Ninth Mill Operators’

Proceedings

Conference 2007

19 - 21 March 2007
Fremantle, Western Australia

The Australasian Institute of Mining and Metallurgy

Publication Series No 1/2007

Published by

THE AUSTRALASIAN INSTITUTE OF MINING AND METALLURGY

Level 3, 15 - 31 Pelham Street, Carlton Victoria 3053 Australia
 © The Australasian Institute of Mining and Metallurgy 2007

All papers published in this volume were refereed prior to publication.

The Institute is not responsible as a body for the facts and opinions advanced
in any of its publications.

ISBN 978 1 920806 61 3

Desktop published by:

Kristy Pocock and Jennifer Hall for
The Australasian Institute of Mining and Metallurgy

Compiled on CD ROM by:

Visual Image Processing
PO Box 3180
Doncaster East VIC 3109
 Contents
Keynote Addresses

Total System Design for Comminution R A Bearman 3

Some Observations Regarding SAG Milling G Lane 9

Flotation

Differential Flotation Commissioning at Telfer M A Benson, C R Headley, 17

S O Hille and N P Jacob

A Model Study of Copper Rougher Recovery Optimisation S R Grano, T Akroyd and 25

at PT Freeport Indonesia M A Mular

An Integrated Expert and Vision System at BHP Billiton D La Rosa, A Cantarutti, 39

Cannington G J Clarke, M Wortley,
C Bouche and O Guyot

Optimising Flotation Innovations and the Role of Plant Test Work P Bourke and S Grönstrand 49

The Application of Ausmelt’s AM28 Alkyl Hydroxamate Flotation T Hughes, K Lee, G Sheldon, 51
Reagent to Fox Resources’ West Whundo Copper Ore at Radio J Bygrave and L Mann
Hill, Western Australia

Operating Variables Affecting the Bubble Size in Forced-Air J E Nesset, J A Finch and 55
Mechanical Flotation Machines C O Gomez

Validation of the AMIRA P9 Flotation Model Using the Floatability R G Coleman, J-P Franzidis 67
Characterisation Test Rig (FCTR) and E V Manlapig

A Correlation Between VisiofrothTM Measurements and the K Runge, J McMaster, 79

Performance of a Flotation Cell M Wortley, D La Rosa and
O Guyot

Improved Flotation Process Control at Newcrest’s Telfer D Baas, S Hille and 87

Operation J Karageorgos

Optimisation of Flotation Circuit Configuration at PT Newmont X Zheng, K McCaffery and 95

Nusa Tenggara R P Halomoan
 Grinding/Mill Liners
Efficiency of the Detritor Mills at Zinifex Century Mine M Gao, L Reemeyer, 105
D P Obeng and R Holmes

Bolting Technology — Understanding and Utilising the Inherent B Docherty 111

Mechanical Properties of Bolts/Nuts

Advanced Mill Relining — A Key Operational Control (Not Simply J Russell 113
a ‘Must Do’ Maintenance Function)

The Effect of Aspect Ratio on the Grinding Efficiency of Open S Morrell 121
and Closed Circuit AG/SAG Mills

Liner Optimisation to Improve Availability of the Ridgeway M Weidenbach and P Griffin 125
SAG Mill

Process Design of a Regrind Facility at the Leinster Nickel D R Seaman, K E Barns, 131
Operations to Improve Concentrator Recovery P J Sharman, S A Thomas
and K J Stokes

Advances in the MillMapper Technology for Grinding Mill Liner J Franke and D Lichti 139
Thickness Gauging

Practical Experience in the Design and Operation of D Royston 147

Semi-Autogenous Grinding (SAG) Mill Liners

Plant Operations
Metallurgical Accounting in the Northparkes Concentrator — W M Jansen, R Morrison 157
A Case Study and R Dunn

Managing Mineral Processing Plants in Australia — Tips and R S Francis and G D Cooper 169
Pitfalls

Process Design Aspects at Bendigo Mining’s Gold Operations E J McLean, T E Hayward, 175
D H English and S J Smith

Beaconsfield Gold Mine — Ironing Out the Bugs R Holder 185

A Review of Processing at the Cannington Ag/Pb/Zn Mine R Alford and G Clarke 195

Design of a Large-Scale Concentrator for Treatment of a Copper J Glatthaar, G Lane, M Phillips 201
Skarn Orebody and T Hayward

Plant Expansions and Process Development at LionOre’s Lake I Mitchell, D Zoetbrood, 211
Johnston Operations P Henderson, S O’Neill
and A Adem
 Process Improvements

Maintaining the Benefit — How to Ensure Mine to Mill Continues A Dance, W Valery, 215
to Work for You A Jankovic, D La Rosa and
S Esen

Improving Fine Copper and Gold Flotation Recovery — A Plant T Rivett, G Wood and 223
Evaluation B Lumsden

Reducing WAD Cyanide at the Discharge Spigot J A Rumball, S D Munro 229

and M L Habner

Evaluating an Alternative to the Gravity Recoverable Gold G K N Subasinghe 235

(GRG) Test

The Importance of Geometallurgical Analysis in Plant Study, D David 241

Design and Operational Phases

Cyanide Measurement and Control for Complex Ores and P L Breuer and J A Rumball 249
Concentrates

New Technology and Approaches in Reagent Development and D R Nagaraj 255

Applications in the Processing of Base and Precious Metals and
Industrial Minerals
Total System Design for Comminution
R A Bearman1
ABSTRACT
Traditionally, the term ‘total system’ design applied to comminution
conjures images of a design that addresses the usual business and
operational drivers, including health, safety and environment (HSE),
capital cost, operating cost, throughput and product size. In terms of the
physical process, the mine to customer type studies also exemplifies an
approach that satisfies a normal ‘total systems’ approach.
As an industry there are drivers that are forcing a re-consideration of
what we regard as our total system. This modification of drivers will
require changes in approach and implementation to provide the best
solutions. Some of the major changes that impact the drivers are shortage
of skilled personnel, changing social attitudes to remote mine site work,
greenhouse gas emissions (GHGE), quality of available orebodies and
geographical location of new mine sites. One approach to the challenges
presented by the changing business environment is to expand the system
envelope around comminution to ensure that comminution, in its widest
sense, helps to meet the new realities.
Background to the new drivers and constraints is provided and some of
the appropriate responses are examined in relation to the impact on the
total system.
INTRODUCTION
Changing business drivers
Over time business drivers do change. At the birth of mining
activities such as flint mining, Neolithic man was not particularly
interested in capital cost, operating cost or health, safety and
environment (HSE). The only driver was survival. Flint was a
necessity to provide fire and arrow heads, so flint as a
commodity gave an improved chance of a survival for the group.
Jumping forward to early metal extraction, again this was done
because of a specific need. When demand started to outstrip
supply, commercial drivers began to come into play, including
the need to compete. The last great change to business drivers
during the 20th century was HSE. With this revolution the ability
to operate with disregard for some of what we now regard as
basic human and environmental principles, disappeared.
As society has evolved, so have the drivers and constraints that
control the way people and business act. It should not therefore
be any surprise that industry is now facing a new phase in the
definition of the business environment.
New business drivers in the early 21st century
There has been much debate in the industry about recent changes
to business drivers and constraints and whether they are real or
imaginary. The main issues identified are:
• generational change in attitude towards mining,
• scale and type of investment in mining R&D, and
• greenhouse gas emissions (GHGE) and energy.
There seems little doubt that generations entering the
workforce now, and further into the future, have a radically
different attitude to work and mining, compared to preceding
generations. There is an increasing resistance to the idea of
fly-in, fly-out (FIFO) and employment at remote sites. In
addition there is also a move away from old World industries, as
1. Principal Adviser – Innovation, Rio Tinto OTX – Innovation CoE,
152 - 158 St Georges Terrace, Perth WA 6000.
Email: ted.bearman@riotinto.com
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
exemplified by mining. This reluctance to enter the mining
industry has been compounded by the inherent boom-bust nature
of mining. Universities and centres of education for the mining
industry have responded to fluctuations in industry demand by
tailoring effort, both in terms of undergraduate study and
post-graduate research. In terms of undergraduate education, the
boom-bust nature of the industry is almost impossible to
accommodate and coupled to the overall drop in interest in
mining, this has led to the closure of many significant
mining/processing departments, particularly in the UK, USA and
Canada. Figure 1 shows the decline in the number of university
programs since 1985.
2
USA
USA UK
UK Canada
Canada
0
Loss / Gain of Course
Australia
Australia South Africa
South Africa Chile
Chile
-2
-4
-6
-8
-10
FIG 1 - Change in number of university mining programs between
1985 and 2003 (Hebblewhite, 2005).
To arrest this decline mining companies are making major
efforts to promote and support establishments, but there are
underlying trends which may make this goal difficult to achieve.
One such trend, clearly seen in Australia, is the drop in school
students choosing to specialise in the traditional sciences. Figure 2
shows the decline in the number of secondary students choosing
to study science subjects since the mid 1970s.
In terms of university-based R&D the picture has also
changed towards a more developmental focus. From an industry
perspective the ugly side of this change is that it was driven by
the industry itself. Between the mid-1990s and the early part of
this decade, the mining industry, in a broad sense, drove an
agenda that valued short-term ‘applied research’ over innovation.
The impact of this has been to drive universities to concentrate
on developmental issues. The reversal of this trend to re-focus on
innovative developments is not a trivial task. Many industry
experts would even argue that by the time universities are
persuaded to lift focus to innovation, the resource cycle will be in
decline and therefore the industry can never catch up.
The counter argument is that it is a matter of focused effort.
Yes, the industry has dropped the ball and to pull all R&D
providers back to an innovation position is unlikely, but by
identifying establishments with innovation potential and funding
these, then there is hope. In conjunction with this re-focus is the
requirement to increase funding of R&D. The latest figures on
the scale of R&D funding from mining companies is not
encouraging, as can be seen in Table 1.
In terms of the type of R&D, the mining and processing
industry has never lacked improvement initiatives, whether this
be in terms of continuous technical improvement, or simply
methods for improving the overall process effectiveness. For
total systems design the total innovation process should simply
3
R A BEARMAN

FIG 2 - Change in final year high school specialisation (Ainley and Underwood, 2003).

4 35
TABLE 1
Total Emissions (t CO2-e / year)
Total Emissions
Comparative R&D funding for industry sectors (DTI, 2006). Total Energy 30

Total Energy (PJ/year)

3
25
Industry R&D R&D
(% profit) (% sales) 20
Aerospace and defence 58.7 4.6 2
15
Automobiles and parts 99.3 4.2
Chemicals 34.4 3.2 10
1
Construction and building materials 18.8 0.9 5
Electricity 8.0 1.0
0 0
Electronics and electrical equipment 73.8 4.8

s
n
n

oc r
g
g /

Pr he
ng
lin ng

tio
io
in

ut
in

in t
ili
au i

ia

M O
H oad

Gas, water and utilities 2.5 0.3

in
M

Ta
fic
m
er

ne
L

m
th

General industrials 16.2 2.4

Be
Co
O

Health care and services 41.1 6.4

FIG 3 - GHGE for a typical open pit mining and processing
Oil equipment and services 11.7 1.8 operation (Batterham, 2004).
Industrial engineering 33.0 2.7
Pharmaceuticals 52.0 14.1 Another important element in the discussion around energy is
Software and computing services 53.3 10.4 the area of embodied energy. Embodied energy refers to the
quantity of energy required to manufacture, and supply to the
Mining 2.9 0.8 point of use, a product, material or service.
Industrial metals 6.6 0.5 Embodied energy can be considered as an accounting
methodology which aims to find the total energy necessary to
deliver a material for use with the calculation, covering everything
be regarded as a spectrum with the here-and-now represented by from the raw material extraction, transport, manufacturing,
continuous improvement and the far field as transformational assembly, installation and finally its disassembly or
change. In between these extremes are a range of titles, including deconstruction.
Incremental Improvement, Innovational Improvement and Step
Every mine is a complex combination of many processed
Change. All of these types of improvement have a place in
materials, each of which contributes to the mine’s total embodied
influencing the total picture.
energy. Expansion, operational changes and maintenance also
add to the embodied energy over a mine’s life.
Greenhouse gas emissions The embodied energy content of a mining operation is small
Undoubtedly one of the major challenges is the issue of compared to the energy used in operating the mine over its life,
greenhouse gas emissions (GHGE). Many countries and but research has shown that in specific areas, this is not always
companies have committed to the various agreements and the case. Embodied energy can be a significant consideration,
initiatives to reduce greenhouse gas emissions. Figure 3 shows specifically comminution; therefore, if the total system is to be
the breakdown of CO2 emissions from a typical open pit mining impacted then this needs to be considered.
and processing operation. It is therefore critical that the design of the total mining system
When considering comminution as a part of the picture is appropriate to the task. This can lead to a variety of
portrayed in Figure 3, it must be remembered that it will not be appropriate concepts depending on the deposit and life cycle, ie:
incremental changes in comminution equipment that will • long life, durable and adaptable;
decrease the overall impact. Rather, it is the total design and
integration that will deliver the greatest reductions. A total • short use, flexible and relocatable; and
system view of comminution impacts much more than just the • consideration of the embodied energy component of
direct amount singled out in Figure 3. operations.

4 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 TOTAL SYSTEM DESIGN FOR COMMINUTION

One classic example of embodied energy is the use of steel
grinding media. A typical ball mill installation treating a soft-
medium copper ore has the following features:
• throughput: 350 t/h,
• power consumption: 5.65 kWh/t, and
• media consumption: 125 kg/h.
In the case of steel grinding media the typical embodied
energy is estimated to be 32 MJ/kg (Hammond and Jones, 2006).
These values can be transformed into an equivalent kWh/t value
for the consumption of the media:
• media consumption per tonne of ore = 125/350 = 0.36 kg/t;
• embodied energy per tonne of ore = 0.36 × 32 = 11.52 MJ/t,
or 11 520 kJ/t; and
• embodied energy consumed through media wear =
11 520/3600 = 3.2 kWh/t.
Given that the power consumption for the installation is
5.65 kWh/t, this means that the embodied energy in the media
adds another 56 per cent to the energy input for grinding. This is
a significant addition to the operational energy footprint of a
comminution operation.
RESPONSES TO NEW DRIVERS
Given the new business drivers and constraints and the potential
impact of comminution on the total system, there needs to be
some changes to how comminution is considered.
Comminution in its purest definition is size reduction and
therefore should be regarded as all size reduction processes from
the extraction of the rock from in situ through to the delivery to a
final size for extraction or sale. In this sense the total design of a
comminution system should actively be looking to:
• reduce and optimise energy input through the replacement, or
reduction in, the numbers of stages of comminution and
associated treatment and handling;
• reduce the amount of material processed;
• utilise equipment and processes that are suited to the
application of automation; and
• minimise the size and footprint – physical and environmental.
Energy usage in the total comminution system
The total mining and processing chain uses significant quantities
of energy. Figure 4 gives a breakdown of energy input at various
stages, excluding the impact of embodied energy.
In most instances the first comminution process is the
so-called chemical comminution that uses explosives. Drill and
blast operations are by nature a batch process that reduces
material to a point where it can be handled. In many cases the
size produced from blasting is matched to some of the largest
material handling and transport equipment available, which
logistically has appeal, but in terms of energy this approach is
open to question. The same thinking has led to increasingly large
excavation and transport equipment. Figure 5 shows the trend in
haul truck size over the last 50 years.
As comminution is all about breaking rock then the total
journey from in situ must start to be considered. Integrated
studies tend to focus on the energy balance between explosive
and mechanical comminution. Why should comminution not
consider alternatives to explosives? Mechanical rock cutting or
continuous mining equipment has been widely researched and
pursued over the years, but seldom as part of an integrated
total design for comminution. There is no reason why such
mechanical cutting devices should not be considered part of the
total picture. The main features of mechanical mining equipment
are:
• type of ground engaging tool and cutting mechanism,
• vehicle to carry cutting device and apply force, and
• product discharge mechanism.
As with all comminution the aim is to exploit as much tensile
breakage as possible, thus avoiding wasted energy through
compressive crushing and shear zone development.
The main cutting tools can be categorised as picks and discs,
with picks used for soft rock and discs deployed in harder
material. There are many versions of the tools designed for
specific machine types and application.
Numerous studies recently have examined the use of
mechanical cutting devices to replace drill and blast. In
considering the total system, the replacement of the explosive

Total Operations
G&A
Total Mill
Other Plant
Process Water
Tailings
Other Processing
Grinding
Crushing
Total Mining
Mine Support
Dewatering
Ore Excavation
Transport
Excavation
Blasting
Drilling
Waste Rock Removal
Handling
Transport
Excavation
Blasting
Drilling

0 5000 10 000 15 000 20 000 25 000 30 000 35 000

kWhe/ktonne
FIG 4 - Energy input per production element (Natural Resources Canada, 2005).

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 5

R A BEARMAN
400
350
Payload (metric tonnes)
300
250
200
150
100
50
0
1950 1960 1970 1980 1990
Year
FIG 5 - Increase in haul truck payload since 1950 (development
in truck payload capacity since 1950 (after Krause, 2001 and
Caterpillar, 2005)).
usage should be considered as simply one of the important
factors. In open pit iron ore mining the breakdown of energy
input is approximately:
• drill/blast: 10 - 15 per cent,
• excavation/loading: 10 - 15 per cent,
• haulage: 40 - 55 per cent,
• processing: 10 - 15 per cent, and
• other: 5 - 10 per cent.
The situation highlights that it is drill/blast/load/haul which is
the greatest consumer of energy, with haulage dominating.
Therefore, considering total system design of comminution
systems it would seem sensible to reduce in situ material to such
a size so that alternatives to trucks can be considered. Many
operations already do this by primary and secondary crushing
down to 80 mm and then using overland conveyers. In trials of
surface, mechanical mining there is evidence to support the view
that they can generate -80 mm material directly from in situ rock
through a cutting process. The knock-on effect is dramatic, in
that material can be conveyed and the requirement for drill, blast,
load, truck haulage and even primary and secondary crushing is
removed. Obviously there are issues to be overcome, including
efficient methods for transferring from the surface miner to a
conveyer based haulage system, but this should not be
insurmountable.
Another integrated comminution and transport approach is
represented by the use of mobile or in-pit crushing and
conveying systems. This is not a new solution, but the scale and
implementation of the systems has developed. One of the major
deployments of large scale mobile crushing and conveying is the
overburden system at the BMA Goonyella Riverside operation.
In this application a P&H 4100XPB electric shovel loads blasted
rock straight into a mobile crushing rig that utilises an MMD
1400 mineral sizer. Although the size reduction targeted is not
huge (P80 = 350 mm) the rig offers the ability to treat up to
10 000 t/h and reduce the material to a size that can be
transported using standard overland conveyers, and distributed
via spreaders (Anon, 2003). Currently this technology is
restricted to soft-medium ore types, but recent work in the
application of sizers to medium-hard material types could see the
expansion of this approach.
Waste removal
A major aspect in the use of energy in mining and processing is
the extraction and subsequent treatment of waste material. Every
6 Fremantle, WA, 19 - 21 March 2007
tonne of waste carried through the total value chain leads to
larger, higher capital spend plant, simply to handle the quantities
of material. Declining grades across the industry has dramatically
increased the magnitude of waste processing. Copper, as an
example, is illustrated in Figure 6.
16
14
Average Copper Grade (%)
12
10
8
6
2000 4
2
0
1750 1800 1850 1900 1950 2000 2050
Year
FIG 6 - Decline in mined copper grades.
Part of the concept of a total systems approach for
comminution is to select and sort the ore from waste/gangue
and/or below-grade ore to allow only valuable material to be
processed. The aim is to produce improvements in the efficiency
of the process and to reduce the unnecessary processing of waste.
This type of selectivity may potentially offer:
• reductions in operating costs due to the decrease in tonnage
processed;
• recovery of previously uneconomical material from the
existing waste stockpiles, which may lead to an extension to
the life-of-mine;
• reduction in utilities such as water and power supply;
• improved utilisation of ore reserves; and
• upgrade of marginal ore.
In mining process streams, materials can be sorted at bulk flow
or at individual particle level. The bulk sorting is aimed at
identification of material properties ‘en-masse’. As such they are
used to discriminate in mass flow environments such as trucks,
conveyor belts and load-haul-dump (LHD) buckets. Bulk sorting
is predominantly limited to low-value products, whereas
individual particle sorting is traditionally restricted to high-value
commodities such as diamonds.
Ore sorting is dependent on the detection of differences in
fundamental mineral properties. Sufficient difference must exist
in the fundamental properties to achieve an efficient ore sort. The
properties that can be used as a basis for ore sorting include:
• Photometric measurements – optical differentiation of waste
from ore, based on the material being induced/fluoresced at a
particular wavelength interval.
• Thermal properties (conductivity and diffusivity). These
fundamental material properties form the basis for potential
selective heating sorting systems.
• Radiometric properties – based on determination of the
natural emission of radioactivity from an ore.
• Optical properties – optical differentiation (brightness and
colour) of waste from ore can be used as a basis for ore
sorting. Optical sorting could equally be included in
photometric sorting.
• Particle density and porosity (gravity separation).
• Electromagnetic properties – measurements of inductive
conductivity or galvanic resistivity can be used to
discriminate between ore and waste.
Ninth Mill Operators’ Conference

• Other surface properties – including chemical breakdown
properties or surface chemical properties (including friction).
Comminution and automation options
Already comminution equipment uses more advanced
automation and control than other areas of the mining value
chain. This is possible through online measurement, machine
health diagnosis and the use of plant-wide management systems.
To make comminution equipment less personnel-intensive
further developments of sensors and condition monitoring are
required, so that maintenance intervals can be better defined and
therefore more predictable. There also needs to be consideration
of equipment size and maintainability. The move to ever-
increasing throughput has led to the development of enormous
grinding mills which, although presenting maintenance challenges,
has dramatically reduced personnel numbers compared to previous
multi-stage crushing circuits.
Modelling and simulation is not always seen as an enabler to
reduce human intervention in comminution, but the link is very
real. One of the major reasons for intervention is maintenance
and adjustment. Modelling and simulation allows for predictive
capability, thus it is possible to predict changes in equipment
performance that are precursors to major maintenance events.
There are numerous examples of such capabilities being
developed. The various mill monitoring devices being developed,
including acoustic and shell-mounted accelerometers are such
devices. With these linked to models of the internal processes
such as discrete element modelling (DEM), changes in
performance can be linked to the mill fundamentals and therefore
a full picture of the mill condition can be achieved.
Without tools to assist in predicting when maintenance or
adjustment is due, or overdue, the effects can be dramatic.
A concept not just confined to comminution, but with
significant ability to ease reliance on site-based operations
personnel, is the remote operations centre. Many mining
companies have these in various forms, but the recent trend has
been to try and make these even more powerful. The main
enhancements to the traditional approach are that the centres will
be:
• truly remote from mine sites – more attractive to personnel;
• integrated multi-control facilities – process integrated with
mining, maintenance, planning and geological functions;
• information management – access to background
information, maintenance manuals, procedures and historical
databases; and
• interactive – improved voice and multimedia abilities
allowing seamless interchange with site personnel.
Minimise the size and footprint – physical and
environmental
Much of the ability to reduce the physical impact of
comminution plants is tied to the ability to treat reduced
tonnages for the same or better generation of product. The two
main components impacted by gross throughput are the number/
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
TOTAL SYSTEM DESIGN FOR COMMINUTION
size of comminution stages and the subsequent scale of disposal
operations. With reference to the section on waste removal the
best option to reduce plant size is to remove waste as soon as
possible from the processing stream. The other option is to use
higher flow rates per unit of equipment, or use comminution
equipment capable of a wider range of size reduction in a single
unit.
CONCLUSIONS
Comminution has always stood in its own right within
processing. Over the years the unified approach to the mine
through to product chain has meant that comminution has
become a more integrated part of the process, but even this is not
enough.
Comminution plant can dramatically impact the stages up and
downstream and in many instances it is these processes that
relate closely to the modified business drivers now in place. Total
system design is therefore an approach that needs to recognise
the impact of comminution and how it can influence the overall
process. The aim of the total system design being to minimise
impact and implement designs/practices that optimise efficiency,
particularly with regard to the realities of energy and human
resources.
ACKNOWLEDGEMENTS
The author would like to thank David Tuppurainen of Rio Tinto
OTX and Dave Bowman and Jahna Speedy from the Rio Tinto
Iron Ore Technology Division for their assistance.
REFERENCES
Ainley, J and Underwood, C, 2003. Australia’s teachers: Australia’s
future, in Review of Teaching and Teacher Education, Volume 3:
Background Data Analysis, DEST, Canberra.
Anon, 2003. Overburden stripping at Goonyella Riverside, Australian
Journal of Mining, January/February:42.
Batterham, R J, 2004. Comminution challenges for Rio Tinto, in
Proceedings Comminution Workshop 2004, 23 March 2004, pp i-v.
Caterpillar, 2005. Caterpillar achieves mining truck milestone as Decatur,
Illinois, plant readies, Number 5000, Release Number: 121PR04 –
for North American Release [online]. Available from:
<http://www.cat.com/cda/components/fullArticle>.
DTI, 2006. The top 800 UK and 1250 Global companies by R&D
investment – Commentary and analysis, 1:16-124.
Hammond, G and Jones, C, 2006. Inventory of Carbon and Energy –
Version 1.5, pp 4-6 (Department of Chemical Engineering,
University of Bath: Bath).
Hebblewhite, B, 2005. MTEC mining engineering education initiatives in
Australia, Presented to SME2005 Conference, Salt Lake City, Utah,
USA, 28 February 2005.
Krause, G, 2001. Large heavy haulers: Is bigger better? Presented to
Learning Seminar No 6, University of Alberta: Edmonton.
Natural Resources Canada, 2005. Benchmarking the Energy Consumption
of Canadian Open-Pit Mines, p 20. Available from: <http://oee.
nrcan.gc.ca/publications/industrial/mining/open-pit/index.cfm?attr=2
4>.
7
Some Observations Regarding SAG Milling
G Lane1
ABSTRACT
This paper summarises some observations regarding the way SAG and
AG mills work based on the last ten years of pilot plant operation,
grinding circuit design and operational reviews for various projects in
Australia and overseas. The main purpose is to promote discussion on key
factors influencing breakage in a SAG mill.
The focus of the paper is on two questions:
• How do mill lifters and liners really work and have an effect on
grinding?
• How does the mill charge composition impact on grinding
performance?
The focus of the paper is on observations of mill behaviour in both
pilot plant trials and plant operation, particularly on Cadia ore. DEM, by
others, and JKSimMet modelling of SAG mill performance have greatly
assisted in forming the ideas expressed in the paper and both are valuable
tools in developing our understanding of the comminution process.
Put in the ‘extreme’ form the major observations and conclusions
expressed in this paper are that:
• the main purpose of mill lifters is to reduce mill liner wear and not to
promote rock breakage;
• the often promoted theory that balls impacting the ‘toe of the charge’
is a dominant form of breakage is incorrect; and
• fines production in a SAG mill is primarily a function of the rock
load mass, with a secondary dependence on ball load for any given
circuit configuration.
INTRODUCTION
This paper is intended to be, in part, mildly controversial to
promote discussion of how SAG mills work. The content is based
on a combination of long-held views developed from pilot trials
by the author over the past 13 years, contemplation of comments
made, and data presented, at SAG 2006 and discussions held
with mill operations and liner system designers.
Some of the data presented dates from pilot trials conducted in
the early 1990s whilst other observations relate to recent
performance of large SAG mills, particularly Cadia’s 20 MW
SAG mill.
AG milling commenced in Australia with mills like the 32 ft
diameter Savage River mill in 1966. Since that time, the
introduction of balls (semi autogenous milling), initially at low
levels, has increased the throughput of mills fed with primary
crushed ore. By the mid 1990s SAG mill ball loads had increased
to 20 per cent and SAG mill diameters were increasing to 40 ft
(Freeport, Cadia). A consequence of this was that the theory
proposed by liner and mill suppliers that the number of lifters
should be equal to (2 × mill diameter – 2) and that 12 degree
leading edges were appropriate, in practice led to ball on shell
impacts at moderate to high mill speeds. At the same time,
efforts to maximise mill throughput led to operation at low total
loads and further exacerbated the potential for mill liner damage
through ball impact. Trajectory calculation software focused on
getting the steel media to impact on the toe or the change – a
sensible approach to increasing mill liner life. Other outcomes
included reduced inter lifter packing and resulted in slightly
increased mill volume and power draw (Karageorgos, 2001).
Two main aspects of SAG mill performance are discussed in
this paper:
1. General Manager Technical Solutions, Ausenco Limited, 2404
Logan Road, Eight Mile Plains Qld 4113.
Email: greg.lane@ausenco.com
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
• the key breakage mechanism in a SAG mill and the role of
mill lining systems, and
• the roll of rock charge mass in determining throughput and
product size.
WHAT IS THE KEY MECHANISM FOR
BREAKAGE IN A SAG MILL?
Current understanding
Royston (2006) described the current understanding of how SAG
mills operate at SAG2006. The following three bullet points
summarise the relevant aspects of his paper.
• Objective of shell lifter design:
The two main process objectives in SAG mill
shell lifter design are to maximise the impact and
rate of delivery of grinding media at the toe of
the charge where most of the rock-breaking
action is assumed to occur and to avoid liner and
ball damage.
• Relationship between mill speed and lifter wear:
Increasing mill speed over the wear life of shell
lifters can maintain a sweet spot in mill
performance by continuing to direct ball impact at
the toe as both the lifter height falls and face
angle increase with wear … It is worth noting that
highly worn shell lifters can still deliver adequate
(though not optimal) milling performance.
• The action of balls in the mill charge:
Within the mill charge, balls have to transfer
energy into rock breakage via attrition and
abrasion through charge motion in the mill … It
is now a common practice, to improve milling
rate, for SAG mills to be run with low charge
levels at maximum ball charge levels. The
outcome is to increase ball-charge participation
by more turnover of the charge per rotation of
the mill and by increasing the ball-to-rock ratio
… With newer mills with higher load capacity,
high ball charge levels (say up to 18 per cent)
have been used. The objective again is to
increase ‘ballcharge participation’ through
increasing the ball to rock ratio, while drawing
maximum power at the maximum allowable total
charge mass.
Other studies on SAG mill performance have concluded that:
• the majority of the contact, shear, energy consumption and
breakage occurs in the ‘kidney’ of the charge in a SAG mill
(see red area in Figure 1, from Nordell, 2002);
• the importance of high energy ‘impact breakage’, where
large rocks are broken into two or more parts by a single
impact event, varies with ore type and mill operating
condition; and
• the majority of the energy consumed in a SAG mill is used in
producing fine (say <2 mm) particles through abrasion and
low energy ‘impacts’.
9
G LANE

• operating high ball load SAG mills at below about 26 per

cent filling and/or at high speed can increase liner damage
(depending on the style of lifter used);
• optimum filling for AG mills for maximum throughput can
be greater than 30 per cent filling, but dependent on the feed
size distribution; and
• increasing mill speed as lifters wear enables throughput to be
maximised.

HOW DOES ROCK CHARGE MASS INFLUENCE

FIG 1 - DEM output illustration – shear power in a SAG mill
(Nordell, 2002).
Rules of thumb
A few ‘rules of thumb’ have resulted from the current
understanding of AG and SAG milling:
• AG mills and low ball load SAG mills benefit from a coarser
feed to maximise throughput as the coarse feed provides the
source of high energy impact media;
• the throughput of high ball load SAG mills:
• increases with the fines content of the SAG mill feed, ie
throughput is inversely proportional to F80 (MacNevin,
1997);
• increases with increasing ball to rock ratio in the mill;
PERFORMANCE?
In the early 1990s an extensive series of pilot plant trials was
completed for a brownfields expansion. The pilot plant test work
for this expansion was conducted on a feed sampled from ROM
ore over many months. The trials predominantly used a single
stage (closed circuit with cyclones) fully autogenous mill
configuration, with and without pebble crushing. Numerous
surveys were completed and a substantial amount of data resulted.
Extensive analysis and modelling of the data was completed.
Figure 2 illustrates the results of JKSimMet modelling of the
pilot plant data. The size distribution of the mill change was a
function of the feed size distribution and the pebble crusher
performance. The mill filling was approximately 25 per cent of
total mill volume for all tests. Pebble crushing only directly
crushed the >10 mm <75 mm fraction. However, the flow-on
affects on the breakage rates of the finer and coarser fractions
were significant. Essentially:
• crush too much in the pebble crusher and the breakage rate of
the 250 micron particles reduces substantially – the practical
outcome of this in the pilot plant was circulating loads of
2000 per cent around the cyclones; and
• crush too little in the pebble crusher and the circulating load
around the cyclones is low, but so is the overall throughput
(with low volume loads of >76 mm material).

4 6
3.5
Breakage Rate 0.25 mm

5
Breakage Rate 4 mm

3
4
2.5
2 3

1.5
2
1
1
0.5
0 0
0 5 10 15 20 25 30 0 10 20 30 40 50
Rock Load %<38.1>12.7 mm Rock Load %<76.2>24.4 mm

5 4
4.5 3.5
Breakage Rate 16 mm

4
Breakage Rate 44 mm

3
3.5
3 2.5
2.5 2
2 1.5
1.5
1
1
0.5 0.5
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Rock Load %<101.6>76.2 mm Rock Load %<202.2>101.6 mm

FIG 2 - Relationship between AG mill charge size distribution and particle breakage.

10 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 SOME OBSERVATIONS REGARDING SAG MILLING

The affect of not having larger rocks in the feed to an AG mill WHAT IMPACT CAN LIFTER DESIGN HAVE ON
can be inferred by the relationship between the proportion of
large rocks and the breakage rate of the critical size material
MILL PERFORMANCE?
(about 44 mm). A low population of large rocks results in a high
population of ‘critical size’ rock. Pebble crushing can assist in Plant experiences
rectifying this situation but feeding AG mills with feed deficient The dominant reasons for SAG mill lifter design optimisation
in large rocks tends to result in lower than optimum throughput have been to reduce liner damage caused by the impact of large
as observed in treating Ridgeway ore (Clements, 2002).
balls on the mill shell and to reduce inter lifter packing.
Similar relationships were observed in using less detailed Practitioners have also sought to improve the efficiency of
analysis of open circuit pilot plant mills on other ores. For any energy utilisation.
given ore type and configuration of the mill (trommel grate
aperture, pebble crusher setting, feed size distribution, etc), the The changes in lifter/liner system design have achieved the
mass of rock in the mill is directly proportional to the proportion desired outcomes with respect to reduced downtime and
of fines (<150 micron) in the trommel undersize (Lane and inter-lifter packing. Where these changes have been on variable
Dunne, 1998). Data with the ball load varying from zero per cent speed mills, there has been the added advantage of optimising
to 12 per cent is given in Figure 3 for pilot trials on the Cadia mill speed as lifter height wears, usually resulting in increased
monzonite ore. Trommel grate aperture, pebble porting and throughput (Figure 5). The increase in specific energy in Figure 5
crushing and rock load in the mill were the significant factors in is accompanied by an increase in mill speed as the lifters wear.
determining the quantity of -150 micron fines in the trommel Interestingly, there is also a slight increase (five per cent) in the
undersize (primary mill product). The ball load varied from zero specific energy as the lifters wear, which would indicate a finer
per cent (for rock loads >650 kg) up to 12 per cent (for a rock product is produced as the liners wear, the mill speed is increased
load of 340 kg). (and possibly the ball charge drops). The last data point is due
Similar data is presented in Figure 4 for a South African ore changes in the mill load to facilitate lifter change out.
piloted in 1995 using a smaller modified pilot plant AG/SAG mill.
These data include pilot plants run in the South African ROM Examples of recent studies on lifter impacts on
manner with high mill fillings of 40 per cent in autogenous mode.
grinding efficiency
The above data illustrates the impact that the rock load mass
and size distribution has on the way SAG and AG mills work. The following examples illustrate recent experiences in trying to
The following section discusses the impact that liner and lifter assess efficiency improvements as a result of mill liner
design has on breakage. modification.

100
SAG Product % -150 m

80 SAG Milling
AG Milling
60

40

20

0
0 200 400 600 800 1000
Rock Load kg

fine feed, 7 mm grate, with Pebble Crush fine feed, 7 mm grate, no Pebble Crush
coarse feed, 10 mm grate, with Pebble Crush coarse feed, 10 mm grate, no Pebble Crush

FIG 3 - Relationship between rock load mass and SAG mill product size (from Lane and Dunne, 1998).

90
2400 10.00
80 2350
9.50
Specific Energy, kWh/t

70 2300 Throughput (t/h)

% - 150 m in product

Throughput (t/h)

2250 9.00
60 Specific Energy (kWh/t)
Bi modal feed 2200
50 2150 8.50
AG Milling 2100
40 SAG Milling
2050 8.00
30 No Pebble Crush, 10 mm
screen 2000 7.50
20 Pebble crush 10 mm screen 1950
No Pebble Crush, 1 mm 1900 7.00
10
screen
0 50 100 150
0
Days Since Lifter Change
0 200 400 600 800 1000
Rock Load, kg
FIG 5 - Performance of the Cadia SAG mill as lifters wear
FIG 4 - Data from pilot plant trials on a South African ore. (data from Hart, 2006).

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 11

G LANE

Hart, Nordell and Faulkner (2006) indicated that plant hourly

throughput increased by only four per cent, after seven years of
liner and lifter development at Cadia, with the bulk of this
improvement due to the use of curved pulp lifters and improved
pebble flow to the pebble crushers. The Cadia liner/lifter system
development program included extensive DEM by Nordell. The
aim of the Cadia work was increasing mill availability through
maximising liner life and this was achieved through the
development of novel lifter profiles incorporating ‘ski-jump’
secondary lifters.
Rajmani (2005), in a paper on work at the Corez Mine,
reported that:
the design of both shell lifters and grate-pulp lifter
assembly are crucial for optimal performance of
the SAG mills. The design of shell lifters, which
control the charge motion thus the breakage field
… the redesign of shell lifters readily resulted in
230 - 370 kW (ten to 15 per cent) reduction in mill
power draw while maintaining the same
throughput level.
A similar improvement in overall circuit operating work index
was also noted. Whilst this conclusion is not supported by
benchscale test work to indicate that the ore characteristics had
not changed, the 26 ft diameter fixed speed SAG mill at Cortez
was fitted with 17 degree face angle lifters that were changed to
28 degree face angle and an improvement in charge density was
likely in this event. Mill throughput increased without a FIG 6 - Mill fitted with grid liners (from Powell).
significant change in mill load after new lifters had been
installed. system previously installed was grossly inadequate (eg caused a
Rajmani’s work was predominately aimed at reducing liner high degree of ball on liner contact and associated damage) or
damage and downtime caused by ball impacts on the shell with the increases are not able to be easily measured during the
resultant breakage of the liners. The improved operating normal course of plant optimisation.
efficiency was a secondary benefit. As noted above, the absence
of supporting ore characterisation data and the high variability in
ore competency makes the reported improvement in efficiency a Purpose of lifters
dubious claim, particularly as the pebble flow to the pebble An interesting personal observation from completing pilot plant
crusher has reduced, the proportion of total circuit (SAG and ball trials in 1.8 m diameter mills and operating large plant AG and
mill) power drawn by the SAG mill decreased and the overall SAG mills is that the diameter of the mill has very little impact
plant operating work index also decreased, all factors pointing to on the specific energy for a given product size and the product
a less competent and softer ore type. sizes for a given set of operating conditions are not substantially
Hollow and Herbst (2006) reported a possible 4.6 per cent different. However, Morrison et al (2006) reported that DEM and
increase in the Fort Knox mill throughput due to lifter profile pilot plant data were used to determine that multi-impact damage
changes. However, the paper again indicates the issues associated is three times more common than single impact damage and the
with measuring improvements due to changes in mill lifter collision energies in the pilot plant mill do not achieve the
profile due to changes in ore characteristics over the 300 days measured threshold energy for impact damage to accumulate.
required for the plant evaluation. Loveday et al (2006) reported the following regarding rock
Liner systems for SAG mills have typically consisted of a breakage in a 1.8 m diameter mill:
‘plate’ component and a ‘lifter’ component, often combined into
a single lifter/liner piece for large mills. However, ROM primary
• ball addition had no affect on large rock breakage rates, and
mills in South Africa and a number of single stage SAG mills • fines production and ‘rock abrasion’ increases at low (69 per
installed in Australia in the 1980s used ‘grid’ liners that formed a cent of critical) mill speeds.
wear protection surface by including media in grids in the lining If small diameter pilot plant mills can not deliver sufficient
system (Figure 6). There were major differences in the mode of energy in one impact and multi-impacts are required for
operation of the grid liner and more traditional plate and lifter breakage, why do plant SAG mills mirror pilot plant
mills, with the grid liner mills operating with higher mill filling, performance? Perhaps high energy impact on the toe of the
and higher mill speed. Mills with grid liners operated at up to charge is not as important as overall ‘mobilisation’ of the charge.
90 per cent of critical speed and at up to 45 per cent mill filling This conclusion, and the higher capital cost of a high aspect mill
compared to 70 per cent to 75 per cent of critical speed and when compared with a low aspect mill of the same installed
25 per cent to 35 per cent mill filling for the typical pancake mill power, suggests we should use low aspect ratio mills rather than
in Australia. high aspect mills. However, the impact that aspect ratio has on
The lack of lifters in the grid liner mills necessitates the higher the material transfer through the grates and pulp lifters has not
mill speeds to achieve a similar movement of the charge to that been discussed in this paper and can be a key factor in aspect
achieved with slower speeds and large lifters. Extensive studies ratio selection.
of a number of these mills have been reported by Powell (1988, There are critical ‘criteria’ for lifter design, mostly related to
2001). avoiding ball on liner shell collision, but maximising mill charge
The general conclusion from the above is that if increases in motion. After this, selection criteria relate to optimising the
SAG milling energy efficiency (as distinct from throughput) are relationship between liner life (mass of steel) and mill volume
possible through optimisation of lifter design, either the lifter loss.

12 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference


This paper proposes that:
• mill liner/lifter profiles have very little impact on SAG mill
specific energy requirements for a given ore as long as key
criteria are met;2
• under these circumstances, the main purpose of lifters is to
minimise impact and abrasive wear of the liner plates, and as
a consequence;
• the concept of maximising the ‘energy efficiency of
breakage’ in SAG mills by maximising impact on the toe of
the charge is not of significant consequence.
ACKNOWLEDGEMENTS
The author wishes to acknowledge the indirect input through
discussion of a number of people, including Steve Hart when at
Cadia and his colleagues, Paul Griffin and Larry Nordell, In
addition, the participants in over a dozen SAG mill pilot
trials, including various personnel from client organisations,
engineering companies, research organisations and laboratories
over the past 15 years contributed to the learning experience.
REFERENCES
Clements, B, 2002. Personal communication.
Hart, S, Nordell, L and Faulkner, C, 2006. Development of a SAG mill
shell liner design at Cadia using DEM modelling, in Proceedings
SAG2006 Conference, Mining and Mineral Process Engineering
(University of British Columbia: Vancouver).
Hollow, J and Herbst, J, 2006. Attempting to quantify improvements in
SAG liner performance in a constantly changing ore environment, in
Proceedings SAG2006 Conference, Mining and Mineral Process
Engineering, pp I-359 (University of British Columbia: Vancouver).
2. The reviewer of this paper suggested that ‘even if it does have an
effect, it is probably too hard to predict the right design prior to
operation so you may as well just go for a proven design for
maximum life.’
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
SOME OBSERVATIONS REGARDING SAG MILLING
Karageorgos, J, 2001. SAG milling at Fimiston, in Proceedings SAG2001
Conference, Mining and Mineral Process Engineering (University of
British Columbia: Vancouver).
Lane, G and Dunne, R, 1998. Ore competency and the design of
FAG/SAG circuits, Paper presented to Comminution 98, Minerals
Engineering Conference, Perth.
Loveday, B, Morrison, R, Henry, G and Naidoo, U, 2006. An
investigation of rock abrasion and breakage in a pilot-scale AG/SAG
mill, in Proceedings SAG2006 Conference, Mining and Mineral
Process Engineering, pp III-379 (University of British Columbia:
Vancouver).
MacNevin, W, 1997. Kidston Gold Mines case study: Evolution of the
comminution circuit, in Proceedings Conference on Crushing and
Grinding in the Mining Industry, Perth, May (IIR Conferences:
Sydney).
Morrison, R, Loveday, B, Powell, M, Djordjevic, N and Cleary, P, 2006.
Applying discrete element modelling to different modes of breakage
in AG and SAG mills, in Proceedings SAG2006 Conference, Mining
and Mineral Process Engineering, pp III-407 (University of British
Columbia: Vancouver).
Nordell, L, 2002. Personal communication.
Powell, M, 1988. A survey of milling and mill lining practice on South
African gold mines, Mintek report No M350, 21 June.
Powell, M and Smit, I, 2001. Startling effect of ball scats removal on
SAG mill performance, in Proceedings SAG2001 Conference,
Mining and Mineral Process Engineering (University of British
Columbia: Vancouver).
Rajamani, R K, Latchireddi, S, Prathy, S K and Patra, T, 2005. Improving
energy efficiency via optimised charge motion and slurry flow in
plant scale SAG mills, Annual report, December, DE-FC26-
03NT41786 University of Utah.
Royston, D, 2006. Current developments in SAG mill liner design, in
Proceedings SAG2006 Conference, Mining and Mineral Process
Engineering (University of British Columbia: Vancouver).
13
Differential Flotation Commissioning at Telfer
M A Benson1, C R Headley2, S O Hille3 and N P Jacob4

ABSTRACT
The redevelopment of the Telfer Gold Mine in the Pilbara region of
Western Australia was primarily based on the construction of a 17 Mtpa
dual train concentrator coupled with open pit and underground mining
operations. The unique copper-gold orebody at Telfer requires the
application of complex flotation conditions normally reserved for base
metal concentrators, where two competing minerals are managed into
separate products. This paper describes the commissioning of the
flotation circuit, the process modes and the master/slave relationship that
flotation has with a variable orebody.

