over the
reduced species depends on the difference between the potential of
the system (Eh) and the Eo (Figure 2). The Eo is the standard
potential and is the potential at which all species are at unit
concentration. For single electron transfer reactions (n = 1), if the
potential is more than 60 mV greater than the Eo, then 90 per cent
of the species considered will probably be in the oxidised
state (Figure 2), assuming equilibrium has been achieved. If the
109
CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION

100 1.2
90 PO
1 2 =1
80 PO PH
0.8 2 = 10 -2 2 =10 -41
0 .6
70
% Oxidised

PH

Eh versus SHE (V)

n=1 0.6 P 2 = 10 -3
60 O2 =
10 -40 1. 6
n=2
50 0.4 PO PH
n=3 2= 2 =10 -4
10 -40 1.6
40 0.2 P
PO H
2 =10 -1
30 2 = 10 - 8 1.6
3.1
0
20 PH
2 =1
10 Eo -0.2
0 -0.4
-120 -80 -40 0 40 80 120
Potential (mV) from Eo -0.6

FIG 2 - Degree of oxidation ( [Oxid]/([Oxid]+[Red])) as a function of
potential difference from the standard potential, Eo, for reactions
having n = 1, 2, or 3 electrons involved.
potential is more than 60 mV less than the Eo, then 90 per cent
of the species considered will probably be in the reduced state
(Figure 2), again assuming equilibrium has been achieved. In the
above case of the Fe2+, Fe3+ couple, over 90 per cent of the iron is
in the Fe3+ oxidation state at a potential 50 mV greater than the
standard half-cell potential.
For systems which are open to the atmosphere, the reduction of
oxygen at the inert electrode is very important. For example,
slurry exiting a ball mill may be in a highly reducing state in situ
of the pulp, but with sampling and stirring the Eh of the sample
can increase due to the introduction of air. The Eh of the plant pulp
can also increase through pumping, cycloning and its introduction
to flotation. The reduction of oxygen can be written as:
2H2O = O2 + 4H+ + 4e-
0.059 PO2 [H+ ]4
Eh = E O + Log
4 [H2O]2
Substituting -pH for [H+], the activity of pure water being
unity, and with Eo = 1.23 v yields:
0.059
Eh = 1.23 − 0.059 pH + Log PO 2
4 is because of:
-0.8
0 2 4 6 8 10 12 14
pH
FIG 3 - Stability limits of water as a function of Eh and pH at 25°C.
Contours show partial pressure of oxygen and hydrogen at
intermediate Eh values.
Finally, it is possible to summarise the various redox and
hydrolysis reactions for a system in the form of an Eh/pH
diagram (Figure 4 in the case of iron). In the case of vertical lines
‘A’ and ‘E’, these represent the equilibrium between Fe3+/Fe
(OH)3 and Fe2+/Fe(OH)2, respectively. These equilibriums are
hydrolysis reactions only and so do not depend upon Eh, only pH.
In the case of horizontal lines ‘B’ and ‘C’, these represent the
equilibrium between Fe3+/Fe2+ and Feo/Fe2+, respectively. These
equilibriums are redox reactions only and so do not depend upon
pH, only Eh. In the case of lines ‘D’, ‘F’ and ‘G’, these represent
the equilibrium between Fe2+/Fe(OH)3, Fe(OH)2/Fe(OH)3 and Feo
/Fe(OH)2, respectively. These equilibriums involve both redox
and hydrolysis reactions and therefore have slopes which are
neither simply horizontal nor vertical. For the case of a ball mill
discharge (Figure 4) in which there is oxidation of iron, redox
reactions involving this species will tend to dominate. This is
reflected in the low Eh (-200 mV, pH 7) of this stream. Fe is
stable at low Eh values and will tend to decrease the Eh.
However, with pumping and cycloning, the introduction of air
into the pulp raises the Eh at the flotation feed sample point. This

At a partial pressure of oxygen of one atmosphere, the PO2 1. the removal of the slurry away from iron in the ball mill; and
term disappears. The equation can be used to plot a line on an 2. the introduction of another couple, O2/H2O.
Eh-pH diagram for any chosen pressure of oxygen (Figure 3). For
potentials greater than the PO2 = 1 atm line (eg pH 10, Eh 1 v), 1.5
water is not stable and oxygen will bubble out of solution. The Eh ‘A’
is dependent upon the partial pressure of oxygen in the system. At 1 Fe 3+
very low potentials (eg pH 10, Eh – 0.7 v) water is not stable and
O2
hydrogen will start to bubble out of solution. ‘B’
‘D’ Fe (OH)3
0.5
Eh versus SHE (V)

It is very useful to plot the experimentally measured values of Flotation

pH and Eh, as shown in Figure 3, from a plant survey or flotation Fe 2+ Feed
test, or across different flotation tests under different conditions to 0
H2 Ball Mill
isolate whether a change in Eh is due to a change in pH or is due Discharge ‘E’
‘F’
to a change in concentration of an oxidising/reducing species -0.5 Fe (OH)2
(Johnson, 1988). For example, Grano, Johnson and Ralston ‘C’
Fe ‘G’
(1997) has measured the Eh of a laboratory mill discharge as a -1
function of pH to demonstrate that the pulp Eh becomes more 0 2 4 6 8 10 12
oxidising with increasing soda ash addition (pH) to the mill. The pH
approach is discussed further in the case study below and in the
next section on the interpretation of chemical measurements. It is FIG 4 - Eh/pH diagram for the Fe system. [Fe]tot = 0.01 M. Also
clear that it is also very important to measure the dissolved shown are values of Eh and pH taken from an Australian gold
oxygen in the pulp. This is discussed further below. mine. The stability limits of water are also shown.

110 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION

It is possible to show that the Eh increase is not due to a change
in pH over these two sample points, as the Eh increases without
any appreciable change in pH. The change in Eh shows there
must also be a change in the dissolved oxygen content of the pulp
(Figure 3). The initially depressed Eh in flotation, and changing
Eh with aeration, have consequences in flotation. Johnson, Jowett
and Heyes (1982) have shown that the initial flotation rate of the
valuable mineral can be retarded due to the low Eh in the initial
stages of flotation. With aeration, the Eh increases above a critical
value for collector adsorption. This is a consequence of the Eh
dependency of sulfide mineral flotation with thiol collectors
(Richardson and Walker, 1985).
Dissolved oxygen
Dissolved oxygen is the next most important variable to
measure. However, its value in a sampled pulp is often not at
steady state. This is because of the continuous removal of
oxygen from the pulp due to mineral and media oxidation. The
introduction of oxygen in flotation systems raises the Eh of the
pulps. Thus, when a sample is removed from a pulp stream, the
pulp is being taken away from the dynamic equilibrium, which
had been established between the removal of oxygen in solution
and the introduction of oxygen. In some cases, it may be
worthwhile to attempt to measure the dissolved oxygen (and Eh)
in situ of the pulp.
The dissolved oxygen electrode consists of a gold cathode and
Ag/AgCl anode, placed in an electrolyte solution. The solution is
contained behind a plastic membrane, usually polyethylene. A
polarising potential of about 800 mV is applied between the two
electrodes. The gold electrode is placed close to the membrane
and because of the polarising potential; oxygen diffusing through
the membrane will be reduced at the gold electrode:
O2 + 2H+ + 2e- = H2O2
The reduction of oxygen will produce a current through the
electrode. A load thermistor situated in the electrode itself
converts this current into a voltage proportional to the oxygen
partial pressure. The thermistor has a temperature coefficient,
which can compensate for changes in ambient temperature,
though this will depend upon the thickness and type of plastic
membrane. It is therefore always important to use the correct
plastic membrane that is matched to the particular electrode. In
some ore systems containing galena/sphalerite and large
quantities of pyrrhotite, the critical variable appears to be
dissolved oxygen, rather than Eh, due to the oxidising influence
of Fe3+ on Eh. In this case, collectors may be used which are able
to adsorb onto galena in a low oxygen environment (eg alkyl or
aryl phosphinates), or alternatively oxygen may be supplied to
the pulp to oxidise pyrrhotite and increase the Eh, allowing
conventional xanthate collector to be used.
Dissolved oxygen demand
The dissolved oxygen demand test (DO demand) has been used to
determine the rate of oxidation of minerals in pulps (Spira and
Rosenblum, 1978). It is a powerful in situ method that yields
information on the average oxidation state of the minerals. A
device to measure the DO demand is shown in Figure 5. The
dissolved oxygen demand is measured in the demand chamber
after a specific purging period in the supply chamber. As
oxidation of the minerals and grinding media proceeds in the
supply chamber, the rate of oxygen removal from the pulp is
reduced (Figure 6). With no initial oxygen purging there is a rapid
depletion of oxygen from the initial value (~6.5 ppm) to virtually
‘nil’ oxygen in solution (~0.8 ppm). With 120 minutes oxygen

Dissolved oxygen probe

Supply chamber

Slurry level
Impeller
Impeller

Gas bubbles

Eh, pH and dissolved oxygen Air inlet line, connected to

probes Valve pressurised air or oxygen

Slurry transfer tube

~150 mm diameter stainless steel frit

with 13 micron pores

Valve
Demand chamber
Slurry discharge tube

FIG 5 - Device for measuring dissolved oxygen demand. The slurry is first introduced into the supply chamber, the agitator turned on, and
oxygen or air purged into the slurry at a known rate. The dissolved oxygen will rapidly reach a steady state value, at which point a sample
of the slurry is transferred to the demand chamber where the dissolved oxygen depletion in solution is measured with time. The rate of
depletion is termed the dissolved oxygen demand. Thus, the rate of removal of oxygen from the pulp can be measured after specific
amounts of oxygen have been supplied to the pulp in the supply chamber.

Flotation Plant Optimisation Spectrum Series 16 111

CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION

8 where:
O2 Concentration (ppm)

7
[O2]o is the initial concentration of oxygen at t = 0 minutes
6 when the oxygen purging gas is turned off
5
[O2]eqm is the concentration of oxygen in the liquid when it is
4
in equilibrium with the atmosphere presented to it,
3 having an approximate value of 8 mg/L when it is in
2 equilibrium with air
1 When k2 = 0 (ie the demand chamber is closed to the
0 atmosphere) the oxygen concentration is simply given by:
0 30 60 90 120 150 180
[O2 ]t = [O2 ]o exp − ( k 1 )t
Aeration Time (minutes) in the Supply Chamber
This equation has been used to fit the experimental datapoints
FIG 6 - Dissolved oxygen concentration as a function of time, shown in Figure 6, and the value of rate constant k1 determined
measured in the demand chamber, after oxygen purging the slurry for the various amounts (ie purging time) of oxygen delivered. It
for different periods of time in the supply chamber. For each is possible to quantify the oxygen delivered with air delivery time
determination of oxygen demand, the initial oxygen concentration in the supply chamber by measuring the rate of oxygen
was set to approximately 6 - 8 mg/L with air purging in the supply dissolution in the liquid phase only at the experimental
chamber, then the oxygen discontinued by transferring a sample of temperature and pressure. The value of this rate constant is
the slurry to the demand chamber and the oxygen concentration plotted against time of air purging in the supply chamber
measured with time. Slurry sample is grinding cyclone overflow. (Figure 8). As the time increases the rate of oxygen depletion
decreases. These experiments also show the effect of adding iron
to the system. Adding iron increases the oxygen demand of the
purging there is slow depletion from the initial value (~8 ppm), slurry. The rate of decrease of oxygen demand with aeration time
which then decreases to a non-zero value. This implies that the is similar for the ore without iron and with iron, suggesting that
rate of oxygen removal by oxidation at that stage is being the rate of removal of oxygen is controlled by media oxidation.
balanced by oxygen being introduced into the solution through its
equilibrium with air. When the oxygen purging gas is 4.0
discontinued for the DO depletion stage then that flow is PCO
First order rate constant (minutes -1)

3.5
discontinued and there is removal of oxygen from solution due to No Fe
fixation on minerals and media debris (rate k1), but also due to 3.0 3kg/t Fe
equilibrium with the atmosphere (rate k2) (Figure 7). When the 6kg/t Fe
2.5
concentration of oxygen in the pulp is greater than the 15kg/t Fe
equilibrium value [O2]eqm, oxygen is removed at a rate of -k2. 2.0
When the concentration of oxygen in the pulp is less than the
1.5
equilibrium value, oxygen is dissolved into solution at a rate of
k2. This analysis assumes the rate of oxygen dissolution is equal 1.0
to the rate of oxygen nucleation. 0.5

d[O2 ]t 0.0
= − k1 .[O2 ]t − k2 .([O2 ]t − [O2 ]eqm) 0 20 40 60 80 100 120 140
dt Aeration time (minutes)

d[O2 ]t
= −( k1 + k2 ).[O2 ]t + k2 .[O2 ]eqm FIG 8 - Rate of oxygen removal from solution (k1) as a function of
dt aeration time in the supply chamber. The aeration time in the
supply chamber can be shown also as the amount of oxygen
Integration yields:
delivered to the pulp solution. The various experiments are for a
plant cyclone overflow (PCO), a sample of ore taken during the
k2 .[O2 ]eqm ⎛ k .[O ]eqm⎞
[O2 ]t = + ⎜[O2 ]o − 2 2 ⎟ exp − ( k 1 + k 2 )t measurement of the PCO and ground in a laboratory mill (no Fe),
( k1 + k2 ) ⎝ ( k1 + k2 ) ⎠ and with various addition of iron to the laboratory mill.

Oxygen Fixed to Mineral

k1 or Media Debris;
DO in O2oxid|t
Oxygen
Solution,
introduced via
O2|t
the Purging
Gas

Oxygen in
Atmosphere in
k2 Equilibrium with
Oxygen in Liquid,
O2atmos|t

FIG 7 - Mass balance on oxygen during the oxygen depletion stage of the dissolved oxygen demand test.

112 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION
Ethylene diamine tetra-acetic acid (EDTA)
extractable metal ions
Ethylene diamine tetra-acetic acid is a powerful chelating reagent.
It can be used as a diagnostic tool for the determination of the
amount of oxidation products on the surface of sulfide minerals
(Kant, Rao and Finch, 1994; Rumball and Richmond, 1996).
HO2C − CH2 CH2 − CO2 H
NH − CH2 − CH2 − NH EDTA
−
O2C − CH2 CH2 − CO2 −
The very high stability of EDTA-metal complexes can be
explained in terms of the values of the formation constant, K, for
the complexes (Table 1). This means that there is a strong driving
force for the formation of metal complexes with EDTA. Note that
complexes of EDTA have an intermediate stability between the
sulfide and hydroxide. This means that the metal hydroxides
adsorbed or precipitated on the mineral surface will be dissolved
with EDTA, but that EDTA will not dissolve the unoxidised metal
sulfide. This statement is true, provided that EDTA does not
catalyse the oxidation of the sulfide minerals. In this method, the
slurry sample is first purged of oxygen to reduce oxidation during
the EDTA extraction stage. A prepurged solution of EDTA is then
added to the slurry to make an overall concentration of ~10-2 M.
It is important that the EDTA concentration be high enough to
ensure the soluble species which can be complexed by EDTA (eg
calcium) do not consume EDTA, such that there is sufficient
available to complex with the products of surface oxidation. The
amount of EDTA metal ions is determined by filtration and
measurement of the solution phase, with mass balancing. It is
very important to quote the EDTA extractable metal ions as both:
1. the concentration of EDTA-metal ions taking into account
both sample dilution and metal ions already in solution prior
to extraction, and
2. the per cent EDTA-metal of the total metal in the head
sample.
TABLE 1
Stability constants, K, for various complexes with various
metal ions.
log K S2- OH- EDTA4-
Cu2+ 36.0 6.3 18.8
Zn2+ 25.1 5.0 16.7
Pb2+ 28.8 6.1 18.0
Fe2+ 17.9 4.5 14.3
Fe3+ 11.8 25.1
Solution ionic species
It is very important to measure the dissolved species as well as
the precipitated species using EDTA. This is because it aids in the
calculation of EDTA-metal ions correctly, but also it is a useful
measurement in its own right. At most alkaline pH values, metal
ions have precipitated and so are usually only measurable by
EDTA extraction. However, elements and species such as Ca, Mg,
Na, Cl, K, SO42-, SxOy2 – (eg thiosulfate) species are all important.
It is important to measure the total dissolved solids (TDS) and to
account for most of the species in solution by mass balance. The
heavy metal cations and others should be measured by filtration
(to 0.45 μm) and acidification by HCl to prevent precipitation
Flotation Plant Optimisation Spectrum Series 16
after sampling. Woodcock and Jones, 1970a, 1970b) have
conducted comprehensive pulp chemistry measurements in many
sulfide flotation plants. In the case of SxOy2 – special preservation
techniques are required to prevent oxidation. It is also possible to
measure carbonate, but this needs to be made locally as carbonate
ion will equilibrate rapidly with the atmosphere. Other important
species include total organic carbon (TOC), which may be a
measure of residual reagents in solution. It is possible to measure
the concentration of collector, frother and polymers in solution
using special techniques. Jones and Woodcock (1984) have
reviewed the measurement and reaction pathways of xanthate
derivates in solution.
ANALYSIS AND INTERPRETATION OF SURVEY
CHEMICAL DATA
In this section, a typical survey and key sample points for
measurements are discussed. The importance of assimilating
laboratory conditions to plant conditions is also outlined.
Following this, a typical example of measurements taken from a
plant and their interpretation will be discussed.
A typical, comprehensive, plant survey should be organised in
the following way:
1. Metallurgical sampling including sampling for per cent
solids and solids composition (bulk assays) at the block level
(ie rougher block, cleaner block) and/or the cell-by-cell level.
This could also include the measurement of lip froth flows
and froth recovery for specific cells (eg froth recovery of
scavenger cells is usually less than rougher cells).
2. Pulp chemical measurements and sampling, including pulp
samples for bulk chemical and/or surface analysis and
process water samples. This could also include specialty
measurements for reagents in solution and/or at the mineral
surface. It should always include the measurement of all
reagent flows during the survey.
3. Cell characterisation measurements, including superficial
gas velocity, bubble size and residence time distribution by
tracer tests. An estimation of energy dissipation in the cell
should also be made (usually in the range of 1 - 10 w/kg of
pulp).
4. Batch flotation tests on process streams for the purpose of
correlating laboratory conditions to the plant and/or to build
a circuit model based on component floatability. The later
batch flotation tests determine the fractions of different
floatability in the feed.
5. Sample of SAG feed and circuit water for the purpose of
laboratory testing. This is not often carried out in
conventional plant surveys, but it is very useful for
optimisation at laboratory scale conducted at a later date.
Obtaining these samples allows the plant conditions to be
mimicked at laboratory scale in grinding and flotation
experiments.
The following section will focus on aspects of items 2, 4 and 5
above.
Pulp chemical measurements during a typical
plant survey
Key points for the pulp chemical measurements (pulp chemical
survey) are shown in Figure 9. It is very important to measure the
primary ball mill discharge Eh and pH as this is usually the point
in the process where the Eh is at its lowest value. This point is
important because this will indicate the extent of media oxidation
and the possibility of low Eh values during collector conditioning
113
CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION
Reagents
Process Process
Water Primary X
Water Cyclone O/F
X Block
Primary
SAG Mill X X X
SAG Ball Mill
Feed
Reagents
Reagents
X = Sampling Point
FIG 9 - Typical points (X) for measuring pulp chemistry values in a flotation plant, showing typical reagent and process water addition points.
and flotation. Low Eh values during flotation may retard the
flotation of minerals (Trahar, 1984). It may be difficult to
measure the SAG mill discharge for reasons of access and safety,
though it can be useful to demonstrate the difference between the
SAG and primary ball mill chemical environments. This is very
important for unit flotation stages in the primary grinding circuit.
It is interesting to note the pulp chemical change across the
primary ball mill by comparing the primary cyclone underflow to
the primary ball mill discharge. This has consequences in the
matching of the laboratory conditions to the plant as discussed
further below.
The cyclone overflow should be separately measured from the
‘conditioned rougher feed’ due to the addition of reagents and
aeration in cycloning and pumping from the primary ball mill
circuit to flotation feed. The ‘conditioned rougher feed’ is the
feed presented to rougher flotation with all reagents added
(including frother), and should also be the focus of batch flotation
experiments during the survey. This is discussed further below.
The regrind ball mill discharge and regrind cyclone overflow
should also be separately measured. If there are any conditioning
stages, such as pulp heating, pH adjustment, aeration, reagent
additions (eg CuSO4, SO2), the change in pulp chemistry across
the conditioning stage should also be measured. The utility of
these measurements in solving processing problems is outlined in
a case study below.
The circuit water should certainly be measured and, if required,
other key streams in the process water circuit (eg tailings return
water, make-up water, mine water, tailings thickener overflow
water, concentrate thickener overflow water, etc) should be
measured. As an example of the importance of the process water
samples, it has been found, for example, that copper emanating
from a thickener overflow stream (which contained [Cu] at 5 ppm
at pH 7) inadvertently activated sphalerite in the lead circuit of a
lead/zinc flotation plant. To minimise this inadvertent copper
activation, only tailings return water, which contained much
lower levels of copper (<0.5 ppm), was used as make-up water in
the lead circuit. Another example of the importance of water
quality is the use of fresh water in the zinc cleaning stages of a
lead/zinc flotation plant to reduce lime consumption to achieve a
set point pH value and, consequently, improve selectivity between
sphalerite and pyrite. The normal process water at this plant is
super-saturated in calcium sulfate (Grano et al, 1997), meaning
that the concentrations of calcium and sulfate exceed the
solubility limit of calcium sulfate, suggesting that calcium sulfate
has and/or is probably precipitated/precipitating in the pulps. This
has a pH buffering effect, which means that higher lime additions
114 Spectrum Series 16
Conditioned
Rougher Feed
X X
Rougher
X Includes regrind
pump box and cyclone
Regrind
Ball Mill
Reagents
X
Reagents
Process Water
X
Cleaner
Block
X
are required to achieve a set pH (~11), and that more calcium
sulfate will precipitate due to lime addition, a source of calcium.
By using fresh water in the cleaning stages, lower lime additions
are required and better selectivity is achieved. In the future, use of
treated water sources for special applications (eg cleaning, fine
grinding) will probably become more common. Another example
is when thickeners are operated with dirty overflows and there is
residual polymeric flocculant adsorbed on the particles in the
thickener overflow. This may interfere with flotation due to the
adsorption of these hydrophilic slime particles onto the value
mineral.
For block surveys, samples and measurements contrasting the
products of the major separation should also be undertaken. This
includes those across the rougher block (ie rougher concentrate,
rougher tailing reporting to final tailing, and rougher feed) and
cleaner block (ie final concentrate, cleaner block tailing reporting
to final tailing, and cleaner feed). EDTA extractions and surface
analysis samples (not discussed here) would contrast the
oxidation state of these streams. EDTA extraction and other
simple measurements can be taken more routinely and correlated
to plant flotation performance. In general, the valuable metal in
the tailing streams has a greater proportion in an oxidised state
than the valuable metal in the concentrate streams, both measured
on a mass of metal basis when using the EDTA technique.
Flotation, under conventional conditions (eg without sulfidisation),
is always selective against oxidised forms. Changing oxidation
levels in the conditioned feed gives rise to changing recoveries in
the plant.
It is very important to measure the reagent flows during the
pulp chemical survey. This should include all reagents and all
addition points (eg collector stage added points, usually in the
scavengers, frothers – often overlooked in surveys, activators and
depressants, including any test reagents). The solution strength of
the reagents should be noted, or failing this, a sample should be
collected for latter analysis, and the reagents additions quoted as
mls/min, g/t of total plant feed and g/t of the specific process
stream to which the reagents are added. This is very important for
any laboratory tests to be conducted on the sample of SAG feed
collected during the survey. A mass balance on reagents should
be attempted, which may identify problems of, for example,
frother adsorption onto particles and high residual frother in the
process tailings and the continuing presence of frother in the
process water (Tsatouhas, Vera and Grano, 2005).
The type of plant survey, which is comprehensive, should be
the beginning point of any optimisation program. It is also
important to undertake some simpler block surveys at key sample
Flotation Plant Optimisation
 CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION
points, both before and after the comprehensive survey to
establish plant stability and the normal operating range of the
plant. These simple surveys may be called quality control surveys
and are aligned to the comprehensive survey in their reporting.
Correlating the laboratory conditions and
performance to the plant
Correlating the laboratory conditions to the plant is extremely
important if any optimisation work is contemplated at laboratory
scale (Pietrobon, Grano and Greet, 2004). Often laboratory
experiments are undertaken under conditions which are far
removed from the plant (eg adding lime in conditioning and not
in the mill in the laboratory, while lime is added to the mill in the
plant). To be more confident about successfully transferring the
results of laboratory studies to plant scale it is, at least, necessary
to mimic the plant pulp chemical conditions and particle size
distribution at laboratory scale. This also has the advantage that it
will focus the initial stages of laboratory investigations onto the
effect of simple parameters, which can be relatively easy to
manipulate at plant scale, and which may acutely affect plant
performance. This could include the effect of water quality or
water source, grind pH, aeration and standard plant reagent
additions, etc. The procedure outlined below also adds more
confidence to the testing of future ores, which have not been
processed through the plant.
During the comprehensive survey, a SAG mill feed sample is
taken (Figure 10) and treated in the normal way to produce
samples suitable for flotation testing. Storage of the crushed and
riffled sample should aim to minimise oxidation, though
subjecting the sample to the same standard laboratory flotation
conditions with storage time may check this, along with routine
EDTA extractions. During the comprehensive survey, flotation
tests on the conditioned rougher feed should also be undertaken
in triplicate under known conditions of the physical variables,
such as cell type and mechanism, gas flow rate (Jg), impellor
speed, scrape rate and bubble size, if possible, etc. These same
physical conditions are used in latter testing of the SAG feed
sample to ensure that differences from the plant to the laboratory
are only due to differences in the feed (properties) presented to
the flotation cell. During the comprehensive survey, a large
sample of circuit water should also be collected and used in
laboratory flotation testing (Figure 10). However, circuit water
samples may not be stable due to the continuing oxidation of
thio-salts and biological activity. In this case, a simulated circuit
water sample should be prepared based on the inorganic
Reagents
Plant Grinding
Circuit
Reagents
Plant Conditioning
Plant Flotation
FIG 10 - Procedure for assimilating the laboratory to the plant conditions of pulp chemistry.
Flotation Plant Optimisation
composition of the circuit water, matching all the major
constituents of the circuit water during the survey including the
total dissolved solids (TDS). Unfortunately, it is very difficult to
match the organic components that may include, for example,
collector decomposition products, bacteria and residual polymer
from dewatering. Comparing these water types may be very
informative as well. Also, removing some, or all, of the organic
components from the circuit water (eg by adsorption onto
activated carbon) may determine the effect of these components
on flotation performance.
Once these samples have been collected and treated
appropriately, the SAG feed, water samples and reagents may be
used at laboratory scale. Special mills may be available (eg
Magotteaux Mill®, Greet et al, 2004) which allow the pH to be
continuously controlled during grinding. This may be important in
matching the laboratory to the plant as the pH of the feed to most
plant primary ball mills (ie the cyclone underflow) is the same pH
as the plant ball mill discharge. Most batch mills have wide
variations in pH from the beginning of grinding to the end, with the
mill discharge pH usually being set as the target pH value. This
special mill (Magotteaux Mill®) also allows continuous purging
with different gases during grinding to ensure that the laboratory
mill discharge Eh is at the same value as the plant ball mill
discharge. Gas purging during grinding and post-grinding may also
be very informative in terms of optimisation. Once grinding is
complete the slurry is conditioned with the same reagents and at
the same addition level as during the plant survey. Laboratory
flotation is then conducted under the same physical conditions as
that in the laboratory flotation tests conducted during the plant
survey. With care, it is possible to duplicate the same performance
as the flotation tests during the plant survey. However, differences
will highlight variables, which sensitively affect plant performance.
These could include:
• grind pH,
• grind Eh,
• flotation pH,
• flotation Eh,
• water quality,
• temperature,
• grinding media,
• reagent quality, etc.
This is the real advantage of this method.
Ore
(SAG Feed)
Process Water Reagents
Laboratory
Grinding
Reagents
Laboratory Conditioning
Laboratory Flotation
Matching Conditions
Spectrum Series 16 115
CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION

The parameters, which require matching, are shown in mineral liberation was the same or greater than for porphyry
Figure 11. As well as Eh and pH, the particle size distribution of sulfide ores. The reason for the low copper concentrate grade and
the ore and individual minerals need to be reasonably matched. If recovery were related to pulp chemistry differences between the
there are significant deviations in the mineral size distribution, porphyry and sulfide skarn ores.
then it may be necessary to analyse the laboratory flotation In the preceding section the stability diagram of iron was
behaviour for both tests on the plant conditioned feed and the discussed. The stability diagram shows the thermodynamically
SAG mill feed on a size-by-size basis. This will also highlight most stable species (ie the species at the highest concentration) at
issues related to particle size. The dissolved oxygen demand and various Eh and pH values while assuming a certain total
EDTA extractable metal ions can also be matched. However, concentration of the species. Attention will now turn to the Eh-pH
these parameters are dependent upon Eh and pH of both grinding diagram for the Cu-S-H2O system (Figure 12). The Eh/pH
and conditioning stages. Matching the values of Eh and pH will measurements at the cyclone overflow from two plant surveys on
go a significant way towards matching dissolved oxygen demand a porphyry ore are also shown (Orwe, Grano and Lauder, 1998).
and EDTA extractable metal ions, provided the SAG feed sample The principal copper-bearing sulfide minerals are chalcocite and
has not oxidised during storage (Pietrobon, Grano and Greet, chalcopyrite. The normal pH for this plant is 11.5. The Eh/pH
2004). It is also important to match the Eh and pH of the mill diagram shows that chalcocite is only stable to approximately Eh
discharge separately from the flotation feed Eh and pH values. 0 mV at pH 11, while above this Eh value it will begin to oxidise.
This is because the Eh during grinding is usually much lower than However, the Eh of the pulp at the cyclone overflow is +150 mV
the Eh during flotation. However, if the Eh during grinding is not (Survey 1), suggesting that chalcocite oxidation may occur.
mimicked then the conditions during grinding are not realistic. Oxidation was confirmed by EDTA extraction and surface
The grind and flotation Eh and pH values should be considered analysis. In this system, further aeration of the pulp in flotation
separately in the study. does not appreciably increase the Eh, suggesting that the pulp is
fully aerated under these high pH conditions (ie DO is high and
FLOTATION the DO demand of the minerals and ore does not outweigh
GRINDING oxygen delivery to the pulp). In Survey 2, NaHS (~100 g/t) was
added to the plant cyclone overflow to achieve an Eh of -50 mV.
This decreased the Eh of the cyclone overflow as shown into the
stability region of Cu2S (Jones and Woodcock, 1978). Oxidation
MILL FLOTATION FLOTATION
would be prevented while oxidation products on chalcocite
DISCHARGE FEED RESPONSE (copper hydroxide) would be reduced to a ‘Cu-S’ phase. With
aeration in the plant flotation cells the Eh increases to +150 mV,
P80 and -38 microns pH mass and water rec., which is probably an acceptable value for collector adsorption
pH Eh grade and recovery
of value and onto chalcocite (assuming the resulfidised chalcocite does not
Eh O2 concentration
O2 concentration O2 demand
other minerals. re-oxidise) (Trahar, 1984). Copper recovery increased with this
EDTA ext. M+ ions treatment though the extent of this increase depended upon
copper mineralogy (chalcocite oxidises more than chalcopyrite
and so the effect is less marked for chalcopyrite dominated
FIG 11 - Parameters which need to be matched from the plant to feeds), the feed particle size (fine chalcocite is more oxidised
the laboratory. than coarser chalcocite), the sulfidisation potential and other
factors.
The Eh/pH stability (predominance) diagram –
interpretation 800
O2
H2 O
To demonstrate the utility of undertaking simple pulp chemical CuO + So
CuS + Cu S Plant
measurements during a plant survey, as well as the importance of 400 2-x Cyclone O/F
correlating laboratory conditions to the plant, an example is Survey 1
Eh / mV

discussed for the remainder of this section. The example comes 0 Cu2 S Cu S + CuO
from a flotation plant which normally treats porphyry copper ore 2-x
(ie copper in chalcocite and chalcopyrite in igneous host rock), H2O
Cu2 S Plant
H2
but for which the proportion of the run of mine (ROM) ore from -400 Cyclone O/F
sulfide skarns is generally increasing throughout the remainder of Survey 2
the mine life. Sulfide skarns emanate from the intrusion of the o o - 2-
Cu + H 2S Cu + HS Cu + S
igneous into the country rock and generally contain higher copper -800
and gold concentrations, but also much higher pyrite and 0 2 4 6 8 10 12 14
magnetite contents (Orwe, Grano and Lauder, 1997). The pH
flotation plant encountered problems when treating feeds with
high sulfide skarn content such as low copper recovery (<75 per FIG 12 - Plant cyclone overflow pulp potential (Eh) as a function of
cent) and low final concentrate grade (<20 per cent Cu). Blending pH in plant surveys on porphyry ore blends (head grade 2.05 per
was only partially effective and cannot be used in the latter stages cent Cu, 5.6 per cent S, Cu/S 0.37, six per cent pyrite) for Survey
of the mine life due to the preponderance of sulfide skarn. The 1 without sulfidisation and Survey 2 with sulfidisation.
low concentrate grade was a consequence of both high pyrite
recovery and also low copper recovery. The normal conditions in Another example of the application of presenting pulp
the plant (lime addition to SAG feed to pH 11 - 11.5 and using a chemical data in an Eh/pH representation is shown in Figure 13.
dithiophosphate collector in the primary ball mill) were Presenting data in this way gives order to the data as it allows
inadequate in the flotation separation of sulfide skarn ores. The changes in Eh, which may have been caused by changes in pH, to
copper-bearing minerals were adequately liberated from gangue be detected. In this example, both plant and laboratory data are
being ~80 per cent fully liberated, across all size ranges, at the shown (Putubu, Grano and Morey, 2001). The Eh and pH
normal grind size of 150 μm in flotation feed. The level of copper experimental values from plant surveys conducted in October (on

116 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION

450 100

Dissolved Oxygen / %
60 Cell 15 Tail
Aeration
30 O2
400 10
Time
H2 O 80 Cell 9 Tail
60
30
Cell 3 Tail
350 3 Laboratory Tests
60
10 Aeration Cell 15 Tail
Eh / mV

Initial pH 10.0 Porphyry Plant

Laboratory Pit Skarn Time Surveys (Oct 2000)
Magnetite Ore 60 Rougher Feed Cell 9 Tail
300 Natural pH 3
30 40
Laboratory
10 Cell 3 Tail
20/11/98 SAG Feed 3
Aeration Cell 15 Tail
Natural pH
250 Time Flotation Feed 20 Rougher Feed
Laboratory Tests Ball Mill D/C (a) Porphyry Ore Feed (b) Skarn Ore Feed
Initial pH 11.5 October 2000 March 2000
200 0
Skarn Plant
Surveys (March 2000) Cell 15 Tail
150 Flotation Feed 0 30 60 90 120 0 30 60 90 120 Time / sec
Ball Mill D/C
Porphyry ore blend Sulphide skarn blends
3 4 5 6 7 8 9 10 11 12 pH 17/10/2000; average for 15/3 to 21/3/2000;
20/11/98 SAG feed sample Pit Skarn Magnetite (PSM) sample Head Grade Head Grade
0.6% Cu, 1.2% S, Cu/S 0.5 ~1.2% Cu, ~4.0% S, ~Cu/S 0.3
Head Grade 2.05% Cu, 5.6% S, Head Grade 1.59% Cu, 10.7% S
Cu/S 0.37, 6% Pyrite Cu/S 0.15, 17% Pyrite

FIG 14 - Dissolved oxygen concentration as a function of time

FIG 13 - The pulp potential (Eh) as a function of pH in plant (dissolved oxygen demand) after the discontinuation of air purging
surveys on porphyry ore blends (17 October 2000) (£) and sulfide for plant rougher streams in surveys on (a) porphyry ore blends
skarn blends (March 2000) (<), as well as for laboratory tests on (17 October 2000; head grade 0.6 per cent Cu, 1.2 per cent S,
20 November 1998 SAG feed sample (head grade 2.05 per cent Cu/S 0.5) and (b) sulfide skarn blends (average for 15 - 21 March
Cu, 5.6 per cent S, Cu/S 0.37, six per cent pyrite) (open symbols) 2000; head grade ~1.2 per cent Cu, ~4.0 per cent S, ~Cu/S 0.3).
and on pit skarn magnetite (PSM) sample (head grade 1.59 per The acid soluble copper and Cu/S ratio is a very important
cent Cu, 10.7 per cent S, Cu/S 0.15, 17 per cent pyrite) (closed parameter in porphyry sulfide ores.
symbols) for aeration tests at different initial grind pH values
(~11.5, ~10 and natural pH). The numerical values are aeration
times in minutes after grinding the ores in the laboratory. depletion from the pulp is an indication of the oxidation rate. The
high dissolved oxygen demand of sulfide skarn pulps is related to
their high active pyrite content. Through processing in the plant,
predominantly porphyry ore) and March (on predominantly skarn the pyrite surfaces become progressively more oxidised,
ore) 2000, are compared to laboratory data for moderate (shown consuming less oxygen (Figure 14). The porphyry ores consume
as 20/11/98 SAG feed) and high pyrite skarn feed types (shown as oxygen to a much lower degree and develop higher Eh values in
PSM). The plant Eh (and pH) values for skarn ore (high pyrite) the pulp for the same pH values. These factors combine to give
surveys are lower than the porphyry ore survey, but also lower better chalcopyrite recovery from porphyry sulfide ores. It is
than the Eh values after laboratory grinding of skarn samples. It important to note that the copper rougher recovery during surveys
should be noted that for high pyrite skarn ores, higher (~4 kg/t) on high pyrite skarn blends (March 2000) varied dramatically
than normal (~1 kg/t) lime additions are required to achieve the between 60 and 80 per cent, while copper rougher recovery
target pH (11.5). In spite of the higher lime additions, the pH during surveys on porphyry ore (October 2000) was between 80 -
value is often lower than the target pH for high pyrite skarns. This 85 per cent. The rapid passivation of sulfide minerals, most
is because of pyrite oxidation, which generates acid in solution. probably pyrite, at alkaline pH values and under laboratory
In the latter cases, laboratory grinding at high pH causes grinding and flotation conditions is demonstrated in Figure 15.
adequate chalcopyrite recovery and selectivity from a high pyrite The high pyrite skarn feed (PSM) requires a longer aeration time
feed (PSM). Surface oxidation of pyrite in the laboratory
experiments assists depression of copper activated pyrite and the 100
promotion of chalcopyrite recovery. Chalcopyrite recovery is 90 20/11/98 SAG Feed
Pit Skarn Magnetite Ore
increased due to increased collector adsorption on its surface due 80
xygen / %

to reduced collector adsorption onto pyrite. It is also useful to 70

note that without pH control, the pH of the pulps generally 60
(a) pH 11.5
decrease with aeration conditioning (Figure 13). In the case of 50
grinding at pH 11.5 and 10 this is accompanied by increases in
olved Ox

Eh, which may be accounted for by changes in pH through the 20/11/98 SAG Feed
90
oxygen/water equilibrium. This is because with aeration the Pit Skarn Magnetite Ore
80
pH/Eh relationship parallels the oxygen/water equilibrium. 70 (b) pH 10.0
10 0
Disso

However, in the case of grinding at natural pH, the Eh increases Aeration

60
Time
after grinding more markedly, suggesting increases in the 50
dissolved oxygen content of the pulp with aeration. The pH/Eh 30 min
relationship does not parallel the oxygen/water equilibrium. In 90 10 min 20/11/98 SAG Feed
this case, the oxygen content of the pulp increases with aeration. 80 3 min
30 min
In these laboratory experiments, pyrite very quickly passivates at 70
alkaline pH values. It is noted that the optimum potential for 60 10 min Pit Skarn Magnetite Ore
chalcopyrite flotation occurs in the Eh range of 200 to 250 mV 50 (c) pH Natural 3 min
(SHE) (Trahar, 1984). When the pyrite is not adequately 40
passivated (at alkaline pH), the pyrite consumes oxygen (and 0 20 40 60 80 100 120 Time / seconds
collector), which retards chalcopyrite flotation in this system.
FIG 15 - Dissolved oxygen concentration as a function of time
(dissolved oxygen demand) after the discontinuation of air purging
Dissolved oxygen demand – interpretation
for laboratory tests on 20 November 1998 SAG feed sample
In general, sulfide skarn ores show much higher dissolved oxygen (closed symbols) and on pit skarn magnetite (PSM) sample (open
demand in the plant rougher cells than porphyry ores (Figure 14). symbols) for aeration tests at different initial grind pH values,
The dissolved oxygen demand is a function of surface oxidation (a) 11.5, (b) 10 and (c) natural pH. Aeration times of three (r),
state and exposure of sulfide minerals. The rate of oxygen ten (£), and 30 (¯) minutes shown.

Flotation Plant Optimisation Spectrum Series 16 117

CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION

6 30

Pyrite Recovery / % Copper Grade / %

pH 6.8
25
pH 7.5
4 20
pH8 pH 9.3
15
pH10 pH 10.6
2 10
pH11 pH 11.5
(a) pH11.5 5 (c)
0 0
80 (b) (d)
40
60 30
40 20
20 Lime after mill 10
Lime in mill
0 0
0 20 40 60 80 100 0 20 40 60 80 100
Copper Recovery / % Copper Recovery / %

FIG 16 - Copper grade/recovery relationship (a, c) and selectivity of copper against pyrite recovery (b, d) for laboratory flotation tests on pit
skarn magnetite (PSM) sample head grade 1.59 per cent Cu, 10.7 per cent S, Cu/S 0.15, 17 per cent pyrite with d80 144 microns 30 g/t
7249 collector added after grinding and pH adjustment, for tests with lime added after milling at natural pH (a, b) and for lime added to
the mill (c, d).

to achieve complete oxidation and that grinding under alkaline
conditions is important for pyrite oxidation.
Following these plant surveys, flotation tests were performed
on samples of SAG feed. Attention turned to the effect of grind
pH on copper recovery and selectivity against pyrite for
laboratory tests of SAG feed (Figure 16). For tests with lime
addition after the mill (Figure 16a, 16b), pyrite readily responds
to collector addition in spite of lime addition to pH 11.5. Under
these conditions, copper also shows low maximum recovery and
rate. This flotation behaviour is characteristic of problems
encountered under normal plant conditions during treatment of
high pyrite skarn ores. In the plant, it is not possible to increase
collector addition to counter low copper recovery due to the
increased pyrite recovery when treating sulfide skarns under
normal plant conditions. Radically different results are obtained
with lime addition to the laboratory mill for pH values above ten
(Figure 16c, 16d). In this case, pyrite recovery is markedly
decreased and copper maximum recovery and rate greatly
increased. It seems clear that without appropriate control of the
grinding environment, in particular pH and Eh, pyrite activation
ensues and copper recovery is very low.
In terms of plant practice, it was recommended that lime
addition be made to the primary ball mill feed, as well as SAG
feed, and that collector addition be made after the grinding circuit.
The objective here was to passivate, or oxidise, pyrite prior to
collector addition so that most of the collector would then be
directed towards chalcopyrite, and not pyrite. It was also
recommended that further work should investigate stage-adding
collector, not only for coarse composite recovery, but also for
pyrite control. The effect of using the first cell in flotation as an
aerator to passivate pyrite, prior to collector addition, also needed
to be investigated for high pyrite skarn feeds. Once conditions are
established which ensure effective pyrite depression, collector
additions may then be generally increased, in accordance with the
higher copper grades of the sulfide skarn ores. Dissolved oxygen
demand and EDTA extraction should be used to monitor pyrite
oxidation and copper activation, and this data correlated to plant
performance. Laboratory tests should continue to focus on
correlating laboratory conditions to the plant, while the effect of
Eh, controlled by gases in the mill and conditioning stages, should
also be investigated. The possibility of increasing the Eh from
+170 mV to +230 mV during the treatment of high pyrite sulfide
skarn blends should also be investigated. Based on the plant Eh/pH
relationship (Figure 13) it is unlikely that increasing the pH to
above 11.5 would be adequate to achieve this desired Eh increase.
CASE STUDY – ZINC REVERSE PROCESS
This case study focuses on determining why a specific zinc reverse
process did not operate efficiently at pilot plant scale (Johnson and
Munroe, 1988). Separation in a zinc reverse flotation process is
achieved after conditioning with heat and sulfur dioxide. These
inputs cause depression of sphalerite and flotation of pyrite in a
relatively high-cost process (Broman, Hultqvist and Marklund,
1985). The zinc reverse process is practised at Brunswick Mining
and Smelting (BMS) Corporation Limited in New Brunswick,
Canada (McTavish, 1980), and previously at Woodlawn Mines,
New South Wales. The principal of the zinc reverse process is to
separate pyrite from a sphalerite-rich stream. The pyrite reports to
the concentrate launder while the sphalerite and minerals, such as
non-sulfide gangue, report to the tailings stream. It is important to
note that even if complete chemical depression of the sphalerite is
achieved in the process, some fine free sphalerite will report to the
pyrite concentrate by entrainment and some sphalerite in
composite particles with pyrite may also be recovered by true
flotation.
The principal attraction of the zinc reverse process is that the
ratio of the pyrite rate constant to the sphalerite rate constant is
much higher than the equivalent ratio for conventional sphalerite
flotation. In other words, the zinc reverse process is a much more
powerful process for separating liberated sphalerite and pyrite
compared with conventional cleaning. In the zinc reverse process,
the ratio of rate constants for pyrite to sphalerite is 10:1 or greater,
while for conventional cleaning the equivalent ratio is typically
1:3.5. Reverse flotation is usually applied to streams in which the
non-sulfide gangue has already been removed by prior
conventional cleaning. Removal of pyrite from a zinc concentrate
(46 - 50 per cent Zn) by the process results in a higher zinc grade
(52 - 54 per cent Zn). The removal of pyrite (and pyrrhotite, if
present) typically lowers the iron assay from eight to ten per cent to
six to 7.5 per cent Fe.
The following discussion details results from an industrial zinc
reverse process at BMS. Further, results from a zinc reverse
process operated at pilot scale on a zinc-rich stream at Mount Isa
Mines, Mt Isa, are presented. This process did not operate
efficiently. However, with careful measurement of the pulp
chemistry and with accompanying interpretation, the process was
modified and the separation efficiency improved. A simplified
circuit diagram for a primary zinc process at an industrial plant is
shown in Figure 17. Three stages of the process are identified as
significant to the pulp chemistry:

118 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION

Zinc Concentrate 400

Pyrite Cleaner
Tailing (from 300 22 oC
Pyrite Cleaner) SO2 Conditioning

Pulp Eh (mV)
SO2 Conditioning
200 Heating 18 oC
25 oC

100 Rougher 80 oC
Flotation
Heat Conditioning 0

-100
3 4 5 6 7 8 9 10 11 12
Upgraded
Air Roughing Zinc Concentrate Pulp pH

Pyrite Rougher
Concentrate
(to Pyrite Cleaner)
FIG 17 - Simplified circuit diagram for the plant zinc reverse
cleaning process.
1. SO2 addition at 18°C,
2. temperature elevation to 80°C, and
3. air addition to the flotation process at 75 - 80°C.
Collection of Eh/pH data from each process allowed
quantification of the pulp chemistry to an important degree.
A typical plant result of the pulp chemistry from BMS is shown
in Figure 18, presented as an Eh-pH diagram. As discussed
previously, this presentation allows changes in Eh due to either pH
changes (water/oxygen equilibria) or changes in the oxidising/
reducing potential of the pulp to be discerned. In this example, the
feed to the zinc reverse process is the third cleaner concentrate, a
stream already close to final zinc concentrate Zn grade, but for
which additional iron rejection was required. The sulfur dioxide
conditioning stage caused a large pH change and a large movement
away from the region of the experimental water/oxygen line
(Figure 18). The movement was essentially parallel to the pH axis
and therefore away from the experimental water/oxygen line. For
the heat conditioning stage, there was a considerable change
towards more reducing conditions in comparison to those after
addition of sulfur dioxide (Figure 18). This change is easily seen in
the Eh-pH diagram because there was no complicating
simultaneous large change in pH. The main feature for the flotation
stage with air was an increase in the Eh value. Unlike the two
previous steps where the Eh values had moved further away from
the experimental water/oxygen line, the Eh values moved back
towards the water/oxygen line (Figure 18).
Attention will now turn to the pilot plant studies at Mt Isa
lead/zinc concentrator. At the commencement of the pilot plant
testing of zinc reverse, only a single pyrite rougher stage was
used (ie no cleaning of the reverse concentrate), and the feed to
the reverse process typically assayed 23 - 24 per cent Zn. In this
case, the zinc reverse process was applied to a plant stream,
FIG 18 - Pulp Eh-pH diagram for the plant zinc reverse cleaning
(BMS).
which contained high levels of slow floating sphalerite. This type
of sphalerite is difficult to upgrade in conventional cleaning
because of its small differential against gangue sulfide minerals.
In reverse flotation, the rate constant for the sphalerite in
conventional cleaning is irrelevant, and therefore the reverse
flotation approach provides one possible solution to the problem
of upgrading this sphalerite.
In the initial stage of the pilot plant study, the pyrite rougher
stage only occasionally performed to expectation, and it was
apparent that it was insufficiently developed to routinely maintain
the expected daily result. The zinc grade was expected to increase
by 0.8 units for each unit of zinc loss, but the observed increase
was only approximately 0.38 (Table 2, large cells). However,
surveys of Eh/pH values in the pyrite rougher bank and
conditioning stages provided very important clues to the cause of
the inconsistent depression of sphalerite (Figure 19). The data
revealed a very large increase in Eh (~+200 mV) between the last
conditioner and the first cell of the pilot plant pyrite rougher bank
(Figure 19, unsuccessful process). This behaviour differed from
the previously described behaviour for plant processes where an
Eh increase of only ~50 mV was observed for the entire pyrite
rougher flotation stage.
The large increase in Eh meant that even the first cell of the
pyrite rougher was destroying the deliberately created chemical
environment in the conditioners. The pulp flowed a horizontal
distance of four metres by gravity from the last conditioner and
the first cell, which was completely enclosed from the atmosphere.
It was reasoned that there was little scope for accidentally raising
the Eh during transportation through the pipe and that the oxygen
in the flotation cell was destroying the reducing environment
created in the conditioners. Therefore, a solution to the problem
(at that time) was the use of smaller flotation cells with weirs at
intervals to lower exposure of the pulp to oxygen in the initial
cells. The use of weirs was necessary to eliminate any possibility
of back mixing of comparatively aerated pulps from the latter
cells. Another possibility was to use nitrogen instead of air as the
carrier gas in flotation.

TABLE 2
Summary of some daily metallurgical results for the pyrite rougher before and after changing the size of the cells in the pilot plant.

Cells Assays/Zinc (%) Zinc recovery (%) Zinc grade increase Zinc grade increase
Feed Concentrate Tailing to concentrate (%) per unit zinc loss
(reverse tailing)
Large 23.7 43.1 16.6 48.7 19.4 0.38
Small 21.9 49.7 10.1 67.6 27.8 0.86

Flotation Plant Optimisation Spectrum Series 16 119

CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION
400
28 oC SO2 Conditioning
300
28 oC
Pulp Eh (mV)
200 Cell 1 80oC (Start of roughing)
100
0 90 oC
90 oC (smaller cells with weirs)
-100
3 4
FIG 19 - Pulp Eh-pH diagram for two configurations of the zinc reverse flotation process.
A robust (ie a consistent performance irrespective of feed
condition) process was obtained at the expected level (Table 2,
small cells) after the change to smaller cells with weirs. Surveys
of the Eh/pH values of the new pyrite rougher bank showed that a
much smaller increase in Eh values (~100 mV) existed between
the last conditioner and the first cell (Figure 19, successful
process). Further, the increase in Eh for the remainder of the pyrite
rougher was quite small ~20 mV. The zinc concentrate (45 - 52 per
cent Zn) from the pyrite rougher concentrate typically assayed 6.5 -
8.5 per cent iron after successful removal of the pyrite. Later
inclusion of cleaning to the process improved its performance
further. There is an indication that the Eh in flotation should not
exceed +150 mV to ensure depression of sphalerite.
This work demonstrated the importance of measuring both Eh
and pH to solve flotation separation problems. It also suggested
that on-line measurement of these variables (at least) may provide
a more stable and robust process. It was also speculated, though
not examined, that use of an inert gas (eg nitrogen) in flotation
may have benefited flotation, by either improving selectivity
and/or reducing the need for pulp heating, which would aid the
process economics.
CONCLUSIONS
This chapter has dealt with simple pulp chemical measurements
which should be taken with most plant surveys. The basic theory,
interpretation and utility of these measurements are demonstrated.
REFERENCES
Broman, P G, Hultqvist, J and Marklund, U, 1985. Experience from the
use of SO2 to increase selectivity in complex sulphide ore flotation,
in Developments in Mineral Processing, Flotation of Sulphide
Minerals (ed: K S Forssberg), pp 277-291 (Elsevier: Stockholm).
Fornasiero, D, Eijt, V and Ralston, J, 1992. An electrochemical study of
pyrite oxidation, Colloids Surfaces, 62:63-73.
Garrels, R M and Christ, C L, 1965. Solutions, Minerals and Equilibria
(Harper and Row: New York).
Grano, S R, Johnson, N W and Ralston, J, 1997. Control of the solution
interaction of metabisulphite and ethyl xanthate in the flotation of the
Hilton ore of Mount Isa Mines Limited, Australia, Minerals
Engineering, 10(1):17-39.
Grano, S R, Lauder, D W, Johnson, N W and Ralston, J, 1997. An
investigation of galena recovery problems in the Hilton Concentrator
of Mount Isa Mines Limited, Australia, Minerals Engineering,
10(10):1139-1163.
120
24 oC
25 oC
Cell 3 72oC (End of roughing)
Cell 1 80oC (Start of roughing)
Data for Unsuccessful Process
Data for Successful Process
5 6 7 8 9 10 11 12
Pulp pH
Greet, C J, Small, G L, Steinier, P and Grano, S R, 2004. The Magotteaux
Mill®: Investigating the effect of grinding media on pulp chemistry
and flotation performance, Minerals Engineering, 17(7-8):891-896.
Johnson, N W, 1988. Application of electrochemical concepts to four
sulphide flotation separations, in Proceedings International
Symposium Electrochemistry in Mineral and Metal Processing II
(ed: P E Richardson and R Woods), vol 88-21, pp 36-48 (The
Electrochemical Society).
Johnson, N W, Jowett, A and Heyes, G W, 1982. Oxidation-reduction
effects in galena flotation: Observations on Pb-Zn-Fe sulphides
separation, Transactions of the Institutions of Mining and Metallurgy,
Mineral Processing and Extractive Metallurgy, 91:C32-C37.
Johnson, N W and Munroe, P D, 1988. Eh-pH measurements for problem
solving in a zinc reverse flotation process, in The AusIMM
Proceedings, 293(3):53-58.
Jones, M H and Woodcock, J T, 1978. Evaluation of an ion-selective
electrode for control of sodium sulphide additions during laboratory
flotation of oxidised ores, Transactions of the Institutions of Mining
and Metallurgy, Mineral Processing and Extractive Metallurgy,
87:C99-C106.
Jones, M H and Woodcock, J T, 1984. Application of pulp chemistry to
regulation of chemical environment in sulphide mineral flotation, in
Proceedings Principles of Mineral Flotation (The Wark Symposium)
(eds: M H Jones and J T Woodcock), pp 147-183 (The Australasian
Institute of Mining and Metallurgy: Melbourne).
Kant, C, Rao, S R and Finch, J A, 1994. Distribution of surface metal
ions among the products of chalcopyrite flotation, Minerals
Engineering, 7(7):905-916.
Kristall, Z, Grano, S R, Reynolds, K, Smart, R St C and Ralston, J, 1994.
An investigation of sphalerite flotation in the Murchison Zinc
concentrator, in Proceedings Fifth Mill Operators’ Conference,
pp 171-179 (The Australasian Institute of Mining and Metallurgy:
Melbourne).
McTavish, S, 1980. Flotation practices at Brunswick Mining, CIM
Bulletin, February, pp 115-120.
Montalti, M, Fornasiero, D and Ralston, J, 1991. UV-visible
spectroscopic study of the kinetics of adsorption of ethyl xanthate on
pyrite, Journal of Colloid and Interface Science, 143:440-450.
Natarajan, K A and Iwasaki, J, 1973a. Practical implication of Eh
measurements in sulfide flotation circuits, Transactions of the
American Institute of Mining, Metallurgical, and Petroleum
Engineers (AIME), 254:323-328.
Natarajan, K A and Iwasaki, I, 1973b. Effect of poisoning of platinum
electrodes on Eh measurements, Transactions of the American
Institute of Mining, Metallurgical, and Petroleum Engineers (AIME),
254:22-28.
Natarajan, K A and Iwasaki, I, 1974. Significance of mixed potentials
in Eh measurements with Pt electrodes, Transactions of
the American Institute of Mining, Metallurgical, and Petroleum
Engineers (AIME), 255.
Spectrum Series 16 Flotation Plant Optimisation
 CHAPTER 6 – CHEMICAL MEASUREMENTS DURING PLANT SURVEYS AND THEIR INTERPRETATION
O’Connor, C T, Dunne, R C and Botelho De Sousa, A M R, 1984. The
effect of temperature on the flotation of pyrites, Journal of
the Southern African Institute of Mining and Metallurgy,
84(12):389-394.
Orwe, D, Grano, S R and Lauder, D W, 1997. Chalcocite oxidation and
its influence on fine copper recovery at the Ok Tedi concentrator,
Papua New Guinea, in Proceedings Fifth Mill Operators’
Conference, pp 85-96 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
Orwe, D, Grano S R and Lauder, D W, 1998. Increasing fine copper
recovery at the Ok Tedi concentrator, Papua New Guinea, Minerals
Engineering, 11:171-187.
Pietrobon, M, Grano, S and Greet, C, 2004. Matching laboratory and
plant performance – A case study of the Elura lead circuit, Pasminco
Australia Ltd, Minerals Engineering, 17(6):811-824.
Putubu, J K, Grano, S and Morey, M, 2001. A pulp-chemical study of
Skarn Ore feed blends at the OK Tedi concentrator, Papua New
Guinea, in Physicochemical Problems of Mineral Processing,
35:125-140.
Rand, D A J and Woods, R, 1984. Eh measurements in sulphide mineral
slurries, International Journal of Mineral Processing, 13:29-42.
Richardson, P E and Walker, G W, 1985. The flotation of chalcocite,
bornite, chalcopyrite and pyrite in an electrochemical flotation cell,
in Congress International Mineralurgie, [C R], 15th, 2, pp 198-210.
Rumball, J A and Richmond, G D, 1996. Measurement of oxidation in a
base metal flotation circuit by selective leaching with EDTA,
International Journal of Mineral Processing, 48(1-2):1-20.
Flotation Plant Optimisation Spectrum Series 16
Sheldon, G P and Johnson, N W, 1988, Galena flotation with
metabisulphite: Solutions to problems from xanthate sulphoxy ion
reactions, The AusIMM Proceedings, 293(3):49-52.
Spira, P and Rosenblum, F, 1978. Application of oxygen demand to
aeration, Canadian Mining Journal, March, 28-29.
Trahar, W J, 1984. The influence of pulp potential in sulphide flotation,
in Principles of Mineral Flotation (The Wark Symposium) (eds: M H
Jones and J T Woodcock), pp 117-136 (The Australasian Institute of
Mining and Metallurgy: Melbourne).
Tsatouhas, G, Vera, M, and Grano, S, 2005. Performance and
characterisation of the froth phase in industrial flotation circuits, in
Proceedings Centenary of Flotation Symposium (ed: G. Jameson),
pp 377-384 (The Australasian Institute of Mining and Metallurgy:
Melbourne).
Woodcock, J T and Jones, M M, 1970a. Chemical environment in
Australian lead-zinc flotation plant pulps: I. pH, redox potentials and
oxygen concentration, in The AusIMM Proceedings, 235:45-60.
Woodcock, J T and Jones, M H, 1970b. Chemical environment in
Australian lead-zinc flotation plant pulp: II collector residuals,
metals in solution, and other parameters, The AusIMM Proceedings,
235:61-76.
Woods, R, 1984. Electrochemistry of sulphide flotation, in Principles of
Mineral Flotation (The Wark Symposium) (eds: M H Jones and J T
Woodcock), pp 91-117 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
121
 HOME

CHAPTER 7

Electrochemical Aspects of Sulfide Mineral Flotation

Ronald Woods

FAusIMM, Visiting Professor, School of Biomolecular and Physical Sciences, Griffith University,
Nathan Qld 4111 and Adjunct Senior Research Fellow, School of Chemistry, Monash University,
Clayton Vic 3800. Email: r.woods@griffith.edu.au

Ronald Woods obtained a University of London PhD from analytical chemistry research carried out with
Professor I M Kolthoff at the University of Minnesota. He joined CSIRO Minerals in 1966 and carried out
research on a range of subjects associated with mineral processing and electrochemical power sources. He
worked at that organisation until 1995, at which time he was Research Manager of Metal Production. He
received the Stokes Medal of the RACI Electrochemistry Division in 1989 and was awarded a DSc (Eng) in
1994 by London University for his work on electrochemical aspects of mineral processing. In more recent
years, he has held a number of research fellowships at universities in Australia and North America carrying
out research on mineral flotation from an electrochemical viewpoint.