INTRODUCTION
Telfer is located within the Great Sandy Desert of Western
Australia, 500 km south-east of Port Hedland (Figure 1). Mining
originally commenced at Telfer in the early 1970s and produced
around six million ounces of gold before, due to high operating
costs, Newcrest Mining Limited (Newcrest) decided to cease
operating the 2 Mtpa processing facility and associated services
in August 2000 and placed the mine on care and maintenance.
Newcrest then evaluated options for Telfer to improve its
economic viability. The results from these evaluations were
reported in Newcrest’s November 2002 Telfer Project feasibility
study. The feasibility study proposed the construction of a new
processing facility with a design capacity of 17 Mtpa. The
economic justification was based on gold and copper recoveries
over the 20 year life of the project.
The Telfer Gold Mine recommenced operations in late 2004
with the construction and commissioning of a new 17 Mtpa
processing plant. The new plant has two crushing, grinding and
flotation trains with common concentrate handling and residue
disposal facilities. The plant recovers coarse gold through a
gravity concentration circuit. Copper and fine gold are recovered
through a series of different flotation modes; bulk, copper only,
sequential and pyrite only flotation.
In the feasibility study each ore block was assigned a preferred
treatment route according to its rock type, copper domain and
mineral grades. The following paper describes how the
variability of the Telfer orebody dictates, in a master/slave
relationship, the flotation circuit configuration required.
BACKGROUND
The feasibility study evaluated the test work undertaken on
different ore types and assigned gold and copper recoveries to
each ore block according to the following:
1. treatment method – processing mode,
2. rock type – lithology,
3. copper domain – weathering profile, and
4. head grade – metal species.
1. Senior Plant Metallurgist, Newcrest Telfer Operations, PO Box
6380, East Perth WA 6892. Email: mark.benson@newcrest.com.au
2. Previously: Telfer Gold Mine, PO Box 6380, East Perth WA 6892.
3. Manager Metallurgy, Newcrest Operations, Hyatt Business Centre,
Level 2, 20 Terrace Road, East Perth WA 6004.
Email: simon.hille@newcrest.com.au
4. Metallurgical Superintendent, Newcrest Telfer Operations, PO Box
6380, East Perth WA 6892. Email: nick.jacob@newcrest.com.au
FIG 1 - Telfer location.
Treatment method
The new processing plant is divided into two parallel treatment
paths (trains) with the following unit processes: crushing,
grinding, gravity, copper flotation and pyrite rougher flotation.
The plant has common copper concentrate handling, tailings
disposal, pyrite cleaner flotation and pyrite concentrate leach
facilities. The flow sheet for the new processing plant is
presented in Figure 2.
The two processing trains are capable of treating different ore
blends with the treatment modes summarised in Table 1. The
preferred treatment method for each ore block depends primarily
on the copper (total and cyanide soluble) and sulfur grade of the
ore and is selected to give the optimum economic outcome.
A gravity gold component is recovered regardless of treatment
method as the gravity circuit is situated in the grinding circuit.
Geology and mineralogy
The stratigraphic setting and copper domains for the principal
open pit ore source, Main Dome, are depicted in Figure 3 and
Figure 4, respectively.
The gold and copper deposits at Telfer are hosted within a
1750 m thick sedimentary sequence comprising the Telfer
Formation, Malu Quartzite and Isdell Formation.
Several styles of mineralisation have been recognised, namely
narrow high-grade reefs, pod-like mineralised bodies, sheeted
vein-sets and large areas of low-grade stockwork mineralisation,
which form the majority of the sulfide resource. The
mineralisation is structurally controlled, but all styles display
stratabound characteristics.
Primary gold mineralisation is associated with pyrite-
chalcopyrite and quartz-dolomite gangue. However, weathering
has strongly modified the mineralisation, to depths ranging from

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 17

M A BENSON et al

Train 1 Train 2
Primary Primary
Crusher Crusher

Train 1 Train 2
Coarse Ore Coarse Ore
Stockpile Stockpile

Train 1 Train 2
Pebble Pebble
Train 1 Train 2
Stockpile Stockpile
Milling Milling
Circuit Circuit
Gravity Conc to dore Gravity Conc to dore

Train 1 Train 2
Copper Copper
Flotation Flotation
Tails Conc Tails Conc Combined
Concentrate
Copper
shipped to
Concentrate
smelter
Bulk Handling

Train 1 Train 2
Rougher Pyrite
Pyrite Rougher
Flotation Flotation Combined
Tails Conc Tails Conc Sequential / Pyrite
CIL
Pyrite Only Cleaning
Flotation

Dore gold
No pyrite produced
Copper Only
concentrate
produced
Combined
Tailings
Disposal

FIG 2 - Telfer processing plant flow sheet.

TABLE 1
Summary of treatment methods.

Flotation products
Copper circuit Pyrite circuit
Bulk High Cu:S Cu concentrate Cu concentrate
Copper only Low Cu:S + high CN sol Cu, low Au pyrite Cu concentrate Nil
Sequential Low Cu:S, low CN sol Cu, high Au pyrite Cu concentrate Pyrite concentrate
Pyrite only Low Cu:S, high Au pyrite, Low CN sol Cu Nil Pyrite concentrate

100 m to more than 200 m, although the boundary between oxide
and primary ore types is irregular. Supergene minerals include
gold with limonite/goethite, malachite and chrysocolla in the
depleted zone, giving way to chalcocite, pyrite, digenite,
covellite, tenorite and cuprite at depth.
The Telfer Project ore sources consist of the following broad
classifications:
1. Main Dome ore, a low-grade auriferous copper sulfide and
pyrite orebody providing the majority of the ore tonnage;
2. West Dome ore, a low-grade auriferous pyrite orebody
containing lesser copper sulfides and higher pyrite values
than the Main Dome orebody;
3. Middle Vale Reef (MVR) ores, comprising remnant reef
ore associated with higher grade massive auriferous copper
sulfide and pyrite mineralisation, with occasional
pyrrhotite, adjacent to the existing underground mine;
4. M Series Reef (MRF) ore, comprising ore associated with
the narrow vein reefs within the Malu Quartzite; and

18 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 DIFFERENTIAL FLOTATION COMMISSIONING AT TELFER

FIG 3 - Telfer project stratigraphy.

FIG 4 - Telfer project copper domains.

5. Telfer Deeps underground mine ore, from the I Series
development, with higher grade massive auriferous copper
sulfide and pyrite mineralisation with occasional pyrrhotite.
The weathering profile varies across the mine. Around Main
Dome, the redox boundary is generally 100 m to 200 m below
the surface although the reef structures and impermeable
quartzite boundaries have a strong effect on the level of
oxidation.
Supergene enrichment and leaching of copper and, to a lesser
more localised extent, gold, occurs throughout the mine. Copper
is, in general, far more mobile than the gold in the weathering
profile, being leached from near-surface areas down to roughly
60 m to 80 m below the surface, where there is a strong zone of
supergene enrichment. The subhorizontal supergene copper
‘blanket’ is generally near or above the oxide/partially oxidised/
sulfide redox boundary.
Localised leaching and enrichment of gold and copper, and
oxidation of sulfides with related occurrence of secondary copper
sulfides, occur in deeper positions along permeable reef horizons
and structures.
Malu Quartzite Member (MQM), Footwall Sandstone Member
(FSM), Malu Sandstone Member (MSM) and MRF ore account
for the majority of the Telfer open pit orebodies, with MQM
being the dominant lithology. In general, copper domain is
strongly correlated with flotation performance and lithology is
associated with physical characteristics. The MVR and E reef
(REF) lithologies are typical of weathered reefs, with low
competency, and have given poorer flotation response during test
work.
THE MASTER/SLAVE RELATIONSHIP
As previously mentioned, the ore fed to the Telfer concentrator
can be classified into four groups according to the preferred

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 19

M A BENSON et al
treatment mode: bulk, sequential, copper only and pyrite only.
The type of ore (rock type and copper domain) being treated
dictates the flotation circuit configuration required, or in process
control language, is the master control element. Processing of ore
in the incorrect treatment mode, either due to operational error or
by misclassification of ore, can have detrimental effects on gold
and copper recovery and copper concentrate grade. In this way
the flotation circuit and the mineral products are slaves to the
mineralogy of the ore blend being fed while accommodating the
market constraints for the products.
Bulk flotation
The simplest of the flotation modes at Telfer, bulk flotation is
defined as a non-selective copper flotation process where all
sulfide species are floated and a bulk copper concentrate is
produced. It is typically characterised by Cu:S ratios >0.7,
generally being high in chalcocite and low in pyrite. There can
be a variety of other copper species present (covellite, digenite,
chalcopyrite, bornite, native copper, malachite, cuprite) but will
typically produce an economic concentrate of >20 per cent Cu.
The majority of the bulk ore comes from the upper open pit
region, which is dominated by the supergene structures.
The bulk mode circuit configuration incorporates flash
flotation with dedicated flash cleaner, copper rougher, copper
scavenger (pyrite rougher) and copper cleaner/recleaner circuits.
The high Cu:S ratio allows for the non-selective flotation
conditions. These conditions include a non-selective sulfide
collector in potassium amyl xanthate (PAX), lime addition to
adjust pH to 9.0 and sodium sulfide (Na2S) to utilise controlled
potential sulfidisation (CPS).
PAX is preferentially added to the mill discharge to allow
conditioning time, servicing both the flash flotation, located in
the grinding circuit recirculating load, and the copper rougher
circuit, which is fed with the grinding circuit cyclone overflow.
Additional PAX can be added to the head of the flash and copper
roughers as required. Na2S for CPS is dosed in the feed to the
pyrite roughers and copper cleaner conditioning tank to
maximise the recovery of weakly sulfidic minerals, copper
carbonates and copper oxides.
Bulk flotation mode results in maximum recovery of gold and
copper whilst maintaining the target concentrate grade. However,
if the ore entering the plant has a lower Cu:S ratio of <0.7, it
becomes very difficult to achieve a saleable copper concentrate,
which may lead to a change in the circuit configuration. When
the ore blend is high in copper oxides and copper carbonates,
whilst still indicating a high copper to sulfur ratio, the flotation
circuit may struggle to achieve the target concentrate grade. This
is where the CPS is of greatest value if applied correctly but can
lead to a dilution of the final concentrate grade. So even within
the selected processing modes the specific mineralogy dictates
the performance of the circuit in a master/slave relationship.
Copper only
One of the more challenging flotation modes at Telfer, copper
only flotation is defined as a selective copper flotation process
where only copper sulfide species are floated to produce a copper
concentrate. The pyrite component of the ore is not economic to
recover for further processing. It is typically characterised by
Cu:S ratios in the range between >0.4 and <0.7 generally being a
mixture of copper sulfides, oxide and carbonates with moderate
to high proportions of pyrite. These copper only ores are
generally characterised by very high proportions of cyanide
soluble copper (CNSCu) which, due to their slow floating nature,
would report to the pyrite concentrate in unfeasible quantity.
Another feature of this processing mode is that pyrite depression
is necessary to produce an economic concentrate of >20 per cent
Cu. The majority of the copper only ore comes from the
transitional area of the upper open pit, which is dominated by the
supergene structures which have been highly weathered.
20 Fremantle, WA, 19 - 21 March 2007
The copper only mode circuit configuration incorporates flash
flotation with dedicated flash cleaner, copper rougher, copper
scavenger (pyrite rougher) and copper cleaner/recleaner circuits.
The range of Cu:S ratios requires selective flotation conditions,
where the amount of pyrite reporting to the final concentrate is
managed through chemical depression. This is achieved by a
combination of a selective thionocarbamate copper collector
(RTD11A) and lime/cyanide addition.
RTD11A is preferentially added to the flash flotation feed with
the copper rougher circuit fed with the grinding circuit cyclone
overflow. Additional RTD11A can be added to the head of the
copper rougher and scavengers as required. The cyanide is dosed
in the flash cleaners and copper cleaner/recleaners to depress
pyrite and maximise the copper concentrate grade.
Copper only flotation mode results in an economic copper
concentrate containing both Cu and Au as the main valuable
minerals. If the ore entering the plant has a lower Cu:S ratio of
<0.4 and is low in CNSCu then it is possible to exploit this
difference by recovering greater amounts of the pyrite by
reducing the strength of the cyanide depression. This may lead to
a change in the circuit configuration. Typically CPS is not
utilised in copper only mode as it leads to large amounts of pyrite
activation and the flotation circuit may struggle to achieve the
target concentrate grade. So even within the selected processing
modes the specific mineralogy dictates the performance of the
circuit in a master/slave relationship.
Sequential flotation
The most difficult flotation mode that Telfer has to offer,
sequential flotation is defined as a selective copper and pyrite
(gold) flotation process where the copper sulfide species are
floated to produce a copper concentrate while the pyrite
component of the ore is depressed using high pH and cyanide.
The pyrite is then activated and floated to recover gold through a
separate processing circuit consisting of cyanide leaching and
cyanide recovery. Sequential ores are typically characterised
by Cu:S ratios <0.4, with a mixture of copper sulfides,
predominately chalcopyrite. They also contain high proportions
of pyrite bearing gold and are typically low in CNSCu. The
majority of the Sequential ore comes from the lower open pit and
Telfer Deeps, which is dominated by the Malu and Isdell
stockwork and reef structures, is primary in nature with minimal
weathering.
Different conditions are required to recover the copper and
pyrite concentrate separately. The copper selective collector,
RTD11A, is added to the mill discharge, head of the flash
rougher, head of the copper rougher and the copper cleaner
circuits. Utilising RTD11A in combination with lime/cyanide
addition to the flash cleaners and copper cleaners allows a
saleable grade copper concentrate to be produced. PAX is not
used in the flash flotation and copper rougher circuits to
minimise pyrite recovery to the copper concentrate. The copper
rougher tail feeds the pyrite rougher circuit where CPS and PAX
addition ensures maximum recovery of pyrite and gold to the
pyrite cleaner circuit. To reduce the copper content of the pyrite
cleaner concentrate, a two-stage cyclone desliming circuit is
used. The deslimed pyrite concentrate feeds the pyrite leach
circuit to recover gold.
Sequential flotation has similar issues to copper only treatment
in that the amount of cyanide required for depression is
dependent on the proportion of pyrite to the copper minerals in
the feed. High quantities of CNSCu or a high CNSCu:Cu ratio in
the feed can result in slow floating copper minerals exiting in the
copper rougher tail and being activated in the pyrite rougher
circuit. Recovery of copper minerals into the pyrite concentrate
results in excessive cyanide consumption in the leach circuit. If
the ore to be fed contains a CNSCu:Cu ratio greater than
0.5 and/or a CNSCu grade above 400 pmm, it is likely that the
Ninth Mill Operators’ Conference

copper reporting to the pyrite concentrate will be too high to
feasibly treat in the leach circuit. The ore may then require bulk
or copper only treatment (depending on Cu:S grade). Hence the
processing team needs to decide which of the minerals is to
dominate the economic decision (master) and adjust the circuit
accordingly (slave).
Pyrite only
One of the challenging flotation modes at Telfer, pyrite only
flotation is defined as a pyrite (gold) flotation process where
there is insufficient copper sulfide species to produce a copper
concentrate. The pyrite is floated into an economic concentrate
to recover gold through a separate processing circuit. It is
typically characterised by Cu:S ratios <0.4 generally being a
mixture of copper sulfides, predominately chalcopyrite, with
high proportions of pyrite bearing gold. Pyrite only ores are
typically very low in cyanide soluble copper. The majority of the
pyrite only ore comes from the lower open pit and Telfer Deeps,
which has been depleted in copper minerals. Like sequential ore
the lithology is dominated by the Malu and Isdell stockwork and
reef structures and is primary in nature with minimal weathering.
Pyrite only mode recovers only a pyrite concentrate, and
therefore a more simplified reagent regime can be employed.
However, the main concentrate production needs to come from
the pyrite roughers to enable recovery through the pyrite cleaners
to the pyrite leach circuit. This requires the majority of the
flotation circuit to be under-utilised as there is no copper
concentrate produced. Both CPS and PAX are applied to the
conditioning tanks at the head of the pyrite rougher circuit to
ensure maximum recovery of pyrite and gold to the pyrite
cleaner circuit.
Pyrite only flotation has some similar issues that exist in
sequential flotation in that a careful watch needs to be
maintained on the amount of CNSCu which reports to the pyrite
concentrate, due to the impact that the copper species will have
on the cyanide consumption in the pyrite leach. Again the
processing team needs to decide which of the minerals is to
dominate the economic decision (master) and the circuit to adjust
accordingly (slave).
COMMISSIONING EXPERIENCE
Bulk flotation commissioning
First ore was processed through Train 2 on 23 November 2004.
At this time there was very limited instrumentation and
construction of Train 1, and parts of Train 2 were continuing.
Some of the critical elements for flotation operation and
monitoring were not available or not adequately commissioned
and calibrated. Train 1 was commissioned on the 18 January
2005.
The commissioning of a new concentrator is usually a
challenging task and at Telfer this was certainly the case for a
number of reasons.
Complex, highly variable ore mineralogy
Bulk flotation ores have many different copper domains and rock
classifications in addition to grade variations. Copper mineralogy
varies from oxides, carbonates, partially oxidised copper sulfides
to readily floatable copper sulfide minerals such as chalcocite,
bornite, covellite and chalcopyrite. This mineralogy variability is
typical of a supergene geological structure, which some refer to
as a ‘mineral soup’. The variable copper grades and mineralogy
result in different flotation characteristics and therefore variable
flotation operating targets, such as pulp densities and reagent dose
rates.
During commissioning the effect of the different copper
domains and rock types on flotation was not well understood. In
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
DIFFERENTIAL FLOTATION COMMISSIONING AT TELFER
addition, the feed could vary dramatically over short periods of
time requiring significant changes to the flotation circuit to
maintain froth stability, pull rates, concentrate grade and
recovery.
Some ore types contained high levels of oxidised clay-type
minerals from the outer siltstone member (OSM) section of the
open pit. With such ores, it was found that the slurry was
excessively viscous at the design flotation density of 35 per cent
solids, resulting in poor air dispersion through the pulp, boiling
and froth instability. As a result the float feed density target had
to be dropped to 30 per cent and at times as low as 28 per cent
solids, requiring significantly higher addition of water through
the grinding circuit. At times the higher water additions resulted
in lower throughputs due to limited supply of make-up water.
Flash flotation commissioning
On plant start-up the flash flotation circuit was not yet available.
As previously mentioned, the flash flotation treats a portion of
the recirculating load (cyclone underflow) in the grinding circuit.
This removes easily floatable minerals at the earliest possible
opportunity preventing over-grinding and therefore reducing
reagent consumption and improving roughing circuit performance.
Approximately 40 per cent of all copper concentrate produced is
from the flash flotation circuit.
The flash flotation circuit consists of an Outokumpu SK1200
rougher cell with the concentrate reporting to two OK6.5TC
flash cleaners producing a final concentrate.
Without the flash flotation circuit, the roughing and
cleaning/recleaning circuits became overloaded in terms of froth
mineral loadings. Increased frother dosages were required to
maintain froth stability in the roughing circuit and throughput
was restricted to avoid high tails metal losses.
Once commissioned, the high levels of metal production from
the flash flotation circuit initially caused some significant
materials handling issues. The concentrate recovered included
significant levels of native copper, which combined with almost
pure copper sulfides and pyrite, resulting in a very high specific
gravity product. Frequent pipeline blockages and high wear on
pump and hopper components occurred transferring the
concentrate over a long and indirect route. A redesign of the
pumping system was necessary, including upgrading the pumps,
redesigning the transfer pipework and incorporation water
injection points. Current operation relies heavily on water
addition and process control to manage the high specific gravity
material over the long pipe run.
Suboptimal process control tuning and
availability of instrumentation
As a result of significant instability in the grinding circuit, the
flotation circuits experienced large flow surges. This would cause
flotation cell pulping and overflows due to the ineffectiveness of
the float level control to deal with such large surges.
The proportional, integral and derivative (PID) settings for the
various control loops within the Telfer concentrator were initially
set at default levels for slurry commissioning. These settings
were changed ‘on the fly’ as the circuit quickly ramped up to
design throughputs with over 3000 m3/h of slurry entering the
flotation circuit. Control of flow control valves for water, air and
reagent addition and level control on hoppers required rapid
tuning to minimise the environmental impact of slurry spillage
and to allow some semblance of steady operation. Bedding down
of the control loops during the commissioning of Train 2 was of
an extremely high priority, not only to ensure environmental
impact (spillage outside of bunded areas) was minimised whilst
recovery targets and concentrate specifications were met, but
also due to the fact that Train 1 was still under construction.
21
M A BENSON et al
Construction activities were being conducted within close
proximity of hoppers and slurry lines, with loss of control having
potential negative ramifications on industrial relations. Early
learnings during the commissioning of Train 2 were immediately
applied to Train 1 before commissioning of this circuit two
months later.
Some of the critical elements for flotation operation and
monitoring were not available or not adequately commissioned
and calibrated. One such piece of equipment now considered
critical to the flotation process at Telfer is the multi-stream
analyser (MSA). The MSA analyses a number of critical streams
throughout the circuit via an X-ray fluorescence elemental
analyser. The outputs of estimated grades for copper, iron, per
cent solids and inferred sulfur can then be used to monitor stream
grades in order to optimise the process.
Poor reliability in the pumps feeding the analyser resulted in
this important control system being unavailable for much of
the first year of operation. During commissioning and the
subsequent period until the MSA was completed, flotation
control was maintained by collecting spot samples of critical
streams for assay every four hours. The assaying had on average
a six-hour turn around time. Often, by the time an issue was
identified in the circuit, it had already been half a shift before
any corrective action could take place, resulting in lower than
desired recoveries and/or concentrate grades.
Resolving the pump issues and the opportunity to calibrate the
streams has now allowed the MSA to be used as an instantaneous
measure of the flotation circuit performance, resulting in much
tighter control over pull rates and the addition of reagents.
The digital froth camera system initially installed at Telfer
proved to be unreliable due to deficiencies in the hardware
architecture utilised and the negative effects of sunlight. This
system was not employed during commissioning or in the first
year of operation. The system required an upgrade and has now
been incorporated into Telfer’s process control. The digital froth
camera system is an important control method used to monitor
concentrate pull rates ensuring they are adequate through the
roughing circuits and manipulate the cell levels as required.
Froth stability issues
The initial commissioning period was punctuated by several
froth stability issues. The main causes were both mineralogy and
chemically driven. The mineralogy issues related more to the
host rock rather than the type of sulfides being recovered. The
highly weathered outer siltstone member (OSM) ore, which hosts
the E reef structure, caused the most significant froth issues. As
previously mentioned the viscous slurry affected the formation of
the froth, whilst the naturally hydrophobic gangue minerals
produce a highly stable froth once formed.
Telfer commenced operation utilising methyl isobutyl carbinol
(MIBC), a weak alcohol-based frother. As Train 2 commenced
operation without the flash flotation circuit, increased mineral
loadings in the copper rougher circuit resulted. The mineral
loading, along with the viscous nature of the OSM ore, required
high addition rates of MIBC to stabilise the froth.
The Telfer circuit was then found to be very sensitive to
frother addition. Increasing frother to the rougher circuit resulted
in ‘over-frothing’ in the cleaner circuit making the cells ‘race’
with uncontrollable mass pull rates. The large proportion of
gangue entrainment in the froth reduced the grade, and the
persistent nature of the froth reduced the efficiency of the cleaner
and recleaner concentrate pumps. Recovery was then impacted
due to lower achievable mass pull.
To resolve frothing issues, close attention was required to ore
blending the limit the proportion of OSM ore. As covered below,
the frother strength and type was altered to provide greater
control and the addition points optimised. The addition rates
were drastically reduced and a higher level of process control
was adapted.
22 Fremantle, WA, 19 - 21 March 2007
Reagent addition and CPS optimisation
Reagent addition rates from the feasibility study test work
provided a starting point for dose rates but were quickly reduced
to gain better control of froth stability without adversely
affecting concentrate grade and recovery. This is common when
scaling up bench top test work flotation reagent regimes to plant
scale operation.
Initial addition rates for MIBC exceeded 30 g/t to ensure
adequate froth stability to the flotation circuits. Alternative
frothers to MIBC were trialled, resulting in the switch to an
alcohol-glycol blend frother, DSF-004. Stronger frothers were
trailed which provided improved flash flotation stability;
however, a compromise was made to reduce adverse overfrothing
in the cleaning circuit. Frother addition rates, DSF004, have
since been reduced to 3 - 10 g/t.
Telfer commenced operation utilising PAX for the non-
selective flotation of sulfide minerals into a bulk copper
concentrate. Initial addition rates were based on feasibility, at
times exceeding 150 g/t. The addition point of the PAX was
found to have a significant effect on froth stability. Addition of
PAX at the head of the flash and copper rougher circuits revealed
detrimental effects on froth stability due to a lack of
conditioning. Addition of PAX to the mill discharge became the
preferred option providing conditioning time prior to both the
copper rougher and the flash circuit, PAX addition was refined to
30 - 50 g/t by being staged throughout the circuit, though a
change in the ore being delivered to the plant meant treatment in
bulk mode, with exclusive use of PAX as the collector, was short
lived.
Treatment of high sulfide feed and ore with a high Cu:S ratio
required the introduction of RTD11A, a copper selective
thionocarbamate collector. Operating in copper only mode
became the norm in the early stages of operation, with RTD11A
added to the mill discharge, as well as the heads of both the
copper and flash roughing circuits. The nature of RTD11A meant
little conditioning time was required and no issues were found
with froth stability.
Controlled potential sulfidisation (CPS) utilises the addition of
sodium sulfide (Na2S) or sodium hydrogen sulfide (NaHS) in
order to reactivate partially oxidised copper sulfide or carbonate
minerals. Coating of the mineral surface with fresh sulfide ions
allowed sulfide collector absorption and flotation recovery. The
CPS at Telfer took considerable effort to understand and
optimise.
Feasibility study copper scavenger CPS was controlled to a
slurry potential of -200 mV Eh. Operating CPS to this Eh set
point in the plant proved to be suboptimal. Instead, operating to a
sulfide ion potential (Es) of -500 mV was determined to be
optimal for the flotation of malachite, the predominant oxidised
copper mineral present in the Telfer orebody. Once this was
understood the measurement of the slurry potential was
converted from standard oxidation reduction potential (Eh) to
that of the sulfide ion only (Es) through the use of a sulfide ion
selective electrode.
Depression of pyrite is critical at Telfer, both to maintain a
saleable copper concentrate, and to ensure the gold bearing
pyrite reports to the leaching circuit via the pyrite roughers. Lime
was initially the only reagent available for this. Lime was added
both as quick lime to the concentrator feed belts and milk of lime
to the flotation circuits. The lime was used to modify pH to 9 to
10 in the roughing circuit and 10 to 11 in the cleaning circuit.
Minor success in depressing the pyrite resulted; however, the
nature of the ore treated and the delay in the construction of the
permanent cyanide dosing circuit required the installation of a
temporary cyanide dosing system. The temporary system dosed
high cyanide addition rates to the flash and copper rougher
circuits to depress pyrite. Once the permanent system was
Ninth Mill Operators’ Conference

available, preferential cyanide dosage to the cleaner circuits
drastically decreased reagent consumption whilst significantly
improving control over pyrite depression. This optimisation is a
change from design but has enabled a higher level of control and
improved performance.
Sequential commissioning
Limited amount of sequential ore feed was available from the
mine in the upper levels of the open pit delaying sequential
flotation commissioning by 14 months. Sequential flotation
commissioning occurred in January of 2006, coinciding with
the production ramp-up of the redeveloped Telfer Deeps
underground operation.
In the initial sequential trial, it proved very difficult to float the
pyrite after successful pyrite depression through the addition of
cyanide and copper flotation. The reason for this was not known
at the time, but was suspected to be a re-activation issue, related
to either pH or Eh. Test work showed:
• excessive cyanide depression results in poor re-activation;
• pyrite re-activation was enhanced at pH 8.0 - 7.0, through the
addition of acid;
• on fresh sulfide ores, a slightly negative pulp Eh (-80 to
-120 mV) in the pyrite conditioning discharge has provided
the best flotation results, highest sulfur recovery and lowest
copper grade;
• residual reagents in the process water (recycled) were the
main chemical variable which impacted all flotation stages
(copper and pyrite); and
• 50 - 60 g/t PAX addition is optimal to recover pyrite but not
leave residual PAX in process water that may float pyrite in
the flash and copper roughers.
On fresh sulfide ores, typical grades of pyrite thickener feed
have been 0.5 per cent Cu and 35 per cent S. Sequential
treatment of oxidised ores (high CNSCu content) result in pyrite
concentrates with a high copper content (>3 per cent) as the
oxidised copper minerals do not float in the copper circuit, but
are activated by CPS and report to the pyrite concentrate.
To reduce the copper content of the pyrite cyanide leach feed,
a two-stage cyclone deslime circuit is used. Analysis of copper
minerals in pyrite concentrate by size shows these minerals are
predominantly in the slimes size fraction. Optimisation of
cyclone cut points for most effective copper rejection is ongoing.
KEY LEARNINGS – COMMISSIONING
Bulk
• Ore variability has a significant effect on flotation operating
parameters; ore blending to maintain steady
characteristics is critical.
• MSA is critical for optimal operation of the flotation process.
• Froth stability needs to be maintained through all ore blends
and different addition strategies need to be applied to
accommodate both ore variability and flotation unit
performance.
• Feasibility study reagent additions were much higher than
required in the processing plant. Bench scale test work
regimes need to be viewed with a high degree of scepticism
and an open-minded approach to the initial operating reagent
regime should be adopted.
• Reagent addition rates have a huge effect on froth stability in
addition to flotation performance. Close control of reagent
addition is required at all times.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
DIFFERENTIAL FLOTATION COMMISSIONING AT TELFER
• Reagent conditioning time is often excluded in final design
in a quest to reduce capital cost; however, when dealing with
a complex chemical flotation process it can be invaluable to
both fast optimisation and process control.
Sequential
• Excessive cyanide depression of pyrite results in poor
re-activation and recovery;
• modifying addition of depressant into the cleaning block
allows for tighter control, reduced reagent consumption and
enhanced performance;
• pyrite re-activation was enhanced at pH 8.0 - 7.0, through the
addition of acid;
• on fresh sulfide ores, a slightly negative pulp Eh (-80 to
-120 mV) in the pyrite conditioning discharge has provided
the best flotation results;
• 50 - 60 g/t PAX addition is optimal to recover pyrite but not
leave residual PAX in process water that may float pyrite in
the flash and copper roughers;
• treatment of oxidised ores in sequential mode results in
excessive levels of oxidised copper minerals in the pyrite
concentrate, as they require CPS to activate, resulting in
uneconomic downstream cyanide leaching;
• residual reagents in the process water (recycled) were the
main chemical variable which impacted all flotation stages
(copper and pyrite); and
• good gold recovery relies on good sulfur recovery due to
gold-pyrite mineral associations.
CURRENT OPERATING PHILOSOPHY
Bulk
At the current point in the project life, there is not a great deal of
bulk ores coming from the open pit. Bulk treatment is the easiest
and most favourable treatment mode and when bulk ores are
treated, the flotation conditions are geared for maximum
recovery with PAX collector addition at higher dose rates and
CPS to recover all floatable minerals. The high Cu:S ratio allows
good concentrate grades at these less selective flotation
conditions.
Copper only
Copper only mode is the least favourable in terms of metal
recovery. Due to concentrate grade constraints, it is not possible
to recover all the pyrite into the copper concentrate, and due to
the presence of oxidised cyanide soluble copper minerals, it is
not possible to produce a separate economically treatable pyrite
feed
concentrate. Therefore, there are gold losses associated with the
pyrite that reports to tailings.
A number of strategies have been employed to minimise such
losses, including adding cyanide depressant to the cleaners only
to maximise free gold recovery and the use of stronger PAX
collector in the scavengers to increase pyrite recovery if
concentrate copper grade allows.
Sequential
To ensure good copper concentrate grades and a clean pyrite
concentrate, it is essential to get a good split between the copper
and pyrite concentrates. This is controlled by correct reagent
additions, pull rates and ore classification and blending.
The decision whether an ore is to be treated in copper only or
sequential mode is an economic one – if the pyrite concentrate
23
M A BENSON et al
produced is economic to leach, the ore is treated in sequential
mode. In addition to the economics, it is necessary to control
copper metal in the pyrite concentrate fed to the leach circuit to
ensure the SART copper and cyanide recovery process is not
overloaded, this is managed by ore blending and correct
classification of sequential ore.
CONCLUSION
The more complex the mineralogy the more processing options
are necessary to accommodate the variability while still achieving
an economic result. The Telfer ore processing circuit is unique in
this way for a copper-gold deposit where the minerals dictate, in
a master/slave relationship, the processing configuration of the
flotation circuit.
The commissioning of the new Telfer concentrator, while
challenging, has been an extremely successful ramp-up of a large
dual train concentrator, exceeding nameplate capacity within the
six weeks of operation which is now consistently processing at
rates in excess of 20 Mtpa. The flotation performance for both
copper and gold has also tracked closely to the feasibility study
estimates, which are primarily based on fixed tailings grade
models.
The key learnings from the commissioning that the
metallurgical team would like to share are:
24 Fremantle, WA, 19 - 21 March 2007
• The importance of ore blending based on mineralogy and
flotation response, not necessarily solely on economic metal
grades.
• The role that process control has in stabilising and optimising
the process. Some consideration should be given to
allocation of a separate process control team, which is
dedicated to this task alone during commissioning.
• The need to approach all bench scale reagent regimes with an
open mind when applied to the processing plant and rely
more heavily on experienced operators and visual indicators
to conduct coarse reagent adjustments.
• Reagent conditioning time can have a major influence on the
effectiveness and quantity of the reagent necessary.
• Close attention needs to be paid to the water balance and the
impact of recycled water on the chemical balance within the
circuit.
ACKNOWLEDGEMENTS
The authors would like to thank Newcrest Mining Limited and
the management at Telfer Gold Mine for permission to publish
this paper. The authors would also like to recognise the entire ore
processing team for their commitment and perseverance through
the challenging times of commissioning to renew Telfer and
optimise a world-class ore processing plant.
Ninth Mill Operators’ Conference
A Model Study of Copper Rougher Recovery Optimisation at
PT Freeport Indonesia
S R Grano1, T Akroyd2 and M A Mular3
ABSTRACT
A fundamental model, which considers the subprocesses of particle
collection by bubbles in flotation, is applied to two comprehensive
flotation surveys in the rougher stages of C3 concentrator, PT Freeport
Indonesia. The flotation surveys focused on the collection of:
• cell-by-cell and size-by-size metallurgical,
• cell hydrodynamic, and
• pulp and surface chemistry information.
The two surveys were conducted at significantly different feed
tonnages, giving rise to significantly different feed particle size
distributions, pulp densities and mean pulp residence times. A series of
model simulations are performed with the aim of quantifying changes in
copper recovery, on both a size-by-size and unsized basis, with changes
in values of the hydrodynamic parameters and hydrophobicity of the
copper sulfide mineral. This was carried out in an effort to identify the
most promising route to increase overall copper recovery in roughing. It
was found that by increasing the hydrophobicity of the copper sulfide
mineral surfaces it became possible to optimise cell hydrodynamics,
which targeted increased fine (-12 µm) copper sulfide mineral recovery
but without adversely affecting coarse (+150 µm) copper sulfide mineral
recovery.
INTRODUCTION
PT Freeport Indonesia (PTFI) operates a large copper/gold
mining and concentration facility, located in Papua, Indonesia.
The copper/gold ore is processed through four different
concentrators at an overall processing rate of approximately
225 K t/d. This discussion refers principally to the C3
concentrator, which includes in its equipment list:
• 10.3 m × 5.2 m SAG mill (12.5 MW),
• two 6.1 m × 9.3 m primary ball mills (6.4 MW),
• 36 Wemco 85 m3 rougher cells,
• a 4.0 m × 7.6 m regrind ball mill (1.9 MW),
• conventional column cells for primary (six 3.66φ × 15.2H m)
and secondary (eight 3.66φ × 15.2H m) cleaning, and
• 12 Wemco 85 m3 cleaner-scavenger cells.
The focus of this paper is on the rougher circuit, which is
described in more detail below. The ore feed comprises
approximately 0.9 - 1.1 per cent copper and 0.5 - 1.5 g/t gold. The
principal copper-bearing minerals are chalcopyrite (~75 per cent
of the total copper), bornite and chalcocite. Pyrite (~4 per cent) is
the principal iron sulfide gangue. The rougher flotation feed
particle size is approximately d80 210 µm, whilst the regrind
cyclone overflow particle size is approximately d80 38 µm. The
rougher copper recovery is approximately 90 - 95 per cent, with
~70 - 80 per cent of the losses in the rougher tailing occurring in
coarse (+150 µm) particle size fractions. There is relatively
minor loss (~6 per cent) of copper in fine size fractions (-12 µm)
of the rougher tailing. Approximately 85 per cent of the copper
1. MAusIMM, Ian Wark Research Institute, University of South
Australia, Mawson Lakes Campus, Mawson Lakes SA 5095.
Email: Stephen.grano@unisa.edu.au
2. Zinifex Century Mine, PO Box 8016, Garbutt BC, Garbutt Qld 4814.
3. PT Freeport Indonesia, 1615 Poydras Street, New Orleans LA 70112,
USA.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
contained in coarse (+150 µm) size fractions of the rougher
tailing is in mineral liberation classes, which contain 20 per cent,
or less by area, copper sulfide mineral, principally as chalcopyrite
in silicate binaries.
This paper focuses on the cell hydrodynamic and copper
sulfide surface conditions required to optimise both fine and
coarse copper recovery from the rougher cells in the C3
concentrator. Clearly, increasing copper recovery across the
rougher circuit needs to principally address recovery of coarse,
chalcopyrite-bearing composites of low chalcopyrite composition.
The low exposure of the chalcopyrite will reduce the overall
hydrophobicity of the composite particle, a point which is
discussed further below. Broadly speaking, the principal reasons
for the reduced recovery of composites particles are:
• detachment of coarse, poorly hydrophobic particles from
bubbles in the collection zone of the flotation cell (ie low
stability efficiency); and
• low froth recovery of attached particles on bubbles across the
froth zone of the flotation cell.
Our principal focus here is on the former possibility, as initial
measurements of froth recovery using the mass balance approach
(Vera, Franzidis and Manlapig, 1999), suggested that the froth
phase is not significantly limiting to overall copper recovery
across the roughers, under normal process conditions (Grano,
Weedon and Akroyd, 2005).
In order to determine the optimum conditions for recovery on
a particle size-by-size basis it is necessary to consider the
fundamentals of the flotation process. A fundamental model is
applied to the data sets derived from two comprehensive surveys
of the rougher circuit. Furthermore, the model is used to quantify
the effect of cell hydrodynamics (ie energy dissipation, bubble
diameter) and copper sulfide surface condition (ie mineral
contact angle) on fine and coarse copper recovery. It is not the
intention here to outline the model in detail, but to summarise its
most salient points. There are a number of key papers that
describe the model in detail (Pyke, Fornasiero and Ralston, 2003;
Duan, Fornasiero and Ralston, 2003). For particle capture to
occur, a bubble and a hydrophobic particle must first collide.
After collision, the liquid film between the bubble and the
particle will drain, leading to a critical film thickness, when
rupture occurs. The three phase line of contact (ie the boundary
between the solid particle, the receding liquid and the advancing
gas phase) then moves across the particle surface until a stable
wetting perimeter is established. This sequence of drainage,
rupture and contact line movement constitutes attachment. Once
the particle is attached to the bubble as an aggregate it can only
be detached if there is sufficient shear (kinetic energy) in the cell.
Thus, the overall process of particle capture by bubbles, the
collection efficiency, may be defined as follows:
E coll = E c E a E s (1)
where:
Ec is the collision efficiency
Ea is the attachment efficiency
Es is the stability efficiency
25
S R GRANO, T AKROYD and M A MULAR
Each of these efficiency parameters is described separately to
develop a mathematical description of bubble-particle capture.
From the collection efficiency, the rate of recovery of particles
from a flotation cell can be determined. The collection zone rate
constant, kc, may be determined by examining the collision
frequency between bubbles and particles, z, the number of
bubbles, Nb, and the number of particles in the cell, Np. The rate
of removal of particles from a flotation cell may then be
calculated as follows:
dN p
= − zN p N b E coll (2)
dt
dN p
= −k c N p (3)
dt
Thus, the collection zone rate constant may be determined as
follows:
k c = zN b E coll (4)
The collision efficiency is mainly controlled by the bulk
hydrodynamics in the flotation cell. The Generalized Sutherland
Equation (or GSE) takes account of both interception and inertial
forces (Dai, Fornasiero and Ralston, 2000). Here, particle size
and density, bubble size and velocity all affect the value of the
collision efficiency. For fine particles (<12 µm), the collision
efficiency often limits their flotation rate, giving rise to reduced
recovery at plant scale. Following collision, the particle must
attach to the bubble. This process is controlled by the pulp and
surface chemistry of the system that, together, ultimately control
the surface hydrophobicity. The attachment efficiency is defined
as the fraction of colliding particles that reside on the bubble for
a time greater than the induction time (Dobby and Finch, 1987;
Dai, Fornasiero and Ralston, 1999). Surface hydrophobicity may
be quantified by the contact angle. In the case of particle
attachment to the bubble, the receding contact angle is most
relevant. In this current paper, Time-of-Flight Secondary Ion
Mass Spectrometry (ToF-SIMS) is used to estimate the copper
sulfide mineral receding contact angle, based on an existing
correlation between contact angle and adsorbed collector
intensity determined by ToF-SIMS for single mineral samples of
chalcopyrite (Duan, Fornasiero and Ralston, 2003). The stability
efficiency (Es) is controlled both by hydrodynamics and surface
hydrophobicity. The probability of bubble-particle aggregate
stability depends on the attachment force between the bubble and
the particle in relation to the external forces in the cell
environment. Schulze (1977) suggested that the average mean
vessel energy dissipation is more important in determining the
mean stability of the aggregates, rather than local energy
dissipation. A force balance around a spherical particle at the
liquid/gas interface (Schulze, 1993) can then be used to
determine the stability efficiency. Calculations suggest that the
stability efficiency becomes important for particles greater than
150 µm. The collision frequency between bubbles and particles,
z, in a turbulent environment is highly dependent on the local
energy dissipation in the cell, bubble and particle size and pulp
viscosity, and is effectively independent of buoyancy-related
forces (eg Abrahamson, 1975). However, in the quiescent zones
of a flotation cell, bubbles and particles will rise and settle due to
buoyancy. These body forces are also included in the collision
frequency calculation used here (Bloom and Heindel, 2002).
Finally, the recovery of particles from the collection zone to
the launder is described. Using the collection zone rate constant,
kc, the particle recovery from the collection zone for a perfectly
mixed cell can be calculated as follows. The mixing regime in
the cell must, of course, be verified. In the case of the PTFI
rougher cells a tracer test was conducted to demonstrate that the
liquid phase was perfectly mixed.
26 Fremantle, WA, 19 - 21 March 2007
k c TMRT
R' p = (5)
1 + k c TMRT
where:
TMRT is the mean pulp residence time in the collection zone
The recovery of particles (both hydrophobic and hydrophilic)
by entrainment (RCellEnt) is calculated using the measured water
recovery, Rw and an entrainment factor (ENT). The entrainment
factor is calculated using an empirical correlation (Savassi et al,
1998). The recovery of particles to the launder compared to those
reporting to the pulp-froth interface is given by the froth
recovery, Rf. Thus, the overall recovery of particles from the
flotation cell, including entrainment and froth recovery may be
determined as shown below:
 R   R  k c TMRT
Rp = 100 1 − CellEnt   f  + RCellEnt (6)
 100  100  1 + k c TMRT
where:
RCellEnt = Rw (ENT) (7)
TMRT =
(
V 1− εg ) (8)
Q
where:
V is the cell effective volume
εg is the gas hold-up
Q is the volumetric flow rate of slurry in the cell feed
This paper focuses on two comprehensive surveys (Survey 1
and Survey 2) in the C3 roughers in which cell-by-cell and
size-by-size metallurgical, hydrodynamic, and pulp/surface
chemistry information were obtained. The two surveys were
conducted at significantly different feed tonnage giving rise to
significantly different feed particle size distributions and pulp
residence times. Survey 2 was used to calibrate the model to the
size-by-size flotation data, while Survey 1 was used to
corroborate the model parameter values. A series of simulations
are then performed with the aim of quantifying the change in
copper recovery, on both a size-by-size and unsized basis, with
changes in values of the hydrodynamic parameters and surface
properties of the copper sulfide mineral. This was carried out in
an effort to identify the most promising route to increase overall
copper recovery in roughing.
EXPERIMENTAL
Plant survey conditions
Two plant surveys were conducted at significantly different feed
tonnages and particle size distribution (Table 1). The C3
concentrator consists of three parallel rougher banks, consisting
of four stages per bank, and 3 × 85 m3 Wemco cells per stage.
The feed tonnage during the surveys in Table 1 is for a single
bank (A) of cells, the bank feed tonnage being calculated from
the C3 SAG mill feed rate, less transfer to C4, and assuming
equal splitting of the slurry between the three rougher banks.
Equal splitting was confirmed by tracer tests on the slurry splitter
system. For Survey 2 at the higher feed tonnage, the pulp density
of the rougher bank feed increased, due to reduced water
additions in grinding at the higher feed tonnage. The increased
pulp density in rougher feed is reflected also in the rougher
tailing pulp density, discussed further below. The increased pulp
density has consequences on coarse copper recovery as discussed
further below. Lime addition to the SAG and ball mill feeds
Ninth Mill Operators’ Conference
 A MODEL STUDY OF COPPER ROUGHER RECOVERY OPTIMISATION AT PT FREEPORT INDONESIA

maintained the pH at 9.5 in rougher feed for both surveys
(Table 1). The primary copper collector, Cytec 7249 (a mixture
of mono- and di-(isobutyl) thiophosphate collector), is added to
the ball mill feeds at an addition rate determined from the copper
units in the rougher feed. The secondary copper collector,
sodium isobutyl xanthate (SIBX), is stage added to each rougher
stage at an addition rate determined from the copper units in the
rougher tailing (Table 1). The alkyl group chain length (R = 4) of
the collectors controls the maximum receding contact angle
attainable on the copper sulfide mineral surface (Sutherland and
Wark, 1955), the consequences of which are discussed further
below. The frother is OTX 140 (Table 1).
Metallurgical sampling and analysis
The metallurgical surveys consisted of measuring the mass flow
rate of concentrate for individual cells and cell groupings. The
cell groupings are discussed further below. Further, samples of
the rougher bank feed, timed individual cell concentrates,
combined bank concentrate, intermediate stage tailings, and
rougher bank tailing (duplicate samples) were also collected. The
samples were weighed, filtered, dried and weighed again to
calculate the pulp per cent solids. Head samples were analysed
for copper, iron, sulfur, silver and gold. The metal assays were
converted to copper sulfide, pyrite and non-sulfide gangue (NSG)
assays assuming that all the copper was in a mixture of
chalcopyrite (90 per cent) and chalcocite (ten per cent) (based on
Mineral Liberation Analysis (MLA) of products), sulfur not
resident in copper sulfide was contained in pyrite (FeS2), and that
the remaining sample mass was as NSG.
Subsamples were also sized using a specific procedure (see
Table 2):
1. subsamples were wet screened at 38 µm;
2. the -38 µm fractions were passed through a 30.2 mm high
pressure (90 psi) pre-cyclone to produce an overflow
fraction -6.8 µm on copper sulfide (Fraction 8 in Table 2);
3. the -38 µm fraction was cyclosized, and the cyclosizer
discharge collected in a centrifuge to produce a fraction
+3.1 - 12.9 µm on copper sulfide (Fraction 7 in Table 2),
and the cyclosizer cone products were collected (Fractions
5 and 6 in Table 2); and
4. the +38 µm fraction was dry sieved at 300, 150, 75, and
38 µm.
The final size ranges on copper sulfide, are shown in Table 2,
as well as the distribution of copper sulfide in the feed to the
rougher bank for Surveys 1 and 2. In the analysis of the data,
some size fractions have been grouped as very coarse, coarse,
intermediate, fine and ultra-fine size fractions (Table 2). Size
distributions within the size fractions were confirmed by a light
scattering technique (Malvern Mastersizer) on separate samples
of chalcopyrite. Table 2 also shows the particle size point used
for plotting of the size-by-size data for copper sulfide and for the
model calculations. The particle size point is based on the
volume average of the lower and upper limits of the particle size
range (Table 2).
The program Banks for Excel was used to mass balance the
streams on a size-by-size basis, and to determine the flows of
minerals and water, recovery of minerals and water, pulp specific
gravity, residence time per cell, and the undistributed rate
constant, k, of each mineral, per size range, per cell grouping.
The individual flow rate of cell concentrates were uniformly
scaled to achieve agreement between the total flow rate of
concentrate and the concentrate flow rate determined from the
copper metal balance over the whole bank and the bank feed
tonnage.
Entrainment is controlled by water recovery (Rw) and the
entrainment factor (ENT). The water recovery values are
designated per flotation cell in the model and are the same values
measured in the plant surveys. The entrainment factor values are
calculated using parameters necessary in an empirical model

TABLE 1
Plant survey conditions.

Survey Rougher bank (A) feed Reagent additions to rougher (g/t)

Feed solids flow % Solids +220 µm Cu pH 7249 SIBX OTX 140
(t/h) (w/w) (%) (%)
1 701 29.7 15.8 1.38 9.5 46 15 17
2 889 40.8 20.1 1.20 9.5 65 15 19

TABLE 2
Particle size fractions.

Designation Size fraction Measured particle distribution of Particle size Distribution of copper sulfide
size fraction point (µm) mineral in feed (%) CLY
Lower end (µm) Upper end (µm) Survey 1 Survey 2 90

Very coarse 1 300 - 305 1.0 7.3 nd

Coarse 2 150 300 212 4.7 13.5 53
Intermediate 3 75 150 102 22.6 19.5 93
4 38 75 60 29.5 23.4 95
5 14.9 45.6 32 15.4 12.5 98
6 6.9 20.9 16.6 11.0 8.6 100
Fine 7 3.1 12.9 10.5 6.3 6.0 100
Ultra-fine 8 0.7 6.8 5.4 9.4 9.3 100

Notes: (1) size fractions 1 to 4 are screen fractions only, the lower and upper end of the size fractions refer to screen sizes, (2) size fractions 5 and 6 are
cyclosizer fractions, (3) size fraction 7 is the centrifuge product, (4) size fraction 8 is the pre-hydrocyclone product, (5) size fractions 4 to 8 lower and
upper end of size fraction refers to laser light diffraction d10 and d90 sizes of chalcopyrite, (6) particle size point refers to the particle size used for both
experimental and model recovery values, (7) CLY90 refers to the per cent of value mineral contained in particles containing greater than 90 per cent
chalcopyrite. The liberation data shown are not from the specific plant survey in this paper. nd = not determined.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 27

S R GRANO, T AKROYD and M A MULAR

(Savassi et al, 1998), (δ) the drainage parameter, and (ξ) the 2.0

Superficial Gas Velocity (cm/s) /

particle size at which the degree of entrainment is 20 per cent.
The value of ξ was taken to be 20 µm, implying that entrainment Bubble Diameter

Bubble Diameter (mm)

would be unimportant for particle sizes significantly greater than 1.6
this size. The principal focus is on the recovery of strongly
hydrophobic copper sulfide for which the contribution by
entrainment to the overall recovery would be expected to be only 1.2
very small. Hence, any errors related to entrainment are likely to
be only small. However, the importance of accurately determining
and modelling entrainment becomes greater for less hydrophobic 0.8
gangue mineral particles. The froth recovery of attached copper Superficial Gas Velocity
sulfide mineral particles (Rf) is assumed to be 100 per cent for
each cell as discussed previously (Table 3). 0.4

TABLE 3
Operating conditions of cells in the rougher bank for use in the 0
fundamental model. 0 1 2 3 4 5 6 7 8 9 10 11 12

Operating parameters Measured Assumed or

Cell (No)
value calculated value FIG 1 - Mean bubble diameter and superficial gas rate in each
Cell volume 85 m3 na rougher cell for Surveys 1 and 2.
Gas hold-up 11 - 17% na
Installed motor power per cell 150 kW na The gas hold-up was measured in each cell at the same
Energy dissipation nm 11 W kg-1 locations as the superficial gas velocity (Sanwani, 2003) (range
of measurements shown in Table 3).
Bubble diameter (average d32 1.3 - 1.7 mm na
across the cells) The energy dissipation was not directly measured, but
calculated according to the following procedure. The flotation
Superficial gas velocity 0.7 - 1.3 cm s-1 na
cells have 150 kW motors installed (Table 3). Assuming no loss
Bubble rise velocity nm 1 cm s-1 of power in transmission, the mean energy dissipation (ε) across
Apparent viscosity of slurry nm 0.2 Pa.s the entire cell volume is approximately 1.1 W/kg of slurry.
Froth recovery - 100%
However, it is very well known that the turbulent energy is not
dissipated uniformly across the cell volume, but is dissipated
Contact angle of liberated sulfide nm 78° principally in a region close to the impeller (Schubert, 1988).
mineral using isobutyl xanthate This is particularly important for large cells, which may have
collector
proportionally larger quiescent zones depending upon the slurry
Notes: na = not applicable, nm = not measured in this study. (1) Froth
rheology and tank dimensions (Schubert, 1988). In this current
recovery was measured in cell 2 at 95 per cent and in cell 12 at 50 per paper, a two-compartment model is adopted, which assumes that
cent, both on an unsized basis, (2) average bubble diameter and all energy dissipation occurs in a region close to the impellor,
superficial gas velocity for all 12 cells is shown. with a ‘dead volume’ in which no energy is dissipated. The key
parameter to determine is the proportion of the cell volume
across which energy dissipation occurs. It is necessary to
Hydrodynamic measurements calculate the mean shear rate, G, within the cell (Camp and
The key hydrodynamic parameters required by the flotation Stein, 1943):
model are:
G = (ε / ν)
1/ 2
(9)
• superficial gas rate,
• bubble diameter,
where:
• gas hold-up,
ν is the kinematic viscosity of the slurry, taken to be 1.4 ×
• energy dissipation, 10-4m 2 s-1 based on a slurry viscosity of 0.20 Pa.s and a
• bubble rise velocity, and rougher flotation feed slurry density of 1400 kg m-3 (Table 3)
• mixing regime. The calculated mean shear rate is approximately 90 s-1.
Referencing to a continuous shear rate distribution/cell volume
For superficial gas rate, the measurements were taken in each
curve for Rushton turbine cell geometry (Koh, 1984), suggests
rougher cell down bank A, at three different positions located at
that, at this mean shear rate, the ‘dead volume’ in the flotation
the same distance from the bottom of the cell (Sanwani, 2003),
cell is likely to be at least 90 per cent for the shielded impellors
using the JKMRC Jg probe (Power, Franzidis and Manlapig,
used in practice. In other words, the local energy dissipation in
2000). At least five measurements of superficial gas rate were
taken at each position and the average was calculated for the the impeller region is at least 11 W/kg. This value of energy
entire cell (Figure 1). The bubble diameter distribution was dissipation was used for each cell for both surveys. The
measured using a UCT bubble size analyser, details of which significance of energy dissipation on the flotation rate is
have been described elsewhere (Tucker et al, 1994). The bubble discussed further below.
diameter distribution was measured at one location (midway The bubble rise velocity was not directly measured, but
between impeller and the cell wall), one metre below the froth calculated according to the procedure discussed below. The
surface. The sauter mean bubble diameter was calculated from bubble (terminal) rise velocity (vb) is normally measured to be in
these measurements (Figure 1). The gas hold-up was measured the region of 15 to 18 cm s-1 for pure water (Sam, Gomez and
using a JKMRC gas hold-up probe, details of which have been Finch, 1996), which has low bulk and surface viscosity. However,
described elsewhere (Power, Franzidis and Manlapig, 2000). for bubbles traveling through a dense pulp in a large flotation cell

28 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 A MODEL STUDY OF COPPER ROUGHER RECOVERY OPTIMISATION AT PT FREEPORT INDONESIA

the bubble rise velocity is likely to be much lower due to the high
bulk viscosity of the slurry and the adsorption of surfactants and
contaminants. In this present study, the effect of bulk viscosity
only on the bubble rise velocity is taken into account. Assuming
that the bubbles behave as low density, hard spheres the bubble
rise velocity using the well-known drag coefficient method may
be used (Lapple and Sheperd, 1940). Here CdReb2 is calculated,
where Cd is the drag coefficient and Reb is the bubble Reynolds
number. The calculated CdReb2 is 5.43 and, using the standard
correlation, Rep may be determined as 0.21. Using the equation
for the bubble Reynolds number (Lapple and Sheperd, 1940), the
bubble rise velocity is calculated as approximately 1 cm s-1
(Table 3).
A residence time distribution test of the rougher bank was
performed using a saturated NaCl solution as a tracer. The tracer
test showed that the flotation cells in the rougher bank behaved
as perfect mixers and thus the recovery calculation shown in
Equation 4 could be used. While the measured residence time is
not a necessary input variable to the model (the residence time is
calculated based on the feed flow rate, the cell volume and the
gas hold-up), it does provide a useful check for some of the
model calculations. The measured residence time per cell (in
Stage 1) determined from the tracer test was 3.0 minutes, while
the calculated residence time for Surveys 1 and 2 was 2.6 and 3.2
minutes respectively in Cell 1 (Table 4).
Surface analysis
ToF-SIMS was used to estimate the receding contact angle on
copper sulfide mineral phases. The ToF-SIMS spectra were
obtained using a PHI Model 21000 TRIF II system equipped
with a gallium liquid metal ion gun in pulse mode. Imaging of
the sample involved mapping for positive and negative ions of
interest. Twenty-five copper sulfide particles were analysed in
the concentrate and tailing to obtain statistical information on the
amount of hydrophobic and hydrophilic species on the surface.
RESULTS
Metallurgical reconciliation
The increased feed tonnage in Survey 2 caused a decrease in
copper recovery, across all size fractions, of approximately
5.1 per cent (Table 4). In spite of the increased feed tonnage
in Survey 2, there was increased mean pulp residence time in
Survey 2 due to the reduced water additions in grinding, and
commensurate increased pulp density in the rougher feed and
tailing streams (Table 4). Consequently, increased residence time
in Survey 2 cannot be a factor to account for the copper recovery
decrease in this survey. There is very close agreement for both
surveys between the feed head copper grade and recalculated
head copper grade, based on the grades and flows of the
individual concentrates and rougher tailing streams (Table 4).
Hydrodynamic results
The mean bubble diameter was consistently between 1.3 to
1.7 mm across the rougher bank (Figure 1). The superficial gas
rate generally increased down the rougher bank from
approximately 0.7 cm/s to 1.2 cm/s. These values are in the
normal range for cells of this type (Power, Franzidis and
Manlapig, 2000). It is important to note that for the calculation of
the fundamental model, discussed further below, that the rate
constant and recovery was calculated for each cell separately,
using the bubble diameter and superficial gas rate separately
measured for each cell. It is also important to note that the
occurrence of slower floating components of copper sulfide
mineral in the latter stages of flotation (cells 7 to 12), as noted
further below, can not be explained by adverse changes in either
bubble diameter or superficial gas rate down the bank of rougher
cells.
Metallurgical results
The flotation results will be considered firstly on an unsized
basis (Figure 2). In increasing the feed tonnage in Survey 2, it is
clear that the proportion of slow floating components increased.
The reasons for the change in floatability are likely to be related
to one or a combination of changes in:
• feed particle size distribution,
• liberation class of the copper sulfide mineral in the size
fractions,
• gas dispersion down the bank of cells, and
• hydrophobicity of the copper sulfide mineral surface.
It is not the purpose here to quantify these differences, though
it is possible to do this with the aid of the fundamental flotation
model. The principal contribution comes from the change in feed
particle size distribution. The decrease in copper recovery with
the increased feed particle size (Table 4) is consistent with
historical correlations of copper rougher recovery with the per
cent +210 microns of rougher feed (Mular and Mosher, 2006).

TABLE 4
Unsized down-the-bank mass balances for Surveys 1 and 2.

Survey 1 Survey 2
Stream Cumulative pulp Solids flow Cu Solids Cumulative pulp Solids flow Cu Solids
mean residence (t/h) (%) (%) mean residence (t/h) (%) (%)
time (min) time (min)
Cell 1 2.6 13.3 29.9 50.7 3.1 15.0 28.6 55.1
Cell 2 5.1 11.1 25.6 37.4 6.3 8.9 24.6 48.9
Cell 3 7.7 9.6 16.0 27.9 9.5 7.5 14.5 30.8
Cell 4 - 6 15.6 31.3 2.0 13.2 19.3 41.3 3.6 17.3
Cell 7 - 9 24.6 12.9 0.9 9.9 30.5 18.9 2.2 15.4
Cell 10 - 12 34.2 21.0 0.4 7.6 42.7 3.6 4.8 14.8
Combined rougher conc 99.2 9.3 13.5 95.1 10.1 20.9
Rougher tail 602.0 0.08 34.9 794.0 0.13 45.4
Recalculated feed 701 1.38 28.5 889 1.20 40.4
Sampled feed 701 1.38 29.7 889 1.20 40.8
Copper recovery (%) 95.0 89.9

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 29

S R GRANO, T AKROYD and M A MULAR

100 Fundamental Model

100 50
Fit to Survey 1
Survey 1
Cell 9 Cell 12 80 40

% in Size Range
80 Cell 6

% Passing
Fundamental Model Survey 2
Recovery (%)

Cell 3 Fit to Survey 2

60
Cell 2
40 Cell 1
20
0 Particle Size ( m)
0 5 10 15 20 25 30
Residence Time (min)
FIG 2 - Cumulative copper recovery, on an unsized basis, as a
function of cumulative mean pulp residence time, for Surveys 1
and 2. The continuous lines are for the fundamental model also
on an unsized basis, determined by addition of the individual
size fraction model recovery values.
Attention will now turn to the flotation data on a size-by-size
basis. In Survey 1 the per cent +220 micron in rougher feed was
16 per cent, a normal value, which increased to 20 per cent in
Survey 2. Increasing the feed tonnage in Survey 2 increased the
proportion of copper in the rougher feed in the coarse and very
coarse size fractions (Table 2, Figure 3). It is these size fractions
that show the lowest recovery, as discussed further below, and
hence there is lower copper recovery, across all size fractions, in
Survey 2. The increased proportion of copper in the coarse and
very coarse size fractions in Survey 2 reduced the proportion of
copper in the intermediate size fractions of the feed, while the
proportion of copper in the fine and ultra-fine size fractions
remained approximately the same in the rougher feed (Table 2,
Figure 3).
As expected, and predicted by the fundamental model
described further below, the very coarse, coarse and fine size
fractions show the lowest recovery for both surveys (Figure 4).
It is important to note that the coarse and very coarse size
fractions show higher recovery in Survey 2 at the same mean
pulp residence time values. The reason for the increase in
recovery for these coarse size fractions is related to one or
combination of:
• an increase in the liberation class of the particle size fractions
at the coarse grind, and/or
• an increase in stability efficiency of the coarse particles due
to the increased pulp density of the feed.
60 30
Survey 1
40 20
Survey 2
20 10
Survey 1
0
Survey 2 0
1 10 100 1000
35 40 45
Survey 1 Cu Distribution
Survey 2 Cu Distribution
Survey 1 Solids Distribution
Survey 2 Solids Distribution
FIG 3 - Solids and copper metal distribution in rougher flotation
feed for Surveys 1 and 2. Solids distribution shown as cumulative
per cent passing, and copper distribution shown as per cent
copper in particle size range.
It is not the purpose here to quantify these differences, though
it is possible to do this with the aid of the fundamental flotation
model. At the high pulp density of Survey 2, the increased pulp
viscosity dampens energy dissipation in the cells increasing the
stability efficiency of coarse composite particles. It is also
important to note that the fine and intermediate size fractions
exhibit lower recovery at the same mean pulp residence times in
Survey 2 conducted at the higher pulp density. In the case of
these size fractions, the collision efficiency and collision
frequency together principally control their flotation rate. At the
high pulp density, the increased pulp viscosity decreases the
collision frequency.
A more important point is that there exists at least two
floatability components of copper in the coarse and very coarse
size fractions as demonstrated by an empirical, floatability
component model fit to Survey 1 and 2 (parameter values shown
in Table 5, continuous lines in Figure 4). It is clear that the
proportion of slow floating components increases with increasing
particle size, and is likely related to a decreasing liberation class
of the copper sulfide mineral (Table 2). For the coarse size
fraction, the CLY90 (defined in Table 2) has previously been
measured to be only 53 per cent while for size fractions below
150 µm, CLY90 is greater than 93 per cent (Table 2). This has
important implications to the modelling of this data, described

TABLE 5
Parameter values for floatability component model of size by size data in Figure 4.

Survey 1 Survey 2
Size range (µm) +300 -300 -150 -75 -12.9 +300 -300 -150 -75 -12.9
+150 +75 +12.9 +150 +75 +12.9
kf (min-1) 0.02 0.2 0.4 0.45 0.19 0.2 0.2 0.25 0.25 0.15
ks (min-1) 0.01 0.05 0.15 0.15 0.11 0.03 0.07 0.15 0.2 0.07
% Fast 0 60 82 85 75 5 60 85 90 70

Note: Per cent fast is percentage of copper sulfide with rate constant, kf.

30 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 A MODEL STUDY OF COPPER ROUGHER RECOVERY OPTIMISATION AT PT FREEPORT INDONESIA
Residence Time (min)
0 5 10 15 20 25 30 35 40 45
1.00
Fraction Remaining (log scale)
0.10
(A) Survey 1
1.00
0.10
(B) Survey 2
0.01
+300 m
--75 +12.9 m
- 300 +150 m
--12.9 m
-150 +75 m
FIG 4 - Fraction remaining (log scale) as a function of cumulative
mean pulp residence time for (A) Survey 1 and (B) Survey 2, and
for various size fractions +300, +150 - 300, +75 - 150, +12.9 - 75,
-12.9 microns. The continuous lines are for a floatability
component model with two components with parameters shown
in Table 5.
further below, using the fundamental model. At this stage, it can
be stated that the copper sulfide minerals in size fractions below
150 µm should behave uniformly with respect to particle contact
angle, assuming that mineral locking only contributes to changes
in particle contact angle with particle size. For size fractions
greater than 150 µm, reduced liberation of the copper sulfide
mineral in composite particles will reduce the overall contact
angle of the copper sulfide bearing particle. Changes in recovery
between the surveys as noted in Figure 4 are also apparent in the
floatability class and rate constant values in Table 5. In
increasing the pulp density from Survey 1 to Survey 2, there is an
increase in the rate of flotation of the coarse and very coarse size
fractions, and a decrease in the rate of flotation of the
intermediate and fine size fractions.
The recovery of the copper sulfide with respect to particle
size exhibits typical behaviour, with its apparent flotation rate
decreasing for particle sizes below and above approximately
60 µm (Figure 5a), the particle size point for the fourth size
fraction defined in Table 2. In practice, and in general, the
decreasing flotation rate of a hydrophobic value mineral with
decreasing particle size below 60 µm does not become a problem
for recovery until the particle size falls below ~12 µm, ie for fine
and ultra-fine particles defined in Table 2. This is because there
is usually sufficient residence time, under the prevailing
hydrodynamic conditions, to recover hydrophobic particles in the
size range of +12 - 75 µm. There is greater than 95 per cent
recovery for this size range at six cells (19.3 minutes in Survey
2), one half the installed flotation capacity (Figure 2). Even for
copper sulfide mineral below 12 µm, their reduced flotation rate
does not greatly impact upon their recovery at 12 cells (42.7
Ninth Mill Operators’ Conference
100
Cell 12
(A)
80
Cumulative Copper Sulfide Recovery (%)
Cell 6
60
Cell 3
40
Cell 2
20
Cell 1
0
Cell 12
(B)
80
Cell 6
60
Cell 3
40
Cell 2
20
Cell 1
0
1 10 100 1000
Particle Size ( m)
FIG 5 - Copper mineral recovery as a function of particle size and
cumulative number of cells for Survey 2. The continuous lines are
for the fundamental model fit for (A) copper sulfide bearing
particles having a contact angle of 78° in all particle size fractions
and (B) copper sulfide bearing particles having a variable contact
angle in each size fraction as shown in Table 7.
minutes in Survey 2) (Figure 2). However, increasing their
flotation rate may have a small impact upon their recovery as
discussed further below.
In the case of particles greater than 60 µm, there is evidence
for both a reducing flotation rate and the occurrence of a
non-floating component, particularly for the very coarse size
fraction (Figure 4 and 5). The occurrence of a non-floating
component of copper sulfide mineral was confirmed in
laboratory batch flotation tests on the rougher tailing during the
survey period, and is due to locking of the copper sulfide mineral
within hydrophilic gangue mineral phases in both the coarse and
very coarse size fractions. The significance of copper sulfide
mineral locking is commented upon further below, but it does
highlight the importance of precisely defining the size-by-
liberation characteristics of the feed, a point that is discussed
further below. This particular flotation plant shows an apparent
high rate of recovery of the copper sulfide mineral. To gain more
confidence in these results, the flotation rate was benchmarked
against flotation rates measured in other large flotation plants,
treating mainly chalcopyrite and with a similar collector type
(Table 6). Table 6 shows that the flotation rate of the copper
sulfide mineral encountered in this study is not unusually high.
Surface analysis to estimate copper sulfide
mineral contact angle
ToF-SIMS measurements suggested a very hydrophobic surface
for fully liberated particles of the copper sulfide mineral with an
adsorbed collector/SO3- ratio greater than 0.01, and therefore an
inferred contact angle of 70° or greater (Duan, Fornasiero and
Ralston, 2003). The contact angle of the value mineral was taken
Fremantle, WA, 19 - 21 March 2007 31
S R GRANO, T AKROYD and M A MULAR
TABLE 6
Flotation rate constant, recovery and cell residence time for the first grouping of cells in plants treating a similar copper sulfide
mineral, on an unsized basis.
Plant Copper sulfide mineral
Rate constant (min-1) Recovery across cell
group (%)
PTFI 0.21 40.0
Plant A 0.24 74.0
Plant B 0.17 47.8
Plant C 0.23 51.4
to be 78° (Table 3), a value characteristic of the particular
collector hydrocarbon chain length (R = 4) used in the plant
(Sutherland and Wark, 1955). For contact angles greater than
70°, both the attachment (Ea) and stability (Es) efficiencies are
very close to unity for particles sizes less than 75 µm, indicating
that the flotation rate will be limited by the collision rate for this
particle size range, rather than particle hydrophobicity, for very
high contact angles. The significance of this statement is
discussed further below. Other studies have also shown that
chalcopyrite has a limiting contact angle close to 78° at high
adsorption densities of collector of the same chain length as used
in this present study (Walker, Walters and Richardson, 1986).
Model fit to Survey 2
Attention will now turn to the fitting of the fundamental model,
outlined in the introduction, to the flotation data of Survey 2.
Specifically, the collision efficiency, Ec, is calculated from the
generalised Sutherland Equation 20 in Pyke, Fornasiero and
Ralston (2003). The attachment efficiency, Ea, is calculated using
Equations 22 - 26 in Pyke, Fornasiero and Ralston (2003). The
particle-bubble stability efficiency, Es, is calculated using
Equations 28 - 34 in Pyke, Fornasiero and Ralston (2003). The
bubble-particle collision frequency is calculated using Equation
26 in Bloom and Heindel (2003), which takes account of body
forces and where the bubble and particle velocity distribution
variances are calculated using Equations 17a and 17b in Bloom
and Heindel (2003).
The calculated recovery values of the fundamental flotation
model are shown as continuous lines in Figure 5, assuming that
the copper sulfide bearing particles have a contact angle of 78°
across all particle size fractions. It is clear that for particles
greater than 75 µm the model overestimates their recovery. This
is certainly related to the reduced level of liberation of copper
sulfide mineral in these coarser size fractions, as previously
discussed. Inferred liberation levels are discussed further below
which adequately model the flotation data for particle sizes
greater than 75 µm.
For particles sizes less than 75 µm very good agreement
between the model and experimental data is apparent. It was
found in the initial studies that using the mean energy dissipation
across the cell (1.1 W/kg) and a bubble rise velocity of 15 to
18 cm/s gave calculated copper sulfide mineral recoveries, which
were much less than the experimental values at the same cell
number. As previously noted, the copper sulfide minerals over
this size range are strongly hydrophobic, and calculations of
attachment (Ea) and stability efficiencies show that both these
parameters are very close to unity over this particle size range
(-75 µm). Therefore, the discrepancy in recovery between the
model calculation using these parameter values, and experiment
could only be accounted by the collision frequency. Key
parameters that determine the collision frequency are energy
dissipation and bubble rise velocity. It was only when the
calculated local energy dissipation (11 W/kg) and bubble rise
velocities (1 cm/s) were used that a close agreement was found
32 Fremantle, WA, 19 - 21 March 2007
Number of cells in Volume per cell (m3) Total pulp residence
group time in cell group
(min)
1 85.0 3.2
2 100.0 8.0
3 38.0 4.2
4 11.3 3.5
for these particle size ranges (-75 µm). In terms of the physical
meaning, it is reasonably suggested that, by the action of the
impellor, all particles are directed through the impeller region
where particle collection by bubbles takes place at a rate
characteristic of this local energy dissipation. For the strongly
hydrophobic (and liberated) copper sulfide mineral particles, less
than 75 µm, the particles apparently remain attached to the
bubbles (Es = 1) and are transported to the top of the flotation
cell. It is important to note that the flotation rate and recovery of
fine copper sulfide (-12 µm) is limited by collision efficiency,
which is controlled through the cell hydrodynamics, rather than
surface hydrophobicity (Ea = Es = 1).
Attention will now turn to copper sulfide mineral contained in
particles greater than 75 µm. In the case of copper sulfide
mineral-bearing particles in the coarse (+150 - 300 µm) and very
coarse (+300 µm) size fractions, assuming a single class of
mineral phase (ie a single contact angle of 78°) did not allow
close agreement of the model with the experimental data
(Figure 5a). Specifically, the model significantly overestimated
the recovery of copper sulfide mineral in these particular size
fractions. To this point, the model did not allow the occurrence of
a non-floating component of copper sulfide mineral. It is
reasonably assumed that this is related to the occurrence of
copper sulfide mineral in composite particles in these coarser
size fractions (Table 2). An approach to adequately model this
flotation behaviour was to assume a distribution of copper sulfide
mineral liberation values in each size fraction of the rougher
flotation feed. Three different liberation classes were chosen:
Class 1 and 2 were both particles containing partially or fully
liberated copper sulfide mineral, and Class 3 were hydrophilic
particles containing completely unliberated copper sulfide
mineral (contact angle equals zero) for which the attachment
efficiency (Ea) is zero (Table 7). Class 3 gives rise to a
non-floating component of copper sulfide. The contact angles of
the composite particles were calculated using the Cassie equation
(Cassie and Baxter, 1944), assuming the contact angle of the
copper sulfide and gangue mineral phases were 78° and 0°,
respectively, in all cases (Table 7).
θ c = cos−1 (f1 cosθ 1 = f2 cosθ 2 + ....+ fn cos θ n ) (10)
where:
f is the area fraction of copper sulfide mineral in the
composite particle
c is composite
1,2,...n represent the mineral components of the particle
The percentage of partially liberated and completely
unliberated copper sulfide mineral increases with increasing
particle size above 75 µm (Table 7). For these coarse unliberated
particles, detachment from bubbles (Es decreases) considerably
reduces their flotation rate due to both their large particle size
and low overall hydrophobicity. Their flotation rate and recovery
Ninth Mill Operators’ Conference
 A MODEL STUDY OF COPPER ROUGHER RECOVERY OPTIMISATION AT PT FREEPORT INDONESIA

is lower due to reduced bubble-particle aggregate stability across
the flotation cell volume. The presence of composite particles
was important only for the coarse and very coarse size fractions
(Table 7). Incorporating different levels of liberation allowed a
very close fit of the model to the experimental data (Figure 5b).
The very close agreement between the experimental data and the
model is also apparent on an unsized basis (Figure 2, Survey 2).
in the feed. There is close agreement between the fundamental
model and Survey 1 on an unsized basis for most cells (Figure 2,
Survey 1). A possible reason for the discrepancy in cells 2 and 3
may be related to the lower pulp density in Survey 1, decreasing
the pulp viscosity, and increasing the bubble particle collision
frequency and rate constant for copper sulfide particles less than
75 µm.

Model fit to Survey 1 Optimising copper recovery on a size basis using

Survey 1 was completed under different feed conditions to
demonstrate that the same model parameters used for Survey 2
may be used to predict the copper sulfide mineral recovery down
the rougher bank of cells. In the approach, the conditions of
Survey 1 were successively applied to the simulation of Survey 2
to quantify contributions to differences in copper recovery due to
differences in:
• feed tonnage and per cent solids as they affect mean pulp
residence time,
• copper sulfide mineral size distribution in the feed, and
• copper sulfide mineral liberation class distribution.
The most important factor, which contributed to differences in
copper recovery, was the copper sulfide mineral size distribution
simulation
A series of simulations were performed to investigate the effect
of process variables on the size-by-size (Figures 6 - 8) and
unsized (Table 8) copper sulfide mineral recovery down the
rougher bank of cells. The process variables considered were:
• energy dissipation (Cases 1 and 2 in Table 8),
• bubble diameter (Cases 3 and 4 in Table 8),
• contact angle of the copper sulfide mineral surface (Cases 5
and 6 in Table 8),
• additional 85 m3 flotation cells (Cases 7 and 8 in Table 8),
and
• combinations of each (Cases 9 and 10 in Table 8).

TABLE 7
Inferred liberation distribution of the copper sulfide mineral in the rougher feed (Survey 2).

Size fraction Liberation class

(1) Partially liberated copper sulfide mineral (2) Partially liberated copper sulfide mineral (3) Unliberated copper sulfide
mineral
Wt % % Lib θc (°) Wt % % Lib θc (°) Wt % θc (°)
1 30 33.2 41 25 27.7 38 45 0
2 75 17.5 30 20 14.6 27 5 0
3 83 13.7 26 15 10 22 2 0
4 98.5 100 78 0 - - 1.5 0
5 99 100 78 0 - - 1 0
6 99 100 78 0 - - 1 0
7 100 100 78 0 - - 0 -
8 100 100 78 0 - - 0 -

Notes: (1) Wt per cent is the percentage of copper sulfide mineral in that liberation class in each size fraction, (2) per cent liberation is area per cent of value
mineral appearing at the surface of the composite particle, (3) θc is the calculated composite contact angle for the liberation class using the Cassie equation.

TABLE 8
Simulations of changes in plant operating conditions and the effect on the recovery of the copper sulfide on an unsized
basis across the rougher bank (12 cells).

Conditions Copper sulfide

Case Energy dissipation Bubble diameter Copper sulfide mineral Additional mineral recovery
(W/kg) (mm) contact angle (°) 85 m3 cells (%)

Base case (Survey 2) 11.0 1.3 78 0 89.9

1 13.8 1.3 78 0 84.8
2 8.8 1.3 78 0 92.9
3 11.0 1.6 78 0 92.0
4 11.0 1.0 78 0 87.4
5 11.0 1.3 85 0 93.4
6 11.0 1.3 90 0 94.2
7 11.0 1.3 85 1 90.3
8 11.0 1.3 85 3 90.7
9 8.8 1.6 85 0 93.4
10 8.8 1.0 85 0 94.0

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 33

S R GRANO, T AKROYD and M A MULAR

100

80
Base Case Energy Dissipation = 11 W/kg
60 Case 1 - Energy Dissipation = 13.8 W/kg

Recovery (%) 40 Case 2 - Energy Dissipation = 8.8 W/kg

20
(A) Energy Dissipation
0

80
Base Case Bubble Diameter = 1.3 mm
60
Case 3 - Bubble Diameter = 1.63 mm
40 Case 4 - Bubble Diameter = 1.0 mm

20
(B) Bubble Diameter
0
1 10 100 1000
Particle Size ( m)
FIG 6 - Simulated copper recovery as a function of particle size for different (A) energy dissipation values of 11 W/kg, 13.8 W/kg and
8.8 W/kg, and (B) bubble diameters of 1.3 mm, 1.63 mm and 1.0 mm. The base case is for fundamental model fit to Survey 2 using
Table 3 parameter values.

100

80
Base Case Cu-S CA = 78o
60
Case 5 - Cu-S CA = 85 o

40 Case 6 - Cu-S CA = 90o

Recovery (%)

20
(A) Copper Sulfide Contact Angle
0

80

60 Base Case No Additional Cells

40 Case 7 - 1 Additional Cell

Case 8 - 3 Additional Cells

20
(B) Additional 85 m3 Cells
0
1 10 100 1000
Particle Size ( m)
FIG 7 - Simulated copper recovery as a function of particle size for different (A) copper sulfide (Cu-S) mineral surface contact angle (CA)
values of 78°, 85° and 90° and for, (B) one and three additional 85 m3 cells. The base case is for fundamental model fit to Survey 2 using
Table 3 parameter values.

For Cases 1 to 6, the relevant hydrodynamic parameter was
adjusted in each cell down the bank. For Cases 7 and 8, the same
hydrodynamic values were used in the additional cells as in the
final cells of the existing rougher bank. It is noted that changes in
recovery as discussed below are based on the model calculations.
For the PTFI system, it is most beneficial to affect positive
changes in coarse copper sulfide mineral recovery, over fine
value mineral recovery, due to the preponderance of coarse
copper sulfide bearing particles in the feed (Table 2, Figure 3).
As an example, decreasing the energy dissipation in Case 2
increases coarse and very coarse sulfide mineral recovery, while
only reducing ultra-fine copper sulfide mineral recovery to a
small extent (Figure 6a). On an unsized basis, decreasing energy
dissipation in Case 2 increases copper sulfide mineral recovery
by three per cent (Table 8). It is noted that decreasing the energy
dissipation will have a positive effect on coarse copper sulfide
mineral recovery by increasing the stability efficiency, while fine
copper sulfide mineral recovery is reduced due to the decrease in
collision frequency, these particles’ flotation rate being principally
controlled by the collision efficiency and bubble-particle collision
frequency. Similar observations are made in relation to the effect
of changes in bubble size, where an increase in bubble size in
Case 3 increased coarse copper sulfide mineral recovery at the
expense of ultra-fine copper sulfide mineral recovery (Figure 6b).

34 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 A MODEL STUDY OF COPPER ROUGHER RECOVERY OPTIMISATION AT PT FREEPORT INDONESIA
100
Recovery (%) 80
60
40
20
0
1 10 100 1000
Particle Size ( m)
FIG 8 - Simulated copper recovery as a function of particle size for different combinations of energy dissipation, bubble diameter and copper
sulfide mineral surface contact angle values. The base case is for fundamental model fit to Survey 2 using Table 3 parameter values.
CA = contact angle.
In this case, increasing the bubble size increases the stability
efficiency of coarse particle size fractions, but decreases the
collision efficiency of fine particle size fractions.
Increasing the collector chain length also benefits coarse and
very coarse value mineral recovery (Figure 7a). Increasing the
hydrophobicity of the copper sulfide mineral surface will not
increase the recovery of copper sulfide mineral contained within
completely unliberated classes (Table 7). However, it will
increase the recovery of copper sulfide mineral in partially
liberated composites. Increasing the copper sulfide mineral
surface contact angle to 85° in Case 5, increases the copper
sulfide mineral recovery by 3.5 per cent on an unsized basis
(Table 8), an increase of similar magnitude to a decrease in
energy dissipation in Case 2, or an increase in bubble size in
Case 3, alone. It is noted that the change in contact angle of the
copper sulfide mineral surface from 78° to 85° is characteristic of
an increase in collector hydrocarbon chain length (to R = 5)
(Sutherland and Wark, 1955). Increasing the contact angle on the
exposed copper sulfide mineral increases the contact angle on the
composite particle overall, and increases the stability efficiency
of coarse composite particles. Increasing the contact angle on
fine copper sulfide has little effect on their recovery, as the
attachment efficiency is already at its maximum value. It is also
worthwhile pointing out that, for the PTFI system, simply
increasing the available residence time by increasing the number
of cells, does not appreciably increase copper recovery on either
a sized (Figure 7b) or an unsized basis (Table 8). Changes in the
operating conditions of the cells or the hydrophobicity of the
copper sulfide mineral need to be pursued to increase copper
recovery.
Increasing the contact angle of the copper sulfide mineral
allows also the bubble size to be reduced in order to pursue
increased ultra-fine copper sulfide mineral recovery. The most
significant, single, process variable to increase ultra-fine copper
sulfide mineral recovery is to reduce the bubble size in Case 4, as
previously noted (Figure 6b). Neither increasing the energy
dissipation nor increasing the copper sulfide mineral contact
angle alone is as effective. However, decreasing the bubble size
alone decreases coarse and very coarse copper sulfide mineral
recovery due to reduced stability efficiency (Figure 6b). By
increasing the contact angle of the copper sulfide mineral, it
becomes possible to target decreases in bubble size to affect
increases in ultra-fine copper sulfide mineral recovery, without
adversely affecting coarse copper sulfide mineral recovery (Case
10, Figure 8). Case 10 gave the highest increase in copper sulfide
mineral recovery of approximately four per cent on an unsized
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
Base Case
o
Case 9 - CA = 85 , Energy Dissipation = 8.8 W/kg
Bubble Diameter = 1.63 mm
o
Case 10 - CA = 85 , Energy Dissipation = 8.8 W/kg
Bubble Diameter = 1.0 mm
basis (Table 8). Increasing the contact angle of the copper sulfide
increases the contact angle on the composite particles overall and
reduces their tendency to detach from bubbles. Under these
modified conditions it is possible to reduce the bubble size,
which otherwise would have decreased the bubble-particle
stability efficiency.
DISCUSSION
The use of a fundamentally derived flotation model allows a
comprehensive view of the rougher bank performance on a
size-by-size basis. The model lends itself for benchmarking plant
flotation data. For example, reasons for the low flotation rate of
particles may be discerned (eg particle surface hydrophobicity
versus cell hydrodynamics). The relative contributions to a low
flotation rate may be quantified, and the effect of common
variables investigated to determine the most profitable avenue to
increase flotation rate. In the case of PTFI, the principal form of
copper loss is as copper sulfide mineral partially locked in
composite particles contained within coarse and very coarse size
fractions of the rougher tailing. These particles have apparently
only low exposure of the copper sulfide mineral at the particle
surface and hence the hydrophobicity of the overall composite
particle is low. For example, the contact angle for coarse (+150 -
300 µm) copper sulfide bearing composites is only 30°,
compared with the contact angle of the exposed copper sulfide
mineral (78°) (Table 7). The low contact angle adversely affects
their flotation rate and recovery due to detachment of the particle
from the bubble in the turbulent environment, giving rise to low
stability efficiency.
The simulations suggest that it is desirable to firstly target
increases in copper sulfide mineral hydrophobicity by increasing
the collector chain length, and following this step, increase fine
copper sulfide mineral recovery by reducing the bubble diameter.
The first step will increase coarse composite recovery, which is
of prime importance to overall copper recovery. The effect of
copper sulfide mineral exposure (ie mineral liberation class) in
the composite particles on their recovery is shown in Figure 9.
In the case of the coarse size fraction, there is virtually zero
recovery of composite particles with a copper sulfide mineral
exposure less than ~15 per cent, for 78° contact angle of the
copper sulfide mineral (Figure 9). The critical surface exposure
increases to ~30 per cent for the very coarse size fraction, also at
78° contact angle of the copper sulfide mineral (Figure 9).
Increasing the copper sulfide mineral contact angle from 78° to
85°, allows composite particles, which contain only ~10 per cent
35
S R GRANO, T AKROYD and M A MULAR

and ~20 per cent copper sulfide mineral exposure to be recovered
in the coarse and very coarse size fractions, respectively. This is
because there is a small increase in the contact angle of the
composite particle overall (Figure 9), and an increase in their
stability efficiency.
Reducing the bubble diameter will increase fine copper sulfide
recovery by increasing the collision efficiency, and is the most
effective method in doing so as previously noted (Figure 6b). It
should also be stressed that adding more cells using the existing
cell hydrodynamic conditions is not an effective means to
increase overall recovery. One method to decrease the bubble
diameter is to increase the frother concentration to the critical
coalescence concentration (Grau, Laskowksi and Heiskanen,
2005). The effect of bubble diameter on model calculations of
the recovery of ultra-fine (-6.8 µm) and coarse (+150 - 300 µm)
composite (0_30 per cent liberation class) copper sulfide bearing
particles is shown in Figure 10. Decreasing the bubble diameter
adversely affects coarse composite recovery at a copper sulfide
mineral contact angle of 78° (Figure 10), due to reduced stability
efficiency, while it increases ultra-fine copper recovery due to
increased collision efficiency. By increasing the copper sulfide
Recovery (%) / Contact Angle (o)
mineral contact angle to 85°, decreases in bubble diameter can be
sought to increase ultra-fine copper recovery without adversely
affecting coarse composite recovery (Figure 10). Increasing the
contact angle of the copper sulfide mineral has little affect on
ultra-fine copper recovery, which is principally controlled by cell
hydrodynamics through the collision efficiency and collision
frequency terms (Figure 10).
CONCLUSIONS
To increase overall copper recovery in roughing at PTFI, at the
current flotation feed size, it is necessary to address copper
recovery in coarse, low-grade composite particles. The most
effective means to increase coarse composite particle recovery is
to increase the contact angle of the exposed copper sulfide
mineral in the composite. Surface analysis suggests that the
contact angle of the copper sulfide mineral is at its maximum
value characteristic of the alkyl chain length of the collector
used. Therefore, it is only possible to increase the contact angle
by increasing the chain length of the collector, increasing the
contact angle of the composite particle overall. A further benefit

100
Composite Contact Angle
Cu-S CA = 85 o
80
o
+150 -300 m Cu-S CA = 78
60 +150 -300 m Cu-S CA = 85
o

40 +300 m Cu-S CA = 78o

Composite Contact Angle
Cu-S CA = 78o +300 m Cu-S CA = 85o
20
Survey 2 Condition

0
0 20 40 60 80 100
Copper Sulfide Mineral in Composite (%)
FIG 9 - Simulated copper recovery in coarse (+150 - 300 µm) and very coarse (+300 µm) copper sulfide (Cu-S) bearing particles as a
function of area fraction of copper sulfide mineral exposed at the composite particle surface. Recovery is at cell 12. Copper sulfide mineral
contact angles of 78° and 85° for coarse size fraction; 78° and 85° for very coarse size fraction. The contact angle of the composite particle
as a function of area fraction of copper sulfide mineral exposed at the composite particle surface also shown, calculated using Equation 10.
Simulations based on Survey 2 conditions. CA = contact angle. Cu-S = copper sulfide.

100

80
Recovery (%)

+150 -300 m 0_30% Lib Class- Cu- S CA = 78o

Survey 2
60
Condition +150 -300 m 0_30% Lib Class- Cu- S CA = 85o
o
-6.8 m 100% Lib Class Cu-S CA = 78
40
o
-6.8 m 100% Lib Class Cu- S CA = 78

20

0
0.8 1 1.2 1.4 1.6
Bubble Diameter (mm)
FIG 10 - Simulated copper recovery in coarse (+150 - 300 µm) copper sulfide (Cu-S) bearing particles (0_30 per cent liberation class) for
copper sulfide mineral contact angles of 78° and 85° as a function of bubble diameter. Recovery is at cell 12. Simulated copper recovery of
ultra-fine (-6.8 µm) copper sulfide bearing particles (100 per cent liberation class) for copper sulfide mineral contact angles of 78° and 85°
as a function of bubble diameter. Simulations based on Survey 2 conditions. CA = contact angle. Cu-S = copper sulfide.