Mechanism of Mineral/Collector Interaction

Electrochemical Potential
Galvanic Interactions
Electrochemical Investigative Techniques
Flotation Recovery as a Function of Potential
Flotation Recovery as a Function of Collector Coverage
The Potential Dependence of Flotation in the Absence of Collectors
Activation and Depressant Action
Selective Flotation Recovery as a Function of Potential
Monitoring Potential in Flotation Practice

ABSTRACT realised that adsorption of collector ions would lead to highly

Flotation is considered from an electrochemical perspective. Electrode
potential is a determining parameter in recovery, and can be monitored and
controlled in the laboratory and the plant. The underlying electrochemical
science is reviewed and its application to plant practice explored.
MECHANISM OF MINERAL/COLLECTOR
INTERACTION
Towards the middle of the 20th century, the doyens of flotation
theory were vigorously debating the mechanism of the
attachment of flotation collectors with sulfide minerals. Taggart
et al, (1930) proposed that attachment occurred ‘by reason of
chemical reactions of well-recognized types’; thus, for example,
xanthates would form lead xanthate on a galena surface. Wark
and Cox (1934) pointed out that the development of a finite
contact angle on sulfide surfaces can occur at lower collector
concentrations than that of the solubility of the metal-collector
compound. Wark and Cox (1934) as well as Gaudin (1939)
considered that the attachment of a collector to a sulfide surface
was best considered as adsorption. But Cook and Nixon (1950)
charged surfaces and proposed that the adsorbed species is the
acid form of the collector ion. Xanthic acids, however, are
unstable. Thus, there were three theories and each had
difficulties; formation of metal thiols didn’t fit thermodynamics,
adsorption of ions didn’t fit charge requirements, and adsorption
of thiol acids didn’t fit stability criteria.
A solution to this dilemma was provided by Nixon (1957). He
had carried out research with Cook on the theory of adsorption of
the acid form of collectors before joining Melbourne University
and working in a Department that was at the forefront of the
development of electrochemical mechanisms for metal corrosion.
With his co-worker, Salamy, he saw another way in which a
neutral surface could be generated, based on corrosion science
concepts. It was proposed that ‘the troublesome electrons’
identified by the Cook school that would lead to excessive charge
on the surface ‘are removed by a reaction with oxygen’. That is,
the interaction of thiol collectors with mineral surfaces comprises
a mixed-potential mechanism in which an anodic oxidation
involving the collector transfers electrons to the mineral and the
simultaneous cathodic reduction of oxygen returns this charge to

Flotation Plant Optimisation Spectrum Series 16 123

CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION
the solution phase. Most sulfide minerals are semiconductors
with high electronic conductivity and hence can readily sustain
such processes. Nixon (1957) suggested that ‘prominent theories
of flotation could be reconciled’ by the electrochemical approach.
This is illustrated in Figure 1 in which the requirement of a
neutral surface species espoused by Cook can be met in different
ways. The anodic reaction can be adsorption as proposed by
Wark, Figure 1a. Also, it can be the formation of a metal thiol
compound as proposed by Taggart. The latter process can occur
as a single step as in 1b or through separate surface oxidation and
ion exchange processes as shown in 1c and 1d. In addition, the
anodic process can be the formation of the dithiolate as suggested
by Nixon (1957) and illustrated in 1e.
Each of these mechanism types has been identified
experimentally and shown to occur in flotation systems. They all
impart hydrophobicity to mineral surfaces. Chemisorption is the
thermodynamically favoured reaction for most mineral/collector
systems, which means it occurs first. As will be shown later, it is
sufficient, in itself, to induce efficient flotation.
Chemisorption involves bonding between collector molecules
and metal atoms in the mineral surface. The attachment of ethyl
xanthate to a galena surface is illustrated in Figure 2. The lower
spheres are sulfur (white) and lead (black) atoms in the galena
surface. The sulfur atoms of the xanthate are smaller than those in
the mineral because they have no ionic character. The xanthate
sulfurs are equally bonded to lead atoms in the mineral surface.
This bonding is similar to that occurring in lead xanthate
molecules, but in chemisorption, the metal atoms remain part of
the mineral lattice. Various spectroscopic investigations have
confirmed this mode of attachment of xanthate to mineral
surfaces (Buckley et al, 2003). With collectors such as
2-mercaptobenzothiazole and butylethoxycarbonylthiourea, bonding
with metal atoms in the mineral surface can occur through both
nitrogen and sulfur atoms (Goh et al, 2006; Hope et al, 2006).
The electrochemical mechanism has two important implications.
Firstly, the potential across the mineral/solution interface will be
an important parameter in determining flotation recovery.
Secondly, the reaction imparting floatability, the anodic process
involving the collector, is amenable to investigation using
electrochemical techniques. From the viewpoint of the flotation
engineer, it is the former of these implications that is more
important and this attribute will be the focus of this chapter. Since
the potential can be measured in flotation cells in the laboratory
and in flotation plants, it can provide the engineer with a
diagnostic tool for developing flotation strategies and for
alleviating flotation problems. Nowadays, most laboratories apply
potential measurement to derive optimum flotation strategies for
operation in the plant. As pointed out by Hintikka and Leppinen
(1995) ‘monitoring of potential with noble metal electrodes is
now almost routinely carried out in laboratory testwork and
full-scale processes’. The reader is referred to the chapter by
Stephen Grano in this publication for information on the practical
aspects of measuring potentials in laboratory cells and in flotation
plants. Early examples of the application of potential
measurements in developing flotation strategies are those of
Johnson (1988) at Mt Isa Mines in Queensland. More recent
examples are those Leppinen et al (1998), in which the influence
of potential on flotation separation in copper and zinc from
complex sulfide ores was considered, and of Smith and Bruckard
(2007) in which the separation of tennantite from chalcopyrite
and bornite was explored.
ELECTROCHEMICAL POTENTIAL
The electrochemical potential is the driving force for processes in
which electrons are transferred between a conducting solid and
species in solution. The web-based Electrochemical Dictionary
124 Spectrum Series 16
(http://electrochem.cwru.edu/ed/dict.htm#e65) says the electrical
potential:
… is somewhat analogous to the difference in
height in a waterfall that causes the water to fall,
or the difference in pressure in a pipeline that
causes the gas to flow.
Electrode potentials are measured against a reference electrode
and are often converted to the standard hydrogen electrode (SHE)
scale. For example, potentials measured against the silver/silver
chloride 4 mol dm-3 potassium chloride reference electrode at
25°C require the substraction of 0.222 V to convert them to the
SHE scale. Similarly, for the calomel electrode at 25°C, 0.244 V
has to be subtracted. Potentials on the SHE scale are often
referred to as the Eh.
The equilibrium potential of an electrochemical process is the
potential at which no overall reaction can take place. As the
potential is taken more positive than the equilibrium value, an
anodic oxidation reaction occurs, in which electrons are
transferred from solution species to the electrode. Conversely, as
it is taken more negative, the reverse, a cathodic reduction
reaction takes place in which electrons are transferred from the
electrode to solution species.
The potential is a mixed potential when it is determined by
two or more different electrochemical processes occurring on
the same electrode surface. Mixed potential systems are non-
equilibrium systems and hence are determined by the kinetics of
potential-determining reactions as well as the thermodynamics.
The situation in which two redox couples are involved is
illustrated in Figure 3.
E1 and E2 in Figure 3 are the reversible potentials of the two
couples and correspond to equilibrium between forward and back
reactions for each of the couples in the absence of the other.
When both are present, the potential will take up a value at which
the two processes give equal and opposite currents so that the net
charge transfer across the interface is zero. This potential is the
mixed potential. It can be seen that the system involves an anodic
oxidation and a cathodic reduction reaction proceeding at finite
rates and hence an overall chemical reaction is taking place.
It should be pointed out that both processes that make up the
mixed potential system can proceed on a homogeneous surface.
The surface is then acting both as a source and sink for electrons.
On the other hand, a heterogeneous surface may have areas of
different activity for the individual processes and in this case the
anodic and cathodic reactions occur on different surface sites.
The potential is the key parameter in determining whether the
important reaction can take place. It can be seen from Figure 3
that, for the oxidation of the species characterised by E1 to take
place, the potential must be greater than E1.
Mixed potential mechanisms are important in a range of
reactions of metallurgical interest, eg corrosion of metals,
cementation (metal displacement) reactions, hydrometallurgical
processing, supergene alteration of sulfide orebodies, spontaneous
oxidation of sulfide minerals in transport and storage, as well as in
sulfide mineral flotation.
The important potential determining flotation is that occurring
across the mineral/solution interface. It is generally measured by
inserting a noble metal electrode into the pulp, in which case it is
often referred to as the pulp potential. It is presumed that sufficient
mineral particles contact the noble metal surface for the measured
potential to be that of the mineral of interest. The solution potential
is often determined in metallurgical systems and is measured by a
noble metal electrode inserted into filtered solution. This is
different from the potential across the mineral/solution interface
and may not relate to flotation recovery. Only when a very strong
redox couple is present in solution will the two potentials be the
Flotation Plant Optimisation
 CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION

-
FIG 1 - Schematic representation of the mixed potential mechanism for the interaction of thiol collectors (X ) with sulfide minerals in
which the anodic process is (a) chemisorption; (b) reaction to form a metal collector compound; (c and d) the reaction in (b) occurring
in two stages – (c) oxidation of the mineral and (d) ion exchange with the collector; and (e) formation of the dithiolate.

Flotation Plant Optimisation Spectrum Series 16 125

CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION
FIG 2 - Schematic of the attachment of chemisorbed xanthate to a
galena surface.
FIG 3 - Schematic current-potential curves showing reversible
potentials of two redox couples and the mixed potential when
both couples are present.
same and this is due to the redox couple determining the potential
of the indicator electrode as well as that of the mineral/solution
interface (Rand and Woods, 1984).
GALVANIC INTERACTIONS
It was pointed out above that a heterogeneous surface may have
areas of different activity for the individual processes in a mixed
potential system and in this case the anodic and cathodic
reactions occur on different surface sites. When two dissimilar
conducting solids are in contact, it is possible for the anodic
process to occur on one solid and the cathodic on the other. This
situation is utilised in corrosion prevention through cathodic
protection of a metal structure such as a ship or a bridge with a
sacrificial anode composed of a more electrochemically active
metal such as magnesium. This manner in which this procedure
operates is illustrated in the schematic shown in Figure 4.
The solid anodic curve represents corrosion of a metal to form
an ionic species or an oxide. The solid cathodic line represents
the reduction of oxygen derived from the atmosphere. The mixed
(corrosion) potential will be at A and it can be seen that corrosion
will be proceeding at the current flowing at point A. When a
sacrificial anode is attached to the corroding metal, oxidation of
this metal occurs to give the dashed curve and the mixed potential
now shifts to B. It can be seen from Figure 4 that, at this
potential, there is a negligible contribution to the anodic current
at the mixed potential from the metal to be protected. Thus,
corrosion of the structure of interest does not take place.
Similar galvanic interactions occur in mineral flotation
systems. Grinding of sulfide ores is a prime example. The
grinding media galvanically protects the sulfide minerals in an
ore from oxidation that would impair interaction with a collector
126 Spectrum Series 16
FIG 4 - Schematic of current/potential relationships explaining
cathodic protection.
in the subsequent float cells. On the other hand, oxidation of the
grinding media results in iron oxy-hydroxy species that inhibit
flotation through coating the minerals of interest with a
hydrophilic layer. Galvanic interaction between grinding media
and sulfide ores has been the subject of many investigations
reported in the literature, from those of Iwasaki and co-workers in
the 1980s (Adam et al, 1984) to those today (eg Grano and
Huang, 2006).
Similarly, sulfide minerals can undergo galvanic interaction
with each other and this can enhance oxidation of a mineral and
affect selective flotation. For example, pyrite has a relatively high
rest potential and also strongly supports oxygen reduction
(Biegler et al, 1977). If a sulfide such as galena is in contact with
pyrite in a flotation pulp, then its oxidation is not only driven by
oxygen reduction on itself, but also that on pyrite and,
consequently, is increased significantly. This can lead to loss in
flotation recovery. Again, this phenomenon has been the subject
of many investigations reported in the literature, from the 1960s
(Majima, 1969) to today (Rao and Finch, 2006).
ELECTROCHEMICAL INVESTIGATIVE
TECHNIQUES
The most useful electrochemical technique for studying the
interaction of collectors with mineral surfaces has proved to be
voltammetry.
In this technique, the potential of an electrode constructed from
the mineral is controlled with a potentiostat and the current
flowing through the electrode is recorded. The potential is
scanned linearly in time traversing regions in which no reaction
occurs to those where collector adsorption takes place.
The range of sensitivity of current measuring techniques allows
submonolayer formation to be determined as well as that of
multilayers. In cyclic voltammetry, the potential is reversed at a
selected value and the scan in the opposite direction allows
electrochemical reactions of products formed on the initial scan
to also be explored.
An example of the application of cyclic voltammetry, that to
investigate the interaction of ethyl xanthate with galena, is shown
in Figure 5 (Woods, 1971). The initial scan was started at -0.5 V
versus SHE and the potential was taken in the positive-going
direction. The anodic current peak that is seen at ~0 V was
identified with chemisorption of a monolayer of ethyl xanthate on
the galena surface. This assignment was later substantiated by
X-ray photoelectron spectroscopy (XPS) (Shchukarev et al,
1994). The increase in anodic current following the anodic peak
arises from oxidation of the collector to lead ethyl xanthate and
diethyl dixanthogen. On the return scan, a cathodic current peak
Flotation Plant Optimisation
 CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION

can be seen at ~ -0.3 V. This peak results from reduction of the

products of anodic oxidation back to galena and ethyl xanthate.
The area under a voltametric peaks is the charge passed in the
electrochemical process and, when the process is adsorption,
gives a measure of surface coverage of xanthate on the mineral
surface. That for the anodic peak in Figure 5 was as expected for
adsorption of a monolayer.

FIG 6 - Flotation recovery of chalcocite with ethyl xanthate as a

function of potential: –o– data from Richardson et al (1984);
–¡– curves from Heyes and Trahar (1979); dashed curve from
Basilio et al (1985). Dotted curve is a cyclic voltammogram at
5 mV s-1 for a chalcocite bed electrode (O’Dell et al, 1984).

also be seen that there is good agreement between the three

groups as to the potential of onset of flotation even though the
experimental conditions employed differed. The dotted curve is a
voltammogram obtained with a chalcocite particle bed electrode
in the flotation cell used by Richardson et al (1984) to record
FIG 5 - Voltammograms for a galena electrode at pH 9.2; scan their flotation recovery data. The characteristics of the
rate 10 mV s-1; ethyl xanthate 9.5 × 10-3 mol dm-3 (solid line), voltammogram are the same as that found with electrodes made
0 (dashed line) from Woods (1971). from single chalcocite specimens. The anodic peak at ~ -0.15 V
arises from the chemisorption of a monolayer of ethyl xanthate
FLOTATION RECOVERY AS A FUNCTION OF on the mineral surface, while the anodic current above 0 V is due
POTENTIAL to interaction with the mineral to form copper(I) xanthate. It can
be seen that the onset of flotation occurs in the potential region of
The mixed potential mechanism implies that flotation will not chemisorption, and that efficient flotation still occurs when
occur if the potential is below the region in which a collector copper(I) xanthate is also formed.
oxidation reaction can occur, and at potentials within this region, As illustrated for galena in Figure 5 and chalcocite in Figure 6,
flotation becomes possible. This mechanism was not generally chemisorbed xanthate occurs before metal xanthate formation,
accepted until the 1980s after investigations had been carried out ie the chemisorbed collector is the thermodynamically more
that demonstrated that flotation recovery was, indeed, a function favourable species. Furthermore, it is sufficient to induce
of potential. The early work in this area was carried out in the flotation.
CSIRO Divisions of Mineral Chemistry and Mineral and Process
Engineering laboratories in Melbourne (both now CSIRO
Minerals), the US Bureau of Mines in Avondale, Maryland and
Virginia Polytechnic Institute and State University.
Gardner and Woods (1973, 1977) designed a particulate bed
electrode flotation cell, and applied it to demonstrate that the
flotation of galena and that of pyrite were potential dependent and
obtained an estimate of the potential of onset of flotation.
Heyes and Trahar modified a Denver flotation cell to allow the
use of nitrogen as the carrier gas and the control the potential of
the mineral particles in the flotation pulp through the regulated
addition of redox reagents. They applied this system to study the
flotation of chalcocite with ethyl xanthate as a function of
potential (Heyes and Trahar, 1979). Richardson et al (1984) and
O’Dell et al (1984) determined the potential dependence of the
flotation of chalcocite with ethyl xanthate in which the potential
was controlled electrochemically. Basilio et al (1985) studied the
same system using a Hallimond tube flotation cell with chemical
control of potential by the addition of redox reagents. The results FIG 7 - Dependence of flotation recovery on conditioning potential;
of these three groups are shown in Figure 6. conditioning time ten minutes; flotation time two minutes; pH 9.2;
It can be seen from Figure 6 that flotation does not take place xanthate 1.44 × 10-5 mol dm-3 for Cu2S and (2 ± 0.1) × 10-5 mol
below ~ -0.2 V and that efficient flotation occurs at ~0 V. It can dm-3 for the other minerals (from Richardson and Walker (1985).

Flotation Plant Optimisation Spectrum Series 16 127

CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION

Figure 7 shows the flotation recovery/potential curves for

chalcocite, bornite, chalcopyrite and pyrite in the presence of
ethyl xanthate reported by Richardson and Walker (1985). It can
be seen that, in each case, the mineral changed from a non-float
to a float condition at a characteristic potential. The potential of
onset of flotation of pyrite is similar to that found by Gardner and
Woods (1973) with their particulate bed system.
Guy and Trahar (1984) applied their modified Denver flotation
cell to the galena/ethyl xanthate system. Flotation of galena from
mixtures with quartz was found to be a function of potential, but
the potential dependence varied with the mode of pretreatment of
the mineral (Figure 8).
The left-hand diagram in Figure 8 shows the potential
dependence after grinding the mineral in a ceramic or a
stainless-steel mill and flotation is observed from the most
negative potential induced. When a mild-steel mill was used
(middle diagram), negligible flotation was observed below about
0 V. The difference between these two relationships could be
explained in terms of the redox conditions in the mill. In the
ceramic and stainless-steel mills, the mineral is free to oxidise by
reaction with atmospheric oxygen, but the mild-steel mill will FIG 9 - Anodic current and flotation response of galena in the
provide cathodic protection and keep the mineral at a low presence of ethyl xanthate following preparation under reducing
potential. A similar situation can be achieved by grinding in the or oxidising conditions from Guy and Trahar (1985).
presence of sodium sulfide. The right-hand graph in Figure 8
shows the results obtained with a stainless-steel mill in the The potential dependence of the flotation of galena has also
presence of 10-2 mol dm-3 sodium sulfide solution. It can be seen been determined in a pilot plant operating with Mt Isa Mines
that the recovery/potential relationship observed is similar to that lead/zinc ore (Johnson et al, 1982) using hydrazine to adjust the
for the mild-steel mill. potential and the results are shown in Figure 10.
Figure 9 shows the correlation made by Guy and Trahar between Comparison of Figures 8 - 10 shows that the potential of onset
their flotation recovery data for galena in the presence of ethyl of flotation in the pilot plant is the same as that in the laboratory
xanthate and the voltammetry of Woods (1971, 1972). It can be investigations in which the mineral was ground under reducing
seen that, for mineral ground under reducing conditions, the onset conditions. Thus, such a pretreatment would appear to be the one
of flotation occurs at the same potential as does the anodic peak
most relevant to flotation practice. It is interesting to note that
due to chemisorbed xanthate (cf Figure 5). When galena is
flotation of galena was found to be initiated at 0.1 V in the
pre-oxidised, some sulfur is lost from the surface as thiosulfate or
sulfate. Thus, when the potential is subsequently taken to low particle bed electrode cell used by Gardner and Woods (1973).
values, the oxidised surface species are reduced to lead metal on Grano et al (1990) also carried out studies on the potential
the surface. On the subsequent potential scan, an anodic peak is dependence of the flotation of galena from Mt Isa lead-zinc ore
observed that occurs at potentials expected for the reaction of lead (Figure 11). It can be seen from the figure that the recovery/
with xanthate to form lead xanthate. In the work of Guy and Trahar potential relationship they obtained is similar to that in Figures 8
for mineral prepared under oxidising conditions, the flotation - 10.
correlates with this current peak and, in this case, lead xanthate It can be seen in Figures 8 and 9 that there is an upper as well
would appear to be the species inducing flotation. The flotation as a lower potential limit of flotation of galena and that it is
recovery increases significantly, however, when the potential is in independent of the pretreatment conditions. A similar upper limit
the region of chemisorption of xanthate on galena. of flotation was observed for the flotation of chalcocite with ethyl

FIG 8 - Flotation recovery of galena in a modified Denver cell at pH 8 in the presence of 10 mg of ethyl xanthate for different
pretreatment conditions from Guy and Trahar (1984).

128 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION
FIG 10 - Flotation response of galena in Mt Isa lead/zinc ore as a
function of potential at pH 8.3 in the presence of ethyl xanthate
(from Johnson, Jowett and Heyes, 1982).
FIG 11 - Potential dependence of galena flotation from Mt Isa
lead/zinc ore with ethyl xanthate at pH 8.1 from Grano et al (1990).
xanthate (Heyes and Trahar, 1979; Basilio et al, 1985). The
results of Heyes and Trahar (1979) are shown in Figure 12; note
that this figure covers a wider potential range than that covered in
Figure 6.
It can be seen from Figure 12 that the potential at which the
upper flotation edge occurs varies with change in pH. For both
galena and chalcocite, the upper limit is close to the potential at
which the metal xanthate is known to oxidise to the metal oxide
and diethyl dixanthogen, and displays the expected pH
dependence. Since dixanthogen is more hydrophobic than metal
xanthates, oxidation to this compound and a metal oxide would
not, alone, be expected to inhibit flotation. Oxidation of the
mineral itself also occurs in the relevant potential range and it is
likely that, once the mineral surface is no longer passivated by a
metal xanthate, oxidation occurs to form significant quantities of
metal oxide and this renders the surface hydrophilic.
FLOTATION RECOVERY AS A FUNCTION OF
COLLECTOR COVERAGE
Voltammetry provides a method of determining collector
coverage as a function of potential under analogous conditions to
Flotation Plant Optimisation Spectrum Series 16
FIG 12 - Flotation recovery of chalcocite with ethyl xanthate as a
function of potential at pH 8 and 11 from Heyes and Trahar (1979).
those used to determine flotation recovery. Figure 13 shows
flotation recovery and surface coverage of chemisorbed xanthate
as a function of potential for three mineral/collector systems. The
flotation recoveries for galena/ethylxanthate in 13a are from Guy
and Trahar (1984) for mineral ground under reducing conditions
and the surface coverage from Buckley and Woods (1991); data
for the chalcocite/ethyl xanthate in 13b is from Woods et al
(1990); the flotation recovery for the chalcocite/diethyldithio-
phosphate system is from Woods et al (1993) and the surface
coverage from Buckley and Woods (1993).
It can be seen from Figure 13 that, for each system, significant
flotation recovery is observed at low thiol coverage, about 50 per
cent recovery at a fractional coverage of <0.2, and maximum
recovery is achieved when there is about half a monolayer of
chemisorbed collector molecules on the mineral surface.
THE POTENTIAL DEPENDENCE OF FLOTATION
IN THE ABSENCE OF COLLECTORS
Some sulfide minerals, such as molybdenite and stibnite are
naturally hydrophobic as a result of cleavage occurring between
planes of sulfur atoms held together only by Van der Waal’s
forces. Such surfaces do not have high affinity for polar water
molecules and adhere readily to air bubbles and the minerals
display floatability in the absence of collectors. Other sulfides
float in the absence of collectors when exposed to mild oxidising
conditions.
Heyes and Trahar (1977) showed that the floatability of
chalcopyrite in the absence of collectors to depend on potential as
shown in Figure 14. In these experiments, flotation was carried
out in their modified Denver cell using nitrogen as the carrier
gas. The potential was controlled by adding redox reagents
(hypochlorite for oxidation and dithionite for reduction) and was
monitored with a platinum electrode inserted in the pulp. Gardner
and Woods (1979) confirmed the potential dependence of self-
induced flotation of chalcopyrite using a Hallimond tube cell in
which the potential was controlled electrochemically. Luttrell and
Yoon (1984) also found the same potential dependence for
chalcopyrite using a Hallimond tube cell with redox reagents for
potential control.
It can be seen from Figure 14 that there is a good correlation
between the potential of onset of flotation and the commencement
of the anodic current. The current arises from oxidation of
chalcopyrite to remove iron and leave an iron-deficient copper
129
CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION
A B
FIG 13 - Comparison of potential dependence of flotation recovery and collector coverage (A) galena/ethyl xanthate,
(B) chalcocite/ethyl xanthate, and (C) chalcocite diethyldithiophosphate.
FIG 14 - Potential dependence of flotation of chalcopyrite in the
absence of collectors: –– • –– • –– Heyes and Trahar (1977);
• • • • Luttrell and Yoon (1984); – – – Gardner and Woods (1979).
Anodic current on a positive-going voltammogram from a
chalcopyrite electrode –––––– Gardner and Woods (1979).
sulfide, which is considered (Buckley and Woods, 1984; Buckley,
Hamilton and Woods, 1985) to be responsible for flotation in the
absence of collectors.
Galena, pyrrhotite and pentlandite are also non-floating under
reducing conditions and floatable in the absense of collectors under
oxidising conditions (Guy and Trahar, 1985; Healy and Trahar,
1989). It was pointed out by Woods (1991) that the potentials
reported by Healy and Trahar for 50 per cent flotation recovery of
these minerals in one minute correlated with the equilibrium
potentials for the oxidation of each mineral to form sulfur. This
suggests that the important species in the self-induced flotation of
sulfides is a sulfur species similar to that formed on chalcopyrite.
Attempts were made by Control International to exploit
self-induced flotation commercially in Arizona copper mines in
130 Spectrum Series 16
the late 1980s (Arbiter and Gebhardt, 1992). Flotation recoveries
in the complete absence of collector failed, however, to match
those achieved in conventional practice and process development
was terminated. A more optimistic application of collectroless
flotation of chalcopyrite was recently reported by Heyes (2007).
A strategy to separate chalcopyrite from chalcocite and bornite
was developed based on work at CSIRO in the late 1970s (Heyes
and Trahar, 1977, 1979). Flotation in the absence of collectors
produced a concentrate that was predominantly chalcopyrite and
the other copper sulfide minerals were then floated following
collector addition. This procedure was devised to maintain the
provision of a copper concentrate with a copper-sulfur ratio
suitable for direct to blister smelting technology as the
chalcopyrite content of the ore is increased as a result of the mine
being converted from underground to open cut operation.
It should be pointed out that self-induced flotation will occur
whenever the potential of the mineral of interest is in the region
of surface oxidation. Thus, self-induced flotation acts as a
complication in collector-induced flotation.
ACTIVATION AND DEPRESSANT ACTION
The separation of one sulfide mineral from another in a complex
ore depends on the ability to selectively depress chosen
components.
Consideration of flotation in terms of a mixed potential
mechanism provides a means of exploring the possible ways in
which flotation can be enhanced or depressed. This will be
illustrated by a discussion of the curves presented in Figure 15.
The solid lines represent collector oxidation (anodic) and
oxygen reduction (cathodic) occurring at a mixed potential A.
This system can be modified in the following ways:
• The mixed potential can be shifted to B if the oxygen current
can be decreased to give curve b1, eg by adding a reagent such
as sulfite that diminishes the oxygen content. In this case the
collector oxidation current becomes zero, no oxidised
collector species are formed, and flotation is depressed.
Flotation Plant Optimisation
 CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION
FIG 15 - Schematic diagram showing the effect of various factors
on the mixed potential of collector/oxygen systems.
• If the collector oxidation current is increased, ie the curve
becomes b2, the new mixed potential B is characterised by
greater currents and flotation will be improved. This can be
achieved by increasing the collector concentration or
substituting a homologue of the collector with longer alkyl
chain length.
• If collector oxidation is inhibited such that the anodic wave
becomes curve c, flotation will be depressed. In this case,
the mixed potential will be moved to point C at which value
negligible collector is oxidised. The effect of cyanide in
depressing pyrite, in which a surface iron cyanide species is
formed is an example of this mechanism.
• Flotation can be depressed by the addition of a species that
oxidises at a lower potential than that for the collector. Thus,
if the additional reaction is represented by curve b2, the mixed
potential B, at which the rate of reaction is equal and opposite
to oxygen reduction, corresponds to zero collector oxidation
and hence no flotation. The use of NaSH to depress
chalcopyrite in copper/moly plants follows this mechanism.
• Flotation can also be depressed if the solution conditions are
changed such that the mineral oxidises in a potential region
below that for the collector. That is, mineral oxidation
becomes represented by curve b2 and the mixed potential,
again, involves negligible collector oxidation. Change in pH
can bring about this situation. Another example is the action
of cyanide on copper minerals in which situation oxidation
of the mineral occurs at lower potentials as a result of the
formation of copper cyanide complexes in solution.
Each of the mechanisms discussed above have been confirmed
by electrochemical investigations. For example Castro and
Lorrondo (1981) validated the mechanism of the depressant
action of cyanide for chalcocite in the presence of propyl
xanthate. These authors correlated measurements of flotation
recovery with the potential of a chalcocite electrode in the
solution. They found that depression of the mineral was
accompanied by a decrease in potential and the appearance of
copper in solution. The dissolution of chalcocite to form a copper
cyanide complex ion shifted the potential below the region in
which xanthate reactions occur.
SELECTIVE FLOTATION RECOVERY AS A
FUNCTION OF POTENTIAL
The early work on the potential dependence of flotation led
researchers to assume that the development of separation
Flotation Plant Optimisation Spectrum Series 16
strategies from differences in recovery/potential relationships
established experimentally for individual minerals would be
uncomplicated. Progress in developing flotation schemes this way
became constrained, however, by the finding (Guy and Trahar,
1985; Gebhardt et al, 1985; Gebhardt and Richardson, 1987) that
the behaviour of mixtures could not be predicted accurately
from single mineral studies due to mineral–mineral interaction.
Nevertheless, flotation recovery of minerals in pulps containing
more than one sulfide mineral is still a function of potential. It is
just that deriving the potential dependence of minerals in
mixtures may not be as straightforward as assuming they are the
sum of the responses of the individual components determined
separately.
Chander and Fuerstenau (1983) utilised the potential
dependence of the flotation of chalcocite with diethyl
dithiophosphate (DTP) to develop an electrochemical strategy for
separating chalcocite from molybdenite. Chalcocite was found
not to float with DTP at potentials less than 0.0 V, but to float
readily at potentials greater than 0.2 V. Molybdenite displays
natural hydrophobicity and its floatability is not affected by
potential. Thus, this sulfide can be selectively floated from
chalcocite at negative potentials. It was found that 66 per cent of
the molybdenite and only 14 per cent of the chalcocite floated
from a 50:50 mixture of the two minerals at -1.2 V. When the
potential was raised to 0.3 V, 100 per cent of the chalcocite, and
only 17 per cent of the molybdenite, floated. It was considered
that the apparent depression of molybdenite at positive potentials
results primarily from competition with the more hydrophobic
constituent, chalcocite.
In recent years, a significant amount of research has been
directed towards developing approaches for the effective removal
of arsenic-containing minerals from copper and nickel ores prior
to smelting since arsenic is a serious environmental hazard and
causes problems in the metal recovery processes. This results in
ores containing arsenic to be subject to severe penalties when
being sold to smelters. The progress made in this area exemplifies
what can be achieved by taking into account the different
potential dependence of flotation recovery of sulfide minerals.
In copper ores, arsenic occurs mainly as enargite (Cu3AsS4)
and to a lesser extent as tennantite (Cu12As4S13) and research has
been carried out to remove these minerals by selective flotation.
Castro et al (2003) determined the flotation response of enargite,
chalcopyrite, chalcocite and pyrite and found (Figure 16) that the
100
90
80
70 Enargite
Chalcopyrite
Recov ery, %
60 Pyrite
Chalcocite
50
40
30
20
10
0
280 300 320 340 360 380 400 420
Potential, m V (SHE)
FIG 16 - The effect of oxidation pre-treatment with hydrogen
peroxide at pH 9.8 on the flotation recovery of energite,
chalcopyrite and pyrite with isopropyl xanthate at pH 6.8
(from Castro et al, 2003).
131
CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION

only mineral that was not strongly depressed at high potentials
was enargite. From these studies, it was concluded that selective
flotation of enargite from copper and iron sulfide minerals is
technically feasible by applying a pre-oxidation stage to depress
copper and iron minerals. This is consistent with the findings of
Fornasiero et al (2001) that selective oxidation with hydrogen
peroxide allows the selective removal of enargite and tennantite
from chalcocite, covellite and chalcopyrite in mixed metal
systems.
Guo and Yen (2005) also studied the influence of potential on
the flotation recovery of enargite and chalcopyrite to assess the
feasibility of enargite removal from a chalcopyrite concentrate.
Their results (Figure 17) showed that chalcopyrite begins to
oxidise quickly at a much lower potential than does enargite.
potential and flotation recovery of single minerals, to develop a
flotation strategy for the separation of enargite from other copper
minerals in the Tampakan copper-gold deposit in Mindanao,
Philippines (Senior, Guy and Bruckard, 2006). The potential
dependence of flotation recovery of enargite reported by Senior,
Guy and Bruckard is similar to that of Guo and Yen for amyl
xanthate shown in Figure 17. The difference in the flotation
threshold for enargite and chalcopyrite was found to be greater,
however, and it was concluded that enargite could be floated from
copper-iron sulfides such as chalcopyrite at pH 8 to 11 at 0 V
versus SHE. A conceptual flow sheet was devised using the
addition of redox reagents and variation in the flotation gas
composition to control potential.
The first stage aims to exploit the difference in the onset of
flotation for chalcocite and cuprite from those of enargite and
chalcopyrite (see Figures 7 and 17). The second stage seeks to
take advantage of the difference in the potential of onset of
flotation of enargite and that of copper-iron sulfides. Any
chalcocite and cuprite not floated in stage 1 will follow the
enargite as the potential in the second stage is in the potential
range for the efficient flotation of these minerals. The third stage
utilises the difference in the potential of the upper flotation edge
between enargite and chalcocite and cuprite (see Figures 12
and 17). It was reported in later studies (Smith and Bruckard,

FIG 17 - Effect of potential on the flotation recovery of enargite and

chalcopyrite at pH 10 in the presence of 7 × 10-5 mol dm-3
potassium amyl xanthate (from Guo and Yen, 2005).

Guo and Yen (2005) found that enargite could be floated

efficiently at a potential higher than +0.2 V versus SCE (0.444
versus SHE) while chalcopyrite was completely depressed at
potentials higher than this value. Selective flotation revealed that
enargite can be successfully removed from chalcopyrite through
controlling the pulp potential in the range +0.2 to +0.55 V versus
SCE, with the most effective separation being achieved at 0.5 V
versus SCE (0.744 versus SHE). These authors noted that the
flotation response of enargite was greater in mineral mixtures
with chalcopyrite than when floated alone. The work of Guo and
Yen illustrates that, although single mineral studies may not give
precise data for mixtures, they can be used to identify conditions
for selective flotation in polymetallic ores.
More recently Leanne Smith (née Shannon) and co-workers,
building on the many years of experience in investigating the
potential dependence of flotation at CSIRO Minerals, have
reported (Kaye, 2006) the development of a flotation process to
separate the arsenic-containing components from other copper
sulfides in ores. By studying the flotation behaviour of individual
minerals as a function of potential, the CSIRO team identified
several electrochemical windows whereby it is possible to
selectively float copper-arsenic minerals from other copper
minerals. The CSIRO flotation process is reported to involve:
… some changes to existing chemicals used in the
separation process as well as possible changes to
the flotation gas, such as the use of nitrogen
instead of oxygen.
Unlike the flotation separation proposals put forward by Castro
et al and by Guo and Yen, it is not restricted to floating under
highly oxidising conditions. FIG 18 - Conceptional flow sheet for a Tampakan arsenic rejection
CSIRO Minerals, in partnership with BHP Billiton, has drawn circuit, based on pulp potential effects from Senior, Guy and
on its extensive databank of experimental relationships between Bruckard, 2006).

132 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION
2007) that one of the separation methods proposed in the work of
Senior, Guy and Bruckard (2006) was successfully tested on
Tampakan ore and enabled the production of a copper concentrate
with sufficiently low arsenic content to be sent to a conventional
smelter for copper metal production.
Smith and Bruchard (2007) showed that it is possible to
selectively separate tennantite from chalcopyrite and bornite in
Northparkes copper-gold ore using pulp potential control. The
conceptual flowsheet developed to treat high-arsenic regions of
the orebody involved producing a bulk copper-arsenic concentrate
using isopropyl xanthate and a dithiophosphate as collector. A
tennantite-rich concentrate was then obtained by floating at pH
12 and an Eh of -120 mV using nitrogen as the carrier gas. A
copper concentrate, constituting about 61 per cent of the copper
content of the ore, with low enough arsenic content to avoid
smelter penalties, could then be produced at pH 12 and the Eh
generated by using air as the carrier gas.
MONITORING POTENTIAL IN FLOTATION
PRACTICE
Copper-moly flotation
The use of nitrogen to replace air in the copper/moly separation
stage of flotation plants was introduced in the mid 1980s in North
and South American plants to minimise oxidation of the sodium
hydrosulfide (NaSH) used to depress copper sulfides. At the same
time, sulfide additions were monitored and controlled by the use
of noble metal electrodes inserted in the pulp. The electrode
becomes coated with sulfur and then responds to hydrosulfide
ions in solution displaying the equilibrium potential of the HS-/S0
redox couple, which is given by the Nernst equation (see:
http://electrochem.cwru.edu/ed/dict. htm#n02):
E (V vs SHE) = –0.065 – 0.0295 pH – 0.295 log [HS-]
Thus, pH needs to be recorded as well as the potential of the
redox electrode for proper control of NaSH additions.
Aravena (1987) reported that NaSH consumption was reduced
by 48 per cent with the introduction of nitrogen at the
Chuquicamata mine in Chile and a further saving of 21 per cent
was achieved when depressant additions were controlled by
electrode measurements.
Controlled potential sulfidisation (CPS)
It is well established that oxidised lead and copper ores can
be activated by treatment with hydrosulfide ions. Jones and
Woodcock (1978) proposed that sulfidisation of oxidised lead
and copper ores should be carried out under potential control
to improve performance. Controlling the potential during
sulfidisation ensures that the oxide component of the ore is
sufficiently converted to sulfide for it to float, while the sulfide ions
remaining in solution are insufficient to depress subsequent
flotation. In both CPS and copper/moly separation, the electrode
responds to solution species, although the solution species are
responsible for control of the potential of the mineral/solution
interface. Woodcock and Jones suggested that silver sulfide
ion-selective electrodes be applied but this proved to be
inappropriate in plant practice and were replaced by noble metal
electrodes (Jones, 1991).
Major improvement in performance results from the use of
CPS compared with that obtained with ‘slug’ addition of NaSH.
In Australasia, CPS, marketed through AMDEL laboratories, is
being, or has been, applied at the Boddington, Cadia, Horseshoe,
Mineral Hill, North Parkes, Ok Tedi, Red Dome, Selwyn and
Telfer mines for the treatment of copper/gold ores containing
reactive sulfides or mixed oxides and sulfides (Jones, 1991;
Woodcock and Hamilton, 1993; Woodcock, pers comm).
Flotation Plant Optimisation
Mekhanobr in St Petersburg, Russia has also marketed a CPS
system (Mashevsky, pers comm) and it has been applied in mines
in Europe and Asia for copper/moly separation and the flotation
of oxidised lead-zinc and copper ores.
Fuerstenau and co-workers (Herrera-Urbina et al, 1999) have
also concluded that the xanthate flotation of sulfidised, mixed
oxide-sulfide ores may be controlled by monitoring pulp
potential.
Outokumpu recently suggested (Heimala and Jounela, 2000)
that djurleite (Cu1.96S) electrodes should be used for CPS
flotation of oxidised copper and cobalt ores with the potential
controlled between -600 mV and -300 mV versus SCE, the
optimum value depending on mineral composition. These authors
reported that effective mixing of NaSH additions was important
in achieving good grades and recoveries.
Sulfide flotation plants
More challenging for the flotation engineer than controlling
sulfide additions is the monitoring and control of the potential of
the mineral/solution interface across the full range of flotation
situations in order to enhance performance and maintain it at a
high level. A number of groups throughout the world have
researched the concept of controlling flotation plants through
continuous measurement of potential in the pulp.
The questions to be answered before a control system can be
made effective are:
• how should the potential be measured, and
• how should the potential be adjusted.
As pointed out above, noble metal electrodes are appropriate
for determining the potential in systems in which a dominant
redox couple determines the potential, such as in monitoring
NaSH additions. Such electrodes can also be used in conventional
flotation pulps for monitoring pulp potential; for example, Hecker
et al (1999) applied platinum electrodes for potential monitoring
in the El Teniente Colon concentrator’s SAG line feeding the
copper flotation rougher circuit. However, the potential measured
by a noble metal electrode in a conventional pulp may not be that
corresponding to a mineral/solution interface of interest. Rand
and Woods (1984) demonstrated that a platinum or gold electrode
could express a potential close to that of a galena electrode
inserted into the same pulp due to continued collision between
the noble metal surface and galena particles. However, this can
not apply to complex ores, because different sulfide minerals in a
pulp can be at different potentials (Ruonala, Heimala and
Jounela, 1997). Also, noble metal electrodes do not always
correspond to sulfide mineral potentials in dilute pulps, such as in
the flotation of platinum group metal ores (Buswell et al, 2002).
It would appear that the most appropriate method of
determining the potential of each sulfide mineral in the flotation
of complex ores is to use electrodes composed of the individual
minerals being floated or depressed. However, the surface of the
mineral electrodes will continuously react with the collector and
other constituents of the pulp and hence will not continue to
respond in the same manner as the freshly ground mineral
particles of the same composition in the flotation pulp.
Outokumpu has addressed this problem (Ruonala, Heimala and
Jounela, 1997) by regenerating sulfide mineral electrodes ‘by
electricity or ultrasound as well as by mechanical cleaning’. It is
difficult to know to what extent these treatments solve the
problem, or if further research is required to develop techniques
to maintain the activity of sulfide mineral electrodes and ensure
that the potential expressed is the same as that of the particular
mineral in the flotation pulp being monitored.
Adjustment of potential can be achieved by modifying the
conditioning time or by changing the composition of the pulp or
Spectrum Series 16 133
CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION
flotation gas. It is interesting to note that Newmont Technologies
Ltd has developed a process, N2Tec®, for the flotation of
refractory gold ores that employs nitrogen as the carrier gas. This
process is employed at the Lone Tree mine and at the Tonkin
Springs gold project (Symmons et al, 1999). Presumably, the use
of nitrogen inhibits the oxidation of auriferous sulfide mineral
fines. Measuring pulp potential would appear to be a useful
method of monitoring such a process.
In most cases, the pulp chemistry that gives good flotation
recovery has been established by conventional methods, and only
small changes in potential are required to maintain flotation
efficiency at the optimum level. A recent paper (Panayatov, 2000)
described technology for direct potential control using electrodes
inserted in the pulp, but regulating pulp chemistry offers a
simpler and more economic approach.
Outokumpu Oy in Finland has progressed further than any
other companies in the application of potential measurement for
flotation control and has introduced its OK-PCF potential
monitoring system in a number of its flotation plants. The system
has been in operation at the Hitura nickel mine since 1984, at the
Vammala nickel and Vihanti Cu/Zn/Pb mines since 1988 and was
installed at the Pyhasalmi Cu/Pb/Zn/FeS2 mine in the early
1990s. At Hitura, the replacement of pH control by potential
control resulted in an improvement in nickel recovery by up to
two per cent depending on ore type (Ruonala, Heimala and
Jounela, 1997). At Vihanti, it is claimed that lime and collector
consumption has been decreased to one third of the values
experienced prior to the installation of the OK-PCF system. It is
claimed that profitability has increased by ten to 20 per cent since
potentials have been monitored.
Outokumpu use electrodes composed of synthetic or natural
sulfide minerals, usually of the composition being floated. For each
valuable mineral, it is claimed (Ruonala, Heimala and Jounela,
1997) that there may be one or more combinations of variables for
the most economic processing, and this requires integration of
on-stream X-ray fluorescence analysis, electrochemical potential
measurements, and collector concentrations. Despite a number of
years of experience in the application of electrode measurements in
flotation plants, Outokumpu has yet to market a potential
monitoring or control system.
Outokumpu insists (Ruonala et al, 1997) that, because different
sulfide minerals can express different potentials in the same slurry,
electrodes prepared from the sulfide minerals present in the
flotation pulp must be used to maximise separation and recovery.
Outokumpu is marketing a plant audit procedure (Sotka et al,
2003; Oravainen et al, 2005) that includes the measurement of
potentials in the flotation plant with mineral electrodes. The
application of Outokumpu’s electrochemical knowledge would
now appear to be focused on its audit procedure rather than on-line
control.
REFERENCES
Adam, K, Natarajan, K A and Iwasaki, I, 1984. Grinding media wear and
its effect on the flotation of sulfide minerals, Int J Miner Process,
12:39-54.
Aravena, J J, 1987. Column flotation applications at Chuquicamata’s
molybdenite flotation plant, in Proceedings Copper ’87 (eds: A Mular,
G Gonzalez and C Barahona), Universidad de Chile, Santiago, vol 2,
pp 155-169.
Arbiter, N and Gebhardt, J E, 1992. Requirements for industrial
collectorless flotation of sulfide minerals, in Proceedings
International Symposium Electrochemistry in Mineral and Metal
Processing III (eds: R Woods and P E Richardson), pp 1-13 (The
Electrochemical Society: Pennington).
Basilio, C, Pritzker, M D and Yoon, R-H, 1985. Thermodynamics,
electrochemistry and flotation of the chalcocite – Potassium ethyl
xanthate system, SME-AIME Annual Meeting, New York, preprint no
85-86.
134 Spectrum Series 16
Biegler, T, Rand, D A J and Woods, R, 1977. Oxygen reduction on
sulphide minerals, in Trends in Electrochemistry (eds: J O’M Bockris,
D A J Rand and B J Welch), pp 291-302 (Plenum: New York).
Buckley, A N, Hamilton, I C and Woods, R, 1985. Investigation of the
surface oxidation of sulphide minerals by linear potential sweep
voltammetry and X-ray photoelectron spectroscopy, in Developments
in Mineral Processing (ed: K S E Forssberg), vol 6, pp 41-60
(Elsevier: Amsterdam).
Buckley, A N, Hope, G A and Woods, R, 2003. Metals from sulfide
minerals: The role of adsorption of organic reagents in processing
technologies, in Solid-Liquid Interfaces Macroscopic Phenomena –
Microscopic Understanding (eds: K Wandelt and S Thurgate), Topics
in Applied Physics, 85, pp 61-96 (Springer-Verlag: Berlin).
Buckley, A N and Woods, R, 1984. An X-ray photoelectron spectroscopic
study of the oxidation of chalcopyrite, Aust J Chem, 37:2403-2413.
Buckley, A N and Woods, R, 1991. Adsorption of ethyl xanthate on
freshly exposed galena surfaces, Colloids Surf, 53:33-45.
Buckley, A N and Woods, R, 1993. Underpotential deposition of
dithiophosphate on chalcocite, J Electroanal Chem, 357:387-405.
Buswell, A M, Bradshaw, D J, Harris, P J and Ekmekci, Z, 2002. The use
of electrochemical measurements in the flotation of a platinum group
minerals (PGM) bearing ore, Minerals Eng, 15:395-402.
Castro, S H, Balttierra, L and Hernández, C, 2003. Redox conditions in
the selective flotation of Enargite, in Proceedings International
Symposium on Mineral and Metal Processing VI (eds: F M Doyle,
G H Kelsall and R Woods), p 27 (The Electrochemical Society:
Pennington).
Castro, S H and Larrondo, J, 1981. An electrochemical study of
depression of flotation of chalcocite by cyanide and iron-cyanide
complexes, J Electroanal Chem, 118:317-326.
Chander, S and Fuerstenau, D W, 1983. Electrochemical flotation
separation of chalcocite from molybdenite, Int J Miner Process,
10:89-94.
Cook, M A and Nixon, J C, 1950. Theory of water-repellent films on
solids formed by adsorption from aqueous solutions of heteropolar
compounds, J Phys Colloid Chem, 54:445-712.
Fornasiero, D, Fullston, D, Li, C and Ralston, J, 2001. Separation of
enargite and tennantite from non-arsenic copper sulfide minerals
by selective oxidation and dissolution, Int J Miner Process,
61:109-119.
Gardner, J R and Woods, R, 1973. The use of a particulate bed electrode
for the electrochemical investigation of metal and sulphide flotation,
Aust J Chem, 26:1635-1644.
Gardner, J R and Woods, R, 1977. An electrochemical investigation of
contact angle and of flotation in the presence of alkylxanthates. II.
Galena and pyrite surfaces, Aust J Chem, 30:981-991.
Gardner, J R and Woods, R, 1979. An electrochemical investigation of
the natural floatability of chalcopyrite, Int J Miner Process, 6:1-16.
Gaudin, A, 1939. Principles of Mineral Dressing, 554 p (McGraw-Hill:
New York).
Gebhardt, J E, Dewsnap, N F and Richardson, P E, 1985. Electrochemical
conditioning of a mineral particle bed electrode for flotation, Report
Invest – US Bureau Mines, RI 8951, 14 p.
Gebhardt, J E and Richardson, P E, 1987. Differential flotation of a
chalcocite-pyrite particle bed by electrochemical control, Miner
Metall Process, 4:140-145.
Goh, S W, Buckley, A N, Lamb, R N and Woods, R, 2006. The ability of
static secondary ion mass spectrometry to discriminate submonolayer
from multilayer adsorption of thiol collectors, Minerals Eng,
19:571-581.
Grano, S R and Huang, G, 2006. Improving the flotation behaviour of a
sulphide ore by controlling electrochemical interactions during
grinding, in ECS Transactions, 2(3):9-20, Electrochemistry in
Mineral and Metal Processing VII (eds: F M Doyle, G H Kelsall and
R Woods) (The Electrochemical Society: Pennington).
Grano, S, Ralston, J and Smart, R St C, 1990. Influence of
electrochemical environment on the flotation behaviour of Mt Isa
copper and lead-zinc ore, Int J Miner Process, 30:69-97.
Guo, H and Yen, W-T, 2005. Selective flotation of enargite from
chalcopyrite by electrochemical control, Minerals Eng, 18:605-612.
Guy, P J and Trahar, W J, 1984. The influence of grinding and flotation
environments on the laboratory batch flotation of galena, Int J Miner
Process, 12:15-38.
Flotation Plant Optimisation
 CHAPTER 7 – ELECTROCHEMICAL ASPECTS OF SULFIDE MINERAL FLOTATION
Guy, P J and Trahar, W J, 1985. The effects of oxidation and mineral
interaction on sulphide flotation, in Flotation of Sulphide Minerals
(ed: K S E Forssberg), pp 91-110 (Elsevier: The Netherlands).
Healy, T W and Trahar, W J, 1989. Challenges in sulfide flotation
fundamental, in Challenges in Mineral Processing (eds: K V S Sastry
and M C Fuerstenau), pp 3-14 (Society for Mining, Metallurgy and
Exploration/The American Institute of Mining, Metallurgical, and
Petroleum Engineers: Littleton).
Hecker, C, Ramirez, J, Beas, E and Cartes, F, 1999. Development of an
on-line Eh-pH electrochemical sensor for the flotation process
control, in Proceedings International Conference Copper
’99 – Cobre ’99 (eds: B Hancock and M Pon), 2:83-94 (The
Minerals, Metals and Materials Society: Warrendale).
Heimala, S and Jounela, S, 2000. Effective flotation of oxidised Cu, Zn,
Ni and Co-minerals, paper presented at Obogashchenije – 2000, St
Petersburg, June.
Herrera-Urbina, R, Sotillo, F J and Fuerstenau, D W, 1999. Effect of
sodium sulfide additions on the pulp potential and amyl xanthate
flotation of cerussite and galena, Int J Miner Process, 55:157-170.
Heyes, G W, 2007. The differential separation of copper sulphides by
flotation, presented to Flotation 07, Cape Town, November.
Heyes, G W and Trahar, W J, 1977. The natural flotatability of
chalcopyrite, Int J Miner Process, 4:317-344.
Heyes, G W and Trahar, W J, 1979. Oxidation-reduction effects in the
flotation of chalcocite and cuprite, Int J Miner Process, 6:229-252.
Hintikka, V V and Leppinen, J O, 1995. Potential control in the flotation
of sulphide minerals and precious metals, Minerals Eng,
8:1151-1158.
Hope, G A, Woods, R, Parker, G K, Watling, K M and Buckley, F M,
2006. Spectroelectrochemical investigations of flotation reagent–
surface interaction, Minerals Eng, 19:561-570.
Johnson, N W, 1988. Application of electrochemical concepts to four
sulfide flotation separations, in Proceedings International Symposium
Electrochemistry in Mineral and Metal Processing II (eds: R Woods
and P E Richardson), pp 131-149 (The Electrochemical Society:
Pennington).
Johnson, N W, Jowett, A and Heyes, G W, 1982. Oxidation-reduction
effects in galena flotation: Observations on Pb-Zn-Fe sulphides
separation, Transactions of the Institutions of Mining and Metallurgy,
Mineral Processing and Extractive Metallurgy, 91:C32-C37.
Jones, M H, 1991. Some recent developments in the measurement and
control of xanthate, perxanthate, sulphide, and redox potential in
flotation, Int J Miner Process, 33:193-205.
Jones, M H and Woodcock, J T, 1978. Evaluation of ion-selective
electrode for control of sodium sulfide additions during laboratory
flotation of oxidized ores, Transactions of the Institutions of Mining
and Metallurgy, Mineral Processing and Extractive Metallurgy,
87:C99-C105.
Kaye, T, 2006. Separation process boosts copper production, Process,
CSIRO research in mineral processing and metal production,
February, p 8.
Leppinen, J O, Hintikka, V V and Kalapudas, R P, 1998. Effect of
electrochemical control on selective flotation of copper and zinc
from complex ores, Minerals Eng, 11:39-51.
Luttrell, G H and Yoon, R-H, 1984. Surface studies of the collectorless
flotation of chalcopyrite, Colloids Surf, A, 12:239-254.
Majima, H, 1969. How oxidation affects selective flotation of complex
sulphide ores, Can Metall Quart, 8:269-273.
Nixon, J C, 1957. Discussion, in Proceedings Second International
Congress of Surface Activity, vol 3, p 369 (Butterworth: London).
O’Dell, C S, Dooley, R K, Walker, G W and Richardson, P E, 1984.
Chemical and electrochemical reactions in the chalcocite–xanthate
system, in Proceedings International Symposium on Mineral and
Metal Processing (eds: P E Richardson, S Srinivasan and R Woods),
pp 81-95 (The Electrochemical Society: Pennington).
Oravainen, H, Heimala, S, Lamberg, P, Liipo, J, Lyyra, M and Sotka, P,
2005. The plant process audit, mineral electrodes and metals
recovery, in Proceedings Centenary of Flotation Symposium,
pp 523-588 (The Australasian Institute of Mining and Metallurgy:
Melbourne).
Flotation Plant Optimisation Spectrum Series 16
Panayatov, V, Kovachev, K, Önal, G, Dovan, J, Diner, H, Bulut, G,
Panayotov, M and Ninova, V, 2000. Technology for copper-zinc
flotation by electrochemical treatment, in Developments in Mineral
Processing, 13, Proceedings XXI International Mineral Processing
Congress (ed: P Massacci), vol B, pp B8b103-108 (Elsevier:
Amsterdam).
Rand, D A J and Woods, R, 1984. Eh measurements in sulphide mineral
slurries, Int J Miner Process, 13:29-42.
Rao, S R and Finch, J A, 2006. Electrochemical studies on heazelwoodite
flotation from Inco matte, in ECS Transactions, vol 2, no 3,
Electrochemistry in Mineral and Metal Processing VII (eds: F M
Doyle, G H Kelsall and R Woods) (The Electrochemical Society:
Pennington).
Richardson, P E, Stout, J V III, Proctor, C L and Walker, G W, 1984.
Electrochemical flotation of sulfides: chalcocite–ethyl xanthate
interactions, Int J Miner Process, 12:73-93.
Richardson, P E and Walker, G W, 1985. The flotation of chalcocite,
bornite, chalcopyrite, and pyrite in an electrochemical-flotation cell,
in Proceedings XV International Mineral Processing Congress,
Cannes, vol II, pp 198-210.
Ruonala, M, Heimala, S and Jounela, S, 1997. Different aspects of using
electrochemical potential measurements in mineral processing, Int J
Miner Process, 51:97-110.
Senior, G D, Guy, P J and Bruckard, W J, 2006. The selective flotation of
enargite from other copper minerals – A single mineral study in
relation to beneficiation of the Tampakan deposit in the Philppines,
Int J Miner Process, 81:15-26.
Shchukarev, A V, Kravets, I M, Buckley, A N and Woods, R, 1994.
Submonolayer adsorption of alkyl xanthates on galena, Int J Miner
Process, 41:99-114.
Smith, L K and Bruckard, W J, 2007. The separation of arsenic from
copper in the Northparkes copper-gold ore using controlled-potential
flotation, Int J Miner Process, 84:15-24.
Sotka, P, Heimala, S, Lyyra, M and Oravainon, H, 2003. Process
optimisation with Outokumpu’s plant audit procedure, extended
abstract, Flotation ’03 Conference, March, Helsinki.
Symmons, G L, Orlich, J N, Lenz, L C and Cole, J A, 1999.
Implementation – Start up of N2Tec flotation at the Lone Tree mine,
SME/AIME Annual Meeting, 1 - 4 March, Denver (Society for
Mining, Metallurgy and Exploration/The American Institute of
Mining, Metallurgical, and Petroleum Engineers: Littleton).
Taggart, A F, Taylor, T C and Knoll, A F, 1930. Chemical reactions in
flotation, Trans Amer Inst Metall Engrs, 87:217-260.
Wark, I W and Cox, A B, 1934. Principles of flotation, I – An
experimental study of the effect of xanthates on contact angles at
mineral surfaces, Trans Amer Inst Metall Engrs, 112:189-244.
Woodcock, J T and Hamilton, J K (eds), 1993. Australasian Mining and
Metallurgy, The Sir Maurice Mawby Memorial Volume, vol 1 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Woods, R, 1971. The oxidation of ethyl xanthate on platinum, gold,
copper, and galena electrodes: Relation to the mechanism of mineral
flotation, J Phys Chem, 75:354-362.
Woods, R, 1972. The anodic oxidation of ethylxanthate on galena
electrodes, Aust J Chem, 25:2329-2335.
Woods, R, 1991. Electrochemistry of sulfide minerals – Stokes medal
address, Chemistry in Australia, 58:392-395.
Woods, R, Kim, D S and Yoon, R-H, 1993. The potential dependence of
flotation of chalcocite with diethyl dithiophosphate, Int J Miner
Process, 39:101-106.
Woods, R, Young, C A and Yoon, R-H, 1990. Ethyl xanthate chemisorption
isotherms and Eh-pH diagrams for the copper/water/xanthate and
chalcocite/water/xanthate systems, Int J Miner Process, 30:17-33.
135
 HOME

CHAPTER 8

Surface Chemical Characterisation for Identifying and

Solving Problems Within Base Metal Sulfide Flotation
Plants
Alan Buckley

Professorial Visiting Fellow, School of Chemistry, University of New South Wales, Sydney NSW 2052.
Email: a.buckley@unsw.edu.au

After obtaining a PhD in solid state physics in 1971, the author carried out research on X-ray photoelectron
spectroscopy (XPS) while a Senior Teaching Fellow in the Chemistry Department at Monash University. On
joining CSIRO in 1975, applied surface science utilising XPS was continued in Sydney, and a long-standing
collaboration with electrochemist (and flotation Living Legend1) Ron Woods, then at CSIRO in Melbourne,
was commenced on sulfide mineral processing research. By 1998, when the author moved from CSIRO to
The University of New South Wales, sulfide mineral surface characterisation by means of XPS had been
augmented by time-of-flight secondary ion mass spectrometry, surface enhanced Raman scattering
spectroscopy, synchrotron XPS and near-edge X-ray absorption spectroscopy. He was instrumental in the
establishment of a soft X-ray spectroscopy facility at the Australian Synchrotron.