36 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 A MODEL STUDY OF COPPER ROUGHER RECOVERY OPTIMISATION AT PT FREEPORT INDONESIA
of increasing the contact angle on the coarse composite particles
is that hydrodynamic conditions can be pursued which may
target increased fine copper recovery, but which would normally
compromise coarse copper recovery. The most effective means to
increase fine copper recovery is to decrease the bubble size. By
increasing the contact angle on the copper sulfide mineral alone,
the overall copper recovery can be increased from 89.9 per cent
to 93.4 per cent at the coarse grind (20 per cent +220 µm). By
increasing both the contact angle on the copper sulfide mineral
and reducing the bubble size the overall copper recovery can be
increased from 89.9 per cent to 94.0 per cent at the coarse grind.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge the sponsors of the P260D
(Mineral Flotation) project and specifically PT Freeport
Indonesia for permission to publish this paper. The authors
would also like to thank Ari Partanen and Arif Priyambodo for
their contribution and support. The authors would also like to
thank Edy Sanwani and the P9M project for the gas dispersion
data, William Skinner for the ToF-SIMS data, and Sergio Vianna,
George Tsatouhas and Tsvetomir Georgiev for their assistance on
site.
REFERENCES
Abrahamson, J, 1975. Collision rates of small particles in vigorously
turbulent fluid, Chemical Engineering Science, 30:1371-1379.
Bloom, F and Heindel, T J, 2002. On the structure of collision and
detachment frequencies in flotation models, Chemical Engineering
Science, 57:2467-2473.
Camp, T R and Stein, P C, 1943. Velocity gradients and internal work in
fluid motion, J Boston Soc Civ Eng, 30:219-237.
Cassie, A B D and Baxter, S, 1944. Wettability of porous surfaces, Trans
Faraday Soc, 40:547-551.
Dai, Z, Fornasiero, D and Ralston, J, 1999. Particle-bubble attachment in
mineral flotation, Journal of Colloid and Interface Science, 217(1):
70-76.
Dai, Z, Fornasiero, D and Ralston, J, 2000. Particle-bubble collision
models – A review, Advances in Colloid and Interface Science,
85(2-3):231-256.
Dobby, G S and Finch, J A, 1987. Particle size dependence in flotation
derived from a fundamental model of the capture process,
International Journal of Mineral Processing, 21(3-4):241-260.
Duan, J, Fornasiero, D and Ralston, J, 2003. Calculation of the flotation
rate constant of chalcopyrite particles in an ore, International
Journal of Mineral Processing, 72(1-4):227-237.
Grano, S, Weedon, D and Akroyd, T, 2005. Freeport plant surveys – C3
concentrator sized data, IWRI confidential report to PT Freeport
Indonesia, March.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
Grau, R A, Laskowski, J S and Heiskanen, K, 2005. Effect of frothers on
bubble size, International Journal of Mineral Processing, 76(4):
225-233.
Koh, P T L, 1984. Compartmental modelling of stirred tank for
flocculation requiring a minimum critical shear rate, Chem Eng Sci,
39:1759-1764.
Lapple, C E and Shepherd, C B, 1940. Calculation of particle trajectories,
Ind Eng Chem, 32:605-617.
Mular, M and Mosher, J, 2006. A pre-production review of PT Freeport
Indonesia’s high pressure grinding roll project, in Proceedings
SAG2006 (Canadian Institute of Mining and Metallurgy: Montreal).
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).
Pyke, B, Fornasiero, D and Ralston, J, 2003. Bubble particle
heterocoagulation under turbulent conditions, Journal of Colloid
Interface Science, 265:141-151.
Sam, A, Gomez, C O and Finch, J A, 1996. Axial velocity profiles of
single bubbles in water/frother solutions, International Journal of
Mineral Processing, 47(3-4):177-196.
Sanwani, E, 2003. Gas dispersion and hydrodynamic characteristics of
selected Wemco 85 m3 flotation cells at PT Freeport Indonesia Mine,
AMIRA P9M project report, JKMRC, Brisbane.
Savassi, O N, Alexander, D J, Manlapig, E V and Franzidis, J-P, 1998. An
empirical model for entrainment in industrial flotation plants,
Minerals Engineering, 11:243-256.
Schubert, H, 1989. The role of turbulence in mineral processing unit
operations, in Challenges in Mineral Processing (eds: K Sastry and
M C Fuerstenau), pp 272-289 (Society for Mining, Metallurgy and
Exploration: Littleton).
Schulze, H J, 1977. New theoretical and experimental investigations on
stability of bubble/particle aggregates in flotation: A theory on the
upper particle size of floatability, International Journal of Mineral
Processing, 4:241-259.
Schulze, H J, 1993. Coagulation and Flocculation (ed: B Dobias),
Chapter 7 (Dekker: New York).
Sutherland, K L and Wark, I W, 1955. Principles of Flotation (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Tucker, J P, Deglon, D A, Franzidis, J-P, Harris, M C and O’Connor, C T,
1994. An evaluation of a direct method of bubble size distribution
measurement in a laboratory batch flotation cell, Minerals
Engineering, 17(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,
12:1163.
Walker, G W, Walters, C P and Richardson, P E, 1986. Hydrophobic
effects of sulfur and xanthate on metal and mineral surfaces,
International Journal of Mineral Processing, 18:119-137.
37
An Integrated Expert and Vision System at BHP Billiton
Cannington
D La Rosa1, A Cantarutti2, G J Clarke3, M Wortley4, C Bouche5 and O Guyot6
ABSTRACT
As part of a broad program to optimise the performance of their grinding
circuit, BHP Billiton Cannington have installed an OCS© Expert system
together with three VisioRock™ online particle size analysis cameras.
The main purpose of the system is to stabilise grind size and feed to the
flotation circuit.
This particular application of OCS© is the first installation to incorporate
Metso Minerals Process Technology-Asia Pacific’s (MMPT-AP) Dynamic
SAG/AG Mill Model, which is able to predict numerous mill parameters
that were previously unable to be measured with the existing plant
instrumentation. Integrated into the logic of the control algorithm are
online vision based, particle size measurements of the AG mill feed and
recycle crusher product.
This paper will describe the history of the project from initiation to
date, the model and OCS© logic configuration, and demonstrate the
significant benefits realised at Cannington mine.
INTRODUCTION
BHP Billiton’s Cannington operation is the largest and lowest
cost silver and lead producer in the world. A great part of its
success has been due to the continuous innovation and
optimisation undertaken in the nine years since commissioning.
This has seen the initial annual throughput to the concentrator
increase from 1.5 Mtpa in 1997 to a record 3.1 Mtpa in 2005.
As part of ongoing optimisation projects at Cannington,
personnel recognised that Advanced Expert Control of the
crushing and grinding circuit would allow further performance
increases of the crushing and grinding circuits. These increases
would be unable to be gained by running the circuits under
manual control. To this end, Metso Minerals Process Technology
– Asia Pacific (MMPT-AP) in Brisbane and Metso Minerals Cisa
(MMC) from Orleans, France, were contracted to design and
commission an expert control strategy based on their OCS©,
(Optimising Control System), and VisioRock™, online image
analysis systems. The control strategy utilises online particle size
measurements from the VisioRock™ system, expert rules based
on both fuzzy and crisp logic as well as model based control
using the Valery Dynamic SAG/AG mill model. This has been
ported as a Dynamic Linked Library (DLL) and is being used by
the soft sensor module of OCS©. The principal goal of the
control strategy is to maximise and stabilise mill throughput
while maintaining the flotation feed grind size to a value such
that recovery is not negatively affected.
1. Manager – Development and Process Engineering, Metso Minerals
Process Technology Asia-Pacific (MMPT-AP), Unit 1, 8 - 10
Chapman Place, Eagle Farm Qld 4009.
Email: David.LaRosa@metso.com
2. Process Control Metallurgist, Cannington Mine, Base Metals, BHP
Billiton, PO Box 5847, Townsville Qld 4810.
3. MAusIMM, Principal Metallurgist, Cannington Mine, Base Metals,
BHP Billiton, PO Box 5847, Townsville Qld 4810.
4. Process Control Engineer, Metso Minerals Process Technology
Asia-Pacific (MMPT-AP), Unit 1, 8 - 10 Chapman Place, Eagle
Farm Qld 4009.
5. Project Manager, Metso Minerals Cisa, 3 Avenue Claude Guillemin,
45060 Orleans Cedex 2, France.
6. Project Development Manager, Metso Minerals Cisa, 3 Avenue
Claude Guillemin, 45060 Orleans Cedex 2, France.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
It was envisaged that, at the time of submitting this paper, the
OCS© rule base and models would be fully commissioned, and
that a substantial amount of performance data would be available
to demonstrate the effectiveness of the system. Limitations in the
amount of ore able to be supplied to the crushing and grinding
circuits by the mine, and delays in the commissioning of the
particle size indicator (PSI) have unfortunately delayed the final
deployment of the control strategy. Nonetheless, significant
performance data from the initial stages of the strategy
development have been collected and these are presented in this
paper. Descriptions of the Dynamic Model and how it will be
used in the final control strategy are also detailed, along with
comparisons of model performance with logged process data.
THE CANNINGTON CIRCUIT
The Cannington silver, lead and zinc deposit, located in
north-west Queensland, was discovered in June 1990. At the
present time, the Cannington ore resource is 42 million tonnes
with 551 ppm of silver, 12.1 per cent lead and 4.7 per cent zinc.
Ore is extracted from underground operations via both open
stoping and bench mining methods, with access via a decline and
production from a hoisting shaft. The Cannington processing
plant was commissioned in October 1997 with an initial
production rate of 1.5 Mtpa (180 t/h).
The 2002 expansion study investigated the addition of a pebble
crusher to the grinding circuit as well as the introduction of
pebble ports to the AG mill. By utilising the pebble crusher for
partial or full pebble crushing, circuit throughput was able to be
increased to 320 t/h. The pebble crusher also allowed another
control option for stabilising the circuit throughput stable due to
variations in feed ore hardness and size. A permanent Metso
HP300 pebble crusher was installed after the study. With
increased pebble crusher capacity and pebble ports in the
discharge grates of the AG mill, throughputs in excess of 350 t/h
were able to be achieved. However, due to the subsequent coarse
flotation feed, these throughputs could not be sustained without
recovery losses. In order to reduce the flotation circuit feed size
and circuit product variations, a grinding circuit expansion
project was then undertaken.
During the design and engineering stage of the expansion
project, an AG mill, pebble crusher and Vertimill® (AVC)
grinding circuit flow sheet was developed. The final circuit is
shown in Figure 1 and consists of the pebble crusher and AG mill
in closed circuit with primary cyclones followed by a secondary
stage of grinding consisting of one Metso VTM 1500 WB
(1.1 MW) Vertimill® in closed circuit with secondary cyclones.
Grinding circuit simulation results (Jankovic, Valery and Shi,
2001; Jankovic and David, 2002) indicated that sustained
throughputs of 350 t/h with a final P80 of 95 microns could be
achieved with the introduction of both pebble crushing and the
VTM 1500 WB Vertimill®.
DESCRIPTION OF THE OCS© AND
VISIOROCK™ SYSTEMS
Metso Minerals Cisa’s Optimising Control System (OCS©) has
been specifically designed for implementing optimising control
strategies in mineral processing applications. OCS© was
developed from the ground up for advanced process control and
39
D LA ROSA et al
FIG 1 - The two-stage AVC circuit at Cannington (David La Rosa).
with the metallurgist in mind. This enables simple configuration
of the control strategy, reduces the software footprint and
subsequent hardware requirements, and increases the speed of
code execution.
Typically OCS© supplements conventional distributed control
system (DCS) or programmable logic controller (PLC) controls.
OCS© reads process information from the PLC or DCS network,
and uses various techniques to determine new set points,
generally several times per minute. The new set points are then
automatically applied at the PLC or DCS level, without
necessitating any action from the plant operators. The set points
calculated by OCS© continuously pursue strategic objectives. For
example, they could maintain a required product quality while
targeting a technical and economical objective. This typically
results in an increase in capacity, product quality and value, as
well as a possible decrease in production costs.
The OCS© software structure makes the application both
powerful and easily maintained. The implementation
methodology and development of the control strategy always
involves metallurgists, operators and other plant personnel, and
therefore promotes ownership, results in realistic control
strategies and allows excellent acceptance of the system.
OCS© overview
OCS© software hosts, in a single software application, a number
of modules necessary for optimally controlling a mineral
processing plant. An OCS© application may involve some or all
of the following modules:
• The Fuzzy Expert module: a real time expert system, with an
inference engine, built-in crisp and fuzzy logic reasoning and
a knowledge base, this is the core of any OCS© application.
• A Process Display module: easily configured in order to
provide one or several user interface(s), such as flow
diagrams, animated process mimics and data entry forms.
The user may graphically develop new interfaces and
manipulate them using simple Visual Basic code.
• A Soft Sensor module with one or more adaptive predictive
models.
• An Optimiser module, with a powerful constrained
sequential quadratic programming (SQP) algorithm.
40 Fremantle, WA, 19 - 21 March 2007
• A Neural Network module.
• A Vision Module, which can act in the following different
modes:
• VisioFroth for froth flotation plants,
• VisioRock for SAG and AG milling or crushing and
screening plants,
• VisioPellet for pelletising plants, and
• VisioTruck to measure run-of-mine (ROM) fragmentation
at the primary crusher.
• A Tag Statistics module, which generates online statistical
information about the process (averages, histograms, etc),
and may be used for evaluation of plant performance.
OCS© communicates with the process it is controlling via
OPC, ODBC, DDE or DAO communications and there are many
third party drivers on the market that allow it to communicate
with PLCs via ModBus, ModBus Plus and other popular
protocols.
In the Cannington installation OCS© reads process variables
and writes set points directly to eight of the Rockwell PLCs at
site. Over 200 tags are read every 30 second cycle in order to
allow a complete picture of the process status. Set points written
back to the PLC network will include the AG mill fresh feed and
water, pebble crusher feed, primary and secondary cyclone
pressure and Vertimill® feed water. This may be extended as the
control strategy matures.
VisioRock™ overview
Online image analysis has been routinely used in mineral
processing plants to monitor the particle size of SAG mill feed
and crusher feed for a number of years (La Rosa et al, 2001).
It was a natural progression to include a particle size analysis
module with OCS© and for the application of ore particles on
conveyor belts, generally feeding grinding mills or as crusher
product, this module is called VisioRock™ (Guyot et al, 2004).
This is a robust and mature technology and Metso Minerals
Process Technology groups have installed over 40 cameras at
more than ten sites globally.
Ninth Mill Operators’ Conference
 AN INTEGRATED EXPERT AND VISION SYSTEM AT BHP BILLITON CANNINGTON
VisioRock™ hardware consists of high quality analogue or
network cameras focused onto an appropriately lit area of the
conveyor belt. Areas of direct sunlight must be excluded from
the field of view of the camera as this can cause errors in the
software delineations.
The computation steps to convert a single colour image to a
user-definable size distribution are complex and include filtering,
contrast enhancement, segmentation, background rejection, fines
identification, calculation of the 2D particle size distribution
followed by a 2D to 3D particle conversion. Additional analyses
such as colour, texture and shape per size class are also
calculated concurrently. The VisioRock™ module can process
over two high resolution images per second or lower resolution
images at between 15 and 30 frames per second. Figure 2 shows
a screen shot of the VisioRock™ screen processing the
Cannington pebble crusher product. The raw video image with
delineation contours are shown on the left while the dynamic
particle size distribution and histogram are shown to the right.
At Cannington there are three VisioRock™ cameras, one on
the underground jaw crusher product, another on the AG mill
feed conveyor before the crushed pebbles join the feed and
another on the pebble crusher product. The jaw and pebble
crusher product cameras are used to monitor product ‘quality’.
Alarms will be generated when out of specification product is
detected, as these can cause disturbances in the grinding circuit
due to coarser than expected product being processed. The AG
mill feed particle size distribution will be used as an input to the
dynamic model.
As an example of how crusher product size can change over
time, Figure 3 shows the P80 for the pebble crusher after a
rebuild. The P80 values can be seen to drift upwards in response
FIG 2 - Typical VisioRock™ screen capture (David La Rosa).
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
to wear in the machine and change as large adjustments were
made to the machine during its re-commissioning. This is
valuable objective data about the crusher performance.
THE USE OF DYNAMIC MODELS IN OCS©
PROCESS OPTIMISATION
OCS© soft sensors typically make use of so-called deep
knowledge about the physics or the chemistry of the process. A
soft sensor uses such knowledge to infer numerous
non-measurable process variables from few measurements. The
core of an OCS© soft sensor within the modelling module is a
dynamic model of the process. The model of the process is
embedded in an adaptive tuning algorithm called an Extended
Kalman Filter (EKF) that dynamically estimates values of
process parameters. The model, coupled with its adaptive tuning
algorithm is a ‘learning’ model.
In some cases, the dynamic model may be linked to an
optimiser specific to the application. At any time, the state of the
process is known from the ‘learning’ model and the optimiser
can use the predictive capabilities of the model to infer optimised
set points to achieve the best possible performance in the near
future.
The combination of soft sensors, modelling modules and the
expert system in OCS© makes it possible to benefit from the
ability of the modelling modules to generate valuable
information about the state of the process and to predict
optimised set points. Another major advantage of this approach
is that the expert system can take into account numerous
constraints and facts that are not handled by the mathematics of
the models to ensure safe optimised control.
41
D LA ROSA et al

The OCS© optimiser continuously looks for the best path to
peak performance using the mathematical models of the soft
sensor module. It takes objectives and constraints into account,
and computes theoretical optimum model-based set points using
an SQP (sequential quadratic programming) algorithm.
Conceptually, these set points can be applied directly, although in
most cases they are delivered to the fuzzy expert module, which
makes realistic use of the model-based outputs. The structure of
the OCS© soft sensor module is shown in Figure 4.
The dynamic model of a process is built up from separate
models of the components or objects in the process. For example
a grinding circuit model may consist of models for feeders, mills,
pumps and cyclones. A process model describes the relationship
between inputs and outputs of a piece of equipment or section of
the process. The model makes a prediction of the output values
using the input values and the model equations and parameters
that describe the physical process.
The output is the point in OCS© where actual measurements
(called observations) are compared to predicted measurements.
The Extended Kalman Filter looks at the difference and adjusts
the parameters depending on the relative confidence of the model
and the observation. New estimates of the model states and/or
parameters are then calculated and produce updated estimates of
the outputs. If we are more certain that the observations are
accurate than the model, the model parameters are allowed to
vary more to make the estimates fit the observations. Conversely,
if we have more confidence in the model than the observations,
the parameters will be changed less. Setting the relative

FIG 3 - Pebble crusher product size over time (David La Rosa).

©
FIG 4 - OCS soft sensor structure (David La Rosa).

42 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 AN INTEGRATED EXPERT AND VISION SYSTEM AT BHP BILLITON CANNINGTON
confidence of the model and observations is a critical step in
tuning the model. When OCS© runs, it combines all the
component models into one EKF, which constitutes a model flow
sheet of the entire process.
The Valery Dynamic AG/SAG mill model
Central to the final control strategy at Cannington is the Valery
Dynamic SAG/AG mill model (Valery, 1997; Valery and Morell,
1995; Valery and Morrell, 1997; Morrell et al, 2001).
The model incorporates the following major features:
• a method to calculate the energy absorbed by rock particles
in the mill,
• more realistic equations for mass transport,
• online prediction of power draw and outermost charge
trajectory,
• a relationship between mill charge composition and breakage
rates based on projected surface areas of the media and target
particles, and
• the effect of slurry content and hold-up on mill behaviour.
This is a population balance model extended with empirical
formulations to calculate power and slurry flow. The model also
estimates the mill charge distribution at every OCS© cycle, in
this case every 30 seconds. The mass balance of each size
fraction is calculated from the fresh feed, existing charge in that
size fraction, material exchange between size fractions as a result
of breakage (through abrasion and impact breakage events) and
product discharge. The model requires mass inputs at each cycle,
eg new feed rate (from weightometer measurements), feed size
distribution (from VisioRock™) and water flow rates. At model
FIG 5 - Outline of the dynamic AG/SAG mill model algorithm (David La Rosa).
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
start-up the state of the initial charge distribution in the mill is
estimated and quickly converges to arrive at a best fit for the
mass balance around the mill.
Once the mass balance is achieved, gross power and its
components; kinetic, potential, cone and no-load powers are
calculated. The Valery Dynamic Model uses Morrell’s power
calculations (Morrell, 1996) that estimate the charge profile,
shoulder and toe angles of the charge, charge radius and slurry
pool formation. This information can be used for visual sanity
checks in the operator display. It also uses Powell’s outermost
charge trajectory (Powell) calculation to provide valuable
feedback on impact of grinding media for the purposes of liner
protection. The slurry flow out of the mill is estimated with an
empirical formulation based on the mass balance around the mill
and the grate characteristics.
Breakage and discharge mechanisms are the two main
components of the dynamic model. Breakage mechanisms are
described by the breakage rate of each size fraction and its
appearance function. For SAG mills, approximations of ball
charge at any time are required to arrive at best estimates of
breakage rates. Impact energy levels of every size fraction are a
function of drop height in the mill and this is calculated from the
estimated charge profile. Mill speed is also an important
component of breakage rates.
The discharge mechanism describes the flow of water and size
fractions finer than the grate size out of the mill and relies on the
maximum discharge rate of the finest size fraction. The
maximum discharge rate is calculated from the slurry flow
estimation.
Actual and calculated power are used in the Extended Kalman
Filter to correct breakage rates and load size distributions. Figure 5
shows a flow diagram of the Valery Dynamic Model.
43
D LA ROSA et al
Soft sensor implementation in OCS©
© mill without approaching overload conditions. In the initial
The OCS HMI (human machine interface) of the Cannington
grinding circuit includes the two grinding stages described
earlier and is shown in Figure 6. For convenience, the primary
and secondary grinding stage circuits are modelled in two
separate soft sensor modules. The soft sensor modules are used
to predict process state information that are not measured online
and which are used in the OCS© expert system advanced control
logic to achieve the final control objectives. These include
product size distributions, ore hardness parameters, mill per cent
solids, etc.
The process model configuration is done entirely through a
graphical interface as shown in Figure 7. The user is able to
select component models from a library of population balance
models developed in VB.net by MMC and then, by connecting
the various inputs and outputs of the component models, builds
the process flow sheet.
OCS© also includes tools to develop custom models in VB.net.
For instance, the AG mill model used at Cannington is a model
customised by MMC and based on Valery’s dynamic model that
was ported from the original Fortran code by MMPT-AP.
Each large coloured block shown in Figure 7 corresponds to a
component model. The component models used for the
Cannington grinding circuit are Feeder, AG mill, Vertimill®,
Screen, Sump with pump, Crusher and Cyclone.
The sensors, or measurements, correspond to the white blocks
in Figure 7. There are two types of measured variables, measured
inputs such as:
• feed belt scale;
• mill inlet water;
• sump dilution water;
• number of cyclones open;
• VisioRock™ D20, D50, D80, D100 for the fresh feed;
• crusher feed;
and measured outputs (or observations) such as:
• VisioRock™ D20, D50, D80, D100 for the crushed pebbles;
• mill power;
• mill bearing pressure;
• mill recycled pebbles; and
• cyclone feed flows, densities and pressures.
The sensors are linked to model inputs (measured inputs) or to
model outputs (measured outputs). The other links in the
graphical model configuration correspond to the ore streams
between component models. For instance, the stream definition
for the AG mill circuit includes 30 size classes, the energy input,
water and total solids flows.
Once the process model is configured, the EKF configuration
and tuning phases start. These two phases are under development
at the present time for the Cannington project. OCS© provides
tools that allow online data to be collected and then replayed
off-line. The soft sensor tuning is then mostly done off-site using
continuously collected historical data. The EKF configuration
consists of selecting the model parameters that will be
continuously adjusted online while the tuning consists of
defining the range and errors (weights or relative confidence) for
each adjusted parameter and measured output (observation).
INITIAL PEBBLE CRUSHER CONTROL
STRATEGY
The performance of the Cannington AG mill is extremely
sensitive to the amount of scats crushed in the pebble crushing
circuit. The removal of these critically sized particles from the
44 Fremantle, WA, 19 - 21 March 2007
primary grinding circuit allows more feed to be processed by the
stages of the project, focus was placed on controlling the pebble
crusher to maximise AG mill feed rate. Figure 8 shows a
schematic of the control logic.
In this strategy the operator enters a bearing pressure set point.
In the short term, this is proportional to the load in the mill.
OCS© then adjusts the feed to the pebble crusher using fuzzy
logic sets to determine how far away and how fast to change the
pebble crusher amps set point to meet this criteria. In this way
the pebble crusher is protected from overload situations by
controlling the amps set point instead of the feed rate.
The simplified OCS© rules for adjusting the pebble crusher
amps setpoint and feed rate to the mill, are:
• if bearing pressure is high THEN increase pebble crusher
amps setpoint,
• if bearing pressure is low THEN decrease pebble crusher
amps setpoint,
• if bearing pressure is at setpoint THEN no change to pebble
crusher amps setpoint,
• if bearing pressure is very high THEN decrease fresh feed
rate to the AG mill, and
• if bearing pressure is very low THEN increase fresh feed rate
to the AG mill.
The pebble crusher control rules respond as follows:
• if pebble crusher amps are high THEN decrease pebble
crusher feed setpoint,
• if pebble crusher amps are low THEN increase pebble
crusher feed setpoint, and
• if pebble crusher amps are at setpoint THEN no change to
pebble crusher feed setpoint.
The definitions of high, low, very high, etc are all fuzzy and
the rules also take into account the rate of change of the process
parameter they are describing. The magnitude of subsequent
setpoint changes will vary according to these criteria. Decrease
and increase changes are similarly described.
This strategy, while quite simple, and requiring less than 20
rules, allowed the AG mill to be taken to areas of operation that
were not considered stable under manual control. Table 1 details
the statistics of mill throughput before and after OCS© control of
the pebble crusher.
TABLE 1
OCS© pebble crusher control results.
Average Standard Sample period
throughput deviation (days)
(t/h) (t/h)
Before OCS© 430 63 30
After OCS© 463 38 40
It was shown that having expert system control on the pebble
crushing circuit significantly decreased the variability in mill
throughput. What was not expected, however, was the magnitude
of the throughput increase that was achieved with this strategy
(33 t/h, or eight per cent). The number of mill overloads that
required a rapid drop in feed tonnes or the dumping of scats was
also reduced while OCS© was in control. It is expected that the
final grind size was coarser under these conditions; however, no
data was collected to quantify this. Even if conservative
assumptions are made about recovery losses due to an increased
final grind size, this simple strategy amounted to a significant
positive benefit to the operation.
Ninth Mill Operators’ Conference
 AN INTEGRATED EXPERT AND VISION SYSTEM AT BHP BILLITON CANNINGTON
FIG 6 - OCS© HMI for the Cannington Grinding Circuit (David La Rosa).
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 45
D LA ROSA et al
FIG 7 - OCS sensor module configuration interface (David La Rosa).
©
46 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference
 AN INTEGRATED EXPERT AND VISION SYSTEM AT BHP BILLITON CANNINGTON

ONGOING DEVELOPMENT OF THE FINAL
CONTROL STRATEGY
As mentioned in the introduction, it was envisaged that the full
control strategy would be commissioned by the time this paper
was written. Ore limitations from the mine have meant that there
is little pressure on the crushing and grinding circuits for
production and it would be difficult to rigorously test the OCS©
system over a wide range of operating conditions. Additionally,
as one of the KPIs of the control system is the final grind size,
delays in commissioning the PSI mean that the final live model
tuning and control strategy development have been unable to be
completed. However, MMPT-AP and MMC control engineers
have spent time since the last site visit replaying captured
historical data through the OCS© projects and tuning the models.
For example, Figure 9 shows a comparison between measured
bearing pressure and estimated mill load for the Cannington AG
mill over a 24-hour period.
Such simulation results have given personnel involved in the
project confidence that the model and optimiser can be used to
further enhance the control strategies at Cannington.
The new strategy will add control of the primary and
secondary cyclone pressure setpoints based on sump level,
modelled per cent solids and pump speed as well as water
addition to the AG mill, Vertimill® and sumps. Most importantly,
the balance between throughput and grind size will be
continuously monitored and adjusted when necessary. This work
is ongoing and the results will be reported when available.
CONCLUSIONS
The installation of the VisioRock™ and OCS© expert control
systems at BHPB Cannington have allowed the operation to
achieve another step change in plant performance. Significant
stable throughput increases have been proven to be achievable,
even with a simple control strategy. AG mill overload events

FIG 8 - Pebble crusher control strategy logic (David La Rosa).

FIG 9 - Comparisons between measured and estimated AG mill load variables (David La Rosa).

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 47

D LA ROSA et al
have been reduced, minimising rehandle costs associated with
the cleaning up of scats dumped on these occasions.
Consequently, pebble crusher utilisation has also increased.
The dynamic model will allow process variables to be
estimated and used where previously not possible and it is
believed that further performance gains will be realised by the
operation once the final model-based strategy is implemented.
REFERENCES
Clarke, G and La Rosa, D, 2006. Changes and upgrades at Cannington, in
Proceedings IIR Crushing and Grinding Conference 2006 (Institute
for International Research).
Guyot, O, Monredon, T, La Rosa, D and Broussaud, A, 2004. VisioRock,
an integrated vision technology for advanced control of comminution
circuits, in Proceedings IIR Crushing and Grinding Conference 2004
(Institute for International Research).
Jankovic, A and David, D, 2002. Further simulations for Cannington
grinding circuit expansion, JKTech report, Job number 02212.
Jankovic, A and Man, Y, 2003. Technical support during commissioning
and optimisation of the new Cannington comminution circuit, Metso
Minerals Process Technology Asia-Pacific report, Job number 32-03.
Jankovic, A and Valery, W, 2004. Optimisation of the new Cannington
comminution circuit, Metso Minerals Process Technology
Asia-Pacific report, Job number 24-04.
Jankovic, A, Valery, W and Clarke, G, 2006. Design and implementation
of an AVC grinding circuit at BHP Billiton Cannington, in
Proceedings SAG2006 – Circuit Design, International Conference on
Autogenous and Semiautogenous Grinding Technology, 2:290-300.
Jankovic, A, Valery, W and Shi, F, 2001. Grinding expansion study at
BHP Billiton Cannington, JKMRC report.
48 Fremantle, WA, 19 - 21 March 2007
La Rosa, D, Girdner, K, Valery, W and Abramson, S, 2001. Recent
applications of the Split-Online Image Analysis System, in
Proceedings Southern Hemisphere Meeting on Mineral Technology
2001, 1:15-19.
Morrell, S, 1996. Power draw of wet tumbling mills and its relationship
to charge dynamics, IMM Trans, C38-C62.
Morrell, S and Valery Jnr, W, 2001. Influence of feed size on AG/SAG
mill performance, in Proceedings SAG2001 – SAG Mill Circuit,
International Conference on Autogenous and Semiautogenous
Grinding Technology, 1.
Morrell, S, Valery, W, Banini, G and Latchireddi, S, 2001. Developments
in AG/SAG mill modelling, in Proceedings SAG2001 – SAG
Mill Circuit, International Conference on Autogenous and
Semiautogenous Grinding Technology, 1.
Powell, M S, 1991. The effect of liner design on the motion of the outer
grinding elements in a rotary mill, International Journal of Mineral
Processing, 31:163-193.
Valery, W, 1997. A model for dynamic and steady-state simulation of
autogenous and semi-autogenous mills, PhD thesis, JKMRC,
University of Queensland, Brisbane.
Valery, W and Morrell, S, 1995. The development of a dynamic model for
autogenous and semi-autogenous grinding, Minerals Engineering,
8(11):1285-1297.
Valery, W and Morrell, S, 1997. A dynamic and steady-state model of
autogenous and semi-autogenous mills, in Proceedings International
Conference Comminution 1998.
Valery, W and Morrell, S, 2001. JKSagCharge – A tool to estimate
AG/SAG mill charge volume and position online, in Proceedings
Comminution 2001.
Ninth Mill Operators’ Conference
Optimising Flotation Innovations and the Role of Plant Test Work
P Bourke’ and S Gronstrand*

ABSTRACT
Optimising flotation designs and innovations is a complex process involving many parameters,
one of which is ensuring sufficient test work is carried out at plant level before these
innovations are launched in the marketplace.
This presentation will describe initial test work phases at various plants and how results and
observations have helped to validate and optimise flotation design innovations for Outokumpu
Technology. Credible, accurate test work at plant level is crucial to ensuring innovations are
proven when launched.
This presentation will include the following:
0 recent test work from Tankcell@-300,the world’s largest flotation unit;
0 further test work of the new FloatForceTMmechanism at Aitik and Siilinjarvi mine sites;
0 comparative results on bubble size with the new FloatForceTMmechanism;
0 plant test work on the flotation of coarse particles;
0 plant results using Outokumpu Technology’s high shear mechanism;
0 results from the innovative auxiliary agitator (FlowBooster); and
0 Dispersed Gas Guidance and current test results.
Test work at plant level, involving flow variables, as well as metallurgy, will ensure
innovations are more than just marketing gimmicks and are proven optimisers in the flotation
process.

1. Global Technology Manager - Flotation Process, Outokumpu Technology Pty Ltd, Level 2, 1 Walker Avenue, West Perth WA 6005.
Email: peter.bourke@outokumputechnology.com
2. Global Technology Manager - Flotation, Outokumpu Technology, Finland. Email: sami.gronstrand@outokumputechnology.com

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 49

The Application of Ausmelt’s AM28 Alkyl Hydroxamate Flotation
Reagent to Fox Resources’ West Whundo Copper Ore at Radio
Hill, Western Australia
T Hughes1, K Lee2, G Sheldon3, J Bygrave4 and L Mann5
ABSTRACT
Ausmelt is supplying commercial quantities of an alkyl hydroxamate
flotation collector AM28 for flotation of oxide and supergene base metal
ores and tailings. AM28 is a non-hazardous product used as a ten per cent
to 50 per cent solution in dilute potassium hydroxide. It is a selective
flotation collector for oxide or oxidised base metal minerals and replaces
conventional sulfidisation for oxide mineral flotation.
At Radio Hill, the AM28 is used in conjunction with traditional sulfide
collectors, which allows simultaneous flotation of oxide and sulfide
copper minerals. The Fox Resources Radio Hill copper concentrator near
Karratha, Western Australia, is currently configured to treat supergene
copper ore sourced from the West Whundo open pit. This Archean VMS
type copper deposit has been deformed and remobilised with uplift,
erosion and weathering to form a significant supergene zone. The sulfide
copper minerals present in the upper zone are digenite and chalcocite
with lesser chalcopyrite and covellite. Oxide copper minerals present
include cuprite, with lesser malachite and azurite. Gangue
minerals include quartz, talc and chlorite.
Plant trials at Radio Hill are reported for supergene West Whundo ore,
where the AM28 was stage added into the circuit at nominally 80 g/t.
Plant data over three months shows the copper recovery increased
consistently from around 50 - 60 per cent to above 70 per cent when
AM28 was added. This has been proven statistically. The grade of the
concentrate was consistent with and without AM28, at nominally 22 -
24 per cent copper.
INTRODUCTION
Base metal flotation concentrates are primarily sourced from
ores containing metal sulfides. Some deposits contain oxide
metal minerals and or partially oxidised sulfides. Metal
recoveries from ores containing supergene and oxide
mineralisation are usually affected by losses of these minerals to
tails, due to the poor efficiency of conventional sulfidisation
techniques and recovery with conventional sulfide collectors.
The availability of a flotation collector which floats tarnished
sulfides and oxide metal minerals represents a considerable
advance in the upgrading of supergene and oxide base metal
deposits.
A review of the application of chelating reagents as flotation
collectors by Hanson and Fuerstenau (1991) reported on the
potential of hydroxamates for the upgrading of oxidised base
metal ores and minerals. To this end, Ausmelt Limited has
developed and patented a range of alkyl hydroxamates (Hughes
2004, 2006) as flotation collector reagents for use with oxidised
and supergene base metal ores and tailings.
1. Principal Geochemical Consultant, Ausmelt Limited, 12 Kitchen
Road, Dandenong Vic 3175. Email: Terry.Hughes@ausmelt.com.au
2. Metallurgical Manager, Ausmelt Limited, 12 Kitchen Road,
Dandenong Vic 3175. Email: Kenneth.Lee@ausmelt.com.au
3. AAusIMM, Manager, Ausmelt Minmet Metallurgical Laboratory,
Unit 3, 8 Yandina Road, West Gosford NSW 2250.
Email: Graeme.Sheldon@tpg.com.au
4. Senior Metallurgist, Fox Radio Hill Pty Ltd, PO Box 1629, Karratha
WA 6714. Email: Jerehmy.Bygrave@foxradiohill.com.au
5. MAusIMM, Production Manager, Fox Radio Hill Pty Ltd, PO Box
1629, Karratha WA 6714. Email: Laurie.Mann@foxradiohill.com.au
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
The most effective reagent is potassium n-octyl hydroxamate
(AM28). It has been manufactured in Melbourne in multi-tonne
quantities using Ausmelt’s environmentally friendly aqueous-
based patented reaction process. The reagent is used as a ten to
50 per cent solution in dilute potassium hydroxide as a flotation
collector for oxidised base metal sulfides, oxide metal minerals,
precious metals (Ag, Au and PGMs) plus copper metal and
lithophile metal oxides.
Laboratory and mine site trials have shown that AM28 can be
used in conjunction with sulfide collectors and is effective for the
flotation of oxide minerals and oxidised sulfide minerals. This
paper will discuss its application to Radio Hill.
The alkyl hydroxamate reagents
Alkyl hydroxamates are defined as anionic collectors, which
have been known for many years (Yale, 1943; Bauer and Exner,
1974). They usually consist of a saturated alkyl chain and AM28
consists of a C8 octyl saturated chain with minimal C10, C12
and C6 content. Hydroxamates behave as weak acids with a pKa
between nine and ten. They are effective in high pH systems
(above 10.5) but can be used to improve metal recovery in pulps
with a pH ranging from 7 to 10. The typical AM28 hydroxamate
structure is shown in Figure 1.
H
N
- K
+
O
O
FIG 1 - Typical AM28 hydroxamate structure.
At Radio Hill, AM28 is used in conjunction with traditional
sulfide collectors to allow the simultaneous flotation of oxide and
sulfide copper minerals.
AM28 replaces conventional sulfidisation for oxidised mineral
flotation and is a non-hazardous product. Extensive toxicological
investigations on AM28 and its predecessor AM2, have
been carried out by NICNAS (National Industrial Chemicals
Notification and Assessment Scheme Australia) to ensure the
safe handling and use of the hydroxamate reagents. This resulted
in NICNAS granting a commercial registration for Ausmelt to
manufacture and distribute AM2 (NICNAS Code is NA/917)
and AM28. Both AM2 (approximately 50:50 C8:C10) and
AM28 (approximately 95 per cent C8) are classified as a ‘non-
dangerous goods’.
The formation and structure of AM28 is discussed in detail
in Hughes (2006). Although fatty hydroxamic acids and
hydroxamate reagents have been the subject of many research
and patent publications, little information is provided on the
structure and mode of the mineral surface attachment of the
reagents. Great care is taken during the manufacture of AM28 to
minimise the content of free fatty acid (FFA), which will react
with soluble calcium ions in the pulp to form a soap-based froth.
51
T HUGHES et al
The successful implementation of the AM28 manufacturing
process depends on the molar proportions of the feedstock
reagents, the control of reaction temperature and the sequence of
reaction additions. The final product contains 45 - 50 per cent
alkyl hydroxamate, which is a 50:50 mixture of the acid and the
potassium salt.
The novel structure of AM28 combined with the saturated C8
hydrophobic tail makes the reagent highly effective as a selective
collector of oxidised metal minerals in the froth flotation
process. In practice the C8 tail promotes bubble attachment with
minimal over-frothing from clays and fines and the hydroxamate
moiety selectively attaches to the oxidised metal surface (eg
Cu(II)) by chelation.
The application of AM28 to copper flotation
Most copper deposits have a mineralogy and geochemistry which
changes significantly with depth. In a typical Cu ore deposit, the
Cu and associated Au or Mo are often significantly enhanced by
oxidative supergene enrichment above the primary chalcopyrite
zone. Cu minerals above this zone are progressively more
oxidised until at or near the surface, fully oxidised Cu species
such as malachite and azurite occur. Cu minerals in the primary
and secondary enrichment zone (eg chalcocite, covellite, bornite,
chalcopyrite and pyrite) can be recovered by conventional
flotation reagents. However, many sulfide minerals, such as
chalcocite (Kuhn, 1968) are readily surface oxidised and may
benefit from the addition of AM28 to the flotation reagent
regime.
In the upper zone of oxidised enrichment oxides, carbonates,
sulfates and hydroxides and metallic Cu can be found. In this
zone, which is often stockpiled for future treatment, the use of
AM28 will allow marketable Cu concentrates to be recovered
using the existing mine plant flotation systems.
RADIO HILL MINE, AUSTRALIA
Fox Resources currently operates the Radio Hill project, which is
located 35 km South of Karratha in the West Pilbara region of
Western Australia. The Radio Hill treatment plant is capable of
producing copper, nickel and zinc concentrates. The plant is
currently configured to treat supergene copper ore sourced from
the West Whundo open pit.
The Whundo and Yannery mining leases cover an area of
approximately 900 ha and are located 10 km south of the Radio
Hill treatment plant. Copper was mined from the Whundo deposit
Float Rougher Cell
Feed
Ball Mill
Feed
Cyclone
Final Concentrate
FIG 2 - Radio Hill plant flow sheet.
52 Fremantle, WA, 19 - 21 March 2007
intermittently since 1911, with the most recent operations
ceasing in 1976. Total production from this deposit was 12 000 t
of direct shipping ore grading 27.4 per cent copper.
The Whundo copper-zinc deposit is located at the contact
between a mafic dominated volcanic sequence and a felsic
volcanoclastic sequence within the Whundo Group. The deposit
is interpreted as a VMS type deposit associated with the
development of a back arc extensional environment during the
Archean. The deposit has been gently folded. Uplift and erosion
during the Archean and more recent times has developed a small
copper supergene zone above the Eastern end of the deposit and
a more significant zinc-copper supergene zone over the Western
end. A small oxide resource to the South is directly associated
with shear remobilised sulfides below the main lode. Subsequent
tectonic events have deformed the deposit into a north-westerly
dipping fold, with minor remobilisation of massive sulfide into
local shears.
The copper minerals present in the supergene ores are
approximately equal amounts of digenite and chalcocite with
lesser amounts of chalcopyrite and covellite. Significant copper
oxide mineralisation has also been identified and includes
cuprite, with lesser malachite and azurite.
The digenite is predominantly present as fine scale veinlets of
coarse pyrite and marcasite whose radial growth habit suggests
former pyrrhotite. The chalcocite occurs as fine veins through
coarse pyrite, but on a coarser scale than the digenite. It also forms
coarse composites with chalcopyrite and non-invaded pyrite. The
chalcopyrite occurs with the chalcocite as above, and the covellite
once rimming chalcopyrite. The main gangue mineral composites
are with sulfides in the coarser grains in quartz. There is talc and
chlorite present that tend to be discrete fines.
THE RADIO HILL FLOTATION CIRCUIT
The flotation circuit at Radio Hill consists of a Rougher cell
(32 m3) followed by a scavenger cell (2.34 m3), with the tail from
the scavenger cell sent to tailings. Froth from both cells is sent to
two cleaner cells (10.35 m3) that are in series to upgrade the
rougher and scavenger concentrate from 12 per cent Cu to a
27.4 per cent Cu final concentrate. The tail from the cleaner cells
is combined with the scavenger tail and sent to tailings. Figure 2
shows a flow diagram of the circuit. Concentrate from the rougher
cell can also be routed straight to final concentrate bypassing the
cleaning circuit. The retention time through the circuit is 16.0
minutes (13.2 minutes in the rougher cell and 2.8 minutes in the
scavenger cell) and 10.2 minutes in both cleaner cells.
Scavenger Cell
Final Tail
Cleaner Cells
Ninth Mill Operators’ Conference
 THE APPLICATION OF AUSMELT’S AM28 ALKYL HYDROXAMATE FLOTATION REAGENT

Under normal operation the circuit operates at 40 tonnes per change in concentrate grades. AM28 has been added to the
hour. This equates to an annualised throughput of 340 000 tonnes rougher feed or feed conditioner (at about 70 per cent of the total
per year. The slurry density at flotation feed is 26 per cent solids plant addition) and the balance to the scavenger feed. Doses
and at the target pH of the circuit is 11.5. The copper grade in the varied between 80 g/t and 120 g/t with the higher doses
flotation feed typically varies in the range of 3.5 to six per cent corresponding to higher head grades.
Cu. Table 1 states the chemical suite that is used on the circuit. Figure 4 shows shift production data at Radio Hill with and
Both copper promoters (AERO 3477 and AM28) are staged to without AM28. It can be seen that over the entire range of head
both rougher and scavenger cells with a majority of collector grades that AM28 has improved copper recovery over the normal
added to the rougher cell. chemical suite. It is interesting to note that without AM28 the
copper recovery decreases as the head grade increases above five
TABLE 1 per cent Cu. It is suspected that this is due in part to low retention
Chemical suite at the Radio Hill copper concentrator. times through the circuit and also that at higher head grades there
is a change in mineralogy to slower floating tarnished sulfides
Addition point and oxidised minerals. This decrease in recovery at higher head
Mill Cyclone Rougher Scavenger
grades was not seen when AM28 was used. Therefore, it appears
feed underflow the AM28 increases the flotation rates of the slower floating,
tarnished and oxidised copper minerals.
pH control (kg/t)
Visual examination of hand panned plant concentrates using a
Quicklime 4.8 simple binocular microscope at site, also clearly shows oxide
Frother (g/t) copper minerals such as cupite in the concentrate when AM28
Cyctec ore prep 45 was being added and the absence of oxide minerals when AM28
X-133-OZ was not added.
Copper promoters (g/t) From general observations and plant operator comments, the
Areo 3477 50 30
froth conditions and stability were also improved when AM28
(di thiophosphate was added.
based) Napier-Munn (1995, 1998) has shown in the past that a
Guar gum 75 statistical linear correlation analysis can be used to compare two
different operating conditions. In this case it is interesting to
AM28 55 25
determine the effect that AM28 has on copper recovery. From
Figure 4 a linear relationship can be seen for when the copper
head grade is between 3.5 to five per cent Cu. Above five per
PLANT TRIALS WITH AM28 AT RADIO HILL cent Cu the relationship does not hold true because of reasons
Before AM28 was added to the circuit the average copper explained previously, therefore the statistical analysis will be
recovery of the circuit was 58.3 per cent. It can be seen in the conducted between the range 3.5 to five per cent Cu. A detailed
snapshot of the plant in-stream analysis output shown in Figure 3 method is discussed in Napier-Munn (1995, 1998).
that when AM28 is not used the final tail copper concentration The relevant statistics from plant data are compared in Table 2.
was around 2.7 per cent Cu, and when AM28 is used the final tail From the analysis it can be concluded that the correlation
copper concentration drops to 2.0 per cent Cu, increasing the coefficients are highly significant. The test for residual mean
recovery to an average of 70.6 per cent. There was an immediate squares is not significantly different so the comparison between
and clear result when AM28 was used on this circuit without any the two operating conditions can proceed. The gradients of each

Flotation
Feed
(%Cu)

Copper
Final
Concentrate
(%Cu)

Final Tail
(%Cu)

AM28 ON AM28 OFF AM28 ON

FIG 3 - In-stream analysis trends of flotation feed, final concentrate and final tail.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 53

T HUGHES et al

Comparison of Regression Lines

90

80

70

Recovery (%)
60

50

40

30

20

10
3 3.5 4 4.5 5 5.5 6 6.5
Feed Grade (%Cu)
AM28 NO AM28 Linear (NO AM28) Linear (AM28)

FIG 4 - Comparison of recovery versus feed grade regression lines with and without AM28.

At a current copper price of A$6900 per tonne Cu, this

TABLE 2 has increased the annual revenue for Radio Hill by over
Comparison of operating condition at the Radio Hill A$6 000 000. AM28 has been and is used routinely at Radio Hill
copper concentrator. for the treatment of West Whundo supergene ores, and is now
considered a standard reagent. The typical AM28 dosage has
Statistic normal AM28 No AM28 varied between 80 g/t and 120 g/t with the higher doses
Number of data points 16 14 corresponding to higher head grades, the annual cost of AM28
for Radio Hill is approximately A$400 000.
Mean Cu recovery (%) 70.62 58.34
Correlation Value 0.53 0.54
coefficient CONCLUSIONS
t/d.f 2.37/14 2.27/12
By the demonstration at Radio Hill, AM28 has significantly
Significance level >95% >95%
increased copper recovery by over 10.9 per cent (and possibly
Residual MS Value 16.04 20.79 over 20 per cent for head grades above five per cent) when
(S2) F/d.f 1.30/12;14 treating a copper supergene orebody, which contains mainly
Significance level Not
oxide minerals of cuprite and lesser malachite and azurite; and
significant sulfide minerals of digenite, chalcocite, chalcopyrite and
covellite. This has been proven statistically.
Intercept Value 46.12 31.45
The AM2/AM28 reagents have the potential to enhance both
Variance 108.17 141.98 mining and processing economics of base metal ore deposits.
Gradient Value 5.45 6.31 Stripping ratios in open cut mining operations could potentially
Variance 5.31 7.73
be significantly reduced with the option to effectively treat the
mineralised oxide orebody using conventional flotation techniques
F/d.f 0.06/1;26 with AM28.
Significance level Not
significant REFERENCES
Separation of Value 10.94
Bauer, L and Exner, O, 1974. The chemistry of hydroxamic acids and
lines Variance 2.61 N-hydroxyimides, Angew Chem Internet Ed 13(6):376-384.
t/d.f 6.77/26 Hanson, J S and Fuerstenau, D W, 1990. The electrochemical and
flotation behaviour of chalcocite and mixed oxide/sulfide ores, in
Significance level >99% Flotation of Sulfide Minerals (ed: K S E Forsburg), International
Journal of Mineral Processing, 33:33-37.
Hughes, T C, 2004. Hydroxamate composition and method for froth
correlation are not significantly different (F <1), but the t-value flotation, Ausmelt Limited, US Patent Application Publication IS
for the mean separation of lines reaches a significance at more 2004/0211933 A1.
than 99 per cent. It can be concluded that AM28 does produce a Hughes, T C, 2006. Preparation of fatty hydroxamates, US Patent
significantly higher copper recovery than the usual non-AM28 Application Publication US2006/7, 049, 452.
operation. The benefit of using AM28 can be estimated by Kuhn, M C, 1968. Doctorial dissertation, Colorado School of Mines.
separation of the lines. In this case AM28 has increased recovery Napier-Munn, T J, 1995. Detecting performance improvements in trials
by 10.9 per cent when the head grade is between 3.5 and five per with time varying mineral processes – Three case studies, Minerals
cent Cu. Above five per cent Cu head grade the benefit of AM28 Engineering, 8(8):843-858.
would be greater than this (typically over 20 per cent) as Napier-Munn, T J, 1998. Analysing plant trials by comparing recovery –
recovery drops with a higher head grade under normal operating Grade regression lines, Minerals Engineering, 11(10):949-958.
conditions, straying away from a linear relationship. Yale, H L, 1943. The hydroxamic acids, Chemical Reviews, 33:209-216.

54 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

Operating Variables Affecting the Bubble Size in Forced-Air
Mechanical Flotation Machines
J E Nesset1, J A Finch2 and C O Gomez3

ABSTRACT 3
Outokumpu TC 130 LDI
The bubble size, together with the gas rate, determines the bubble surface 2.5
area flux in flotation machines and, therefore, impacts directly on the
kinetics of flotation. Understanding the variables that influence bubble 2

size is one of the keys to understanding flotation recovery. Surprisingly,
these relationships are poorly understood, largely due to the difficulties
associated with gas dispersion measurements in industrial environments.
The McGill gas dispersion sensors (bubble size, gas rate, gas holdup) are
capable of reliably making such measurements. This paper reports on a
study, using the McGill sensors in a 0.8 m3 Metso RCSTM pilot cell,
which investigated the effect on bubble size of frother type and
concentration, gas rate, temperature and rotor speed. A Denver laboratory
cell was used to investigate the effect of elevation above sea level. Frother
concentration proved to have the predominant influence on bubble size,
and the effect of frother type can be normalised as a function of the
critical coalescence concentration (CCC). Measurements made at many
plant sites worldwide are compared to the model developed using the
Metso RCSTM unit.
INTRODUCTION
Shortly after the industrial advent of froth flotation at the Broken
Hill operation just over 100 years ago, T A Rickard (Rickard,
1916) observed:
… we know that the key to the flotation process is
to be found, not in the oil, the acid, or the
apparatus, but in the bubbles.
Today, with the advent of new measurement technologies and
computational techniques, the industry is finally getting around
to better defining the role of bubbles in flotation. Part of the
problem has been the lack of robust sensors capable of making
sufficiently precise measurements in a reasonable time frame in
the industrial environment. The recent development of gas
dispersion sensors by the group at McGill University (Gomez
and Finch, 2002) has helped to accelerate the study of the role
bubble size plays in flotation at operating plants (Pyecha et al,
2006; Hernandez-Aguilar, Thorpe and Martin, 2006; Coleman,
Urtubia and Alexander, 2006).
The impetus behind the work reported here was a series of
earlier studies (Nesset et al, 2005; Nesset et al, 2006) in which
the McGill gas dispersion sensors were used to measure bubble
size, superficial gas velocity and gas holdup in a suite of
industrial flotation plants. The findings from those studies,
shown in Figure 1, suggested that the Sauter mean diameter, the
D32, was well represented by the following relationship:
1. PhD Candidate, McGill University, Department of Mining, Metals
and Materials Engineering, M H Wong Building, 3610 University
Street, Room 2480, Montreal QC H3A 2B2, Canada.
Email: nessetech@videotron.ca
2. Professor, McGill University, Department of Mining, Metals and
Materials Engineering, M H Wong Building, 3610 University Street,
Room 2600, Montreal QC H3A 2B2, Canada.
Email: jim.finch@mcgill.ca
3. Senior Research Associate, McGill University, Department of
Mining, Metals and Materials Engineering, M H Wong Building,
3610 University Street, Room 2580, Montreal QC H3A 2B2, Canada.
Email: cesar.gomez@mcgill.ca
D32 (mm)
Bateman TC 50 Impala
1.5 Metso RCS 30 Impala
Outokumpu 28U Raglan
1
0.5
0
0 0.5 1 1.5 2 2.5 3
Jg (cm/s)
FIG 1 - The D32 versus Jg relationship from earlier plant studies.
D32 = DO + CJg
n
(1)
where:
Jg is the superficial gas velocity
Do is the bubble size at Jg = 0
n is a constant
C is a parameter dependent on the system chemistry and
bubble generation mechanism
This relationship and the notion of a finite, creation bubble
size (Do) at Jg = 0 that appears to be similar (0.5 - 0.6 mm) for all
the flotation plant systems investigated was intriguing and called
for further study in a more controlled environment outside the
plant.
The current investigation was therefore launched to more fully
investigate the relationship between bubble size and the main
operating variables of industrial flotation systems, viz frother
concentration and type, air rate, impeller speed (power intensity),
pulp temperature and gas density. The latter two are often not
deliberately controlled but could be important as many plants
operate in remote regions at extreme temperatures or at altitudes
were the air density is considerably less than at sea level. The
method selected was to employ an available pilot sized tank-cell
supplied by Metso Minerals (a fully-instrumented 0.8 m3 pilot
unit with a Svedala mechanism) that was sufficiently large to
avoid the problems of scaling that result from using much
smaller laboratory machines. Figure 2 shows the unit in position
in the McGill gas dispersion/hydrodynamics laboratory.
PREVIOUS RESEARCH
Characterising the bubble size distribution
A population of bubbles in an aerated and agitated vessel will
have a (number) frequency distribution of bubble sizes that may
exhibit any of a variety of shapes; normal, lognormal or bimodal
(Bailey, Gomez and Finch, 2005; Hernandez-Aguilar and Finch,
2005) being the most common. As evident from Figure 3, the
surface area and volume distributions (by bubble size) can vary
significantly from the number distribution for a typical

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 55

J E NESSET, J A FINCH and C O GOMEZ

The D32 is widely used in reactor engineering and in flotation

since it represents the equivalent surface area of the bubble
distribution for a given volume of gas. Since the D32 is fairly
insensitive to the number of very small bubbles in a distribution
(they contribute little to the total surface area) the D10 is a better
measure when the total number of bubbles is an important
parameter. This has been demonstrated by Nesset et al (2005)
with a variety of plant measurements showing that the number of
small bubbles (minus 1 mm) is important when collecting very
small particles (minus 10 µm). Both D10 and D32 are indicated for
the distribution shown in Figure 3. Note the large difference in
their values with D10 being 0.99 mm and D32 being 1.71 mm for
the example given, chosen to highlight the point.

Bubble surface area flux and the link to flotation

recovery
As noted, a key parameter in flotation is the concept of bubble
surface area flux, Sb, having units of surface area of bubbles
leaving the liquid phase per unit of cross-sectional area per unit
of time, s-1. The relationship for Sb is (Finch and Dobby; 1990):

Sb = 6 Jg D32 (4)

where:
Jg (cm/s) is the superficial gas velocity (the volumetric flow rate
of air per unit cross-sectional area of the cell)

TM 3
D32 is in units of cm (for D32 in mm the constant in
FIG 2 - Metso RCS 0.8 m pilot cell in position in McGill’s gas Equation 4 becomes 60)
dispersion/hydrodynamics laboratory. It has been demonstrated (Gorain, Franzidis and Manlapig,
1997; Heiskanen; 2000; Hernandez, Gomez and Finch, 2003)
0.3
that there exists a linear relationship between the flotation rate
Number constant, k (s-1), and Sb according to:
Surface
Volume k = PSb Rf (5)
0.2
Frequency

where:
Rf is the froth recovery factor
0.1 P is the mineral floatability parameter, both unit-less
Since flotation can be treated at first approximation as a first
order kinetic reaction, the inverse relationship between bubble
size D32 and flotation recovery through the Sb and k parameters is
evident.
0.0
0.1 D10 = 0.99 mm 1.0 D32 = 1.71 mm 10.0 Modelling D32 in aerated stirred tanks
Db (mm)
The ability to predict and model D32 based on the main flotation
FIG 3 - Typical bubble size distribution showing the number, variables would clearly be advantageous in terms of process
surface area and volume distributions as well as D10 and D32. understanding and optimisation. There currently exists no model
for D32 specifically for industrial flotation; however, relationships
have been developed for use in the aerated stirred reactors
lognormal plot. Any number of measures of mean diameter can commonly used in the chemical process industries. These
be used to represent a population of bubbles (Kelly and reactors differ in some significant ways from flotation machines,
Spottiswood, 1982); however, the most commonly used in so the models developed for their use may not be applicable. As
aerated stirred-tank reactor engineering and flotation are the a starting point, however, a brief review will be made.
arithmetic mean, D10, and the Sauter mean, D32, defined as
follows: The analysis presented by Hinze (1955) forms the basis for
most of the subsequent work and is based on a balance between
n inertial (disruptive-turbulent) forces and stabilising (surface
∑d i tension) forces. There exists, therefore, a critical Weber No:
D10 = 1
(2)  ρN 2 D 3 
n  N We = 
 σ 
n

∑d 3
i where:
D32 = 1
n
(3) ρ and σ are fluid density and surface tension
∑d
1
2
i
N and D are impeller rpm and diameter

56 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 OPERATING VARIABLES AFFECTING THE BUBBLE SIZE IN FORCED-AIR MECHANICAL FLOTATION MACHINES

and hence bubble diameter, below which a bubble will remain All of this work has maintained that a strong relationship
stable. It has since been shown (Martinez-Bazan, Montanes and (exponent 0.6) exists between the D32 and the surface tension at
Lasheras, 2000) that the D32 is directly related to this critical the liquid gas interface. More recent workers have disputed that
diameter. Hinze established the following relationship for D32: this is the case in systems were a chemical environment exists
that strongly opposes bubble coalescence after initial formation
  (Parthasarathy, Jameson and Ahmed, 1991). Machon, Pacek and
  Nienow (1997) concluded that there was no correlation between
σ 0.6
D32 ∝   (6) surface tension and bubble size based on tests in an aerated
   0.4  stirred reactor for aqueous electrolyte and alcohol solutions, and
P
   ρ 0.2  that the critical Weber number approach to the analysis of these
  VL   systems was inappropriate. In later work by Alves et al (2002)
they concluded that there is a clear difference between
where: coalescing and non-coalescing systems (addition of electrolyte
and surfactant), and that the non-coalescing system exhibited a
 P represents the specific energy dissipation rate and relates
  D32 dependence on P/VL having an exponent from -0.52 to -0.37,
 VL  to the turbulence in the system
depending on location relative to the turbine impeller. This
agrees roughly with the theoretically derived exponent of -0.4 of
P represents the power input Equation 6. They also found a similar exponent for coalescing
systems at very low Jg values (0.2 cm/s) and concluded that the
VL is typically taken as the volume of liquid, but,
addition of surfactant or electrolyte did not alter the formation
alternatively, could be the volume swept by the impeller
size of bubbles, only the subsequent coalescence process. They
The analysis was further refined by Calderbank (1958) who also concluded that the effect of P/V (ie turbulence) on bubble
incorporated gas holdup and viscosity into Equation 6 to give: size is significantly lower for the non-coalescing systems.
Furthermore, they could find no relationship between bubble size
  and gas rate for the non-coalescing system (due to the
  0.25
dominating influence of electrolyte/surfactant); however they did
σ  ε α µG 
0.6

D32 ∝
   ( )   (7) find a good correlation between gas rate (Q/D2) and D32 for
µL 
0.4 G

  P  ρ 0.2  coalescing systems.

  VL  
where:
εG is the gas holdup
µL, µG are the liquid and gas viscosities respectively
The fitted α parameter was found to be 0.65 for an aliphatic
alcohol and 0.4 for an electrolyte solution. Gas holdup, ε G, has
been found to be linearly related to the superficial gas velocity, Jg
(over a limited range, which often defines the operating range of
a flotation cell), both in columns (Finch and Dobby, 1990) and
mechanical flotation machines (Dahlke, Gomez and Finch, 2005;
Nesset et al, 2006), thus Equation 7 does indirectly link D32 to
the superficial gas velocity.
The analogous parameter in aerated stirred tank reactors to the
Sb parameter in flotation (Equation 4) is the specific interfacial
area, a, the surface area of gas per unit volume of the occupied
liquid volume in the vessel, given by:
a = 6 ε G / D32 (8)
Overall, these studies on aerated stirred tank reactors provide
some valuable information about the importance of electrolyte
and surfactant additions in preventing bubble coalescence and
thereby strongly influencing bubble size. They have also served
to point out that, once sufficient surfactant or electrolytes are in
the aqueous system, surface tension is no longer the driving force
behind stabilising the bubble. Many studies have shown the
importance of the power intensity, P/VL, and to a lesser extent,
liquid viscosity and gas density. In comparison to flotation
machines, these aerated stirred tanks operate at lower superficial
gas velocities (typically well below 1 cm/s) and with lower
power intensity (typically below 1 kW/m3). Given the absence of
stators surrounding the impellers in these reactors, and given the
narrow region of high shear stresses in the impeller/stator gap of
flotation machines, the direct applicability of these models to
flotation machines must not be assumed. Al Taweel and Cheng
(1995) noted this in test work on a laboratory Denver cell and
concluded that large differences in energy dissipation rates exist
within the impeller/stator gap compared with the remainder of
the tank volume.

In an effort to avoid issues of bubble coalescence that had been Modelling D32 in flotation machines
a factor in many of the earlier studies, Sridhar and Potter (1980) In his classic review of the early work on flotation machine
found that in the cyclohexane-air system, across a wide range scale-up, Harris (1976) noted that frother had a greater influence
of pressure, temperature and surface tension, the following on bubble size than any of the machine variables under
relationship could be fitted: manipulation. At the time there was no adequate method for
-1 measurement of bubble size in industrial machines and the
  comment went largely unnoticed. With the advent of flotation
  0.5 0.16 column technology in the 1970s and 80s, much work was done
σ   Jg   ρ G   Po 
0.6
a∝
  U   ρ   P 
    (9) by the group at McGill and the University of Toronto (Dobby
   0.4
P
  ρ  0.2 b o and Finch, 1986) on characterisation of gas-sparged flotation
  VL   columns. The method of establishing bubble size was by drift
flux analysis, an inferred method that relies on measurement of
gas holdup to calculate the bubble size (Banisi and Finch, 1994).
Note the introduction of the terms Ub (bubble rise velocity), ρ o In the water-air system with frother added, Finch and Dobby
(air density under standard conditions) and Po (total power input)
(1990) proposed an equation for bubble size as a function of
in order to maintain dimensional coherence for the relationship.
If one assumes ε G ∝ Jg (as referenced previously) then Equation superficial gas velocity, namely:
9 can be inverted to give an expression for D32 using Equation 8.
Db = CJg
n
In this case, D32 becomes proportional to Jg0.5 and (ρ ο / ρ G )0.16. (10)

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 57

J E NESSET, J A FINCH and C O GOMEZ
The exponent n was found to be 0.25 for most porous sparger
types, while the C parameter depends primarily on frother
concentration and also corrects for the effective active area of the
sparger. At the University of Cape Town a photometric device for
measuring bubble size by extracting slurry samples through a
narrow capillary tube (the UCT bubble size analyser, Randall et
al, 1989) was used to measure bubble size in a laboratory column
as a function of different variables including air rate, temperature,
viscosity, particle size and pulp density (O’Connor, Randall and
Goodall, 1990). The data were fitted to power-law relationships
similar to Equation 10. The exponent, n, for bubble size versus
gas rate was 0.4 and 0.47 respectively for air-water and
air-water-solids, and n = 0.26 for the effect of viscosity in the
air-water system. These tests were in the absence of frother. In
another series of tests, the authors note (same reference) the
overwhelming effect of frother on bubble size, and that various
xanthate collectors, alone or in combination with frother, had no
additional effect on bubble size.
Several workers attempted to model the hydrodynamic
performance of particular laboratory flotation machines
including Sawant et al (1981), Nonaka, Inoue and Imaizumi
(1982), O’Connor, Randall and Goodall (1990), and Al Taweel
and Cheng (1995). They all found that the power intensity had a
measurable relationship with either gas holdup or the measured
bubble size (O’Connor). It is suspected that these small
laboratory machines do not scale well to the larger machines
since measurements are made in very close proximity (a few
centimetres) to the impeller/stator region, with its high
recirculation rate of both fluid and gas back into the high shear
region, compared to larger industrial units where the fluid flows
and bubbles leave the impeller/stator region. Hernandez-Aguilar,
Gomez and Finch (2002) have shown the sensitivity between
sampling location and measured bubble size distribution in a
Denver laboratory machine.
The first attempt at modelling the multiple effects of process
and flotation machine variables in industrial mechanical
forced-air machines using direct measurement of bubble size (the
UCT method) in plant environments was performed by Gorain,
Franzidis and Manlapig (1997). Their work related the surface
area flux (Sb) to the variables tested on a pilot flotation machine
with interchangeable rotor/stator parts. The resulting model
showed (Gorain, Franzidis and Manlapig, 1999):
Sb ∝ Sb (Q / A) As P80
c d e
(11)
where:
b, c, d, e are fitted constants
S is the peripheral impeller speed
Q is the volumetric air flow rate
A is the cell cross-sectional area
As is the aspect ratio of impeller diameter to height
P80 is the 80 per cent passing size of the solids in the cell
Their data fitting resulted in an R2 of 0.81 and yielded the
following parameters, b = 0.44, c = 0.75, d = -0.10 and e = -0.43.
Examination of Equation 11 reveals that the Jg measured to
yield Sb (left side) and Q/A (right side) are essentially the same
variable, measured independently, of air rate per unit of surface
cross-sectional area of the cell. By substituting Jg for Q/A, and
Equation 4 for Sb, Equation 11 can be rearranged to solve for D32
giving:
 J0.25 
D32 ∝  0.44 0.1 −0.43 
g
(12)
 S As P80 
58 Fremantle, WA, 19 - 21 March 2007
Gorain, Franzidis and Manlapig (1999) recognised that frother
concentration plays the largest role in bubble size, although it
does not appear in the equation. They concluded that given the
good correlation of the data fitting, the frother concentration
must have been sufficiently high and the variation sufficiently
little in these plants so as not to greatly influence the measured
bubble size. Although not specifically designed as a model of
bubble size, the re-arranged Gorain relationship is the only one
that predicts D32 for air rate and key variables in mechanical
flotation machines. Note that P80 can be considered a measure of
pulp viscosity and S relates directly to power intensity.
The importance of considering frother in the bubble generation
process is evident from the observation of Harris and has been
demonstrated and commented on by several researchers recently
(Cho and Laskowski; 2002; Laskowski; 2003; Gélinas and Finch,
2005; Nesset et al, 2006). The notion that frother addition
reduces surface tension and hence the bubble size is an erroneous
one that still prevails (Gupta and Yan, 2006). One current
mechanism considers the effect of frother (and electrolyte) on
ordering the local water structure, which increases resistance to
water drainage and hence retards coalescence (Finch, Gélinas
and Moyo, 2006).
The review of the literature has pointed out that there is
currently no specific model that incorporates the key
manipulated and process variables for the prediction of bubble
size in forced-air, mechanical flotation machines. At best, there
are models from aerated stirred reactor design that can give an
indication; however, these systems differ in some important
aspects. In addition to the absence of stators, they operate at
different levels of gas rate and power intensity and (typically)
without the addition of surfactants. This work attempts to address
this shortcoming.
EQUIPMENT AND EXPERIMENTAL PROCEDURE
The Metso Minerals 0.8 m3 RCSTM pilot flotation cell was
selected as it provided for power measurement and variable
speed drive as well as being sufficiently large to accommodate
mounting of the McGill gas dispersion sensors. The installed unit
is shown in Figure 4, including the inset that shows the immersed
gas rate and gas holdup sensors. The arrangement is schematically
illustrated in Figure 5. Note the positioning of all sensors at the
same cross-sectional reference plane. The testing protocol calls
for all measurements to be taken at a location sufficiently above
the turbulent impeller/stator region (25 cm above top of stator)
but below the froth interface (20 cm below), thereby representing
conditions of the bubble flux as it is about to exit the liquid
(pulp) phase. All three sensors were positioned symmetrically on
the radius (15.7 cm from the wall representing 73 per cent of the
distance from the centre of the cell) at the same elevation
GAS HOLD -UP
BUBBLE VIEWER
GAS RATE
FIG 4 - Metso pilot cell showing position of McGill bubble viewer
and, inset, the gas rate (Jg) and gas hold-up (εg) sensors.
Ninth Mill Operators’ Conference
 OPERATING VARIABLES AFFECTING THE BUBBLE SIZE IN FORCED-AIR MECHANICAL FLOTATION MACHINES
Air flow
FCV
RPM
Torque
Bubble size
P Pentanol
Froth phase
Gas
velocity
X-sectional area
reference plane
Gas holdup
Water phase
•frother
Impeller Stator
FIG 5 - Schematic of the Metso pilot cell, manipulated variables
and sensor arrangement.
(50 cm) from the cell bottom, a location that had been confirmed
through test work as giving consistent Jg and Db values.
The details of the instruments and measurement techniques
have been well described elsewhere (Gomez and Finch, 2002;
Hernandez-Aguilar, 2005; Gomez et al, 2006) and will be
reviewed only briefly here. The measurement of Jg relies on
sensing volumetric change in a closed tube through a pressure
sensor thus giving a slope of pressure versus time. It was
established through testing that 15 repeat measurements of slope
were sufficient to yield steady mean and variance values for Jg
(relative standard deviation 2.4 per cent). Bubble size distribution
is measured using a sampling tube and visualisation chamber
with backlighting from which a video camera captures bubble
images. Typically, 10 000 individual bubble images are
processed by the commercial software (Northern EclipseTM,
Empix Imaging) and deposited into an Excel file in order to
generate the distributions. The gas holdup (ε g) sensor is
comprised of two tubes that measure conductivity. One tube is
open and reports conductivity for the aerated liquid while the
second tube is partly restricted at the bottom and reports
conductivity for gas-free liquid; the ratio of the readings permits
the void fraction to be calculated. Both Jg and ε g measurement
data were captured using commercial software (iFixTM, GE
Fanuc Automation).
The Metso cell has a nominal capacity of 800 litres; however,
a standard filling volume of 700 litres was established. Montreal
tap water was used and, except for the temperature tests, the cell
was filled the day before a test to allow the water to equilibrate to
room temperature, typically about 16 - 18°C. Air was controlled
via a rotameter valve (rough adjustment) on the supply line and
fine adjusted via the Jg measurements to achieve the desired set
point. Bubble size, Jg and ε g measurements were taken
simultaneously. The tests conducted to establish the effects of
gas density were run on a standard Denver laboratory flotation
machine with mass flow meters and controllers on the air and
helium lines feeding the cell. It should be noted that laboratory
tests were conducted on the two-phase water-gas system only,
while the reported plant results are for three phase water-solids-
air.
Table 1 lists the five frother types and suppliers. These
represent four of the major chemical groups of frothers. Pentanol
was added to the list of four commercially available products to
extend the range of molecular weight and carbon-chain length.
(Plant surveys had also indicated that short chain alcohols can
enter as impurities in commercial xanthate supplies (Gélinas and
Finch, 2005)). The standard frothers used for the tests
investigating gas rate were Dowfroth-250 (D-250) and methyl
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
TABLE 1
Five frothers used in the laboratory testing.
Frother Description Supplier Mol weight
Simple alcohol Fisher 88
MIBC Methyl isobutyl carbinol Dow 102
D-250 Polypropylene glycol Dow 235 - 265
alkyl ether
F-140 Mixture aldehydes Flottec Mixture of C8 -
and ketones C22, typical
200 - 250
F-150 Polypropylene glycol Flottec 410 - 440
isobutyl carbinol (MIBC), while D-250 was the standard frother
for the tests on impeller speed, water temperature and gas
density. The frother tests were conducted at two levels of
superficial gas velocity, Jg = 0.5 and 1 cm/s. Frother additions
were incrementally and sequentially made in order to increase
the concentration levels tested. For each new addition level, ten
minutes of agitation in the cell without air preceded the
measurements. This was found to be critical in order to ensure
complete mixing as the MIBC and F-140, in particular, did not
dissolve readily.
The tests for varying impeller speed were designed to establish
the effect of power intensity on bubble size. It was confirmed
through testing that power (P) increase on the Metso cell did
indeed follow the expected cubic power law relationship as a
function of impeller speed (N) (ie P ∝ N3, non-aerated and
aerated). A doubling of impeller speed was possible on the unit,
thereby covering impeller tip speeds from 4.6 to 9.2 m/s,
providing an almost eight-fold increase in power intensity.
Typical impeller operating tip speeds on industrial machines are
from 5 - 7 m/s. The impeller speed tests were run at two levels of
Jg, 0.5 and 1 cm/s.
Varying gas density tests were included in an effort to simulate
elevation since many plants are located well above 3000 m, some
approaching 5000 m in Bolivia and Peru. Mixing air and helium
(He) proved a convenient means of obtaining gas mixtures with
the desired density representing different altitudes. Individual
mass flow meters and controllers for air and He were used to
feed the mixture through an inline mixer before being introduced
to the Denver flotation machine. Mixtures between zero per cent
and 100 per cent He were tested; however, only a mixture of 60
per cent He is required to simulate an altitude of 5000 m, where
air density is about half that at sea level. Only data for Jg values
of 0.25 and 0.3 cm/s were used, as the ability of the Denver
machine to fully disperse the gas at higher rates (above 0.5 cm/s)
was suspect. Other test conditions were D-250 at 5 ppm and
impeller speed of 1600 rpm (5.9 m/s tip speed).
The tests to establish the effect of water temperature served a
dual purpose as the range of water temperature (3 - 41°C)
introduced a correspondingly significant change in the (dynamic)
water viscosity (1619 to 641 µPa-s). The changes in both water
density (999.9 to 991.8 kg/m3) and surface tension (.07522 to
.06943 N/m) over this range are very small in comparison with
viscosity changes, so the test became, de-facto, an investigation
of the effect of viscosity over a temperature range that is
representative of what flotation plants in many parts of the world
experience. These tests were run at a Jg of 1 cm/s (at 5 ppm of
D-250) and temperature was allowed to either increase (if
starting below room temperature) or decrease (if starting above
room temperature). Hot water was used for the warmer tests and
the naturally cold Montreal tap water of April/May was used for
the colder tests. In total, four series of tests were conducted and
the results combined.
59
J E NESSET, J A FINCH and C O GOMEZ

Pentanol
RESULTS 5
Measured
4 CCC95
Frother type and concentration Model
32.3

D32 (mm)
The results of increasing frother addition for the five types tested 3
at Jg = 0.5 cm/s are shown in Figure 6. It is evident that all
2
frother types exhibit a similar exponential decrease in D32 with
increasing concentration, appearing to reach a limiting value at 1
about 0.9 mm. An exponential decay model has therefore been
fitted to the data as shown in the figure. The trend lines obtained 0
for Jg = 1 cm/s (not shown) were almost identical. These curves 0 10 20 30 40 50
are very similar to those obtained by others (O’Connor, Randall FROTHER ADDITION (ppm)
and Goodall, 1990; Laskowski, 2003). Laskowski (2003)
introduced the concept of the critical coalescence concentration MIBC
(CCC), the concentration at which the limiting bubble size (ie the 5
size before any coalescence occurs) is reached. It is a useful 4
CCC95
Measured
concept, but difficult to identify on a decay curve, so an 10.4

D32 (mm)
exponential curve fit was used to identify a more useful measure, Model
3
the CCC95, the concentration for which 95 per cent of the
decrease in D32 has been achieved (Hernandez-Aguilar and 2
Finch, 2005). The data for all five frothers at the two Jg levels
was normalised using the individual CCC95 values identified for 1
each frother test. Figure 7 shows the normalised plot and the 0
fitted curve having the following expression: 0 10 20 30 40 50
FROTHER ADDITION (ppm)
D32 = D1 + A exp[− B(PPM / CCC95)] (13)
D-250
where: 5
CCC95
Dl is the limiting D32 as concentration → ∞ (Dl = 0.907 mm) 4 8.35 Measured
D32 (mm)

A is the difference between Dl and the D32 at 0 ppm frother 3 Model

(A = 3.07 mm)
2
B is the decay constant (B = 3.13)
1
Table 2 lists the individual CCC95 values obtained for the ten
test series. The CCC95 values for Jg = 0.5 cm/s are seen to be 0
somewhat lower than for the Jg = 1 cm/s values. The reason for 0 10 20 30 40 50
the apparent effect of Jg on CCC95 may be related to the wider FROTHER ADDITION (ppm)
overall bubble size distribution at the higher Jg (ie a reflection on
the fact that D32 is not a perfect measure of the overall bubble F - 140
size distribution). Other factors may also influence the CCC95, 5
perhaps decreasing frother concentration with time during the Measured
test, which would be more pronounced at the higher Jg. 4
CCC95 Model
D32 (mm)

3 7.66
TABLE 2
CCC95 values established for the five frother types at Jg 2
levels of 0.5 and 1 cm/s.
1
Frother CCC95 (ppm)
0
Jg = 0.5 Jg = 1 Average
0 10 20 30 40 50
Pentanol 25.59 34.03 29.81 FROTHER ADDITION (ppm)
MIBC 10.45 12.47 11.46
F-150
D-250 8.35 13.86 11.11 5
F-140 7.66 11.48 9.57 CCC95 Measured
4
F-150 3.74 5.26 4.50 3.74 Model
D32 (mm)

3
Superficial gas rate 2
These tests were conducted at five frother concentration levels
1
across a range of Jg values from 0.05 cm/s to 1.5 cm/s. The
selected frother concentrations corresponded to 0, 50, 75, 90 and 0
99 per cent of the CCC as determined from Equation 13. The 0 10 20 30 40 50
resulting data are presented in Figure 8 for both MIBC and FROTHER ADDITION (ppm)
D-250. The curves represent the least-squares best-fit to Equation
1 (D32 = Do + CJg ), the relationship observed by Nesset et al
n
FIG 6 - The effect of frother addition on D32 for the five frother
(2005) for flotation cells in plant environments. It is immediately types (Jg = 0.5 cm/s). The model curves define the CCC95
evident that, while the curves show very good agreement with assuming an exponential decay relationship for D32.

60 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 OPERATING VARIABLES AFFECTING THE BUBBLE SIZE IN FORCED-AIR MECHANICAL FLOTATION MACHINES

5 Impeller speed
4
Jg = 0.5 Impeller rotational speed on the Metso cell could readily be
Jg = 1 varied across a broad range (1188 to 2376 rpm) corresponding to
impeller tip speeds of 4.6 - 9.2 m/s, a range that covers the
D32 (mm)

3 Model
operational range of mechanical flotation cells as reported by
2 Harris (1976); Fallenius (1976); Arbiter (1999) and Deglon,
Egya-Mensah and Franzidis (2000). Typical impeller tip speeds
1 fall in the lower range of 5 - 7 m/s.
Figure 9 shows the effect of increasing impeller tip speed on
0 D32 for both varying frother concentration (0, 2.5, 5 and 10 ppm
0 2 4 6 8 10 12 D-250) and Jg (0.5 and 1 cm/s). The data show that D32 is largely
NORMALISED FROTHER ADDITION (ppm/CCC95) unaffected across this doubling of the tip speed, which
5 effectively increases the power intensity by a factor of eight. This
was an unexpected finding based on the previous studies on
4 Jg = 0.5 aerated stirred tanks and laboratory flotation cells (Al Taweel
and Cheng, 1995; Parthasarathy Jameson and Ahmed, 1991). At
D32 (mm)

Jg = 1
3 the lower end, 4 m/s is close to the operational limit of the cell.
Model
Careful examination of the figure does show a very slight
2 increase in D32 as impeller speed approaches the 4 m/s limit;
however, the effect is judged to be so small as not to be
1
operationally significant. Since D32 is simply one measure of the
0
size distribution, it is prudent to check on possible changes to the
overall bubble size distribution. Figure 10 shows the full
0 0.5 1 1.5 2
(number) frequency distributions for each of the frother
NORMALISED FROTHER ADDITION (ppm/CCC95) concentrations (at Jg = 0.5 cm/s). It can be seen that the overall
distributions remain very similar, although it is noted that for the
FIG 7 - The normalised curve for frother addition (ppm/CCC95) lower tip speed (4.6 cm/s and 5.7 cm/s) the distributions are
versus D32 for all five frother types at both Jg = 0.5 and 1 cm/s. slightly different toward the smaller-size end of the distribution,
Top shows full x-axis scale while lower shows expanded scale. the portion which has a proportionally small impact on the D32.
The results are therefore felt to be conclusive, and illustrate what
may be an important difference between aerated stirred tanks
MIBC
10.0 without stators, and mechanical flotation machines (where a
CCC0 region of very high shear is created in the gap between impeller
CCC50
and stator).
CCC75
CCC90 5
D32 (mm)

CCC99
1.0 0 ppm 0 ppm Jg = 1
4
2.63 ppm 2.5 ppm Jg = 1
D32 (mm)

5.16 ppm 5 ppm Jg = 1

3
8.74 ppm 10 ppm Jg = 1
17.6 ppm 0 ppm Jg = 0.5
0.1 2 2.5 ppm Jg = 0.5
0 0.5 1 1.5 5 ppm Jg = 0.5
Jg (cm/s) 1 10 ppm Jg = 0.5

D-250 0
10.0
CCC0
4 6 8 10 12 14
CCC50
CCC75
IMPELLER TIP SPEED (m/s)
CCC90
D32 (mm )

1.0
0.1
0 0.5 1 1.5
Jg (cm/s)
FIG 8 - Jg versus D32 for MIBC and D-250, fitted model curves
and experimental data.
the data points, the Do values (the notional D32 creation size for
Jg = 0) are not a constant as postulated in the earlier work, but
decrease with increasing frother concentration. Similarly, the
parameter C is also dependent on the frother concentration. The
exponent n was found to be almost the same for both MIBC and
D-250 and a value of n = 0.5 was found suitable for both frother
CCC99 FIG 9 - D32 versus impeller rotational tip speed for increasing
0 ppm frother addition (0, 2.5, 5, 10 ppm D-250) and Jg values of 0.5
2.41 ppm and 1 cm/s. The gray arrow indicates typical cell operating
4.82 ppm range of 5 - 7 m/s.
8.01 ppm
16.02 ppm
Gas density
The approach that varying the gas density can be used to
simulate elevation relies on the assumption that the type of gas
(He mixed with air in this case) will not affect the bubble
creation process. Others studying aerated bio-reactors at
pressures from one to 20 atmospheres (0.1 to 2 MPa) have shown
this assumption to be valid (Wilkinson and van Dierendonck,
1990). The data presented in Figure 11 showing D32 versus the
gas density (ρg, ie air-helium mixture) relative to 100 per cent air
(ρo) at standard conditions (0.1 MPa, 25°C) was fitted to an
equation of the form:
[ ]
n

types. D32 ∝ ρ o ρ g (14)

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 61

J E NESSET, J A FINCH and C O GOMEZ

20 1.6

4.6 m/s Model

1.5
Frequency (%)

15 5.7 m/s
Test 1 Jg = 0.25
6.9 m/s
1.4

D32 (mm )
Test 2 Jg = 0.25
10 8.0 m/s 0 ppm
9.2 m/s Test 3 Jg = 0.3
1.3
Test 4 Jg = 0.3
5
1.2
0
1.1
0.1 1.0 10.0
Db (mm) C B A
1.0
25 0 0.2 0.4 0.6 0.8 1
4.6 m/s GAS DENSITY RELATIVE TO STANDARD CONDITIONS
20
Frequency (%)

5.7 m/s
15 6.9 m/s FIG 11 - D32 versus the gas density (ρg) relative to air at standard
8.0 m/s 2.5 ppm conditions (ρo) of one atmosphere and 25°C. Jg as indicated and
10 9.2 m/s 5 ppm D-250. A, B and C indicate density at elevations of 1000,
5 3000 and 5000 m respectively.

D32 ∝ [ η1 ηo ]
0.1 1.0 10.0 n
(15)
Db (mm)
30
4.6 m/s The value of n = 0.777 shows there is a strong dependence on
25 bubble size of the water viscosity. As noted earlier, the other
5.7 m/s
Frequency (%)

20 6.9 m/s properties of water (density and surface tension) vary little over
8.0 m/s this range and so are unlikely to account for the observed change
15 9.2 m/s
5 ppm in D32. The data also exhibit a small deviation from the modelled
10 curve shown in Figure 12, with D32 flattening somewhat in the
range 10 - 20°C, before increasing again at low temperatures. It
5 is not known if this is simply an artifact of the tests or a real
0 phenomenon of viscosity effect in this region. Across the overall
0.1 1.0 10.0 range the data fit was quite acceptable having an R2 of 0.92. The
D b (mm) effect on diameter of varying gas/air temperature within the
70 bubble volume has been raised; however, the effect on diameter
is likely very small (less than five per cent) over the 40-degree
60
4.6 m/s range based on the ideal gas law. Figure 12 shows that the
Frequency (%)

50 measured D32 doubles in value across this temperature span so

5.7 m/s
40 6.9 m/s volumetric changes affected by the temperature change would
8.0 m/s
10 ppm clearly be negligible.
30
9.2 m/s
20 3.0 1900

10 4 Test Series

VISCOSITY (uPa-s)
0 2.5 Water viscosity
D32 (mm )

1400
0.1 1.0 10.0 D32 viscosity model
Db (mm) 2.0

900
FIG 10 - Bubble size (number) frequency distributions for the 1.5
data of Figure 9 at Jg = 0.5 cm/s. Increasing frother addition
as indicated from top to bottom.
1.0 400
0 10 20 30 40 50 60

The data for the four-test series (Jg of 0.25 and 0.3 cm/s) TEMPERATURE (Co )

yielded a best-fit value of n = 0.114, suggesting that the effect of

gas density is relatively small, less than ten per cent increase in FIG 12 - D32 and (dynamic) water viscosity versus temperature for
D32 at 5000 m compared to sea level. Jg = 1 cm/s and 5 ppm D-250.