Abstract
Introduction
Surface Chemical Characterisation – General Considerations
Specific Surface Chemical Characterisation Techniques and Their Application
Concluding Remarks
Acknowledgements
References

ABSTRACT
The principal surface chemical characterisation techniques of relevance to
the identification and solution of processing problems within base metal
sulfide flotation plants have been described. The main emphasis is on
conventional X-ray photoelectron spectroscopy (XPS), the most widely
applied spectroscopy for obtaining ex situ surface chemical analyses. The
complementary information that can be obtained by synchrotron XPS and
by X-ray absorption spectroscopy is also described. Other important
ex situ characterisation techniques covered are static secondary ion mass
spectrometry and to a lesser extent infrared (IR) spectroscopy, with
particular attention given to the need to obtain information from several
techniques. General considerations for the preparation of specimens for
the ex situ techniques are discussed, including the relative merits of
investigating single particle rather than pulp specimens. Off-stream, but
in situ (wet), surface chemical characterisation techniques are outlined.
Examples of the application of the various techniques have been
presented.
1. Twenty-three Living Legends of flotation were declared at the 2005
Centenary of Flotation Symposium in Brisbane.
INTRODUCTION
As discussed in preceding chapters, the floatability of a particle in
a flotation cell depends on its hydrophobicity, and hence on its
surface chemical composition. The separation of a valuable ore
component from gangue, or the selective flotation of one valuable
component relative to others, normally requires one component to
be significantly more hydrophobic than the other constituents (or
the converse for reverse flotation). Thus it can be important to
know the surface chemical composition of not only the component
to be concentrated, but also the other metal sulfide components or
gangue that should report to the tails. Of course flotation has
been an industrial process for at least a century, and for much of
this period surface chemical characterisation of the particles in
flotation pulps has not been carried out or even feasible. Advances
in flotation technology have usually arisen from empirical
developments rather than as a result of surface chemical
knowledge. However, when a flotation process fails to deliver the
required recovery or grade, or when there is some other need to
understand flotation behaviour, then surface characterisation
becomes not an option but a necessity.

Flotation Plant Optimisation Spectrum Series 16 137

CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
The surface chemical composition of relevance in flotation
systems is usually that of mineral particles either bearing organic
adsorbates, such as thiol collectors, or being organic
adsorbate-free but altered by interaction with oxygen, added
reagents other than collectors such as activators, or normal
constituents of flotation pulps such as metal ions. In the first
category, adsorbed collector coverage can range from the
submonolayer of chemisorbed thiol that is just sufficient to impart
hydrophobicity, to a multilayer of metal thiolate and/or of
dithiolate. In the second category, the altered surface can vary
from one involving only the outermost atomic layer of the sulfide
mineral lattice, such as the interaction of Cu ions with pyrite (eg
Wang, Forssberg and Bolin, 1989), to one involving several
atomic layers, such as the interaction of Cu ions with sphalerite
(eg Finkelstein, 1997). Thus, in both categories, the altered
surface layer to be chemically characterised can vary in thickness
from one atomic layer (~0.2 nm) to several or even many atomic
layers (typically 1 - 5 nm), notwithstanding the fact that it is only
the outermost atomic layer that will have significant influence on
the hydrophobicity. Characterisation of the former calls for
extreme surface sensitivity at the limit of available technology,
while that of the latter is manageable using conventional
laboratory surface analysis techniques such as X-ray
photoelectron spectroscopy (XPS). The advanced approaches,
which usually require access to major infrastructure such as a
synchrotron, have become well-established for flotation research
in recent years, and are now at the stage that they should be
considered for flotation problem-solving when appropriate.
This chapter is concerned with the surface chemical nature of
mineral particles, but readers should be reminded that surface
physical aspects, including microstructure and particle shape, can
also influence floatability. Of course the chemical and physical
aspects are often inextricably linked when the crystal structure is
particularly anisotropic. For example, the basal planes of
molybdenite have been shown, as anticipated, to have lower
reactivity than the edges (Buckley et al, 1995), and a similar
reactivity difference is expected for covellite. In these cases, the
particle size will influence the ratio of edges to basal planes, and
hence the composition of surfaces predominantly encountered by
the bubbles in a flotation cell.
SURFACE CHEMICAL CHARACTERISATION –
GENERAL CONSIDERATIONS
To a large extent, the surface chemical characterisation technique
or techniques applied in any given problem-solving situation will
depend on the relevant altered surface layer thickness, the
feasibility of investigating samples that are truly representative of
the particles in the flotation environment under consideration, and
the time scale in which the information is needed. These general
considerations will be addressed before specific characterisation
techniques are described in detail.
Sample selection is of crucial importance, and hence deserves
serious attention. It goes without saying that any flotation pulp
sample for any characterisation, whether it be bulk or surface,
must be representative of the majority of particles in that part of
the flotation cell or circuit under investigation. The issues
peculiar to surface chemical characterisation are those beyond
representative sampling to those that concern the preservation of
the surface chemical composition, until and during the actual
characterisation measurements, without significant inadvertent
change. Ideally, surface characterisation would be carried out
on-stream; that is, while the mineral particles were in the flotation
cell or circuit, as is the case for analysis of some bulk properties.
However, on-stream surface chemical characterisation is rarely
feasible, although it is achievable in principle with one or two of
the in situ techniques described towards the end of this chapter.
In most cases the characterisation measurements must be made
138 Spectrum Series 16
off-stream, and very often ex situ (dry rather than wet) in which
case the considerable problem of sample preservation and
handling after collection arises.
The collection and subsequent handling of industrial flotation
pulp samples suitable for meaningful surface chemical
characterisation have been considered in detail by Smart and
co-workers at the University of South Australia (Smart, 1991).
Their methodology involves the initial purging of the pulp sample
with nitrogen or argon within 15 min of collection to reduce
dissolved oxygen levels from typically 8 - 40 ppm to less than
0.1 ppm before sealing and freezing within 30 min of sealing.
Immediately before surface characterisation, for example by
XPS, the sample is thawed and, if necessary, deslimed by several
successive applications of a procedure which consists of shaking
(but not ultrasonic agitation), decanting of the supernatant
solution containing the dispersed ultra-fine particles, and
replacement of this pulp solution immediately after filtering
through a 0.1 μm filter. A small quantity of the deslimed slurry is
then inserted into the spectrometer on an electrically conducting
sample stub and the solution evaporated in the sample
introduction chamber of the spectrometer to achieve a vacuum of
approximately 10-5 Pa within 10 min. These workers maintain
that before the mineral particles are exposed to a non-aqueous
environment by evaporation of the solution, essentially all air has
been removed from the chamber, thereby minimising extraneous
oxidation or other alteration of the surface of the particles. In
demonstrating the validity of the procedure, it was found that
deposition of soluble species on to the mineral particles is not
usually a problem. In cases where it is a problem, the actual pulp
liquor is changed for a dilute solution having a similar pH and Eh,
the latter being established with sodium dithionite stabilised with
sodium carbonate for reducing potentials and with hydrogen
peroxide for oxidising potentials. Cooling of specimens prepared
in this way to retain volatile surface species was also carried out,
and water retention by the sample was found not to be a problem,
notwithstanding some complication of spectral interpretation
arising from different degrees of charge-shifting from different
components in the specimen. Surface characterisation of
specimens handled using this procedure was shown to be valid
provided the storage time prior to analysis was less than three
months (Smart, 1991).
One alternative to the surface characterisation of mineral
particles from industrial or pilot scale flotation feeds,
concentrates and tailings is the study of ground sulfide particles
treated under flotation-related conditions in the laboratory.
Simulated flotation pulps prepared in this way can be extremely
useful for determining the effect of a particular variable, such as
Eh, that might be difficult to control even in a pilot plant. Another
alternative is to move still further towards the laboratory
environment by investigating the surface of single mineral
particles. This step might be taken in circuit design or to attempt
to solve a plant problem that potentially requires extremely fine
control of reaction conditions, such as an initial surface that is
completely unoxidised or unaltered. The single mineral particle
approach is the one normally adopted in research when a generic
flotation problem is being investigated. There are many
advantages of investigating either fracture or abraded single
particle surfaces over fine particle specimens, including better
control or knowledge of the surface atomic plane presented to the
aqueous phase, better control of the electrical potential (and
hence enhanced spectral resolution for some techniques), better
temperature control (and hence retention of adsorbates), greater
compatibility (less out-gassing) for those techniques that require
the specimen to be under UHV, easier mounting (anchoring) of
the specimen especially for those techniques for which the
specimen does not remain horizontal, and a ‘flatter’ specimen
surface for techniques where that is an advantage or a necessity
Flotation Plant Optimisation

(eg angle-resolved near-edge x-ray absorption fine structure
(NEXAFS) spectroscopy or IR external reflection spectroscopy).
It is important to note that the surface characterisation of single
mineral particles does not necessarily imply the study of systems
containing only a single mineral component. For example, if the
selective separation of two sulfide minerals (such as pentlandite
and pyrrhotite) is being investigated, systems containing both
minerals should be included in those studied (Trahar, Senior and
Shannon, 1994) to allow for any interactions between the two
sulfides not already covered by the use of a relevant pulp liquor
as the aqueous phase. The possible influence of particle size on
floatability must also be kept in mind (Senior, Shannon and
Trahar, 1994).
Apart from specimen mounting differences, the surface
characterisation principles are largely the same for both particles of
floatable size and single piece specimens. In general, this situation
is predicated on the fortuitous reality that most sulfide minerals
have either metallic conductivity (eg pentlandite) or are low
band-gap semiconductors (eg galena, chalcopyrite) with only a few
being large band-gap semiconductors (eg sphalerite). The main
exception to the similarity arises in characterisation techniques
involving a variation of the angle between the specimen surface
and either a source or a detector. In such cases, special procedures
must be adopted to minimise the typically ‘rough’ surface of a
multi-particulate specimen. For the UHV-based techniques, single
piece specimens are normally held in place on a conducting metal
sample stub by means of metal screws or clips. Particulate
specimens are usually either pressed into pure indium foil or
mounted on double-sided adhesive Cu tape.
For research on single particle surfaces, a decision must be
made on whether to use abraded or fracture surfaces, or if
possible, both. It is clear that for most sulfides, the chemical
composition of a surface abraded in air and washed free of swarf
is different from (and typically sulfur-rich relative to) that of a
fracture surface. This does not mean that an abraded surface is
necessarily inferior, as because of the comminution processes
commonly used, the surface of sulfide mineral particles in a
flotation pulp may be more closely represented by abraded than
fracture surfaces. Ideally, both abraded and fracture surfaces
should be used in any investigation; however, unless the problem
being addressed appears to be one that might be strongly
influenced by any prior alteration of the surface, then it is more
realistic to restrict experiments to abraded surfaces only. There
are a number of difficulties involved in the use of fracture
surfaces, including the acquisition of sufficient mineral of the
desired composition, the conchoidal fracture rather than cleavage
along an ‘easy’ plane exhibited by a majority of sulfides, and the
tendency to fracture along interfaces already weakened by small
inclusions of other phases or weathering products (although that
in itself may be of relevance in the flotation of some minerals).
On the other hand, the main difficulty involved in using abraded
surfaces is the additional complexity introduced by the need to
interpret the observed behaviour of altered surfaces. Indeed, in
order to obtain meaningful data there is sometimes the need to
fracture surfaces under the flotation-related aqueous phase (eg
collector solution).
Separation by flotation takes place in an aqueous medium,
therefore it is legitimate to question the general validity of surface
characterisation effected in the absence of the aqueous phase (ie
ex situ), including under ultra high vacuum. This is tantamount to
questioning whether removal of the pulp liquor and replacing it
with an inert gaseous phase or ultra high vacuum alters the
surface chemical composition of the mineral particles. There is
substantial evidence to indicate that provided precautions are
taken to avoid loss of moderately volatile species such as
dixanthogen from the surface to a vacuum, removal of the
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
aqueous phase in this way does not have a noticeable effect on the
surface chemical composition. This evidence includes the
agreement of in situ and ex situ surface enhanced Raman
scattering (SERS) spectroscopic data, the agreement of cyclic
voltammograms determined before and after emersion, and the
correlation of surface composition and floatability (eg Zachwieja
et al, 1989). Photon beam damage, which may actually be
damage caused mostly by electrons ejected by the photons, can
sometimes be a problem when intense sources such as lasers or
synchrotrons are used; however, such damage is usually minor
compared with that caused by electron or ion beams.
Accordingly, charged particle beams are rarely used in surface
chemical characterisation unless the technique is inherently
destructive as in secondary ion mass spectrometry (SIMS).
The loss of moderately volatile species from the surface of
mineral particles is a possibility that must always be kept in
mind when ultra high vacuum-based surface characterisation
techniques are being used. Although not widely believed when
first noted (Buckley and Woods, 1984), the complete loss from a
specimen surface of a low coverage of even elemental sulfur by
sublimation to an ultra high vacuum is possible. Sublimation can
be prevented by cooling the specimen to well below ambient
temperature with a flow of liquid nitrogen; however, all
condensable material in the vapour phase must be removed to
prevent its deposition on to the specimen surface before cooling
is commenced. Removal of this material, including water vapour,
is usually effected by thorough flushing with dry nitrogen or
argon in an insertion chamber in a procedure similar to that
described by Kartio et al (1992). The specimen must continue to
be maintained at a low temperature when transferred to the
analysis chamber. As an alternative to cooling, it is possible
(although not common) to incorporate differential pumping into
some surface analysis equipment so that the local pressure near a
specimen can be maintained at ~1 mbar while the other parts of
the equipment remain under UHV (Shchukarev, 2006).
For on-stream surface characterisation, essentially contemp-
oraneous information is feasible provided there is someone present
with the expertise to interpret the raw data, or the data is processed
by an ‘intelligent system’. For off-stream surface chemical
characterisation using laboratory techniques, it is possible to obtain
same-day information if the equipment is in a corporate laboratory,
or same-week information if an external service provider is
involved. For characterisation requiring a major facility such as a
synchrotron, the waiting time is usually much longer, as access to
that facility is normally allocated several months in advance on a
competitive, peer-reviewed basis. However, most major mineral
processing countries (with the possible exception of those in
Africa) already have, soon will have, or can readily gain access to,
a synchrotron. Well-established synchrotrons in the United States,
Sweden, Brazil, China, Russia and France are particularly
important in this regard. Most synchrotrons welcome approaches
from the mineral processing sector, and many recently constructed
synchrotrons (including those in Canada and Australia) have been
justified on an economic and scientific case that depended
significantly on the minerals sector. Beam access is typically at no
cost to the user if the research can be published in the open
literature, but synchrotrons often have industry liaison personnel,
and affordable arrangements for beam access at short notice can
usually be made for urgent measurements. As explained later in
this chapter, synchrotron radiation techniques will be used
increasingly for surface chemical characterisation of minerals
because the X-ray energy can be tuned to provide either maximum
surface sensitivity (essential for advanced flotation research and
problem-solving) or maximum elemental sensitivity in electron
spectroscopy, and to enable complementary X-ray absorption
spectroscopy.
139
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
SPECIFIC SURFACE CHEMICAL
CHARACTERISATION TECHNIQUES AND
THEIR APPLICATION
The surface chemical characterisation techniques covered in this
chapter are discussed from the point of view of a minerals
processing engineer. In particular, only generic and ‘need-to-
know’ details of the instrumentation are described, and the
fundamentals of each technique are presented in no greater depth
than the ‘working knowledge’ level. In these two aspects, then,
some fine details may be glossed over and subtleties ignored in
the knowledge that the cited references are available to redress
any imbalance. By contrast, an attempt will be made to treat
interpretation of the data in most of its complexity, as proper data
evaluation is essential in the context of problem-solving or
process optimisation, and it is always preferable if a plant
engineer and research provider can understand the same
technological language. The application to flotation research of
some of the techniques described below has been reviewed
previously (eg Smart et al, 1998, 2003).
The principal ex situ techniques
X-ray photoelectron spectroscopy (XPS) and X-ray
excited Auger electron spectroscopy (XAES)
It is appropriate to start with conventional XPS and to describe
it in most detail, as this spectroscopy is indisputably the
‘workhorse’ ex situ surface chemical characterisation technique.
XPS is widely applicable, and at least at a cursory level, the data
are relatively straightforward to interpret. Not surprisingly then,
at least one electron spectrometer is available in most major cities
and large corporate laboratories where flotation research is
carried out, and the expertise to interpret the data is
correspondingly well distributed. Commercially manufactured
instruments have been available since about 1970, and are now at
a mature stage of development. Nevertheless, in their standard
form, these spectrometers still do not have a glove box around the
load lock and not all have the ability to cool a specimen before
evacuation in the load lock. Succinct descriptions of the
fundamentals of, and instrumentation for, XPS are available in
surface analysis handbooks (eg Briggs and Seah, 1990a) and in
journals (eg Szalkowski, 1977; Chaney, 1987). There have been
previous reviews of the application of XPS to flotation research
(eg Buckley, 1994); however, there have been considerable
advances in the area in recent years.
Essentially, an electron spectrometer consists of a high-
resolution electron energy analyser to determine the energy
distribution of electrons ejected from a specimen by an incident
monochromatic X-ray beam. A vacuum is required to ensure a
sufficiently long inelastic mean free path (IMFP) for the electrons
to pass through the analyser and be detected, but so that the
vacuum quality is sufficiently high for the specimen surface to
remain clean, the vacuum is almost always an ultra high vacuum.
Thus, a spectrometer is based on an ultra high vacuum system
typically constructed of non-magnetic stainless steel. An
alternative arrangement in which the local pressure for the
specimen is ~1 mbar and differential pumping maintains UHV in
other parts of the spectrometer (Ruppender et al, 1990) has
significant advantages as well as some disadvantages but is not
widely used and is not considered further in this chapter. The
photon source is usually of energy 1486.6 eV produced by X-rays
from an aluminium metal anode and rendered monochromatic by
means of a quartz crystal monochromator. Although of greater
linewidth, non-monochromatised 1253.6 eV X-rays from a Mg
anode are also used, especially when Auger electron peaks
interfere with photoelectron peaks ejected by Al X-rays. The
electron energy analyser is usually of hemispherical design with
140 Spectrum Series 16
~150 mm mean radius preceded by an electron lens and with a
solid state detector typically consisting of either a number
(usually three to ten) of individual channel electron multipliers, or
a multichannel plate. The specimen is inserted into the
spectrometer via a load lock and subsequently transferred to the
analysis chamber where it is mounted on a manipulator of some
kind. An electron spectrum is normally obtained by maintaining a
fixed pass energy (eg 20 eV) for the hemispherical analyser, and
incrementing a potential to retard the kinetic energy of the
emitted electrons to this pass energy value.
An X-ray photoelectron spectrum is a plot of the intensity of
detected electrons as a function of their kinetic energy. In most
cases, it is the binding energy of the electrons from each core
level that is of primary interest; consequently it is traditional to
show the spectrum abscissa as binding energy increasing from
right to left, rather than kinetic energy (the actual independent
variable) increasing from left to right. The counter-current
relationship between kinetic energy and binding energy exists
because a photoelectron binding energy, measured relative to the
Fermi level, is simply the exciting photon energy (usually
1486.6 eV) minus the photoelectron kinetic energy minus the
spectrometer work function (typically ~4 eV). A survey electron
spectrum excited by 1486.6 eV photons from a chalcocite fracture
surface prepared in air is shown in Figure 1 as a typical example
of a wide scan.
Cu 2p
Cu LMM
Cu 3p
S 2p
Cu 3s
C 1s
S 2s
O 1s
1200 1000 800 600 400 200 0
Binding Energy (eV)
FIG 1 - Survey electron spectrum from a chalcocite fracture
surface prepared in air.
In the spectrum shown in Figure 1, photoelectron peaks from
the Cu 2p, Cu 3s, Cu 3p, S 2s and S 2p core levels and from the
valence band are observed. A peak from S 1s electrons is not
observed because their binding energy (~2470 eV) is greater than
the photon energy. There are two well-resolved Cu 2p peaks,
Cu 2p1/2 and Cu 2p3/2, because the spin-orbit splitting (~20 eV) is
much greater than the available resolution. The corresponding
splitting of the S 2p1/2,3/2 doublet is not obvious in the wide scan
because it is only 1.19 eV, but it is evident in high resolution
narrow scans. The Cu 3p splitting is also small (2.47 eV)
compared with the experimental linewidth. The expected relative
intensity of the 2p doublet components is [2×(1/2)+1]:[2×
(3/2)+1] = 1:2. The Cu 2p peaks are more intense than the Cu 3p
Flotation Plant Optimisation

because the photoionisation cross-section at a photon energy of
1487 eV for the former is significantly greater than for the latter,
notwithstanding the fact that there are six 2p and six 3p electrons
in each Cu atom. Peaks from Cu LMM Auger electrons are also
observed, the most intense Cu L3M4,5M4,5 peak appearing at an
‘effective binding energy’ of 569 eV or, more correctly, at a true
kinetic energy near 1487 – 569 = 918 eV. Auger electrons are not
ejected directly by photons; rather, they are emitted as part of one
of the processes by which core electron vacancies are filled, the
other process being X-ray emission (fluorescence). Thus, Auger
electrons have a kinetic energy that is independent of the energy
of the photons that created the core electron vacancies by
photoemission. Auger electron emission is more probable than
X-ray fluorescence for low atomic number elements.
Cu L3M4,5M4,5 Auger electrons are produced when a vacancy in
the L3 (2p3/2) level is filled by an electron from the M4,5 (3d3/2,5/2)
levels and another electron from the M4,5 levels is ejected. It will
be shown later that these X-ray excited Auger electrons can be
extremely useful for sulfide mineral systems; however, it should
be noted that ‘Auger electron spectroscopy’ (AES) usually
denotes Auger electron emission brought about by the creation of
core level vacancies by means of an electron beam, as distinct
from the photon beam in XAES. Electron excited AES is
practically useless for surface chemical characterisation of sulfide
minerals because of the concomitant chemical alteration brought
about by the primary electron beam, although it can be of use for
surface elemental analysis. Finally in Figure 1 it can be seen that
small peaks from C 1s and O 1s electrons are discernible, the C
and O most probably adsorbed from the atmosphere after the
mineral had been fractured. The ubiquitous nature of C and O at
the surface of air-exposed surfaces can prove to be a nuisance
when attempts are made to detect collector adsorption by
monitoring photoelectron peaks from those elements. The
adventitious C can also be somewhat useful as an indicator of any
charging of the specimen surface layer. The C 1s binding energy
of the hydrocarbon peak from such contamination should be close
to 285.0 eV relative to the Au 4f7/2 binding energy of 84.0 eV for
Au metal used in calibration.
A wide scan electron spectrum, then, can reveal which
elements are present at the surface, and their relative
concentrations within the depth analysed (typically 1 - 5 nm),
provided those surface concentrations are greater than the
detection limit, which is of the order of one atomic per cent. The
elements present can be identified because the various core
electron binding energies and Auger electron kinetic energies are
essentially unique for each element. Hydrogen has no closed shell
electrons, and consequently photoelectrons from H cannot be
identified as arising from that element. However, the great
strength of XPS (and XAES) arises from the fact that the core
electron binding energies (and Auger kinetic energies) for a
particular element do vary to a small extent to reflect different
chemical environments of that element. In most cases, this
‘chemical shift’ is large enough to allow adequate discrimination
of different oxidation states, with the binding energy increasing
with increasing oxidation state or ‘positive charge’, but not so
large that the origin of a photoelectron peak becomes uncertain.
Thus a survey scan is usually no more than an overview of the
chemical composition of a surface; a detailed surface chemical
characterisation necessitates the determination of a series of
narrow, high resolution scans from which the position of each peak
to ±0.05 eV, its intensity (area above background), and its shape
(including linewidth) can be obtained. In many cases, an electron
spectrum will consist of a partially resolved superposition of
several components, and curve fitting will be required to extract the
binding energy, intensity and shape for each component.
The most important narrow scan for the surface character-
isation of sulfide minerals is that covering the S 2p region. The
S 2p spectrum from a fracture surface of the semiconductor
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
chalcocite is shown in Figure 2 as an example. Although there are
12 S sites (and 24 Cu sites) in the Cu2S unit cell, the S 2p
spectrum is an uncomplicated doublet with no obvious line
broadening arising from the distribution of electronic
environments in the lattice. The 2p3/2 binding energy of
161.55 eV is within the range observed for most monosulfide
minerals, which is from 160.6 eV for galena to 161.7 eV for
millerite. By contrast, the S 2p spectrum from a chalcocite
surface abraded in air and subsequently washed with water
(Figure 2) does not appear to comprise one uncomplicated
doublet. It is certainly different from the spectrum for the fracture
surface, most probably because of the superposition of a second
doublet from an altered, S-rich (Cu-deficient) environment at a
binding energy near 162.3 eV arising from loss of Cu during the
abrasion and washing. Similar, atypically high S 2p3/2 binding
energies are observed for unaltered heazlewoodite, Ni3S2
(162.35 eV) and for the five-coordinate S in pentlandite,
(Ni,Fe,Co)9S8 (162.2 eV). The corresponding values for the
disulfides such as pyrite and vaesite (NiS2) are normally in the
range 162.4 to 162.7 eV. In between those values for the disulfides
and that for elemental sulfur, S8 (163.6 eV) are the 2p3/2 binding
energies for the oligosulfides (Buckley et al, 1988) and severely
metal-deficient sulfide lattice environments (Buckley and Walker,
1988). The values for S-O species such as sulfite (~166.5 eV) and
sulfate (~169 eV) are significantly higher than those for S with a
zero or negative oxidation state.
S 2p
Abraded
Fractured
168 166 164 162 160
Binding Energy (eV)
FIG 2 - S 2p spectrum from chalcocite fracture surface (lower
spectrum) and surface abraded in air and subsequently washed
with water (upper spectrum).
The S 2p spectrum from essentially unaltered covellite
(Figure 3) is much more complicated than that for chalcocite for
at least two reasons. Firstly, there are two quite different S
environments in the unit cell of covellite (CuS but often
represented as Cu2S.CuS2), a monosulfide giving rise to a doublet
with a 2p3/2 binding energy of 161.6 eV, and a disulfide giving rise
to a doublet of twice the intensity of the former at 162.4 eV. The
monosulfide 2p1/2 component is buried underneath the disulfide
2p3/2 component. Secondly, covellite has metallic conductivity;
consequently the lineshape of individual components will be
asymmetric, with a ‘tail’ on the high binding energy side arising
from the high density of states at the Fermi level (Doniach and
Sunjic, 1970). XPS data analysis software packages usually have at
least one asymmetric lineshape that can be used to help fit such
spectra. To further complicate an already complicated situation,
there may also be broad contributions to the spectrum at 2 - 3 eV
higher binding energy from each principal component arising from
energy loss processes. Fortunately, such energy loss contributions
are usually of low intensity relative to the principal photoelectron
141
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
FIG 3 - S 2p spectrum from covellite.
peaks, but their possible presence must always be kept in mind
when subtleties are introduced into spectral interpretation. From a
practical point of view, the best way to approach this complexity is
to interpret changes in spectra resulting from a flotation-related
treatment.
The Cu 2p and Cu LMM spectra for several sulfide minerals
and Cu metal are shown in Figure 4. It can be seen that there are
only subtle differences in the Cu 2p spectra for all the sulfides,
notwithstanding the significant differences in structure. In
particular, the 2p3/2 binding energy for each sulfide is in the
relatively narrow range 931.95 to 932.4 eV, compared with the
value of 932.6 eV for Cu metal. It is not unusual for metal
sulfides of the same formal oxidation state to have core electron
binding energies that are similar and also close to, or even slightly
less than, those for the corresponding metal. For example, the
Ni 2p3/2 binding energies for NiS, (Ni,Fe)9S8, Ni3S2 and Ni metal
are 853.15, 852.9, 852.8 and 852.7 eV, respectively. All Cu
sulfides are Cu(I), even CuS and CuS2. By contrast to the quite
similar Cu 2p spectra, it is evident from Figure 4 that there is a
much greater difference in the Cu LMM spectra for the various
sulfides. Such differences in X-ray excited Auger electron spectra
can be extremely useful, especially when investigating collector
adsorption on Cu- and Ag-containing sulfides, and high
resolution narrow scan Cu LMM and Ag MNN spectra should
always be collected at the same time as the Cu 2p and Ag 3d
photoelectron spectra.
FIG 4 - Cu 2p and Cu LMM spectra from (a) copper metal; (b) chalcocite; (c) covellite; (d) bornite and (e) chalcopyrite.
142 Spectrum Series 16
It should be noted that the analysis depths associated with the
S 2p and Cu 2p electrons are quite different. For excitation
by 1486.6 eV X-rays, S 2p photoelectrons would have a kinetic
energy (KE) of about 1487 – 162 = 1325 eV while the
corresponding Cu 2p value would be 1487 – 932 = 555 eV.
The IMFP for electrons in solids increases approximately in
proportion to KE beyond a minimum at about 50 eV, so the Cu
2p photoelectrons would be significantly more surface sensitive
than the S 2p. While this is of no consequence for a surface layer
that is uniform throughout its depth, it can be of crucial
importance for non-uniform surface layers, for example in a
comparison of the Cu activation of pyrite and sphalerite. The
analysis depth associated with Cu 3p photoelectrons, which have
a binding energy near 77 eV, is closer to that for S 2p electrons.
Thus the analysis depth for conventional XPS varies across the
binding energy range, the greatest surface sensitivity being at the
highest binding energy (lowest kinetic energy) end.
An appreciation of the magnitude of the analysis depth and its
variation with photoelectron energy is fundamental to the
meaningful application of XPS to flotation-related systems. As
discussed in the introduction, for flotation it is the chemical
composition of the outermost atomic layers that is of most
relevance, whereas the analysis depth for XPS varies from about
1 nm to 5 nm. At best, then, XPS has adequate surface sensitivity,
but away from that maximum sensitivity, the contribution to a
spectrum from the outermost atomic layers becomes a minor one.
This can best be illustrated for sulfide minerals by a consideration
of the ‘surface states’ formed in the outermost atomic layers when
bonds are broken during fracture. The S 2p spectrum from pyrite
(Figure 5) obtained by means of conventional XPS at a photon
energy of 1487 eV consists predominantly of one doublet at a
2p3/2 binding energy of 162.4 eV. However, there appears to be a
small contribution from another doublet near 161.7 eV, and it will
become clear from what follows that this contribution is from
disulfide anions at the fracture surface that are no longer in full
coordination in the crystal lattice. The main point at this stage is
to note that the contribution to the spectrum from the surface
states is small.
It is difficult to provide a precise value for the analysis depth.
The intensity of photoelectrons escaping from a surface through a
distance x in a specimen is proportional to (1 – e-x/λ), where λ
is the IMFP. Thus, 63 per cent of the intensity would arise from
x = λ, 86 per cent would arise from x = 2λ, and 95 per cent would
Flotation Plant Optimisation

FIG 5 - S 2p spectrum from a pyrite fracture surface determined
using XPS at a photon energy of 1487 eV and using SXPS at a
photon energy of 206 eV (Schaufuß et al, 1998a).
arise from x = 3λ. In practice, a thickness of 3λ is usually taken
as the thickness analysed, but as explained below, the relationship
between that thickness and the surface layer depth analysed
depends on the photoelectron take-off angle (θ).
Can the surface sensitivity of XPS be enhanced? The answer is
yes, but not easily. The analysis depth associated with any
particular photoelectrons (eg S 2p) is determined not only by their
kinetic energy (and hence by the exciting photon energy) but also
by how the specimen surface is oriented relative to the electron
detector. If the electron detector is aligned with the normal to the
specimen surface (an electron take-off angle of 90°), then the
detected electrons can escape through the surface layer of
maximum thickness, and so the analysis depth will be maximised
h h
e-
63% of
analysis
depth
FIG 6 - Schematic illustration of the effect of electron take-off angle (θ) on X-ray photoelectron spectroscopy (XPS) analysis depth.
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
(Figure 6). On the other hand, if the electron take-off angle is
very small, the detected electrons will have passed furthest
through the specimen for a given depth below the surface, and the
analysis depth will be minimised. However, only if the specimen
surface is a relatively smooth single particle one will lowering the
electron take-off angle lead to a marked enhancement of surface
sensitivity. In practice, the minimum electron take-off angle is
~15°, but even at that angle, the analysis depth is four times less
than for 90°. For fine particulate specimens, lowering the take-off
angle has a much smaller effect.
The other way to enhance the surface sensitivity of XPS is to
reduce the photon energy so that the photoelectron kinetic energy
is reduced. If there weren’t a need to also minimise the X-ray
source linewidth and complexity, reducing the photon energy
would simply be a matter of replacing the usual Al or Mg anode
in the X-ray source with an anode made of another suitable metal
and utilising an X-ray line other than Kα, such as the 192.3 eV
Mξ line from Mo, but most alternative sources would require the
development of a new monochromator (Krause, 1971). Rather
than go down that path, it is preferable to use a synchrotron as a
source of X-rays, as in addition to making SXPS possible, the
high flux of photons of tunable energy and aligned polarisation
allows angle-resolved NEXAFS spectroscopy to be carried out.
The application of these techniques to flotation-related sulfide
mineral systems is covered in sections that follow. Before doing
so, however, a brief comment on spatial resolution and imaging
XPS is pertinent. Some degree of spatial resolution in XPS can be
achieved either by using a small X-ray spot or by accepting
electrons from only a small area of the surface. Most modern
spectrometers can be operated in imaging mode, but that
capability is rarely used. Spatial resolution of better than 10 μm is
achievable in imaging mode, and areas of less than 150 μm in
diameter can be analysed in high energy resolution spectrometer
mode; however, for rough mineral surfaces, the minimum
analysis area would be closer to 500 μm in diameter.
Synchrotron X-ray photoelectron spectroscopy
(SXPS)
SXPS for surface characterisation requires an appropriate soft
X-ray beam-line and a suitable end-station. It is preferable for the
beam-line to cover as much of the soft X-ray energy range as
possible so that the S 2p and the metal core levels can be studied
both at maximum surface sensitivity and at much lower surface
sensitivity, to enable surface and subsurface information to be
obtained for the same specimen. The ideal situation is to be able
to assemble a non-destructive chemical depth profile for the S
e-
143
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
and metal(s) in the sulfide, and for any other key elements in
surface modifiers such as adsorbed collectors. In this regard it
should be noted that depth profiles obtained by ion beam
sputtering rarely provide accurate chemical information because
of alteration by the ion beam. An X-ray energy range of 100 to
1500 eV would be considered good for sulfide mineral studies,
and one covering 100 eV to 2500 eV (which also encompasses
the S K-edge near 2470 eV) would be considered ideal. In the
soft X-ray range, the photon energy is normally selected by
means of a grating monochromator. It is not unusual for more
than one grating to be required to span the full energy range, and
for the X-ray linewidth and flux to vary across the energy range
covered by each grating. Thus, photon energy tunability comes at
the price of resolution varying with energy, but the available flux
is usually sufficient to maximise resolution when necessary. The
X-ray ‘spot’ size on the specimen depends on the beam-line, but
it is rarely bigger than 3 × 0.5 mm and is typically 0.5 × 0.1 mm.
Because of the high flux, specimen charging can be a problem
with poorly-conducting materials (such as pure sphalerite), even
when a low energy electron flood gun for charge compensation is
employed. Also because of the high flux, the possibility for beam
damage to the specimen is greater than in conventional XPS, but
in practice, it is more likely that the high flux will be used to
improve resolution (by narrowing slits, for example) than to
reduce data acquisition time.
Synchrotron X-ray beams are horizontal, therefore an end-
station suitable for SXPS and NEXAFS spectroscopy is essentially
a high-quality XPS instrument with the electron energy analyser
oriented so that its lens is horizontal, with an optimally positioned
NEXAFS partial yield detector, with the specimen mounted on a
five-axis manipulator, and with the specimen handling facilities
necessary for investigating mineral fracture surfaces bearing
adsorbed collector species. The Australian Synchrotron Research
Program (ASRP) soft X-ray spectroscopy end-station, which was
designed to enable, inter alia, minerals processing research is
shown in Figure 7. Such an end-station allows liquid nitrogen
cooling in the entry lock, preparation chamber and analyser
chamber to retain any moderately high vapour pressure reaction
products on a specimen surface, has the entry lock within a glove
box that can be flushed with an inert gas, and has a double-blade
cleaver for the preparation of mineral fracture surfaces under UHV.
Specimen transfer between the entry lock and various chambers is
via a rotary distribution mechanism.
FIG 7 - The Australian Synchrotron Research Program (ASRP)
soft X-ray spectroscopy end-station.
The enhancement of surface sensitivity achieved by ensuring
that the photoelectrons of interest are ejected with a kinetic
144 Spectrum Series 16
energy of ~50 eV is illustrated by the S 2p spectrum for a pyrite
fracture surface obtained using photons of energy 206 eV
(Figure 5). At that photon energy, the surface species produced on
creation of the fracture surface under UHV are clearly evident.
In addition to the surface S22- species with a 2p3/2 binding energy
near 161.7 eV, S2- surface species near 161 eV arising from broken
disulfide bonds are also obvious. Interestingly, when a pyrite
fracture surface is exposed to oxygen, the S2- surface species react
rapidly to produce sulfate, whereas the concentration of the
S22- surface species remains approximately constant during the
subsequent increase in sulfate formation (Schaufuß et al, 1998b).
This behaviour was interpreted by Schaufuß et al (1998b) as being
consistent with oxidation in patches.
Another example of the enhanced surface sensitivity afforded
by SXPS is illustrated by the S 2p spectra determined at photon
energies ranging from 800 to 210 eV for the same Cu-activated
fracture surface of relatively pure sphalerite (Figure 8). It can
be seen that at a photon energy of 800 eV, the altered S
environments at the Cu-activated surface are barely evident,
whereas at 210 eV, the surface sensitivity is sufficient to reveal
that there are two high binding energy components. The
sulfur-rich composition of the outermost layers, which may arise
as a result of limited oxidation after the Cu-activation, would be
expected to impart hydrophobicity to the treated sphalerite.
(a) (b)
172 168 164 160 172 168 164 160
Binding Energy (eV) Binding Energy (eV)
(c) (d)
170 166 162 158 170 166 162
Binding Energy (eV) Binding Energy (eV)
FIG 8 - S 2p spectra for Cu-activated sphalerite determined at
photon energies of (a) 800; (b) 400 eV; (c) 260 eV and (d) 210 eV.
One of the most convincing demonstrations of the additional
information that can be obtained by progressively increasing the
photon energy from the value corresponding to maximum surface
sensitivity to obtain a non-destructive chemical depth profile has
been provided by Szargan et al (1997) for 2-mercaptobenzo-
thiazole (MBT) adsorbed on galena. The S 2p spectrum
determined at photon energies of 210 - 480 eV revealed (Figure 9)
that a chemisorbed monolayer of MBT was covered by the
multilayer species ‘(MBT)2’, or more correctly 2,2’-dithiobis
(benzothiazole). All four S atoms in the dithiolate have a 2p3/2
binding energy near 164 eV, similar to that for the endocyclic S in
chemisorbed MBT, whereas the value for the exocyclic S in
chemisorbed MBT is near 162 eV. Fortuitously, the S 2p doublet
for the galena substrate is at a binding energy sufficiently low to
not interfere too badly with the S 2p peaks from the collector.
Flotation Plant Optimisation

FIG 9 - S 2p spectra for 2-mercaptobenzothiazole (MBT)
adsorbed on galena determined at different photon energies
(Szargan et al, 1997).
Another advantage of the ability to select the photon energy is
the possibility to vary relative photoionisation cross-sections. For
example, in XPS at 1487 eV, the S 2p:Pb 4f photoionisation
cross-section ratio is 0.07:1 (Goldberg, Fadley and Kono, 1981), so
that the Pb 4f intensity is much greater than the S 2p (Figure 10).
This in itself would not be a problem, but because of the disparate
cross-sections, the background under the S 2p spectrum is
complicated by energy loss structure associated with the Pb
4f spectrum situated only 20 eV to lower binding energy. At a
photon energy of 200 eV, near maximum S 2p surface sensitivity,
the cross-section ratio is 4.4:1, so that now the S 2p spectrum is
much more intense than the Pb 4f and it can be difficult to extract
chemical information from the Pb 4f peaks. However, at a photon
energy of 250 eV, the cross-section ratio is close to 1:1 while both
S 2p and Pb 4f photoelectrons remain quite surface sensitive.
Of course a major advantage of the ability to tune the photon
energy is that X-ray absorption spectroscopy (XAS) becomes
feasible. For surface science and/or for the soft X-ray region, it
is the near-edge structure in the X-ray absorption spectrum that
is of interest.
Near-edge X-ray absorption fine structure (NEXAFS)
spectroscopy
NEXAFS spectroscopy (Stöhr, 1992), also known as X-ray
absorption near-edge structure (XANES) spectroscopy, can be a
powerful surface and near-surface characterisation technique in
its own right, but is even more powerful when it complements
XPS or SXPS data. The fine structure that is observed in XAS
can extend to several hundred eV above the absorption edge, but
it is the structure in the near-edge region, typically 20 eV above
the absorption edge, rather than the extended structure that is of
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
Pb 4f
S 2p
165 160 155 150 145 140 135
Binding Energy (eV)
FIG 10 - S 2p/Pb 4f spectrum from galena determined at a photon
energy of 1487 eV.
more interest for flotation-related systems because in the soft
X-ray range there will almost certainly be interference from other
absorption edges in the extended structure region.
Determination of an X-ray absorption spectrum in the soft
X-ray range is rarely effected directly in transmission mode
because it would require very thin specimens, and that
arrangement would certainly be inappropriate for sulfide
minerals. Instead, X-ray absorption is measured indirectly, by
monitoring the intensity of electrons or photons emitted by the
specimen as a result of the X-ray absorption and subsequent
de-excitation processes. A ‘partial yield detector’, consisting of
three grids in front of a multichannel plate, is used for this
purpose, and it is positioned in an analysis chamber so that the
electron energy analyser can be used at the same time.
Absorption can also be monitored by measuring the ‘drain’
current required to replace the electrons given off by the
specimen. Because both the penetration depth of the incident
X-rays and the escape depth of the fluorescent X-rays are long
(>100 nm), measuring absorption via the ‘fluorescence yield’ will
provide subsurface (essentially bulk) information. A partial yield
detector will measure the fluorescent X-rays given off by all
elements with core binding energies lower than the incident
photon energy; consequently a total fluorescence yield (TFY) is
obtained. If an energy-dispersive detector, such as a liquid-
nitrogen cooled ultra-low energy germanium diode, is used a true
fluorescence yield for the edge under investigation will be
obtained. Measuring absorption via the yield of electrons of
energy greater than about 50 eV can provide information that is at
least as surface sensitive as that obtained in XPS. This partial
electron yield (PEY) is dominated by Auger electrons, but the
Auger electrons can be singled out for detection by using the
electron energy analyser to obtain a true Auger electron yield
(AEY). The drain current is less surface sensitive because it must
replace all electrons given off, including the many electrons
that have energies below 30 eV and that have a long IMFP.
Accordingly, the drain current is effectively a total electron yield
(TEY), and provides near-surface as distinct from true surface
information. Although the TFY and TEY spectra do not provide
surface information, they should be obtained at the same time as
145
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
the PEY or AEY spectrum either to confirm that any differences
observed are indeed due to surface effects, or to contribute to any
non-destructive chemical depth profile partially assembled from
SXPS data.
At the same time as indirectly measuring X-ray absorption by
the specimen (I), the photon flux incident on the specimen (I0) is
monitored using a high transmission mesh, usually Au, Cu or Ni,
to allow correction for any absorption by the beam-line optical
components or any instability in the synchrotron itself. An X-ray
absorption spectrum shown is usually the variation of I/I0 with
photon energy, where I/I0 is referred to as the normalised
absorption intensity. The TEY and TFY Cu L2,3- edge absorption
spectra for air-exposed Cu metal and Cu sulfide minerals are
shown in Figure 11 as typical examples of NEXAFS spectra.
The leading peaks in an X-ray absorption spectrum involve
electronic transitions from a filled core level to unfilled bound
states of appropriate symmetry in the conduction band (above the
Fermi level). Dipole selection rules, viz Δl = ±1, normally apply
so that the leading peaks in the Cu L2,3-edge spectra shown in
Figure 11 would be expected to correspond to transitions from the
Cu 2p3/2 (L3) level to unfilled Cu d-states or s-states in the
conduction band. Transitions to states beyond the unfilled bound
states can also produce peaks because of interference between the
excited electron and the electrons of the neighbouring atoms.
These peaks constitute the fine structure that extends well above
the absorption edge, but as explained above it is the near-edge or
NEXAFS spectrum that is of most relevance to surface
characterisation of flotation-related systems. In Figure 11, the
peak about 20 eV above the first intense absorption peak is its
Cu L2 counterpart, ie transitions from the Cu 2p1/2 core level to
the same unfilled d- or s-states above the Fermi level.
A NEXAFS spectrum is more difficult to interpret than a
photoelectron spectrum because of the strong influence of the
final unfilled states as well as the initial core levels. Interpretation
is carried out either by comparison of spectra with those from
similar materials of known structure, or by simulation of
absorption spectra based on calculations of the densities of
unfilled states. Both approaches are useful, but it can be
dangerous to rely only on the former. For investigating modified
sulfide mineral surfaces, reliable interpretation of absorption
spectra usually requires ab initio calculations to be carried out
using codes such as FEFF8 or WIEN2k. Such calculations are
beyond the scope of this chapter, but any research provider
utilising NEXAFS data for flotation-related problem-solving
FIG 11 - Total electron yield (TEY) and total fluorescence yield (TFY) Cu L2,3-edge near-edge X-ray absorption fine structure (NEXAFS)
spectra from air-exposed Cu and Cu sulfides.
146 Spectrum Series 16
should use spectral simulation to support interpretation of the
experimental data. An interpretation of the spectra shown in
Figure 11 has been provided by Goh et al (2006b), but one brief
illustrative example will be given here. It can be seen from
Figure 11d that the leading Cu L3 peak for bornite is of lower
intensity relative to the higher energy peaks in the TEY spectrum
than in the TFY spectrum, whereas the spectra for chalcopyrite
show no comparable difference (Figure 11e). Could this be
because air-exposure of the bornite has led to rapid migration of
Fe to the surface where it has formed an over-layer of
Fe(III)-oxygen species as indicated by XPS studies (eg Buckley
and Woods, 1983) while significantly less oxidation of the
chalcopyrite surface has occurred? There is little alternative here
other than to simulate spectra from a bornite lattice that is
Fe-deficient, and to compare the simulated and experimental
spectra. When this is carried out for a bornite structure in which
both Fe sites in the unit cell are filled, in which one is vacant and
in which both are empty, it is found that the relative intensity of
the leading Cu L3 peak does indeed decrease with increasing
Fe-deficiency (Figure 12).
In Figure 11, it can be seen that a small initial peak near
931.3 eV is evident in the TEY but not the TFY spectra for Cu
metal and chalcocite. This small peak arises from a low
Absorption
(c)
(b)
(a)
930 940 950 960 970 980 990
Photon Energy (eV)
FIG 12 - Simulated Cu L3 near-edge X-ray absorption fine structure
(NEXAFS) spectra for a bornite lattice with Fe sites (a) all vacant;
(b) half vacant; (c) all filled.
Flotation Plant Optimisation

concentration of Cu(II)-oxygen species at the surface of the two
materials. A similar peak is not evident for air-exposed bornite,
since notwithstanding its greater reactivity, the oxidation products
include Fe-oxygen rather than Cu-oxygen species. It can also be
seen from Figure 11 that the leading Cu L3 absorption peak for
the sulfides is in the range 932.4 - 932.7 eV and the corresponding
peak for Cu metal is at 933.7 eV. Thus for the Cu L2,3-edge, the
leading absorption peak energy does not increase with oxidation
state as might be expected from the Cu 2p binding energies. The
reason is that Cu(II) is nominally a d9 system, whereas the 3d
shell is nominally filled for Cu(0) and Cu(I); consequently there
will be lower energy unfilled states available for the Cu(II) case.
Another point that should be emphasised is that unlike the
situation for XPS, the relative intensity of absorption peaks is not
necessarily proportional to the concentration of the species giving
rise to the peaks. The relative intensity of the Cu(I) and Cu(II)
peaks is a prime example, the latter being 25 times that of the
former for the same concentration (Pattrick et al, 1997), again
because of the greater availability of unfilled d-states for Cu(II)
than for Cu(I).
It was noted in the XPS section above that synchrotron
radiation is polarised, and that this feature can be exploited in
NEXAFS spectroscopy. Angle-resolved NEXAFS spectroscopy
can provide additional surface chemical information when a
specimen surface is relatively flat, and when the molecules in the
surface layer are aligned or oriented relative to the specimen
itself. Thus in practice, this approach is only relevant for single
particle specimens. A description of the application of
angle-resolved NEXAFS spectroscopy to a collector adsorbed on
a sulfide surface has been given by Mayer et al (2003).
Essentially, absorption will be greatest when the plane of
polarisation of the X-ray beam is parallel to the relevant orbitals
of the element under investigation. For adsorbed collector
molecules that are preferentially aligned, it may be necessary to
vary the polar and azimuthal angles of the specimen to achieve
maximum absorption for any particular bond in the adsorbate.
Further examples of angle-resolved NEXAFS will be given in the
section on multi-technique characterisation.
Static secondary ion mass spectrometry (static SIMS)
Conventional XPS is the surface chemical characterisation
technique most likely to be found in a surface science laboratory,
and if there is a second instrument in that laboratory to augment
the XPS equipment, it is likely to be a secondary ion mass
spectrometer. In secondary ion mass spectrometry, a primary
beam of positive ions incident on the surface of a specimen ejects
positive ions, negative ions, electrons and neutral species, and the
mass-to-charge ratio (m/z) of those positive and negative
secondary ions is determined (Briggs and Seah, 1990b;
Benninghoven, 1994). The primary beam can consist of inert gas
ions, or more typically ions such as 69Ga+ ions from a liquid 69Ga
source. These ions can be quite energetic, 25 keV being not
unusual, and SIMS is an inherently destructive technique. SIMS
can be static or dynamic. In dynamic SIMS, the primary ion dose
on the specimen is high, and the surface and subsurface layers of
the specimen are progressively sputtered or etched to provide
steady-state elemental composition information as a function of
depth below the surface. The information is not chemical, as the
ion beam can alter the chemical species in the material
underlying the layers removed by sputtering. In static SIMS, the
total ion dose on a specimen area analysed is limited to
1013 ions.cm-2, so that statistically, no primary ion hits the same
part of the surface twice. Thus, static SIMS is no more gentle
than dynamic SIMS; the fundamental difference is that only the
outermost surface layer that has not been damaged by the ion
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
beam is analysed in static SIMS, and hence static SIMS is highly
surface sensitive. Typically, a primary ion beam of diameter
~0.2 μm would be rastered over a 150 × 150 μm area of a
specimen surface. In static SIMS, the secondary ion m/z is usually
determined by means of a time-of-flight mass analyser, in which
case the technique is called ToF-SIMS. Not surprisingly, then, the
technique is a UHV-based one, but surprisingly, not all
spectrometers have the facility to cool the specimen before and
during analysis. In ToF-SIMS, the primary ion source is operated
in pulsed mode, and a typical repetition rate of 5 kHz gives a
mass spectral range to m/z 1440 with resolution of at least 0.05.
There are several general comments that should be made about
static SIMS. The first is that the primary ion beam usually ejects
more neutral species than ions; consequently there is no
necessary relationship between the abundance of an ion peak at a
particular m/z and the surface concentration of the species giving
rise to that peak. Some ions are more stable than others, and for a
given specimen, the intensity of peaks from more stable ions are
expected to be greater than the intensity of peaks from less stable
ions. In particular, some species are more stable as positive ions
than negative, or vice versa, therefore it is almost always better to
determine both positive and negative secondary ion mass spectra.
For an organic adsorbate such as a collector, it is possible that
only collector fragment ions will be detected rather than the
parent ion (of mass equal to the complete collector radical). In
practice, and with good reason, it is assumed that negligible
interaction of ions occurs between ejection and detection, so that
if an ion of a species is detected, then that species is most
probably present at the specimen surface as a discrete entity or as
a fragment of a larger entity. The exception to this assumption is
the loss or gain of hydrogen. Hydrogen is often a common
constituent of the vacuum in the vicinity of a material under
primary ion bombardment, and it is not unusual to observe both
positive and negative ions that have an m/z one or two atomic
mass units (amu or Da) greater than expected. These observations
are not artefacts of an incorrectly calibrated m/z scale, but it is
important that the m/z scale is calibrated across as much of the
range as possible with ions of unambiguous m/z.
The unambiguous identification of some ions can depend on
the m/z resolution; however, for most of the ions encountered in
systems related to sulfide mineral flotation, an m/z resolution of
0.05 is usually sufficient. The identification of metal-containing
ions is often facilitated by the diagnostic stable isotopic ratio
that is reflected in the cracking pattern. For example, as can be
seen in Figure 13, Cu-containing ions give rise to doublets with
components of intensity ratio approximately 2:1 separated by
2 amu reflecting the 63Cu to 65Cu ratio of 69:31, and Ag-
containing ions produce doublets with components of
approximately equal intensity separated by 2 amu reflecting the
107
Ag to 109Ag ratio of 52:48. Ions containing Pb give rise to
triplets of intensity ratio ~1:1:2 separated by 1 amu because the
206
Pb to 207Pb to 208Pb ratio is 24:23:52. The natural abundance of
the 204Pb isotope is only 1.48 per cent. Identification of a peak at
m/z 28, on the other hand, would require a mass resolution better
than 0.05, as CO would be at m/z 28.01 and C2H4 would be at
m/z 28.05.
Secondary ion mass spectra can be tedious to interpret, but they
contain valuable chemical bonding information, and some of the
findings from ToF-SIMS data for sulfide mineral/collector
systems have been surprising. For example, it is evident from
Figure 13 that Cu(DTP)2– ions were detected for multilayer
diethyl dithiophosphate (DTP) adsorbed on chalcocite, but not for
monolayer coverage. This observation suggested that the
multilayer species might either be, or contain, Cu(DTP)2, which
was not expected as XPS data had indicated that the multilayer
147
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
FIG 13 - Negative secondary ion mass spectra for monolayer (upper spectrum) and multilayer (lower spectrum) diethyl dithiophosphate
(DTP) adsorbed on (a) chalcocite and (b) silver sulfide.
species was CuDTP or (CuDTP)n. The data from both techniques
could only be rationalised if the Cu(DTP)2 ions arose from
contiguous CuDTP molecules adsorbed at the surface, or if the
multilayer was predominantly CuDTP but Cu(DTP)2 was a minor
component. As noted above, the Cu L3-edge NEXAFS spectrum
is particularly sensitive to the presence of Cu(II), and a small
Cu(II) peak was discernible in the Cu L3-edge NEXAFS
spectrum for multilayer DTP on chalcocite, an observation
consistent with the presence of a small concentration of
Cu(DTP)2 (Goh et al, 2006c). The situation for DTP adsorbed on
silver sulfide was similar, and fortuitously because of the
presence of the low concentration of Cu or Ag dithiolate in the
multilayer, it was concluded that differentiation of monolayer and
multilayer coverage by ToF-SIMS alone would be expected for
all thiol collector/sulfide systems.
Because of the relatively small primary ion beam spot size,
ToF-SIMS imaging is feasible to determine the distribution of a
particular species across the surface of particles from a flotation
pulp sampled as described near the beginning of this chapter. In a
typical application, Hart, Biesinger and Smart (2006) discuss 256
× 256 pixel secondary ion images covering 100 × 100 μm areas of
a slurry sample pressed into indium foil, and the statistical
methods used to identify the different mineral phases present.
Scanning tunneling microscopy (STM) and atomic
force microscopy (AFM)
These are considered to be techniques that are useful in a research
context, but not widely applicable for identifying and solving
flotation plant problems. Examples of their application in sulfide
mineral flotation research have been described by Smart et al
(2003).
148 Spectrum Series 16
Infrared (IR) reflection spectroscopy
Although the penetration depth of IR radiation is of the order of
the wavelength, ie ~1 μm, effective submonolayer sensitivity for
IR spectroscopy may be possible for adsorbates such as collectors
on mineral substrates when no more than submonolayers are
adsorbed. The two main IR reflection techniques are the
well-known attenuated total reflectance (ATR) and the less
well-known but potentially more powerful external reflection
spectroscopy (ERS), also known as reflection-absorption
spectroscopy. Application of ERS for investigating the interaction
of collectors with sulfide mineral surfaces has been described in
detail by Mielczarski and co-workers (Mielczarski, Mielczarski
and Cases, 1996; Mielczarski and Mielczarski, 2005) and the
underlying theory for non-metallic substrates has been treated by
Buontempo and Rice (1993). The spectra can be obtained using
an FTIR spectrometer equipped with a polariser before the
sample to provide a p- or s-polarised beam, and a mercury
cadmium telluride (or similar) detector. The reflected intensity
from the mineral/adsorbate interface as a function of wavelength
is usually normalised by the reflected intensity from the
adsorbate-free surface.
There appear to be two disadvantages of ERS for sulfide
flotation problem-solving rather than research-only applications.
The first disadvantage is that mineral slab samples (typically 20 ×
15 × 10 mm) with one face highly polished appear to be required.
Arguably, the surface of a sulfide mineral that has been highly
polished, ultimately with a fine abrasive such as 0.03 μm alumina,
would be significantly altered. The second disadvantage, which
could be considered an advantage in a research context, is the
complexity of the spectra. An external reflection spectrum
depends on the polarisation and angle of incidence of the IR
beam, the thickness and orientation of the adsorbed layer, and the
Flotation Plant Optimisation
 CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION

optical properties of the sulfide mineral substrate and adsorbate.

Of course in that complexity lies the valuable information, but S 2p
correct interpretation of the experimental data requires the
calculation of simulated spectra for comparison. For each
specimen, spectra determined at three different incident angles
and two polarisations are typically required for a self-consistent
interpretation.
Monolayer
Surface enhanced Raman scattering (SERS)
spectroscopy Untreated
Like STM and AFM, Raman spectroscopy and SERS
spectroscopy are ‘niche’ techniques that are more useful in a
research context than for flotation plant problem-solving. Raman
spectra are complementary to IR spectra, in that transitions
between vibrational energy levels are involved in both, but
different selection rules apply. FT-Raman spectra are usually
determined with a laser source (eg of wavelength 1064 nm) and a Multilayer
cooled solid-state detector. The spectroscopy is rendered surface
sensitive by suitably ‘roughening’ the surface of a metal substrate.
FT-SERS is now a well-established technique for the examination
of adsorbed layers, including self-assembled monolayers, at 168 167 166 165 164 163 162 161 160
electrochemically roughened Ag, Au and Cu surfaces, and it was Binding Energy (eV)
used in conjunction with XPS in a flotation-related context to
verify that xanthate chemisorbed on Ag without decomposition FIG 14 - S 2p spectra for 2-mercaptobenzothiazole (MBT)
(Buckley et al, 1997). By contrast, the applicability of SERS adsorbed on chalcocite.
spectroscopy for investigating alteration of, and adsorption on,
sulfide mineral surfaces has only recently been demonstrated.
Cu 2p Cu(LMM)
Promising results have been reported by Hope and co-workers on
evaporative ‘decoration’ with gold of an MBT layer adsorbed on
chalcopyrite to induce SERS (Hope et al, 2006), and also on gold
decoration of chalcopyrite prior to interaction with MBT
(Buckley et al, 2006). In the latter case, chemisorption of MBT
Untreated
on the Au decoration was found to be not a problem, as Untreated
adsorption on the mineral sulfide occurs at a potential lower than
for the Au. Monolayer

Multi-technique characterisation Monolayer

While some of the ex situ techniques described above are

Multilayer
individually quite powerful, for many problem-solving
applications, several techniques need to be applied to obtain Multilayer

sufficient information, especially when adsorbed flotation

collectors are involved. As an example of the complementary 960 955 950 945 940 935 930 905 910 915 920 925

information obtained from the application of XPS, SXPS, Binding Energy (eV) Kinetic Energy (eV)

NEXAFS spectroscopy, ToF-SIMS and SERS spectroscopy, the

adsorption of MBT on abraded surfaces of single-piece chalcocite
will be considered (Buckley et al, 2006; Goh et al, 2006a).
As is usually the case, conventional XPS alone provides a large
part of the information needed. No more than monolayer
coverage of chemisorbed MBT is evident from the S 2p spectrum
(Figure 14) because of a low intensity 2p3/2 component near
164 eV arising from the endocyclic S atoms in MBT, together
with Cu LMM Auger and Cu 2p spectra (Figure 15) that are
indistinguishable from those for a fresh chalcocite surface
(Buckley and Woods, 1993; 1995). Determining the precise S 2p
binding energy for the exocyclic S atoms bonded to the Cu atoms
in a chalcocite surface is complicated by the proximity of the
substrate S 2p3/2,1/2 peaks at ~161.6 and 162.8 eV. The onset of
multilayer adsorption is revealed by an additional component
near 915 eV in the Cu LMM spectrum but without a resolved
Cu 2p component (Figure 15). The S 2p spectra from substantial
multilayers of adsorbed MBT indicate that the predominant
multilayer species is the metal thiolate rather than (MBT)2. The
S 2p components near 162.4 and 163.85 eV are of approximately
equal intensity, and more intense than the component from the
chalcocite substrate. The S 2p:Cu 2p intensity ratio is consistent
FIG 15 - Cu 2p and Cu LMM spectra for 2-mercaptobenzothiazole
(MBT) adsorbed on chalcocite.
with the multilayer S:Cu ratio being no greater than 2:1, and
hence consistent with the multilayer species being predominantly
CuMBT rather than Cu(MBT)2. The N 1s spectrum for
monolayer coverage consists of a single peak at 399.0 eV. The
spectrum for multilayer coverage is quite similar; in this case a
single peak at 399.1 eV. The N 1s binding energy of ~399 eV is
very close to the value (398.9 eV) determined for the PbS/MBT
system by Szargan et al (1997), and suggests a similar role for the
N in MBT in the interaction between the collector and the
relevant metal atom.
In SXPS, S 2p spectra from chemisorbed and multilayer MBT
on abraded chalcocite surfaces obtained at photon energies of 205,
250 and 300 eV (Figure 16) clearly reveal the adsorbed collector.
At photon energies near 200 eV, corresponding to S 2p optimum
surface sensitivity, the doublets arising from the MBT S atoms are
at their greatest intensity relative to the doublet from S in the
chalcocite, and the components fitted at these energies provide the
most reliable binding energy data for chemisorbed MBT. The

Flotation Plant Optimisation Spectrum Series 16 149

CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION

* * o
15
o
90

Absorption
Monolayer

Multilayer

395 400 405 410 415 420

Photon Energy (eV)

FIG 17 - Angle-dependent N K-edge near-edge X-ray absorption

fine structure (NEXAFS) spectra for 2-mercaptobenzothiazole
FIG 16 - Synchrotron S 2p spectra for 2-mercaptobenzothiazole
(MBT) adsorbed on chalcocite.
(MBT) adsorbed on chalcocite at (a) monolayer and (b) multilayer
coverage.
σ* peaks are of significantly lower intensity relative to the π*
continuum structure for a 15° angle of incidence than for 90°,
intensity ratio of the two components should indicate whether the
with the difference greater for the lower energy (=N-) π* peak.
MBT is adsorbed parallel or inclined to the substrate surface; in the
This behaviour is similar to that observed for MBT adsorbed on
former case, the intensities should be equal, whereas in the latter
CdS, and is indicative of a tilted upright orientation (Mayer et al,
case, the intensity of the component from the endocyclic S, which 2003). It also indicates that the surface prepared by gentle
would be near to the collector/air interface, should be slightly abrasion was indeed sufficiently flat to result in a significant
greater than that of the exocyclic S, which would be at the ‘average’ alignment. The corresponding C K-edge spectra are
solid/collector interface. However, such information is difficult to also consistent with an aligned chemisorbed monolayer and a
extract for abraded surfaces without also determining the S 2p randomly oriented multilayer (Goh et al, 2006a).
spectrum for the untreated chalcocite at each photon energy, as the
The Cu L2,3-edge spectrum for multilayer MBT on chalcocite
corresponding abraded chalcocite spectrum cannot be fitted
and the spectrum for a fresh chalcocite surface untreated with
adequately without the inclusion of a high binding energy
MBT are shown in Figure 18. The Cu L3 peak near 931 eV in the
component to account for the surface layer altered by the abrasion.
spectrum for the multilayer is indicative of the presence of some
As the photon energy is increased, the analysis depth increases, so
Cu(II) in the material adsorbed on the surface of the chalcocite.
that any intensity difference for the two S components arising from
inclination of the chemisorbed MBT should decrease. Again,
extraction of alignment information from spectra obtained at
different photon energies is hindered by the high binding energy
component.
Nitrogen K-edge NEXAFS spectra for MBT adsorbed on Multilayer
chalcocite, determined with the incident photon beam normal and
at 15° to the substrate surface, are shown in Figure 17. Each
Absorption

spectrum exhibits two relatively narrow absorption peaks near 399

and 401 eV as well as a broad peak near 407 - 408 eV. On the basis
of the considerable groundwork carried out by Szargan and Untreated
co-workers (eg Szargan et al, 1997; Mayer et al, 2003), the peaks
observed near 399 and 401 eV can be assigned to N 1s transitions
to (=N-) and (=Nδ+≈) 2pπ* states, respectively, and the peak
near 407 - 408 eV to transitions to 2pσ* states (where * denotes
antibonding). The absorption energies are similar to those reported
for MBT chemisorbed on CdS and for Cd(MBT)2 (Mayer et al,
2003). The observations are consistent with some interaction
between the N and metal atoms for both the monolayer and
multilayer in each system.
It can be seen from Figure 17 that there is no difference in the 930 935 940 945 950
N K-edge spectra for multilayer MBT on chalcocite determined Photon Energy (eV)
with the mineral surface at a polar angle of 90° or 15° to the
incident photon beam. This is as expected, since no preferred FIG 18 - Cu L2,3-edge NEXAFS spectrum for multilayer
alignment of the copper thiolate relative to the chalcocite 2-mercaptobenzothiazole (MBT) adsorbed on chalcocite
substrate is expected. By contrast, for monolayer coverage, both and the chalcocite surface prior to treatment.

150 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION

The presence of some Cu(II) in the multilayer is consistent with on-stream, although a few of the in situ techniques could be
the detection of some Cu(MBT)2 by ToF-SIMS, and a similar adapted for on-stream application. Also in this context, the
situation to that already described for the corresponding surface chemical composition must be inferred from some of the
DTP/chalcocite system. in situ measurements somewhat more indirectly than for most of
In the negative secondary ion mass spectrum from a chalcocite the ex situ techniques. The exceptions are the molecular
surface bearing no more than a monolayer of MBT (Figure 19), specificity of IR and Raman spectroscopies for organic
MBT– ions were evident but CuMBT– ions were not. Copper- adsorbates. In general, the in situ techniques are used less
containing CuMBT fragment ions were observed at low intensity, frequently than the ex situ techniques for flotation plant
however. For multilayer coverage, Cu(MBT)2– ions were evident, problem-solving, and hence are described here in less detail.
but (MBT)2– ions were not observed. Cu(MBT)2 fragments were
evident at m/z 255 and 257 as well as at 287 and 289 in multilayer Electrochemical techniques
but not monolayer spectra. Low intensity peaks arising from
The application of electrochemical techniques in flotation research
Cu(MBT)2H– and Cu(MBT)2H2– ions were also observed in the
has been described by numerous authors, including Chander and
spectra from multilayers. In the positive ion spectra, for both the
Gebhardt (1989). The most widely used electrochemical technique
monolayer and multilayer, CuMBT+, Cu(MBT)H+ and [CuMBT
is voltammetry, especially when rotating ring disc electrode and
fragment]+ ions were detected, although the peaks were relatively photoelectrochemical measurements are included, but less
weak for the monolayer. Weak peaks from [Cu(MBT)2 common transient response techniques such as AC impedance
fragment]+ ions were observed only for the multilayer. Some spectroscopy and chronoamperometry can be informative for some
[CuMBT fragment]+ ions contained S and N, but others, systems. The application of linear potential sweep (uni-directional)
principally CuNCH2+, contained N without S, suggesting that at or cyclic (bi-directional) voltammetry for investigating the
least a proportion of the interaction of MBT with chalcocite oxidation of sulfide minerals, and the interaction of sulfide
involved both the N and exocyclic S. However, there was no minerals with collectors and other surface modifiers such as
convincing evidence that this interaction was with adjacent Cu hydrosulfide ions, has been described by Woods in the section on
atoms, as Cu2(MBT) parent ions were not detected, and fragment Electrochemical Investigative Techniques in Chapter 7 and
ions that must have arisen from Cu2(MBT) were of extremely low elsewhere in more detail (Woods, 1996). In cyclic voltammetry, the
abundance. On the other hand, the absence of these ions does not variation of the current that flows as the potential applied to the
necessarily mean that interaction was not with adjacent Cu atoms. mineral electrode is varied in a cyclical scan, reveals the potential
Hope and co-workers (Buckley et al, 2006) have investigated at which a reaction (eg chemisorption) occurs at the electrode, and
the chalcocite/MBT system by means of SERS spectroscopy after the extent of that reaction, but it does not unambiguously establish
Au decoration of the mineral surface. The band arising from the the identity of the reaction products. The reaction products are
NCS ring stretch vibration of MBT was observed at a similar deduced from the potential at which the reaction occurs, the effect
Raman shift for MBT chemisorbed on a Cu metal electrode, and of pH on that potential, the inferred reactions of the products on the
downshifted from the corresponding position for CuMBT. The reverse scan, and, if a rotating ring disc electrode arrangement is in
SERS spectra confirmed that MBT chemisorbed on chalcocite is use, the effect of rotating the electrode to disperse soluble products.
bonded to Cu atoms in the mineral surface through exocyclic S One weakness in this procedure is that inferred reactions are
usually based on bulk rather than surface reaction potentials. Thus
atoms.
voltammetry is most powerful when used as an in situ adjunct to an
ex situ technique, such as XPS, that does identify reaction products
The principal in situ techniques more directly.
In this context, in situ is understood to mean while still wet with It should also be noted that in cyclic voltammetry, electrode
the flotation-related aqueous phase, and not necessarily reactions are driven at sweep rates of the order of 10 mV.s-1,

-
CuS
-
CuNC
- No Cu 2 (MBT) fragments No Cu(MBT)2-
-
CuSCN
0.20
0.4 -
0.3
CuSC 0.15

0.2 0.10

0.1 0.05

0.0 0.00
90 95 100 105 110 115 120 125 250 260 270 280 290 300
Peak intensity / MBT – intensity

(MBT Ð S)
- MBT-

10
-
9
S- Cu(MBT)2 fragments Cu(MBT)2 -
8 0.12
7 0.10
6 0.08
- 0.06
5 (MBT – C6H 4S) 0.04
4 0.02
3 0.00
2 250 260 270 280 290 300

1
0
20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200

FIG 19 - Negative secondary ion mass spectra for monolayer (upper spectra) and multilayer (lower spectra) 2-mercaptobenzothiazole
(MBT) on chalcocite.

Flotation Plant Optimisation Spectrum Series 16 151

CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
therefore it is conceivable that a reaction that takes place during a
potential scan may be different from the reaction that takes place
spontaneously but slowly at open circuit. Also, it is possible that a
reaction that takes place under the steady-state conditions of
multiple scan cyclic voltammetry might be different from the
corresponding reaction that occurs during the first potential scan.
Indeed, when the oxidation behaviour of pyrite and pyrrhotite
fracture surfaces was investigated by means of cyclic voltammetry,
it was found that, commencing at the rest potential, the first and
second positive-going scans for pyrite were similar, whereas they
were different for pyrrhotite (Buckley, Hamilton and Woods,
1988). However, additional information concerning the formation
of a metal-deficient lattice during initial oxidation was obtained
from interpretation of those observations.
Voltammetric investigations of the interaction of sulfide
minerals with flotation collectors have revealed a prewave on the
positive-going scan at a potential below the reversible value for
the formation of the metal thiolate for numerous sulfide/collector
systems. In each case, the charge associated with the prewave
corresponds to that expected for the charge-transfer adsorption of
a monolayer of the collector (Buckley, Hope and Woods, 2003).
Infrared spectroscopy
Water is a strong absorber of IR radiation, therefore the in situ
application of IR spectroscopy is not straightforward.
Nevertheless, mineral sulfide/collector systems have been
investigated by ERS under potential control in specially designed
cells (eg Bozkurt et al, 1996) that enable the mineral electrode to
be positioned close to (within 100 μm of) the cell window during
the recording of single reflection spectra. Calcium fluoride is one
cell window material that has been used. However, in most of this
work, as discussed in the consideration of ex situ techniques, the
requirements of a moderately large area (~10 × 10 mm) and
highly polished mineral specimen surface are a disadvantage.
As an alternative to ERS, internal reflection spectroscopy, or
ATR, has been used with a finely ground mineral in carbon paste
working-electrode pressed against a Ge ATR crystal (eg
Leppinen, Basilio and Yoon, 1989). These measurements have
been extremely informative in revealing the identity of the
adsorbed species for particular Eh/pH conditions, and it is
conceivable that they could be of use in flotation plant
problem-solving applications.
Raman/SERS spectroscopy
Raman scattering spectroscopy has some advantages over IR
spectroscopy in that water is a weak (but not negligible) Raman
scatterer, and glass cells can be used. In situ SERS spectroscopic
investigations of the interaction of flotation collectors with
electrochemically-roughened Au, Ag and Cu metal surfaces have
been carried out under potential control (Hope et al, 2006). Of
more relevance for investigating flotation plant problems, the
demonstration that Au decoration of a chalcopyrite surface to
impart a SERS response could be carried out prior to interaction
with MBT solution (Buckley et al, 2006) has raised the
possibility of determining in situ SERS spectra of other sulfide
mineral/collector systems.
Soft X-ray absorption spectroscopy
In situ soft X-ray absorption spectroscopy with specimens under
potential control behind thin silicon nitride windows is feasible
(except at the N K-edge), but measurement of absorption would
most probably have to be via fluorescence yield, thereby making
the technique a near-surface one at best. Nevertheless, it is
expected that progress will be made in this area in the future.
152 Spectrum Series 16
CONCLUDING REMARKS
The reader may have gained the impression that short shrift was
given to some surface chemical characterisation techniques, and
excessive emphasis given to others. In one sense this is true, but it
must be kept in mind that this chapter addressed those surface
chemical characterisation techniques of most use for identifying
and solving problems within base metal sulfide flotation plants. It
is not through any bias of the author, who has very little
involvement in the day-to-day activities of a surface analysis
facility, that the ex situ techniques of XPS and to a lesser extent
ToF-SIMS have become the almost universal starting point for
surface chemical problem-solving. These spectroscopies have a
successful track record, and accordingly, the emphasis in this
chapter has been on them. Might the attention given to synchrotron
techniques have been the result of personal bias? Perhaps, but the
synchrotron-based spectroscopies of SXPS and XAS deliver on
information needed to complement that from XPS and static SIMS,
especially the information from SXPS, which has the surface
sensitivity essential for the identification of some flotation
problems. The application of SXPS and XAS can only increase as
more synchrotrons become available in countries with vibrant
minerals processing industries.
ACKNOWLEDGEMENTS
Access to synchrotron beam time was made possible by support
from the Australian Synchrotron Research Program, which is
funded by the Commonwealth of Australia under the Major
National Research Facilities Program. Use of the Taiwan Light
Source, and assistance by soft X-ray beam-line staff Yaw-wen
Yang and Liang-Jen Fan at the National Synchrotron Radiation
Research Center in Hsinchu, are gratefully acknowledged. The
author is also grateful for access to The University of New South
Wales (UNSW) Surface Science and Technology infrastructure
established by Rob Lamb, for stimulating discussions with Ron
Woods, to Prof Rüdiger Szargan for Figures 5 and 9, and for the
opportunity to carry out research with Siew Wei Goh while she
was a UNSW graduate student. Most of the spectra shown in this
chapter were determined by Siew Wei.
REFERENCES
Benninghoven, A, 1994. Chemical analysis of inorganic and organic
surfaces and thin films by static time-of-flight secondary ion mass
spectrometry (TOF-SIMS), Angewandte Chemie International
Edition England, 33:1023-1043.
Bozkurt, V, Brienne, S H R, Xu, Z, Rao, S R, Butler, I S and Finch, J A,
1996. Development of in situ external reflection spectroscopy
technique for adsorbed films, Minerals Engineering, 9:351-355.
Briggs, D and Seah, M P, 1990a. Practical surface analysis, Auger and
X-ray Photoelectron Spectroscopy, second edition, volume 1, 657 p
(Wiley: Chichester).
Briggs, D and Seah, M P, 1990b. Practical Surface Analysis, Ion and
Neutral Spectroscopy, second edition, volume 2, 756 p (Wiley:
Chichester).
Buckley, A N, 1994. A survey of the application of X-ray photoelectron
spectroscopy to flotation research, Colloids and Surfaces A,
93:159-172.
Buckley, A N, Hamilton, I C and Woods, R, 1988. Studies of the surface
oxidation of pyrite and pyrrhotite using X-ray photoelectron
spectroscopy and linear potential sweep voltammetry, in Proceedings
International Symposium Electrochemistry in Mineral and Metal
Processing II (eds: P E Richardson and R Woods), pp 234-246
(Electrochemical Society: Pennington).
Buckley, A N, Hope, G A and Woods, R, 2003. Metals from sulfide
minerals: The role of adsorption of organic reagents in processing
technologies, in Solid–Liquid Interfaces, Topics in Applied Physics
(eds: K Wandelt and S Thurgate), 85:61-96 (Springer: Berlin).
Flotation Plant Optimisation

Buckley, A N, Laajalehto, K, Suoninen, E and Woods, R, 1995.
Flotation-related surface reactivity of molybdenite, in Processing of
Hydrophobic Minerals and Fine Coal (eds: J S Laskowski and G W
Poling), pp 29-42 (Canadian Institute of Mining, Metallurgy and
Petroleum: Montréal).
Buckley, A N, Parks, T J, Vassallo, A M and Woods, R, 1997.
Verification by surface-enhanced Raman spectroscopy of the
integrity of xanthate chemisorbed on silver, International Journal of
Mineral Processing, 51:303-313.
Buckley, A N and Walker, G W, 1988. Sulfur enrichment at sulfide
mineral surfaces, in Proceedings XVI International Mineral
Processing Congress (ed: E Forssberg), pp 589-599 (Elsevier:
Amsterdam).
Buckley, A N and Woods, R, 1983. An X-ray photoelectron spectroscopic
investigation of the tarnishing of bornite, Australian Journal of
Chemistry, 36:1793-1804.
Buckley, A N and Woods, R, 1984. An X-ray photoelectron spectroscopic
study of the oxidation of galena, Applied Surface Science,
17:401-414.
Buckley, A N and Woods, R, 1993. Underpotential deposition of
dithiophosphate on chalcocite, Journal of Electroanalytical
Chemistry, 357:387-405.
Buckley, A N and Woods, R, 1995. Identifying chemisorption in the
interaction of thiol collectors with sulfide minerals by XPS:
Adsorption of xanthate on silver and silver sulfide, Colloids and
Surfaces A, 104:295-305.
Buckley, A N, Wouterlood, H J, Cartwright, P S and Gillard, R D, 1988.
Core electron binding energies of platinum and rhodium
polysulfides, Inorganica Chimica Acta, 143:77-80.
Buckley, F M, Hope, G A, Watling, K M and Woods, R, 2006. Use of an
evaporated gold SERS substrate to detect 2-mercaptobenzothiazole
on sulfide mineral surfaces, ECS Transactions, 2:49-60 (The
Electrochemical Society: Los Angeles).
Buontempo, J T and Rice, S A, 1993. Infrared external reflection
spectroscopy of adsorbates on dielectric substrates: Determining
adsorbate orientation in Langmuir monolayers, The Journal of
Physical Chemistry, 98:5825-5834.
Chander, S and Gebhardt, J E, 1989. Application of electrochemical
techniques in sulfide mineral flotation, in Challenges in Mineral
Processing (eds: K V S Sastry and M C Fuerstenau), pp 90-108
(Society of Mining Engineers: Littleton).
Chaney, R L, 1987. Recent developments in spatially resolved ESCA,
Surface Interface Analysis, 10:36-47.
Doniach, S and Sunjic, M, 1970. Many-electron singularity in X-ray
photoemission and X-ray line spectra from metals, Journal of
Physics C: Solid State Physics, 3:285-291.
Finkelstein, N P, 1997. The activation of sulphide minerals for flotation:
A review, International Journal of Mineral Processing, 52:81-120.
Goh, S W, Buckley, A N, Lamb, R N, Fan, L-J and Yang, Y-W, 2006a.
XPS, static SIMS and NEXAFS spectroscopic investigation of thiol
adsorption on metals and metal sulfides, ECS Transactions,
2:107-119 (The Electrochemical Society: Los Angeles).
Goh, S W, Buckley, A N, Lamb, R N, Rosenberg, R A and Moran, D,
2006b. The oxidation states of copper and iron in mineral sulfides,
and the oxides formed on initial exposure of chalcopyrite and bornite
to air, Geochima et Cosmochim Acta, 70:2210-2228.
Goh, S W, Buckley, A N, Lamb, R N and Woods, R, 2006c. The ability of
static secondary ion mass spectrometry to discriminate submonolayer
from multilayer adsorption of thiol collectors, Minerals Engineering,
19:571-581.
Goldberg, S M, Fadley, C S and Kono, S, 1981. Photoionization
cross-sections for atomic orbitals with random and fixed spatial
orientation, Journal of Electron Spectroscopy, 21:285-363.
Hart, B, Biesinger, M and Smart, R StC, 2006. Improved statistical
methods applied to surface chemistry in minerals flotation, Minerals
Engineering, 19:790-798.
Hope, G A, Woods, R, Parker, G K, Watling, K M and Buckley, F M,
2006. Spectroelectrochemical investigations of flotation reagent –
Surface interaction, Minerals Engineering, 19:561-570.
Kartio, I, Laajalehto, K, Suoninen, E, Karthe, S and Szargan, R, 1992.
Technique for XPS measurements of volatile adsorbed layers:
Application to studies of sulphide flotation, Surface Interface
Analysis, 18:807-810.
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 8 – SURFACE CHEMICAL CHARACTERISATION
Krause, M O, 1971. The Mζ X rays of Y to Rh in photoelectron
spectrometry, Chemical Physics Letters, 10:65-69.
Leppinen, J O, Basilio, C I and Yoon, R H, 1989. In-situ FTIR study of
ethyl xanthate adsorption on sulfide minerals under conditions of
controlled potential, International Journal of Mineral Processing,
26:259-274.
Mayer, D, Hallmeier, K H, Zerulla, D and Szargan, R, 2003. SXPS and
XANES studies of interface reactions of organic molecules on
sulfide semiconductors, in Solid–Liquid Interfaces, Topics in Applied
Physics (eds: K Wandelt and S Thurgate), 85:97-112 (Springer:
Berlin).
Mielczarski, E and Mielczarski, J A, 2005. Infrared spectroscopic studies
of galvanic effect influence on surface modification of sulfide
minerals by surfactant adsorption, Environmental Science and
Technology, 39:6117-6122.
Mielczarski, J A, Mielczarski, E and Cases, J M, 1996. Interaction of
amyl xanthate with chalcopyrite, tetrahedrite, and tennantite at
controlled potentials, Simulation and spectroelectrochemical results
for two-component adsorption layers, Langmuir, 12:6521-6529.
Pattrick, R A D, Mosselmans, J F W, Charnock, J M, England, K E R,
Helz, G R, Garner, C D and Vaughan, D J, 1997. The structure of
amorphous copper sulfide precipitates: An X-ray absorption study,
Geochima et Cosmochim Acta, 61:2023-2036.
Ruppender, H J, Grunze, M, Kong, C W and Wilmers, M, 1990. In situ
X-ray photoelectron spectroscopy of surfaces at pressures up to
1 mbar, Surface Interface Analysis, 15:245-253.
Schaufuß, A G, Nesbitt, H W, Kartio, I, Laajalehto, K, Bancroft, G M and
Szargan, R, 1998a. Reactivity of surface chemical states on fractured
pyrite, Surface Science, 411:321-328.
Schaufuß, A G, Nesbitt, H W, Kartio, I, Laajalehto, K, Bancroft, G M and
Szargan, R, 1998b. Incipient oxidation of fractured pyrite surfaces in
air, Journal of Electron Spectroscopy, 96:69-82.
Senior, G D, Shannon, L K and Trahar, W J, 1994. The flotation of
pentlandite from pyrrhotite with particular reference to the effects
of particle size, International Journal of Mineral Processing,
42:169-190.
Shchukarev, A, 2006. XPS at solid-solution interface: Experimental
approaches, Surface Interface Analysis, 38:682-685.
Smart, R StC, 1991. Surface layers in base metal sulphide flotation,
Minerals Engineering, 4:891-909.
Smart, R StC, Amarantidis, J, Skinner, W, Prestidge, C A, La Vanier, L
and Grano, S, 1998. Surface analytical studies of oxidation and
collector adsorption in sulfide mineral flotation, Scanning
Microscopy, 12:553-583.
Smart, R S C, Amarantidis, J, Skinner, W M, Prestidge, C A, La Vanier,
L and Grano, S R, 2003. Surface analytical studies of oxidation and
collector adsorption in sulfide mineral flotation, in solid-liquid
interfaces (eds: K Wandelt and S Thurgate), Topics in Applied
Physics, 85:3-60 (Springer: Berlin).
Stöhr, J, 1992. NEXAFS Spectroscopy, 403 p (Springer: Berlin).
Szalkowski, F J, 1977. The characterization of surfaces by electron
spectroscopy, Journal of Colloid and Interface Science, 58:199-215.
Szargan, R, Uhlig, I, Wittstock, G and Roßbach, P, 1997. New methods
in flotation research – Application of synchrotron radiation to
investigation of adsorbates on modified galena surfaces,
International Journal of Mineral Processing, 51:151-161.
Trahar, W J, Senior, G D and Shannon, L K, 1994. Interactions between
sulphide minerals – The collectorless flotation of pyrite,
International Journal of Mineral Processing, 40:287-321.
Wang, X, Forssberg, K S E and Bolin, N J, 1989. Adsorption of
copper(II) by pyrite in acidic to neutral pH media, Scandinavian
Journal of Metallurgy, 18:262-270.
Woods, R, 1996. Chemisorption of thiols on metals and metal sulfides, in
Modern Aspects of Electrochemistry (eds: J O M Bockris, B E
Conway and R E White), volume 29, pp 401-453 (Plenum: New
York).
Zachwieja, J B, McCarron, J J, Walker, G W and Buckley, A N, 1989.
Correlation between the surface composition and collectorless
flotation of chalcopyrite, Journal of Colloid and Interface Science,
132:462-468.
153
 HOME

CHAPTER 9

Laboratory Flotation Testing – An Essential Tool for

Ore Characterisation
Kym Runge

Manager – Flotation Process Technology, Metso Minerals (Australia) Limited, Unit 1, 8 - 10 Chapman
Place, Eagle Farm Qld 4009. Email: kym.runge@metso.com

Dr Kym Runge is a mineral processing engineer specialising in flotation circuit analysis using modelling and
measurement techniques. She currently works for Metso Minerals as a flotation specialist both consulting to
industry and managing the development of Metso’s flotation related product line. Her expertise in flotation
was developed during a ten year period at the Julius Kruttschnitt Mineral Research Centre. During this time
she initially studied for a PhD in flotation modelling and later worked full time as a researcher and
consultant. She was instrumental in the development of JKSimFloat, which is used extensively in the
industry for flotation circuit modelling and analysis. She also has three years industrial experience from
when she worked at Pasminco’s Broken Hill flotation concentrator.

Abstract
Introduction
Batch Laboratory Testing Procedures
Techniques Available for Interpretting Batch Laboratory Flotation Test Results
Areas of Application
Advantages and Limitations of Using Batch Flotation Testing for Circuit Diagnosis
Concluding Remarks
References
Appendix – Worked Examples

ABSTRACT
The properties of the ore have a significant impact on the ultimate grade
and recovery achievable in a flotation circuit. To maximise recovery, the
proportion of particles containing the valuable mineral which are floatable
should be maximised. To maximise the grade, the flotation rate of the
valuable mineral should be significantly higher than the flotation rates of
the gangue minerals in the ore. This is often referred to as ‘maximising
the selectivity’.
Laboratory batch flotation, in which a 1 - 2 kg sample of ore is floated
using a standard set of operating conditions, is the standard technique used to
assess ore floatability. Traditionally these tests have been performed
extensively to determine the ultimate flotation design, screen potential
reagents, determine optimum feed grind size and predict the change in
performance of different ore types. These tests have usually been performed
using the feed ore sample to the process but they are increasingly being used
to characterise the floatability of internal flotation circuit streams and ‘map’
how the floatability is changing around a flotation circuit and across different
processes.
Key ore floatability parameters (flotation rate constants and proportion
of mineral in different floatability classes) can also be derived from the
laboratory test flotation response. The values of these ore floatability
parameters can be evaluated and compared to rank different ore types or
reagent suites. These parameters are also suitable for predicting, through
mathematical modelling, the grade-recovery relationship achievable in a
multistage flotation circuit.
There are various graphical and modelling techniques that have been
developed to interpret laboratory batch flotation test results. In this chapter,
the recommended procedure for performing these laboratory flotation tests
and the various graphical and modelling analysis methods used to
interpret these test results will be outlined.
INTRODUCTION
To achieve separation in the flotation process, those particles
containing the valuable mineral need to float and they need to
float more quickly than those containing predominantly the
gangue minerals. The particle size, degree of liberation of the
minerals and the hydrophobicity of the mineral surface will in
large dictate a mineral’s recovery rate. Grinding and chemical
addition are weapons in the metallurgist’s armoury used to
modify particle properties for a particular ore type in an attempt
to achieve maximum separation.

Flotation Plant Optimisation Spectrum Series 16 155

CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION
A vast array of techniques can be used to predict the optimum
grind size and chemical suite to use for a particular ore.
Ultimately, there is always a need for small-scale testing to
validate the results of these analyses or to explore, screen or
prioritise different options. This is where laboratory batch
flotation testing and the various techniques that can be used to
analyse these batch test results come in useful. They are relatively
easy and inexpensive to perform in comparison to detailed
liberation or chemical speciation analyses and they go some way
to replicating the process conditions that will be present in the
full-scale processing plant.
The aim of this chapter is to outline how these ore
characterisation tests should be performed and demonstrate the
various graphical and modelling techniques that can be used to
interpret the results produced from these tests.
BATCH LABORATORY TESTING PROCEDURES
The standard procedure
There are various designs of laboratory-scale flotation cells which
can be used to perform the laboratory batch flotation tests. My
preference is to use a bottom driven laboratory cell allowing
impeller speed and air rate adjustment, an example of which is
shown in Figure 1. This type of cell, used with a fixed depth
scraper to regularly remove the entire froth phase during
concentrate collection, can produce reproducible test results even
between different operators.
FIG 1 - A bottom driven laboratory flotation cell supplied by
Runge Engineering.
A typical test involves washing an ore pulp sample into the
laboratory cell. Make-up water is added to achieve the desired
volume. The impeller is started and set to the desired speed.
Collector and depressant (if required) are then added and an
appropriate conditioning time is allowed to elapse. At the instant
the air flowmeter is set to the appropriate air rate, the timing of
the test commences. The concentrate produced from the cell
flows and/or is scraped into trays and at designated time intervals
the trays are removed and replaced. Water is continually added to
the cell to maintain the desired froth level. The test continues
until the froth is barren and no more particles are reporting to the
froth in significant quantities. It is recommended that one more
concentrate be collected after this point to ensure the end point of
the test can be established.
All concentrate samples and the final tailing sample are
weighed wet and dry and assayed for the elements of interest.
These results are used to calculate the rate of mineral recovery as
156 Spectrum Series 16
100
Mineral Recovery (%)
80
60
40
20
0
0 2 4 6 8 10 12 14
Time (min)
FIG 2 - Typical mineral versus recovery time curve obtained from a
laboratory batch test.
a function of time, usually on a cumulative basis. A typical
cumulative mineral recovery curve is depicted on Figure 2.
As an aside, the water in the concentrate versus the water in the
flotation cell will be used to estimate recovery by the entrainment
mechanism. Thus known quantities (or no) water should be used
to wash concentrate into the tray.
Samples to be tested can be collected directly from an operating
flotation circuit, referred to as ‘hot samples’, or prepared in the
laboratory – grinding to produce the appropriate particle size or
adding reagent to achieve the required pulp chemistry. The time
between sample collection and flotation should be minimised to
prevent ageing or oxidation of particle surfaces prior to testing.
The weight or size of sample added to the laboratory flotation
test should be chosen to achieve the required pulp density. Pulp
density is a variable which often affects batch flotation test
results. As it is easier to add water (rather than remove it), it is
recommended that all samples be floated at a low per cent solids
(ie less than ten per cent solids). The use of a low per cent solid
also minimises the risk of bubble saturation or overloading. In
tests where there is a high proportion of floatable solids
performed using a low air rate and at a high pulp density, bubble
surfaces can become saturated, resulting in low recovery rates.
This condition should be avoided as the result is no longer a
function of only the ore but the air rate used in the test.
The water used to dilute the samples and added during a test to
maintain froth depth should be as similar to that used (or to be
used) in the plant as possible. Water properties (ie ph, Eh,
temperature, dissolved oxygen and dissolved molecules) will
affect the composition of reagent and deposition products on
particle surfaces, which ultimately affects ore floatability. Frother
(as long as it has no collecting properties) should also be added to
all make-up water in sufficient quantities to maintain a constant
bubble size in the pulp and minimise bubble breakage within the
froth phase. For most frothers, a concentration of 20 ppm within
the make-up water will be sufficient for this purpose. Distinct
additions of frother in the middle of a test should be avoided as
they can result in a spike in flotation recovery.
It is important that all tests are performed using the same air
rate, impeller speed, froth depth and froth scraping rate so that
any change in a batch test result is a consequence of a change in
the ore characteristics rather than to any change in the
environment in which it is floated. The air rate chosen should
be high to prevent bubble saturation and maintain a fluid
non-collapsing froth. Air rates of 10 - 18 L/min are usually
appropriate for tests performed using a 5 L cell. Impeller speed
should be sufficient to keep all solids in suspension but not that
high that molecules loosely deposited on surfaces are stripped off.
An impeller speed within 750 to 1000 rev/min is recommended.
Flotation Plant Optimisation
 CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

Ideally the froth should be removed the instant it is formed so
that differences in results can be attributed to changes in pulp
flotation kinetics rather than to any change in the flotation froth
characteristics. As a practical compromise, it is recommended
that two to three centimetres of froth be maintained throughout
the test and a scraper of sufficient depth be used to remove all
but 0.5 cm of this froth at regular time intervals. Make-up water
should be added regularly to maintain a constant froth depth as
solids and water report to concentrate. A fast froth scraping rate is
desirable but one stroke every ten seconds is usually a speed that
allows the execution of other tasks (addition of water and
changing of trays).
The time intervals over which each concentrate is collected
should be chosen such that the shape of the mineral recovery time
curves can be well established for the particular operating
conditions chosen to do the tests. A test of a slow floating
platinum ore, for instance, may require 45 minutes to an hour to
reach completion and thus the first concentrate can be collected
after two minutes of flotation. A faster floating copper porphyry
flotation test, however, may finish in five minutes and a first
concentrate may need to be collected after 20 seconds.
Depending on one’s circumstances, some deviation may be
required from the recommended test conditions (eg use of high
per cent solids to obtain sufficient concentrate samples for assay,
samples left to sit for long periods). Tests should be performed
to establish whether any adverse effect arises from the
proposed change.
It is clear from these graphs that a change in flotation time
(Figure 3a) simply increases or decreases flotation recovery but
does not result in changes in separation efficiency. A steepening
of the separation curve, however, results from refloating the
concentrate (Figure 3b). This occurs because of the increase in
the fast to slow ratio of particles in the feed to subsequent stages.
The net result of this improvement in separation efficiency is an
improvement in the concentrate grade or purity achievable at a
particular mineral recovery.
Lock cycle batch flotation experiments provide a better insight
into the expected grade and recovery achievable from multistage
processing of an ore. These tests are performed such that they
replicate, on small scale, the full-scale flow sheet. To incorporate
the effect of recycle streams, they are often performed multiple
times with the recycled streams from a previous test added at the
appropriate place in the subsequent test. Tests are performed until
the mass flow in all samples converges to a constant value.
Figure 4 shows a schematic of a lock cycle test performed to
replicate a rougher/cleaner flotation circuit in which the cleaner
tail is recirculated back to the feed.
New Feed
Recycled Tail

Refloatation and lock cycle tests

The grade versus recovery relationship derived from a standard New Feed
batch laboratory flotation test will not be that achievable in a
multiple stage flotation circuit. Multiple stage flotation results
Recycled Tail
in a more efficient separation between particles of different
floatability. Figure 3 demonstrates this point by showing the
flotation separation curves for a batch flotation test (calculated
using flotation modelling techniques) performed over different
time periods and for tests in which the concentrate has been Circuit Tailing
refloated. The flotation separation curve is a plot of recovery
New Feed
versus the flotation rate of a particle in a particular flotation
process. If a flotation process was a perfect separator, the
separation curve would be a vertical line – complete recovery of
particles to the right of the line and no recovery of particles to the Circuit Concentrate
left. The probabilistic nature of the flotation process, however,
results in non-ideal separation. FIG 4 - Lock cycle test schematic.

100 100
90 90
80 80
Recovery (%)
Recovery (%)

16
70 70
60 60
8 50
50
Refloatation
40 40
4
30 30
20 20
10 10
0 0
0.001 0.01 0.1 1 10 0.001 0.01 0.1 1 10

Flotation Rate (min-1) Flotation Rate (min -1)

(a) (b)

FIG 3 - Recovery of particles with different flotation rates (a) after four, eight and 16 minutes of flotation and (b) after one, two and three
refloatation stages of eight minutes in duration.

Flotation Plant Optimisation Spectrum Series 16 157

CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

The disadvantage of these types of tests is that they are time- 100
consuming to perform and they result in a single grade and

Gangue Mineral Recovery (%)

recovery value rather than the relationship between the two
values derived from the single stage test. To investigate an 80
alternative flow sheet or a change in the residence time of a stage,
another experiment needs to be performed. 60
A more time- effective option for estimating multistage circuit
grade and recovery is to model the data from a single stage 40
experiment (or a lock cycle test). This model can be used to
mathematically predict circuit grade and recovery achievable Increasing Selectivity
from different circuits. The procedures to perform this type of 20
analysis will be outlined later in the chapter.
0
TECHNIQUES AVAILABLE FOR INTERPRETING 0 20 40 60 80 100
BATCH LABORATORY FLOTATION TEST Valuable Mineral Recovery (%)
RESULTS
FIG 6 - Mineral selectivity curves with the important features
Mineral recovery, grade versus recovery and denoted.
selectivity curves
The 45 degree line on this plot represents the points at which
Mineral recovery, grade/recovery and selectivity curves are all
no selectivity occurs – the recovery of valuable mineral is equal
graphs that can be constructed from the information collected
to the gangue mineral recovery and no separation is achievable.
during a single batch flotation experiment. These curves are ideal
The further to the right of this line the selectivity curve lies, the
for comparing the results of different tests. better. Different selectivity curves should be created for each of
The mineral recovery curve displays the cumulative mass of the gangue minerals of interest as different minerals often exhibit
mineral (on a percentage basis) recovered at each time interval different degrees of selectivity with the valuable mineral.
during a test. The slope of this curve is an indicator of the rate of The overall measure of selectivity in the batch flotation test
flotation recovery and the recovery value at which the curve is the grade (or purity) of the concentrate produced. The
asymptotes is an indicator of the proportion of floatable mineral cumulative concentrate grade versus cumulative recovery curves
in the tested sample (Figure 5). (Figure 7) produced from different tests can be compared to
determine those conditions which result in the best selectivity
with respect to all minerals.
100
100
Mineral Recovery (%)

Cumulative Recovery to

80 Improving Overall Selectivity

Increasing Proportion of
80
Concentrate (%)

Floatable Mineral
60
Increasing Flotation Rate 60

40 40

20 20

0
0 0 5 10 15 20
0 2 4 6 8 10 12 14
Cumulative Concentrate Grade (%)
Time (min)

FIG 5 - Cumulative mineral recovery curve with the important FIG 7 - Cumulative recovery versus cumulative concentrate grade
features denoted. produced during a batch flotation experiment.

It is desirable that the valuable mineral in a sample exhibit fast
flotation kinetics (steep slope) and the proportion of non-floating
mineral in the sample be minimised. The proportion of
non-floating valuable mineral is an indicator of the amount of
mineral that is either in particles too large or small for flotation,
locked with non-floating particles or with a hydrophilic surface
due to insufficient collector coverage or depressant coatings.
Regardless of the reason, non-floating mineral is not recoverable
in a flotation circuit and should be minimised if circuit recovery
is to be maximised.
Flotation is a separation process and thus it’s important that the
valuable mineral is not only recoverable but that it’s recoverable
to a greater extent than the gangue mineral. Selectivity curves are
ideal for this type of evaluation. On these curves, the recovery of
the valuable mineral is plotted against the recovery of a mineral
of interest (Figure 6).
These curves do not, however, give an estimate of the grade
and recovery achievable in a multiple stage flotation circuit. Lock
cycle tests or modelling are better suited to this task. These
curves do give a good relative measure of selectivity. An
improvement in selectivity in these tests should result in a better
grade at a particular recovery at full scale.
Entrainment assessment techniques
Gangue can report to the concentrate either attached to bubbles
(due to its own hydrophobicity or because of an association with
a hydrophobic mineral) or entrained with the water phase. It is
important to be able to differentiate between these two recovery
mechanisms as ultimately this distinction may affect the strategy
that is adopted to reduce gangue recovery. This section outlines
techniques that can be used to estimate the degree of entrainment
in a laboratory batch flotation test.

158 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

Entrainment is a consequence of particles suspended in the

pulp phase following the water into the froth and ultimately into 2.5
the concentrate. The mass of a mineral entrained into the Gangue to Water Ratio in Concentrate
Ratio
concentrate is proportional to the water flow to concentrate 2.0 Gangue to Water Ratio in Pulp
and can be estimated using Equation 1 (Johnson, 2005). The
classification function (Cf,i) is a factor to account for the degree of 1.5

Ratio
drainage with respect to the water flow and must be estimated to Maximum C f,i Limit
perform the calculation. It is known to fall between zero and one; 1.0
zero representing the condition of total drainage where there is no
entrained recovery of particles regardless of the water recovered, 0.5
C f,i Estimate
and one representing the condition where there is no drainage and
the concentration of particles in the water in the concentrate (due 0.0
to entrainment) is the same as in the pulp phase. Cf,i is known to 0 5 10 15
be strongly dependent on size with the coarser particles Time (min)
exhibiting a high degree of drainage (Cf,i = 0) and the fine
particles usually following the water phase (Cf,i = 1).
FIG 8 - Cf,i calculated using Equation 1 versus time in a batch
Fi = C f, i ωi, pulp Fw (1) flotation test.

where:
70
Fi is the flow of a component i into the concentrate due to
entrainment 60

Gangue Recovery (%)

Cf,i is the classification function for component i 50 Overall
ωi,pulp is the mass to water ratio of component i in the pulp 40
phase Entrainment
30 Estimate
Fw is the flow of water to the concentrate
20
Thus to calculate the recovery by entrainment in a batch test,
the water to solids ratio in the pulp, water flow to concentrate and 10
an estimate of Cf,i is required. The water flow to concentrate can 0
be measured and the average water to solids ratio in each stage 0 5 10 15
can be estimated based on knowledge of the cell volume and the Time (min)
particle mass balance. The challenge is to estimate the degree of
drainage in the system.
If one of the minerals is known to be non-floating, an inversion FIG 9 - Curves used to assess the predominant gangue recovery
of Equation 1 can be used to calculate a Cf,i estimate for the other mechanism in a batch flotation test.
minerals in the ore. This, unfortunately, is not usually the case
with even the non-floating minerals locked to some extent with
the valuable and exhibiting floatability. Johnson (2005) published compared to the overall recovery curve for that mineral
typical values of Cf,i for siliceous non-sulfide gangue (Table 1). (Figure 9). The difference in these two curves is the proportion of
These values coupled with knowledge of the particle size the mineral recovered due to true flotation (attachment to
distribution of a particular sample can enable a reasonable bubbles).
estimate of Cf,i to be calculated.
Ore floatability mapping in a flotation circuit

TABLE 1
Typical values of Cf,i (after Johnson, 2005).
Size fractions (μm)
-11 -16 +11 -23 +16 -33 +23 -44 +33
0.77 0.46 0.30 0.24 0.13
Alternatively, Cf,i can be estimated using results from the batch
test by recognising that towards the end of an experiment, the
predominant particle recovery mechanism is entrainment. If one
calculates the ratio of mineral to water in both the concentrate
and pulp and plots this value for each stage of a batch experiment
(Figure 8), it usually decreases, asymptoting at a particular value
towards the end of the test. This value is an estimate of the Cf,i
value. The larger values at the beginning are because recovery is
due to entrainment and flotation in these earlier stages.
Once a Cf,i estimate has been obtained and the gangue
entrainment flow has been calculated (Equation 1), a cumulative
entrainment recovery curve can be constructed which can be
Traditionally batch flotation tests have only been performed using
samples representative of a flotation circuit feed. Much can be
learnt, however, by floating the samples collected from different
streams of an operating flotation circuit.
During flotation the faster floating particles will report to the
concentrate streams, whereas the slower floating particles will
report predominantly to the tailing streams. This results in
different streams possessing very different ore floatability
properties. As an example, Figure 10 shows the copper recovery
achieved in batch flotation tests performed using different streams
from Rio Tinto’s Northparkes flotation circuit.
This information can then be assessed in terms of mineral
recovery and selectivity. A traditional rule of thumb is to
recirculate streams to a point of similar assay. A better philosophy
is to recirculate streams to a point of similar floatability.
The final tailing is of special importance as it is discarded
and any valuable mineral contained in this stream is a lost
opportunity. High valuable mineral recovery rates coupled with
good selectivity in a tailing stream batch test is an indication of
insufficient residence time in the circuit.

Flotation Plant Optimisation Spectrum Series 16 159

CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

This direct comparison is difficult, however, when there are

100 Rougher Feed
Rougher Con
multiple feed or product streams. In many cases it is also
impossible to obtain ‘the desired’ samples from the flotation
Copper Recovery (%)

80 Rougher Tail
Scavenger Con process. In these cases, the mineral recovery rate of a ‘combined
60 Scavenger Tail feed’ or ‘combined product’ stream can be estimated by
Cleaner Feed
mathematically combining the experimental results of the
Cleaner Con
40 standard laboratory batch flotation tests on two or more streams
Cleaner Tail
Recleaner Feed (Equation 2) based on their relative flows in the circuit.
20
Recleaner Con
Recleaner Tail n
0
0 2 4 6 8
Clnr Scav Tail
Combined Stream, t minutes
∑F
s=1
Stream s
Mineral j × R Stream s, t minutes
Mineral j
R Mineral j = n
(2)
Time (min)
∑F
s=1
Stream s
Mineral j

FIG 10 - Copper recoveries achieved in the batch tests performed

on different streams in the Northparkes flotation circuit.
Streams with poor valuable mineral to gangue mineral
selectivity are appropriate targets for regrinding or depressant or
collector addition. Batch flotation of different streams in a
flotation circuit enables the appropriate streams for this type of
modification to be delineated from those that would not benefit
from this type of processing.
Regular batch laboratory flotation test mapping during circuit
surveys will provide a record of ore floatability properties –
enabling the comparison of this property between different surveys.
Nodal analysis comparisons
The nodal analysis technique was developed to compare the
floatability of particles before and after processes (ie nodes)
within a flotation circuit (Runge et al, 1997). It involves
comparing the mineral recovery rate and selectivity achieved in
batch laboratory flotation tests of the feed and product streams of
a node. A node is defined as a point in a circuit where streams are
either combined, separated or altered in some way (Figure 11). A
node can be a single unit such as a flotation cell, a regrind mill or
a pump sump, a combination of units (eg a bank of cells) or even
an entire circuit with the feed being split into the final concentrate
and final tailing stream.
where:
Stream s
F Mineral j is the flow rate of mineral j in stream s
Stream s, t minutes
R Mineral j is the cumulative recovery of mineral j in stream
s after t minutes of flotation in the standard
laboratory batch flotation test
Take for example a flotation column. A column has a single
feed but two product streams. Equation 2 can be used to
recombine the concentrate and tailing stream to produce a single
‘combined’ product stream result. The combined stream
floatability can then be compared to the feed stream floatability
to assess whether there has been any change across the unit.
Figure 12 shows the galena recovery rate measured in a batch
flotation test of the feed, concentrate and tailing streams of an
industrial column. It also shows the calculated ‘combined
product’ stream which, in this case, is very similar to the column
feed. It can therefore be concluded that floatability does not
change significantly across this unit. The difference in galena
recovery rates in the concentrate and tailing stream is attributed to
the faster floating galena-containing particles concentrating in the
concentrate streams and the slower floating galena-containing
particles concentrating in the tailing streams. Using this technique
in a number of different flotation circuit operations, it has been
shown that ore floatability is not often affected in a full-scale
flotation cell (Runge et al, 2006).