Liquid temperature
Figure 12 shows both the standard curve of (dynamic) water
viscosity versus temperature (secondary y-axis) for 1 - 50°C, and
the measured D32 values (primary y-axis) for the four-test series
across the range of 3 - 41°C. Also plotted is the best-fit curve
relating D32 to the water viscosity (ηl) for a given temperature
relative to the viscosity at 20°C (ηo). The fitting yielded a value
of n = 0.777 for the following equation:
Comparison with plant data
Bubble size, D32 and D10
The opportunity presents itself for comparing the fitted models
for D32 versus superficial gas velocity and frother concentration,
with data collected from the plant surveys. The MIBC model
curves for Jg (Equation 1 having n = 0.5) for frother levels

62 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 OPERATING VARIABLES AFFECTING THE BUBBLE SIZE IN FORCED-AIR MECHANICAL FLOTATION MACHINES
relative to CCC of 0, 50, 75, 90 and 99 per cent are compared to
data from six plants in Figure 13. It is seen that the plant data
exhibit similar shaped curves to the laboratory generated models,
with all data falling within the boundaries of maximum and
minimum bubble size (CCC0 and CCC99). The model and the
plant data presented are for forced air mechanical flotation
machines, but even when data for self-aerating machines are
included in the plot (not shown) the fit within the model
boundaries is equally good. This agreement suggests that the
model based on a two-phase test is applicable to the three-phase
industrial data.
10.0
CCC0
CCC50
CCC75
CCC90
D32 (mm)
CCC99
1.0 Impala Bateman
Impala Metso
Raglan
Lac des Iles
WMC-LNO
Escondida-LC
0.1
0 0.5 1 1.5 2 bring performance of new plants up to design, and to continually
Jg (cm/s)
FIG 13 - Frother addition model (MIBC) versus plant data for D32
versus Jg. Model curves are indicated for frother levels of 0, 50,
75, 95 and 99 per cent of CCC.
Figure 13 can be used to indicate where a plant is operating
with respect to frother dosage. It was not possible to obtain
actual plant measurements of frother concentration during the
surveys; however, as a baseline the data for Raglan and WMC
can be assumed to approach the 100 per cent frother CCC level
since their pulps are highly saline solutions, between 25 and
120 g/L salt. It has been demonstrated that electrolyte solutions
exceeding an ionic strength of approximately 0.4 (equivalent to
23 g/L NaCl) act in an equivalent manner to frother
concentration at 100 per cent CCC (Quinn, Gomez and Finch,
2006). It can be seen from Figure 13 that the Raglan and WMC
data fall very close to the CCC99 model line. It is noted that all
of the data fall at concentrations at or above the CCC75 curve,
suggesting that none of the plants are operating at dramatically
reduced frother concentrations, but that there may be room for
some improvement for several. The effect of the pulp
temperature at the various plants has not been incorporated into
the data of Figure 13, and will be an obvious factor as
demonstrated earlier in this paper. This will account for some of
the differences between plants, but again, likely to be a
secondary effect as most plant pulps tested fall into the 20 -
35°C range.
Bubble surface area flux, Sb
The data presented in Figure 13 can be readily converted to Sb,
the bubble surface area flux, a parameter that links directly to
flotation kinetics (Equation 5). This is shown in Figure 14.
Again, the agreement with plant data is striking. The maximum
(CCC99) and minimum (CCC0) boundaries shown in Figure 14
serve to represent the true limits of Sb for these types of flotation
machines, since bubble size cannot be any larger than CCC0 nor
effectively smaller than CCC99. Figure 14 is therefore seen as a
useful tool for assessing the performance of a plant and for
indicating the opportunity for improving flotation kinetics
through improving the Sb. Clearly there is some room for
improvement of Sb at several of the operations.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
100 CCC0
CCC50
80 CCC75
CCC90
Sb (1/s)
60 CCC99
Impala Bateman
40 Impala Metso
Raglan
Lac des Iles
20 WMC-LNO
Escondida-LC
0
0 0.5 1 1.5 2
Jg (cm/s)
FIG 14 - Frother addition model (MIBC) versus plant data for D32
versus Sb. Model curves are indicated for frother levels of 0, 50,
75, 95 and 99 per cent of CCC.
DISCUSSION
The need for a competent bubble size model for flotation
machines goes beyond simply having an equation for predicting
bubble size. In today’s economic climate the need to rapidly
improve the performance of existing plants, is a matter of
survival. Robust models are required in all aspects from
equipment and process design, to plant optimisation, to
troubleshooting, benchmarking and seeking opportunities for
improvement. This work has sought to correct an observed
deficiency in the toolbox available to flotation process engineers.
The current literature offers models for related aerated stirred
tanks without stators, and while there are a few basic models that
have been developed for laboratory flotation machines, the need
to build an understanding of how both chemistry and machine
variables affect bubble size on full-size machines was clearly
evident. This work has taken some initial steps, starting with the
two-phase water-air system. That the plant data all falls within
the range of the model (Figures 13 and 14) suggests that the
effects of solids may be of secondary importance to the primary
variables of frother concentration (relative to the CCC) and
superficial gas velocity (Jg). Some effect of solids on bubble size
can be anticipated, either through the primary process of creation
or secondary processes such as coalescence, but it also seems
possible that a portion of this effect could be the way in which
particles, especially fines, alter the fluid viscosity. This work has
demonstrated that increasing water viscosity has a major impact
on D32 (Equation 15) so the way in which solids influence
viscosity may impact bubble size as well. Extending the test
range for viscosity effect on bubble size beyond that achievable
by simply changing water temperature would clearly be of
interest.
The strong influence of chemistry in determining bubble size
is revealed by Equation 13, showing the effect of frother
concentration (relative to the CCC), and also by the Do and C
parameters in the D32 - Jg relationship (of Equation 1), both of
which are closely linked to frother concentration. It would seem,
therefore, that the chemistry of the system is establishing both
the initial (creation size, Do) bubble size as well as the rate of
bubble size change with increasing gas rate. This may be
somewhat of a departure from current views that see frother as
inhibiting coalescence after initial bubble creation, but not the
creation size itself (Alves et al, 2002; Laskowski, 2003). Figure 8
shows the strong effect of frother concentration on the initial
bubble size created. In further support of this notion that the
frother is strongly influencing the initial bubble size, is the
finding that increasing power intensity by a factor of eight
(Figure 9) did not change the bubble size distribution in any
significant way. There is no doubt that such a marked increase in
power intensity has increased turbulence and shear but it has had
63
J E NESSET, J A FINCH and C O GOMEZ
no measurable impact on the bubble size created; that appears to
be a role that has been assumed by frother addition, even
relatively small concentrations.
The lack of effect of power intensity on bubble size in this
mechanical flotation system compared to the demonstrated effect
of P/V on D32 in aerated stirred tanks (Equations 6, 7, 8) points to
the important role of the stator as a difference between the two
systems. The possibility can be raised that as the rotational speed
is increased, and the pumping rate increases correspondingly,
that only bubbles having a certain size (or larger) have the
buoyancy required to escape the fluid streamlines. The
measurements were all made at a reference cross-sectional area
located below the froth interface and not directly in the region of
the impeller/stator. From a steady-state perspective, however, this
notion does not sit well, as it implies that the very small bubbles
must eventually coalesce in order to escape, and this is a process
known to have a small probability of occurrence in systems
containing frother. One is therefore left to conclude that the
measured effect of no change in bubble size is real since any
changes to the bubble size distribution would have been detected,
even at the reference plane above the turbulent region of the
rotor/stator.
Interestingly, the relationships for the effect of Jg and gas
density on D32 developed in the current work (Equations 1 and
14), having exponents (n) of 0.5 and 0.114 respectively, agree
quite well with the work of Snidhar and Potter (1980) for aerated
stirred tanks where they found exponents of 0.5 and 0.16
(Equation 9) for non-coalescing systems.
The comparison between the two-phase models and the plant
data (Figures 13 and 14) are surprisingly good given the known
differences between the systems as well as the fact that
comparing data between plant environments and machine types
and sizes does require some additional correction. For example,
reporting bubble size for a reference plane in one machine
requires a correction for the hydrostatic depth at which the
bubbles are created if a true comparison is to be made with
another machine, particularly if the machines have their
rotor/stators at different depths. The details are provided by
Gomez et al (2006) and have not yet been applied here. Other
relatively small corrections may also need to be applied to
account for such things as relative size of rotor to tank diameter
and effective cross-sectional area between machines. Such
corrections are relatively small and will not alter the observations
or conclusions arrived at here. It is also intended to combine the
various models presented (frother concentration, superficial gas
rate, gas density and liquid viscosity) into a single relationship.
This will be the subject of a future communication.
A powerful outcome of this work will be the ability for
plant engineers to benchmark their operations and compare
performance to the model data in a manner such as that presented
in Figures 13 and 14. This will provide a quantifiable measure of
the upside expected from improvements to cell hydrodynamics
through frother type and addition rate, Jg adjustment, or changes
to pulp viscosity. In spite of the fact that, to-date, there are very
few in-plant measurements of frother concentration to correlate
with the gas dispersion data, the situation will change with the
advent of a recently developed colorimetric analysis technique
(Gélinas and Finch, 2005). The data collected to date does
suggest, as indicated by Figures 13 and 14, that many plants tend
to operate at levels of 75 to 90 per cent of the CCC level. This
may not be surprising since higher levels, although advantageous
in the recovery-focused rougher and scavenger circuits, may be
disadvantageous in the selectivity-focused cleaner circuits where
the build-up of high frother concentrations can result in
uncontrollable frothing. Given the demonstrated importance of
frother chemistry to flotation, there is an obvious need for
development of new, multi-functional frothers or frother
combinations that can control both the bubble size and froth
stability functions of frothers. The work presented here should
help in the assessment of any such new developments.
64 Fremantle, WA, 19 - 21 March 2007
CONCLUSIONS
1. A set of models has been developed for predicting the
Sauter mean D32 bubble size in mechanical forced-air
flotation machines based on testing in the air-water system
using a 0.8 m3 Metso RCSTM pilot tank cell. Frother
concentration (relative to the CCC), superficial gas velocity
and viscosity were found to be the main variables, while
gas density had a secondary effect. Surprisingly, the
rotational speed of the impeller had no effect on bubble size
over the typical, full operating range of the flotation
machine.
2. The frother and gas rate models correlate well with a wide
selection of plant data collected worldwide. The ‘fit’ is
surprisingly good given the expected differences between
the two-phase model system and the three-phase plant
environment. It is postulated that the models can be used to
effectively benchmark plant operations and project the
potential upside for improving the hydrodynamic
performance.
3. The analysis suggests that frother concentration has a major
effect on initial bubble creation size. Combined with the
observation that power intensity (rotational speed) has little
or no effect on bubble size, the work has highlighted that a
significant difference exists in the bubble generation
process between aerated stirred tanks and flotation
machines in which a stator is present.
ACKNOWLEDGEMENTS
The authors would like to acknowledge the assistance of the
following individuals and organisations who contributed to this
work: the many graduate students and researchers of the McGill
gas dispersion/hydrodynamics team who participated in the
numerous plant surveys, and in particular, Michel Leroux, Wei
Zhang and Seng How Kuan who conducted the laboratory
testing. Particular thanks are due to SGS Lakefield Research and
NSERC for funding (JEN) and to Metso Minerals for supplying
the Metso RCSTM pilot unit. Thanks are also due to Frank
Cappuccitti of Flottec Inc, who continues to guide us through the
understanding of frothers, and has been an enthusiastic supporter
of McGill’s work in this area from the beginning. The ongoing
development of the gas dispersion technology would not be
possible without the continued support of NSERC and the
AMIRA P9 project. The final thanks goes to the operating sites
that support the work and permit us to disrupt their operations in
the expectation that they, as well as science, will ultimately
benefit.
REFERENCES
Al Taweel, A M and Cheng, Y H, 1995. Effect of surface tension on
gas/liquid contacting in a mechanically-agitated tank with stator,
Trans Inst Chem Eng Research Design, 73(A6):654-660.
Alves, S S, Maia, C I, Vasconcelos, J M T and Serralheiro, A J, 2002.
Bubble size in aerated stirred tanks, Chem Eng J, 89:109-117.
Arbiter, N, 1999. Development and scale-up of large flotation cells, in
Advances in Flotation Technology (eds: B K Parekh and J D Miller),
pp 345-352 (The Society for Mining, Metallurgy and Exploration:
Littleton).
Bailey, M, Gomez, C O and Finch, J A, 2005. Development and
application of an image analysis method for wide bubble size
distributions, Minerals Engineering, 18:1214-1221.
Banisi, S and Finch, J A, 1994. Reconciliation of bubble size estimation
methods using drift flux analysis, Minerals Engineering, 7(12):
1555-1560.
Calderbank, P H, 1958. The interfacial area in gas-liquid contacting with
mechanical agitation, Trans Instrum Chem Eng, 36:443-463.
Ninth Mill Operators’ Conference
 OPERATING VARIABLES AFFECTING THE BUBBLE SIZE IN FORCED-AIR MECHANICAL FLOTATION MACHINES
Cho, Y S and Laskowski, J S, 2002. Effect of flotation frothers on bubble
size and foam stability, Int J Miner Process, 64:69-80.
Coleman, R G, Urtubia, H E and Alexander, D J, 2006. A comparison of
BHP-Billiton’s Minera Escondida concentrators, in Proceedings 38th
Annual Meeting of the Canadian Mineral Processors (division of
CIM), pp 349-370.
Dahlke, R, Gomez, C O and Finch, J A, 2005. Determining operating
range of a flotation cell from gas holdup vs gas rate, Minerals
Engineering, 18(9):977-980.
Deglon, D A, Egya-Mensah, D and Franzidis, J P, 2000. Review of
hydrodynamics and gas dispersion in flotation cells on South African
platinum concentrators, Minerals Engineering, 13(3):235-244.
Dobby, G S and Finch, J A, 1986. Particle collection in columns – Gas
rate and bubble size effects, Can Metall Quart, 25(1):9-13.
Fallenius, K, 1976. Outokumpu flotation machines, in AM Gaudin
Memorial Volume (ed: M C Fuerstenau), Vol 2, pp 838-861 (The
Society for Mining, Metallurgy and Exploration: Littleton).
Finch, J A and Dobby, G S, 1990. Column Flotation, 180 p (Pergamon
Press).
Finch, J A, Gélinas, S and Moyo, P, 2006. Frother-related research at
McGill University, Minerals Engineering, 19(6-8):726-733.
Gélinas, S and Finch, J A, 2005. Colorimetric determination of common
industrial frothers, Minerals Engineering, 18(2):263-266.
Gomez, C O, Acuña, C A, Girgin, E H and Finch, J A, 2006. Bubble
distributions in laboratory flotation machines: Measurement and
evolution with increasing gas velocity, in Proceedings International
Mineral Processing Congress, Istanbul, pp 579-586.
Gomez, C O and Finch, J A, 2002. Gas dispersion measurements in
flotation machines, CIM Bulletin, 95(1066):73-78.
Gorain, B K, Franzidis, J P and Manlapig, E V, 1997. Studies on impeller
type, impeller speed, and air flow rate in an industrial scale flotation
cell – Part 4: Effects of bubble surface area flux on flotation kinetics,
Minerals Engineering, 10(4):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,
Minerals Engineering, 12(3):309-322.
Gupta, A and Yan, D S, 2006. Mineral Processing Design and
Operations: An Introduction, p 559 (Elsevier: Amsterdam).
Harris, C C, 1976. Flotation machines, in AM Gaudin Memorial Volume,
(ed: MC Fuerstenau), Vol 2, pp 753-815 (The Society for Mining,
Metallurgy and Exploration: Littleton).
Heiskanen, K, 2000. On the relationship between flotation rate and
bubble surface area flux, Minerals Engineering, 13(2):141-149.
Hernandez, H, Gomez, C O and Finch, J A, 2003. Gas dispersion
and de-inking in a flotation column, Minerals Engineering,
16(6):739-744.
Hernandez-Aguilar, J R and Finch, J A, 2005. Validation of bubble sizes
obtained with incoherent imaging on a sloped viewing window,
Chemical Engineering Science, 60:3323-3336.
Hernandez-Aguilar, J R, Gomez, C O and Finch, J A, 2002. A technique
for the direct measurement of bubble size distribution in industrial
flotation cells, in Proceedings 34th Annual Meeting of the Canadian
Mineral Processors (division of CIM), pp 389-402.
Hernandez-Aguilar, J R, Thorpe, R and Martin, C J, 2006. Experiences in
using gas dispersion measurements to understand and modify
metallurgical performance, in Proceedings 38th Annual Meeting of
the Canadian Mineral Processors (division of CIM), pp 387-402.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
Hinze, J O, 1955. Fundamentals of the hydrodynamic mechanism of
splitting in dispersion processes, AIChE J, 1(3):289-295.
Kelly, E G and Spottiswood, D J, 1982. Introduction to Mineral
Processing, pp 27-29 (J Wiley and Sons).
Laskowski, J S, 2003. Fundamental properties of flotation frothers, in
Proceedings 22nd International Mineral Processing Congress, Cape
Town, pp 788-797.
Machon, V, Pacek, A W and Nienow, A W, 1997. Bubble sizes in
electrolyte and alcohol solutions in a turbulent stirred vessel, Trans
Instrum Chem Eng, 75(A):339-348.
Martinez-Bazan, C, Montanes, J L and Lasheras, J C, 2000. Bubble size
distribution resulting from the break-up of an air cavity injected into
a turbulent water jet, Physics of Fluids, 12(1):145-148.
Nesset, J E, Gomez, C O, Finch, J A, Hernandez-Aguilar, J R and DiFeo,
A, 2005. The use of gas dispersion measurements to improve
flotation performance, in Proceedings 37th Annual Meeting of the
Canadian Mineral Processors (division of CIM), pp 401-421.
Nesset, J E, Hernandez-Aguilar, J R, Acuña, C, Gomez, C O and Finch,
J A, 2006. Some gas dispersion characteristics of mechanical
flotation machines, Minerals Engineering, 19(6-8):807-815.
Nonaka, M, Inoue, T and Imaizumi, T, 1982. A microhydrodynamic
flotation model and its application to the flotation process, in
Proceedings 14th International Mineral processing Congress,
Toronto, Vol III, pp 9.1-0.19.
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 Pro, 28:139-149.
Parthasarathy, R, Jameson, G J and Ahmed, N M, 1991. Bubble break-up
in stirred vessels predicting the Sauter mean diameter, Trans I Chem
E, 69(A):295-301.
Pyecha, J, Sims, S, Lacouture, B, Hope, G and Stradling, A, 2006.
Valuation of a Microcell™ sparger in the Red Dog column flotation
cells, Minerals Engineering, 19(6-8):748-757.
Quinn, J, Gomez, C O and Finch, J A, 2006. Effect of salt type and
concentration on gas dispersion properties, in Interfacial Phenomena
in Fine Particle Technology, 6th UBC-McGill-UA International
Symposium on Fundamentals of Mineral Processing (eds: Z Xu and
Q Liu), pp 417-428 (Canadian Institute of Mining and Metallurgy:
Montreal).
Randall, E W, Goodall, C M, Fairlamb, P M, Dold, P L and O’Connor,
C T, 1989. A method for measuring the sizes of bubbles in two and
three-phase systems, Journal of Physics E: Scientific Instruments,
22:827-833.
Rickard, T A, 1916. The Flotation Process, 364 p (Mining and Scientific
Press: San Francisco).
Sawant, S B, Joshi, J B, Pangarkar, V G and Mhaskar, R D, 1981. Mass
transfer and hydrodynamic characteristics of the Denver type
flotation cells, Chem Eng J, 21:11-19.
Sridhar, T and Potter, O E, 1980, Interfacial areas in gas-liquid stirred
vessels, Chem Eng Sci, 35:683-691.
Wilkinson, P M and van Dierendonck, L, 1990. Pressure and gas density
effects on bubble break-up and gas hold-up in bubble columns, Chem
Eng Sci, 45:2309-2315.
65
Validation of the AMIRA P9 Flotation Model Using the Floatability
Characterisation Test Rig (FCTR)
R G Coleman1, J-P Franzidis2 and E V Manlapig2

ABSTRACT AMIRA P9 FLOTATION MODEL AND MODELLING

The Floatability Characterisation Test Rig (FCTR) is a self-contained, METHODOLOGY
highly instrumented, mobile pilot plant. It was developed as part of the
AMIRA P9 project, primarily to determine ore floatability characteristics Flotation model
and to develop and validate flotation plant modelling, scale-up and
simulation methodologies. The AMIRA P9 flotation model considers the kinetic response of
This paper describes the use of the FCTR in validating the AMIRA P9 a mineral in a flotation cell (as quantified by the first order rate
flotation model and modelling methodology. In 2001 the FCTR was constant, k) to arise from the ore characteristics (related to the
installed and operated in parallel with WMC’s (now BHP Billiton’s) floatability of the minerals in the ore) and the environment in
Kambalda Nickel Operation. The FCTR was operated using four different which flotation takes place (related to the design and operating
circuit configurations of increasing complexity. Metallurgical surveys, conditions in the flotation machine in which the ore is
cell characterisation measurements and batch flotation tests were processed). The environment in which flotation takes place is
performed around each circuit. This allowed the calibration of a number divided further into a pulp phase characteristic and froth phase
of flotation models using a combination of data from one, two, three or
characteristic.
four circuit configurations.
The calibrated flotation models were then validated using predictive The P9 model for the recovery of component i of mineral j
validation. To do this, data around a further five FCTR circuit (Ri,j) is given below (Savassi, 1998) for a perfectly mixed
configurations were collected (three of which are presented in this paper). continuously operated single cell. This equation includes the
The performance of each of the five circuit configurations was predicted effect of entrainment and water recovery to the concentrate.
using JKSimFloat V6.1 and the parameters from the various calibrated
flotation models. The predicted performance was compared to the Pi , j ⋅ S b ⋅ τ ⋅ R f ⋅ (1 − Rw ) + ENT ⋅ Rw
experimentally determined performance of each circuit. Ri , j = (1)
The validation of the AMIRA P9 model was shown to be largely (1 + P i, j )
⋅ S b ⋅ τ ⋅ R f ∗(1 − Rw ) + ENT ⋅ Rw
successful except in the case where the reagent addition (depressant) to
the validation circuit was different to that in the calibration circuits. The where:
results also provided information regarding the required complexity of the
calibration circuit and how many calibration circuits are required to Pi,j floatability of component i of mineral j
obtain an accurate prediction of circuit performance.
Sb bubble surface area flux (bubble surface area/time/cell
These results have increased the confidence in the AMIRA P9 flotation
modelling methodology and have provided important information on
cross-sectional area)
what data must be collected to allow the accurate prediction of plant Rf froth recovery (ratio of the mass of mineral reporting to
performance using the floatability component modelling approach. the concentrate to the mass of mineral reporting to the
pulp-froth interface)
INTRODUCTION τ residence time in the flotation cell
Over the past decade, the flotation research groups at the Julius ENT degree of entrainment (unsized basis)
Kruttschnitt Mineral Research Centre – JKMRC (Brisbane, Rw water recovery to the concentrate
Australia), University of Cape Town (Cape Town, South Africa)
and McGill University (Montreal, Canada) have been developing The total recovery of each mineral j is the cumulative product
a methodology for optimising flotation plants using actual plant of the recovery of each component i of the mineral and the
data together with a semi-empirical subprocess model. The proportion of each component i of the mineral in the feed to the
project is being conducted as part of the Australian Minerals flotation cell:
Industry Research Association (AMIRA) P9 Project, titled n
‘The Optimisation of Mineral Processes by Modelling and R j = ∑ mi , j ⋅ Ri , j (2)
Simulation’. i=1

The Floatability Characterisation Test Rig (FCTR) has been

developed as part of the AMIRA P9 Project, primarily to
determine ore floatability characteristics and to develop and
validate flotation plant modelling, simulation and scale-up
methodologies within that project. This paper describes the use
of the FCTR for the validation of the AMIRA P9 flotation model
and modelling methodology at WMC’s (now BHP Billiton’s)
Kambalda Nickel Operation (KNO).
1. Flotation Manager, JKTech Pty Ltd, The University of Queensland,
Isles Road, Indooroopilly Qld 4068. Email: r.coleman@jktech.com.au
2. Professor, Julius Kruttschnitt Mineral Research Centre, The
University of Queensland, Isles Road, Indooroopilly Qld 4068.
where:
mi,j proportion of component i of mineral j in the feed to the
flotation cell
Modelling methodology
The P9 flotation modelling methodology is presented in Figure 1
(Alexander et al, 2000). The methodology has been described
previously in the literature (Alexander and Morrison, 1998;
Alexander et al, 2000; Harris et al, 2002; Schwarz et al, 2006)
and applied successfully to existing flotation circuits (Alexander
and Wigley, 2003; Alexander, Bilney and Schwarz, 2005;
Schwarz and Kilgariff, 2005; Coleman, Urtubia and Alexander,
2006; Schwarz, Alexander and Coleman, 2006).

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 67

R G COLEMAN, J-P FRANZIDIS and E V MANLAPIG
FCTR at Steady State ((Chemistry/Grind)
(Chemistry/Grind)
Chemistry/Grind)
Entrainment True Flotation
ENT cell
m R wm Sb m Rf m ?m PjFCTR
cell cell
cell cell
cell cell
cell cell
cell
j ==mineral
mineral
m = cell
FCTR Model
Circuit
Validate model
Estimate or predict
New PLANT
New Simulation
circuit
Circuit
Add Cell/Bank
Change Operating Condition
FIG 1 - AMIRA P9 flotation modelling methodology
(after Alexander et al, 2000).
Floatability rates and mass fractions are estimated by using a
non-linear least squares fitting routine on survey data, laboratory
batch flotation test data and measurements or estimates of the
flotation model parameters. Once the flotation model has been
calibrated, it requires validation, which is typically achieved by
successfully predicting the response of the circuit to a change.
Once validated, the flotation model can be used to simulate
changes in circuit configuration or operating conditions.
Although the methodology has been applied to existing
flotation circuits, this paper presents the first detailed validation
of the model and modelling methodology.
THE FCTR
The FCTR is a compact flotation pilot plant able to treat up to
2 t/h of feed solids and consists of six 150 L rougher/scavenger
cells and six 50 L cleaner/recleaner cells. The unit is mobile,
easily installed at plants and large enough for the results obtained
from it to be relevant to full-scale operation. The plant is also
easily changeable into a wide range of circuit configurations. The
FCTR is an ideal vehicle for the production of modelling-quality
flotation data as it is highly instrumented, automatically
controlled, and has the facility for full stream sampling on every
stream. The original design and characteristics of the FCTR have
been presented previously (Rahal, Franzidis and Manlapig,
2000a, 2000b; Rahal et al, 2000; Coleman et al, 2003; Coleman,
Franzidis and Manlapig, 2004).
FCTR TEST WORK
The FCTR was installed and commissioned at KNO in
September 2001. KNO is located 50 km south of Kalgoorlie in
the goldfields of Western Australia. The first ore processing was
performed in 1967. In 1997 a new concentrator was built and
commissioned. Today, the plant receives ore from several mines
in the vicinity and treats approximately 1.3 million tonnes of ore
per year. Typical nickel recovery is 92 per cent at a head grade of
three per cent nickel to produce concentrate grades of
approximately 13 per cent nickel. Historical mineralogical
analyses have shown that the main sulfide minerals present in the
KNO ore are pentlandite, pyrrhotite and chalcopyrite. The
non-sulfide gangue is comprised mainly of talc and quartz.
A comprehensive test program was performed, which included
FCTR model calibration and predictive model validation.
Predictive validation tests are based on the comparison of model
predictions to independent plant data that has not been used in
the fitting of the model. A description of the model calibration
and validation test work is presented in the following sections.
68 Fremantle, WA, 19 - 21 March 2007
FCTR model calibration
Circuit configurations
Data was collected around four FCTR circuits of increasing
complexity. The four FCTR circuits are shown in Figure 2.
Circuit 1 is a basic flotation circuit consisting of a bank of six
rougher cells and a bank of six cleaner cells. The concentrate from
the rougher cells is pumped to the cleaner bank where the cleaner
concentrate is the final product. The tails from the rougher and
cleaner banks are discarded and there are no recycle streams.
Circuit 2 consists of a bank of six rougher cells, a bank of four
cleaner cells and a bank of two recleaner cells. The concentrates
from the cleaner cells are pumped to the recleaner bank and the
recleaners produce the final concentrate. The recleaner bank tails
are recycled to the head of the cleaner bank. The tails from the
rougher and cleaner banks are again discarded.
Circuit 3 is similar to Circuit 2 but is more complex as it
contains two recycle streams. The cleaner tails in Circuit 3 are
recycled to the head of the last two rougher cells (scavenger
bank). This is aimed at increasing the recovery of valuable
mineral although it is at the expense of an increased circulating
load, hence a reduced residence time for flotation.
Circuit 4 is similar to Circuit 3. However, in Circuit 4 the first
rougher concentrate is pumped directly to the recleaner bank
instead of to the cleaner bank. As the first rougher concentrate
contains a large fraction of the fast floating material, it can be
beneficial to remove this material from the cleaner bank and
clean it in the recleaner bank instead. Removing this material
from the cleaner bank increases the residence time available in
the cleaners to treat the slower floating material.
Flotation feed from the KNO circuit was used for all test work.
In each circuit, collector, frother and activator reagents were
added to the KNO flotation feed prior to being pumped to the
FCTR. Depressant (guar) was added to the cleaner and recleaner
feed streams to minimise unwanted talc flotation.
Metallurgical surveys
Metallurgical sampling was performed around each of the FCTR
circuits to determine the circuit performance and to establish the
mass balance for each circuit. This data was used to obtain the
flotation model parameters.
Approximately one hour (three residence times) after
steady-state operation had been achieved, samples of the feed,
concentrate and tails were taken from each of the cells. For the
feed and tails, full stream samples were obtained using three-way
valves on each cell. Concentrate samples were obtained from the
concentrate pump discharge. FCTR feed and cell concentrate flow
rates were also measured. The flotation control system provided
sufficient control of the circuit to allow sampling of the streams
without altering the performance of the circuit during sampling.
For each circuit, a total of three to five metallurgical surveys
were performed under identical operating conditions (as far as
possible). This allowed the repeatability of the survey results to be
established and quantified. This is important for estimating the
measurement errors for the mass balance and model regression.
All samples were assayed for Ni, Fe, MgO, S, Cu and SiO2.
The elemental assays were then converted to mineral assays
using an element-to-mineral conversion table.
Laboratory batch flotation tests
Batch flotation tests were performed simultaneously with the
FCTR surveys, on stream samples taken around specified nodes
in the circuit. The samples for batch flotation testing were taken
independently of the mass balance survey samples, and were
floated immediately to minimise any oxidation or ageing effects.
Ninth Mill Operators’ Conference
 VALIDATION OF THE AMIRA P9 FLOTATION MODEL USING THE FLOATABILITY CHARACTERISATION TEST RIG (FCTR)

CIRCUIT 1 CIRCUIT 2

FEED 6 x ROUGHER CELLS FEED 6 x ROUGHER CELLS

ROUGHER ROUGHER
TAILS TAILS

COMBINED
RO CONC DEPRESSANT

COMBINED DEPRESSANT
RO CONC
2x 4 x CLEANER CELLS
6 x CLEANER CELLS CLEANER
TAILS RECLNR CLEANER
CELLS TAILS
RECLNR
TAILS
COMBINED
DEPRESSANT CLEANER
FINAL CONC FINAL CONC CONC

CIRCUIT 3 CIRCUIT 4

FEED 2x FEED 1x 3x 2x
4 x ROUGHER ROUGHER ROUGHER ROUGHER
ROUGHER
CELLS CELLS ROUGHER CELL CELLS
CELLS
TAILS ROUGHER
TAILS

COMBINED ROUGHER 1
COMBINED
RO CONC CONC
DEPRESSANT ROUGHER DEPRESSANT
DEPRESSANT CONC
DEPRESSANT
2x 4 x CLEANER CELLS 2x 4 x CLEANER CELLS
RECLNR RECLNR
CELLS CLEANER CELLS CLEANER
RECLNR TAILS RECLNR TAILS
TAILS TAILS
COMBINED COMBINED
CLEANER CLEANER
FINAL CONC FINAL CONC
CONC CONC

FIG 2 - FCTR circuit configurations for flotation model calibration at KNO.

A bottom-driven JKMRC flotation cell with a volume of five
litres was used for all of the batch flotation tests, which were
performed under identical conditions. No reagents were added to
the samples before the tests were conducted. The cell was
operated using a shallow froth depth to minimise the effect of the
froth phase, and a froth recovery of 100 per cent was assumed in
all subsequent calculations. Concentrate was collected in
separate trays at various time intervals while maintaining the
constant operating conditions. After collecting the last sample,
the remaining sample in the batch cell (tails) was emptied into a
bucket and the mass of each tray and the tails sample were
measured.
Gas dispersion measurements
Gas dispersion characteristics (air hold-up, superficial gas
velocity and bubble size) were measured to characterise the
operation of the FCTR cells and to allow the estimation of
bubble surface area flux values for each operating cell.
The air hold-up was determined by isolating each cell
individually and measuring the difference in pulp level between
the cell operating normally and not operating. This was done by
simultaneously stopping the impeller, shutting off the air to the
cell, stopping the feed to the cell, and closing the tails valve. The
superficial gas velocity was calculated as the air flow rate in the
cell divided by the cell cross-sectional area. The bubble size was
measured using the UCT bubble size analyser (Tucker et al,
1994). Bubble size measurements were performed in roughers 1,
2, 5 and 6 and cleaners 1 and 6.
Froth recovery measurements
Froth recovery was measured using the mass balance method
(Alexander, Franzidis and Manlapig, 2003). This requires samples
of cell feed, concentrate, tail, a sample of the pulp below the
pulp-froth interface and from the froth surface. Froth recovery
was measured once in each FCTR cell for each circuit. The froth
recovery samples were taken at the same time as the first survey
around the FCTR.
FCTR model validation
Circuit configurations
For the predictive validation tests, data were collected around a
further five FCTR circuits. The results from three of the five
validation circuits will be discussed in this paper. The three
FCTR model validation circuits discussed in this paper are
shown in Figure 3.
Circuit 5 consists of a bank of six rougher cells, a bank of four
cleaner cells and a bank of two recleaner cells. The concentrates
from the cleaner cells are pumped to the recleaner bank and the
recleaners produce the final concentrate. The tails from the
rougher, cleaner and recleaner banks are discarded and there are
no recycle streams.
Circuit 6 is similar to Circuit 3 and consists of a bank of six
rougher cells, a bank of four cleaner cells and a bank of two
recleaner cells. The concentrates from the cleaner cells are
pumped to the recleaner bank and the recleaners produce the
final concentrate. The recleaner bank tails are recycled to the
head of the cleaner bank. The cleaner tails are recycled to the
feed of the third rougher cell. This is aimed at increasing the
recovery of valuable mineral although it is at the expense of an
increased circulating load, hence a reduced residence time for
flotation. The tails from the rougher bank are again discarded.
Circuit 7 consists of a bank of five rougher cells, a bank of six
cleaner cells and a single recleaner cell. In this circuit, the last of

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 69

R G COLEMAN, J-P FRANZIDIS and E V MANLAPIG

CIRCUIT 5 CIRCUIT 6

FEED FEED 22 xx
6 x ROUGHER CELLS ROUGHER
ROUGHER 44 xx ROUGHER CELLS
ROUGHER CELLS
CELLS
CELLS
ROUGHER ROUGHER
ROUGHER
TAILS TAILS
TAILS
COMBINED
RO CONC DEPRESSANT
COMBINED DEPRESSANT
DEPRESSANT
DEPRESSANT RO CONC

2x 44 xx CLEANER
CLEANER CELLS
CELLS 22 xx CLEANER CELLS
44 xx CLEANER CELLS
RECLNR RECLNR
CLEANER CLEANER
CLEANER
CELLS CELLS
CELLS
TAILS RECLNR TAILS
TAILS
RECLNR TAILS
COMBINED COMBINED
COMBINED
TAILS CLEANER
CLEANER CLEANER
FINAL CONC CONC FINAL CONC CONC
CONC

CIRCUIT 7

FEED
ROUGHER CELLS
5 x ROUGHER CELLS
ROUGHER
TAILS

COMBINED
DEPRESSANT
RO CONC
DEPRESSANT

1x 66 xx CLEANER
CLEANER CELLS
CELLS
RECLNR
CELL CLEANER
(ROUGHER) RECLNR TAILS
TAILS COMBINED
CLEANER
FINAL CONC CONC

FIG 3 - FCTR circuit configurations for FCTR flotation model validation at KNO.

the six rougher cells is operated as a recleaner cell. In all other
circuits using a recleaner bank, the effective volume of each
recleaner cell is 35 litres (70 litres total recleaner capacity for
two cells). In this case, as a larger rougher cell is used as a
recleaner cell, the effective volume of the recleaner cell is
125 litres. Therefore, the recleaning capacity in this circuit is
almost doubled. The residence time in the cleaner circuit is also
increased as the two cells used previously for recleaning
(Circuits 5 and 6) are used as additional cleaners.
In all the validation circuits depressant (guar) was added to the
cleaner and recleaner feed streams to minimise unwanted talc
flotation.
Metallurgical surveys
Metallurgical samples were taken around each of the FCTR
validation circuits to determine the circuit performance and to
establish the mass balance for each circuit. Three repeat
metallurgical surveys were performed for Circuit 5 and two
repeat surveys for Circuits 6 and 7. The samples were collected
and assayed using the same procedure as that used in the model
calibration surveys.
MODEL CALIBRATION RESULTS
The surveys around the calibration circuits were first mass
balanced. The ore floatability characteristics for each calibration
circuit were estimated using the mass balanced survey data in
conjunction with the laboratory batch flotation test results and
the measured model parameters (the degree of entrainment was
calculated using the technique described by Savassi (1998) using
the recovery to the concentrate of a liberated non-floating tracer
present in the ore). A number of flotation models were calibrated
using data from one, two, three or four calibration circuits.
From the estimated floatability characteristics it was found
that the sulfide minerals were best represented as consisting of
two floating fractions (fast and slow) and a non-floating fraction,
and the non-sulfide gangue was best represented as consisting of
a slow floating fraction and a non-floating fraction. Pentlandite
consisted mainly of fast-floating material and a small proportion
of slow-floating and non-floating material. Pyrrhotite consisted
mainly of slow-floating material. Chalcopyrite consisted mainly
of fast-floating material. Non-sulfide gangue contained a small
proportion of slow-floating material but was predominantly
non-floating.
Where data from two, three or four calibration circuits were
used to derive the ore floatability characteristics, it was assumed
that the feed ore floatability characteristics were the same for
each circuit. A single set of floatability characteristics was
successfully fitted to the combined survey and batch flotation
test data from each circuit. No significant differences in the
floatability characteristics were found with an increasing number
of calibration circuits.
MODEL VALIDATION RESULTS
The validation was performed using the feed floatability
characteristics determined from the various calibration models
discussed above, in conjunction with estimates of the model
parameters. JKSimFloat V6.1 was used to simulate the
performance of each validation circuit. A number of different
methods were used to predict the performance of the validation
circuits. The prediction results for two of the methods will be
discussed in this paper. In prediction method 1, the water flow to
the concentrate in each cell was predicted using an empirical
water recovery model based on the flow rate of solids to the
concentrate. In prediction method 2, experimental water flows
were used.

70 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 VALIDATION OF THE AMIRA P9 FLOTATION MODEL USING THE FLOATABILITY CHARACTERISATION TEST RIG (FCTR)

It was important to be able to compare the predictions of each Circuit 5

model to the experimental data and to each other; therefore a
standard measure of accuracy of the predictions was required.
Sets of theoretical data were produced for this purpose. Each
hypothetical data set consisted of data points that were relative
increments (such as ±5 per cent, ±10 per cent, ±15 per cent, etc)
of the experimental data. This was applied to all experimental
data points to produce the hypothetical data sets. Each
hypothetical data set was then compared to the experimental data
and the error was calculated for each. This produced a
relationship between the relative error at each data point and the
prediction error. The relative error at each data point could then
be calculated given the actual error from the predictive validation
based on the various models tested.
The predictive validation results for each circuit are presented
in the sections below.
Overall prediction
For Circuit 5, the average relative error at each data point ranged
from six per cent to 13 per cent for all predictions. The models
based on Circuit 3 produced the best predictions for both
prediction methods. This is possibly due to Circuit 3 having a
similar circuit configuration to Circuit 5 especially in terms of
the recleaner tails recycle.
Figures 4 and 5 show the comparisons between the
experimental and predicted mineral recoveries and grades for the
predictions that produced the highest average relative error per
data point and lowest average relative error per data point,
respectively. These figures also show the 95 per cent confidence
intervals calculated from experimental standard deviations.

100 100

90 90

80 80

70 70
Predicted Recovery (%)
Predicted Recovery (%)

60 60

50 50

40 40
High NiS Low NiS
30 30
High FeS Low FeS
20 20
High CuS Low CuS
10 10
High NSG Low NSG
0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
A Experimental Recovery (%) B Experimental Recovery (%)

FIG 4 - Comparison of experimental and predicted mineral recoveries for predictions producing the highest (A) and lowest (B) average
relative errors (Circuit 5).

100 100
High NiS Low NiS
90 90
High FeS Low FeS
80 80
High CuS Low CuS
70 70
High NSG Low NSG
Predicted Grade (%)
Predicted Grade (%)

60 60

50 50

40 40

30 30

20 20

10 10

0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100

A Experimental Grade (%) B Experimental Grade (%)

FIG 5 - Comparison of experimental and predicted mineral grades for predictions producing the highest (A) and lowest (B) average
relative errors (Circuit 5).

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 71

R G COLEMAN, J-P FRANZIDIS and E V MANLAPIG

Table 1 lists the gradients and correlation coefficients for the lower than the predictions based on the empirical water recovery
presented relationships. model (prediction method 1). There is also less than two per cent
The following observations can be made from these results: difference in the average relative error between predictions based
on experimental water recoveries using a single calibration
• The prediction of pentlandite and chalcopyrite recovery and circuit, and predictions based on the empirical water recovery
grade is accurate and within the experimental standard model using four calibration circuits. This suggests that with an
deviation for most streams in the circuit. improved water recovery model, fewer calibration circuits would
• The prediction of pyrrhotite recovery is accurate and within be required to achieve the same prediction accuracy.
the experimental standard deviation for most streams in the
circuit, although the pyrrhotite grade is over-predicted for Circuit 6
most of the streams in the circuit.
• There is a significant improvement in the prediction of Overall prediction
pyrrhotite grade in the results of the model producing the
lowest average relative errors. The average relative error at each data point ranged from nine
per cent to 45 per cent for all predictions and the simulator was
• The prediction of non-sulfide gangue recovery is accurate generally unable to predict the performance of Circuit 6
and within the experimental standard deviation for the accurately. The models based on Circuit 3 again produced the
majority of streams in the circuit. As the majority of the best predictions for both prediction methods. This is expected as
non-sulfide gangue reports to the concentrate via Circuit 3 is the closest representation of Circuit 6, in terms of
entrainment, the accurate prediction of non-sulfide gangue circuit configuration, of all of the four calibration circuits.
indicates that the method for estimating the degree of
The sources of error in the predictions were examined to
entrainment is suitable for the simulations.
determine the reasons for the generally poor prediction of circuit
performance. It was found that over 65 per cent of the errors in
TABLE 1 the prediction were due to the poor prediction of pyrrhotite
Gradients and correlation coefficients of predictions of Circuit 5 recovery in some streams. If the prediction of pyrrhotite recovery
performance with the highest and lowest average relative errors. was removed from the error analysis, the average relative error at
each data point would fall to between six per cent and 18 per
Measurement Highest average Lowest average cent.
relative error relative error One possible reason for the poor prediction of pyrrhotite is the
Gradient r2 Gradient r2 addition of guar during the operation of Circuit 6. The current
modelling methodology cannot account for the effect of reagent
Recovery Pentlandite 0.98 0.98 1.00 1.00
addition in the estimate of the ore floatability rate. Therefore, if
Pyrrhotite 0.99 0.97 1.00 0.98 the ore floatability parameters are determined at a specific
Chalcopyrite 0.99 0.99 1.00 1.00 reagent addition rate (or range of rates), a circuit that operates
NS gangue 1.03 0.99 1.01 1.00 outside this specific range of reagent addition rates may not be
accurately predicted. The ore floatability parameters determined
Grade Pentlandite 1.08 0.96 1.04 0.99 from Circuits 1 to 4 were calibrated for guar addition rates
Pyrrhotite 1.18 0.88 1.05 0.92 between 30 to 65 g/t (average of 46 g/t) to the cleaner feed
Chalcopyrite 1.05 0.98 1.03 1.00 stream and 25 to 36 g/t to the recleaner feed stream (note: guar
addition rates based on the flow rate of the cleaner and recleaner
NS gangue 0.94 0.90 0.99 0.98
feed streams). The guar addition rates to the cleaner and
recleaner feed streams in Circuit 6 were 25 g/t and 26 g/t,
Comparison of models and prediction methods respectively. Therefore, the guar addition rate to the cleaner feed
stream in Circuit 6 was approximately 46 per cent lower than the
Table 2 shows the average relative errors based on the number of average addition rate to Circuits 1 to 4.
calibration circuits and prediction method used for all models Although guar is added primarily to depress talc from the
tested. concentrate, iron sulfide (pyrrhotite) would also be affected.
Nickel and copper sulfide would also be affected but to a much
TABLE 2 lesser extent. It is therefore possible that the poor prediction of
Average relative errors for the prediction of Circuit 5 performance. pyrrhotite in Circuit 6 is due to the different guar addition rates
used in the calibration of the ore floatability parameters
No of calibration Average relative error (%) compared to those used in the prediction circuit.
circuits Prediction method 1 Prediction method 2 Figures 6 and 7 show the comparisons between the
experimental and predicted mineral recoveries and grades for the
1 11.1 10.0 predictions that produced the highest average relative error per
2 9.1 8.3 data point and lowest average relative error per data point. Table 3
3 8.5 7.9 lists the gradients and correlation coefficients for the presented
relationships.
4 8.1 7.6
The following observations can be made from these results:
The prediction results show that the average relative error • There is a significant difference in the prediction of
decreases as the number of calibration circuits increases. pentlandite recovery between the models producing the
However, the maximum difference in the average relative error highest and lowest average relative errors. The prediction of
between a single calibration circuit and four calibration circuits pentlandite recovery is accurate and within the experimental
is less than three per cent. This improvement in the accuracy of standard deviation for all streams in the circuit for the model
prediction must be weighed up against the increased test work producing the lowest average relative errors.
that must be performed to calibrate four circuits. • The prediction of pentlandite grade is accurate for both the
On average, the relative error based on experimental water models producing the highest and lowest average relative
recoveries (prediction method 2) was approximately one per cent errors.

72 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 VALIDATION OF THE AMIRA P9 FLOTATION MODEL USING THE FLOATABILITY CHARACTERISATION TEST RIG (FCTR)

100 100
High NiS Low NiS
90 90
High FeS Low FeS
80 80
High CuS Low CuS
70 70

Predicted Recovery (%)

Predicted Recovery (%)

High NSG Low NSG

60 60

50 50

40 40

30 30

20 20

10 10

0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
A Experimental Recovery (%) B Experimental Recovery (%)

FIG 6 - Comparison of experimental and predicted mineral recoveries for predictions producing the highest (A) and lowest (B)
average relative errors (Circuit 6).

100 100

High NiS Low NiS

90 90
High FeS Low FeS
80 80
High CuS Low CuS
70 70
Low NSG
Predicted Grade (%)

High NSG
Predicted Grade (%)

60 60

50 50

40 40

30 30

20 20

10 10

0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
A Experimental Grade (%) B Experimental Grade (%)

FIG 7 - Comparison of experimental and predicted mineral grades for predictions producing the highest (A) and lowest (B) average
relative errors (Circuit 6).

• The prediction of chalcopyrite recovery and grade is accurate
and within the experimental standard deviation for most
streams in the circuit.
• The prediction of pyrrhotite recovery is poor (gradient = 0.44)
for the model producing the highest average relative errors
with the majority of points outside the experimental standard
deviation. For the model producing the lowest average relative
errors the prediction of pyrrhotite recovery is accurate and
within the experimental standard deviation for all streams in
the circuit. In the case of the model producing the highest
average relative errors, the poor prediction of pyrrhotite
recovery may be related to the addition of guar as discussed
previously, or to the under-prediction of water recovery.
• The pyrrhotite grade is under-predicted for the model
producing the highest average relative errors.
• Non-sulfide gangue recovery is under-predicted for the
model that gives the highest average relative errors and
accurate for the model the gives the lowest average relative
errors. The non-sulfide gangue grade is over-predicted for the
model producing the highest average relative errors (gradient
= 1.07). The over-prediction in non-sulfide gangue grade is
linked with the under-prediction of the floatable minerals
grades.
Comparison of models and prediction methods
Table 4 shows the average relative errors based on the number of
calibration circuits and prediction method used for all models
tested.
The prediction results show that the average relative error
decreases as the number of calibration circuits increases.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 73

R G COLEMAN, J-P FRANZIDIS and E V MANLAPIG

TABLE 3 approximately five per cent lower than predictions based on the
empirical water recovery model using four calibration circuits.
Gradients and correlation coefficients of predictions of Circuit 6
performance with the highest and lowest average relative errors. This further indicates that with an improved water recovery
model, fewer calibration circuits would be required to achieve
Measurement Highest average Lowest average the same (or better) prediction accuracy.
relative error relative error
Gradient r2 Gradient r2 Circuit 7
Recovery Pentlandite 0.84 0.80 0.98 0.98
Pyrrhotite 0.44 0.19 1.04 0.98 Overall prediction
Chalcopyrite 0.99 0.93 1.02 0.97 The average relative error at each data point ranged from five per
NS gangue 0.87 0.94 1.00 1.00 cent to 16 per cent for all predictions. The models based on
Circuit 2 produced the best predictions for both prediction
Grade Pentlandite 0.90 0.94 0.94 0.98
methods. This is expected as Circuit 2 is the closest
Pyrrhotite 0.90 0.70 1.00 0.93 representation of Circuit 7, in terms of circuit configuration, of
Chalcopyrite 0.94 0.96 0.99 0.98 all of the four calibration circuits.
NS gangue 1.07 0.94 1.01 0.98 Figures 8 and 9 show the comparisons between the
experimental and predicted mineral recoveries and grades for the
predictions that produced the highest average relative error per
TABLE 4 data point and lowest average relative error per data point. Table 5
Average relative errors for the prediction of Circuit 6 performance. lists the gradients and correlation coefficients for the presented
relationships.
No of calibration Average relative error (%)
circuits Prediction method 1 Prediction method 2 TABLE 5
1 22.4 17.5 Gradients and correlation coefficients of predictions of Circuit 7
2 22.2 16.0 performance with the highest and lowest average relative errors.

3 22.1 15.2
Measurement Highest average Lowest average
4 22.2 15.2 relative error relative error
Gradient r2 Gradient r2
However, the maximum difference in the average relative error Recovery Pentlandite 1.02 0.95 1.01 1.00
between a single calibration circuit and four calibration circuits Pyrrhotite 1.08 0.87 1.02 1.00
is less than 2.5 per cent. Again, this improvement in the accuracy Chalcopyrite 1.00 0.98 1.02 1.00
of prediction must be weighed up against the increased test work
that must be performed to calibrate four circuits. NS gangue 0.97 0.99 1.00 1.00
On average, the relative error based on experimental water Grade Pentlandite 0.86 0.88 1.02 0.99
recoveries was approximately six per cent lower than the Pyrrhotite 0.98 0.67 1.04 0.96
predictions based on the empirical water recovery model. In this Chalcopyrite 0.86 0.95 1.04 0.99
case, the average relative error of the predictions based on
NS gangue 0.98 0.71 0.98 0.98
experimental water recoveries using a single calibration circuit is

100 100

90 90

80 80

70 70
Predicted Recovery (%)

Predicted Recovery (%)

60 60

50 50

40 40

High NiS Low NiS

30 30

High FeS Low FeS

20 20
High CuS Low CuS
10 10
High NSG Low NSG
0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
A Experimental Recovery (%) B Experimental Recovery (%)

FIG 8 - Comparison of experimental and predicted mineral recoveries for predictions producing the highest (A) and lowest (B) average
relative errors (Circuit 7).

74 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 VALIDATION OF THE AMIRA P9 FLOTATION MODEL USING THE FLOATABILITY CHARACTERISATION TEST RIG (FCTR)

100 100

90 90

80 80

70 70
Predicted Grade (%)

Predicted Grade (%)

60 60

50 50

40 40

High NiS Low NiS

30 30
High FeS Low FeS
20 20
High CuS Low CuS
10 10
High NSG Low NSG
0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
A Experimental Grade (%) B Experimental Grade (%)

FIG 9 - Comparison of experimental and predicted mineral grades for predictions producing the highest (A) and lowest (B) average
relative errors (Circuit 7).

The following observations can be made from these results:

• The prediction of pentlandite recovery is accurate and within
the experimental standard deviation for most streams in the
circuit.
• Pentlandite grade is under-predicted (gradient = 0.86) for the
model producing the highest relative error. There are
significant differences in the predictions of pentlandite grade
using the models producing the highest (gradient = 0.86) and
lowest (gradient = 1.02) average relative errors, respectively.
• The prediction of chalcopyrite recovery is accurate with most
points within the experimental standard deviation.
• Chalcopyrite grade is under-predicted (gradient = 0.86) and
over-predicted (gradient = 1.04) for the models producing the
highest and lowest relative errors, respectively.
• Pyrrhotite recovery is over-predicted for the model that gives
the highest relative error, with most points outside the
experimental standard deviation. Pyrrhotite recovery is
predicted accurately for the model producing the lowest
relative error.
• The pyrrhotite grade is predicted accurately for all streams in
the circuit. As with Circuit 6, this may again be related to the
rate of guar addition to the cleaner feed stream in Circuit 7
(56 g/t). This is approximately 21 per cent higher than the
average guar addition rate to the cleaner feed streams in
calibration circuits 1 to 4.
• The prediction of non-sulfide gangue recovery and grade is
accurate and within the experimental standard deviation for
the majority of streams in the circuit.
Comparison of models and prediction methods
Table 6 shows the average relative errors based on the number of
calibration circuits and prediction method used for all models
tested.
The prediction results again show that the average relative
error decreases as the number of calibration circuits increases.
However, the maximum difference in the average relative error
between a single calibration circuit and four calibration circuits
is less than five per cent. Again, this improvement in the
accuracy of prediction must be weighed up against the increased
test work that must be performed to calibrate four circuits.
TABLE 6
Average relative errors for the prediction of Circuit 7 performance.
No of calibration Average relative error (%)
circuits Prediction method 1 Prediction method 2
1 10.7 10.0
2 8.3 8.0
3 7.5 7.3
4 7.0 6.8
On average, the relative error based on experimental water
recoveries was approximately 0.5 per cent lower than the
predictions based on the empirical water recovery model. There
is also less than three per cent difference in the average relative
error between predictions based on experimental water
recoveries using a single calibration circuit, and predictions
based on the empirical water recovery model using four
calibration circuits. This again suggests that with an improved
water recovery model, fewer calibration circuits would be
required to achieve the same prediction accuracy.
IMPORTANT FACTORS FOR MODELLING
A number of important factors have been highlighted from the
validation of the AMIRA P9 flotation model and modelling
methodology.
Reagent addition
Reagent addition was the most important operating parameter in
this test work as, unlike bubble surface area flux or froth
recovery, for example, it does not appear as a directly measurable
parameter in the flotation model. The current floatability
modelling methodology cannot account for the effect of reagent
addition in the estimate of the ore floatability rate. If ore
floatability parameters calibrated at a specific reagent addition
rate are used to predict the performance of a different circuit
configuration, the reagent addition in the new circuit should be
within the range under which the floatability parameters were
calibrated. This will have a significant effect on the prediction of
the performance of the minerals most affected by the addition of
the reagent.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 75

R G COLEMAN, J-P FRANZIDIS and E V MANLAPIG
Flotation model parameters
The accurate estimation or measurement of the flotation model
parameters (bubble surface area flux, froth recovery, residence
time, degree of entrainment and water recovery) is critical for the
accurate prediction of circuit performance.
This work showed a significant difference between the
predictions based on the two different methods of estimating the
water flow. As expected, the predictions using experimental
water flows were significantly more accurate in most circuits.
Predictions based on experimental water recoveries with a single
calibration circuit were within three per cent of, or more accurate
than, the predictions based on the empirical water recovery
model with four calibration circuits. With an improved water
recovery model, fewer calibration circuits would be required to
achieve the same prediction accuracy.
Complexity of the model calibration circuit
This work showed that, of the four calibration circuits, the most
accurate prediction of the performance of the three validation
circuits was achieved using either Circuit 2 (two-stage cleaning
circuit with a single recycle stream) or Circuit 3 (two-stage
cleaning circuit with two recycle streams). Circuit 1, which
consisted of a single-stage cleaning circuit and no recycle
streams, was unable to predict circuit performance accurately. As
all validation circuits were two-stage cleaning circuits, vital
operating information was missing from the circuit based only on
a single stage of cleaning.
The design of the model calibration circuit is therefore critical.
It must consist of flotation banks that treat material of a similar
type to that which must be predicted (eg cleaning or recleaning
stages).
Furthermore, the complexity of the circuit will affect the
stability of circuit operation. A highly complex circuit with many
recycle streams may be difficult to control hence the model
calibration measurements may not be representative of the actual
operation of the circuit.
Number of model calibration circuits
The accuracy of the prediction of circuit performance was found
to improve with the number of circuits used in the model
calibration. This could be due to an increase in the amount of
data used in the model calibration and to the inclusion of a wide
selection of data from circuits of increasing complexity.
However, this improvement in the accuracy of prediction must be
weighed up against the increased test work (in terms of cost,
analysis time, etc) that must be performed to calibrate more than
a single flotation circuit.
CONCLUSIONS
The AMIRA P9 flotation model and modelling methodology has
been successfully validated using the FCTR. Overall, the
predictions of the circuit performance were accurate and within
experimental standard deviations for most streams in the circuits.
The average relative error at each data point from all predictions
ranged from five per cent to 45 per cent. The predictions of the
key variables of pentlandite and chalcopyrite recovery were
accurate.
The prediction of pyrrhotite recovery and grade produced the
largest errors. If the prediction of pyrrhotite performance was
excluded from the analysis, the average relative error at each data
point from all predictions ranged from four per cent to 18 per
cent. The prediction of pyrrhotite recovery appeared to be
dependent on the addition of guar to the cleaner and recleaner
feed streams of the circuit to be predicted. When the guar
addition rates were significantly different to the range of addition
rates used in the calibration circuits, the prediction of pyrrhotite
recovery was inaccurate. These errors were reduced when
experimental water recovery values were used.
76 Fremantle, WA, 19 - 21 March 2007
These results have shown that the predictive validation is
expected to be accurate, even for a circuit with several recycle
streams, as long as the operating conditions of the circuit to be
predicted are within the range of the operating conditions under
which the model was calibrated. Future research should be
undertaken to investigate how far accurate predictions can be
made outside the range of operating conditions. For now it is
recommended that the calibration circuit contain flotation banks
that treat material of a similar type to that which must be
predicted (eg cleaning or recleaning stages).
ACKNOWLEDGEMENTS
The authors would like to thank Dorr-Oliver Eimco and GL&V
for making the FCTR available to the AMIRA P9 project. The
authors would like to thank the staff at Kambalda Nickel
Operation for their assistance with the test work at KNO. The
authors would also like to thank Dan Alexander, David Seaman
and Brigitte Seaman from the JKMRC for their assistance in the
operation of the FCTR at KNO. Finally, the authors would like to
acknowledge the AMIRA P9 project and its sponsors for the
funding which made this test work possible.
REFERENCES
Alexander, D J, Bilney, T and Schwarz, S, 2005. Flotation performance
improvement at Placer Dome Kanowna Belle Gold Mine, in
Proceedings 37th Annual Canadian Mineral Processors Conference,
pp 171-201 (The Canadian Institute of Mining, Metallurgy and
Petroleum: Montreal).
Alexander, D J, Franzidis, J-P and Manlapig, E V, 2003. Froth recovery
measurement in plant scale flotation cells, Minerals Engineering,
16:1197-1203.
Alexander, D J and Morrison, R D, 1998. Rapid estimation of floatability
components in industrial flotation plants, Minerals Engineering,
11:133-143.
Alexander, D J, Runge, K C, Franzidis, J-P and Manlapig, E V, 2000. The
application of multi-component floatability models to full-scale
flotation circuits, in Proceedings Seventh Mill Operators’
Conference, pp 167-177 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
Alexander, D J and Wigley, P, 2003. Flotation circuit analysis at WMC
Ltd Olympic Dam Operation, in Proceedings Eighth Mill Operators’
Conference, pp 41-52 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
Coleman, R G, Esvelt, M, Franzidis, J P and Manlapig, E V, 2003. The
Floatability Characterisation Test Rig as a pilot plant for flotation
testing – An operator’s viewpoint, in Proceedings Eighth Mill
Operators’ Conference, pp 81-86 (The Australasian Institute of
Mining and Metallurgy: Melbourne).
Coleman, R G, Franzidis, J P and Manlapig, E V, 2004. Using the
Flotation Characterisation Test Rig for industrial flotation plant
design, in Proceedings Metallurgical Plant Design and Operating
Strategies 2004, pp 135-149 (The Australasian Institute of Mining
and Metallurgy: Melbourne).
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 Canadian Mineral Processors Conference,
pp 349-370 (The Canadian Institute of Mining, Metallurgy and
Petroleum: Montreal).
Harris, M C, Runge, K C, Whiten, W J and Morrison, R D, 2002.
JKSimFloat as a practical tool for flotation process design and
optimisation, in Proceedings SME Mineral Processing Plant Design,
Practice and Control Conference, pp 461-478 (The Society for
Mining, Metallurgy and Exploration: Littleton).
Rahal, K R, Franzidis, J-P and Manlapig, E V, 2000a. Flotation plant
modelling and simulation using the Floatability Characterisation Test
Rig (FCTR), in Proceedings MINPREX 2000, pp 339-344 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Rahal, K R, Franzidis, J-P and Manlapig, E V, 2000b. The application of
the Floatability Characterisation Test Rig (FCTR) in flotation
research, in Proceedings Seventh Mill Operators’ Conference, pp
339-344 (The Australasian Institute of Mining and Metallurgy:
Melbourne).
Ninth Mill Operators’ Conference
 VALIDATION OF THE AMIRA P9 FLOTATION MODEL USING THE FLOATABILITY CHARACTERISATION TEST RIG (FCTR)

Rahal, K R, Nelson, M G, Lelinski, D, Gritton, K S, Manlapig, E V and
Franzidis, J-P, 2000. Design and operation of a Floatability
Characterisation Test Rig, in Proceedings SME Annual Conference
(The Society for Mining, Metallurgy and Exploration: Littleton).
Savassi, O N, 1998. Direct estimation of the degree of entrainment and
the froth recovery of attached particles in industrial flotation cells,
PhD thesis, University of Queensland, Brisbane.
Schwarz, S, Alexander, D J and Coleman R G, 2006. Development of a
hybrid floatability model, in Proceedings XXIII International
Minerals Processing Congress, Vol III, pp 1860-1865 (International
Minerals Processing Congress: Istanbul).
Schwarz, S, Alexander, D J, Whiten, W J, Franzidis, J-P and Harris, M C,
2006. JKSimFloat v6: Improving flotation circuit performance and
understanding, in Proceedings XXIII International Minerals
Processing Congress, Vol III, pp 1717-1722 (International Minerals
Processing Congress: Istanbul).
Schwarz, S and Kilgariff, B, 2005. Flotation optimisation studies of the
Perilya Broken Hill lead circuit, in Proceedings Centenary of
Flotation Symposium, pp 913-922 (The Australasian Institute of
Mining and Metallurgy: Melbourne).
Tucker, J P, Deglon, D A, Franzidis, J-P, Harris, M C and O’Connor, C T,
1994. An evaluation of a direct method of bubble size measurement
in a laboratory batch flotation cell, Minerals Engineering,
7(5/6):667-680.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 77

A Correlation Between Visiofroth™ Measurements and the
Performance of a Flotation Cell
K Runge1, J McMaster2, M Wortley3, D La Rosa4 and O Guyot5
ABSTRACT
Metso Minerals Cisa have developed Visiofroth™ – a system which
calculates parameters of the froth surface using vision from remote
cameras mounted about individual flotation cells. The parameters
calculated from this system can be utilised within process control
strategies to achieve stabilisation or optimisation of the flotation process.
As part of a JKMRC AMIRA P9 field campaign, froth vision was
collected from various cells operated at different air rates and froth depths
within Rio Tinto’s Northparkes copper roughing and scavenging circuit.
At the same time, samples were collected to determine the cell
metallurgical performance at each operating condition.
These froth images have been processed using the Visiofroth™ system
with froth velocity, bubble size, froth collapse rate and various froth
colour parameters calculated. These parameters have been compared to
the measured concentrate flow and concentrate grade produced at each
condition. This analysis shows that the concentrate tonnage flow is
strongly correlated with froth velocity whereas the concentrate grade is
best correlated using froth velocity in combination with a bubble stability
parameter which affects the bubble size distribution on the top surface of
the froth.
This paper summarises the results from this analysis as well as an
overview of the Visiofroth™ system and the parameters that are
measured by the software.
INTRODUCTION
Traditionally operators have used the visual appearance of the
froth to make decisions about when and what changes need to be
made to optimise flotation circuit performance. With the
development of inexpensive and efficient camera technology and
faster computer processing speeds, it is now possible to capture
and measure various characteristics of the froth in a timely
manner. These measurements can be used within a process
control strategy to stabilise and optimise flotation cell operation.
These systems have the advantage that they can be ‘on the job’
all of the time and provide quantitative rather than vague
qualitative measures of the froth appearance.
Visiofroth™ is a froth image analysis system that has been
developed by Metso Minerals Cisa. It can calculate various
parameters associated with the froth phase, including the froth
velocity, various froth colour parameters, bubble size distribution
1. Manager – Flotation Process Technology, Metso Minerals Process
Technology Group – Asia Pacific, Metso Minerals, Unit 1, 8 - 10
Chapman Place, Eagle Farm Qld 4009.
Email: kym.runge@metso.com
2. Graduate Process Technology Engineer, Metso Minerals Process
Technology Group – Asia Pacific, Metso Minerals, Unit 1, 8 - 10
Chapman Place, Eagle Farm Qld 4009.
Email: jaclyn.mcmaster@metso.com
3. Process Control Engineer, Metso Minerals Process Technology
Group – Asia Pacific, Metso Minerals, Unit 1, 8 - 10 Chapman
Place, Eagle Farm Qld 4009. Email: michael.wortley@metso.com
4. Manager – Process Control and Information Engineering, Metso
Minerals Process Technology Group – Asia Pacific, Metso Minerals,
Unit 1, 8 - 10 Chapman Place, Eagle Farm Qld 4009.
Email: david.larosa@metso.com
5. Product Development Manager, Metso Minerals Cisa Process
Technology Group, 3 Avenue Claude Guillemin, 45060 Orléans
Cedex 2, France. Email: olivier.guyot@metso-cisa.com
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
and bubble collapse rates. These systems are being installed in
many operations around the world. The predominant control
philosophy employed by these systems involves manipulation of
froth velocity to maintain metallurgical targets (eg grade or
recovery).
There is scope for better utilisation of the parameters measured
by Visiofroth™. Improved understanding of:
1. the relationship between the measured parameters and the
metallurgical performance of the cell; and
2. how the various operational parameters (eg air rate, froth
depth, frother concentration, reagent dosage rate) impact on
these measured parameters,
will lead to more effective process control and optimisation.
In this paper, Visiofroth™ is used to analyse froth images
collected from rougher and scavenger cells operated at different
air rates and froth depths. Parameters derived from this analysis
are evaluated to determine if they can be related to the measured
grade and recovery produced at each operating condition.
THE VISIOFROTH™ SYSTEM
Visiofroth™ is an image analysis system for online measurement
of flotation froth properties. Visiofroth™ camera and lighting
assemblies are typically mounted directly above the flotation cell
just inside the cell lip. The camera signals are relayed back via
ethernet cable and optical fibre to a central processing computer,
typically located in the plant control room.
This central processing computer uses the image stream from
each camera to calculate various froth specific parameters.
Currently the system calculates the froth velocity, froth bubble
size distribution, various colour and brightness indices, froth
collapse rate and image stability. Froth texture matching
algorithms and various statistical parameter calculations (eg one
hour average, rate of change) are also provided within the
Visiofroth™ software.
Within the control room, the flotation circuit operators are
provided with images of the froth during processing and the
average value of each froth parameter after each processing
cycle. Time based configurable graphs are also supplied within
the software, which enables the operator to view and assess the
changes in the froth parameters as a function of time.
As Visiofroth™ is a module of OCS©, Metso Mineral Cisa’s
expert system, further processing of the froth parameters is easily
accomplished. Froth parameters can be used, often in
conjunction with other sensor information (eg on stream
analysis), to calculate new operational set points, which can be
outputted directly to the plant’s DCS or PLC. The usual control
strategy currently employed is to maximise froth velocity (and
thus cell recovery) through air and froth depth adjustment whilst
not falling below concentrate grade targets or exceeding other
process constraints. The system can also generate automatic
alarms associated with abnormal froth parameters (eg excessive
froth velocity).
In the following sections, the Visiofroth™ parameters that are
to be investigated within the study presented in this paper will be
described and the method of calculation will be outlined.
79
K RUNGE et al
Froth velocity
The speed at which the froth moves towards the launder is a
parameter often used by operators to assess the mass pull and
recovery off a flotation cell. Visiofroth™ measures the velocity
in both the x and y direction, as well as the resultant velocity
over a specified region of froth. A modified fourier transform
comparative technique is used to assess the displacement in both
the x and y dimensions of consecutive images. This displacement
coupled with the frame collection rate is used to calculate a
velocity. The reported velocity is an average calculated from
processing up to 30 frames per second over a typical ten second
cycle period. The algorithm produces a robust velocity estimate
even in poor lighting conditions.
Bubble size distribution
The texture or bubble size distribution is often used by operators
as a barometer of concentrate quality or as an indicator of the
need for a reagent change. Visiofroth™ uses state-of-the-art
watershed techniques to delineate the bubbles within the image.
The equivalent circle diameter of these bubbles is then calculated
and used to assign each bubble to a bubble size distribution class.
This bubble size count in each class is reported within the system
and can be used to calculate a number of different bubble size
distribution parameters (eg percentage of bubbles in a size class,
average bubble size).
Watershed techniques are suitable for bubble delineation of
froth images because the pixels that delineate the border of
bubbles are usually darker than those in its centre. For these
types of segmentation problems, watershed is a technique that
produces stable segmentation with continuous segmentation
boundaries (Gonzalez and Woods, 2002) and is thus being used
in a number of froth vision systems (Sweet et al, 2000; Hargrave
and Hall, 1997). Visiofroth™ also enables adjustment of the
parameters associated with the watershed algorithm to enable
optimisation of the delineation of bubbles in the image. Froths
with a wide size distribution will, however, exhibit some areas of
over-segmentation and/or some areas of under-segmentation,
regardless of the parameters chosen. The number of
pixels/bubble (which will be related to the zoom setting on the
camera) has been found to affect the result of the segmentation
algorithm.
The segmented image and the bubble size distribution are
displayed, pictorially, within the software. Different colours can be
assigned to the bubbles of different size on these images. Results
of a Visiofroth™ bubble segmentation is shown in Figure 1.
Colour and brightness descriptors
Operators often visually assess the concentrate grade produced
from a cell by the colour of the froth. There are a variety of
different colour models/descriptors that can be used to quantify
the ‘colour’ of a pixel or the colour distribution across an image.
A B
FIG 1 - Results of the Visiofroth™ bubble segmentation algorithm: (A) original image, (B) segmented image and (C) colour differentiation
of bubble size.
80 Fremantle, WA, 19 - 21 March 2007
Visiofroth™ analyses a defined area of the image and calculates
the parameters associated with three different colour models: the
RGB colour model, the HSV colour model and the LAB colour
model. The average colour descriptors of the image are reported
as well as the proportion of pixels within a subset of the colour
descriptors.
The RGB colour model is based on the Cartesian coordinate
system where colour is defined by its position in the RGB colour
cube. The origin is black, with the extremities of the three axes
being red, green and blue. The point furthest from black, passing
through the cubic centre, is white. The remaining three points are
magenta, cyan and yellow. A colour is defined by the values of
red, green and blue it possesses.
The HSV colour model provides an alternative method of
defining colour through its analysis of hue, saturation and
intensity – parameters more akin to a human’s perception of
colour. Hue is related to a pixel’s true colour (eg green, blue or
orange) and is represented as an angular displacement from the
red axis. Saturation is a measure of how much white light is
present in a hue and ranges between zero and 100 per cent. For
example, red is a fully saturated colour whereas pink (red and
white) is less saturated. Finally, value is a measure of a colour’s
intensity or brightness – which is also reported as a value
between zero and 100 per cent. A value of zero for any colour is
white whereas a value of 100 of any colour is black.
Like the HSV colour model, the LAB colour model attempts
to decouple colour from its intensity. It is also designed to
associate similar looking colours closer to each other within the
colour space. The luminance (L) is a value representing the
intensity of the colour and is reported as a value between zero
and 100 per cent, extending from black to white. The Lab ‘a’
parameter is an axis spanning from 100 per cent red to 100 per
cent green and the Lab ‘b’ parameter is an axis spanning from
100 per cent yellow to 100 per cent blue.
Bubble collapse rate
The rate at which bubbles coalesce within and on the surface of
the froth has been shown to have a strong impact on the recovery
and drainage within the froth phase (Zheng, Franzidis and
Manlapig, 2004; Hatfield, 2006). It has been shown to be
strongly affected by the frother concentration, hydrophobicity of
the attached mineral particles and bubble loading (Hatfield,
2006).
Visiofroth™ reports a measure potentially associated with the
surface bubble collapse rate. The bubble collapse rate parameter
is calculated as the dynamic difference between bubble size
histograms derived from two consecutive frames. It is reported as
the ‘per cent change in bubble surface area per frame pair’.
This parameter is related to the size and presence of bubbles
measured by the system. It will also be affected by velocity as
the bubble size histogram will potentially change as bubbles
move out of the field of view.
C
Ninth Mill Operators’ Conference
 A CORRELATION BETWEEN VISIOFROTH™ MEASUREMENTS AND THE PERFORMANCE OF A FLOTATION CELL

EXPERIMENTAL • An in-pulp sampler was used to collect a sample of the cell

The experimental data presented in this paper was collected
during an AMIRA P9 field campaign conducted by JKMRC
personnel (Runge, 2002). The experimental work was undertaken
within the Module 2 copper roughing circuit of Rio Tinto’s
Northparkes operation during July and August of 2001. This
circuit consists of four Dorr Oliver 1000 cells in the rougher
bank and four Dorr Oliver 1000 cells in the scavenger bank.
These cells are square with concentrate overflowing three (first
three roughers) or two peripheral flotation lips. Reagent is added
prior to roughing and scavenging and the cleaner circuit tail
recycles back to the head of the scavenger (Figure 2).
At the time of the study, the operating objective of this circuit
was to maximise the recovery of copper whilst producing a
concentrate of acceptable copper grade for downstream
processing. Copper in the feed was low (<1 per cent) and was
present in a number of copper bearing minerals (bornite,
chalcopyrite and tetrahedrite). The gangue was predominantly
silicates. Pyrite was the predominant sulfide gangue. The solids
in the feed stream to this circuit had a P80 in the order of 100
microns.
Experimental test work involved operating the first rougher,
the third rougher, the first scavenger and third scavenger at
different air rates and froth depths. At each condition, a digital
video camera was mounted above the cell in the direction of flow
of the concentrate and five to ten minutes of froth vision was
recorded. The following samples were also collected during this
time to ascertain the metallurgical performance of the cell:
• Cell feed was either subsampled from the circuit feed pipe or
collected as an in-pulp sample from the preceding cell.
• A cereal container was used to collect timed samples from
different points along the cell lip.
contents/tailing.
• A bucket lid was gently placed on the surface of the froth and
the sample collected was washed into a sample bucket. This
sample is referred to as the ‘top of froth sample’ and is
considered to be a measure of the grade of particles attached
(rather than entrained) in the concentrate (Vera, 2002). This
sample was not collected at all cell conditions.
All samples collected during the test program were weighed to
determine solids and water content and analysed by an X-ray
fluorescence spectroscopy technique to determine their copper,
iron, sulfur, silica, aluminium and calcium assay.
METALLURGICAL ASSESSMENT OF FLOTATION
CELL PERFORMANCE
Data collected from the campaign were mass balanced to
determine the overall flow and recovery achieved at each cell
condition.
The mass balanced results from this study showed that both air
rate and froth depth have a significant impact on the concentrate
grade and recovery achieved from a particular flotation cell. As
was expected, an increase in air rate or a decrease in froth depth
usually resulted in an increase in copper recovery and a decrease
in copper concentrate grade. Anomalously low cell recovery
rates were measured at high air rates in the first rougher. This
result has also been observed by Hadler, Barbian and Cilliers
(2006) who attributed this result to poor froth stability at high
air rates. Figure 3 shows the copper concentrate grade versus
recovery measured in the first rougher and third scavenger at the
different operating conditions.

Rougher Feed Cleaner Scavenger

Tailing
Collector Collector
NASH Scavenger
Frother Rougher Scavenger Tailing

Rougher Scavenger
Concentrate Concentrate
FIG 2 - Northparkes rougher and scavenger flotation circuit.

1st Rougher 3rd Scavenger

60 35
30 Top of Froth Grade
Copper Grade (%)
Copper Grade (%)

50
25 Concentrate Grade
40
20
30
15
20 10
Top of Froth Grade
10 Concentrate Grade 5
0 0
20 30 40 50 60 0 5 10 15 20
Copper Recovery (%) Copper Recovery (%)
A B
FIG 3 - Copper grade versus recovery measured in (A) the first rougher and (B) the third scavenger cell.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 81

K RUNGE et al

Interestingly, there is a significant difference between the

concentrate grade and the top of froth grade for all flotation cells 50.0

Concentrate Grade (%)

(examples of which are shown in Figure 3). This difference Rougher 1
40.0
is attributed to low-grade entrained particles reporting to Rougher 3
concentrate and diluting the higher grade particles that report to 30.0 Scavenger 3
concentrate attached to the bubbles.
Scavenger 1
Silica was found to be predominantly recovered to concentrate 20.0
entrained with the water. Silica flow to concentrate is strongly
10.0
correlated to water flow in all cells (Figure 4).
0.0
14.0 0.0 10.0 20.0 30.0 40.0 50.0
12.0 Water Flow to Concentrate (%)
Silica Flow (t/h)

10.0
8.0 1st Rougher FIG 5 - Relationship between concentrate grade and water flow in
the different rougher and scavenger cells analysed during the
6.0 3rd Rougher
campaign.
4.0 1st Scavenger
2.0 3rd Scavenger
0.0 50.0

Concentrate Grade (%)

0.0 10.0 20.0 30.0 40.0
Water Flow (t/h)
FIG 4 - Relationship between water flow and silica flow in the
different cells of the rougher/scavenger circuit.
High air rates and shallow froth depths were found to be the
conditions that result in maximum water flow to concentrate thus
resulting in high entrainment dilution and low concentrate grade.
This is particularly evident in the third scavenger results where
the concentrate grade is very much lower than the top of froth
grade at high recovery rate conditions (Figure 3b).
It was found that the concentrate grade in a particular cell was
strongly related to both top of froth grade and flow of water to
concentrate (Figures 5 and 6). This information will be used in the
subsequent analysis to facilitate interpretation and development of
the froth parameter correlations.
CORRELATION BETWEEN METALLURGICAL
PARAMETERS AND VISIOFROTH™
PARAMETERS
The five to ten minutes of video collected at each cell condition
was processed using Metso’s Visiofroth™ system to calculate
froth velocity, bubble size, froth collapse rate and the various
froth colour parameters. Each froth parameter in turn was then
compared to the key metallurgical performance indicators of the
cell – concentrate grade, top of froth grade, solids flow rate and
water flow rate. Copper recovery rate was not analysed
independently as it is a consequence of solids rate and concentrate
grade.
A summary of the outcomes of this work is given in the
following sections.
Correlations associated with flow
Froth vision systems are being used in industry to control mass
flow from flotation cells (van Olst et al, 2000; Holtham and
Nguyen, 2002; Kewe, 2003). Hatfield, Bradshaw and de Jager
(2003) showed that the measured froth velocity can be correlated
to concentrate mass yield.
In this study, the froth velocity calculated by Visiofroth™ was
compared to the measured solids and water flow rate produced
from the different rougher and scavenger flotation cells. It was
found that on a particular day of operation, the flow of solids and
water was related to the measured froth velocity (Figure 7).
50.0
40.0
30.0 Rougher 1
20.0 Rougher 3
Scavenger 3
10.0
Scavenger 4
0.0
0.0 20.0 40.0 60.0
Top of Froth Grade (%)
FIG 6 - Relationship between concentrate grade and top of froth
grade in the different rougher and scavenger cells analysed
during the campaign.
This relationship, however, was not linear and was found to
change from day to day. For example, the data collected from
scavenger three was collected over two consecutive days. The
relationship between froth velocity and both solids and water
flow is clearly different between the two different days of
measurement (Figure 8).
As discussed by Hatfield, Bradshaw and de Jager (2003), mass
flow from a particular flotation cell will not only be a function of
froth velocity but also the froth height above the lip weir and the
bubble loading (grams/bubble surface area of bubble) and these
parameters could potentially change with a change in ore type or
reagent dosage rate. Laser techniques are now available for
measuring the height of froth above the weir. It would be
interesting to test whether this information coupled with a froth
velocity measurement would enable a more universal prediction
of solids flow from a cell on different days of operation (in
contrast to the scatter in the relationship between solids flow and
velocity found in this study (Figure 9)).
The results of this analysis indicate that the philosophy of
increasing or decreasing froth velocity to increase or decrease
mass flow from a cell is sound. However, it does indicate that
absolute froth velocity, itself, is not the ideal target for the
control system with mass flow off a cell potentially changing for
a particular froth velocity under different feed conditions. It is
better to change froth velocity set point to meet a target grade or
recovery measured or estimated using auxiliary sensors (eg
on-stream analysis).
Correlations associated with grade
Operators often use the colour or froth texture to assess
concentrate quality or to decide on an appropriate response to a

82 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 A CORRELATION BETWEEN VISIOFROTH™ MEASUREMENTS AND THE PERFORMANCE OF A FLOTATION CELL

45.0 40.0
40.0 Solids 35.0 Solids
35.0 30.0
Water Water
30.0
Flow (t/h)

Flow (t/h)
Solids + Water 25.0 Solids + Water
25.0
20.0
20.0
15.0 15.0
10.0 10.0
5.0 5.0
0.0 0.0
0.0 5.0 10.0 15.0 20.0 25.0 0.0 2.0 4.0 6.0 8.0 10.0
Froth Velocity (cm/sec) Froth Velocity (cm/sec)
A B

FIG 7 - Relationship between solids and water flow and froth velocity in (A) the first rougher and (B) the third scavenger.

35.0
Solids 8/8/2001
30.0
Solids 9/8/2001
25.0
Water 8/8/2001
Flow (t/h)

20.0
Water 9/9/2001
15.0

10.0

5.0
Rougher 1 − 40.9% Cu
0.0
0.0 2.0 4.0 6.0 8.0 10.0 12.0

Froth Velocity (cm/sec)

FIG 8 - Relationship between solids and water flow and froth

velocity in the third scavenger on different days of measurement.

25.0
Solids Flow to Concentrate

20.0
Rougher 3 − 25.5% Cu
15.0
(t/h)

10.0

5.0

0.0
0.0 5.0 10.0 15.0 20.0 25.0 30.0
Froth Velocity (cm/sec)
Rougher 1 Rougher 3 Scavenger 1 Scavenger 3
Scavenger 1 − 13.3% Cu
FIG 9 - Relationship between solids flow and froth velocity
measured in all the cells during the field campaign.

disturbance in the circuit. In the Northparkes rougher and

scavengers, there was a distinct change in both the colour and the
texture of the froth down the flotation bank. Figure 10 shows
images of the froth taken from the first rougher, third rougher,
first scavenger and third scavenger on the same day. Also shown
is the respective copper grades of concentrate produced from
these cells during image collection.
As the concentrate grade decreased down the bank or within a Scavenger 3 − 2.7% Cu
cell operating at higher recovery rates, the concentrate typically
lost its metallic lustre and very low grades were typified by a FIG 10 - Images of the froth collected 7 August 2001.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 83

K RUNGE et al

greyish white froth. High-grade froths (>35 per cent copper)

usually consisted of mono-sized large bubbles with none or very 160.0
few small sized bubbles. Intermediate grade froths (ten to 30 per 140.0
cent copper) consisted of large bubbles interspersed with smaller 120.0

Hue (degrees)
sized bubbles. The lower the grades of these intermediate froths, 100.0
the smaller the size of the interspersed bubbles. The low-grade 80.0
froths (<5 per cent copper) consisted of small, essentially 60.0
mono-sized bubbles. 40.0
The various colour and textural parameters calculated by 20.0
Visiofroth™ were therefore investigated to see whether they 0.0
could be correlated to concentrate grade. Parameters were not
0.0 10.0 20.0 30.0 40.0 50.0 60.0
only compared to the overall concentrate grade but also to the top
of froth grade. Colour for instance may be better correlated to Top of Froth Grade (%)
what is on the top surface of the bubbles rather than the net result Rougher 1 Rougher 3 Scavenger 1 Scavenger 3
of attachment and entrainment in the overflowing concentrate.
It should be noted that a change in camera zoom and
brightness settings occurred halfway through the campaign. This FIG 11 - Comparison of the hue of the HSV colour model and the
change had an effect on both the derived colour parameters and top of froth grade measured in the froth (zoom setting 2).
bubble segmentation, which affected bubble size. Comparisons
often had to be performed with the data split according to zoom. correlated to the concentrate grade than the RGB colour model.
Top of froth grade samples were only collected during the second This finding is similar to that reported previously (Bonfazi et al,
half of the field visit and were therefore all performed with the 2001; Morar et al, 2005).
camera operated at the second zoom setting.
Correlations between grade and bubble size
Correlation between grade and the colour
parameters During analysis of the video footage from each cell condition,
Visiofroth™ segments the image and calculates the number of
Both the concentrate grades and top of froth grades were bubbles in a series of bubble size classes. This information was
compared to the various average colour model parameter used to calculate an average bubble size parameter, calculated by
calculated from the froth footage. Each model exhibited one or summing the number of bubbles in a particular size class by the
two parameters which showed some relationship to these grades: average size of that class and then dividing by the total number
in the RGB colour model it was the blue colour coordinate, in the of bubbles. This particular parameter was found to be strongly
HSV model it was the hue and in the Lab model it was the Lab linearly correlated to the top of froth grade (Figure 12). It was
‘a’ and Lab ‘b’ model parameters (Table 1). also, to a lesser extent, correlated with concentrate grade, but this
correlation is thought to be due to the fact that concentrate grade
TABLE 1 in this system is itself a function of the top of froth grade
(Figure 6).
Linear correlation coefficients obtained from relating concentrate
and top of froth grade to various colour model parameters
(zoom setting 2). R2 = 0.4779
20.0
Number of Correlation coefficient (R2)
Average Bubble Size (cm)

Colour Parameter R2 = 0.7162

model observations Concentrate Top of froth
15.0
copper assay copper assay
RGB Green 25 0.017 0.024
10.0
colour Red 25 0.080 0.098
cube R2 = 0.7951
Blue 25 0.39 0.38 5.0
HSV Hue 25 0.63 0.64
Saturation 25 0.34 0.19 0.0
Value/Intensity 25 0.10 0.13 0.0 10.0 20.0 30.0 40.0 50.0 60.0
Lab Luminance 25 0.058 0.068 Copper Grade (%)
Lab a 25 0.47 0.42 Con Grade (Zoom 1) (24 observations)
Lab b 25 0.63 0.56 Con Grade (Zoom 2) (30 observations)
Top of Froth Grade (Zoom 2) (25 observations)

These relationships, however, all exhibited considerable
scatter, an example of which is shown in Figure 11, and therefore
would only be useful for a rough estimate of the grade being
produced from a cell.
Although the froth was illuminated during image collection
and black plastic was used to minimise ambient light
interference, there were noticeable differences in the brightness
of the images during the course of a day of measurement. All
colour models suffer from the fact that their colour parameters
are affected by the degree of luminescence. It is interesting to
note that the HSV model and Lab model, which attempt to
differentiate between colour and brightness, were better
FIG 12 - Comparison of the average bubble size with the
concentrate and top of froth grades measured in the different cells
during the study.
Correlations between grade and collapse rate
The best correlation, however, was obtained by comparing top of
froth grade with Visiofroth™’s bubble collapse rate parameter
(Figure 13). This result is similar to that found by Morar et al
(2006) who analysed images from the Northparkes rougher using
Smartfroth software. In their analysis they found the correlation
peak, which they use as a measure of bubble collapse rate, was

84 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 A CORRELATION BETWEEN VISIOFROTH™ MEASUREMENTS AND THE PERFORMANCE OF A FLOTATION CELL

Prediction of concentrate grade using collapse rate

Collapse Rate (% per frame pair)

16.0
14.0
and froth velocity
R2 = 0.9088 The metallurgical assessment of the Northparkes cell data showed
12.0
that the concentrate grade, although related to the top of froth
10.0 grade, was also a function of the flow of water to concentrate.
8.0 Water flow is related to froth velocity and it is thus hypothesised
R2 = 0.7931 that a better prediction of concentrate grade could result from an
6.0
equation comprising of both the collapse rate as a measure of the
4.0 top of froth grade and froth velocity as a measure of solids and
2.0 water flow. This idea is referenced from Morar et al (2006) who
was the first to use this approach. They showed that the use of a
0.0 correlation peak (which was shown to be related to the top of froth
0.0 10.0 20.0 30.0 40.0 50.0 60.0 grade) and velocity (which was assumed to be related to
Copper Grade (%) entrainment recovery) gave a much more accurate concentrate
grade prediction than the use of either parameter in isolation.
Con Grade (Zoom 1 & 2) (58 observations) In this study, a similar result is obtained. An equation of the
Top of Froth Grade (Zoom 2) (25 observations) form shown in Equation 1 was better able to predict concentrate
grade (Figure 15) than any of the measured froth parameters used
FIG 13 - Comparison of the concentrate grade and top of froth in isolation. It should be noted that different parameters were
grade with the measured collapse rate measured in the different required to fit the data from each cell. The outlier points are
cells during the study. considered cell conditions at which flooding (ie pulping over the
lip) was occurring.
very strongly correlated with the top of froth assay. These two 1
studies suggest that in this system, the rate at which bubbles a concentrate = (1)
a Collapse Rate + b Velocity + c
collapse in the froth is governed primarily by the grade (and thus
hydrophobicity) of the attached particles. The relationship
developed in this study is reasonably robust with data collected 50.0
from different cells operated on different days of operation
Predicted Concentrate

falling on the same straight line. The bubble collapse rate 40.0
parameter was also found to be independent of the zoom setting Rougher 1
Grade (%)

of the camera. 30.0 Rougher 3

Bubble collapse rate, or its inverse froth stability, is an 20.0 Scavenger 1
important parameter as it has a strong effect on both the degree of Scavenger 3
drainage and recovery of attached particles that occurs within the 10.0
froth phase (Hatfield, 2006). The derivation of a single
0.0
relationship between collapse rate and the copper attachment
grade is an interesting result because it implies that bubble 0.0 10.0 20.0 30.0 40.0 50.0
collapse rate is only a function of the hydrophobicity of the Actual Concentrate Grade (%)
particles and is unrelated to the number of particles on the bubble
surface. Bubble loading (mass per surface area of bubble) is
thought to be a factor affecting froth stability in the froth (Hatfield, FIG 15 - Actual concentrate grade versus a prediction based on
2006), but in this system it would be expected to change as a collapse rate and froth velocity.
function of air rate and would be definitely expected to decrease
down the bank. No effect of this change is observed in the results. Potential use of concentrate grade correlations
More work is planned to further investigate this observation.
Both concentrate grade and top of froth grade are potential
One of the unexplained results that arose from the analysis is targets for a process control strategy. The top of froth grade is a
an inverse relationship between the average bubble size and the measure of the quality of the floatable mineral feeding the
bubble collapse rate (Figure 14). Bubble coalescence is expected subsequent stages of flotation and will have a large bearing on
to have an effect on bubble size on the surface but one would the ultimate circuit concentrate grade. The overall concentrate
expect a high bubble coalescence rate to coincide with a large grade is a measure of the relative quantities of attached and
bubble size. The opposite trend was observed. entrained copper flow. In a circuit with high ultimate entrainment
recovery, this measure is also of significance. Currently, the
16.0 determination of the value of these parameters that result in
optimum circuit performance would be a trial and error process.
(% per frame pair)

14.0
Collapse Rate

12.0 As with other variables in a flotation process, operators learn

R2 = 0.7209 the values of these parameters, which result in optimum
10.0
8.0 performance. Potentially, in the future, modelling algorithms
6.0 could establish the concentrate target grades which best suit a
4.0 particular circuit condition.
2.0 R2 = 0.8331
More work is required to determine how resilient the derived
0.0 grade correlation relationships are to changes in ore type and
0.0 5.0 10.0 15.0 20.0 25.0 30.0 reagent addition used in the process, before they could be used
Average Bubble Size (cm) with confidence in a process control strategy.