100
Galena Recovery (%)

90
80 Column Feed
70
60 Column Con
50 Column Tail
40
30 Column Con +
20 Column Tail
(a) (b)
10
0
0 10 20 30 40

Time (min)

FIG 12 - Galena recovery versus time for the feed, concentrate,

(c) (d)
FIG 11 - Examples of nodes in a flotation circuit. (a) Flotation cell;
(b) tower mill; (c) pump sump; (d) flotation circuit.
The experimental feed and product mineral recovery rates or
mineral selectivity curves derived from laboratory batch flotation
tests can be compared directly when there is only one feed stream
and one product stream from a node.
tailing and combined product of an industrial-scale flotation
column (after Runge et al, 1997).
Regrinding and reagent addition are employed in a flotation
circuit specifically to alter particle floatability. Often the
effectiveness of these processes is assessed by expensive and
often inconclusive on-off testing. Comparison of ore floatability
using nodal analysis techniques can enable a direct assessment of
the effectiveness of these particle modification operations. In
Figure 13a for example, it is obvious that staged reagent addition
has resulted in a significant increase in the proportion of floatable
copper in the circuit. In Figure 13b it is obvious that regrinding

160 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION
Chalcopyrite Recovery (%)
100
80
60
40
Feed to Reagent Node
20
Product of Reagent Node
0 0
0 2 4 6 8 10
Time (min)
FIG 13 - Nodal analysis comparative graphs: (a) mineral recovery before and after collector addition, (b) chalcopyrite/pyrite selectivity
before and after regrinding.
has not resulted in an improvement in the chalcopyrite/pyrite
selectivity. These types of observations provide the metallurgist
with invaluable information regarding the effect of an ore
modification process.
These techniques will be demonstrated in a worked example
included at the end of this chapter. For a more detailed example,
the reader is referred to Runge, Franzidis and Manlapig, 2004 in
which these techniques are used to assess the effect of staged
reagent addition, regrinding and lime addition on ore floatability
in an industrial circuit.
Ore parameter estimation techniques
The mineral recovery response measured in a laboratory batch
flotation experiment can be used to calculate ore floatability
parameters – numbers which represent the rate and extent to
which a mineral will float. These parameters can be used as a
means of comparison of different batch test experiments or within
a flotation model for performance prediction.
Flotation is considered a first order kinetic process, recovery
being a function of the time in the process (t) and a flotation rate
constant (k). In a batch laboratory flotation environment, recovery
can be calculated using Equation 3.
R = 1 − exp( − k t )
The flotation rate constant is a function of the operating
conditions in the cell and also the size, mineralogy and surface
chemical speciation of the particles being floated. There are a
number of different types of particles in an ore sample and each
will float with a different flotation rate – thus an ore sample
exhibits a distribution of floatabilities. There are no methods
currently available for measuring each particle’s individual rate.
Instead the problem is simplified by assuming the form of the
floatability distribution. The two methods most commonly used
involve assuming a continuous shape for the distribution (eg
rectangular distribution (Klimpel, 1980)) or splitting a mineral
into different floatability components. In this chapter, derivation
and use of multiple component parameters will be discussed. For
those wanting more information regarding shaped distributions, the
reader is referred to a paper written by Chander and Polat (1994).
Mineral recovery in a batch flotation test using multiple
components can be calculated using Equation 4.
n In a similar vein, laboratory batch flotation tests can be
R = ∑ m i (1 − exp( − ki t))
i= 1
Flotation Plant Optimisation
100
Combined Feed
Pyrite Recovery (%)
80 Product
60
40
20
0 20 40 60 80 100
Chalcopyrite Recovery (%)
(a) (b)
The rate of each component (ki) and the proportion of the
mineral in that class (mi) are the ore floatability parameters of the
system that can be derived from batch laboratory flotation test data.
This involves calculating the ‘best’ set of these parameters that
minimise the sum of squares difference between the recoveries
calculated using Equation 4 and that measured in the laboratory
batch flotation experiments. This difference can, optionally,
be weighted by the standard deviation of the experimental
data (Equation 5). These problems are easily set up on an
Excel spreadsheet using the Solver add-in to perform
the minimisation.
( Rj.calculated − Rj.experimental )2
SSE = ∑ 2
(5)
j SDj.experimental
The number of components that can be derived to describe a
mineral’s floatability is dependent on the amount of data
available to do the parameter estimation. Unfortunately the data
from a single batch laboratory flotation test is only sufficient to
determine, with statistical confidence, the parameters associated
with a simple representation of the system (one floating and one
non-floating component). More complex representation requires
an increase in the experimental data set size.
(3) Harris (1998) recognised that this data set size could be
increased by using information collected from laboratory batch
flotation tests performed using different streams of a circuit.
A circuit by its very nature results in a redistribution of
the floatability components – fast floatability components
predominantly reporting to the concentrate streams and the slower
and non-floating components predominantly reporting to the
tailing streams. If we assume that the feed consists of a number of
floatability components, the other streams are also made up of
these same floatability components, just in a different ratio.
Analysis involves assuming the rate of each floatability
component in the different batch tests is the same and that the
mass of each floatability component is conserved across different
nodes in the process. These constraints enable the data to
parameter ratio to be significantly increased and enable
parameters of multicomponent systems to be derived with
statistical confidence. The measured distribution of floatability
components in the different streams of an industrial flotation
circuit derived using this type of analysis is shown in Figure 14.
(4) performed with multiple stages or with recycle streams to
increase the number of datapoints available for parameter
Spectrum Series 16 161
CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

88.9 3.8 3.8 1.3

5.8 3.7 4.2 3.7
5.3 5.3 5.3 5.3

85.1
2.5
2.1
0.5
0.0
0.0 0.0
11.7 0.5
2.7 0.0
80.7
0.0
1.8 11.7
0.0 2.2
0.0
4.8 Legend
1.9 6.9 Fast Floating
0.0 0.3 Slow Floating
0.0 Non Floating

FIG 14 - Mass in different copper floatability components around the Northparkes industrial flotation circuit (based on 100 units of
copper in the feed).

estimation. Thus multiple ore floatability parameter estimation
can be performed without the need to collect samples from an
operating flotation circuit.
A step by step worked example of the ore parameter derivation
technique will be provided at the end of this chapter.
Comparison of ore floatability parameters
Ore floatability parameters provide the metallurgist with numbers
that can be used to quantitatively assess and compare the
floatability of ore in different laboratory batch flotation tests.
For optimal separability in a flotation circuit, the valuable
mineral flotation rate should be high and the proportion of slow
or non-floating valuable mineral should be minimised. Gangue
flotation rates should ideally be low and a large proportion of the
gangue mineral should be non-floatable. It will not be possible to
separate mineral components of similar flotation rate so the ratio
of the slow floating gangue to slow floating valuable mineral
flotation rate is also an important parameter.
Hay and Rule (2003) characterised the ore floatability of the
ore feeding a number of different platinum concentrators using a
two floating component model. They found that the concentrate
recovery and grade were strongly correlated with the fraction of
fast floating valuable mineral with a proportion of the slow
floating fraction being recoverable when the ratio of the slow
valuable mineral to gangue flotation rate ratio was above a
threshold value.
Ore floatability component mapping, as illustrated in Figure 14,
can be used in conjunction with the mineral recovery and
selectivity graphical information to assess the appropriate
destination of recycle streams and the target streams for
regrinding, collector or depressant addition.
circuit scenarios. They can replace the need for complex lock
cycle tests, giving estimates of the ultimate grade and recovery
achievable in multistage flotation.
The aim of simulation is to calculate the distribution of
floatability components across a proposed circuit. Beginning with
the feed and the first unit in the circuit, recovery in the product
streams are calculated. This calculation sequence is continued for
each unit process, the product of earlier processes forming the
feed to subsequent processes. Recycle streams during the first
iteration are set to zero and multiple iterations of circuit
calculations are performed until the mass in these recycle streams
converges to a constant value. Once convergence has been
achieved, the mass of the different floatability components of
each mineral in each stream can be added together. These mineral
masses can be used to calculate stream mineral grade (mineral
mass divided by total mass in stream) and recovery (mass with
respect to feed).
At each process in the circuit, recovery equations are used to
calculate product component flows. Wherever streams are mixed,
for instance, the mass of each floatability component in the
product stream is equal to the addition of the mass of this same
floatability component in the feed streams.
The recovery equation used for a flotation process is dependent
on its particle residence time distribution. For a semi-batch
process (like that performed in the laboratory batch flotation test)
recovery of each component is calculated using Equation 3.
Conventional mechanical cells behave as perfect mixers and thus
have a different recovery equation (Equation 6).
kτ
R= (6)
1 + kτ

Simulation using ore floatability parameters
The advantage of ore floatability parameters over the traditional
graphical techniques, are that they can be used for prediction.
Once determined, they can be used to rapidly investigate many
The flotation recovery equation can be extended to include
parameters associated with entrainment and the cell operating
variables (eg air rate, froth depth). Flotation cell water recovery
equations are also required to calculate the circuit water flow
information required for cell residence time calculations. For

162 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

more information on these types of simulations and the methods

Final Concentrate Grade (%)

required to determine the parameters of these relationships, the
reader is referred to Harris et al, 2002.
A simplification to these more complex models can be used to
assess the separability of a particular ore in a particular flow
sheet. The effect of the cell operating variables and residence
time can be lumped into a single parameter, the cell operating
constant, θ (Loveday, 1966). Flotation recovery of a component
is a function of only this parameter and its particular batch
flotation test rate constant (Equation 7).
k batch θ
R=
1 + k batch θ
Multiple simulations are performed for a particular flow sheet,
with the cell operating constants, θ, randomised in each stage.
This randomisation is a representation of the range of different
cell volumes/cell operating conditions which could be used to
achieve flotation separation. It is equivalent to the flotation
operator slowing or speeding up the process to achieve the
desired concentrate grade and recovery at each stage.
The resulting final concentrate grade/recovery numbers from
these simulations can be plotted. Figure 15 shows examples of the
type of curves generated using this type of analysis. These curves
were generated using ore floatability constants derived from
batch laboratory experiments applied to a three stage flotation
circuit. The result is not a single grade versus recovery
relationship, as changes in the cell operating constant in one stage
result in both a change in recovery but also a change in the
separation efficiency of the circuit.
These curves can be generated to compare the floatability of
different ore types (Figure 15b) or to compare the ability of
different flow sheets to separate the minerals in a particular ore
sample (Figure 16).
Simulations required to create these curves can be performed
on an Excel spreadsheet. Alternatively, packages such as
JKSimFloat and Limn© have been designed to enable these
types of simulations to be performed rapidly in a user
friendly environment.
Estimation of full-scale residence time
requirements
The time it takes for a mineral to be recovered in a batch laboratory
flotation test can be used to estimate the cell volume and the
number of cells at full scale required to achieve this same recovery.
Recovery rates in the batch cell environment are usually
Final Concentrate Grade (%)
60
58
56
54
52
50
48 Three Stage
46 Four Stage
44
42
40
(7)
40 60 80 100
Circuit Recovery (%)
FIG 16 - Circuit grade and recovery simulated for the same ore
processed in a three and four stage flotation circuit.
significantly higher than those measured at full scale due to the
higher degree of turbulence and the higher froth recoveries
achievable at small scale. A scale-up factor (C) can be introduced
to the full-scale recovery equation to account for these differences
and relate full-scale recovery to the rate measured in a batch
laboratory flotation test (kbatch) (Equation 8).
Ckbatch τ
R= (8)
1 + Ckbatch τ
The cell scale-up factor is a complex function of the operating
conditions at full scale and those used in the batch laboratory
flotation test. It will also be strongly dependent on the ore and its
effect on the full-scale froth characteristics. It is the author’s
recommendation that this scale-up factor be determined by
comparing batch and full-scale kinetics in an existing concentrator
processing a similar ore. The use of globally quoted scale-up
factors (eg batch test rate is twice the rate at full scale) is not
recommended as this factor changes significantly with the
operating conditions used in both the batch test and full-scale cells.
Once the batch test ore floatability parameters and the scale-up
factor have been established, simulations can be performed to
determine the number of cells at a particular volume that are
required to achieve the desired bank recovery. Additional
capacity is usually factored into this analysis to account for any
error and the usual increase in capacity (over design) often
demanded from a flotation circuit over time.

Final Concentrate Grade (%)

56 56
54 54
52 52
50 50
48 48
46 46
44 44
42 42
40 40
40 50 60 70 80 90 100 40 50 60 70 80 90 100
Circuit Recovery (%) (a) Circuit Recovery (%)
(b)

FIG 15 - Circuit grade and recovery simulated in a three stage flotation circuit. (a) For a particular ore type, (b) for the same ore subject to
different reagent/regrinding schemes.

Flotation Plant Optimisation Spectrum Series 16 163

CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION
AREAS OF APPLICATION
Laboratory batch flotation testing can be used as a diagnostic tool
throughout the development and life of a flotation circuit.
Even before a project has commenced, batch laboratory
flotation of drill core samples provide the information required to
make decisions regarding a project’s viability. Throughout the
development of a flotation circuit flow sheet, batch tests are used
to screen and determine the appropriate reagent suite and dosage
rates, the target grind size and an estimate of the circuit residence
time requirements.
From the day a flotation circuit is commissioned, the
metallurgist strives to better understand the process and
implement changes which will improve or optimise operation.
Batch laboratory flotation testing is one of a suite of tools used in
this process. New reagents are almost always screened first in
these small tests before implementation at full scale.
After implementation of a change in grind size, reagent or circuit
change at industrial scale, batch flotation tests can be used to assess
the effect of the change on mineral recovery rates and selectivity.
The ore feeding a flotation circuit is continually changing as
rock is extracted from different areas of the mine. Batch testing is
a valuable tool for comparing the floatability of different ores or
predicting the metallurgical performance that will be achieved
when a particular ore is processed.
Flotation modelling techniques published in the literature
(Harris et al, 2002; Runge et al, 2001; Coleman, Urtubia and
Alexander, 2006) are using batch testing to determine ore
characterisation properties. These models can be used for
flotation circuit design, flotation circuit diagnosis and simulation
of alternative operating strategies. They enable an estimation of
the overall change in grade and recovery achievable in a
particular process fed an ore with a particular mineral selectivity.
In summary, laboratory batch flotation testing is an essential
tool, which when used appropriately, helps a metallurgist to
design new flotation processes, develop strategies for improving
a flotation circuit and ultimately assess the effectiveness of
these strategies.
ADVANTAGES AND LIMITATIONS OF USING
BATCH FLOTATION TESTING FOR CIRCUIT
DIAGNOSIS
Small-scale batch laboratory tests are relatively cheap and easy to
perform. Almost all concentrator laboratories are equipped with
laboratory flotation cells and the associated equipment required
to prepare and process the samples produced from a test. They
also have the advantage that, to a significant degree, they
replicate the flotation process that occurs at full scale.
Batch tests also have the advantage that they enable the ore
floatability to be analysed in isolation to those effects related to
circuit operation. Poor flotation recovery measured in a circuit
survey can be due to a number of different effects: poor ore
floatability, insufficient residence time, use of overly deep froth
depths, to name a few. Batch laboratory tests are usually
performed using the same operating conditions and thus any
change between tests can be attributed solely to a difference in
ore floatability. Batch tests performed in conjunction with circuit
surveys can help decouple ore floatability and circuit operation
effects and highlight the cause of flotation recovery loss.
What is often forgotten when analysing batch flotation test
results, however, is that they are a good measure of pulp
floatability but sometimes a poor indicator of full-scale froth
phase performance. The froth in a small-scale laboratory test is
only a couple of centimetres in depth and is continually removed.
164 Spectrum Series 16
In a full-scale cell the froth can be ten to 50 cm in depth and there
are usually significant transportation distances between a point on
the surface of the froth and the nearest launder. The froth
produced in a small-scale cell is often very different in bubble
size, texture and viscosity to that observed when the same ore is
processed at full scale. This can result in a particular reagent
suite, producing a superior pulp floatability (and thus batch test
result) but poor recovery in a full-scale cell because of its effect
on movement of the froth phase.
The primary assumption associated with batch laboratory test
analysis is that pulp selectivity in the small scale is similar to
what would be achieved in the full-scale flotation machine. It’s
likely this is a valid assumption in mechanical cells but less
likely in alternative flotation technologies. The low degree of
turbulence in a column, for instance, usually results in the poor
recovery of coarse particles which are recovered quickly in a
batch flotation test.
Another risk during batch laboratory flotation testing is the
removal of loosely deposited surface coatings. The higher
agitation energies which exist in a small-scale cell can sometimes
remove molecules which would remain in situ in the full-scale
flotation environment.
These limitations, as with most diagnostic techniques, mean
that those conclusions drawn from batch laboratory testing should
be verified through full-scale trials.
Much can be gained by performing batch laboratory flotation
tests in conjunction with other measurements. Batch laboratory
flotation tests indicate the overall result of a change. Sizing,
mineralogical and chemical analysis tools can be used to
determine the reason for a particular result, often suggesting
alternative batch laboratory experiments to trial.
CONCLUDING REMARKS
In this chapter, batch laboratory testing procedures and the
modelling and graphical techniques that can be used to analyse
the results of these tests have been reviewed.
It has been demonstrated that batch laboratory flotation tests
have a role to play throughout the lifetime of a particular flotation
circuit operation. They can be used during the design of new
flotation processes to develop strategies for improving a flotation
circuit and, ultimately, to assess the effectiveness of these strategies.
They have the advantage that they are relatively cheap and easy
to perform and to an extent, replicate the full-scale flotation
process. They enable an assessment of ore floatability without the
complication of changing cell operating variable effects. Effects,
due to differences between the small-scale and full-scale cell,
should be kept in mind when interpreting results.
The value of laboratory batch flotation testing is greatly
enhanced when performed in parallel with other techniques that
can assist with data interpretation.
REFERENCES
Chander, S and Polat, M, 1994. In quest of a more realistic flotation
kinetics model, in Proceedings IV Meeting of the Southern
Hemisphere on Mineral Technology and III Latin American Congress
on Froth Flotation (eds: S Castro and J Alvarez), pp 481-500.
Coleman, R G, Urtubia, H E and Alexander, D J, 2006. A comparison of
BHP Billiton’s Minera Escondida flotation concentrators, in
Proceedings 38th Annual Meeting of the Canadian Mineral
Processors, pp 349-370 (Canadian Institute of Mining, Metallurgy
and Petroleum: Montreal).
Harris, M C, 1998. The use of flotation plant data to simulate flotation
circuits, in Proceedings SAIMM Mineral Processing Design
School (Southern African Institute of Mining and Metallurgy:
Johannesburg).
Flotation Plant Optimisation
 CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

Harris, M C, Runge, K C, Whiten, W J and Morrison, R D, 2002.

JKSimFloat as a practical tool for flotation process design and 100
optimization, in Mineral Processing Plant Design, Practice and

Galena Recovery (%)

Control, vol 1, pp 461-478 (Society for Mining, Metallurgy, and
Exploration: Littleton).
Hay, M P and Rule, C M, 2003. Supasim: A flotation plant design and
analysis methodology, Minerals Engineering, 16(11):1103-1109.
Johnson, N W, 2005. A review of the entrainment mechanism and its
modelling in industrial flotation processes, in Proceedings Centenary
of Flotation Symposium (ed: G J Jameson), pp 487-496 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Klimpel, R R, 1980. Selection of chemical reagents for flotation, in Mineral
Processing Plant Design, Ch 45, pp 907-934 (The American Institute of
Mining, Metallurgical, and Petroleum Engineers: Littleton).
Loveday, B K, 1966. Analysis of froth flotation kinetics, Trans IMM,
75:C219-C225.
Runge, K C, Dunglison, M E, Manlapig, E V and Franzidis, J-P, 2001.
Floatability component modelling – A powerful tool for flotation
circuit diagnosis, in Proceedings Fourth International Symposium on
Fundamentals of Minerals Processing, pp 93-107 (Canadian Institute
of Mining, Metallurgy and Petroleum: Montreal).
Runge, K C, Franzidis, J P and Manlapig, E V, 2004. Optimisation of
flotation pulp selectivity utilising nodal analysis, in Proceedings 36th
Annual Meeting of the Canadian Mineral Processors, pp 575-597
(Canadian Institute of Mining, Metallurgy and Petroleum: Montreal).
Runge, K C, Harris, M C, Franzidis, J P and Manlapig, E V, 2006.
Conservation of floatability across processes in industrial flotation
circuits, in Proceedings XXIII International Mineral Processing
Congress (eds: G Onal, N Acarkan, et al), vol 3, pp 1909-1914
(Promed Advertising Agency).
Runge, K C, Harris, M C, Frew, J A and Manlapig, E V, 1997. Floatability
of streams around the Comino Red Dog lead cleaning circuit, in
Proceedings Sixth Annual Mill Operators’ Conference, pp 157-163
(The Australasian Institute of Mining and Metallurgy: Melbourne).
APPENDIX
WORKED EXAMPLES
Example 1 – regrind evaluation
A tower mill has been incorporated into the Eureka lead circuit
to regrind lead rougher concentrate in an effort to improve
liberation and thus the grade achievable in the lead flotation
circuit concentrate.
Batch laboratory flotation tests were performed on the rougher
concentrate and the reground rougher concentrate (after reagent
addition) to assess whether this unit is having the desired effect.
Six timed concentrates and a tailing sample were produced
from each test. Table A1 outlines the mineral grade measured in
each of these samples.
This data was then used to calculate the cumulative grade and
recovery achieved as a function of time in these tests. Figure A1
shows the grade versus recovery curve and Figure A2 shows the
galena gangue selectivity curves.
80
60 Rougher Concentrate
Reground Rougher
40
Concentrate
20
0
40 50 60 70 80
Galena Grade (%)
FIG A1 - Galena grade versus recovery achieved in batch flotation
tests of the lead rougher concentrate before and after regrinding.
In contrary to what was expected, regrinding has not resulted in
an improvement in the floatability of the ore. At low galena
recoveries, the grade recovery relationship is significantly poorer
in the ‘after regrinding’ batch test. At high galena recoveries, the
relationship in the two tests is very similar.
This result is primarily due to deterioration in the pyrite-galena
selectivity. Galena recovery rate has changed very little by the
regrinding process, whereas there has been a significant increase
in pyrite recovery (Figure A3). There is very little difference in
the galena-sphalerite selectivity and a marginal improvement in
galena-gangue selectivity.
The grinding environment often results in changes in the
condition of particle surfaces – iron hydroxide coatings, for
instance, can deposit on all surfaces during grinding with mild
steel balls. Some change is occurring to the pyrite particles in this
grinding system, resulting in an increased recovery rate. Surface
chemical speciation of particle surfaces before and after grinding
would almost certainly explain the reason for this observed result
and maybe suggest strategies to reverse the effect.
Under the current chemical conditions, however, regrinding of
lead concentrate should be discontinued.
Example 2 – circuit survey comparison using
nodal analysis
Lime is used periodically in the zinc cleaner scavenger flow sheet
(Figure A4) of the Eureka concentrate to depress pyrite. Debate
reigns as to whether this lime results in an improvement in
performance. To try and resolve the dispute between those
advocating lime addition and those against it, two circuit surveys
of the zinc cleaner scavenger flow sheet were performed – one
with lime on and one with lime off. Batch laboratory tests were
performed to measure the floatability of the feed, concentrate and

TABLE A1
Mineral grade of samples produced from batch flotation testing of a lead concentrate before and after grinding.

Sample Time Before regrinding After regrinding

(min) Galena Sphalerite Pyrite NSG Galena Sphalerite Pyrite NSG
Concentrate 1 0 - 0.5 76.71 14.08 4.02 5.20 72.80 14.86 6.77 5.58
Concentrate 2 0.5 - 1.5 68.66 20.80 5.73 4.81 68.54 17.67 7.66 6.13
Concentrate 3 1.5 - 4 48.42 31.44 9.12 11.03 57.27 25.34 9.52 7.87
Concentrate 4 4-8 25.88 45.83 16.35 11.95 28.29 44.10 15.60 12.00
Concentrate 5 8 - 12 18.17 50.83 21.40 9.60 15.99 50.99 19.02 14.01
Concentrate 6 12 - 20 12.54 51.77 23.49 12.21 12.19 50.05 20.80 16.96
Tailing 4.72 53.02 21.25 21.01 5.75 45.20 22.84 26.21

Flotation Plant Optimisation Spectrum Series 16 165

CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

70 70
Sphalerite Recovery (%)

60 60 Rougher Concentrate

Pyrite Recovery (%)

Rougher Concentrate
50 50 Reground Rougher
Reground Rougher Concentrate
40 40
Concentrate
30 30
20 20
10 10
0 0
0 20 40 60 80 100 0 20 40 60 80 100
Galena Recovery (%) Galena Recovery (%)

60
Rougher Concentrate
50
NSG Recovery (%)

Reground Rougher
40
Concentrate
30

20

10

0
0 20 40 60 80 100
Galena Recovery (%)

FIG A2 - Galena versus gangue mineral selectivity before and after regrinding of the lead rougher concentrate.

60 100
Galena Recovery (%)

80
Pyrite Recovery (%)

40 Rougher Concentrate
60
Rougher Concentrate Reground Rougher
40 Concentrate
20 Reground Rougher
Concentrate 20

0 0
0 5 10 15 20 25 0 5 10 15 20 25

Time (min) Time (min)

FIG A3 - Mineral recovery versus time before and after regrinding of the lead rougher concentrate.

final tailing of the circuit. Regrinding and staged reagent addition
is also used within this section to optimise the zinc gangue
selectivity in the circuit.
Results from the survey indicate that, although similar grade is
achieved, zinc recovery is significantly lower in the survey
performed when lime is added to the circuit (Table A2). This
survey information alone fails to resolve the dispute. Those
advocating lime addition suggest that the feed floatability and
how the circuit was operated was probably very different during
the two surveys.
Table A3 shows the cumulative mineral recovery data
generated from the feed, concentrate and tailing sample batch
laboratory flotation tests.
The feed mineral recovery rate curves of the various minerals
can be compared to assess whether feed floatability is different
sphalerite, galena and non-sulfide mineral recovery rates are very
similar in the feed batch flotation tests. The pyrite recovery rate is
low and thus sphalerite/pyrite selectivity of the feed is superior in
the lime survey (Figure A5). It can therefore be concluded that
the low recoveries in the lime survey are not due to poorer feed
ore floatability.
Lime is supposed to improve sphalerite/pyrite selectivity by
depressing pyrite in preference to sphalerite. The change in
sphalerite/pyrite selectivity in the two circuit surveys can be
assessed using nodal analysis techniques. The circuit is a node
with a single feed and two product streams. The two product
stream results can be combined using Equation A1 to create an
estimate of the combined product floatability.
RConcentrate
× RConcentrate, t
+ RTail
Mineral j × R Mineral j
Tail, t

=
Mineral j Mineral j
RCombined
Mineral j
Product, t
(A1)
between the two surveys. A review of Table A3 shows that the 100

166 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

Activator Collector
To Final Tail
Stage 1 Sump

Lime Activator
Regrind

Collector
Stage 2

To Final Concentrate
Sump

Cleaning Columns

FIG A4 - Eureka zinc cleaner scavenging circuit.

the sphalerite/pyrite selectivity in the feed and product streams of

100 this particular circuit. Sphalerite/pyrite selectivity in the survey
when lime is not added, improves dramatically across the circuit. It
Pyrite Recovery (%)

Survey - No Lime Added

80 can therefore be concluded that regrinding and the reagent addition
Survey - Lime Added
60
during this survey have had a positive effect on the ability to
separate sphalerite and pyrite. In contrast, sphalerite, pyrite
40 selectivity deteriorates in the survey in which lime is added.
20
This deterioration in sphalerite/pyrite selectivity is due to a
decrease in the sphalerite recovery rate (in contrast to an increase
0 when lime is not added to the circuit) and a small but similar
40 50 60 70 80 90 100 decrease in the recovery of pyrite (Figure A7).
Sphalerite Recovery (%) Use of nodal analysis in this example has shown that lime
addition in the circuit does not result in appreciable improvements
in sphalerite/pyrite selectivity. The deterioration in valuable to
FIG A5 - Sphalerite/pyrite selectivity measured in the feed batch gangue mineral selectivity will have had a significant impact on
tests performed during the two circuit surveys. the ability to separate minerals in this circuit. Adjustments in
circuit operation would have occurred to maintain grade but at a
where: cost of losing zinc recovery.
As a confirmation of this conclusion, nodal analysis was also
RConcentrate
Mineral j is the recovery of mineral j to concentrate performed before and after lime addition in this circuit (Runge
Tail
R Mineral j is the recovery to tailing of mineral j et al, 2004). These tests show that lime addition dramatically
decreases the floatability of all minerals, but sphalerite
RCombined
Mineral j
Product, t
is the cumulative recovery of mineral j in the in particular.
combined product after t minutes of flotation
It thus can be concluded that lime addition is clearly not the
The results of this calculation are shown in Table A4. solution to improving pyrite/selectivity in this circuit. Batch
Once the combined product floatability has been calculated, feed laboratory testing performed in conjunction with survey work
and product mineral recoveries can be compared. Figure A6 shows enabled a more definitive analysis of circuit operation.

TABLE A2
Metallurgical performance achieved during each survey.

Details of survey Concentrate grade (%) Recovery to concentrate (%)

Galena Sphalerite Pyrite NSG Galena Sphalerite Pyrite NSG
No lime added 6.2 79.7 7.0 7.2 57.8 86.2 11.4 5.5
Lime added 5.0 79.0 8.0 8.0 33.0 61.5 8.4 4.2

Flotation Plant Optimisation Spectrum Series 16 167

CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

TABLE A3
Cumulative mineral recovery achieved in the feed, concentrate and tailing batch tests performed during the two circuit surveys.

Cumulative time Mineral cumulative recovery in feed batch test

(min) No lime addition Lime addition
Galena Sphalerite Pyrite NSG Galena Sphalerite Pyrite NSG
0.33 22.5 33.3 10.8 7.5 23.2 36.4 9.7 10.3
1 49.9 63.4 28.2 19.8 49.0 66.6 22.2 21.8
2 66.8 79.1 43.3 29.9 65.7 80.7 33.5 31.8
4 82.4 89.2 58.0 40.9 79.3 90.4 48.3 42.0
8 89.5 94.3 69.6 51.1 87.7 94.8 62.1 52.2
12 92.4 96.1 75.5 57.6 91.2 96.5 69.7 58.9

Cumulative time Mineral cumulative recovery in concentrate batch test

(min) No lime addition Lime addition
Galena Sphalerite Pyrite NSG Galena Sphalerite Pyrite NSG
0.33 32.2 37.5 18.9 16.6 19.3 22.9 15.0 16.6
1 68.6 74.9 42.1 38.2 46.8 53.8 37.7 36.9
2 86.0 90.4 59.3 52.7 65.2 73.2 52.5 52.3
4 94.7 97.3 76.2 68.7 80.4 87.8 67.6 67.7
8 98.0 99.1 89.0 81.1 91.1 96.0 81.5 83.0
12 98.7 99.5 91.8 83.8 94.6 98.0 87.4 88.5

Cumulative time Mineral cumulative recovery in tailing batch test

(min) No lime addition Lime addition
Galena Sphalerite Pyrite NSG Galena Sphalerite Pyrite NSG
0.33 13.5 24.2 8.4 6.2 13.8 18.7 7.6 5.9
1 31.7 53.0 23.0 11.6 33.9 41.2 19.0 13.8
2 46.8 68.4 35.9 17.5 49.5 56.0 29.9 20.0
4 61.0 77.1 48.1 25.7 62.9 67.0 40.7 25.7
8 73.0 84.1 62.6 35.0 75.2 75.9 53.8 34.2
12 77.4 86.0 67.2 40.1 80.2 79.3 60.6 39.7

TABLE A4
Cumulative mineral recovery of the combined product streams from the two circuit surveys.

Cumulative time Mineral cumulative recovery in combined product

(min) No lime addition Lime addition
Galena Sphalerite Pyrite NSG Galena Sphalerite Pyrite NSG
0.33 24.3 35.7 9.6 6.8 15.6 21.2 8.2 6.3
1 53.0 71.9 25.2 13.1 38.2 49.0 20.6 14.7
2 69.5 87.4 38.6 19.4 54.7 66.6 31.8 21.4
4 80.5 94.5 51.3 28.1 68.7 79.8 43.0 27.5
8 87.4 97.0 65.6 37.5 80.4 88.2 56.2 36.3
12 89.7 97.6 70.0 42.5 85.0 90.8 62.9 41.8

Example 3 – ore floatability parameter modelling
of laboratory batch test data
There is a concern that a future ore to be fed to the Eureka
concentrator will be more difficult to treat in the lead circuit than
the current ore. Liberation analysis has indicated it has a more
complex lead mineralogy.
A multiple stage batch laboratory flotation test was performed
using the future ore to enable its ore floatability parameters to be
determined and compared to those of the current ore. Grind size
and reagent dosage rates were similar to those used in the plant.
The test involved floating a sample of the future ore for eight
minutes with two samples of concentrate collected – one from the
first two minutes (denoted rougher concentrate) and one from the
two to eight minute time period (denoted scavenger concentrate).
The rougher concentrate was then refloated in a cleaner stage
where concentrates were collected from zero to 20 seconds, 20 to
40 seconds, 40 to 60 seconds and from one to two minutes. The
scavenger concentrate was then combined with the tailing from
the cleaner stage and refloated in a cleaner scavenger stage for six
minutes. Concentrates were collected from zero to 0.5 minutes,
0.5 to one minute, one to two minutes and two to six minutes. A
diagram depicting the stages of the test and samples produced is
shown in Figure A8.

168 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

Survey - No Lime Added Survey - Lime Added

100 100

Pyrite Recovery (%)

Pyrite Recovery (%)

80 80
Feed Feed
60 Combined Product 60 Combined Product

40 40

20 20

0 0
40 50 60 70 80 90 100 40 50 60 70 80 90 100
Sphalerite Recovery (%) Sphalerite Recovery (%)

FIG A6 - Sphalerite/pyrite selectivity measured in the feed and product of the two circuit surveys.

100 100
Sphalerite Recovery (%)

Pyrite Recovery (%)

80 80

60 Feed No Lime 60
Feed Lime
Feed No Lime
40 Product No Lime 40
Feed Lime
Product Lime
20 20 Product No Lime
Product Lime
0 0
0 5 10 15 0 5 10 15
Time (min) Time (min)

FIG A7 - Sphalerite and pyrite recovery rates measured in the feed and product of the two circuit surveys.

Rougher Scavenger TABLE A5

(2 minutes) (6 minutes)
New Feed Tailing Multiple stage batch test results.

Cleaner Clner Scav Sample Solids % Lead % Gangue

(2 minutes) (6 minutes) (g)
Cleaner
Cleaner concentrate 1 13.4 75.9 24.1
Tailing
Cleaner concentrate 2 10.0 74.0 26.0
Cleaner concentrate 3 7.5 71.9 28.1
Concentrates
Cleaner concentrate 4 13.5 67.3 32.7
FIG A8 - Representation of multiple stage batch laboratory flotation Cleaner scavenger concentrate 1 9.2 55.8 44.2
test – samples produced denoted by a red circle.
Cleaner scavenger concentrate 2 6.7 50.5 49.5
Cleaner scavenger concentrate 3 8.9 42.8 57.2
The samples produced from the test were dried and weighed
and analysed for lead content – the results of which are shown in Cleaner scavenger concentrate 4 14.5 25.7 74.3
Table A5. Cleaner scavenger tailing 19.6 7.2 92.8
It is decided to represent the ore floatability of the system by Tailing 2096.7 0.9 99.1
three lead components (fast, slow and non-floating lead) and Recalculated feed 2200.0 3.1 96.9
three gangue components (fast, slow and non-floating gangue).
Three components are the maximum number that could be
derived with confidence from the experimental data available. The calculation is performed in three tables. The first table
Gangue is a term to refer to the weight in each sample that is contains the parameters to be derived from the fitting exercise.
‘not lead’. To determine the flotation rate and proportion of lead The values first input into the table are an estimate. Note that the
and gangue in each of the floatability components, an Excel rate and mass fraction of the non-floating component are not
spreadsheet was set up to calculate the weight and assay of the fitted parameters. The rate of the non-floating component is zero
experimental samples based on a flotation model. Solver, an and the proportion of material in the non-floating component is
add-in tool to Excel, was then used to determine the ore one minus the fraction in the other components.
floatability parameters that minimise the difference between the The second table calculates the recovery of each component in
measured and calculated sample information. each stage of the batch test (Rs). Recovery is a function of the
Table A6, Table A7 and Table A8 show the calculation tables flotation rate of the component (ki) and the time of flotation (t)
that were set up in Excel to perform the calculations. (Equation A2).
Table A9 shows the calculation of the weight of lead in each
batch test sample based on any set of lead floatability parameters. Rs = 1 − exp ( − ki t ) (A2)

Flotation Plant Optimisation Spectrum Series 16 169

CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

TABLE A6
Lead flotation model Excel spreadsheet.

Parameter table Fast Slow Non

Lead flotation rates (min-1) 0.932 0.201 0.000
Proportion of lead in feed in each component 0.642 0.099 0.258

Stage recovery matrix Time (min) Calculated recovery in stage

Fast Slow Non
Rougher 2 84.5% 33.1% 0.0%
Scavenger 6 99.6% 70.1% 0.0%
Cleaner 1 0.333 26.7% 6.5% 0.0%
Cleaner 2 0.333 26.7% 6.5% 0.0%
Cleaner 3 0.333 26.7% 6.5% 0.0%
Cleaner 4 1 60.6% 18.2% 0.0%
Cleaner scavenger 1 0.5 37.3% 9.6% 0.0%
Cleaner scavenger 2 0.5 37.3% 9.6% 0.0%
Cleaner scavenger 3 1 60.6% 18.2% 0.0%
Cleaner scavenger 4 4 97.6% 55.3% 0.0%

Component distribution matrix Weight of component in sample

Batch test sample Fast Slow Non Total
Feed 43.84 6.78 17.64 68.26
Rougher concentrate 37.05 2.25 0.00 39.29
Scavenger concentrate 6.77 3.18 0.00 9.95
Tailing 0.03 1.35 17.64 19.02

Cleaner feed (rougher concentrate) 37.05 2.25 0.00 39.29

Cleaner concentrate 1 9.89 0.15 0.00 10.03
Cleaner concentrate 2 7.25 0.14 0.00 7.39
Cleaner concentrate 3 5.31 0.13 0.00 5.44
Cleaner concentrate 4 8.85 0.34 0.00 9.19
Cleaner tail 5.75 1.50 0.00 7.25

Cleaner scavenger feed (SC + CT) 12.51 4.68 0.00 17.19

Cleaner scavenger concentrate 1 4.66 0.45 0.00 5.11
Cleaner scavenger concentrate 2 2.93 0.41 0.00 3.33
Cleaner scavenger concentrate 3 2.99 0.70 0.00 3.68
Cleaner scavenger concentrate 4 1.89 1.73 0.00 3.62
Cleaner scavenger tailing 0.05 1.40 0.00 1.44

Combined cleaner concentrate 31.30 0.74 0.00 32.05

Combined cleaner scavenger concentrate 12.47 3.28 0.00 15.75
Total concentrate 43.77 4.03 0.00 47.79

Total tailings 0.07 2.75 17.64 20.47

In the final table, the weight of each component in each batch
test sample throughout the duration of the test is calculated. The
weight in each component in the test feed is established by
multiplying the total weight of lead in the feed (an input) by the
proportion of lead in each component as specified in the first
table. The weight in the concentrate of each stage is determined
by multiplying the weight in the feed to the stage by the stage
recovery (Rs). The weight in the tail of each stage is the feed
minus the concentrate weight. These calculations are performed
in sequence, with either the concentrate or tailing (or both in the
case of the cleaner scavenger feed) of previous stages becoming
the feed to subsequent stages.
The total weight of lead reporting to each concentrate or tailing
stream in the batch test is then calculated by summing the
weights of its floatability components.
The identical calculation sequence is performed to determine
the mass of gangue in each of the batch test samples according to
a set of gangue specific floatability components (Table A7).

170 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

TABLE A7
Gangue flotation model Excel spreadsheet.

Parameter table Fast Slow Non

Gangue flotation rates 0.654 0.099 0.000
Proportion of gangue in feed in each component 0.009 0.030 0.961

Stage recovery matrix Time (min) Calculated recovery in stage

Fast Slow Non
Rougher 2 73.0% 17.9% 0.0%
Scavenger 6 98.0% 44.7% 0.0%
Cleaner 1 0.333 19.6% 3.2% 0.0%
Cleaner 2 0.333 19.6% 3.2% 0.0%
Cleaner 3 0.333 19.6% 3.2% 0.0%
Cleaner 4 1 48.0% 9.4% 0.0%
Cleaner scavenger 1 0.5 27.9% 4.8% 0.0%
Cleaner scavenger 2 0.5 27.9% 4.8% 0.0%
Cleaner scavenger 3 1 48.0% 9.4% 0.0%
Cleaner scavenger 4 4 92.7% 32.6% 0.0%

Component distribution matrix Weight of component in sample

Batch test sample Fast Slow Non Total
Feed 19.70 63.15 2050.89 2133.74
Rougher concentrate 14.37 11.32 0.00 25.70
Scavenger concentrate 5.22 23.18 0.00 28.40
Tailing 0.11 28.65 2050.89 2079.65

Cleaner feed (rougher concentrate) 14.37 11.32 0.00 25.70

Cleaner concentrate 1 2.81 0.37 0.00 3.18
Cleaner concentrate 2 2.26 0.35 0.00 2.62
Cleaner concentrate 3 1.82 0.34 0.00 2.16
Cleaner concentrate 4 3.59 0.96 0.00 4.55
Cleaner tail 3.89 9.29 0.00 13.18

Cleaner scavenger feed (SC + CT) 9.11 32.47 0.00 41.58

Cleaner scavenger concentrate 1 2.54 1.56 0.00 4.11
Cleaner scavenger concentrate 2 1.83 1.49 0.00 3.32
Cleaner scavenger concentrate 3 2.27 2.77 0.00 5.04
Cleaner scavenger concentrate 4 2.28 8.70 0.00 10.98
Cleaner scavenger tailing 0.18 17.95 0.00 18.13

Combined cleaner concentrate 10.49 2.03 0.00 12.51

Combined cleaner scavenger concentrate 8.93 14.52 0.00 23.45
Total concentrate 19.42 16.55 0.00 35.96

Total tailings 0.29 46.60 2050.89 2097.78

The weight of lead and gangue calculated in each stream can be SD = 1 if assay > 9% (A3)
used to calculate the total weight of solids and the lead and gangue SD = assay / 10 + 0.1 if assay < 9%
assay of each batch test sample. This calculated assay is then
compared to the actual assay measured during the test (Table A8). The squared error associated with each experimental datapoint
To calculate the total sum of squares error associated with the is calculated using Equation (A4). The total sum of squares error
system, the error of each experimental datapoint must be estimated. is the sum of the squared errors.
For this exercise, the Whiten function is used to estimate the ( d calculated − d experimental )2
standard deviation of the assays (Equation A3) and a relative error SE = 2
(A4)
of 2.5 per cent is used as the error of the solid weights. SD experimental

Flotation Plant Optimisation Spectrum Series 16 171

CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

TABLE A8
Calculation of the sum of squares error associated with the model fit.

Calculated data Experimental data Standard deviation Squared error

Batch test sample Assay Solids Assay Solids Assay Solids Assay Solids
Lead Gangue (g) Lead Gangue (g) Lead Gangue (g) Lead Gangue (g)

Feed 3.10 96.90 2202.00

Rougher concentrate 60.46 39.54 64.99
Scavenger concentrate 25.94 74.06 38.34
Tailing 0.91 99.09 2098.67 0.85 99.13 2096.7 0.19 1.00 52.4 0.09 0.00 0.00

Cleaner feed (rougher concentrate) 60.46 39.54 64.99

Cleaner concentrate 1 75.94 24.06 13.21 76.7 24.09 13.4 1.00 1.00 0.3 0.58 0.00 0.22
Cleaner concentrate 2 73.84 26.16 10.00 73.1 25.97 10.0 1.00 1.00 0.3 0.54 0.04 0.00
Cleaner concentrate 3 71.56 28.44 7.60 71.0 28.09 7.5 1.00 1.00 0.2 0.31 0.13 0.10
Cleaner concentrate 4 66.85 33.15 13.74 65.0 32.67 13.5 1.00 1.00 0.3 3.43 0.23 0.36
Cleaner tail 35.48 64.52 20.43

Cleaner scavenger feed (SC + CT) 29.25 70.75 58.77

Cleaner scavenger concentrate 1 55.45 44.55 9.22 58.3 44.16 9.2 1.00 1.00 0.2 8.10 0.15 0.03
Cleaner scavenger concentrate 2 50.07 49.93 6.65 48.2 49.51 6.7 1.00 1.00 0.2 3.49 0.18 0.02
Cleaner scavenger concentrate 3 42.23 57.77 8.72 41.1 57.17 8.9 1.00 1.00 0.2 1.28 0.36 0.39
Cleaner scavenger concentrate 4 24.81 75.19 14.60 24.0 74.35 14.5 1.00 1.00 0.4 0.65 0.71 0.06
Cleaner scavenger tailing 7.38 92.62 19.57 7.5 92.79 19.6 0.82 1.00 0.5 0.02 0.03 0.00

Combined cleaner concentrate 71.92 28.08 44.56

Combined cleaner scavenger 40.18 59.82 39.20
concentrate
Total concentrate 57.06 42.94 83.76

Total tailings 0.97 99.03 2118.24

Sum of squares error 21.52

TABLE A9
Stage batch test ore floatability results.

Parameter Future ore Current ore

Fast Slow Non Fast Slow Non
Lead flotation rates 0.93 0.20 0.0 0.99 0.16 0.0
Proportion of lead in each component in feed 0.64 0.10 0.26 0.71 0.08 0.21
Gangue flotation rates 0.65 0.10 0.0 0.62 0.07 0.0
Proportion of gangue in each component in feed 0.009 0.03 0.96 0.007 0.02 0.97

Once the spreadsheet has been set up in the way described
above, Solver (an add-in tool in Excel) can be used to determine
the lead and gangue parameters which minimise the total sum of
squares error. For this example, Solver is able to find a set of
parameters which result in a good fit to the experimental data.
In Table A9, the parameters derived from the analysis above
are compared to those derived from a similar test performed using
the ore currently processed by the plant.
The flotation rates of the different components in the two
different ores are similar. The ore does have poorer floating
characteristics than the current ore. The proportion of floatable
lead is significantly lower and the proportion of floatable gangue
has increased. This is an indication that the future ore is less
liberated at the grind size tested.
The flotation model that was developed to determine the ore
parameters can also be used to simulate various different batch
test configurations – predicting the change in grade and recovery
that would be produced if the test was run in a different way.
To use the model, the parameters of the ore are input into the
parameter table and the time of each stage is input into the
stage recovery matrix. The lead grade and lead recovery of
the combined concentrate (cleaner and cleaner scavenger
concentrate) is calculated by the model. Using the parameters
derived for the current ore and the future ore, the model was used
to predict the combined concentrate lead grade and recoveries
that would be achieved if different times were used in the various
stages of the batch test. Table A10 show the different simulations
that were performed and the results. Figure A9 shows a graphical
comparison of the results derived for the two ores.

172 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 9 – LABORATORY FLOTATION TESTING – AN ESSENTIAL TOOL FOR ORE CHARACTERISATION

TABLE A10
Multiple stage batch test simulation results.

Model simulation Batch test stage time Future ore Current ore
Rougher Scavenger Cleaner Cleaner Lead grade Lead Lead grade Lead
scavenger recovery recovery
Base case 2.00 6.00 2.00 6.00 57.06 70.02 70.44 74.72
Simulation 1 2.00 6.00 0.50 1.50 66.69 55.63 77.69 61.67
Simulation 2 4.00 12.00 2.00 6.00 52.54 71.41 66.97 75.92
Simulation 3 6.00 18.00 2.00 6.00 50.52 71.78 65.05 76.32
Simulation 4 12.00 24.00 2.00 6.00 48.91 72.04 63.26 76.61
Simulation 5 12.00 24.00 4.00 12.00 43.01 73.71 57.32 78.28
Simulation 6 12.00 24.00 1.00 3.00 55.64 68.08 69.04 73.31
Simulation 7 6.00 18.00 1.00 3.00 57.03 67.83 70.34 73.08
Simulation 8 2.00 6.00 1.00 3.00 62.21 66.21 74.16 71.71
Simulation 9 2.00 6.00 0.50 1.50 66.69 55.63 77.69 61.67
Simulation 10 1.00 3.00 0.50 1.50 69.76 52.34 79.80 58.69
Simulation 11 1.00 3.00 1.00 3.00 66.11 62.97 76.87 69.06
Simulation 12 1.00 3.00 2.00 6.00 62.02 66.60 73.98 72.02

90
The ore floatability of the future ore, as measured in the batch
80 flotation test, is clearly poorer than that currently treated. The
Lead Recovery (%)

70 lead recovery achievable at a particular grade is ten per cent lower

60 in the future ore simulations than in the current ore simulations.
50 Future Ore Increased recoveries may be possible at a finer grind. Additional
40 Current Ore test work would be required to substantiate this hypothesis.
30
20
10
0
0 20 40 60 80 100
Lead Grade (%)

FIG A9 - Simulated batch test lead grade and recovery of the

future and current ore.

Flotation Plant Optimisation Spectrum Series 16 173

 HOME

CHAPTER 10

Designing and Analysing Plant Trials

Tim Napier-Munn

FAusIMM(CP), Julius Kruttschnitt Mineral Research Centre, The University of Queensland, 40 Isles
Road, Indooroopilly Qld 4068. Email: t.napier-munn@uq.edu.au

Professor Tim Napier-Munn is a metallurgist who has spent nearly 40 years in the practice and
management of applied R&D in mineral processing and related disciplines. He has Bachelors and PhD
degrees from Imperial College, London, and a Masters degree from the University of the Witwatersrand in
Johannesburg. He had an 11-year career with De Beers in South Africa and taught mineral processing (and
statistics) at Imperial College for three years. He joined the Julius Kruttschnitt Mineral Research Centre at
the University of Queensland in 1985. He was appointed JKMRC Research Director in 1992, Director in
1997 and the inaugural Managing Director of JKTech Pty Ltd in 2001. He retired from the latter two roles in
2004, and now divides his time between working with the JKMRC, consulting, teaching and writing a book
on statistics. In 2009 he gave his course on Statistics for Metallurgists for the 100th time.

Abstract
Introduction
The Problem Statement
Error and How to Manage it
Comparing Means
Comparing Several Conditions
Factorial Designs
Modelling Your Way Out
Changes in Process Mean – The Cusum Chart
Evolutionary Operation
Running Plant Trials
Which is the Best Method?
References
Further Reading
Appendix – Critical Values for the Outlier Test

ABSTRACT eliminating outliers. If a formal design cannot be implemented, or it is

This chapter deals with the principles of running plant trials. The purpose
of a plant trial is (usually) to determine whether a change in some process
condition will lead to an improvement in performance. Detecting the
presence and magnitude of any change in process performance is difficult
against the inevitable background of process noise (error) if the correct
procedures are not followed. The source and management of error is
discussed, and then several desirable experimental methods are described,
including the paired t-test, the analysis of variance (ANOVA) and the
randomised block design. Other designs discussed briefly include the
factorial experiment and evolutionary operation (EVOP). A formula is
given for the number of trials required for a t-test, and a procedure for
desired to analyse historical data, then modelling is the appropriate method,
and several approaches are described. Cusum charts are recommended as a
powerful way of examining data collected in time sequence. The chapter
finishes with a consideration of some of the practical aspects of doing a
plant trial.
There really is no alternative to the adoption of a formal experimental
design or, if all else fails, a rigorous analysis of the data from some other
experimental activity. Experience repeatedly shows that a systematic
approach to conducting plant trials will produce the correct answer in the
minimum time and at minimum cost. Any other approach is very likely to
take much longer and cost more to produce the wrong answer, or at best to
produce no answer at all.

Flotation Plant Optimisation Spectrum Series 16 175

CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS
INTRODUCTION
The great statistician R A Fisher, whose work and philosophy
underpin many of the methods described in this chapter, once said:
To call in the statistician after the experiment is
done may be no more than asking him to perform
a post-mortem examination. He may be able to
say what the experiment died of.
Fisher was a brilliant mathematician who worked for 20 years
at the famed Rothamsted Agricultural Research Station in
England, and the experimental methods which he developed for
Rothamsted’s scientists were practical, effective and have stood
the test of time. The problems which agricultural scientists face
are very similar to those encountered by metallurgists charged
with a policy of ‘continuous improvement’ and the consequent
testing of new reagents, equipment or process circuits. The data
are very noisy and the experiments often difficult to do.
This chapter is a plea to follow Fisher’s implicit advice. You
don’t necessarily have to call in a statistician, but you do have to
plan the experiments properly. Many laboratory experiments and
plant trials are conducted in the naïve view that hard work and
enthusiasm are adequate substitutes for planning, rigour and
thought. The protagonists throw themselves into an experiment in
which hope triumphs over method, and the answer is
optimistically expected to emerge from the fog like advice from
the Oracle. As a consequence they often lead to no clear result or,
worse still, the wrong result. The irony is that the right way to do
things is often cheaper than the wrong way, and leads to the
correct result more quickly. There is therefore no real excuse for
not doing a good experiment.
In the past the Eureka concentrator was no exception. Figure 1
shows the daily recovery from a plant trial conducted some years
ago to test the efficacy of an expensive new reagent. The average
zinc recovery with the new reagent was found to be seven per
cent more than with the old. This is an astonishing improvement
and duly excited the metallurgical staff no end.
90
85
80
75
Zn Recovery (%)
70
65
60
Reagent 1 Reagent 2
55 R = 70% R = 77%
50
45
40
0 20 40 Day 60 80
FIG 1 - A plant trial to test a new reagent.
Would you, dear reader, looking at this simple graph of the
recovery time series, support the view that Eureka should invest
in the new reagent? I hope not. There is good visual evidence that
the recovery was climbing steadily even before the new reagent
was introduced, and in fact one could take the view that the
improvement was actually impeded by the new reagent, even
though the average recovery under the new condition was much
higher than that under the old.
176
The key point to take from this all-too familiar scenario is that,
in the absence of other information, there is no way in which any
change in metallurgical performance can be unequivocally
attributed to the new experimental condition. This is the
fundamental flaw in such ‘before-and-after’ experiments, and it
cannot be avoided, except by doing some special analysis which
is discussed further below. A much better option is to adopt a
simple experimental protocol to deal with the problem illustrated
in Figure 1 and to use the corresponding methods of data
analysis, rather than suffer the indignity of Fisher’s wrath and,
more importantly, the waste of time and money (and sometimes
credibility) which failure brings. The good news is that
appropriate protocols and analysis methods exist which are
relatively easy to implement and so maximise the chance of a
successful outcome. And once the design is chosen, the data
analysis is straightforward.
This chapter describes the most accessible experimental
designs and data analysis procedures. In most cases the analysis
can be carried out using Excel’s excellent statistical toolbox, and
reference is made to the appropriate routines. The methods can be
applied to laboratory experiments, pilot plant tests and full-scale
plant trials, and no distinction is made between these other than
some discussion at the end of the chapter about the particular
practical problems of conducting plant trials. The quality system
Six Sigma has much in common with the techniques described.
This chapter is not a textbook on statistics. It is intended to
introduce metallurgists to the use of good practice in
experimental design and analysis to assist process optimisation.
Discussion is confined to principles, and demonstration of the use
of Excel in some of the more widely-used methods. For a more
comprehensive discussion and the underlying mathematics, the
interested reader is referred to references and further reading.
THE PROBLEM STATEMENT
Mineral processing experiments are usually carried out to
determine whether there is a benefit in introducing some change
in process operating conditions, or to identify and quantify the
effect of key process variables on metallurgical performance.
Examples are:
• Testing a new reagent in a flotation plant.
• Testing an alternative piece of equipment, or equipment
setting.
• Testing a new circuit configuration.
• Answering the question: does feed rate affect recovery?
• Answering the question: are two parallel circuits performing
the same way?
Other examples of the need for appropriate experimentation and
data analysis include the assessment of sampling or assay methods,
and the calibration of instruments including OSA/ISA systems.
As one of the greatest experimenters of all, Robert Hooke,
famously said in 1683:
The object of experiment is improvement, and an
experiment should lead to something beyond itself.
In the metallurgist’s world this process often starts in the
laboratory with experiments to test a new idea, or a new reagent
or piece of equipment recommended by a vendor. If these
experiments are promising, the next step is often a full plant trial
to test whether the benefit can be realised in practice.
A plant trial is usually the only way of determining with
confidence whether a process benefit really exists; improvements
seen at the laboratory or pilot scales often evaporate when
Spectrum Series 16 Flotation Plant Optimisation

implemented at the full production scale. However, plant trials are
expensive to conduct, and can interfere with normal plant
operation. They may even reduce plant performance for a while.
It is therefore very important to conduct the trial in such a way
that a firm conclusion is reached (one way or the other) in the
minimum time. The problem, as in all experiments, is to
understand and deal with the basic experimental error. In plant
trials, as often as not, this is apparent mainly as a variation of
performance with time. Figure 2 shows the actual daily variation
in recovery from historical plant data. A one per cent recovery
improvement may be worth several million dollars a year in a
large operation, but how is one to detect it in data which varies
over a 25 per cent range? The answer is: easily, if you do the
right experiment.
85
80
Metal Recovery (%)
75 70% ?
70
65 69%
60
55
0 10 20 30 40 50 60
Day Number
FIG 2 - Daily variation in recovery.
ERROR AND HOW TO MANAGE IT
Accuracy and precision
Every data point has uncertainty associated with it. This is true
whether it is an assay, a feed rate, or a value calculated from other
data using a formula such as estimating recovery or yield from a
two-product formula. This uncertainty is given the term error
(electronics engineers call it noise). We hope that the error is
random, but systematic error due to mistakes, incorrect sampling,
miscalibration of instruments, etc can cause bias in any result.
Random and systematic errors are reflected in the terms precision
and accuracy respectively, which mean different things. Precision
implies the natural variation in the measurements we make,
which we hope to minimise by being careful but will always be
there. Accuracy implies getting the right answer. Precision is
measured by the standard deviation (or its square, the variance),
and its management in various forms is the basis of the statistical
methods discussed in this chapter.
In experiments of the kind to be discussed, error can generally
only be estimated by replicating1 the measurement or experiment,
ie by doing it more than once under the same conditions. The
formula for the standard deviation is:
s=
∑(x i − x )2
(1)
n −1
1. Strictly speaking one should make a distinction between repeats,
which are repeats of a particular measurement, and replications,
which are repetitions of experiments or parts of experiments.
Similarly there is a distinction between repeatability, which for
example measures the variability of a particular assay method within
a particular laboratory, and reproducibility which measures the
variability between laboratories.
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS
where:
xi = data point i
x = the mean or average
n = number of data points
(n-1) is called the degrees of freedom
(The units of s are the same as those of the original data.)
For n = 1, s is indeterminate. A single result therefore carries
no information about precision, and, one might argue, in those
cases where little if anything is known about the experimental
system being studied, no information of any kind. So it is
important to get into the habit of performing replicates. A plant
trial run over many days or weeks (or even months) produces
many replicates. Laboratory tests can also be done with many
replicates. In assessing the value of a particular quantity (recovery
in a laboratory float test for example) replication directly reduces
variance, thus making real differences between conditions (eg
reagents) easier to detect against the background noise of
experimental error.
Where does error come from?
Sampling
Every sample taken, whether of ore, slurry, solution, concentrate or
any other product, has inherent uncertainty associated with it.
Samples should be taken according to the proper sampling
protocols. Gy’s theory and the associated methodologies are an
important guide.
Laboratory practice
Variations in operator, machines, batch of reagent, method,
ambient conditions and many other sources will contribute to error.
Assay
The baseline of error is nearly always an assay: chemical,
mineralogical or particle size. The repeatability of assay methods
should be well characterised by the assay chemist or metallurgist.
Metallurgical accounting and mass balancing
Metallurgical accounting and the associated mass or material
balancing should be carried out according to correct and rigorous
protocols. This includes the use of redundant data (more data than
the minimum needed to close the balance) to get the best estimate
in a least squares sense of all important quantities, including
assays, mass flows and computed quantities such as grades and
recoveries. Rigorous mass balancing will require the entry of
estimates of the standard deviation of all measured quantities to
ensure the optimum weighted balance. Chapter 3 discusses these
issues in more detail.
The plant
The plant is home to the biggest variabilities of all, usually driven
by feed changes over time, but also by poor feed distribution,
shift variations, parallel circuit differences, ambient conditions,
reagent and many other factors. Figure 2 is a good example of the
trends, cycles and random noise evident in concentrator recovery
over a two month period. Figure 3 shows a consistent bias in
recovery between two parallel flotation circuits; Line 1 operating
at about 1.3 per cent higher recovery than Line 2, and Figure 4
shows a statistically significant negative impact of feed rate on
recovery.
177
CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS

95 symbol H0) to be true unless the data ‘prove’ otherwise. The null
hypothesis can be any quantitative relation you wish to test, but
90 by far the commonest is that there is no difference between the
quantities being compared. Thus for two mean values X 1 and X 2 ,
Cu Recovery (%)

85
the null hypothesis H0 is that X 1 − X 2 = 0. The alternative
80 hypothesis, H1, is that the difference is not zero, ie there is a real
difference between the two means: X 1 − X 2 ≠ 0. 2 We will only
75 accept H1 if there is a sufficiently low risk in doing so, and it is the
70 great power and ingenuity of these statistical methods that allows
Line 1 the risk to be quantified and rational decisions made on that basis.
65 Line 2 Statistical tests of significance generate a probability, P, which
60 is interpreted as the risk of being wrong in rejecting the null
hypothesis, ie concluding that there is a difference when in fact
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Day
there isn’t, which statisticians call a Type I error. 100(1 - P) is the
confidence level of the test. So if the test returns a P-value of
0.05, we infer that there is a five per cent chance of being wrong
FIG 3 - Differences in parallel flotation circuit performance.
in concluding there is a difference, or, conversely, we are 95 per
cent confident that there is a difference. P will always exceed zero
so there is always a risk of being wrong – the only certain things
100
in life are death and taxes!
A question which always arises is: what level of risk (or
Cu recovery (%)

95 conversely, confidence) should we choose as a hurdle rate for

90
R = -0.31
85
80
200 250 300 350 400 450
Feed Rate (t/h)
FIG 4 - A negative impact of feed rate on recovery.
All these effects conspire to increase the variability of plant
data, making it essential to adopt the correct experimental design
procedures to be able to detect differences (improvements) that
are in general small compared to the background noise.
Hypothesis testing and risk
Of course we work to limit the variability of our data, by good
experimental practice and careful measurement. But there will
always be a residual error that gets in the way of comparing one
condition against another. We therefore need a rigorous way of
comparing values that takes into account the magnitude of the
uncertainty in those values. Hypothesis testing or significance
testing is the approach used for this purpose.
The hypothesis test is rather like a court case in which the
accused is always presumed innocent until proven guilty by the
evidence. The accused does not have to prove themself innocent;
rather the prosecution has to prove them guilty. This is a very
important distinction, and one which is followed in the hypothesis
test. We generally assume the null hypothesis (denoted by the
2. Strictly speaking, we distinguish between H1: X 1 − X 2 ≠ 0, and H1:
X 1 − X 2 > 0, the first being a ‘two-sided’ test and the second a
‘one-sided test’. More of this later.
3. If the Data Analysis option is not visible, go to: Tools > Add-Ins, and
check the Analysis ToolPak box.
4. Excel also provides a routine for unequal variances. ‘Unequal’ means
statistically different (usually determined using an F-test). In most
cases variances can be considered ‘equal’ even though they may look
different. Using the ‘unequal’ routine reduces the confidence in any
observed difference.
making decisions? This will depend on the consequences of the
decision. The lower the value of P we choose the less chance we
have of making a mistake but also the less positive decisions we
will make. If we were doing a reagent trial in the plant and the
new reagent was much more expensive than the old, we would
need to be more certain of a difference before recommending the
change than if there were no real downside to the decision.
If you have nothing else to go on, the commonest hurdle
rate for P used by most experimenters is 0.05, following
recommendations originally given by Fisher in 1926. However,
there is no magic bullet. It would probably be unwise to choose P
to be any more than 0.1 as this gives a one in ten chance of being
wrong. If you were in a casino with a 90 per cent chance of
success in each game, you would never leave! But here we are
dealing with a different situation: you want to minimise the
chance of failure; in fact you would prefer never to fail, but as we
know P > 0, we want to limit the chance of failure as much as
possible, especially if the decision turns out to be career-limiting.
Now we will consider a number of different methods of
making decisions about differences in quantities of interest. All
are based on the principle of the hypothesis test.
COMPARING MEANS
The two-sample t-test
This tests the difference between the averages (means) of two
samples of data, which need not be of the same size. An example
is the testing of two different ore sources or two different reagents
using a standard laboratory float test. Example 1 (Table 1) shows
a test of a new zinc collector at Eureka, in which a single sample
of ore was ground, split ten ways, and four float tests done with
the old collector (A) and six with the new (B). The null
hypothesis is that the difference in means is zero. Excel’s
two-sample t-test is invoked as follows: Tools > Data Analysis3 >
t-test: Two-Sample Assuming Equal Variances4 and complete the
dialogue box. The result is as shown in Table 1 (data on the left,
Excel output on the right).
The important numbers to interpret are means and variances,
hypothesised mean difference and P(T< = t).
Means and variances
Excel does not compute the difference in means, which is the key
piece of information (it is this value which is being tested using
the t-test), but this is easily done.

178 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS

TABLE 1 level. For example, if a 90 per cent two-sided confidence interval

Example 1 – two-sample t-test for laboratory flotation test.
is required, α is 0.10.
In example 1, CI95 per cent = ±1.28 per cent, ie the difference in
t-Test: Two sample assuming equal variances mean recoveries is 1.15 ± 1.28 with 95 per cent confidence. There
A B is no contradiction in the fact that this interval includes zero, even
Mean 86.5 87.65 though our t-test has told us that we are more than 95 per cent
confident that the difference is not zero. The t-test was a
Variance 0.366667 0.971
one-sided test and the confidence limits here are two-sided.
Observations 4 6 In fact it can often be useful to form a one-sided confidence
Pooled variance 0.744375 limit. This is done by using 12 α in Equation 2 and the Excel
A B Hypothesized mean difference 0 function TINV. For Example 1, the one-sided 90 per cent
85.8 86.5 df 8 confidence limit is 0.78, and we could interpret this as saying: we
are 90 per cent certain that the recovery improvement does not
87.1 89.0 t stat -2.06494
exceed 1.15 + 0.78 = 1.93 per cent. If a cost-benefit analysis
86.2 87.2 P(T< = t) one-tail 0.0364 requires the difference to be at least two per cent to get a return
86.9 86.8 t critical one-tail 1.859548 on the increased cost, then one might be justified in not
88.5 P(T< = t) two-tail 0.072799 proceeding with the new collector even though it was shown to be
87.9 t critical two-tail 2.306004
better than the old.
The two-sample t-test is not always appropriate for plant trials
because it implies that the data have been not been collected
under one of the protocols suggested below. For example, they
Hypothesised mean difference may have been collected in the ‘before and after’ fashion
The default value in the dialogue box is zero (ie H0 is that there is illustrated in Figure 1. If so, even though the t-test may show a
no difference between the collectors), but it could be any value real difference, there is no guarantee that the difference is due to
the experimenter wishes to enter in the dialogue box. In most the variable being investigated rather than some other factor
cases zero is appropriate. which was not controlled by the experimental design.

P(T< = t) The paired t-test

P(T< = t) is the chance (risk) of a Type I error, ie saying there is a
difference when in fact there isn’t one. A low P means that the
calculated value of t is unlikely to have arisen by chance and that
the difference is therefore statistically significant. Whether it is
one- or two-sided (‘tail’) will depend on how the objective of the
experiment has been defined. In general, if we are seeking a
change in a particular direction (eg an improvement) it is
one-sided. If the question is ‘are the means different?’ (in either
direction) it is two-sided. As can be seen from the Excel output, a
two-sided problem has a higher risk and lower confidence than a
one-sided problem.
In this experiment the comparison is one-sided as we are
seeking an improvement. P = 0.036, ie there is a 3.6 per cent
chance of being wrong in rejecting H0, or alternatively, we are
96.4 per cent confident that the difference is not zero (and our
best estimate of the difference is 87.65 - 86.50 = 1.15 per cent). Is
this enough? This is up to the experimenter and will depend upon
the consequences of the decision, but in most cases if P < 0.05 we
would conclude that the difference is real.
A useful statistic to calculate is the confidence interval, CI, for
the difference in means, given by:
1 1
CI = tα s + (2)
n1 n2
where:
tα = t-value for a two-sided P-value of α
s = pooled variance (from the Excel output)
n1 = number of results in sample 1
n2 = number of results in sample 2
tα can be obtained from tables of the t-distribution, or from the
Excel function TINV(α, df) where df = degrees of freedom = n1 +
n2 - 2, and α is the two-sided P-value for the chosen confidence
Pairing is a form of blocking to deal with common variability, and
is always to be preferred if the experiment can be set up to
achieve it. The main advantage of the paired t-test is that the
standard deviation due to error is greatly reduced because much
of it is essentially removed by taking the difference of two
performance values in which common variation is subtracted out.
The test is therefore more sensitive and smaller differences can be
detected. It is also easier to attribute any observed difference
unequivocally to the factor being investigated.
Paired values must be alike in all senses other than the factor
being studied. In the assay laboratory, an example would be a
comparison of two assay instruments in which a number of
samples, which could be of different origin, are each divided in
two and one aliquot tested in each instrument. In plant trials
pairing can be achieved in two ways:
1. using parallel circuits or machines; and
2. sequential pairs in time, ie on-off (more of this later).
Where parallel circuits are available, each circuit can be run on
one condition. The difference in daily performance (eg recovery)
is then calculated and the paired t-test applied daily until a
conclusion is reached. The problem is that parallel circuits are
rarely identical; a common problem is biased feed. Any
difference between the circuits will bias the difference it is
desired to detect. The best way to deal with this is to switch the
treatment on a random basis (Excel has a random number
generator, but tossing a coin would do just as well) (see Table 2).
The t-test is then applied to the differences. The test can be
done each day and the experiment terminated when a conclusion
is reached or when the prescribed number of pairs has been
completed (see below). The significance of t can then be
determined from the Excel routine by inspecting P, exactly as for
the two-sample t-test.
The confidence limits on the mean difference d are given by:
d ± t α sd / n (3)

Flotation Plant Optimisation Spectrum Series 16 179

CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS

TABLE 2 Number of trials needed

Typical arrangement for paired trial in parallel circuits. It is very important to have some idea of how long an experiment
is likely to take, ie how many data points must be collected to
Day Circuit A Circuit B Difference
achieve the purpose. The number of trials can be calculated, given
1 T N T-N the difference it is required to detect and the risks which the
2 N T T-N experimenter wishes to take.
3 T N T-N However the risks are in fact of two kinds. We have already
4 N T T-N encountered the Type I error which is the risk of saying that there
has been a change in mean when in fact no such change has
5 T N T-N occurred, ie rejecting the null hypothesis incorrectly. This is
6 etc etc etc controlled by choosing an appropriately small value of P, which
for the Type I error we will now call α.
T = trial condition, N = normal condition It is easy to see that there exists another potential trap into
which the unwary experimenter could fall, the error of asserting
that no change has occurred when in fact there is a real
where: difference; this is called a Type II error. In trials to achieve
sd = standard deviation of the differences process improvement, the Type II error is an opportunity lost.
Simply doing a t-test controls only Type I errors. When the
n = number of pairs chosen significance level is exceeded, the experimenter is
tα has the same meaning as in Equation 2 justified in concluding that a change has occurred, ie that the null
hypothesis is false. However, when no significant difference is
found he is not equally justified in asserting that the null
Example 2 hypothesis is true. This can only be done by conducting sufficient
Two flotation circuits are operated in parallel. Each day, one experiments to control Type II errors at some chosen confidence
circuit is run with the usual collector (A) and the other with a new level, β.
collector (B). The order is switched randomly between the The decision options and their respective errors are illustrated
circuits each day to avoid any systematic differences between the in the ‘truth table’ (Table 4). The test (ie decision) is made at a
circuits. The experiment is run for 12 days, and the daily recovery probability level of α (one- or two-sided).
is monitored. It is required to determine whether the new The total number of experiments required in a t-test to detect a
collector gives a significantly improved recovery. The results difference as large as D with the Types I and II errors controlled
were as shown in Table 3. at confidence levels of a and b respectively is (for large samples):
The mean difference is 0.88 per cent. The question (as always)
is: is that difference real, or has it arisen by chance due to 2( z α + z β ) 2 s2
n= (round n up to a whole number) (4)
experimental error? Excel provides a routine to do the analysis: D2
Tools > Data Analysis > t-Test: Paired Two Sample for Means.
Again this is a one-tailed test as we are seeking an improvement.
TABLE 4
P(T< = t) = 0.075. Thus we are 92.5 per cent confident that the
Truth table for decision-making.
difference is not zero. Is this sufficient confidence? If the new
collector were more expensive, we may wish to improve our Decision Truth about the population
confidence by doing more trials (see below for a formula for the H0 true H0 false
number of trials needed). (no difference) (true difference)
The power of the paired test is illustrated by using Excel’s Reject null hypothesis H0 Type I error Correct decision
two-sample t-test to analyse the same data. P(T< = t) = 0.31 (not P =α P = 1 -β
significant) and we would have to accept the null hypothesis and Accept null hypothesis H0 Correct decision Type II error
thus erroneously conclude that Collector B was no better than A. P = 1 -α P =β

TABLE 3
Example 2 – paired t-test for plant flotation trial.

Day Collector A Colllector B t-test: paired two sample for means

1 84.2 85.8 A B
2 71.9 75.0 Mean 79.03333 78.15
3 77.8 76.6 Variance 22.06788 16.88818
4 80.1 82.5 Observations 12 12
5 75.6 77.7 Pearson correlation 0.90725
6 70.8 70.0 Hypothesised mean difference 0
7 81.3 85.0 df 11
8 82.6 80.6 t stat 1.544139
9 78.5 77.1 P(T< = t) one-tail 0.07541
10 79.2 81.5 t critical one-tail 1.795885
11 80.6 82.2 P(T< = t) two-tail 0.15082
12 75.2 74.4 t critical two-tail 2.200985

180 Spectrum Series 16 Flotation Plant Optimisation


where:
zα = normal deviate at a confidence level of α (one or
two-sided)
zβ = normal deviate at a confidence level of β (always
one-sided)
s = expected standard deviation
n is the size of each sample for the two-sample t-test, and the
number of pairs for the paired t-test. For small samples n should
be multiplied by the ratio (f + 3)/(f + 1) where f is the degrees of
freedom: f = 2n - 2 for the two-sample t-test and n - 1 for the
paired t-test. zα and zβ are the standard normal deviates for
probability levels α and β. These can be obtained from tables
of the normal distribution, or from the Excel function
NORMSINV(1-P) where P is α or β one-sided (use (1 - P)/2 for
two-sided α), or from Table 5. Note that for the paired t-test, the
leading two in Equation 4 is omitted, and n = number of pairs.
After carrying out n experiments, the t-test is then conducted at
a confidence level of α (taking the usual account of whether the
test is one or two-sided). If P(t) ≤ α, then the alternative
hypothesis is accepted (there is a real difference) with a
probability α of being wrong (Type I error). If P(t) > α, then the
null hypothesis is accepted (no real difference of size D) with a
probability β of being wrong (Type II error). 1 - β, the probability
of correctly concluding that there is a difference, is called the
power of the test.
Some textbooks provide power curves (Mason, Gunst and
Hess, 1989; Montgomery and Runger, 1999) or tables (Davies,
1967), allowing the experimenter to read off the number of trials
required from an assumption of α, β and the expected
signal-to-noise ratio D/s.
The balance of α and β is important. Because the number of
experiments needed to control both types of error can be large,
and increases as α and β decrease, β is sometimes chosen to be
somewhat larger than α to limit the number of tests. Thus if we
choose α = 0.05 (95 per cent confidence level) then we might
choose β = 0.2 (80 per cent confidence level), since it is usually
more important to avoid concluding that there is a real difference
when there really is not (Type I error, controlled by α) than to
assert no difference when one really exists (Type II error,
controlled by β). In trials of new operating conditions, for
example, this gives more weight to retaining the status quo.
Example 3
Equation 4 can be used to determine the number of paired trials
to be conducted in the flotation trial of Example 2. s is here the
standard deviation of the paired differences, sd.5 It is desired to
control the risk of incorrectly concluding that there is a process
improvement when there is not at five per cent (α = 0.05) and to
5. If the sd of the paired differences is not known from previous trials,
then an estimate is s √2, where s is the sd of the raw data (eg daily
recovery). This assumes no benefit from subtracting out common
variation and is therefore conservative. In mineral processing plant
trials sd can often be substantially less than this.
Tail
0.3 0.2
One-sided 0.524 0.842
Two-sided 0.842 1.282
Flotation Plant Optimisation
CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS
have a 90 per cent chance of detecting a recovery improvement as
large as one per cent – the power of the test (β = 0.1). From
previous experiments it is believed that sd ≈ 2. How many pairs of
tests should be conducted? From Table 1, one-sided z0.05 = 1.645,
one-sided z0.10 = 1.282. Substituting directly in Equation 4 with
D = 1 and s = 2, and with the correction for degrees of freedom,
gives n = 37 pairs.
It is easy to use Equation 4 to construct general curves relating
the number of trials required to the signal-to-noise ratio D/s for a
range of confidence levels for one-sided and two-sided tests.
Figure 5 shows such curves, from which it is clear that as the
signal-to-noise ratio declines (either the error increases or the
difference required to detect it decreases, or both) the number of
trials required increases rapidly.
FIG 5 - Number of trials required as a function of expected
signal-to-noise ratio.
The reference distribution
One of the problems in using parametric tests such as the t-test
regression analysis (see below) is that these methods make
assumptions about the independence of the data and the normality
of residuals that are not always fulfilled in mineral processing.
The non-parametric Reference Distribution method overcomes
this problem as it makes no assumptions as to random sampling
or data independence (Box, Hunter and Hunter, 1978; Cavender,
1993; Bruey and Briggs, 1997; Napier-Munn, 2008).
The method is based on the assumption that a body of
consistent data is available representing the condition of the
process before the trial starts (the ‘reference distribution’). The
characteristics of the trial data are then compared with those of
the reference distribution to determine the probability with which
they are likely to have arisen, assuming there has been no change
in the process. If that probability is small (eg ≤ 0.05), then it is
concluded that a real change has occurred. So essentially it is
another significance test. However, in this case the probabilities
are measured directly from the historical data.
Suppose that the trial period gives us ten results, with a mean
recovery three per cent more than the mean of the last ten results
immediately before the change. The question, as ever, is ‘is this
TABLE 5
Values of z.
α or β
0.1 0.05 0.025 0.01
1.282 1.645 1.960 2.326
1.645 1.960 2.326 2.576
Spectrum Series 16 181
CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS
difference significant (real) or has it arisen by chance?’. This
can be re-stated rather more formally as ‘what is the probability
with which such a difference would arise, if no change had
occurred?’. This is answered directly by interrogating the
reference distribution.
Suppose that the RD comprises 100 sequential data points. We
can then look at the 100 - 10 + 1 = 91 sequential sets of ten
results (1 - 10, 2 - 11, 3 - 12, etc) and see how many times one set
of ten (eg 1 - 10 or 2 - 11) is followed by another set of ten (eg
11 - 20 or 12 - 21) with a mean recovery higher by three per cent.
If this occurs rarely in the RD (eg ≤5 per cent of times), then the
trial result must be regarded as significant.
The problem with this method is the key assumption that the
reference distribution is stable and represents the constant
condition of the process before the change. This is an assumption
which is often not met – the process mean moves around and it is
difficult to define what is the appropriate RD. The cusum chart
(see below) can help to determine whether the RD is
homogenous. Also, it cannot help with the nuisance influence of
covariates such as feed grade. But it is free of the statistical
assumptions required by many of the other methods.
Outliers
Outliers may be cynically regarded as inconvenient points, ‘wild’
results that do not appear to belong to the rest of the data set or
perhaps conflict with our prejudices. They are best identified by
plotting the data. They can often have a significant effect on
hypothesis tests, mainly due to the substantial increase in
standard deviation which they cause. As a general rule, however,
outliers should not be rejected without some formal justification.
For example, if it is found that a mistake was made in collecting
the data, such as a biased sample, or a miscalibrated instrument,
then the relevant point(s) can be rejected. If, however, no such
reason can be found but suspicion persists that they do not
belong, a formal test can be invoked.
Several such tests have been devised, but there is no consensus
amongst the statistical community and certainly no magic bullet.
Mason, Gunst and Hess (1989) recommend the Grubbs test for a
normally distributed sample whose mean is of interest. This test
is more powerful than some others and can reject more than one
outlier at a time. The test statistic is a ratio of sums of squares,
( x i − x )2 , the SS of the data without the outlier(s) being divided
by the SS of all the data with the outlier(s), giving a ratio less
than one. The column of data should first be sorted using the
Excel Data > Sort option. The outlier(s) will be at the top or
TABLE 6
Example 4 – the one-way ANOVA in comparing operators.
Op 1 Op 2 Op 3 Op 4 ANOVA: Single factor
67.3 60.5 63.5 60.7
66.2 58.9 64.9 64.7 Summary
63.3 54.1 66.4 67.3 Groups
64.6 64.3 Op 1
Op 2
Op 3
Op 4
ANOVA
Source of variation
Between groups
Within groups
Total
182 Spectrum Series 16
bottom of the column depending on whether they are the highest
or lowest values and how the sort has been done. The Excel
function DEVSQ (array) is then used to calculate the two sums of
squares, with and without the outliers. The ratio of these is then
compared with Table A26 in Mason, Gunst and Hess (1989) with
n = total number of data points and k = number of outliers. (The
values in the table have been multiplied by 103 to avoid the
confusion of decimal points.) If the calculated ratio is less than
the tabulated value at the desired α-value then the outlier(s)
should be rejected.
COMPARING SEVERAL CONDITIONS
The one-way analysis of variance (ANOVA)
The analysis of variance is a powerful procedure for partitioning
the sources of variance in an experiment, and judging their
significance by comparing them against an estimate of error
variance. There are several versions, used in many different
applications.
The one-way ANOVA is in a sense a multiple t-test; in fact it is
identical to the two-sample t-test for two means. Its strength,
however, lies in its capacity to consider several sources of
variation (several means) simultaneously. It is useful in
comparing multiple ore sources in future ores testing, or perhaps
a number of different collectors on a single ore. The ‘one-way’
implies that there is only one source of variation other than
experimental error. Several repeat tests are conducted on each
condition being investigated (eg each ore source or collector).
Example 4
In a large feasibility study, several operators in different
laboratories are involved in laboratory float tests. It is important
to ensure that the operators do not produce statistically different
results. To establish whether they all give similar results, a single
ore sample is therefore divided and each of four operators
conducts a number of repeat tests on the sample. The results are
shown on the left side of Table 6. On the right is the Excel output
invoked by Tools > Data Analysis > Anova: Single Factor. The
full block of data is selected, including the labels in the first row,
and the ‘Labels’ box in the dialogue box is checked.
The summary gives the number of tests for each operator, and
the corresponding mean recoveries and variances. The ANOVA
table provides the key information to judge whether there are
differences between the operators (see Table 6). This is done by
comparing the mean squares (MS) of the ‘Between Groups’ and
Count Sum Average Variance
4 261.42 65.35393 3.222062
3 173.45 57.81624 10.82865
3 194.78 64.92636 1.993553
4 256.97 64.24125 7.376275
SS df MS F P-value F crit
118.557 3 39.51887 6.880095 0.00855 3.7083
57.4394 10 5.743942
175.996 13
Flotation Plant Optimisation
 CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS

‘Within Groups’ variation, which have units of variance. The
‘Between Groups’ variance is due to real differences between the
operators, and the ‘Within Groups’ variance is experimental error,
measured from the repeats done by each operator. The
comparison is made by forming the ratio of the two MS values,
which is called the F-ratio (after Fisher who invented it). The
P-value then gives the Type I error risk. In this case P = 0.009, ie
there is less than one per cent risk in saying that the operators do
indeed give different results, or conversely, we are more than
99 per cent sure that they are different.
A glance at the summary shows that Operator 2 produced a
lower mean recovery than the other three, and therefore some
mentoring or re-training might be in order.
A useful piece of additional information is the MS for ‘Within
Groups’, which is the experimental error variance based on the
repeats. The square root of this value (2.4 per cent) is the error
standard deviation which can be used in planning future
experiments, eg in Equation 4.
Randomised block design – the two-way ANOVA
The purpose of this experimental design is to eliminate (or rather
separately identify) the effect of an uncontrolled nuisance
variable, such as feed condition or parallel circuit, by blocking
this variable. For a simple unreplicated two-condition experiment
this is the same design as the paired t-test, and analysis by an
ANOVA will give the same answer. However, it is more
appropriate where three or more conditions are being compared.
Time can be thought of as a proxy for feed condition, and the
variable blocked in a concentrator trial is therefore often time,
though it can also be circuit in plants which have parallel circuits.
In the time case, the trial is divided up into time blocks of
appropriate duration and each condition tested at least once in
each block, the order being randomised within the block.
Replication within a block is desirable to improve the estimate of
experimental error, though it is not essential. If the conditions to
be tested are A, B and C, the experiment might look like this:
ABC BAC CBA BCA ACB
⎯ ⎯ ⎯ ⎯ ⎯ ⎯⎯→ Time
Each block might be three days, so each condition is tested for
a day, in random order.
Analysis is by a two-way ANOVA, either unreplicated or
replicated, both of which are available in Excel. Note the
important design criteria that each treatment is tested at least once
in each block in random order. If these criteria are not followed,
the ANOVA analysis is invalid. Randomisation can be done using
Excel’s routine ‘Random Number Generation’.
Example 5
A trial is conducted on a concentrator with three parallel closed
grinding circuits. The purpose is to increase the fineness of grind
by using an alternative cyclone geometry (a combination of
vortex finder and apex). After doing simulations and consulting
the manufacturer, three likely geometries are selected for test,
plus the current arrangement. It is thought likely that the circuits
do not operate exactly the same but it is important that the
conclusions apply to all the circuits. A randomised block design
is therefore run, with ‘circuit’ the blocked variable and each of the
four geometries being tested in random order once in each
circuit. A composite sample of cyclone overflow is collected for
each circuit/geometry combination and the per cent -75 μm
determined.
The results are shown on the left side of Table 7. On the right is
the Excel output invoked by Tools > Data Analysis > Anova:
Single Factor. The full block of data is selected, including the
labels in the first row and column, and the ‘Labels’ box in the
dialogue box is checked.
Now there are three sources of variation: the geometries (rows),
the circuits (columns) and error. The null hypotheses are that the
four geometries give the same product size, and the three circuits
also give the same product size. The P-values for the F-tests show
that both hypotheses can be rejected: the geometries and the
circuits are significantly different. The geometry effect is
significant at >95 per cent confidence (P = 0.043). We can
therefore choose the best cyclone with the finest product (from
the summary – Cyclone C).
The circuit effect is significant at 93.6 per cent (P = 0.064),
which is probably sufficient in a trial of this kind to conclude that

TABLE 7
Example 5 – the two-way ANOVA in comparing cyclone geometries.

Cyclone Circuit 1 Circuit 2 Circuit 3 ANOVA: Two-factor without replication

A 73.8 74.2 69.0
B 68.9 75.5 65.2 Summary Count Sum Average Variance
C 79.2 78.5 73.8 A 3 217 72.33333 8.373333
D 72.5 69.6 70.0 B 3 209.6 69.86667 27.22333
C 3 231.5 77.16667 8.623333
D 3 212.1 70.7 2.47

Circuit 1 4 294.4 73.6 18.23333

Circuit 2 4 297.8 74.45 13.69667
Circuit 3 4 278 69.5 12.49333

ANOVA
Source of variation SS df MS F P-value F crit
Rows 95.9367 3 31.97889 5.139464 0.042715 4.75706
Columns 56.0467 2 28.02333 4.50375 0.063904 5.14325
Error 37.3333 6 6.222222
Total 189.317 11

Flotation Plant Optimisation Spectrum Series 16 183

CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS
the difference is real, as suspected. Inspection of the summary
shows that Circuit 3 is producing a coarser product than the other
two. This bonus information can be used to explore why this is so
and perhaps correct Circuit 3 so that an additional process
improvement can be achieved.
The square root of the error MS (2.5 per cent -75 μm) is the
experimental standard deviation for the trial.
Note that if the experiment had not been blocked, ie not
arranged as a formal randomised block design, the block variation
(columns) would not have been distinguishable from error. The
consequence of this can be simulated by combining the block and
error sums of squares (93.38), dividing by the sum of their
degrees of freedom (8) to give a new estimate of the error MS
(11.67) and then dividing this into the geometry MS to give a new
F-value (2.74). The significance of this can be tested using the
Excel function FDIST(F, df1, df2) where df1 = degrees of freedom
for the geometry effect and df2 = degrees of freedom for error.
This returns P(F) = 0.13, and we would have erroneously
concluded that the geometry had no significant effect.
FACTORIAL DESIGNS
The factorial is a widely-used and powerful experimental design,
and is comprehensively covered in the literature (eg Davies,
1967; Box, Hunter and Hunter, 1978; Mason, Gunst and Hess,
1989). It is the basis of the EVOP process optimisation method
discussed briefly below, but is most widely used in the laboratory
environment. It will therefore not be dealt with in detail here; the
reader is referred to the literature.
The key feature of a full factorial is that each factor (variable)
is investigated at every level of every other factor, ie a
measurement is made at every factor combination. The number of
experiments needed is cxn, where x is the number of levels, n is
the number of factors and c is the number of replicates of the
whole experiment. If three factors are investigated at two levels
each without replication, the number of tests required is 23 = 8.
A 23 factorial can be thought of as a cube in space, with each of
the three dimensions representing one of the three factors, A, B
and C. A test is run at each of the eight corners of the cube (see
Figure 6).
FIG 6 - A 23 factorial experiment: three factors each at two levels.
The analysis falls into two parts: the calculation of the actual
effects (how each factor affects the response), and testing of the
significance of each effect using an ANOVA. Some estimate of
error is required for the ANOVA – either a full replication of the
experiment, or a number of centre point runs at the centre of the
experimental space (as shown in Figure 6), or some other source.
No specific routines are available in Excel for the analysis of
184 Spectrum Series 16
factorials, though a two-level factorial can be analysed by a
simple method suitable for spreadsheets, called Yates’ analysis
(Box, Hunter and Hunter, 1978; Mason, Gunst and Hess, 1989).
The two-level factorial only detects linear relationships between
factors. This is satisfactory in many cases. However, three or
more levels are required to detect non-linear effects. A specialist
statistics package is best used both for the design and analysis of
the factorial and its many variants.
The design is not limited to the same number of levels for each
factor. However, a disadvantage is that many tests are required if
the situation calls for several levels and/or several factors, the
number of tests increasing exponentially with the number of
factors. In such cases subsets of the full design can be used,
called fractional factorials. Special designs such as the central
composite rotatable design (CCRD) are also available. The
CCRD is able to investigate quadratic effects for the development
of response surface models in many less tests than would be
required for a full three-level factorial; it has been successfully
used in mineral process pilot plant experiments.
The factorial has important advantages over other less formal
approaches to experimentation, including:
• It is mathematically the most efficient way of estimating the
factor effects. This is particularly useful in collecting data
for modelling (see below).
• It is necessary to detect interaction, in which the effect of
one factor depends systematically upon the level of one or
more other factors.
• It can incorporate blocking and randomisation, which are
important weapons in dealing with error to improve the
ability of the experiment to detect important effects.
• The conclusions hold over a wide range of conditions.
MODELLING YOUR WAY OUT
A formal design such as the paired trial or randomised block
design is always to be preferred as it will generally provide a
robust conclusion in the shortest time. If, however, for whatever
reason, it is not possible to do a formal trial, then it is possible to
use modelling to assess the significance of any new condition
which is being investigated. Modelling is also helpful for historical
data not collected from a planned trial (Napier-Munn, 2008).
Dealing with the effect of an uncontrolled variable
– intervention analysis
It often happens that there is one particular variable which is
known to influence the performance criterion of interest, but over
which the experimenter has no control. A good example is the
way feed grade generally drives recovery. One way to deal with
such a variable is to use its influence to develop a simple model
of the performance variable of interest (eg recovery) in terms of
the nuisance variable (eg feed grade) and to add the presence or
absence of the trial condition (eg a new reagent) as a second
variable. This is called Intervention Analysis when used by
economists to analyse time series data (Box, Hunter and Hunter,
1978). (The method is essentially equivalent to the comparison of
two regression lines (trendlines in Excel-speak) described by
Napier-Munn, 1998.)
The model looks like this:
Y = a0 + a1X+ a2T + ε (5)
where:
Y = performance variable of interest, or dependent variable
(eg recovery)
Flotation Plant Optimisation
 CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS

X = uncontrolled variable (eg feed grade)
T = a dummy variable representing the trial condition: T = 0
for the old condition and T = 1 for the new condition6
a0 , a1 and a2 are coefficients to be estimated from the data
ε represents the error (noise)
X and T are independent variables. The null hypothesis here is
that the trial condition does not influence recovery, ie that a2 = 0.
Example 6
A plant trial was conducted to test a new flotation reagent. An
appropriate protocol (eg a paired trial) could not be conducted, so
data using the new reagent were compared with a historically
equivalent period when using the plant’s usual reagent.
Equation 5 is fitted to the data by invoking the Excel regression
tool: Tools > Data Analysis > Regression, which estimates the
linear parameters (coefficients) in the model using the least
squares method. The Y-range in this case is the column of
Recovery, and the X-range the contiguous block of feed grade
and trial condition, including the labels. Table 8 shows part of the
data set on the left, and the Excel output on the right (excluding
the 95 per cent confidence limits for clarity).
The key parts of the output can be interpreted as follows. The
significance in the ANOVA (refer Table 8) (1) is the P-value for a
test of whether the whole regression is significant. A value much
less than 0.01 is desirable, and certainly less than 0.05. The trial
coefficient (2) is an estimate of the magnitude of the effect which
the trial condition has had on recovery: in this case an increase of
1.48 per cent. (Note that this will not be equal to the difference in
the mean values, which in this case is 1.8 per cent, because it is
essentially a mean weighted towards the centre of gravity of the
data.) It is essential as always to test whether this is significant,
using the corresponding standard error (3) and t-value. This is
done as usual with the P-value (4), in this case 0.046, which is
6. If the trial variable has several quantifiable levels (eg collector
concentration) then of course it can be represented as a normal
quantitative variable.
significant at the 95 per cent level of confidence. (Similar
P-values are available for the grade coefficient (5) and intercept,
both of which are highly significant.)
Note that Excel returns a two-sided P-value for the coefficients.
If a one-sided value is required, as in this case (we are looking for
an improvement), then use the Excel function ‘= TDIST(t, df, 1)’.
t is the value of t for the trial coefficient (2.02 in this case), df is
the degrees of freedom for the residual (6) (109 in this case), and
‘1’ implies a one-sided test (use ‘2’ for a two-sided test). In this
case TDIST returns the value P(t) = 0.023; so we are more than
97 per cent confident that the new reagent gives a higher recovery
than the old.
The standard error for the trial coefficient (3) can also be used to
calculate one-sided or two-sided confidence limits, as before. The
95 per cent two-sided confidence interval is approximately 1.96 ×
standard error = ±1.4 per cent. So we are 95 per cent confident that
the trial conditions produces an increase in recovery, and our best
estimate of that increase is 1.5 per cent ±1.4 per cent.
Models with many variables
The intervention analysis approach described above can be
extended to many process variables.
Example 7
Historical data for 114 values of daily feed rate and feed grade
were available from two periods running with a different flotation
reagent. Trial = 0 indicates the period running with the old
reagent and Trial = 1 the period with the new (see Table 9).
Recovery is the dependent variable (Y) and feed rate, feed
grade and ‘trial’ (the dummy variable) are the independent
variables (X1, X2, T). Choosing Excel’s Tools > Data Analysis >
Regression will produce the basic output as shown in Table 10
(leaving out the ANOVA for clarity).
The P-value (two-sided) for the trial coefficient (1) is 0.310,
and the one-sided P-value calculated from the function TDIST is
0.155. Neither are statistically significant, and the main
conclusion of the trial is that the new reagent has no significant
effect on recovery.

TABLE 8
Example 6 – regression model of a plant trial with intervention analysis.

Feed grade Trial Recovery Summary output

0.364 0 66.9
0.320 0 63.7 Regression statistics
0.453 0 66.7 Multiple R 0.59745195
0.413 0 75.0 R square 0.35694883
0.617 0 76.8 Adjusted R square 0.34514973
0.503 0 82.9 Standard error 3.86107058
0.500 0 73.4 Observations 112
0.518 0 72.6
… … … ANOVA
0.560 1 68.8 df SS MS F Significance F
0.493 1 68.1 Regression 2 901.9914438 450.996 30.2522 3.54308E-11
0.590 1 75.8 Residual 109 6 1624.957399 14.9079 1
0.697 1 73.3 Total 111 2526.948843
0.560 1 73.6
0.570 1 74.6 Coefficient Standard error t stat P-value
0.520 1 66.0 Intercept 53.4035743 2.180237461 24.4944 4E-46
0.537 1 71.0 Grade 32.4636222 4.40265655 7.37364 3.4E-11 5
0.580 1 74.7 Trial 2 1.47612985 0.731053365 3 2.01918 0.04592 4

Flotation Plant Optimisation Spectrum Series 16 185

CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS

the standard deviation of the residuals, the difference between the

TABLE 9 observed and predicted values of recovery. 1.96 × standard error
Example 7 – data from plant trial. gives approximately the 95 per cent confidence interval of the
predictions of the model, in this case ±2.1 per cent.
Day Feed rate Feed grade Trial Recovery
(t/h) (%) (%) Regression models of this kind are helpful but users should be
1 354 1.63 1.00 90.1
aware of their limitations and dangers:
2 364 1.71 1.00 90.3 • They are simplistic models and should not be over-
interpreted. In particular they will not apply to different
3 343 1.62 1.00 90.5
process conditions and so should not be extrapolated.
4 324 1.75 1.00 91.8
• The model may be biased if terms (eg other significant
5 365 1.70 1.00 89.9 process variables) have been omitted.
6 373 1.39 1.00 87.6
• Outliers should be removed where appropriate, and the
… … … … … regression re-run. These can be identified by plotting
109 352 2.06 0.00 92.8 observed Y (on the y-axis) versus predicted Y (on the
110 353 1.87 0.00 91.4 x-axis) – this is a useful plot anyway, giving a visual
indication of the goodness-of-fit. The points should be
111 337 1.95 0.00 93.1
scattered around the 45° line, the closeness of the points to
112 343 2.05 0.00 93.3 the line indicating the quality of fit. Outliers will lie outside
113 328 1.93 0.00 93.0 the general cluster of points. A more rigorous way of
114 334 1.86 0.00 92.1 rejecting them is to select Standardized Residuals from the
Regression dialogue box. These are residuals divided by a
standard error and form a kind of t-test. Any value greater
TABLE 10 than about 2.5 can be rejected with 99 per cent (two-sided)
Multi-variable regression model of a plant trial with
confidence; though don’t forget that a large data set would
intervention analysis. be expected to have one per cent of its residuals of this value
or greater by chance alone.
Summary output
• The least squares method is based on a number of
assumptions. One of the most important is that the residuals
Regression statistics (predicted – actual values) are normally distributed; this can
Multiple R 0.81970252 be visually checked using a histogram (the Excel Regression
dialogue box permits you to produce a column of residuals
R square 0.67191222
and predicted values). Another is that the errors are
Adjusted R square 0.66296437 2 independent; this assumption is violated when the data are
Standard error 1.06188161 3 serially correlated, and the time relationship would need to
Observations 114 be accommodated in the model (see below).
Regression analysis is a big subject and interested readers should
Coefficient Standard error t Stat P-value consult appropriate texts for further details (eg Box and Draper,
1987; Box, Hunter and Hunter, 1978; Chatfield, 1983; Draper and
Intercept 88.5948393 1.478259002 59.932 8E-86
Smith, 1998; Montgomery and Runger, 1999; Snedecor and
t/h -0.0193845 0.003091221 -6.271 7.2E-09 Cochran, 1968).
Feed grade 5.49063873 0.439740137 12.486 8.1E-23
Trial -0.216325 0.212147581 -1.02 0.31011 Time series models and intervention analysis
1 Data plotted in time such as those shown in Figures 1 and 2 can
be modelled as a time series. These models exploit the fact that
such data are often ‘serially correlated’ – that is, each result is to
However, the model is quite interesting in its own right. The
some extent dependent on the previous results (like the weather!).
model itself is: The degree of dependence is measured by ‘autocorrelation’,
defined similarly to the correlation coefficient, R (the square root
Recovery (per cent) = 88.59 − 0.019 (feed rate) + (6) of R2 which is given by Excel’s ‘Add Trendline’). There is a
5.49 (feed grade) family of time series models known as Box-Jenkins models
which take this approach.
This simple process description gives us additional useful A simple TS model that has been found useful in mineral
information from the trial. An increase of 0.5 per cent in feed processing time series is the autoregressive model of order 1
grade increases recovery by about 2.7 per cent, and an increase of (AR(1)) which simply says that the current result is dependent to
50 t/h decreases recovery by about one per cent. some extent on the previous result (Napier-Munn and Meyer,
Excel provides statistics to judge the quality of the model. The 1999):
adjusted R2 (2) is the coefficient of multiple determination
(adjusted for the degrees of freedom), which is a goodness-of-fit yt = μ + φ(yt-1 - μ) + et (7)
criterion describing the proportion of the total data variation
accounted for by the model (the balance would be due to missing where:
terms, model inadequacies and error). R2adj = 66 per cent is high
yt = performance (eg recovery) in current time period (eg
for plant data of this kind in which values of less than 50 per cent
today)
are usual, and suggests that the model gives reasonable
predictions. This is confirmed by the standard error (3), which is yt-1 = performance in previous time period (eg yesterday)

186 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS

φ = autocorrelation coefficient (0 ≤ φ ≤ 1) which measures
how strongly yt depends on yt-1
μ = process mean: the value about which the process is
randomly oscillating
et = error for time t: a measure of the residual error after the
model is fitted
Equation 6 can be rearranged in the form of a linear expression
which can be easily fitted by Excel’s regression tool or ‘Add
Trendline’ in Chart.
y t = μ(1 − φ ) + φ( y t − 1 )
Here yt is the column of daily data and yt-1 is this column
displaced down by one cell. (The first yt value and the last yt-1
value must be eliminated to produce two columns with the same
number of data points.) a0 (the intercept of the fitted line) = μ(1 -
φ), and a1 (the gradient) = φ, and so μ = a0 / (1 - a1).
Now we invoke Equation 5 to perform intervention analysis on
the time series model:
Note that the concentrate grade (t - 1) coefficient is 0.45. This is
a measure of the dependency of each day’s concentrate grade (t) on
the previous day’s figure (t - 1), and is also highly significant. This
confirms the validity of using the AR(1) time series model.
CHANGES IN PROCESS MEAN –
THE CUSUM CHART
In analysing plant trials, it is always worth plotting the
performance data (eg recovery) as a simple time series, as shown
in Figure 1. This will often show trends, cycles or other features
that are not apparent in statistical tests. These features are helpful
(8)
in guiding the subsequent analysis of the data. Figure 7 shows the
performance of a precious metal concentrator over five months in
which an improvement in recovery and a reduction in variability
in the last month are clearly apparent; this was the result of
introducing a new control system.

yt = a0 + a1(yt-1) + a2T (9)

where again:
T is the presence or absence of the trial condition
The process shown in Example 6 is then followed.

Example 8
One type of experiment which is difficult to do on a plant scale is
FIG 7 - Time series of plant performance.
the testing of grinding media, which cannot be switched easily to
conform to a particular experimental protocol such as a paired
t-test. In this example a study was made of the effect of two types However, process changes are generally more subtle and less
of grinding media on final concentrate grade from the flotation apparent than this. In such cases a cumulative sum chart can often
plant. A homogeneous period using the first type of media was identify changes in process mean more clearly. A ‘cusum’ chart is
taken as representing that condition (T = 0). The new media was a time sequence plot of the cumulative sum of the current value
then progressively added to the mill and a second period of data minus some mean value plus the previous cusum. In symbols:
collected once the mill was fully charged with the new media
(T = 1). Three hundred and ten days of data were available in all. C t = C t − 1 + R1 − μ (10)
The data and Excel Regression output were as shown in Table 11.
The Media coefficient is -1.31 per cent and its two-sided where:
P-value 1.8E - 06, which is highly significant. We may therefore Ct = cusum at time t
conclude that moving from Media 0 to Media 1 reduced the
concentrate grade by 1.3 per cent. Ct-1 = cusum at time t - 1

TABLE 11
Example 8 – modelling a plant trial as a time series.

Media Conc grade (t-1) Conc grade (t) Summary output

0 34.34 33.27
0 33.27 32.86 Regression statistics
0 32.86 33.11 Multiple R 0.6092497
0 33.11 30.44 R square 0.3711852
0 30.44 29.99 Adjusted R square 0.3670887
… … … Standard error 2.0754444
1 28.50 28.39 Observations 310
1 28.39 28.79
1 28.79 28.50 Coefficient Standard error t stat P-value
1 28.50 27.23 Intercept 16.531569 1.535578721 10.766 3.83E-23
1 27.23 29.85 Media -1.312924 0.26970263 -4.868 1.81E-06
1 29.85 29.55 Conc grade (t-1) 0.4530328 0.050493461 8.9721 2.95E-17

Flotation Plant Optimisation Spectrum Series 16 187

CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS
Rt = value (eg recovery) at time t
μ = some mean or target value (often the process mean over
the period)
Figure 8 (left) shows a normal time series of mill P80 over an
18 month period from a gold plant. Apart from the usual noise,
there appear to be three distinct periods of prevailing P80, a low
leading to a high and then back to a low again. On the right is the
cusum series of the data, using a mean value μ = 174 μm (the
mean of the whole series). The points of change are now very
clear, and careful inspection of the chart shows that the changes
occurred on days 130 and 404. This information can now be used
forensically to explore the reason(s) for the change.
In cusum charts, the absolute value of the cusum at any point is
not important. It is the gradient of the line over a characteristic
period which indicates the prevailing process mean over that
period, and changes in gradient indicate changes in process
mean. A change in gradient can be from positive to more or less
positive, negative to more or less negative, or more dramatically
(as in Figure 8) from negative to positive or vice versa. If the
gradient over a given period is negative then the prevailing mean
is less than μ, if positive then the prevailing mean is more than μ,
and if flat then the prevailing mean is equal to μ.
In constructing a cusum chart, therefore, the choice of μ is
important, and it is usually worth trying several to see which
gives the most obvious changes in gradient. This is easy if
Equation 10 is implemented in a spreadsheet. Typical choices are
the total mean of the data (as in Figure 8, for which the cusum of
the final point will always be zero), some target value for the
process being sought, or the mean for the period before a known
change is made to a process variable.
In many cases, changes in gradient can be assessed by eye.
However, rigorous methods are available for determining if the
change in gradient at any given point is statistically significant
(Himmelblau, 1970; Montgomery and Runger, 1999).
EVOLUTIONARY OPERATION
Evolutionary Operation, or ‘EVOP’, has been widely used in
some process industries and has been advocated for mineral
processing (Mular, 1971; Mular and Bull, 1973) but never seems
to have caught on. It is not clear why this should be, as its
features suit the characteristics of concentrators rather well.
EVOP is a process optimisation procedure in which simple
factorial designs are used to explore and approach the optimum
performance in a noisy process. The range of each factorial is
small to prevent damaging interference to the process. The basic
experiment is replicated until effects become statistically
significant, and the design follows the contours or response
surface of the process until it detects a maximum. It is one of a
broad class of optimisation methods called hill climbing
300
250
P80 (micron)
200
150
100
50
0 -2000
0 100 200 300 400
Day
FIG 8 - Time series and cusum chart of daily mill P80.
188
techniques, and has a number of variants including SSDEVOP:
simplex self-directed EVOP. It is easy to implement and to
calculate, and is promoted as a method in which plant operators
can play an important role. It does, however, require a commitment
to the process over long operating periods. The method is covered
in detail in Mular and Bull (1973) and is also discussed in Box,
Hunter and Hunter (1978) and Himmelblau (1970).
RUNNING PLANT TRIALS
There are probably three main ways to conduct an effective full
plant trial:
1. in pairs of some kind, so that the paired t-test can be used to
eliminate the effect of any uncontrollable variable such as
feed conditions or parallel circuit;
2. if more than two conditions are to be compared, then use a
randomised block design, in which the effect of the
unwanted variable (eg time as a proxy for feed condition) is
removed by blocking; and
3. without a formal design, using models to assess any
differences.
Of these, the paired t-test is probably the most efficient if only
two conditions are being tested. The randomised block design is
appropriate where more than two conditions are being trialled.
Modelling using intervention analysis will generally require more
experiments than the other two, and is normally used only where
practicalities prevent the use of a formal design, or in the analysis
of historical data. Some comments on each follow.
Clearly it is no good running for a time (a day, week or month)
with one condition, then switching to the other condition for the
same period, and then comparing mean performance over the two
periods, as was done in the experiment illustrated in Figure 1.
Other factors may have changed with time which would interfere
with the comparison, most notably ore type or grade. The error
would also be high and uncontrolled. The paired trial or
randomised block design are intended to deal with this situation.
In both cases, it is necessary to choose a duration for each
replicate of the experiment, eg for each member of the pair.
Selection of an appropriate time period will depend on:
1. How the plant does its metallurgical accounting and
reconciliation.
2. Material residence time in the process (the period must be
selected to be long compared to mean residence time; at
least three residence times are required, and preferably
much more – five to ten).
3. The period of large-scale changes and trends in process
performance, such as those evident in Figure 2. The
replicate times should be short compared to this period.
3000
2000
Cusum of P80
1000
0
-1000
500 0 100 200 300 400 500
Day
Spectrum Series 16 Flotation Plant Optimisation

Common durations for a replicate are a shift or a day. A day is
preferred, as shift performance often varies systematically over
the 24-hour cycle, and a day is usually long compared with mean
residence time but short compared with long-term trends.
However, the longer the period, the longer (and more expensive)
is the total duration of the experiment.
The switching order in the trial should be randomised. A
problem arises if it is difficult to switch condition, eg when a
major item of equipment is to be included and excluded, although
this is often easier than it seems if a little thought is given to the
problem. In such cases the switch can be made less frequently but
the trial will have to proceed for longer to ensure a time bias does
not accumulate.
In some cases identical parallel circuits are not available. Here
the pairs are generated sequentially in time. It is often considered
that after switching a condition the plant requires a full period (eg
a day) to settle down, and the performance of the day following
any change should therefore be ignored. The order of each pair
should also be randomised. This produces a sequence of groups
of three or four days: four days when the first day of the group is
different to the last day of the previous group, and three days
where the first day is the same as the last day of the previous
group and no day is required for stabilisation. A typical sequence
is as shown in Table 12. The pairs used in the analysis are
underlined.
TABLE 12
A paired trial including stabilisation periods.
1 NNTT 2 NNTT 3 TNN 4 NTT
5 NNTT 6 TNN 7 NTT 8 NNTT
9 TNN 10 NTT 11 TNN 12 NTT
13 TNN 14 TTNN 15 NTT 16 NNTT etc
Groups numbered in sequence; N = normal and T = trial condition
If 16 ‘pairs’ were required to detect a difference in this case,
then the experiment would run for 55 days, assuming no breaks.
A checklist
In planning a plant trial (or any experiment), work through the
following checklist:
1. Define the objective(s) of the trial; write them down and be
specific. Why are you doing it? What will be your actions
depending on the outcome?
2. Formulate the objectives in experimental terms. What
experiments need to be conducted to achieve the objectives?
3. What response(s) and factors are to be selected for
investigation? What do we know about the random variation
of the response(s) – ie experimental error? Conduct pilot-
scale tests if necessary.
4. Over what range do we need to alter the factors? Are we
interested in knowing factor interaction? Is there any
unwanted or uncontrollable variable whose effect must be
eliminated?
5. Taking into account all these considerations, plus the time,
money and labour available, and the possible effect
on current operating processes, select an appropriate
experimental design.
WHICH IS THE BEST METHOD?
The most powerful designs for plant trials are the parametric
methods such as the paired t-test or randomised block design.
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS
‘Powerful’ means that they are more sensitive and will detect
smaller differences faster than other methods. The only
restrictions on their use are:
1. practical – the process cannot accommodate the design, and
2. statistical – the assumptions of randomness and independence
are violated.
Problem 1 can often be managed with a little lateral thinking.
Problem 2 can also be managed, as the t-test in particular is very
robust to departures from these assumptions, and experience
shows that the data pairs (as distinct from the original data) very
often exhibit properties indistinguishable from those required; in
particular they are not serially correlated. Also, a modified
version of the paired t-test has been developed (Napier-Munn and
Meyer, 1999) which uses the AR(1) model to deal with the
problem of serially correlated data.
If there is any doubt, raise the confidence level at which you
are prepared to make a decision, thus reducing the risk of making
a Type 1 error.
If the parametric methods cannot be used, then you have to
model your way out, or use the reference distribution (Napier-
Munn, 2008). The latter is advocated by some, but as noted above
it is susceptible to the assumption that the chosen RD is
characteristic of the (constant) process before the trial. Models
have the advantage that they accommodate other variables which
are known to affect the response, such as feed grade or time, so
improving the precision of the comparison. They also don’t need
a formal design and can sometimes throw up other useful
information. However, again they depend on the validity of the
assumptions underlying the method of least squares, such as
normally distributed and independent residuals. And again,
experience in mineral processing suggests that these assumptions
are not as wildly infringed as some suggest. A disadvantage of
modelling is that the effect of a trial condition can only be
unequivocally established if all other material effects are included
in the model. This is not always possible.
So here is the preferred order:
1. paired t-test, or randomised block if there are more than two
conditions;
2. regression model with important process variable(s) and
trial as a dummy variable;
3. time series model with trial as a dummy variable
(intervention analysis); and
4. reference distribution.
The cusum chart can be a useful way of spotting or illustrating
process changes visually. And process optimisation procedures
such as EVOP may be worth a try.
Some final thoughts
1. Plan the trial, stick to the plan, analyse the data rigorously,
and then make a decision. The decision might be that there
is no decision, or to continue the trial, but there must be an
outcome to every experiment.
2. Always plot the data, particularly time series data. Potential
outliers should show up clearly, as should trends and cycles
in the data which may have important implications.
3. Replicate, replicate, replicate. You cannot make a comparison
between conditions unless you have some idea of error.
4. It is usually all too easy to follow local tradition or to cut
corners, particularly in plant trials. Traditions often grew up
in an age of ignorance, and cutting corners, especially in
regard to replication, is almost always wrong. Doing the
189
CHAPTER 10 – DESIGNING AND ANALYSING PLANT TRIALS
right experiment will always produce the right answer in the
minimum time (and at the minimum cost). A little lateral
thinking will usually solve the apparently intractable
practical problems of getting the plant environment to
conform to the required experimental design. Cutting
corners, such as not replicating, is a seductive alternative to
rigour in experimental design. History tells us that this
always leads to a non-optimal outcome, sometimes to the
wrong outcome, and often to no outcome at all. And those
who forget history are condemned to repeat it.
5. Most of the data analysis methods described in this chapter
can be handled in Excel. For the more complex
experimental designs, such as large factorials or the CCRD,
it is recommended that you invest in one of the specialist
statistics software packages both for designing the
experiment and analysing the data. However, be aware that
there is a substantial overhead involved in coming to grips
with such packages, and retaining the skill of using it
over time.
6. And finally, ponder the well-known quotation:
An unsophisticated forecaster uses statistics
as a drunken man uses lamp-posts – for
support rather than for illumination
attributed to the Scottish man of letters Andrew Lang. In
conducting and interpreting our experiments, let us seek the
illumination of wisdom and eschew the lamp-post
of prejudice.
REFERENCES
Box, G E P and Draper, N R, 1987. Empirical Model-Building and
Response Surfaces (John Wiley and Sons: New York).
Box, G E P, Hunter, W G and Hunter, J S, 1978. Statistics for
Experimenters (John Wiley and Sons: New York).
Bruey, F and Briggs, D, 1997. The REFDIST method for design and
analysis of plant trials, in Proceedings Sixth Mill Operators’
Conference, pp 205-207 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
Cavender, B W, 1993. Review of statistical methods for the analysis of
comparative experiments, SME Annual Meeting, Reno, February,
Preprint 93-182 (Society for Mining, Metallurgy, and Exploration:
Littleton).
190 Spectrum Series 16
Chatfield, C, 1983. Statistics for Technology, third edition (Penguin:
Harmondsworth).
Davies, O L (ed), 1967. Design and Analysis of Industrial Experiments
(Oliver and Boyd: London).
Draper, N R and Smith, H, 1998. Applied Regression Analysis, third
edition (John Wiley and Sons: New York).
Himmelblau, D M, 1970. Process Analysis by Statistical Methods (John
Wiley and Sons: New York).
Mason, R L, Gunst, R F and Hess, J L, 1989. Statistical Design and
Analysis of Experiments (John Wiley and Sons: New York).
Montgomery, D C and Runger, G C, 1999. Applied Statistics and
Probability for Engineers (John Wiley and Sons: New York).
Mular, A L, 1971. The selection of optimization methods for mineral
processes, in Proceedings Decision-Making in the Mineral Industry,
special volume 12, pp 442-450 (Canadian Institute of Mining,
Metallurgy and Petroleum: Montreal).
Mular, A L and Bull, W R, 1973. Mineral Processes: Their Analysis,
Optimization and Control, course manual, 1969 - 1973.
Napier-Munn, T J, 1998. Analysing plant trials by comparing recovery-
grade regression lines, Minerals Engineering, 11(10):949-958.
Napier-Munn, T J, 2008. Some practical problems in running statistically
valid plant trials, and their solution, in Proceedings Metallurgical
Plant Design and Operating Strategies (MetPlant) 2008, pp 249-258
(The Australasian Institute of Mining and Metallurgy: Melbourne).
Napier-Munn, T J and Meyer, D H, 1999. A modified paired t-test for the
analysis of plant trials with data autocorrelated in time, Minerals
Engineering, 12(9):1093-1100.
Snedecor, G W and Cochran, W G, 1968. Statistical Methods, sixth
edition (Iowa State University Press: Oxford).
FURTHER READING
Cochran, W G and Cox, G M, 1966. Experimental Designs, second
edition (John Wiley and Sons).
Davies, O L and Goldsmith, P L (eds), 1972. Statistical Methods in
Research and Production, fourth edition (Oliver and Boyd).
Napier-Munn, T J, 1995. Detecting performance improvement in trials
with time-varying mineral processes – three case studies, Minerals
Eng, 8(8):843-858.
Flotation Plant Optimisation
 HOME

CHAPTER 11

Economics and Communication

Joe Pease

MAusIMM, Chief Executive, Xstrata Technology, Level 4, 307 Queen Street, Brisbane Qld 4000.
Email: jpease@xstratatech.com.au

Joe has degrees in Metallurgical Engineering and Economics. He spent 21 years in operations in Mt Isa,
Canada and Alaska, in a variety of research, production and business analysis roles. He was manager of
the Mount Isa lead zinc and copper concentrators for eight years, and lead smelter manager for two years.
He was involved in several large capital projects to substantially improve metallurgical performance. He was
involved in the development and application of new technologies to improve processing efficiency, including
the Jameson Cell and IsaMill™. Since 2003 Joe has been Chief Executive of Xstrata Technology, which
markets innovative processing technologies and expertise to the global minerals industry. Joe is a Board
Member of the industry research groups AMIRA International, the Julius Kruttschnitt Mineral Research
Centre and the Ian Wark Institute Advisory Board.

Abstract
Evaluation Techniques
Three Common Techniques – Net Present Value, Internal Rate of Return and Payback Period
Communication
The Presentation
A Case Study
References

ABSTRACT
This chapter is about the final stage of getting projects approved –
economic evaluation and communication to ‘sell’ the project. This is an
exciting stage, and is easy compared with the work you have done to get
here. But don’t underestimate it – many poor projects are approved, and
many good projects are rejected because of mistakes made at this stage.
Not numerical mistakes, but mistakes in assumptions or approach.
Economic calculations are easy, but are completely dependent on the
quality of data and assumptions. And the approach to ‘selling’ a project is
different from the rigorous engineering approach you have needed to get to
this point.
The chapter starts with a discussion on some of the simple evaluation
techniques to assess projects. More importantly, it highlights the dangers in
these techniques – their simplicity can conceal the unavoidable uncertainty
in your data. The really hard work in the economic evaluation is getting the
right data and knowing its limitations.
We then look at how to increase the chance of getting a good project
approved. This is where engineering meets marketing and psychology. The
directors are clever, hard-working people, but maybe none of them are
metallurgists. They are busy, may not really understand what you are
talking about, and they need to ration capital between an overwhelming list
of competing projects. You have been trained in the scientific method – you
won’t quote anything that isn’t backed up by hard data. Some of the other
groups competing for funds aren’t as rigorous – but they do fantastic
PowerPoint presentations. While you were working weekends to collect
data, they were playing golf with one of the directors.
So how do you make sure that the company considers your project fairly
and help it make the right decision? The last section gives you a few hints.
Most important is that the communication process starts in the earliest
stages of the project – indeed, it starts with selecting the right project to
work on in the first place.
EVALUATION TECHNIQUES
This section is deliberately brief. There are many well qualified
texts to demonstrate the theory and mechanics of economic
evaluation methods and risk analysis. Many calculators have built
in net present value (NPV) and internal rate of return (IRR)
functions, and computer spreadsheets are well suited to laying out
an economic evaluation, and completing it with an NPV and IRR
function. These techniques can be easily researched elsewhere.
Your important task is to get the right numbers to use in the
evaluation – this can take years, whereas the economic model will
take hours. Spend your time making sure you look at the right
project, in the right way, and collecting accurate numbers. Almost
certainly your company will have a financial analyst responsible
for reviewing your project according to the company standards.
You don’t have to do the analyst’s job – but you can help them do
it. Understand how they are going to use your numbers and what is
important to their analysis. Encourage them to ask you tough
questions – better that you deal with them now than be caught out
at the Board Meeting. Gaining the respect and support of the
financial analyst is an important part of the communication process
– if they believe in your project you have an influential ally.

Flotation Plant Optimisation Spectrum Series 16 191

CHAPTER 11 – ECONOMICS AND COMMUNICATION
THREE COMMON TECHNIQUES – NET PRESENT
VALUE, INTERNAL RATE OF RETURN AND
PAYBACK PERIOD
Texts often summarise these as the three most common
techniques used for evaluation. It isn’t a choice of which one to
use, rather all of them look at the same numbers in a slightly
different light to give some different perspectives. Table 1 shows
a simple example analysed by all three methods. Sensitivity
analysis is useful to show the effect of the various assumptions,
both so you know how robust the project is to changes, and to
demonstrate which numbers are really important to get right.
Net present value (NPV)
Net present value (NPV) is generally regarded as the most correct
economic evaluation method. But remember that NPV is always
trumped by GIGO – if your assumptions or data are wrong, then
your answer is garbage, no matter what method you use or how
detailed your spreadsheet is.
NPV represents the value in today’s money of all the future
cash flows associated with the project. All future cash flows are
‘discounted’ back to today’s value using a ‘discount rate’. Usually
the discount rate will be set by your company. It may include:
• the cost of borrowing,
• inflation rate, and
• an additional factor to cover risk and margin.
The company’s cost of borrowing is effectively the interest rate
your company pays – ie if interest rate is eight per cent then if
you borrow $100 you will owe $108 next year. Or to look at it the
other way, $108 next year is worth $100 today.
Further, if the inflation rate is four per cent, then you will need
$104 next year to buy what $100 would buy you today. This
inflation effect adds to the interest rate effect, so people usually
do their evaluations on the basis of ‘real dollars’. All future cash
flows are converted to ‘real’ dollars (ie expressed in today’s
dollars), and the discount rate the company provides should
account for this. The alternative is to use ‘nominal dollars’, which
means future dollar values include inflation, but they then need to
be adjusted by inflation so you can compare dollars from
different years on an ‘apples and apples’ basis. Whether you use
real or nominal dollars, you must ensure all your numbers are
consistent. For example, if you get a metal price assumption from
corporate, is it quoted in real or nominal dollars?
After the interest and inflation factors, companies add an
additional risk factor to the discount rate. If a project will only
return inflation and borrowing cost, then why bother risking
money on it? The risk factor will depend on how much margin
the company thinks is acceptable to reward it for the risk, and
how risky it thinks the project is. Companies often use this to
‘ration’ capital between projects – there are too many proposals to
go ahead, so they set a ‘hurdle rate’ so that only the higher
returning projects get approved (more accurately, only the
projects that claim they will get a higher return get approved!)
Whatever policy your company has, it will be expressed in the
discount rate that you will have to use.
The base case
Whatever analysis technique you use, you will be analysing your
project against the ‘base case’. Often the base case is ‘do
nothing’, to compare doing the project with current performance.
But think carefully – is the base case really ‘do nothing’? Maybe
there is a better alternative. For example, assume the project you
are considering is to install extra grinding power to improve
liberation and recovery, and the project will cost you $5 million
192 Spectrum Series 16
for five per cent recovery increase. Before you analyse the NPV
of spending $5 M and getting five per cent higher recovery, you
must check if there are any other alternatives. Is the current
grinding circuit performing as well as it can? Can you improve
grind size by reducing media size, replacing cyclone components,
better use of grinding control, improving operating practices? What
if spending $0.5 M on these items can get you three per cent
recovery? In that case, you need to first fix the grinding circuit for
$0.5 M, then analyse the remaining gains – can you now justify the
$5 M project for the remaining two per cent recovery?
This sounds obvious, but a depressing amount of projects are
proposed (and approved) that do not rigorously challenge the base
case. It is easier and more interesting to buy new equipment
rather than the hard work of improving what we already have. It
is your job to get the best out of your equipment first – ‘no new
toys until you are using all your old toys properly!’ Ultimately
this will pay you back – you want to build a reputation as
someone who spends money wisely – someone who wouldn’t ask
for capital unless they really need it. And management will
believe your project will deliver because you operate existing
equipment well. It is better to propose one really good project
than five ordinary projects – especially if the five ordinary
projects are approved and you have to deliver!
Checklist and conventions for NPV spreadsheets
• Make sure you include all relevant cash flows – eg extra
operating costs, extra maintenance costs, depreciation value
of capital equipment, scrap value or resale value or
rehabilitation value at the end of the project.
• Make sure you know whether every number is real or nominal
dollars.
• Include tax – check your company’s convention and tax rates.
• Make sure the discount rate you use is the right one, eg if
your cash flows are real, after tax dollars, then the discount
rate must be the one to suit real, after tax dollars.
• Finally, never quote an NPV without at least three qualifiers
– what discount rate was used, was it real or nominal dollars,
was it before or after tax? To say the project has an NPV of
$10 M is utterly meaningless. NPV depends entirely on the
discount rate used – the same set of cash flows can give an
NPV of $10 M at 15 per cent discount rate (real, after tax),
but an NPV of $100 M at ten per cent discount rate. So you
need to say that the project has an NPV of $10 M at a
discount rate of 15 per cent in real, after tax dollars.
Some golden rules for NPV
While NPV is a correct technique, remember some golden rules:
• Compare the right options. The project NPV depends
entirely on you analysing the right project, against the right
base case, using the right numbers.
• NPV doesn’t capture everything. Some things can’t be put
into money terms, eg a project may cause harm to the
environment, or may damage your company’s reputation.
You can’t put a value on these but you can’t ignore them.
Perhaps you can add to the capital cost to ensure there will
be no adverse consequence – your project cost may go from
$5 M to $7 M, but you have to be able to assure the Board
that there are no adverse consequences compared with the
base case.
• NPV may not capture the relative risk of projects. Sensitivity
analysis can help to quantify the risk of error in the
assumptions (Figure 1). Tools are also available to ascribe a
range of outcomes against each variable in the analysis,
Flotation Plant Optimisation
Flotation Plant Optimisation

TABLE 1
Simple financial analysis speadsheet.

No Item Year
This example uses single years. You may have to break
0 1 2 3 4 5 6 - 13 14 15
early years into months to capture cash flow details.
All dollar values estimated in real dollars (ie in today’s dollar values).
1 Capital cost -10 000 000 Capital cost includes all items from your analysis in the Sample
Financial Analysis section, including capital spares. State the
accuracy of the estimate, eg ±10 per cent, ±30 per cent.
2 Additional -500 000 Budget to use extra resource and for contingencies during
commisioning costs commissioning.
3 Additional -1 800 000 -2 000 000 -2 000 000 -2 000 000 -2 000 000 etc -2 000 000 0 Includes concentrator, filtering and drying, transport, etc.
operating costs
4 Ramp-up profile 81.3% 100% 100% 100% etc 100% 0% This example assumes 50 per cent of benefit in first three months,
75 per cent for next three months, then 100 per cent. The ramp up
profile will relate to project complexity, quality of engineering, and
quality of commissioning.
5 Increased 12 188 15 000 15 000 15 000 etc 15 000 0 Ultimate increased annual tonnes, 5000 × ramp up profile (Row 4).
concentrate tonnes
produced
Spectrum Series 16

6 Net value of one 500 500 450 400 etc 400 Calculated from your estimate of concentrate quality, and company
tonne of extra assumptions of metal prices, exchange rates, treatment charges,
concentrate penalties, transport costs, etc. In this example the company forecasts
prices to fall in year 4.
7 Net sales revenue 6 093 750 7 500 000 6 750 000 6 000 000 etc 6 000 000 Row 5 × Row 6.
of extra concentrate
8 Net salvage value 750 000 Estimate. After discounting, it has little effect on project value.
(after
decommissioning

CHAPTER 11 – ECONOMICS AND COMMUNICATION

costs)
9 Net operating cash -10 000 000 -2 300 000 -4 093 750 5 500 000 4 750 000 4 000 000 etc 4 000 000 750 000 Rows 1 + 2 + 3 + 7 + 8. This is the net operating cash flow before
flow before tax tax. Rows 10 to 12 now calculate the effect on tax paid.
10 Allowable -1 500 000 -1 500 000 -1 500 000 -1 500 000 -1 500 000 Stops in Depreciation rate will vary for different equipment – get accounting
depreciation on year 7 advice. This example assumes 15 per cent. Don’t forget to stop
new equipment depreciation when full value has been written off – in this case after
6.7 years.
11 Extra operating -3 800 000 2 593 750 4 000 000 3 250 000 2 500 000 etc 4 000 000 750 000 Row 10 + Row 11, ie operating cash flow less allowable
profit before tax depreciation. It is this amount that tax is paid on.
12 Extra tax paid 1 140 000 -778 125 -1 200 000 -975 000 -750 000 etc -1 200 000 -225 000 Assuming company tax rate of 30 per cent. The positive tax in Year
1 assumes the company is profitable, so the the loss will reduce tax
paid on other profit.
13 Cash flow after tax -10 000 000 -1 160 000 3 315 625 4 300 000 3 775 000 3 250 000 etc 2 800 000 525 000 Row 9 + 12.
(real dollars,
undiscounted)
14 Cumulative cash -10 000 000 -11 160 000 -7 844 375 -3 554 375 230 625 3 480 625 etc 46 302 813 46 827 813
flow (after tax,
193

undiscounted)
CHAPTER 11 – ECONOMICS AND COMMUNICATION

NPV, $ M

The project has paid back its invested cash about four years after the
15

Sum across Row 16. You could also use the NPV function on
Row 13. However, writing down values in Row 16 shows the
Change in revenue

impact of variables, and helps you see anomolies or errors.

12 – eg production,
recovery, prices
Change in capital or
operating cost (lines 9
For example in year 3, divide Row 13 by 1.153.
overlap)

For example in year 5 divide Row 13 by 1.35.

6

Use IRR function on Row 13.

1 year commissioning delay

0
-30% -20% -10% 0% 10% 20% 30%
money is spent.

-3 % change in assumption

FIG 1 - Sensitivity of net present value (15 per cent) to key

assumptions.

along with a probability distribution. This provides a

64 520

10 257

probability distribution of final outcomes. This can be

15

useful, but don’t be fooled; it is still only as good as the

assumptions used. Too often we underestimate capital cost,
underestimate commissioning time, and overestimate gains –
395 720

71 113

whether we estimate a single value or a probability

Simple financial analysis speadsheet.

14

distribution, they are just as wrong. It is better to spend your

time making sure your estimates are realistic and defensible,
your project is robust, and developing contingency plans to
TABLE 1 cont ...

deal with surprises.

6 - 13

etc

etc

• NPV may favour short-term projects over long-term

company building projects. This is especially a problem if a
high discount rate is used to ration capital. At a discount rate
1 615 824

875 319

of 15 per cent (real, after tax) one dollar earned in 15 years

5

time is worth only 12 cents today; at 30 per cent discount

All dollar values estimated in real dollars (ie in today’s dollar values).

rate it is worth only two cents. That is, a 30 per cent discount
rate assigns very little value to anything that happens beyond
2 158 369

1 321 732

about ten years time. This favours projects with low up-front
Year
4

capital, short project duration and early cash flows. It

penalises longer term or complex projects with a high capital
investment and slower returns, even if they result in
2 827 320

1 957 214

excellent cash flow in ten years. So strategic, company

building projects may be overlooked. You can’t change your
3

company’s policy, and don’t even think about ‘fiddling’ the

numbers to improve the NPV. But if you think your project
2 507 089

1 961 908

is low risk and fundamental to build future strength, you

should be honest and argue that a lower discount rate may be
2

appropriate. You may not be successful, but give the Board

the opportunity.
• NPV says nothing about your relative competitiveness. The
-10 000 000 -1 008 696

-892 308

NPV analysis uses assumed metal prices to calculate project

1

margin. But any project that increases production will look

good at high prices. What about when prices drop – then it is
-10 000 000
11 543 000

more important that you control, or reduce, the unit cost of

986 100
4 years

33%

metal production. You should work out the unit cost of the
0

extra metal your project will produce – does it increase or

decrease the average unit cost? If it increases, ask the
question ‘Do I really want to spend capital to become less
NVP (30 per cent)
NVP (15 per cent)

competitive?’ Try to find a way to both increase revenue and

Payback period

Internal rate of
discounted cash

discounted cash
At 15 per cent

At 30 per cent

return (IRR)

reduce unit cost.

discount rate:

discount rate
Item

Internal rate of return (IRR)

flow

flow

Having done NPV, then IRR is simple – it is simply finding out

what discount rate would make the NPV of the project zero. For
No

16

18
15

17

19
20

example, a project may have an NPV (real, after tax) of $15 M at

194 Spectrum Series 16 Flotation Plant Optimisation


15 per cent, and an IRR of 31 per cent. That is, even if the
company discount rate (or hurdle rate) for projects was 30 per
cent, the project would still have a (slightly) positive NPV and
should be approved.
Though it is closely related to NPV, IRR adds a different
dimension. You can compare the relative risk/reward of projects.
For example, consider the two projects in Figure 2:
• Project A costs $80 M, and has an NPV (real, after tax at
15 per cent discount rate) of $39 M, for an IRR of 28 per
cent; and
• Project B costs $20 M, and has an NPV (real, after tax at
15 per cent discount rate) of $23 M, and an IRR of 43 per cent.
119 M
Project NPV
$39 M
$80M
$43 M
$20 M
NPV of Net
Capital NPV of Net
Revenue
Capital
Revenue
Internal Rate of Return : 28% Internal Rate of Return : 43%
Project A Project B
FIG 2 - Which is the better project?
Assume both proposals have been prepared professionally, with
good quality data. Which is the better project? On purely NPV
analysis we would say Project A – it has $14 M higher NPV than
Project B; but consider two other factors:
• What is the risk/return, and the unavoidable uncertainty in
the analysis?
• Many projects will be competing for limited capital.
While Project B has a lower NPV, it only risks a quarter of the
capital. It has a higher return on the capital spent, as shown by the
higher IRR. While both proposals have been prepared
professionally, there are always factors out of our control, eg
metal prices may drop, treatment charges and exchange rates will
change, the mine tonnes and grade may be different from
forecast. So perhaps Project B is more ‘robust’ than Project A –
you may not make as much money, but you are less likely to go
broke if things change.
If this sounds too conservative, look at it from the Board’s
perspective. They have lots of projects to consider, not just yours.
They can’t approve everything, even if they are all good projects
– capital is limited, so are the resources to execute the projects
(engineers, project managers, etc). Assume the company can only
consider $80 M of projects. Should it do Project A or Project B?
We still don’t know. What if other departments have also
proposed projects like Project B? For $80 M capital on four
projects, the company can now return an NPV of $92 M, much
better than the $34 M from Project A. This also hedges the risks
across four projects instead of one, each of which has room for
error before it loses money.
Flotation Plant Optimisation
CHAPTER 11 – ECONOMICS AND COMMUNICATION
Payback period
Payback period is a simple way to rank projects – how many
years does it take for the project to pay back what we spent on it?
With sophisticated spreadsheet modelling so easy, the simple
payback period is not commonly used. You need the same
information to calculate it as you do for NPV and IRR (ie future
cash flows). However, it is useful to calculate it anyway as it gives
you another perspective of the project and risk – a project that pays
back in three years is inherently less risky than one that takes ten
years to pay back – even if metal prices change dramatically in five
years time you have already got your money back. On the other
hand, over-emphasis on payback period may cause you to only
consider simple short-term projects, not more fundamental projects
that will strengthen your operation but take longer to deliver.
Sample financial analysis
The project
You propose to install additional grinding power to improve
liberation, and improve recovery of lead, copper and zinc by
five per cent each. The capital cost is $10 M, and the project will
Project take 18 months to execute. You estimate it will take three months
NPV
$23 M
after start-up to deliver full recovery gains.
Work you have already done before the financial
evaluation
• Circuit surveys backed up with sizing and mineralogical
work to determine the theoretical benefits of extra grinding.
You have quantified the reasons minerals are misplaced in
your circuit – ie of the diluents in concentrate, and values in
tails, how much is due to:
• inadequate liberation,
• entrainment with water, and
• poor flotation selectivity (after accounting for entrainment).
• Laboratory and/or pilot scale work to demonstrate the gains
from extra grinding. This work has been backed up with
quantitative mineralogy.
• Evaluated different grind sizes and determined the cost and
benefit of each, to determine the ‘optimum’ grind size.
• Evaluated different places to apply the grinding power – do
you need to grind the whole stream, or can you achieve
90 per cent of the gain by grinding 50 per cent of the stream.
You may need to compare different options to choose the
best capital versus return (best ‘bang for your buck’).
• Evaluated all test results in terms of the effect on
grade/recovery curve. Determined what the project will
achieve – increased grade at the same recovery, increased
recovery at the same grade, or a combination of both.
• Considered the impact on impurities in concentrate, eg
silica, MgO, halides, mercury, etc – anything that may be
relevant to your smelter/refinery/customer.
• Worked with geologists and mine planners to ensure that:
• the ore you did your test work on was representative, and
• you understand future ore supplies (grade and
mineralogy) and how this might influence the project
returns (and how it might influence the base case).
• Having chosen your target grind size, you have then done
test work and simulation to choose what grinding and
classification equipment is needed.
• Determined whether the additional grinding will require
extra flotation equipment.
Spectrum Series 16 195
CHAPTER 11 – ECONOMICS AND COMMUNICATION
• Designed for all downstream impacts – froth handling
(launders and sprays), froth pumping, thickening, filtering,
concentrate drying, transportable moisture.
• Checked for all infrastructure impacts and costed for
additional if necessary – power supply, power distribution,
control system capacity, supply and reticulation of water,
compressed air, instrument air, gland water, blower air, etc.
• Checked with customers whether the finer concentrate will
have any influence on their volumes or treatment charges.
• Checked that you are comparing against the right ‘base
case’. Is your current grinding/classification circuit running
as well as it can? Or do you need to get it running better
before you can evaluate the benefit of extra grinding?
• Done adequate engineering evaluation for your cost
estimate. Your engineers will advise the accuracy of this
estimate, eg ±30 per cent, ±20 per cent, etc.
• Included recommended initial spares holding as part of the
capital cost.
• Estimated maintenance and operating costs – power, water,
air, reagents, maintenance consumables, operating and
maintenance labour, instrument labour. Will your project
require additional maintenance or operating people?
• Estimated a start-up profile for your project. What is the
profile for cost increase? What is the profile for revenue
increase? What resources are necessary to achieve the start-
up profile? Commissionings always need more resource than
normal operating levels. If you don’t plan for them, you will
have a slower start up profile.
• Considered any impacts on safety and environment. You will
have done a ‘hazop’ or risk analysis study in line with your
company’s procedures, and will have allowed for any
mitigating actions that resulted from this study.
COMMUNICATION
There is nothing worse than a good project that isn’t approved
because of poor communication and presentation. As the saying
goes, presentation isn’t everything, it is the only thing.
Actually, there is one thing worse – bad projects that do get
approved because they were sold by great communication and
presentation. So add another saying – never let style replace
substance. There is nothing you can do about the bad projects
with slick presentation – it is the Board’s job to try to recognise
them. They are human too, so they will get it wrong sometimes.
Don’t worry about that. Just make sure you get your numbers
right, and that your project will definitely deliver if it is approved.
Then you have the confidence to add some style to your
substance. The Board likes style, but most of all they like to back
a proven winner, so a track record of delivering on your promises
will outweigh a slick presentation.
Communication or presentation?
The above paragraph refers to the Board presentation. In fact, this
is misleading – the real communication work should be done long
before you present to the Board. The Board presentation should
be a formality – they should be sold on your project before you
even walk into the room. We will consider several stages
of communication.
Step 1 – choose the right project
This might sound trite, but it is the most important step. You are
going to spend a considerable part of your life on this project –
doing the research, doing test work, analysing alternatives, doing
196 Spectrum Series 16
designs and costings, then ‘marketing’ it. If you succeed, you will
be involved in detailed design, scheduling, training, implemen-
tation, and trying to run your plant while allowing for cut-ins and
installation shutdowns. You will have to commission it, solve any
problems, and make sure it delivers performance. Meanwhile you
will still be doing your day job, and trying to spend some time
with your family and friends. This project could dominate your
life for the next three years. Your reputation will depend on its
success. Are you really sure it is worth it?
In deciding whether you really want to ‘marry’ this project, ask
yourself some questions:
• Is this project addressing a symptom or a root cause? For
example, our project of installing extra grinding addresses
the root cause of poor mineral liberation. An alternative
project to try to improve flotation by changing flotation
impellors or reagents may help alleviate some of the
symptoms, but doesn’t address the root cause. If you treat
symptoms only you will just end up with a more
complicated plant. You want to simplify plant operation, not
complicate it. So do your homework and determine the root
cause. If flotation performance is poor, find out how much is
due to poor liberation, how much is due to poor flotation
selectivity, and how much is due to entrainment.
• Have I really considered all the alternatives? Am I getting
the best out of my current equipment? Are there other
projects that can deliver some of the gains more simply and
cheaply?
• Am I prepared to stake my reputation on the project? If not,
find another project. If you can’t, then be happy – your plant
is running well, and you can save the company’s money for
important projects elsewhere.
By answering these questions honestly, you know how to
justify the project. If you still believe in the project, then you will
have the passion and commitment to convince others.
Step 2 – do your homework
Let’s say it again. Get your numbers right. Be rigorous. The
financial analysts and the Board can’t check your metallurgy.
They don’t know how well your tests were designed, how
accurate your samples were, whether you drilled down to the root
cause. That is your job and they expect you to do it. Maybe the
person proposing another project cut corners and did a fraction of
the work so their numbers are ‘rubbery’ and the Board won’t
know the difference. Too bad; there is nothing you can do. All
you can do is get your own numbers right, and be prepared to
stake your reputation on your project. Ultimately you will
develop a track record of success.
Step 3 – develop champions and momentum
You can’t do this by yourself. You will need a ‘critical mass’ to
make this project succeed – research people, engineering,
maintenance, operations people. The wrong word from any one
of these groups can kill the project. Talk to the maintenance
people early – will they be able to maintain the extra equipment?
Can you design the project to make their job easier? What sort of
equipment do they prefer, what spares should you buy, what
maintenance access should you design in? What about the
operators and operating superintendent – are you making their job
easier or harder, more interesting or just overwhelming? How can
you get them involved in the design – are they happy with the
access, the sample points, the training, the clean-up points? Can
the project solve some of their existing problems? What about the
senior metallurgist who has been pushing an alternative project
for years without success. How do you make him an ally, rather
Flotation Plant Optimisation

than try to undermine your project? Can you incorporate some of
his ideas, and let him share in the credit? Remember the most
important rule in any organisation: you can achieve anything you
like so long as you don’t mind who gets the credit.
If you involve these people early enough, they can get
ownership of the project – they will want it to happen, and this
will build momentum. If it really is a good project and you have
involved enough people, eventually the project will develop
momentum and appear ‘inevitable’.
Step 4 – cover all the angles
Remember that senior management and the Board have to paint
on a much broader canvas than you. They will be concerned with
issues of corporate governance, public image, share market
perceptions, safety performance, environmental issues,
greenhouse strategy, strategic direction, debt levels, employee
relations, competitors, unit cost reduction, external regulations
and audits, etc. Put yourself in their shoes – what might they need
to know about your project besides the recovery improvement?
Can you design it to improve performance in several areas, not
just recovery and profit? It’s hard not to support a project
that increases profit, reduces unit cost, simplifies the plant,
improves housekeeping and safety and improves environmental
performance.
Maybe this just takes some lateral thinking. On first
appearances the Board might think that installing extra grinding
power will increase the ‘greenhouse footprint’. Think again.
Aren’t you going to increase recovery by five per cent, on ore that
has already been mined, crushed and ground? If you calculate
total energy consumption from mine to concentrate, divided by
the tonnes of concentrate produced, you may be producing metal
for less energy per tonne. That is no surprise to you – your job as
a metallurgist is to increase the efficiency of processing. But you
may need to calculate it and communicate it in different terms so
others can understand the efficiency increase.
Step 5 – disarm the torpedos
This step is to address the negative issues that you couldn’t deal
with positively in Steps 3 and 4. It is better to be positive, but
every organisation has negative people. Some of them have the
power to derail your proposal with a single comment – a
project-killing torpedo. Seek them out. Spend time with them, try
to make them positive. If they are still negative, don’t waste
energy trying to change them, but make sure you understand their
concern and deal with it proactively. Disarm the torpedo before
they fire it, rather than go into damage control after it has hit your
project. The torpedos could be about anything. Some examples
for our grinding project are listed below, but you need to do
your homework in your own company to know where they might
come from:
• Why would we want to grind finer, most of our losses are in
the fines fraction, won’t we just make it worse?
• You might improve recovery, but you will never be able to
filter the finer concentrate.
• Our customers are already complaining that our concentrate
is too fine; you won’t be able to sell it if it is finer.
• The finer tailings will make the tailings dam unstable, or
will generate more acid in the dam. This could shut us
down!
• If you ran your plant properly you wouldn’t need ‘new toys’
to increase recovery.
• You have ‘gold plated’ the design. I could do the same
project for half the cost.
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 11 – ECONOMICS AND COMMUNICATION
• The ore is going to change next year so your test work is
meaningless.
• Spending the same money in the mine/smelter/IT system
would yield double the benefits.
We didn’t say these are fair, or right. But they exist, they are
dangerous, and you need to prepare for them.
Step 6 – get high level champions
Perhaps this is covered in Step 3, but it is important enough to
discuss separately. You need high level support, the earlier the
better. Explain the project to senior management, try to get them
excited and involved early. Find out what they need to sell it to
their bosses. They have a lot of experience dealing with company
‘politics’ and getting projects approved. Don’t worry if it seems
your boss will get all the credit – if the project makes your boss
successful, then you and your department will be successful too.
Get the high level contacts to work on the parts of the
organisation that you can’t influence – maybe the mine manager,
the general manager, the Board members. Get their feedback on
what concerns they have, what issues need to be addressed, what
torpedos are out there?
Step 7 – timing and patience
The communication discussed in Steps 1 to 6 is not an event; it is
a continuous process that weaves through the entire project
development. You have probably had to justify spending time and
research money on this project from the very start, so use that
time to build the momentum in case the research turns into a
viable project.
And try, try again. If you have chosen the right project to work
on, then it is worth fighting for. People need ‘soak time’ – they
may really be hoping that a cheaper alternative is found, or that
the problem goes away. They may need time to get used to the
idea. As in sport, timing can be just as important as skill. It might
be a great project but a bad time to be pushing it – the company
may have a lot of other demands on capital and there may be a
bunch of other excellent projects that will be approved first. Don’t
despair, if it is a good project it will have its day. Use the time to
continue building understanding and momentum, so when the
opportunity opens you are ready to strike.
Step 8 – be passionate, be committed, and be
prepared to walk away
If you don’t really believe in the project, how can you sell it to the
Board? If you aren’t prepared to stake your reputation on it, then
walk away from it. Cut your losses – sure you have spent a lot of
time on it already, but don’t waste your time and reputation by
proceeding with it and trying to make it work. One of the failures
of human nature is that we never recognise those who prevent
fires as much we recognise the fire fighters. We build so much
momentum behind a project that we really want it to happen. Yet
often we should award the medals to the cool heads that walk
away from poor projects.
But if you have faced this question honestly, and still believe in
the project, then your passion will show, and you should have all
the answers for the Board.
There is another time you need to walk away. You have done a
thorough job planning the project, you are sure of your costings,
scheduling and the plant design. You ask for $50 M and
18 months to execute the project. The rate of return will be 30 per
cent. The Board says they will approve it if you do it for $30 M in
12 months. Find out what can be done to the costs and schedule,
and what the risks are. Maybe the risks are acceptable. Or maybe
197
CHAPTER 11 – ECONOMICS AND COMMUNICATION

you aren’t prepared to live with the risks – perhaps the new 90
design will be a maintenance and operating nightmare. In that 80
case, tell them you would rather walk away from the project than
do it badly. That you can’t guarantee the returns. This is not a 70
bluff, you have to be serious. Either they will see your passion, 60
and decide to accept your 30 per cent return. Or they will accept
50 East
your offer to walk away from the project – but you live to fight
another day, and you have saved them from a badly executed 40 West
project that would have wasted $30 M, several years of your life, 30 North
and your reputation.
20
THE PRESENTATION 10
By now the presentation should be easy. You have done all the 0
hard work, you have answered the questions and disarmed the 1st Qtr 2nd Qtr 3rd Qtr 4th Qtr
torpedos. You have built champions at all levels of the
FIG 3 - A victory of style over substance ... the PowerPoint default
organisation. People have had a long time for the ideas to ‘soak template.
in’. Momentum has been built, and everyone expects the project
is inevitable. You have put yourselves in the shoes of your boss,
the senior management, and the Board, and have considered what 100
they are interested in and what they need to know.
There are plenty of books and courses on presentation skills, so 80 East
we won’t repeat them here. If your project has rigour and
substance, and you have passion, then you will get the
60
presentation right. Just follow some simple rules:
• Keep it brief. These are very busy people with lots on their 40
North
mind. They don’t have to know the details of your job –
West
that’s what they pay you for. Tell them what they need to
know to do their job. 20
• Keep each slide simple and concise. A few bullet points per
slide to make the key points. 0
• Structure – tell ‘em what you’re gonna tell ‘em, tell ‘em, 1 2 3 4
then tell ‘em what you just told ‘em. Quarter
• Avoid fancy effects, sound effects, lots of colour and fonts.
FIG 4 - Let the data speak simply! The same data without the
They are looking for professional work explained simply
‘chart junk’.
and eloquently, not a Hollywood production.
• Remember a significant amount of people are red-green
colour blind. Choose simple colours with good contrast.
A CASE STUDY
• Use pictures and graphs to communicate concepts wherever
possible. Look for ‘the picture that tells a thousand words’.
…or, two pictures that said a thousand words
• Keep your graphs simple – get rid of all unnecessary grids,
legends, titles, etc. The graph is designed to communicate a This example comes from the Mount Isa lead zinc concentrator in
few points – use the minimum amount of ‘ink’ to emphasise the 1980s. Our ore had become progressively finer grained, and
just these points. An excellent and enjoyable text is The recoveries dropped accordingly. We tried everything to arrest the
Visual Display of Quantitative Data, by Edward R Tufte decline – circuit changes, reagent changes. Because recovery and
(1983). He shows the power of simple graphs that remove revenue was down, we did what all good operators do – we
‘chart junk’ – remove every bit of ink that doesn’t increased tonnage. Which made grind size coarser. We made
communicate something useful about the data, eg more concentrate, but metallurgy was on a downhill spiral.
unnecessary grid lines, shading, legends, borders and special We did an enormous amount of work to define the problem –
effects. A particular annoyance is the use of two- or laboratory work, future ores testing, pilot plant work, reduced
three-dimensional graphs to portray one dimensional data, tonnage trials in the plant. We did huge amounts of quantitative
eg using prisms to represent a single measure like recovery. optical mineralogy, then automated mineralogy via QEM*SEM,
Not only is this unnecessary, it is misleading, since the eye
which was just emerging. The evidence was absolutely
perceives a volume rather than a linear measure, thereby
overwhelming – we had to grind finer.
magnifying the difference in numbers. Software packages
like PowerPoint and Excel make it easy to stylise your data. From the time we had irrefutable evidence, it took us about five
Resist the temptation! For example, Figure 3 shows the years to get the project approved. There were two barriers – one
standard chart template offered by PowerPoint. Figure 4 we knew about, the other we didn’t:
plots the same data using Tufte’s principles on chart junk. • We knew that money was tight, there were a lot of
Which one lets the data speak simply and quickly? competing projects, and the Board had to impose a very high
• Put yourself in their shoes. They are smart people, but hurdle rate of return to ration available capital. At times
maybe none of them are metallurgists. You are so familiar projects needed a 100 per cent rate of return. We knew our
with mineral ‘liberation’ that you don’t even consider it project gave a guaranteed 30 per cent return, but there was
jargon. But maybe they get a mental picture of women no way we could do a project of this scale for a 100 per cent
burning bras, or the liberation of Iraq. return.

198 Spectrum Series 16 Flotation Plant Optimisation

 CHAPTER 11 – ECONOMICS AND COMMUNICATION

• In hindsight we may have overcome this problem if it hadn’t
been for the second barrier that we didn’t know – the Board
had no idea what we were talking about!
The communication gap
Time after time we would present progress to the Board, and they
would tell us to do more work. Can’t you just change the circuit?
We hear that columns have improved recovery by five per cent at
ACME mining, why don’t you put in some columns? A supplier
told me they have a new collector that improved recovery by
five per cent at XYZ, why don’t you try that? We were stunned –
we had put enormous resources into quantitative mineralogy and
test work. We had just given them a comprehensive presentation
on a mineralogical circuit balance, showing what was happening
to different composites in every part of the circuit. We had an
ocean of mineralogical data that proved beyond any doubt that
the problem was composite particles. We had shown countless
graphs and tables to demonstrate this – the distribution of
sphalerite binaries with galena, pyrite and gangue for each size
fraction in the lead concentrate, zinc concentrate and tailings. But
the Board sent us back to look at columns and reagents! Did they
think that reagents ground composites? Don’t they know that
columns deal with entrainment, not composites? So we tried
harder, did even more detailed work. Next time we would wow
them with irrefutable scientific evidence.
You can guess the problem. We had done so much work, we
were so close to the data that we overwhelmed the Board with
data. It never occurred to us that they didn’t have a mental image
of what we were talking about. It wasn’t the Board that was
stupid, it was us. When the penny finally dropped, we tore up all
the old presentations and developed a stunningly simple one, as
summarised in Figures 5 and 6. These two slides did in ten
minutes what five years of presentations couldn’t. They look
dated now; these were the days before PowerPoint. But it shows
that you don’t need a fancy PowerPoint presentation; you have to
get your numbers right, then think about how to communicate
them to your audience. They need to be able to see the forest, not
a million trees.
The first of these slides is a simplification – it conveys no
quantitative information about the mineralogy. But it conveyed
the key concept. We left this slide on the screen for a few
minutes, and talked about the options we had for these particles:
• if we recover them, we lose money; and
• if we don’t, we lose recovery.
Immediately the Board said, ‘Well of course you are going to
have to grind it’. No one mentioned collector or columns.
We followed up Figure 5 with Figure 6. This added the
quantitative rigour to the mental picture we had just given. While
we had an ocean of quantitative mineralogical data, we needed to

Zinc losses in final tailing

Period 10 1989/90

Particle size
300 microns

Key:
1. ZnS = sphalerite (grey)
2. Gangue = non-sulfide gangue
3. PbS = galena (white)

Particle size
300 microns

Key:
1. ZnS - easily liberated piece of sphalerite
2. ZnS - more difficult to liberate
3. Gangue (dark)
4. Pyrite (spherulitic)

These particles are coarse, but we do not have the grinding capacity to grind them finer.
These particles assay less that 30 per cent zinc, so we would lose money if we put them
in a concentrate.

Over ten per cent of the zinc in plant feed reports to

final tailings in particles like these.
FIG 5 - The picture that said a thousand words.

Flotation Plant Optimisation Spectrum Series 16 199

CHAPTER 11 – ECONOMICS AND COMMUNICATION

Zinc Recovery is Determined by Liberation

75

70

Recovery to Zinc concentrate

65
Per cent

60

55
Sphalerite Liberation

50
1982 1983 1984 1985 1986 1987 1988 1989 1990 1991

Month/Year
FIG 6 - The rigorous data to support the big picture (Figure 5).

summarise it to communicate the root problem. This is a common
problem with quantitative mineralogical analysis – it is a very
powerful tool, but it is very easy for the trees to obscure the
forest. So we summarised just one aspect of the data – the overall
average zinc liberation achieved in the plant each month. This
was a fraction of the mineralogical data we had – but it showed
the forest. Liberation had been declining for ten years – and our
recovery had inexorably declined with it.
From Figure 5 they understood what a composite particle
looked like and why it needed to be ground. Figure 6 showed the
size of the problem. From then on, approval was a formality. The
project was a great success.
REFERENCES
Tufte, E R, 1983. The Visual Display of Quantitative Data (Graphics
Press: Connecticut).

200 Spectrum Series 16 Flotation Plant Optimisation

 HOME

CHAPTER 12

Operational Geometallurgy
Dean David

MAusIMM, Process Consultant, GRD Minproc Limited, Level 14, 140 St Georges Terrace, Perth WA
6000. Email: dean.david@minproc.com.au

Dean gained a Bachelor of Applied Science (Metallurgy) from the SAIT (now the University of South Australia)
in 1982. For 14 years he worked for CRA (now Rio Tinto), including six years at Argyle Diamonds in both
project and operational roles. Dean joined JKTech as a consultant in 1996, then managed JKTech Consulting
from 1999 to 2003. He joined GRD Minproc as their Process Manager in 2003 and switched to Process
Consultant (with GRD Minproc) in 2005. Dean has conducted many test work, pilot plant and process
implementation projects and managed day-to-day plant operations. Dean’s experience covers hundreds of
projects across the globe and across the mineral spectrum. Expertise areas include geometallurgy,
beneficiation, comminution, classification, physical separations (flotation, gravity, magnetic), sample selection
culminating in test program design and interpretation leading to plant design and and optimisation.

Introduction
Definitive Issues
Available Data
Making Process Sense of the Databases
Planning
Troubleshooting
Expansions
Data Analysis

ABSTRACT on opposite sides of the earth. Within a single orebody it is possible

The orebody is the only asset that mine has at its disposal to generate
revenue. Nature rarely provides neat and consistent orebodies so it is
essential that process engineers understand orebody variability and how this
interacts with the proposed mine plans. A fundamental difficulty in
understanding the orebody is that most of the available information is
geological in nature and not readily accessed by process engineers. To make
matters worse the metallurgical information is typically sparse and
disconnected from the geology. This chapter is an attempt to unlock some
geological mysteries and provide tools for the process engineer to link the
geological and metallurgical data to their maximum advantage, especially
when attempting to optimise base metals flotation plants.
INTRODUCTION
Geometallurgy can be defined as the marriage of geological and
metallurgical concepts in order to provide useful inputs to process
design and operational planning. The focus of geometallurgy is
entirely on the process outcomes but the basis of geometallurgy
should be firmly in the geology.
Orebodies are naturally occurring phenomena with little
consistency from location to location. Orebodies situated within a
few hundred metres can be as geologically different as orebodies
to have primary igneous rock sources, igneous intrusions, faulted
zones, metamorphosed alterations, oxidised zones, alluvial
accumulations and much more. From a geological viewpoint the
various aspects of each orebody can be defined with scientific
definitions and statistical accuracy using well-established
conventions. As with any science, the terminology used in geology
is specific to the discipline and meets its own requirements.
The consequent problem for the process engineer is interpreting
the science with the aim of finding a place for the geological
information within the metallurgical world. It must always be
remembered that geological data is primarily aimed at satisfying
geological needs and not process needs and, therefore, much of the
available geological information is not going to be used by the
process engineer. The process engineer must carefully and
selectively examine the relationship between geological data and
process response before committing to a geometallurgical strategy.
DEFINITIVE ISSUES
Geometallurgy, like any study topic, requires a problem definition
before embarking upon an analysis. The process engineer must

Flotation Plant Optimisation Spectrum Series 16 201

CHAPTER 12 – OPERATIONAL GEOMETALLURGY
define problems and issues that, if solved, generate value for the
organisation. Geometallurgy becomes important when it is
identified that the key to understanding or solving a process
problem lies in an improved understanding of the source and
nature of the ore being fed to the plant.
Examples of geologically based issues that regularly arise for
process engineers in base metal sulfide operations include:
• inability to predict and manage semi-autogenous grinding
(SAG) mill feed rate variability,
• inability to predict and manage concentrate grades,
• inability to predict and manage minor elements in concentrates,
• simultaneous control of plant feed grade and throughput rate,
• planning for throughput expansions, and
• planning for new oretypes or orebodies to be treated in an
existing plant.
The economic importance of an issue is determined by
estimating the value that could be realised if the issue was
eliminated or addressed successfully. The potentially redeemable
value becomes the basis for justifying the resources that need to
be invested to solve the problem. For example, stabilising an
erratic SAG mill feed rate typically results in an average
throughput benefit of five per cent. A five per cent throughput
benefit achieved through operational stabilisation typically
translates to a five per cent revenue improvement with little or no
real capital or operating cost penalty. Additional revenue at little
or no cost equals profit. Assuming a two year payback is
acceptable to management it should be justifiable to invest seven
to ten per cent of annual revenue in a single solution that provides
a five per cent throughput benefit by stabilising the SAG mill feed
rate. However, this is unrealistic and all investment decisions are
tempered by risk assessment and by other projects competing for
attention and the same slice of the profit cake. For example, on a
plant having an annual revenue of A$100 M it may be realistic to
expect management to commit A$1 M to a project with good
potential to increase SAG mill throughput by five per cent.
Establishing an economic case is relatively easy compared to
providing the practical pathway to solving the problem. Often,
the opportunity to apply a large sum of resources to a problem
leads to the implementation of an expensive and generic
off-the-shelf solution. On occasions this is appropriate and
successful but in many instances it is not, and it could even have
an overall negative economic impact. A greater degree of success
is often achievable by understanding the problem both in detail in
breadth, and then applying a solution fit for the purpose at hand.
The first step to understanding the problem in detail is defining
the goal. In this example the goal is stabilised SAG mill feed rate.
What are the definitions of instability and stability in this
context?
Instability, at its worst, involves stop-start operation where it
should be continuous. More commonly instability is seen as a
boom-bust operational cycle where the plant is pushed to its
throughput limits and then beyond, resulting in the need to
significantly reduce throughput to bring the circuit back under
‘control’. Often the ‘boom-bust’ throughput variations exceed
15 per cent of average feed rate and can be as large as 30 per
cent. The average throughput that results is usually more than ten
per cent below the maximum achievable in the circuit. As well as
a loss of feed tonnes, the other major effects of this instability are
erratic flotation feed stream properties including F80 values, per
cent solids and volumetric flow. The economic effects are a loss
of feed tonnage coupled with a loss of flotation performance
resulting in less that optimal recoveries and grades. A general rule
for troubleshooting any process is that any upstream instability
202 Spectrum Series 16
that has the definite potential to cause problems should be
addressed before attempting to solve apparent in-stream or
downstream problems.
Stability is then defined by ‘smooth operation’ of the plant.
However, when is a process operating smoothly? Smooth
operation of a ball mill circuit may be seen as continuous
operation with the feed rate varying by one to two per cent on an
hourly basis. However, operation of a SAG mill is usually
accepted as being smooth if hourly throughput variations are
within four to eight per cent. The reason for the difference is that
the SAG mill throughput rate has a greater dependence on ore
properties than does the ball mill.
The next step is to determine what is causing the problem.
There are a number of possible causes for an unstable SAG mill
feed rate, including:
• inappropriate manual control actions,
• inappropriate automatic control actions,
• poor feed rate control capability,
• poor or no control of multiple stockpile reclaim feeders,
• lack of variable mill speed capability, and
• poor grinding circuit equipment design.
None of these examples relate primarily to the orebody
geology and each should be dealt with using proven process
operation, process design and engineering practice. If any of
these issues is a problem then elimination may achieve the target
throughput improvement without entering the realm of
geometallurgy. These are relatively easy problems to solve as the
technical solutions are readily available from vendors,
engineering companies and experienced process engineers.
If, however, the engineering problems are addressed but the
feed stability problem persists then the primary cause of
variability is probably related to either the hardness or the
competence of the ore. Achieving feed rate stability will then
only be possible if a method is available to predict and manage
the hardness and competence of the ore that is delivered to the
plant for processing. A prerequisite for understanding what is
being delivered to the plant is an understanding of what makes an
ore competent or incompetent with respect to SAG milling and
what makes it hard or soft with respect to fine grinding.
The operating process engineer will have available a number of
historically developed devices for understanding the
comminution properties of the ore ranging from simple to
complex. The earliest of these methods would have been
developed by process engineers during the design phase of the
project. Subsequently, operating process engineers would have
arrived at their own conclusions about the main drivers of plant
throughput and developed new or modified models. The job of
the process engineer is to determine which of the available
predictive methods provides real information about the ore
properties and which methods are ineffective.
The ultimate proof of the method is in its use. The definitive
question is ‘Is it possible, using the chosen method, to
successfully predict the SAG mill feed rate ahead of time for a
range of plant feed situations?’ If not then the method is either
wrong or incomplete. If the method can successfully predict feed
properties then there is probably a management issue that is
causing the feed variability, such as poor implementation of the
predictive method or poor control over ore sources and feed
blending.
On the assumption, often true, that the prediction of ore
properties is ineffective in practice, it is then necessary to
determine if a predictive method can be developed at all. It is here
that the process engineer enters the realm of geometallurgy.
Flotation Plant Optimisation

AVAILABLE DATA
The starting point for any geometallurgical investigation is the
geological model for the orebody and especially its relationship
to the mine model. The geological model contains drill data while
the mine model contains ore and waste blocks derived from the
drill data. Together this is the biggest geometallurgical asset for
the process engineer but, for some, it can be the biggest liability
in such an investigation. Two characteristics of drill data libraries
make them difficult for process engineers to work with: their size
and their complexity.
Geological data libraries (drill libraries) are much larger than
the databases process engineers usually work with, such as tables
of metallurgical test results. Geological databases can contain
data derived from 20 000 m of drilling during the early
exploration phase of a project and during the operational phases
the database can grow to more than 200 000 m. At the simplest
level the database usually contains multi-element assays for every
metre or two metres of drill intersection. However, the database is
also likely to contain information related to lithology (rock type),
alteration characteristics, colour, mineralisation, veining,
competence (RQD), point load index (PLI, a measure of rock
strength), specific gravity (SG), positional survey and any number
of other measures depending on the ore type involved.
The size of such a database makes it difficult to manage and
presents challenges when it comes to isolating useful information.
The tools used to manage drill databases are geologically based
and rarely suit the requirements or experience of process
engineers. Another complicating factor in the operational stage of
a project is that a large portion of the database may be irrelevant
as it represents ore that has already been mined.
The complexity of the database is the second negative factor
for process engineers. The complexity in terms of the multiplicity
of data sets has been mentioned previously. More problematic
than the number of data sets is the variety of bases on which those
data sets are arranged. For example, the assays will be determined
on a regular one or two metre interval basis but the lithology is
likely to be provided as irregular intervals, the lengths of which
relate to the extent of each lithological unit. Within the geological
database there may be five or more different data set tables, each
with their own individual interval set. Further complication can
happen if the geological model has been built up by a number of
successive companies or geologists, each with their own ideas
about what data is important and what is not.
To make the drill library database accessible to process
engineers it is best to use geological software to generate a single
suitable database. Packages such as GenSys can de-survey
geological data sets so that all properties can be reported on a
common spatial coordinate (ie XYZ) basis rather than on a
multiplicity of hole and depth bases. The ideal outcome is for each
individual record of the geometallurgical database to have a unique
spatial address and include every measured data type that has any
chance of being useful. The process engineer can then analyse the
single database using appropriate tools such as spreadsheets.
An accessible data set is only useful to the process engineer
when coupled with additional information about the orebody and
how it is being processed. In an operating mine the economic
limits of the orebody will be well understood by the geologists
and the plant ore delivery schedule is defined by the short-term,
medium-term and long-term mine plans.
A spectrum of mine-related data will exist with relevance
ranging from the imminent to the distant future, the available
mine planning data probably falls into the time-based categories
of imminent, short term and long term as per below.
• Imminent – ore that is ready to be processed within the next
week and is likely to be either blasted, ready for blasting or
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 12 – OPERATIONAL GEOMETALLURGY
in stockpiles ahead of the primary crusher. This ore is the
subject of the daily production planning processes on site.
This ore can be well defined in terms of actual grade through
assay of blasthole drill cuttings. Comminution is either based
on the test result for the closest available metallurgical sample
or based on assumptions about the properties of ore types.
Factors that impede or blur the process engineer’s
understanding of imminent ores include medium-term and
long-term ROM stockpiles, blending stockpiles for plant feed,
lack of a blasthole sample program, lack of blasting (ie free
dig ore), poor in-fill drilling coverage, multiple or numerous
ore sources (typically a problem for complicated underground
mines) and a high degree of short distance variability in the
orebody (as may be seen in vein type deposits).
• Short term – each mining operation has a short-term
planning process to make decisions about the upcoming
waste and ore allocations. The basis for short-term decision-
making is usually in-fill drilling conducted during the
construction and operational phases of the project. In-fill
drilling is designed to fill the gaps between the exploration
drillhole data and allow final ore/waste decisions to be made
about each block in the mine model. It is likely that mine
block sizes (the volume of ore considered to have similar
properties and is assumed to be mined as a unit) will be
reduced in size to better match the upcoming production
schedule. The in-fill drilling data set will certainly provide
grade and problem element data and, in some cases, it may
provide comminution data. The mine block properties are
recalculated using all available drill information and blocks
are then designated as ore or waste.
The most significant complicating factor influencing the
process engineer’s understanding of short-term planning data
is the mathematics used to derive the block properties. It is
essential that the process engineer understand the source of
the data underpinning short-term planning and understand
how it has been manipulated on its way to becoming block
data. Manipulations can include truncation of high-grade
spikes, data smoothing, data kriging (especially for
exploration data that is used to define block properties
because new measurements have not been conducted during
infill drilling, often comminution data falls into this category)
and data exclusions. The ideal data sources for the process
engineer are the short-term planning block predictions and
the raw drill data used to define the block. The full extent of
smoothing is then evident and the ore variability expectations
can be assessed by the process engineer.
• Long term – the long-term mine plan is essentially an
extension of the final operational plan arising out of the
exploration phase. The ore included in the long-term plan is
usually based more on geological orebody models than high
density drill data. The presence or absence of ore in a
particular location, its associated grade and recovery, its
lithology and alteration characteristics are all much less
certain than the data available for ore that is to processed
within the week or mined within the month. Much of the ore
will not even classify as proven under geological reporting
regulations. The long-term plan is typically updated
annually, firstly to reflect new drilling that is designed to
convert probable ore into proven ore and secondly to reflect
actual mining progress.
The long-term plan is most useful for gaining an under-
standing of the overall orebody and mine shapes,
documenting the proportions of the major ore types that are
expected in future years and understanding trends with time
and location for critical measures such as grade, impurities,
hardness and competence.
203
CHAPTER 12 – OPERATIONAL GEOMETALLURGY
When using block model data it is important to note a number
of mathematical facts:
1. The properties assigned to an ore block are dependent
upon the measured properties of a few drill intersections.
It is possible that there are no drill intersections within a
particular mining block.
2. Geostatistical techniques, especially kriging, are used to
estimate the distribution of geological properties between
drill holes.
3. Some ore block properties may have been calculated
using complex equations that may have an unintended
outcome.
4. Some ore block properties will be dependent upon the
location of the block relative to a geological interface (eg
above or below water table, oxidised versus fresh ore
zone, etc).
5. Some of the block properties, or even the underlying data,
may have been filtered or truncated to avoid illogical or
extreme outcomes.
A good check when first encountering a block model is to
look at both the model and the underlying geological data in
the same orebody location and ensure that there is a
reasonable relationship between the two. In general, the
predictive methods are reliable but it is always advisable to
verify information, especially where important decisions are
being made.
The weakest link in the available data for geometallurgical
analysis is invariably the metallurgical data set. It is rarely, if ever,
practical to analyse the orebody metallurgically to anywhere near
the data density that is available geologically. For example, an
orebody of 100 Mt with 10 000 m of intersections identified
within ore has a geological data density of 100 samples per
million tonnes of ore. Assuming there are three major lithologies
that have been tested metallurgically and each lithology has had
50 variability samples tested for flotation properties and
grindability, the metallurgical data density is only 1.5 samples per
million tonnes of ore. At 5 Mt/a plant throughput (for example)
each year’s production is effectively based on the results from 7.5
metallurgical test samples compared to reliance on 500 metres of
drill intersections.
One ore block may represent a few days to a few weeks of ore
so the block model is typically not useful for short-term planning.
Usually a subset of the block model is populated with infill drill
data and then smaller mining blocks are chosen for medium-term
planning purposes. A subset of the medium-term plan can then be
populated with blasthole assay data for production grade control
purposes. In a few instances the metallurgical database is
continually updated with preproduction testing. Generally, the
testing gap between geology and process continues to widen as
the project advances. Fortunately the amount of available process
data begins to catch up once production begins. It is at this point
where the adequacy, or otherwise, of the metallurgical predictions
is revealed.
It is important for the process engineer to become familiar with
the planning processes and the available data sources on site.
Perhaps more importantly, the process engineer should become
acquainted with the owners of the various databases, the
individual geologists and mine planners. Databases always have
quirks and skeletons so it is best to have an expert assist you to
navigate to the data sources essential for your problem solving.
It is not necessary, or desirable, for process engineers to become
users of mining or geological software. Usually specific data is
required to solve a problem and it is preferable to extract the
204 Spectrum Series 16
necessary data from the sources and analyse it off-line using
familiar tools such as spreadsheets. Although it may be useful to
learn how to use the specialised software it is not normally
efficient. All modern geological and mining software has excellent
data export and interchange capabilities that should be exploited.
MAKING PROCESS SENSE OF THE DATABASES
The disparity between the geological and process databases
presents a problem and an opportunity. The problem is that the
metallurgical database is comparatively small and it is not going
to get larger very quickly. The opportunity that exists is the
potential to link the databases together and make some
metallurgical use of the geology.
Linking the metallurgical database to the geological cannot
progress without a critical mass of metallurgical data. There is no
real point proceeding if all that is available is a few tests on each
of the major ore types. It is especially problematic if these few
tests do not reveal unique characteristics about the ore types. In
many operations there is a library of process test data that is
available for interpretation but may not be gathered together in
one place. The starting point should be the available data and
making the most of it rather than embarking on a new test
program.
One criticism of existing metallurgical data that is often heard
is that because the test was done many years ago the results may
not be reliable. In a few isolated instances, and only because of
specific circumstances, this may be a valid criticism. However, in
general this is a foolish first position to have as you may be
ignoring the most valuable data resource you have. Saying that
test data is unusable because of age implies that one of the
following is true:
• the test procedure used was at fault or is irrelevant to the
existing flow sheet,
• the operator that conducted the test was at fault,
• the ore tested has no relevance to current or future operations,
• the way the ore was tested (ie inappropriate blending)
devalues the results, and
• there is no information about what the sample represents
geologically.
In the absence of a clear problem it is advisable to assume all
past test results are valid until proven otherwise. Collect all past
test data and bring it into the geometallurgical analysis.
Historical metallurgical data is often characterised by its
diversity of form. Each process engineer or laboratory
superintendant is likely to have had their own ideas about the test
methodology, test products generated, the assays conducted and
the calculations performed. It is essential to get back to raw data,
especially test conditions, product masses, process times and
original assay data. It is also worth noting that there may be
comments by operators that are relevant and that even
identification of the operator can be a valuable piece of data.
Once the body of test data is identified the metallurgical
database can be built. Although time consuming it is usually
better to be comprehensive rather than selective when compiling
the database. Critical factors for interpreting the test work often
emerge during the analysis process so it is better to include all
possible raw data. It is not necessary to incorporate calculated
data as this can be derived as required. The metallurgical database
should also include reference data such as test numbers, dates,
sample numbers and ore types. As the aim is to link back to the
geological database is also essential to incorporate any links that
may exist back into the geological database such as drill hole
numbers and downhole depths. Try to standardise the data so that
Flotation Plant Optimisation

it is possible to analyse the metallurgical database using
spreadsheet or database tools. It is recognised that this can be a
daunting task where an extensive library of tests exists. However,
the usefulness of the metallurgical database alone is enough to
justify the investment, regardless of the multiplied benefits once
the geometallurgical linkages are established.
Establishing linkages to the geological database is entirely a
case by case process. Both the metallurgical database structure
and the geological database structure are likely to be unique so it
is not possible to list the methods by which they can be linked.
However, a number of aspects of each database are standard and
provide a starting point.
Geological databases typically contain:
• drill hole numbers;
• references to properties (assays, lithology, etc) by depth;
• hole locations (collar positions);
• hole directional information (dip, downhole survey, etc); and
• ore properties – possibly including:
• assays,
• lithology,
• alteration,
• SG,
• magnetic properties,
• radiometric properties,
• mineralisation (ore and gangue minerals),
• RQD,
• fracture frequency (FF), and
• point load index (PLI), etc.
In order of priority, the most important information after the
positional data is the set of chemical assays followed by the
lithology. Where there is limited comminution data from
metallurgical testing, then drill core structural information (RQD,
FF) and the strength data (PLI) become more important.
Note that the above description is typical of a database that has
not been subject to metallurgical influence. If detailed
metallurgical testing has been conducted on the orebody for
geometallurgical purposes some of the following may appear
either in the drill database or in the mining block model:
• drop weight index (DWI),
• SAG power index (SPI),
• bond (or equivalent) work indices,
• liner abrasion indices (eg Ai),
• ore abrasion index (ta),
• mill feed F80 predictions,
• recovery predictions for products,
• concentrate grade predictions,
• SAG throughput rate (in an existing operation), and
• achievable grind P80 (with existing mills), etc.
If such information does exist then a few checks need to be
made to ensure the data is useful in an operational sense. For
example, in an existing operation it should be relatively simple to
check if the geometallurgical predictions are useful to the
operators and planners. If the data is ignored by the operators
then it is either a poor predictor of performance or systems have
not been put in place to ensure that the metallurgical information
is made available to those that need it. The latter is often the case
when the data has been collected during the design period for the
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 12 – OPERATIONAL GEOMETALLURGY
purpose of selecting major equipment but no operationally useful
systems were implemented at plant start-up.
If it is found that the geometallurgical predictions do not match
operational reality then there is something wrong with either the
data that is being used, the interpretation method or the method of
distributing the data across the geological databases.
Examples of inappropriate use of metallurgical data include:
• the use of Bond ball mill work indices to predict SAG or AG
mill throughput rates;
• the adoption of unverified relationships between ore
measurements and operational outcomes;
• inappropriate assignment of metallurgical ore properties to
geological categories, such as lithologies, where there is no
proven linkage between the two; and
• predictive methods that ignore key factors, for example
prediction of copper concentrate grade based on head grade
without reference to copper mineralogy.
In a modern operational environment there is usually enough
data available to verify or discredit these predictive systems. If
not, then systems should be implemented and instruments
installed to allow, at a minimum, the verification of the predictive
information that is meant to drive decision-making in both the
mine and the plant. The saying ‘if you can’t measure it, you can’t
manage it’ is especially true when it comes to geometallurgy.
Unfortunately, in many instances, it appears that measurement is
being done but because the measurements or the interpretation
are inappropriate the consequential management is misdirected.
Where there is no metallurgical data in the geological
information systems the geometallurgist must endeavour to
insert some in a useful manner. The success of establishing
geometallurgical links will depend on the type, quantity and
quality of the metallurgical data, the spatial nature of the
metallurgical samples and the ability to link the metallurgical
data to geological properties and locations.
Here is a theoretical example of establishing a set of useful
geometallurgical linkages for a copper mine, having mixed
copper sulfide mineralogy and using a SAG/Ball mill
comminution circuit, which experiences significant swings in
throughput rate.
• Decide what metallurgical predictions are necessary for
effective management of the process. In this example the
following predictions are required:
• plant feed rate based on both SAG and ball mill
limitations,
• copper concentrate grade,
• copper recovery, and
• gold recovery.
Note that fundamental metallurgical inputs such as feed grade
and ore type are assumed to be already available because they
are geological and mining model outputs rather than based on
specific metallurgical test work or interpretation.
• Decide what measurements are useful for predicting plant
performance.
• JK Axb has been proven to be an effective predictor of
SAG throughput at the site so a test such as DWI can be
used to economically establish a metallurgical database.
• BWI has been shown to be useful for predicting ball mill
performance well so either a full BWI test or a simple
modified BWI will be used.
• Copper concentrate grade is strongly related to the
copper mineralogy so either directly measuring
205
CHAPTER 12 – OPERATIONAL GEOMETALLURGY
mineralogical mix or inferring mineralogical mix from
Cu:S ratio are to be used. As the geologists have included
copper sulfide mineralogical proportion estimates in the
database and sulfur assays are available both methods of
predicting concentrate grades will be used and the most
appropriate will be chosen after a period of verification.
• Copper recovery is also strongly related to mineralogy
and it is known that recoveries are lower when chalcocite
and covellite are present. It is also related to oxidation,
which is a function of depth, alteration and ore type.
Copper recovery is also well predicted by simple
laboratory flotation tests on samples. As copper recovery
has been highly variable and prediction has been only
moderately successful from the geology a simple rougher
flotation test will be carried out to estimate recovery.
• Gold recovery is variable and difficult to predict. No
current methods are reliable. The rougher flotation test
will also be used to attempt to predict gold recovery.
• Choose calculation methods for the predictions.
• The plant feed rate prediction will be based on a
calculation that determines if the SAG or ball mill is
going to limit production based on DWI and BWI and
applying operating constraints such as available milling
power, maximum acceptable flotation feed F80 values and
maximum acceptable flotation circuit feed rate.
• An established relationship between plant feed Cu:S ratio
and final concentrate grade will be applied to the
geologically based block model Cu and S assays. A
second prediction will be based on the mineralogical
information logged by the geologists. Inconsistencies
will be flagged for potential investigation.
• For 30 days the plant feed will be sampled hourly to
provide shift composites. Shift composites will be
subjected to a standard grind and flotation test. The
flotation test Cu and Au recoveries will be compared to
the shift recoveries in an attempt to develop copper and
gold flotation recovery predictors. In addition the ore
source for the shift or period will be noted. Additional
flotation tests will be conducted using freshly drilled
diamond core samples to firm up the relationship with
ore from known locations and geological characteristics.
• Implement the predictive framework.
• The predictions can be implemented in one of two
locations. The first, and most comprehensive is in the
geological database. The second, and most convenient, is
in the mining block model.
• Adding the predictive framework to the geological
model invokes considerable complexity because
the useful process outcomes of the predictive
calculations will not be needed in the geological
model but in the form of short-term planning outputs
from the mine block model.
• Adding the predictive framework to the block model
is simpler because the block model is numerically
smaller than the geological model and has been
smoothed to eliminate problematic data.
• Adding the predictions into the geological model
requires that one or more of the many data tables be
expanded to include the predictive calculations. The data
table to be expanded depends upon the inputs to the
calculations. For example, if the calculations are assay
based then the assay table should be expanded. If the
calculations are ore type based then it may be necessary
to expand the lithology table.
206 Spectrum Series 16
• The prediction of SAG mill feed rate will be made
using the DWI and BWI measurements.
• It has been found that DWI is related to lithology
and to depth so each lithology will be assigned a
base DWI value which will be increased in
proportion to depth.
• Similarly, the BWI value has been found to be
related to lithology and depth with the exception
that one lithology is considerably harder in the
north than the south. This lithology will be
divided in north and south types and each have its
own base BWI and depth relationship.
• As a first estimate the SAG feed will be the
minimum throughput rate based on DWI versus
available SAG power and BWI versus available
ball mill power. The estimate will be capped to
maximum and minimum throughput rates.
• Copper concentrate grade predictions will be made in
two independent databases, the assay database and
the mineralogy database.
• In the assay database a column will be inserted
which applies a calibrated theoretical relationship
between the Cu:S ratio and final concentrate
grade.
• In the mineralogy database a column will be
added that uses the stoichiometric copper
proportion of each mineral to calculate an overall
Cu concentrate grade.
• Both grade predictions will be carried through
the predictive process and reported separately.
• Copper recovery predictions will initially be made in
the assay database on the basis of constant tails
grade. This relationship will be updated following the
results of the planned flotation test program.
• Gold recovery predictions will also be made in the
assay database on the basis of constant tails grade
and will be subsequently updated.
• The block model divides the orebody into mineable
quantities of ore (blocks) to which properties are
assigned on the basis of internal or nearby drill hole data.
Including the desired predictions into the block model
requires that the necessary information is carried through
from the geological model for the calculations.
• For SAG mill feed rate prediction it is necessary to
carry DWI, BWI and SG data. It is also necessary to
carry lithology as this determines the equation to be
applied to the data.
• For copper concentrate grade prediction it is
necessary to carry per cent Cu, per cent S and copper
mineralogy.
• For copper and gold recovery prediction it is
necessary to carry the per cent Cu and g/t Au values.
• Add the necessary calculations into the block model
set-up so that the predictions are made and can be used
immediately in block value calculations.
Assuming the set-up is mathematically valid and there is a
enough data to make the predictions statistically useful it is now
possible to add the desired predictions of SAG throughput, copper
concentrate grade, copper recovery and gold recovery to the
short-term planning outputs. Equally important is setting up a
comparison to see how good the predictions matched the actual
production outcomes.
Flotation Plant Optimisation

PLANNING
The most obvious use of the predictive data is for short, medium
and long-term metallurgical planning. Applying the predictions to
the long-term planning process should reveal future constraints
on throughput or production that will need to be accommodated.
Predictions of this nature feed into the capital planning process
and also provide early warning of systemic problems such as
hardening ore or changing mineralogical mixes.
Using such tools the process engineer can participate more
fully in the long-term planning of the overall project. For
example, hardening ore without a compensating grade change
will lead to a revenue reduction that must be counteracted.
Counteraction measures range from plant optimisation to plant
expansions or even to new plant construction. Marketability of
the concentrate is influenced by its grade, another of the possible
predictions. A trend to lower grade concentrates may lead to early
warnings for customers, a reduction in revenue expectation,
planning for increased throughput or sourcing of new customers
for the concentrate.
Equally important can be the educational aspect that arises
from geometallurgical planning. Provided the predictions are
shown to be reliable the mining engineers and geologists learn
what is important to the process plant and are able to contribute
more effectively to the overall site optimisation. For example, the
geologists may change the way they look at the orebody and
report ore characteristics. Alternatively, the mining engineers may
introduce, or modify, blending practices to smooth the process
plant operation by avoiding the delivery of undesirable ore
packages.
Without an integrated predictive geometallurgical system it is
difficult to participate fully in the site planning process beyond
the fundamental requirement of grade and quality targets.
TROUBLESHOOTING
Troubleshooting is an ongoing responsibility of production
process engineers. Any system that assists in the troubleshooting
process not only makes life easier but also improves the company
bottom line by solving problems faster. An integrated predictive
geometallurgical system would facilitate rapid troubleshooting by
providing a consistent basis for sourcing and analysing data. In
addition, a reliable integrated system would have a significant
benefit of having established trust with the mining and geological
departments and also with the site management. Not only would
it be possible to generate the necessary data to build a case for
change, but it should also be possible to get rapid agreement on
implementation.
More typically there is no available system where data can be
instantly sourced to back up metallurgical arguments. In these
situations it is necessary to carry out ad-hoc geometallurgical
investigations to complement a particular investigation. Examples
may include tracking down periodic recovery drops, reducing
concentrate grade variability or tracking down the source of an
impurity such as arsenic or fluorine.
An ad-hoc geometallurgical investigation would typically
follow a similar pattern to that used to set up an integrated
system. Even though the particular investigation is unique it is
advisable to set up generic tools that can be used in future
investigations. Begin by tracking back as reliably as possible
from the symptom of the problem to the source. For example,
high arsenic in concentrate is related to high arsenic in feed and
arsenic in feed may be the result of any one of a number of
geological features, such as particular ore types or alteration
zones. Tracing back from the problem to the source allows the
issue to be managed by flagging the geological source of the
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 12 – OPERATIONAL GEOMETALLURGY
problem and implementing a suitable solution. This could include
blending the high arsenic ores with low arsenic ores or it could
involve implementing a processing solution such as depressing
the arsenic minerals during flotation by the use of cyanide.
An ad-hoc investigation still requires a good level of
understanding of the geological information and a good working
relationship with the geologists and mining engineers. It also
requires a full understanding of the factors that influence the ability
to establish a link between the symptom and the source. Common
complications arise from the delays and mixing caused by storage
systems such as stockpiles, fine ore bins, ROM storage and
concentrate storage. These necessary devices can make it virtually
impossible to isolate the source of a problem in the mine. It may be
necessary to arrange for extended runs on a single ore type to
ensure the ore source is unequivocal during an investigation.
Other complications can arise from generic property
assignment to ore types. In these instances it is necessary to test
the assumptions by real testing. For example, if a particular ore
type is assumed by others to have a constant BWI value then take
five samples of that ore (as defined by the geologists) from
different locations and test them. Repeat the exercise on other
important ore types. If the results support the assumption then
continue to use it. If they disprove the assumption then it may be
necessary to modify site systems and re-educate people
depending on this assumption. Where the assumption is found to
be invalid, and it is necessary to rely on the BWI value in
reaching a solution to the problem at hand, then the situation
must be corrected. Additional BWI tests on ore samples specific
to the investigation are easily justified.
Statements such as ‘all this ore type is the same’, or ‘there is
not much variability in the orebody’ must be tested by
examination of available data or by collecting more. In most
cases these statements prove to be false and can derail
investigations at the start.
EXPANSIONS
Plant expansions are special cases where geometallurgy is
essential to achieving the intended outcome. Many expansions
have been less than successful because the ore to be treated did
not meet the expectations of the designers. This is a direct result
of a poor understanding of the future orebody and the future ore
properties.
The first rule with expansions is to check all assumptions that
drive the process design. If the expansion is throughput related it
is essential to check that tests have been carried out on samples
representative of the ore that is to be treated by the expanded
plant. In the absence of supporting data it is not adequate to
assume that because the ore blend will be 30 per cent X and
70 per cent Y that the properties can be predicted from past
experience. Take multiple samples of X and Y from the ore zones
to be treated by the expanded plant and subject them to all the
necessary tests to prove the design assumptions.
To check if this additional geometallurgical expense is
necessary it is relatively simple to repeat the financial
calculations with ten per cent lower throughput than predicted or
five per cent lower recovery.
An expansion decision must be based on a reasonable
guarantee of a return from the ensuing revenue stream. The same
principles that are applied to exploration companies, proving the
presence of the ore and establishing its grade, are the first steps in
justifying an expansion. In most cases this is well established.
However, in some cases the ore in question may lie in the inferred
category and will need to be elevated to a higher status before
proceeding. Note that if the ore is inferred then it follows that all
available ore properties are inferred.
207
CHAPTER 12 – OPERATIONAL GEOMETALLURGY
If the expansion is based on treating future ores according to
established lithological categories then check that lithology has
been a useful indicator of process properties in the past. If this
has not been the case then the future will be no different to the
past and it will be essential that metallurgical tests are conducted
on samples of the future ores.
When carrying out a metallurgical test program on the future
ore there will be a set of obvious tests to perform and there will
be some less obvious tests. Perform the obvious tests
(comminution, flotation, abrasiveness, etc) and then carry out
some speculative tests, such as full elemental analysis or
mineralogical analysis of concentrates. For reference it is best to
carry out the same tests on current production samples. These
types of tests are simple and inexpensive checks designed to
identify if there are any new problems that may arise from the
future ore. If problems are confirmed or indicated by the tests
then a more comprehensive test program may be required to
identify mitigation strategies.
Another possibility in preparing for an expansion is to conduct
a plant trial on the future ore. This is highly desirable if it is
known that the future ore differs from the past ore. The existing
plant is the ideal pilot plant facility. The ability to conduct a
meaningful plant trial is contingent upon ore access and ore
volume availability. A plant trial has additional benefits such as
providing concentrate samples for customer evaluation and
providing a clearer evaluation of plant areas that are dependent
upon concentrate availability, such as regrinding, thickening and
filtration. Another less obvious benefit is that the plant operators
are exposed to the future ore. It is usually the operators that
notice operational effects first and the crew should be well
prepared before and debriefed after the trial.
At the implementation stage of an expansion project the major
risk areas must be mitigated to the satisfaction of the disciplines
on site and to the satisfaction of the management team. If the
factors listed below have been established definitively then the
expansion has been underpinned geometallurgically:
• future ore tonnage and grade established,
• future ore metallurgical response is known,
• acceptable concentrates can be sourced from future ore,
• future ore throughput properties are known and have been
used in comminution design, and
• the variability of the future ore is understood.
DATA ANALYSIS
Effective and efficient analysis of geometallurgical data is
essential to any geometallurgical exercise. It is often necessary to
manipulate very large databases and spreadsheets and summarise
them into meaningful sources of information. It is also necessary
to correlate properties from the geological and metallurgical
realms so that trends and indicators can be established. It is in the
data sets that the important information resides, the difficult part
is extracting it. Even more difficult can be communicating the
information to others.
The breadth of available geological data has been described
previously. Also described has been the necessity to develop
analysis tools useful to process engineers. Unfortunately it is not
possible to provide a one-size-fits-all solution to geometallurgical
data analysis as the forms of geological and mining data are as
diverse as the orebodies they describe.
The fist step is to organise the data into a form that can be
manipulated meaningfully. One method (that preferred by the
author) is to arrange all the available geological data into a single
spreadsheet table against a common set of depth intervals. In
addition the data is de-surveyed to convert drillhole numbers and
208 Spectrum Series 16
depths into 3D spatial (ie X, Y, Z) coordinates. This is best done
with the assistance of a geologist, and with direction as to which
data is required the task can be completed in hours or minutes.
Once the database is available it is then possible to begin detailed
analysis.
Spreadsheet packages, such as Microsoft Excel, contain many
useful statistical and data manipulation tools. Probably the
most useful tools for manipulating large and unwieldy
geometallurgical databases are the ‘Pivot Table’ tool and the
‘AutoFilter’ tool. Excel tutorials will not be provided here but all
readers that deal with large data sets of any kind are strongly
encouraged to become familiar with these tools. They have the
ability to extract data and trends from million-cell databases with
a few keystrokes.
Before making use of the data it is essential that each column of
data be checked for unusual features or errors. For example,
sometimes numerical data is actually present as text in the
spreadsheet. Often text information has a minimum length and
what appears to be a three letter code is three letters followed by
ten spaces. Such data anomalies confuse the analysis and the
analyser. Check for data artefacts inserted by the geologists, for
example representing missing data with ‘9999’. Check for entry
errors such as a mix of ppm values in a per cent column or per cent
values greater than 100. There are infinite possibilities for
problematic data in databases. The general rule is if unexpected
results appear during analysis then go back and check if the data
that you thought you were incorporating is in the appropriate form.
It is usually necessary to add additional data columns to the
geological databases. For example, ore and waste are rarely
defined in the geological data because these are mining terms. A
data column that classifies the intervals into simple ore and waste
sets according to a reasonable grade cut-off level is easily added.
A consequent sort on ‘ore’ then removes all the material that
process engineers are not interested in.
In banded or vein orebodies it may be necessary to re-interpret
the ore and waste definitions to match the reality of mining
limitations. For example, ensure that 1 m waste intervals within
an ore band are considered as ore because they will not be
selectively excluded. Similarly, if the minimum mining cut width
is 4 m in a vein deposit then include some waste either side of
significant ore intersections. These limitations are best discussed
with the miners before being implemented in the database. Once
implemented it is essential that a reality check is carried out to
ensure that the effect of the data manipulation is consistent
whenever it is invoked. Check between ten and 20 instances
through the database of any new data manipulations.
In databases compiled by different geological teams over time
there are likely to be duplicate assay columns or lithology
columns. Decisions must be made as to how this is reduced to a
single consistent column describing that property.
Another common manipulation is to round the X, Y, Z data into
useful intervals. For example, it is much more useful to know that
ore is within the depth range RL 100 to RL 150 m than it is to
know that the ore RL is precisely 123.27 m. This then allows the
database to be analysed by depth slices or slices in the north or
easterly directions. It also allows the ore to be examined as
blocks. Blocks defined by this rounding method are usually much
larger than the mining model blocks and are best distinguished by
the term ‘Metblock’.
Another useful addition to the database is to indicate those drill
intersections samples that have been utilised in the relevant
metallurgical test programs. Again, this can be tedious, but it
provides an excellent basis for evaluating the representative
nature of past sampling and is a guide to selection of samples for
future test programs.
If it is not possible to correct data to something meaningful,
then a number of alternatives exist. One is to eliminate the
Flotation Plant Optimisation

offending data record or column by deletion (or removal to a
place so it is then known to have been removed). Another option
is to ‘fix’ the data, such as by averaging the corresponding values
for the records above and below. Another is to move a decimal
point if it is unequivocal that the reported value is not real but is
the result of a typographical slip. Any manipulations should be
noted and the information returned to the geologists. This is
usually valuable information that can lead to improvements in the
integrity of the geological database.
Once verified the database is ready to use. A number of
guidelines are useful here:
1. Identify the ore that is important to the metallurgical issue
being addressed.
Analysing ore that will never be mined or will be mined
15 years from now is rarely useful. Identify, with the
assistance of mining engineers, the ore that impacts on the
decision or problem at hand and then concentrate on it. One
method may be to add an additional column to the database
and flag the important ore with a one or a TRUE. The data
can then be sorted on this basis. Another alternative is to
assign a future year number to ore intervals but this is much
more complicated.
2. Identify the properties that are of most importance to the
metallurgical issue.
These properties then become the subject of data analysis
and guide the analysis to the important aspects of the data
set. For example, in a copper flotation plant the copper
grade is obviously important but so is the sulfur grade, the
arsenic grade and probably the gold grade. Using the data
tools mentioned above, it is then simple to make these
measurements the subject of the analysis.
3. Use the data tools to focus on the important ore and the
important properties.
Using pivot tables it is possible to quickly construct a table
constrained to ore in the relevant time frame and examine
average Cu, Au, S and As grades and identify the number of
intervals identified in each category. It is also possible to
view this data in plan on the orebody by arranging it within
the rounded X and Y data axes categories. It is then possible
to step down through the orebody by selecting an individual
RL value or a set of RL values.
Once the data is in this analysis form it is then possible to
examine the orebody from a process perspective. For example:
Flotation Plant Optimisation Spectrum Series 16
CHAPTER 12 – OPERATIONAL GEOMETALLURGY
• trends of arsenic with location, depth or ore type can be
quickly extracted;
• trends in Cu:S with time can be examined; and
• specific locations for extracting new metallurgical test
samples can be identified and a click will instantly drill
down into the database and show the relevant drill
intersections in that particular Metblock.
It is also possible to use graphical and statistical techniques to
search for correlations between factors, for example plot per cent
As against per cent Au on an X/Y plot to see if gold and arsenic
are related. If a relationship is suspected within a particular ore
type then use the pivot table drill down facility to extract the
relevant records then plot the specific data comparison again.
Tools such as correlation tables can be used to search large data
sets for significant linear correlations between large numbers of
pairs of data columns. Multiple linear regression can be applied
to see if a particular property is a function of a number of others.
Regression tools are particularly useful when investigating the
relationship between the limited metallurgical data set and the
extensive geological data set. Assuming the drill samples that
were subject to metallurgical tests have been correctly identified,
it is possible to extract the geological data for those intervals from
the database, calculate the averages of the interval properties and
correlate them to the measured metallurgical properties. In this
manner it may be possible to identify geological predictors of
metallurgical properties.
Ideally the metallurgical test samples will have been made up
of contiguous intervals of single lithologies, this is much more
useful than single test samples being formed as composites from
across the orebody. Especially problematic is when the only
available data relates to composites formed from across the
orebody and from a mix of lithologies. Both these situations
make it almost impossible to establish any linkages between
geological and metallurgical properties.
One of the most rewarding outcomes for the process engineer
from such an analysis is the deeper understanding that is gained
about the orebody. With this understanding the process engineer
is well equipped to contribute to discussions on the future of the
operation and to converse meaningfully and factually with the
geologists and mining engineers on site. If nothing else is gained
then the improved communication lines across the disciplines is
usually worth the effort.
209
‘The great tragedy of science – the slaying of a beautiful hypothesis by an ugly fact.’
Anonymous

‘. . . to those who taught me that ‘hope is not a method’, and to those

whom I trust will learn to know and think about their process . . .’
 FOREWORD
This book maintains a long tradition by the Australasian Institute of Mining and Metallurgy of publishing
learned works in flotation. Previous landmark publications by the Institute on this important subject are
Principles of Flotation by I W Wark in 1938, Principles of Flotation by K L Sutherland and I W Wark in 1955
and Principles of Mineral Flotation The Wark Symposium, edited by M H Jones and J T Woodcock in 1984.
It is fitting that there has been a strong continuing interest in the science and engineering of the flotation process
in Australia as the industrial practice of sulfide mineral flotation was first developed at Broken Hill in New
South Wales at the beginning of the 20th century. This exciting story is lucidly told in Professor Alban Lynch’s
forthcoming book The History of Flotation, planned for publication by the Institute in 2010. The importance of
flotation to the mineral industry can be measured by Professor Geoffrey Blainey’s evaluation that ‘(i)n the last
thousand years of metallurgy it stands with the cyanide process, and the Bessemer process as one of the three
1
greatest advances’.
I have been a practitioner in the flotation field since 1970 and have been fortunate to see it from different
aspects. From starting in production, then working with some of the early efforts to measure and control the
process, going into applied research to support concentrator operations, dabbling in new devices such as the
development of the Jameson cell followed by exposure to fundamental research on surface chemistry, flotation
cell hydrodynamics and mathematical modelling I am currently a consultant looking at a wide range of
commodities treated by the process in different countries. Flotation is a fascinating subject covering a wide
range of scientific and engineering disciplines that has never ceased to interest me and provide intellectual
stimulus.
There has been a continuing theme during my time in the field in the application of new technologies to
improve our understanding of the process. In Australia this technical understanding has come from research and
academic institutions such as the Commonwealth Scientific and Industrial Organisation (CSIRO), University of
Melbourne, the Julius Kruttschnitt Mineral Research Centre (JKMRC) at the University of Queensland, the Ian
Wark Research Institute at the University of South Australia and the Chemical Engineering Department of the
University of Newcastle. Mining companies both supported research and technology transfer through the
Australian Minerals Industry Research Association (AMIRA) and did significant work in their own right.
From a personal view, in the late 1960s to early 1970s on-stream analysis produced real-time data for the
control of production plants; in the 1970s computers enabled modelling, simulation and control of the process
while in the 1980s new techniques of surface analysis showed what was really happening on the surfaces of
mineral particles though to some observers this may have proved Wolfgang Pauli’s observation that while solids
are the creations of God, surfaces are the work of the Devil! Advances in automated mineralogy, first the
QEMSCAN, followed by the MLA, coincidentally both Australian developments, allowed collection of detailed
quantitative mineralogical data to supplement the traditional method of point counting using an optical
microscope. In the 1980s there was a revival of interest in the ‘chemical engineering’ approach to transport
processes in the pulp and froth phases brought on by the introduction of column flotation exemplified by the
work of Professor Jim Finch and Glen Dobby at McGill University in Canada. This culminated in the important
concept of bubble surface area flux of Gorain at the JKMRC. Throughout this time the size of conventional
flotation machines increased by a factor of over 200, especially rapidly in the last 15 years with more novel
types used in particular applications.
However, the fundamentals of the flotation process have not changed and I exhort people in the field to
remember two fundamental points. These are:
• Flotation is not a metaphysical process; though some of the earlier publications particularly
focusing on chemistry may have made it appear so.
• Flotation outcomes are completely describable by using the axiom of size-by-size mineral
particle behaviour by liberation class.
My career-long journey of discovery of the technical literature of the art of flotation started as a new graduate in
Applied Chemistry without any specific raining in mineral processing, attempting to read the seminal works in
the subject then available in the excellent technical library of Mount Isa Mines Limited. As a notorious
bibliophile I have encompassed most of the important publications though biased towards the ‘practical’ side.
What is missing in the English language literature of flotation is a single up-to-date volume covering all aspects
of the ‘practical art’ as required by a mineral processing engineer either operating or supporting a flotation
concentrator. Books written to date have usually focused on the ‘science’ or chemistry of the process to the
detriment of ‘practical’ aspects with much useful information on the latter being either in monographs or
scattered through many technical papers written for journals and symposia. Even in the age of the internet,
going through this large volume of material is a daunting prospect. In my opinion the only comparable volume
to this book is the venerable Froth Flotation 50th Anniversary Volume published in the USA nearly 50 years
2
ago. Even so it was not as oriented to practical plant problem solving as Flotation Plant Optimisation.
This book is essentially a ‘how to’ manual and will be particularly useful as a conversion course for mineral
processors such as those whose undergraduate training has been primarily in chemical engineering. Its approach
is based on the relevant design and execution of a plant survey followed by the steps required to analyse the
results to a suitable level of detail.
Chris Greet has assembled a great cast of contributors for this book, all authorities in their fields representing
the leading edge of accumulated current knowledge on relevant topics such as applied mineralogy,
geometallurgy, flotation chemistry, process analysis, mass balancing, experimental design and characterisation
of industrial flotation machines. I know most of the individual authors and can attest to their mastery of their
particular areas of expertise. It has been a privilege to have worked with them over the years and participated in
the occasional ‘technical highs’ on the science and application of flotation that have made working in the field
so rewarding.
I commend the book to readers and can assure them that understanding the contents and applying the
knowledge gained to real mineral processing situations will result in improved metallurgical outcomes.
Peter Munro
Senior Principal Consulting Engineer
Mineralurgy Pty Ltd

1. Blainey, G, 1963. The Rush that Never Ended: A History of Australian Mining (Melbourne University Press: Melbourne).
2. Fuerstenau, D W (ed), 1962. Froth Flotation 50th Anniversary Volume (American Institute of Mining, Metallurgical and Petroleum Engineers, Inc:
New York).
 ACKNOWLEDGEMENT
I was reminded recently by my colleague, Jessica Kinal, of the day that the concept for this book was born. We
were visiting a mine in Indonesia when, to use Jessica’s word, I had just ‘bombarded’ this poor metallurgist
with a detailed description of the benefits of recovery-by-size, liberation analysis, pulp chemistry, circuit
configuration and general flotation metallurgy. It was at this point that Adrian Zarantonello suggested that I
should write a book sharing my knowledge.
With the seed planted, and knowing that there are people better qualified than I to write the various chapters I
set about asking for help. I am humbled and grateful to those that contributed. My heartfelt thanks goes to: Bill
Johnson, Rob Morrison, Alan Butcher, Greg Harbort and Sarah Schwarz, Stephen Grano, Ron Woods, Alan
Buckley, Kym Runge, Tim Napier-Munn, Joe Pease and Dean David. Without their knowledge and wisdom the
value of this book would be much diminished.
I am appreciative of Magotteaux’s financial support for this project. The vision Philippe Steinier had in the
early days of the value of this project and how he convinced our masters in Belgium that it was the right thing to
do must be acknowledged. I must also thank my Manager Vincent Werebrouck for his continued support and
friendship, particularly during the harsh times we have experienced recently.
The many metallurgists I have had the privilege to meet and discuss various aspects of flotation with over the
years are too numerous to mention. However, their challenging questions, their thirst for knowledge and their
desire to improve their plant are the motivation behind this publication. I trust that you will gain value from this
work.
The Australasian Institute of Mining and Metallurgy warrant special thanks for sticking with this project.
Special thanks must go to Angie Spry, Jenni Stiffe and Kristy Pocock. It’s been a long road, and I am grateful
for their support, professionalism and their nagging. Next time maybe a cattle prod would come in handy!
Thank you for your patience.
Finally, I would like to thank my family for their love and support, particularly my kids.
Dr Christopher J Greet
 SPONSOR

The Australasian Institute of Mining and Metallurgy would like to thank

Magotteaux for their generous support of this volume.
 MONOGRAPH SERIES
1 • Detrital Heavy Minerals in Natural Accumulates George Baker 1962
2. • Research in Chemical and Extraction Metallurgy Ed: J T Woodcock, 1967
A E Jenkins
and G M Willis
3. • Broken Hill Mines - 1968 Ed: M Kadmanovich 1968
and J T Woodcock
4. • Economic Geology of New Zealand Ed: G J Williams 1974
5. • Economic Geology of Australia and Papua New Guinea - 1 Metals Ed: C L Knight 1975
6 • Economic Geology of Australia and Papua New Guinea - 2 Coal Ed: D M Traves 1975
and D King
7. • Economic Geology of Australia and Papua New Guinea - 3 Petroleum Ed: R B Leslie 1976
H J Evans
and C L Knight
8. • Economic Geology of Australia and Papua New Guinea - 4 Industrial Ed: C L Knight 1976
Minerals and Rocks
9. Field Geologists’ Manual
1st Edition Ed: D A Berkman 1976
2nd Edition and W Ryall 1982
3rd Edition Ed: D A Berkman 1989
4th Edition Ed: D A Berkman 2001
10 • Mining and Metallurgical Practices in Australasia Ed: J T Woodcock 1980
(the Sir Maurice Mawby Memorial Volume)
11. • Victoria’s Brown Coal - A Huge Fortune in Chancery Ed: J T Woodcock 1984
(the Sir Willis Connolly Memorial Volume)
12. • Australasian Coal Mining Practice
1st Edition Ed: C H Martin 1986
2nd Edition Ed: C H Martin 1993
and A J Hargraves
3rd Edition Ed: R J Kininmonth 2009
and E Y Baafi
13. • Mineral Deposits of New Zealand Ed: Dr D Kear 1989
14. Geology of the Mineral Deposits of Australia and Papua New Guinea Ed: F E Hughes 1990
15. The Rocks Speak H King 1989
16. • Hidden Gold - The Central Norseman Story J D Campbell 1990
17. • Geological Aspects of the Discovery of Some Important K R Glasson 1990
Mineral Deposits in Australia and J H Rattigan
18. • Down Under - Mineral Heritage in Australasia Sir Arvi Parbo 1992
19. Australasian Mining and Metallurgy Ed: J T Woodcock 1993
(the Sir Maurice Mawby Memorial Volume) and K Hamilton
20. • Cost Estimation Handbook for the Australian Mining Industry Ed: M Noakes 1993
and T Lanz
21. • History of Coal Mining in Australia Ed: A J Hargraves, 1993
(the Con Martin Memorial Volume) R J Kininmonth,
C H Martin
and S M C Saywell
22. Geology of Australian and Papua New Guinean Mineral Deposits Ed: D Berkman 1998
and D Mackenzie

Copies of all books currently in print can be obtained from

The Institute office - Tel (03) 9662 3166

Key: • Out of print

23. Mineral Resource and Ore Reserve Estimation — Ed: A C Edwards 2001
The AusIMM Guide to Good Practice
24. Australian Mineral Economics Ed: P Maxwell and P Guj 2006
25. Geology and Exploration of New Zealand Mineral Deposits Ed: A B Christie and 2006
R L Brathwaite

Copies of all books currently in print can be obtained from

The Institute office - Tel (03) 9662 3166

Key: • Out of print

 PUBLICATIONS OF THE AUSTRALASIAN INSTITUTE
OF MINING AND METALLURGY

CONFERENCE, SYMPOSIUM AND MISCELLANEOUS

1972 1/72 SI * Project Evaluation and Management, Melbourne
2/72 * Annual Conference, Newcastle
1973 1/73 S2 * Subsidence, Illawarra
2/73 S3 * Mine Filling, North West Queensland
3/73 S4 * Transportation, Sydney
4/73 S8 * Mine Fires, Southern Queensland
5/73 * Annual Conference, Western Australia
1974 1/74 S5 * Support in Pillar Extraction, Illawarra
2/74 S6 * Recent Technical and Social Advances in the North Australian Minerals Industry,
North West Queensland
3/74 S7 * Pellets and Granules, Newcastle and District
4/74 * Annual Conference, Southern and Central Queensland
1975 1/75 S9 * People and the Mining Industry - The Future, Broken Hill
2/75 S 10 * Occupational Safety in Mines, Southern Queensland
3/75 S 11 * Australian Black Coal, lllawarra
4/75 * Annual Conference, South Australia
1976 1/76 S13 * Landscaping and Land Use Planning as Related to Mining Operations, Adelaide
2/76 * Design and Construction of Tunnels and Shafts, Melbourne - 2 nd Australian
Tunnelling Conference
3/76 S 14 * Thick Seam Mining by Underground Methods, Central Queensland
4/76 S 15 * Sampling Practices in the Mineral Industries, Melbourne
5/76 * Annual Conference, lllawara
1977 1 /77 S 16 * Apcom ‘77, Brisbane
2/77 S 18 * Coal Borehole Evaluation, Southern Queensland
3/77 S 17 * Underground Operators’ Conference, Broken Hill
4/77 * Annual Conference, Hobart
1978 1 /78 S 19 * Mill Operators’ Conference, North West Queensland
2/78 S20 * Rock Breaking Equipment and Techniques, Melbourne
3/78 * International Resource Management, Canberra
4/78 * Annual Conference, Townsville
1979 1/79 S21 * Utilisation of Steelplant Slags, lllawarra
2/79 S22 * Estimation and Statement of Mineral Reserves, Sydney
3/79 * Annual Conference, Perth
1980 1/80 * Annual Conference, New Zealand
2/80 S23 * Australia/Japan Extractive Metallurgy Symposium, Sydney
3/80 S/24 * Occurrence, Prediction and Control of Outbursts in Coal Mines, Southern
Queensland
4/80 S25 * Management in the Mining Industry, Melbourne

Copies of all books currently in print can be obtained from The Institute office
Tel (03) 9662 3166 or Fax (03) 9662 3662

* = Out of print
The ‘S’ numbers in the third column refer to an older identifying number for Symposia, the numbers preceeding the ‘S’ number signify the new
publication ordering number.
1981 I/81 S26 * International Blast Furnace Hearth and Raceway Symposium, Newcastle
2/81 * Fourth Australian Tunnelling Conference
3/81 S27 * Ignitions, Explosions and Fires, Wollongong
4/81 * Annual Conference, Sydney
5/81 S28 * Strip Mining 45 Metres and Beyond, Central Queensland
1982 1/82 S29 * Off Highway Truck Haulage Conference, Newman
2/82 S30 * Mill Operators’ Conference, North West Queensland
3/82 S31 * Underground Operators’ Conference, West Coast Tasmania
4/82 * Annual Conference, Melbourne
5/82 S32 * Carbon-ln-Pulp Technology for the Extraction of Gold, Perth and Kalgoorlie,
(Reprinted 1988)
6/82 S33 * Seam Gas Drainage with Particular Reference to the Working Seam, Wollongong
1983 1/83 S34 * Computers in Mining, Southern Queensland
2/83 * Annual Conference, Broken Hill
3/83 S35 * Project Development Symposium, Sydney
4/83 S37 * Ventilation of Coal Mines, Wollongong
5/83 S40 * Principles of Mineral Flotation (The Wark Symposium), Adelaide
1984 1/84 S36 * Metallurgy Symposium, Melbourne
2/84 S38 * Coal and Mineral Sizing, Wollongong
3/84 * Annual Conference, Darwin
4/84 S39 Gold Mining, Metallurgy and Geology, Kalgoorlie
1985 1/85 S41 * Smelting and Refining Operators’ Symposium, North Queensland
2/85 S42 * Underground Operators’ Conference, Kalgoorlie
3/85 * Annual Conference, Brisbane
4/85 S43 * Scientific and Technological Developments in Extractive Metallurgy (G K
Williams Memorial Volume), Melbourne
1986 1/86 * l3th Congress The Council of Mining and Metallurgical Institutions, Singapore, 6
Volumes
2/86 S44 * Selective, Open Pit Gold Mining Seminar, Perth
3/86 S45 * Ground Movement and Control Related to Coal Mining, Wollongong
4/86 S46 * Australia: A World Source of Illmenite, Rutile, Monazite and Zircon Conference,
Perth
5/86 S47 * Second Project Development Symposium, Sydney
6/86 S48 * Large Open Pit Mining Conference, Newman
7/86 S49 * Education and Research for the Mineral Industry for the Future, Melbourne
8/86 * The AuslMM 10 Year lndex
1987 1/87 * Vl Australian Tunnelling Conference: Bore or Blast, Melbourne
2/87 S50 * Risk and Survival Seminar, Canberra
3/87 * Annual Conference, Newcastle: Coal Power ‘87
4/87 S51 * Research and Development in Extractive Metallurgy, Adelaide
5/87 * Leslie Bradford Golden Jubilee Oration
6/87 S52 * Mining and Environment: A Professional Approach, Brisbane
7/87 * Pacrim ‘87, Gold Coast, Queensland
8/87 S53 * Dense Medium Operators’ Conference, Brisbane

Copies of all books currently in print can be obtained from The Institute office
Tel (03) 9662 3166 or Fax (03) 9662 3662

* = Out of print
The ‘S’ numbers in the third column refer to an older identifying number for Symposia, the numbers preceeding the ‘S’ number signify the new
publication ordering number.
 9/87 S54 * Equipment in the Minerals Industry: Exploration Mining and Processing
Conference, Kalgoorlie
10/87 S55 * Resources and Reserves, Sydney
11/87 * South Australia’s Mining Heritage
1988 1/88 S56 * 21st Century Higher Production Coal Mining Systems Symposium, Wollongong
2/88 * The Second International Conference on Prospecting in Arid Terrain, Perth
3/88 S57 * Third Mill Operators’ Conference, Cobar
4/88 S58 * Underground Operators’ Conference, Mount Isa
5/88 * Fourth International Mine Ventilation Congress, Brisbane (Proceedings and
Addendum volume)
6/88 * Annual Conference, Sydney: Minerals and Exploration at the Crossroads: The
International Outreach
7/88 S59 * Second AuslMM Mineral Heritage Seminar, Sydney
8/88 S60 * Economics and Practice of Heap Leaching in Gold Mining Workshop, Cairns
9/88 * Third International Mine Water Congress, Melbourne
10/88 S61 * Explosives in Mining Workshop, Melbourne
1989 1/89 * Mineralogy and Petrology, Sydney, February
2/89 * Second Large Open Pit Mining Conference, Latrobe Valley Vic
3/89 * NQ Gold ‘89 Conference, Townsville Qld
4/89 * Annual Conference, Perth-Kalgoorlie: Education, Training and Professional
Development; Industrial Minerals; Project Development/Processing
5/89 * Mineral Fuel Alternatives and the Greenhouse Effect, July 1989
6/89 Non-ferrous Smelting Symposium: 100 Years of Smelting and Refining
Operations in Port Pirie, SA September 1989
7/89 * Dewatering Technology and Practice Conference, Brisbane October 1989
8/89 MINVAL ‘89, Mining and Petroleum Valuation 1989, Sydney September 1989
1990 1/90 * Ore Reserve Estimates - The Impact on Miners and Financiers, Melbourne, March
1990
2/90 Annual Conference, The Mineral Industry in New Zealand, Rotorua New
Zealand, March 1990
3/90 Pacific Rim Congress, Gold Coast Qld, May 1990
4/90 * Mining Industry Capital and Operating Cost Estimation Conference, Sydney,
June 1990
5/90 * Third International Symposium on Rock Fragmentation by Blasting, Brisbane,
August 1990
6/90 * Sir Edgeworth David Memorial Oration, May 1990
7/90 * Mine Geologists’ Conference, Mount Isa, October 1990
1991 1/91 * Fourth Mill Operators’ Conference, Burnie Tas, March 1991
2/91 * World Gold ‘91, Cairns Qld, April 1991
3/91 Mining Industry Optimisation Conference, Sydney, June l991
4/91 * PNG Geology, Exploration and Mining Conference, Rabaul, June 1991
5/91 Qld Coal Symposium, Brisbane, August 1991
6/91 * Reliability Production and Control in Coal Mines, Wollongong, September 1991
7/91 * Fifth AuslMM Extractive Metallurgy Conference, Perth, October 1991

Copies of all books currently in print can be obtained from The Institute office
Tel (03) 9662 3166 or Fax (03) 9662 3662

* = Out of print
The ‘S’ numbers in the third column refer to an older identifying number for Symposia, the numbers preceeding the ‘S’ number signify the new
publication ordering number.
1992 1/92 * Enviromine Australia, Sydney NSW, March 1992
2/92 The AuslMM Annual Conference, ‘The State-of-the-Art - A Product of 100 Years
of Learning’, Broken Hill NSW, May 1992
3/92 ‘Energy, Economics and Environment’ Gippsland Basin Symposium, Melbourne,
June 1992
4/92 Arnold Black Mineral Heritage Oration
5/92 The Man from ASARCO: a life and times of Julius Kruttschnitt
6/92 5th Underground Operators’ Conference, Ballarat, July 1992
7/92 * 11th International Conference on Ground Control in Mining, Wollongong, July
1992
8/92 * Third Large Open Pit Mining Conference, Mackay, August 1992
9/92 * Extractive Metallurgy of Gold and Base Metals Conference, Kalgoorlie, October
1992
10/92 * Sampling Practices in the Minerals Industry, Mount Isa, November 1992
11/92 * Rehabilitate Victoria, Latrobe Valley, November 1992
1993 1/93 Mining People - A Century
2/93 The AuslMM Centenary Conference, Adelaide, March 1993
3/93 XVIII International Mineral Processing Congress, Sydney, May 1993
4/93 * Narrow Vein Mining Seminar, Bendigo, June 1993
5/93 * International Mining Geology Conference, Kalgoorlie, July 1993
6/93 Vlll Australian Tunnelling Conference, Sydney, August 1993
7/93 * World Zinc ‘93 - International Symposium, Hobart, October 1993
1994 1/94 1994 AuslMM Student Conference, Brisbane, April 1994
2/94 * PNG Geology, Exploration and Mining Conference, Lae, PNG, June 1994
3/94 No Two The Same by Bert Mason
4/94 Sixth Extractive Metallurgy Conference, Brisbane July 1994
5/94 * 1994 AuslMM Annual Conference, Darwin, August 1994
6/94 * 4th Large Open Pit Mining Conference, Perth, September 1994
7/94 * Recent Trends in Heap Leaching, Bendigo, September 1994
8/94 * Maintenance in the Mining and Metallurgical Industries,Wollongong, October
1994
9/94 * Fifth Mill Operators’ Conference, Roxby Downs, October 1994
10/94 Mineral Valuation Methodologies 1994, Sydney, October 1994
11/94 Victorian Mining Week Conference, Melbourne, October 1994
1995 1/95 1995 AuslMM Annual Conference, Newcastle, March 1995
2/95 Sir Maurice Mawby Memorial Oration
3/95 World’s Best Practice in Mining and Processing Conference, Sydney, May 1995
4/95 APCOM XXV 1995 Conference, Brisbane, July 1995
5/95 Mineral Valuation Methodologies 1994, Sydney, October 1994 (revised)
6/95 * EXPLO 95 Conference, Brisbane, September 1995
7/95 Underground Operators’ Conference, Kalgoorlie, November 1995
8/95 * Young Professionals’ Conference, Mt Isa, October 1995
9/95 * PACRIM ‘95 Congress, Auckland, New Zealand, November 1995
10/95 * Ethics, Liability and the Technical Expert, Sydney, December 1995

Copies of all books currently in print can be obtained from The Institute office
Tel (03) 9662 3166 or Fax (03) 9662 3662

* = Out of print
The ‘S’ numbers in the third column refer to an older identifying number for Symposia, the numbers preceeding the ‘S’ number signify the new
publication ordering number.
1996 1/96 * 1996 AusIMM Annual Conference, Perth, March 1996
1a/96 * 1996 AusIMM Annual Conference Supplementary Volume, Perth, March 1996
2/96 * Ethics, Liability and the Technical Expert, Sydney, March 1996
3/96 * Entrepreneurs and Partners, Sydney, July 1996
4/96 * Contract Operators’ Conference, Kalgoorlie, October 1996
5/96 Asia/Pacific Mining Communications Summit, Singapore, November 1996 -
Withdrawn
6/96 * Nickel ‘96, Kalgoorlie, November 1996
1997 1/97 1997 AusIMM Annual Conference, Ballarat, March 1997
2/97 * World Gold ‘97 Conference, Singapore, September 1997
3/97 Sixth Mill Operators’ Conference, Madang, PNG, October 1997
4/97 Gem 97, Madang, PNG, October 1997
5/97 * Contract Operators’ Conference, Brisbane, Qld, October 1997
6/97 Third International Mining Geology Conference, Launceston, Tas, November
1997
7/97 Mindev 97 - The International Conference on Mine Project Development,
Sydney, November 1997
8/97 1997 AusIMM Travelling Technology Forum, Singleton, NSW, March 1997
1998 1/98 * MINEFILL ‘98 - The Sixth International Symposium on Mining with Backfill,
Brisbane, Qld, April 1998
2/98 * AusIMM’98 - The Mining Cycle, Mount Isa, Qld, April 1998
3/98 * Seventh Underground Operators’ Conference, Townsville, Qld, June/July 1998
4/98 Mine to Mill Conference, Brisbane, Qld, October 1998
5/98 Third Regional APCOM - Computer Applications in the Minerals Industries
International Symposium, Kalgoorlie, WA, December 1998
6/98 COAL98 - First Australasian Coal Operators’ Conference, Wollongong, NSW,
February 1998
1999 1/99 10th Australian Tunnelling Conference, Melbourne, Vic, March 1999
1a/99 10th Australian Tunnelling Conference Keynote Addresses and Asia–Pacific
Forum, Melbourne, Vic, March 1999
2/99 Students and Young Professionals Conference, Perth, WA, July 1999 (N/A)
3/99 ICARISM ’99 Conference, Perth, WA, September 1999
4/99 PACRIM ’99 Congress, Bali, October 1999
5/99 EXPLO ’99 Conference, Kalgoorlie, WA, November 1999
2000 1/2000 Southern Africa - Australia Mineral Sector Synergies Symposium, Canberra,
ACT, March 2000
2/2000 After 2000 - The Future of Mining, Sydney, NSW, April 2000
3/2000 4th International Mining Geology Conference, Coolum, Qld, May 2000
4/2000 Young Leaders 2000, Sydney, NSW, July 2000
5/2000 MINPREX 2000, Melbourne, Vic, September 2000
6/2000 Seventh Mill Operators’ Conference, Kalgoorlie, WA, October 2000
7/2000 MassMin 2000, Brisbane, Qld, October - November 2000
2001 1/2001 Strategic Mine Planning 2001, Perth, WA, March 2001
2/2001 AusIMM Youth Congress 2001, Brisbane, Qld, May 2001
3/2001 International Heavy Minerals Conference, Fremantle, WA, June 2001
4/2001 EXPLO 2001, Hunter Valley, NSW, October 2001

Copies of all books currently in print can be obtained from The Institute office
Tel (03) 9662 3166 or Fax (03) 9662 3662

* = Out of print
The ‘S’ numbers in the third column refer to an older identifying number for Symposia, the numbers preceeding the ‘S’ number signify the new
publication ordering number.
 5/2001 Mineral Asset Valuation Issues 2001, Sydney, NSW, October 2001
6/2001 Coal2001, Coal Mining Operators’ Geotechnology Colloquium, Wollongong,
NSW, February 2001
2002 1/2002 Young Leaders 2002, Kalgoorlie, WA, March 2002
2/2002 Metallurgical Plant Design and Operating Strategies, Sydney, NSW, April 2002
3/2002 CMMI Congress, Cairns, Qld, May 2002
4/2002 Green Processing Conference, Cairns, Qld, May 2002
5/2002 Eighth Underground Operators’ Conference, Townsville, Qld, July 2002
6/2002 AusIMM 2002 Conference, 150 Years of Mining, Auckland, New Zealand,
September 2002
7/2002 Iron Ore Conference, Perth, WA, September 2002
8/2002 Value Tracking Symposium, Brisbane, Qld, October 2002
9/2002 Coal2002, Third Australasian Coal Operators’ Conference, Wollongong NSW,
February 2002
2003 1/2003 Twelfth International Symposium on Mine Planning and Equipment Selection,
Kalgoorlie, WA, April 2003
2/2003 Young Leaders 2003, Brisbane, Qld, April 2003
3/2003 Sixth International Conference Acid Rock Drainage, Cairns, Qld, July 2003
4/2003 Eighth Mill Operators’ Conference, Townsville, Qld, July 2003
5/2003 Mining Risk Management Conference, Sydney, NSW, September 2003
6/2003 Water in Mining 2003, Brisbane, Qld, October 2003
7/2003 Fifth Large Open Pit Conference, Kalgoorlie, WA, November 2003
8/2003 Fifth International Mining Geology Conference, Bendigo, Vic, November 2003
9/2003 Coal2003, Fourth Australasian Coal Operators’ Conference, Wollongong NSW,
February 2003
2004 1/2004 New Leaders 2004, Ballarat, Vic, April 2004
2/2004 Second International Green Processing Conference, Fremantle, WA, May 2004
3/2004 EXPLO 2004, Perth, WA, July 2004
4/2004 Metallurgical Plant Design and Operating Strategies, Perth, WA, September 2004
5/2004 PACRIM 2004 Congress, Adelaide, SA, September 2004
6/2004 Bac-Min Conference, Bendigo, Vic, November 2004
7/2004 Orebody Modelling and Strategic Mine Planning, Perth, WA, November 2004
8/2004 Coal2004, Fifth Australasian Coal Operators’ Conference, Wollongong, NSW,
February 2004
2005 1/2005 Ninth Underground Operators’ Conference, Perth, WA, March 2005
2/2005 Coal2005, Sixth Australasian Coal Operators’ Conference, Brisbane, Qld, April
2005
3/2005 New Leaders 2005, Brisbane, Qld, April 2005
4/2005 Second World Conference on Sampling and Blending, Sunshine Coast, Qld, May
2005
5/2005 Centenary of Flotation Symposium, Brisbane, Qld, June 2005
6/2005 Eighth International Mine Ventilation Congress, Brisbane, Qld, July 2005
7/2005 Hoist and Haul Conference, Perth, WA, September 2005
8/2005 Iron Ore Conference, Fremantle, WA, September 2005
9/2005 First Extractive Metallurgy Operators’ Conference, Brisbane, Qld, November 2005

Copies of all books currently in print can be obtained from The Institute office
Tel (03) 9662 3166 or Fax (03) 9662 3662

* = Out of print
The ‘S’ numbers in the third column refer to an older identifying number for Symposia, the numbers preceeding the ‘S’ number signify the new
publication ordering number.
2006 1/2006 Disposal of Mining Waste Symposium, Sydney, NSW, March 2006
2/2006 New Leaders 2006, Kalgoorlie, WA, April 2006
3/2006 Third International Green Processing Conference, Newcastle, NSW, June 2006
4/2006 Coal2006, Seventh Underground Coal Operators’ Conference, Wollongong,
NSW, July 2006
5/2006 Australia’s Uranium, Adelaide, SA, July 2006
6/2006 Sixth International Mining Geology Conference, Darwin, NT, August 2006
7/2006 AusIMM New Zealand Branch 2006 Conference, Waihi, New Zealand, August
2006
8/2006 Metallurgical Plant Design and Operating Strategies, Perth, WA, September 2006
9/2006 International Mine Management Conference, Melbourne, Vic, October 2006
10/2006 Water in Mining 2006, Brisbane, Qld, November 2006
2007 1/2007 Ninth Mill Operators’ Conference, Fremantle, WA, March 2007
2/2007 New Leaders 2007, Brisbane, Qld, May 2007
3/2007 Australia’s Uranium, Darwin, NT, May 2007
4/2007 Project Evaluation Conference, Melbourne, Vic, June 2007
5/2007 AusIMM New Zealand Branch 2007 Conference, Christchurch, New Zealand,
August 2007
6/2007 Iron Ore Conference, Perth, WA, August 2007
7/2007 EXPLO 2007, Wollongong, NSW, September 2007
8/2007 6th Large Open Pit Mining Conference, Perth, WA, September 2007
9/2007 World Gold 2007, Cairns, Qld, October 2007
10/2007 Uranium Reporting Workshop, Adelaide, SA, November 2007
2008 1/2008 Coal2008, Eighth Underground Coal Operators’ Conference, Wollongong, NSW,
February 2008
2/2008 Tenth Underground Operators’ Conference, Launceston, Tas, April 2008
3/2008 13th Australian Tunnelling Conference, Melbourne, Vic, May 2008
4/2008 Sampling 2008 Conference, Perth, WA, May 2008
5/2008 AusIMM International Uranium Conference, Adelaide, SA, June 2008
6/2008 The AusIMM New Leaders’ Conference 2008, Wollongong, NSW, July 2008
7/2008 MetPlant 2008, Perth, WA, August 2008
8/2008 Ninth International Congress for Applied Mineralogy (ICAM) 2008, Brisbane,
Qld, September 2008
9/2008 Narrow Vein Mining Conference 2008, Ballarat, Vic, October 2008
10/2008 First International Future Mining Conference and Exhibition 2008, Sydney,
NSW, November 2008
11/2008 PACRIM Congress 2008, Gold Coast, Qld, November 2008
2009 1/2009 Coal2009, Ninth Underground Coal Operators’ Conference, Wollongong, NSW,
February 2009
2/2009 Orebody Modelling and Strategic Mine Planning, Perth, WA, March 2009
3/2009 Project Evaluation 2009 Conference, Melbourne, Vic, April 2009
4/2009 The AusIMM New Leaders’ Conference 2009, Brisbane, Qld, April 2009
5/2009 AusIMM International Uranium Conference, Darwin, NT, June 2009
6/2009 Sustainable Development Indicators in the Minerals Industry (SDIMI) 2009,
Gold Coast, Qld, July 2009

Copies of all books currently in print can be obtained from The Institute office
Tel (03) 9662 3166 or Fax (03) 9662 3662

* = Out of print
The ‘S’ numbers in the third column refer to an older identifying number for Symposia, the numbers preceeding the ‘S’ number signify the new
publication ordering number.
 7/2009 Iron Ore 2009 Conference, Perth, WA, July 2009
8/2009 Seventh International Mining Geology Conference 2009, Perth, WA, August
2009
9/2009 AusIMM New Zealand Branch Conference 2009, Queenstown, New Zealand,
August 2009
10/2009 Water in Mining 2009, Perth, WA, September 2009
11/2009 Tenth Mill Operators’ Conference, Adelaide, SA, October 2009

Copies of all books currently in print can be obtained from The Institute office
Tel (03) 9662 3166 or Fax (03) 9662 3662

* = Out of print
The ‘S’ numbers in the third column refer to an older identifying number for Symposia, the numbers preceeding the ‘S’ number signify the new
publication ordering number.
 SPECTRUM SERIES
1. Making the Mount Isa Mine, 1923 - 1933 Don Berkman 1996
2. History of Drilling Graham McGoggan 1996
3. The Cobar Mineral Field – A 1996 Perspective Warren Cook 1996
Andrew Ford
Julian McDermott
Peter Standish
Craig Stegman and
Therese Stegman
4. Towards 2000 – Resource to Reserve Inputs Seminar, Melbourne, Vic 1997

5. Towards 2000 – National Conference on Ironmaking Resources and Reserves 1997

Estimation, Perth, WA
6. Towards 2000 – The Resource Database Towards 2000, Wollongong, NSW 1997
7. Towards 2000 – Ore Reserves and Finance, Sydney, NSW 1998
8. Towards 2000 – Assessment of Reserves in Low Rank Coals, Morwell, Vic 1997
9. Towards 2000 – Ore Reserve Reconciliation Workshop, Darwin, NT 1997
10. Towards 2000 – Gold and Nickel Ore Reserve Estimation Practice Seminar 1998
11. Towards 2000 – Resource/Reserves Estimation Practice in the Central West 1999
New South Wales Mining Industry, Cobar, NSW
12. Field Guide for Geoscientists and Technicians
1st Edition 2004
2nd Edition 2007
13. The Extractive Metallurgy of Zinc Roderick J Sinclair 2005
14. Orebody Modelling and Strategic Mine Planning – Uncertainty and Risk Roussos Dimitrakopoulos
Management Models
1st Edition 2005
2nd Edition 2007
15. The Extractive Metallurgy of Lead Roderick J Sinclair 2009
16. Flotation Plant Optimisation – A Metallurgical Guide to Identifying and Ed: Christopher J Greet 2010
Solving Problems in Flotation Plants