Zoom 1 (30 observations) Zoom 2 (28 observations) CONCLUSIONS

Visiofroth™ is a froth vision analysis system which can measure
FIG 14 - Relationship between the average bubble size and parameters which are related to the flow and grade of particles
collapse rate measured by the Visiofroth™ system. produced in the concentrate from a flotation cell.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 85

K RUNGE et al
In this Northparkes case study, solids and water flow from a
particular cell on a particular day was related to the measured
froth velocity. Thus it is appropriate to use the velocity measured
by Visiofroth™ as a means of increasing or decreasing recovery
to achieve a particular processing target. These types of control
loops have the advantage that they can respond to a change more
quickly than one dependent on on-stream analysis alone.
The grade of attached particles on the surface of the froth was
linearly correlated with the bubble collapse rate. This
relationship included data collected from the different cells
operated at different air rates and froth depths on different days
of measurement and thus over the period of the field visit proved
relatively robust. Overall concentrate grade was best predicted
using an equation incorporating the collapse rate and velocity
parameters. These correlations, potentially, allow the calculation
of concentrate purity targets for use in a process control
optimisation strategy.
This study has verified the findings of Morar et al (2006) who
showed that it is better, when investigating froth image data, to
consider the froth as made up of two distinctly different types of
particles – those attached to the bubbles and those entrained in
the plateau borders within the froth phase. Delineation of the
problem in this way, allows correlations to be developed which
would otherwise, potentially, have been hidden.
Bubble collapse rate in the froth was found to be solely a
function of the grade of attached particles and therefore would
seem independent of particle froth loading. This result may have
important implications as we move towards improved
understanding of the froth phase and how we should manipulate
variables to achieve optimal performance.
ACKNOWLEDGEMENTS
The authors would like to acknowledge Eddy Sanwani, David
Seaman, Cesar Gomez, Jorge Torrealba, Brigette Comley, Marco
Vera, Michael Rosenfield and Esther Soden who were involved
in the JKMRC campaign from which the data used in this paper
was collected. We’d also like to acknowledge the support of the
Northparkes metallurgists and technical assistants who provided
assistance with the test work. Northparkes is also thanked for
giving permission for this data set to be analysed by Metso
Minerals and for reviewing the analysis results.
REFERENCES
Bonfazi, G, Serranti, S, Volpe, F and Zuco, R, 2001. Characterisation of
flotation froth colour and structure by machine vision, Computers
and Geosciences, 27:1111-1117.
Gonzalez, R C and Woods, R E, 2002. Digital Image Processing, Second
edition (Prentice Hall, Incorporated: New Jersey).
86 Fremantle, WA, 19 - 21 March 2007
Hadler, K, Barbian, N and Cilliers, J J, 2006. The relationship between
froth stability and flotation performance down a bank of cells, in
Proceedings XXIII International Mineral Processing Congress,
Volume 1, pp 516-523, Istanbul, Turkey, 3 - 8 September.
Hargrave, J M and Hall, S T, 1997. Diagnosis of concentrate grade and
mass flowrate in tin flotation from colour and surface texture
analysis, Minerals Engineering, 10(6):613-621.
Hatfield, D P, 2006. Factors affecting froth surface descriptors and the
implications for froth structure and flotation performance, PhD thesis
(unpublished), University of Cape Town, Cape Town, South Africa.
Hatfield, D P, Bradshaw, D and de Jager, G, 2003. The relationship
between concentrate yield and descriptors from a machine vision
system in a platinum flotation application, in Proceedings XXII
International Mineral Processing Congress, pp 929-936, Cape Town,
South Africa, 29 September - 3 October.
Holtham, P N and Nguyen, K K, 2002. On-line analysis of froth surface
in coal and mineral flotation using JKFrothCam, International
Journal of Mineral Processing, 64(2-3):163-180.
Kewe, T, 2003. Advances in flotation process control at Porgera Gold
Mine using JKFrothCam imaging technology, in Proceedings Eighth
Mill Operators’ Conference, pp 87-92 (The Australasian Institute of
Mining and Metallurgy: Melbourne).
Morar, S H, Forbes, G, Heinrich, G S, Bradshaw, D J, King, D, Adair,
B J I and Esdaile, L, 2005. The use of a colour parameter in a
machine vision system, Smartfroth, to evaluate copper flotation
performance at Rio Tinto’s Kennecott Utah Copper Concentrate, in
Proceedings Centenary of Flotation Symposium, pp 147-151 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Morar, S H, Hatfield, D P, Barbian, N, Bradshaw, D J, Cilliers, J J and
Triffett, B, 2006. A comparison of flotation froth stability
measurements and their use in the prediction of concentrate grade, in
Proceedings XXIII International Mineral Processing Congress,
Volume 1, pp 739-744, Istanbul, Turkey, 3 - 8 September.
Runge, K C, 2002. Northparkes field visit preliminary report part 2:
Rougher and scavenger cell characterisation, Confidential internal
AMIRA P9M flotation report 02/09 (unpublished).
Sweet, C G, Bradshaw, D J, Cilliers, J J L, Wright, B A, de Jager, G and
Francis, J J, 2000. The extraction of valuable minerals from mined
ore, Smartfroth Adams & Adams Patent Attorneys, Pretoria, A&A
Ref: V13676 (1-11) 1999, Full Patent October 2000.
van Olst, M, Brown, N, Bourke, P and Ronkainen, S, 2000. Improving
flotation plant performance at Cadia by controlling and optimising
the rate of froth recovery using Outokumpu Frothmaster™, in
Proceedings Seventh Mill Operators’ Conference, pp 127-135 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Vera, M A, 2002. A touch of froth: How bubble–particle aggregates take
the strain, PhD thesis (unpublished), University of Queensland,
Brisbane, Australia.
Zheng, X, Franzidis, J-P and Manlapig, E V, 2004. Modelling of froth
transportation in industrial flotation cells. Part I: Development of
froth transportation models for attached particles, Minerals
Engineering, 17:981-988.
Ninth Mill Operators’ Conference
Improved Flotation Process Control at Newcrest’s Telfer
Operation
D Baas1, S Hille2 and J Karageorgos3
ABSTRACT
With the increased variability of today’s large-scale mining operations,
consistent flotation performance is important to improving and sustaining
overall recovery through the circuit. The Newcrest Telfer Mine has
implemented advanced process control technology to obtain stable level
operation and consistent mass pull in the copper rougher and scavenger
flotation circuits. The final result has been an overall reduction in
variability, which impacts on both copper and gold recovery. This paper
describes the methodology and implementation of the improvements in
control at the Newcrest Telfer Mine as being an essential part of the
business for minerals processing.
BACKGROUND TO NEWCREST MINING
LIMITED’S TELFER OPERATION
Newcrest Mining Limited’s Telfer operation is a gold-copper
mine located in the Pilbara region of Western Australia.
The operation has three different production pathways,
namely: dump leach to produce dore; copper flotation to produce
copper concentrate; and pyrite flotation to produce a pyrite
concentrate, which is leached to recover dore and a SART circuit
to recover copper and cyanide.
The 20 Mtpa treatment plant is designed to provide optimum
process routes for ore types from different mine sources with
1. Principal Process Control Engineer, Manta Controls, 1 Sharon Place,
Grange SA 5022. Email: dirk.baas@mantacontrols.com.au
2. Manager Metallurgy, Newcrest Operations, Level 2, 20 Terrace
Road, East Perth WA 6892. Email: simon.hille@newcrest.com.au
3. Senior Principal Process Control Engineer, Manta Controls, 1 Sharon
Place, Grange SA 5022.
Email: john.karageorgos@mantacontrols.com.au
FIG 1 - Train 1 flotation circuit layout.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
different metallurgical characteristics. As a consequence, the
flotation circuit has four different processing modes; bulk,
copper only, sequential and pyrite only flotation.
DESCRIPTION OF THE FLOTATION CIRCUIT
The concentrator consists of two trains to enable different
processing modes to be in operation in each train simultaneously.
Train 1 has been designed to process predominately sequential
ores and has a separate pyrite cleaning circuit unlike Train 2.
Train 2 is designed to treat predominately bulk and copper
only ores but has the flexibility to treat sequential ores and utilise
the pyrite cleaning circuit if required by the ore blend. Both
trains have a flash rougher and flash cleaning circuit, the product
of which is combined with the final copper concentrate. The
flash flotation circuit is not discussed in this paper.
The remainder of the flotation circuits are identical and will be
the primary focus of this paper, being:
• copper roughers – four pairs of 150 m3 cells (total eight) per
train,
• pyrite roughers – three pairs of 150 m3 cells (total six) for
Train 1 and two pairs of 150 m3 cells (total four) in Train 2,
• copper cleaners – three by 36.6 m3 cells per train,
• copper cleaner scavengers – three by 36.6 m3 cells per train,
• copper re-cleaners – three trough style 11.33 m3 cells per
train, and
• pyrite cleaners – eight by 36.6 m3 cells for Train 1.
The copper roughers
The objective of the copper roughers is to maximise the recovery
of copper concentrate from the ore. Each pair of cells shares a
87
D BAAS, S HILLE and J KARAGEORGOS
FIG 2 - Train 2 flotation circuit layout.
single level sensor and a pair of level control valves located on
the discharge of the second tank in series. Each cell has an
individual air injection control valve and froth analyser (camera)
to measure the froth behaviour on top of the flotation cell.
The copper rougher feed and copper rougher tails streams are
sampled and analysed by multi-stream analyser (MSA) to
determine metal grades.
The copper cleaners and scavenger cleaners
The objective of the copper cleaners is to upgrade the copper
rougher concentrate to generate a product sufficient for the
recleaning circuit. The objective of the scavenger cleaners is to
minimise losses of valuable mineral once it has reported to the
cleaning block.
The concentrate from the copper roughers is pumped to the
copper cleaner circuit. Each cleaner and scavenger cleaner cell
has individual level and air injection control. The copper rougher
concentrate and the cleaner scavenger tail concentrate and tails
are sampled and analysed by the MSA for metal grades.
The copper recleaners
The purpose of the recleaners is to bring the grade of the copper
concentrate to the required market specification. The recleaners
consist of three cells linked together sharing one level sensor.
The air injection control has one flow controller and three
manual valves allowing the operator to manually adjust the split
of air between the three cells depending on the desired pull-rate
and copper grade being produced. The recleaner concentrate is
also sampled through the MSA for metal grades.
The pyrite roughers
The objective of the pyrite roughers is to maximise the recovery
of pyrite concentrate from the ore. Their design is the same as
the copper roughers. The concentrate from Train one’s pyrite
roughers goes to the pyrite cleaning circuit, which is not
discussed in this paper. Train two’s concentrate is feed to the
copper cleaners.
PROCESS CONTROL PHILOSOPHY – TELFER
When commissioning a new plant, classical process control
philosophy is made up of several key phases, these being;
88 Fremantle, WA, 19 - 21 March 2007
commissioning, stabilisation and optimisation. The duration of
each of these phases will depend on the complexity of the
process and the available technical resources to drive the
improvement.
The Telfer metallurgical team had outlined a simple
approach in line with the classical philosophy during the
pre-commissioning phase of the project, which maximised the
use of available personnel. These steps for flotation were
outlined as:
• Commission – ensure all instruments are functioning and
critical control functions operate. This was accomplished
through a series of factory acceptance testing (FAT) and site
acceptance testing (SAT).
• Stabilisation – commenced with simple field loop tuning of
the circuit. This was followed by circuit characterisation,
which determined key upstream and downstream influences
on major components.
• Optimisation – firstly develop a feed forward responsive
control to process disturbance, followed by application of the
froth cameras and lastly incorporation of the MSA to control
product quality.
After the commissioning of both processing trains had been
successfully completed and the stabilisation process begun,
Manta Controls were invited to work with the Telfer
metallurgical team to continue the stabilisation and optimisation
phases of the process control philosophy.
The Manta Controls hierarchy of controls identifies all the
elements required to achieve the objectives of the optimisation.
The hierarchy of control is similar to a pyramid; if one of the
lower level elements is not available the final objective may not
be achieved. (see Figure 3).
This version of control hierarchy was natural fit with Telfer
Process control philosophy and adapted as a logical way for the
two parties to effectively interface.
LEVEL 1 – FINAL CONTROL ELEMENTS
The first step at Level 1 is to audit all available instrumentation
operation, electrical and electronic performance to ensure that it
is performing within specification. This is time consuming but
essential before any higher level control can be performed.
Ninth Mill Operators’ Conference
 IMPROVED FLOTATION PROCESS CONTROL AT NEWCREST’S TELFER OPERATION

Copper Copper
recovery grade

Level 4 Control Strategy – Optimisation and Constraint Control

Section balancing Grade recovery

Level 3 Control Strategy – Regulatory Control A

Pulp chemistry Surface Cell coordination

chemistry Balance

Level 2 Control Strategy – Regulatory Control B

Cell flow Cell stabilisation

coordination

Level 1 Control Strategy – Final Control and Measurement Elements

pH, ORP, Su
Collectors,
Hopper pump Froth or bubble Air injection Froth depth Grinding Concentrate and
depressants, circuit
activator flow rate velocity flow rate throughput tails grade

FIG 3 - The Manta Controls’ hierarchy of control.

This was then followed by critical circuit parameter
assessment, which involved system step testing and field
response measurements. Based on this work the first element at
Level 1 that needed to be addressed was the response of dual dart
valves on each flotation cell. It was determined that their
performance can be improved as follows.
Dual valve control – split range
There is a small amount of dead band in the positioners, making
it difficult for the dart valves to be positioned at better than
one per cent. To minimise the effect of dead band, the first valve
opens from zero to 100 per cent as the control signal changes
from zero per cent to 50 per cent and the second valve opens
from zero per cent to 100 per cent as the control signal changes
from 50 per cent to 100 per cent. This method halves the effect
of dead band.
Valve linearisation
The trends of the level controllers displayed a non-linear
response due to the shape of the dart and the seat. This was very
pronounced at outputs of 40 per cent to 60 per cent where the
controller changes from the last 20 per cent of the first valve to
the first 20 per cent of the second valve. The overall control loop
responses were improved by linearising the controllers and final
control elements throughout each controller’s range.
The combination of both dual valve split ranging and valve
linearisation has enabled the flotation circuit to perform in a
consistent stabilised fashion with effective response to level
deviations.
LEVEL 2 – REGULATORY CONTROL B
Cell stabilisation
If the level control valve in an upstream cell opens up, it allows
more slurry into the downstream cell, thus increasing that cell’s
level, which in turn opens the second cell’s level control valve.
This means a disturbance in the first cell will be seen in each
cell thereafter. Depending on the individual cell tuning the
disturbance, more often than not, is amplified. This was
particularly noticeable during plant start-ups where the operator
would use manual control to prevent overflows.
To resolve this issue a feed forward signal has been added
from the first cell’s level controller to the second cell’s level
controller and so on through the whole rougher circuit. In this
way any disturbance on the first cell will not have any impact on
any of the subsequent cells, as the amplitude of the upstream
disturbance is adjusted in the downstream level controller.
The improvements in the level stability are shown in Figure 4
and Figure 5. The trends are of the levels of two adjacent
flotation cells during a normal start-up procedure. Both cell’s set
points are changed by the same amount, but the second cell
experiences a greater disturbance caused by the first cell’s
change in throughput. After the stabilisation, split range and
valve linearisation were implemented the cells experienced a
smaller disturbance even though the step change to the set point
was 50 mm greater than the previous trend.
Cell flow coordination – froth production rate
The cell flow coordination can only be applied to flotation cells
that are equipped with a means to measure froth rate being
‘pulled’ (produced) by the flotation cells. This is currently
configured on the rougher circuit.
Froth analyser (camera) data
The froth analyser provides a lot of information that can be used
for control and includes:
• bubble X and Y velocity and vector;
• bubble colour made up of three signals – red, green, blue
(RGB);
• statistical bubble size information;
• bubble load;
• bubble stability; and
• picture brightness.
The quality of the information was assessed to identify what
information could be useful for control and six weeks of data
was sufficient to perform the assessment. By establishing
correlations between plant performance and the froth analysers’
data, the most suitable information for control purposes was
identified.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 89

D BAAS, S HILLE and J KARAGEORGOS
500
400
300
200
100
2/06/05 11:50 2/06/05 11:59 2/06/05 12:08
FIG 4 - Level before stabilisation.
500
400
300
200
100
17/11/06 21:20 17/11/06 21:29 17/11/06 21:38
FIG 5 - Level after stabilisation.
The bubble velocity in the X axis proved to be a reliable
measurement and was representative of the flotation cells’
performance. The bubble velocity provides an indicative figure
of the mass going over the lip of the flotation cells. A weakness
of the bubble velocity measurement is that it does not take into
account the amount of gangue that is entrained in the froth
reporting to the concentrate launder.
The froth from all the copper rougher cells is fed into a single
hopper and then pumped to the copper cleaner circuit. The
hopper pump speed is used to control the hopper’s level and also
provides a good indication of the mass of concentrate being
pulled off the roughers. By combining the bubble velocity from
each cell with the pump speed it is possible to get an indication
of concentrate mass production.
Pull rate control
Disregarding the flotation reagent effects within the cell for the
sake of simplification, there is a direct correlation between
decreasing the froth depth (raising the level of the slurry) and the
bubble velocity of the froth going over the lip of the flotation
cells. Therefore the shallower the froth depth the faster the pull
rate. To control the pull rate of each cell, the bubble velocity is
used to determine the required level.
Using the bubble velocities alone to control the pull rate of the
roughers is not the ideal solution as the pull rate for any given
bubble velocity over the flotation cell’s lip will vary depending
on the load of the bubbles and the amount of gangue that is
entrained with the bubbles. The copper rougher hopper pump
speed gives a good indication of the mass being pulled by the
90 Fremantle, WA, 19 - 21 March 2007
Cell 1 SV mm
Cell 1 PV mm
Cell 2 SV mm
Cell 2 PV mm
2/06/05 12:17 2/06/05 12:26
Cell 1 SV mm
Cell 1 PV mm
Cell 2 SV mm
Cell 2 PV mm
17/11/06 21:47 17/11/06 21:56
roughers, but not the grade of the concentrate produced. If more
concentrate is being pulled, the hopper level rises and the pump
speed increases to maintain the level. This is not an exact
measurement, but can be used to prevent the roughers pulling
more concentrate than the pump can cope with and ensure that
the roughers are not pulling too little concentrate. However, the
weakness of this control is that the grade of the product is not
known even though consistent mass is achieved.
The hopper pump speed is used as a master signal to tell all the
bubble velocity controllers to pull a little harder and reduce the
froth depth if the pump speed is below set point, or slow down
the total pull rate by increasing the froth depth if the pump speed
is above set point.
LEVEL 3 – REGULATORY CONTROL A
Cell flow coordination
Each pair of rougher cells shares a level controller. Without
operator intervention, in most cases, only one cell per bank ends
up pulling concentrate. To ensure each cell in the bank is pulling
the same quantity of concentrate the cells must be balanced. The
only way to do this is to balance the pull rate between each cell
using air addition.
Balancing the twin cells
One cell can end up pulling a lot harder than the other for a
number of reasons:
• the air injection balance between the two cells,
Ninth Mill Operators’ Conference
 IMPROVED FLOTATION PROCESS CONTROL AT NEWCREST’S TELFER OPERATION

• the percentage mineral in the first cell will be slightly higher
than the second cell thus forming a more stable froth bed,
• the flow rate of slurry through the cells tends to raise the
level in the first cell with respect to the level in the second
cell, and
• relative hydraulic head differentials between paired cells
with upstream and downstream effects.
Using the difference in bubble velocity between the two cells
alone cannot achieve a steady balance. Understanding why the
two cells became unbalanced enabled the implementation of one
part of the Manta Froth Cube that uses multiple inputs and one
output as listed in Table 1.
This part of the Froth Cube balances the two cells effectively
and ensures both cells continually pull concentrate.
TABLE 1
Twinned cells balancing control elements.
Pulp chemistry pH control
The pH is modified in the Telfer circuit to assist in the selective
flotation of copper sulfides from pyrite and to reduce the effects
of oxidised species on the pulp chemistry. The pH is controlled
via two controllers, one that adds quick lime to the feed of the
grinding circuit and one that adds milk of lime to the copper
rougher feed. The idea is to use the milk of lime to react to
short-term changes (trim addition) in the pH and the quick lime
to respond to the long-term changes (step addition) in pH. These
loops were decoupled to prevent the loops oscillating and the
quick lime addition was linked to the mill feed rate. By
decoupling the two control loops the pH control improved during
a mill start-up as shown in Figure 6 and Figure 7. The figures
indicate the relationship between the mill feed rate and the
copper roughers’ pH.
In Figure 7 the mill start-up was not as smooth, yet the copper
roughers’ pH was more stable.

Inputs Output ANALYSIS OF THE IMPROVEMENTS

Bubble velocity of cell 1
Bubble velocity of cell 2
Air injection flow rate to cell 1
Grinding circuit throughput
Density of the slurry in the roughers
Air injection flow rate to cell 2.
The following areas of control were significantly improved
during the stabilisation phase:
• copper roughers pH control,
• copper roughers and pyrite roughers level control, and
• cell balancing control.

1600 10

1400
8
1200

1000 6

800

4
600

400
Mill Feed Rate t/h 2
200 Rougher pH PV
Rougher pH SV
0 0

9/03/06 9:00 9/03/06 9:36 9/03/06 10:12 9/03/06 10:48

FIG 6 - Copper rougher pH before.

1600 10

1400
8
1200

1000 6

800

4
600

400
Mill Feed Rate t/h 2
200 Rougher pH PV
Rougher pH SV
0 0

23/10/06 4:45 23/10/06 5:21 23/10/06 5:57 23/10/06 6:33

FIG 7 - Copper roughers pH after.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 91

D BAAS, S HILLE and J KARAGEORGOS

pH control
The stabilisation of the pH has also contributed to better
recovery.
Time to start-up
The time taken to get the rougher cells to recover effectively
during start-up has also been reduced.
Figure 9 shows start-up of the roughers after being optimised.
The flotation froth depths upon start-up are at 400 mm, then drop
to 350 mm and are controlled from there on. The flotation depths
reach their optimum operating levels within two hours of start-up;
however, concentrate is produced within 30 minutes of re-start.
By looking at the rougher concentrate hopper pump speed, we
can see that with the automated start-up sequence control
(Figure 9) significantly more production is realised than in the
manual mode (Figure 8) of operation. During the optimised
start-up the flotation froth depths are continually being adjusted
to maximise the mass pulled from the roughers. This appears
more unstable than the manual start-up but is achieving the
objective of maintaining the hopper pump speed around 75 per
cent. The time taken to reach the optimum level has been
reduced by at least 50 per cent with the optimised control.
MSA samples of the concentrate and tails grades of Train 2
copper roughers were analysed over a three-day period. These
results were plotted on a histogram. Figure 10 and Figure 11 show
the copper rougher tail grade results are more consistent after the
optimisation. The statistical analysis of the before and after data
showed the copper roughers tails mean reading improved from
0.078 per cent Cu to 0.073 per cent Cu and the standard deviation
improved from 0.017 per cent Cu to 0.007 per cent Cu.
The copper rougher concentrate grades shown in Figure 12 and
Figure 13 indicate that the standard deviation in copper rougher
concentrate has also decreased.
After optimisation the concentrate grades are more consistent.
The concentrate mass being pulled from the roughers has also
increased. Data taken over a 30 day period before and after the
optimisation indicated the concentrate mass recovered by the
copper roughers increased by 23 per cent. The results indicate
that with the control hierarchy in place, the copper roughers
operate with a higher level of repeatability in their performance.
The higher mass of concentrate delivered to the cleaner circuit
has ensured that the cleaning circuit is operating at maximum
capacity.
Key benefits
The added benefits of the stabilisation have been:
• the flotation circuits are easier to operate;
• the operators can spend less time monitoring the flotation
cells enabling them to spend more time on grade reagent
control, thereby enhancing production further;
• more predictable flotation circuit performance; and
• faster production recovery from a shutdown situation.

450
Cell 1 SV mm
400 Cell 2 SV mm
Cell 3 SV mm
350 Cell 4 SV mm
Hopper Pump %
300

250

200

150

100

50

0
8/03/06 14:35 8/03/06 16:05 8/03/06 17:35

FIG 8 - Copper roughers start-up before optimisation.

450 Cell 1 SV mm
Cell 2 SV mm
400 Cell 3 SV mm
Cell 4 SV mm
350 Hopper Pump %

300

250

200

150

100

50

0
7:40:00 9:10:00 10:40:00

FIG 9 - Copper roughers start-up after optimisation.

92 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 IMPROVED FLOTATION PROCESS CONTROL AT NEWCREST’S TELFER OPERATION

40
20

30
15

Frequency
Frequency

20
10

10

5
0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0
Train 2 Copper rougher concentrate grade %Cu
0
0.01 0.03 0.05 0.07 0.09 0.11 0.13 0.15
Train 2 Cu tails grade %Cu
FIG 13 - Copper roughers concentrate grade after optimisation.

FIG 10 - Copper roughers tails grade before optimisation.

and concentrate from each part of the flotation circuit and the
information extracted can be used to control the grade produced
20 throughout the flotation circuit. The grade can be altered by the
air injected into the flotation cells or by manipulating the reagent
addition to the circuit. Once the preferred method has been
15
identified by the plant metallurgists, then the final level in the
Frequency

control hierarchy can be implemented.

10

CONCLUSION
5
The merging of the Telfer Process Control Strategy with the
Manta Controls hierarchy of controls has proved an effective
0 method of systematically stabilising the flotation circuit at
0.01 0.03 0.05 0.07 0.09 0.11 0.13 0.15
Train 2 Cu tails grade %Cu Newcrest Mining Limited’s Telfer operation.
The stabilisation of the flotation circuits reduced the start-up
FIG 11 - Copper roughers tails grade after optimisation. time and achieved the required concentrate flow rates sooner.
Further improvements can be realised by automation of the
start-up procedure.
40
The optimisation of the copper roughers has improved the
flotation circuit’s robustness. The grade achieved throughout the
30 copper rougher and pyrite roughers is more consistent and is less
Frequency

affected by variation in mineralogy and the grinding circuit

20 throughput changes. The copper rougher flotation circuit
continually adjusts the froth depth and air injection to maintain
10 the grades and mass recovery required. The concentrate mass
recovered by the copper roughers has increased by 23 per cent.
0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0
Train 2 Copper rougher concentrate grade %Cu
ACKNOWLEDGEMENTS
The authors wish to thank the management of Newcrest Mining
FIG 12 - Copper roughers concentrate grade before optimisation. Limited’s Telfer operation for permission to publish this paper.
The optimisation process was aided by the valuable input
received from the Flotation Process Technicians and the Telfer
FUTURE IMPROVEMENTS Metallurgical Team. Without their feedback on the behaviour of
With the flotation circuit stabilised and balanced, the control the flotation circuit under various conditions the identification of
system’s performance can be accurately analysed. This a number of key correlations would have been difficult.
information will enable the optimisation phase or the fourth level The operations team is also acknowledged for their patience
of the control hierarchy to be implemented. This level utilises the and understanding during the number of step tests performed to
information provided from the MSA. The MSA samples the tails identify the dynamics of the process.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 93

Optimisation of Flotation Circuit Configuration at PT Newmont
Nusa Tenggara
X Zheng1, K McCaffery2 and R P Halomoan3

ABSTRACT TABLE 1
Ranges of mill throughput and flotation feed conditions found
Two column flotation cells have been installed as single-stage rougher in the plant.
cleaners at Batu Hijau. The objective was to provide additional residence
time in the cleaner circuit to maintain maximum recovery during high
Range Target/average
mill throughput and high feed grade periods. Installation of the columns
has allowed the relatively coarse, well-liberated rougher concentrate and Mill throughput TPD 100 000 - 170 000 140 000
the relatively fine, low-grade scavenger concentrate to be treated Feed % Cu 0.35 - 1.2 0.69
separately, improving metallurgical performance in the cleaner circuit.
Plant results show that the columns produce a high-grade concentrate at Feed % pyrite 1-8 2
an average stage copper recovery of approximately 80 per cent. There is Flotation feed P80 micron 150 - 300 220
also indication of increases in both overall plant copper recovery and final
concentrate copper grade. In addition, the operability and stability of the Flotation feed % solids 30 - 40 40
cleaner circuit has greatly improved.
Based on a typical plant monthly composite sample of
INTRODUCTION flotation feed, approximately 65 per cent of the chalcopyrite and
PT Newmont Nusa Tenggara, also known as Batu Hijau, is a bornite particles are fully liberated and an additional 12 per cent
copper/gold operation. It was initially designed to process of the particles are greater than 90 per cent liberated. Table 2
120 000 tonnes of ore per day with a head grade of 0.69 per cent shows the association of chalcopyrite and bornite with other
copper and 0.45 g/t gold. Since commissioning in 1999, a minerals in the composite feed sample.
number of modifications have been made to the grinding circuit. A significant portion of chalcopyrite and bornite is locked in
These initiatives have resulted in increased mill throughput up to binary composites with other copper minerals or pyrite. As pyrite
140 000 tonnes per day (Burger et al, 2006). is also floatable, it is expected that most of the copper-bearing
Periodically, run-of-mine ore grade can exceed 0.9 per cent particles are fast-floating with the exception of ultra-coarse and
copper. During periods of high mill throughput and high head ultra-fine particles.
grade, the cleaner flotation circuit becomes overloaded reducing Figure 1 shows laboratory batch flotation test results of
overall plant recovery (Gatto, 2004). Overloading also occurs different domain ores at Batu Hijau at a constant head P80 of 220
during periods of high pyrite ores. To rectify this, a decision was microns, similar to the typical flotation feed size in the plant
made to expand the cleaner circuit by installing two conventional (Halomoan, 2006).
column cells as single-stage rougher cleaners (McCaffery, 2005).
These new columns were commissioned in early September, 100
2006. 90
With the additional capacity provided by the two columns, the 80
new flotation circuit is expected to maintain recovery during high
Cu Recovery (%)

70
mill throughput and feed grade periods and also to improve 60
operational stability and metallurgical performance.
50
This paper describes the design criteria of the columns and
40
compares flotation circuit performance before and after the cleaner
30
circuit expansion. Options for additional circuit modifications to
further improve metallurgical performance are also discussed. 20
10
ORE CHARACTERISTICS AND FLOTATION FEED 0
0 2 4 6 8 10 12
CONDITIONS
Flotation Time (min)
The Batu Hijau orebody is a large copper/gold deposit. The main
copper minerals are chalcopyrite and bornite. Pyrite is the main FIG 1 - Laboratory batch flotation test results of 11 main
floatable gangue mineral. Ore grade and mineralogy tend to vary domain ores at Batu Hijau.
significantly with location in the orebody. Variations also exist in
physical properties such as hardness and fracture. This variability
has a strong influence on mill throughput and grinding product Copper recoveries ranging from 63 per cent to 86 per cent
particle size. Table 1 presents the ranges of mill throughput and were achieved after the first minute of flotation. Regardless of
ore type, recovery quickly approached a plateau after three
flotation feed conditions found in the plant.
minutes flotation. The laboratory test results indicate relatively
fast floating kinetics of the copper minerals in the ore, attributed
1. Senior Metallurgist, Process Metallurgical and Technical Services, to the high degree of sulfide liberation.
PT Newmont Nusa Tenggara, Indonesia, Locked Bag 544, Victoria
Park WA 6979. Email: xiaofeng.zheng@newmont.com INITIAL FLOTATION CIRCUIT CONFIGURATION
2. Superintendent, Process Metallurgical and Technical Services, PT The initial flotation circuit design was based on ore mineralogy
Newmont Nusa Tenggara, Indonesia. whilst taking into account variability in mill throughput and
3. Metallurgist, Process Metallurgical and Technical Services, PT grinding product size. Figure 2 is a schematic representation of
Newmont Nusa Tenggara, Indonesia. the initial flotation circuit flow sheet.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 95

X ZHENG, K McCAFFERY and R P HALOMOAN

TABLE 2
Mineral association (per cent weight) in a typical composite flotation feed sample.

Chalcopyrite Bornite
+150 150 75 75 45 45 20 -20 +150 150 75 75 45 45 20 -20
Biotite 1.25 0.16 0.02 0.21 0.25 1.18 0.44 0.07 0.00 0.00
Bornite 2.88 4.68 4.30 3.44 0.00
Chalcocite 0.04 0.00 0.33 0.32 0.00 0.03 0.00 0.13 0.00 0.00
Chalcopyrite 6.07 11.05 10.09 7.65 0.00
Chlorite 2.76 1.06 1.47 0.12 0.33 1.33 1.75 0.23 0.00 0.79
Plagioclase 23.53 5.88 1.22 0.67 0.60 21.24 7.04 2.64 0.18 3.75
Pyrite 5.31 6.50 3.20 0.29 0.25 0.38 0.30 0.22 0.00 0.00
Quartz 7.59 1.27 0.25 0.00 0.40 6.31 3.99 0.63 0.00 1.01
Sericite 4.24 3.33 1.40 0.47 0.00 6.41 0.21 0.22 0.00 0.00
Smectite 5.35 3.29 0.06 0.00 0.46 6.33 0.66 0.56 0.12 0.70
Free surface 44.47 71.09 85.60 94.00 97.41 48.62 71.88 84.19 90.32 93.76

Fresh Feed

Rougher Scavenger

Final Tailings

Polishing
Mill Regrind
Mill

3rd Cleaner 2nd Cleaner

Cleaner Scavenger 1st Cleaner

Final Concentrate
FIG 2 - Schematic representation of the previous flotation circuit flow sheet.

The circuit consists of five parallel banks of roughers/
scavengers. Each bank has ten Wemco 130 m3 flotation cells
with the first two cells acting as roughers and the remaining eight
cells as scavengers. The residence time in the plant
rougher/scavenger bank is normally greater than 25 minutes,
with a rougher residence time of approximately five minutes.
The entire rougher concentrate stream is reground in a tower
mill (known as the polishing mill at Batu Hijau) before being fed
to the cleaner circuit. There is no classifier in the polishing mill
circuit. The polishing mill product P80 can vary between
70 microns and 90 microns depending on the feed size (80 -
120 microns), stream flow rate (150 - 300 m3/h) and density (30 -
40 per cent solids).
A separate regrinding circuit consisting of two tower mills and
a hydrocyclone cluster (known as the regrind mills at Batu Hijau)
is set up for grinding the scavenger concentrate. Scavenger
concentrate reports to the hydrocyclones so that only the coarse
fraction from the hydrocyclone underflow is reground. The
closed circuit arrangement is designed to produce a final
grinding product P80 of about 45 microns.
The cleaner circuit provides multiple stages of cleaning to
minimise entrainment and maximise final concentrate grade,
especially for the scavenger concentrate. Froth washing is
applied to the final stage of cleaning to further reduce entrained
particles in the concentrate.
The final tailings of the cleaner circuit are returned to the head
of the rougher circuit. The recycle provides an opportunity to
recover valuables in the cleaner tailings when there is a flow
surge through the cleaner circuit and/or the cleaner circuit is not
stable. Open circuiting this tailings stream can be an option with
high pyrite ores.
The initial design targets for copper recovery in the roughing/
scavenging stage and the cleaning stage were 91 per cent and
98 per cent, respectively. The overall plant recovery of copper
from fresh feed was designed to be 89 per cent.
INITIAL FLOTATION CIRCUIT PERFORMANCE
Figure 3 shows copper recovery in the individual rougher,
scavenger and cleaner circuits as well as the overall plant

96 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 OPTIMISATION OF FLOTATION CIRCUIT CONFIGURATION AT PT NEWMONT NUSA TENGGARA

A 100
Rougher Scavenger
100% 95
90

Cu Recovery (%)
90%

80%
85
80
70%
Cu recovery

75
60%
70
50% 65
40% 60
10 20 30 40 50 60 70 80
30%
A Feed Flow (Cu t/h)
20%

10%
100
0%
Shift-by-shift production data 90
B

Cu Recovery (%)
Overall Cleaner
100% 80

98% 70

96%
60
94%
<5000 t/h
Cu recovery

50 5000 - 6000 t/h

92%
>6000 t/h
90% 40
0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 1.10
88% B Feed Grade (%Cu)
86%
FIG 4 - Effect of mill throughput and feed grade on copper
84%
recovery. (A) Copper recovery versus feed flow rate of metal
82% copper, (B) copper recovery versus feed grade at three different
mill throughput ranges.
80%
Shift-by-shift production data

Mill throughput is clearly influenced by feed grade with a

FIG 3 - Comparison of individual circuit recoveries and overall
plant recovery. (A) Rougher and scavenger recoveries,
(B) cleaner and overall plant recoveries.
recovery on a shift-by-shift basis during a normal month of
operation. The average mill throughput during this month was
136 850 tonnes per day with the feed grade of 0.80 per cent
copper.
Under normal operation, the rougher recovers 60 per cent to
80 per cent of the copper in the feed and the scavenger recovers
an additional ten to 30 per cent of the copper in the feed.
Although mass pull and recovery splits between the rougher and
the scavenger vary from time to time, the total recovery in the
rougher/scavenger circuit remains relatively consistent. The
scavenger circuit appears to have sufficient capacity to recover
the valuables when disturbances occur in the rougher circuit.
Based on plant monthly data, average copper recovery in the
cleaner circuit is generally between 95 per cent and 97 per cent,
although short-term fluctuation in the cleaner circuit can be
frequent, as seen in Figure 3b. Nevertheless, statistical analysis
shows no significant correlation between the cleaner recovery
and the overall plant recovery under normal feed and plant
operating conditions. This is likely due to the recycle of the
cleaner tailings to the rougher circuit and there being sufficient
capacity in the rougher and scavenger circuits at a normal mill
throughput and feed grade.
Figure 4 shows daily recovery of copper at different mill
throughput rates and feed grades in the past three years before
the columns were commissioned. The combined effect of mill
throughput and feed grade is given by metal copper flow rate in
the feed in Figure 4a. To illustrate the effect more clearly,
average recoveries by feed grade increments (every 0.1 per cent)
are used in Figure 4b.
higher mill throughput rate generally being achieved at higher
feed grades. It is also evident that copper recovery increases with
an increase in feed grade.
Interestingly, copper recovery tends to decrease slightly as the
mill throughput rate increases. Loss in copper recovery at a high
mill throughput and relatively low feed grade (ie below 0.7 per
cent copper) is likely attributed to relatively coarser grinding and
poorer liberation rather than limited flotation circuit residence
time. Losses at a higher mill throughput and high feed grade (ie
above 0.9 per cent copper) are likely due to insufficient flotation
circuit capacity.
During August and September 2004 when the average mill
throughput was 152 000 tonnes per day with an average feed
grade of 0.86 per cent copper, the plant copper recovery was
88.2 per cent. This was significantly lower than that achieved at a
similar feed grade but a lower mill throughput. During this peak
production period, recoveries in the rougher, scavenger and
cleaner circuits based on individual circuit feeds were 50.4 per
cent, 80.3 per cent and 87.5 per cent, respectively. The
significantly lower cleaner recovery implies that the cleaner
circuit was overloaded. It is also found that correlation between
the cleaner recovery and the overall plant recovery became
statistically significant during this peak production period
(Figure 5).
The above phenomenon can be explained. When the cleaner
circuit is overloaded, recycle of the cleaner tailings results in
floatable valuable minerals building up in the rougher and
scavenger circuits. If the condition lasts significantly longer than
the circuit residence time, the rougher and scavenger circuits will
eventually reach their capacity limit and recovery will start to
decrease. Therefore, overall plant recovery becomes dependent
on recovery in the cleaner circuit during high mill throughput
and feed grade periods.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 97

X ZHENG, K McCAFFERY and R P HALOMOAN

100%
TABLE 3
95% Summary of column design criteria (all data referring
Overall plant recovery

90%
to one column only).

85% Column dimensions

y = 0.142x + 0.7708
80%
R 2 = 0.3689
Diameter 4.3 m
75% Height 14.0 m
70% Cross-sectional area 14.5 m2
65% Volume 203 m3
60% Effective volume 165 m3
60% 65% 70% 75% 80% 85% 90% 95% 100% Total lip length 41.3 m
Cleaner recovery
Column operating conditions
Froth depth 60 cm
FIG 5 - Effect of cleaner circuit capacity on overall plant recovery.
Air rate 150 m3/h
As noted previously, all the flotation cells at Batu Hijau are Air pressure 400 kPa
Wemco mechanical cells. Air is self-induced in the Wemco cell Superficial gas velocity (Jg) 1.15 cm/s
and air flow available in the cell is relatively independent of feed Sauter mean bubble diameter (d32) 2.5 mm
conditions. This limits the ability to increase mass pull and
recovery in these cells and means that additional capacity to Bubble surface area flux (Sb) 27.5 m2/m2/s
increase overall circuit recovery can only be obtained by adding Air hold-up 15.0 %
more flotation cells. Wash water 20 m3/h
Based on the above plant observations and analysis, a proposal
Superficial bias (Jb) 4.2 cm/min
to install additional flotation cells in the cleaner circuit was
initiated to improve recovery during high mill throughput and Cell residence time 42.1 min
high feed grade periods. Froth residence time 52.3 s
Column feed conditions
COLUMN DESIGN CRITERIA
Column feed solids mass flow rate 90 t/h
The choice of flotation columns over mechanical cells was based
Column feed water mass flow rate 210 t/h
on three main considerations. Firstly, column cells are more
selective in floating a relatively narrow range of liberated Column feed slurry density 30 %
medium-size hydrophobic particles. Secondly, column cell can Column feed slurry volumetric flow rate 236 m3/h
be operated with a deep froth and wash water to minimise
Feed grade 20 % Cu
entrainment. Hence, using column cells to float rougher
concentrate will likely produce high-grade concentrate at a Feed P80 75 µm
reasonably high recovery in a single-stage of cleaning. Finally, Column metallurgical performance
column cells are also more flexible with air rate, froth depth and
Concentrate grade 35 % Cu
wash water addition adjustments. This will provide better control
during operation. Recovery 80 %
Two column cells were installed and commissioned at Batu Mass pull 45.7 %
Hijau in early September, 2006. Some key design criteria are Bubble load 68.6 kg/m3
listed in Table 3.
Carrying capacity 2.83 t/h/m2
The two column cells run in parallel under identical operating
conditions. Column concentrate forms one part of the final Maximum carrying capacity 7.88 t/h/m2
concentrate and column tailings report to the second cleaner Lip carrying capacity 1.00 t/h/m
bank. The rest of the cleaner circuit configuration remains
unchanged at this stage. With the installation of the columns, the
coarser, more liberated rougher concentrate is now being treated Wash water is added to the froth from a tray 50 cm above the
in a single stage separately from the finer, lower grade scavenger top surface of the column. The design was based purely on
concentrate. Figure 6 is a schematic representation of the new maintenance considerations. The amount of wash water is
flotation circuit configuration. currently limited to 30 m3/h due to its impact on froth stability.
With the limited amount of wash water, the columns are operated
Metallurgical improvement at a large positive bias.
Finally, it is worth mentioning that the current working carrying
Based on plant production data over two months after column
capacity (2.8 t/h/m3) of the columns is significantly below their
commissioning, the two column cells achieved an average
potential. It is predicted that the columns will have sufficient
concentrate grade of 34.4 per cent copper at an average stage
capacity to maintain the current recovery level at higher mill
recovery of 79.3 per cent. In comparison, the average final
combined concentrate grade was 31.3 per cent during the same throughputs and feed grades, up to 16 000 tpa and 0.9 per cent
time period. Silica content in the final combined concentrate and copper.
the column concentrate was 9.8 per cent and 7.0 per cent, Figure 7 compares the overall metallurgical performance
respectively. before and after the column commissioning.
The column cells are normally operated at an air rate between It is evident in Figure 7 that concentrate grade-recovery points
120 m3/h and 140 m3/h per column. Maximum air rate available have shifted to a better regime since the column commissioning.
is 170 m3/h. Stage recovery of greater than 90 per cent has been Figures 8 and 9 further examine the post column effect on the
achieved in a number of shifts. The control strategy for column overall plant copper recovery and final concentrate copper grade,
operation is to target silica content at no more than seven per cent while taking into account the effect of feed grade and mill
in the column concentrate. throughput.

98 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 OPTIMISATION OF FLOTATION CIRCUIT CONFIGURATION AT PT NEWMONT NUSA TENGGARA

Cleaner Scavenger Tailings

Rougher 1-2 Scavenger 3-10

Final Tailings
Feed Cleaner
Regrind 1st Cleaner
Scavenger
Cyclones

Polishing Rougher
Mill Cleaner
Regrind
Columns 3rd Cleaner
Mills
2nd Cleaner

Final Concentrate 1 Final Concentrate 2

FIG 6 - Flotation circuit configuration after cleaner circuit upgrade.

In Figure 8, there appears to be an improvement in the overall
plant copper recovery after the column commissioning. However,
statistical verification of the improvement has not been achieved
for this limited data set due to the relatively large data dispersion.
In Figure 9, the improvement in final concentrate copper grade
It has also been observed in the plant that the two additional
columns tend to stabilise the cleaner circuit. The impact caused
by disturbances from the rougher and scavenger circuits on the
100
Overall Plant Copper Recovery (%)

is more evident. Table 4 shows the comparison of linear 95

regression lines for final concentrate copper grade before and
90
after the column commissioning.
The results in Table 4 confirm that there is a linear correlation 85
between the final concentrate copper grade and the feed flow rate 80
of copper both before and after the column commissioning. Tests
for slopes and intercepts suggest that the two lines effectively 75

have the same slope and are unlikely to intercept. The results 70
also indicate that the average 1.38 per cent increase in the final 2006 Before Columns
concentrate copper grade since the column commissioning is 65 2006 After Columns
statistically significant. 60
The plant has not yet experienced high mill throughputs and 0 10 20 30 40 50 60
high feed grades since the column commissioning; however, the Feed Flow (Cu t/h)
new cleaner circuit should have sufficient capacity to maintain
the cleaner stage recovery and the overall plant recovery during FIG 8 - Comparison of overall plant copper recovery before and
peak production periods as have occurred in the past. after column commissioning.

2006 Before Columns 2006 After Columns 40

40
Final Concentrate Grade (%Cu)

Final Concentrate Grade (%Cu)

35
35

30 30

25 25

20 20

2006 Before Columns

15 15
2006 After Columns

10 10
60 65 70 75 80 85 90 95 100 0 10 20 30 40 50 60
Overall Plant Copper Recovery (%) Feed Flow (Cu t/h)

FIG 7 - Comparison of overall metallurgical performance before FIG 9 - Comparison of final concentrate copper grade before
and after column commissioning. and after column commissioning.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 99

X ZHENG, K McCAFFERY and R P HALOMOAN

TABLE 4
Comparison of linear regression lines for final concentrate copper grade before and after column commissioning.

Cu TPH Conc % Cu Cu TPH Conc % Cu Statistic Before (A): n After (B): t Total Pooled
27.17 26.90 48.48 38.07 N 240 59 299
20.69 28.17 40.64 31.01 ΣX 6015 2047 8062
19.97 27.61 30.88 32.28 Mean X 25 35 27
18.36 28.56 30.35 31.14 ΣY 6770 1832 8602
20.90 29.91 21.52 32.25 Mean Y 28 31 29
27.61 29.13 24.77 34.35 Σ x∧2 10 658 3328 18 386 13 985
28.60 27.45 24.35 30.07 y∧2 1498 264 2146 1762
29.91 26.71 26.90 30.64 Σxy 1713 418 3432 2131
35.89 28.20 29.72 31.03 Pool s∧2 5
27.87 29.70 29.81 29.56
24.39 28.50 31.06 29.97 Ao = 24.178 26.697
19.94 27.66 32.63 31.43 V(Ao) = 0.324 1.403
18.90 28.76 35.36 32.03 A1 = 0.161 0.126 0.152
19.16 26.59 31.23 31.50 V(A1) = 0.00048 0.00111 0.000347
19.67 27.39 32.89 29.27 R(A) = 0.429 t(A) = 7.322
22.73 26.01 26.69 30.69 D.f.(A) = 238 p(t{A}) = 3.7497E-12
21.62 28.54 25.69 28.35 R(B) = 0.446 t(B) = 3.766
21.92 27.21 32.54 34.53 D.f.(B) = 57 p(t{B}) = 0.000395
20.79 28.19 19.46 30.28
23.16 28.00 22.22 26.98 Analysis of variance – A
31.70 31.75 26.65 29.38 Source D.F SS MS F (p)
28.32 31.78 29.33 29.27 Regre 1 275.4 275.4 53.61
21.49 31.18 43.20 29.68 Residual 238 1222.6 5.1 3.75E-12
35.46 31.82 35.64 29.16 Total 239 1498.0 -
48.77 29.66 37.57 32.45
30.65 28.50 39.75 32.67 Analysis of variance – B
21.40 30.14 33.28 32.34 Source D.F. SS MS F (p)
22.48 26.66 33.88 32.06 Regre 1 52.5 52.5 14.2
25.05 28.19 32.74 30.36 Residual 57 211.0 3.7 0.000395
27.56 29.35 43.48 31.28 Total 58 263.5 -
29.47 30.17 44.89 32.07
21.01 28.64 43.64 31.43 Test for residual MSs
16.53 28.73 29.86 30.19 F = 1.387 D.F. = 238 57
15.16 25.35 28.37 33.09 p(F) = 7.06E-02
18.15 25.59 23.47 27.31 Test for slopes
21.01 24.76 35.49 31.12 F = 0.644 D.F. = 295
17.72 28.06 47.06 33.68 p(F) = 0.422733
24.59 27.61 35.04 29.75 Test for intercepts
23.21 26.57 43.58 32.09 F = 14.1262 D.F. = 296
24.52 26.67 39.87 31.30 p(F) = 0.000206
20.80 26.59 49.48 32.99 Test for vertical distance (y)
27.43 28.29 27.03 30.19 d(x/y) = -1.380 V(d) = 0.135
22.08 29.61 29.88 29.41 t = -3.756 D.F. = 295
22.15 26.70 38.43 32.39 p(t) = 0.000104 (one-sided)
28.97 28.07 35.20 28.44
27.43 27.25 34.86 29.25
25.68 28.44 40.29 29.36
23.55 28.20 47.02 33.42
20.88 28.90 36.93 33.57
16.34 24.11 32.77 34.46

100 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 OPTIMISATION OF FLOTATION CIRCUIT CONFIGURATION AT PT NEWMONT NUSA TENGGARA

TABLE 4 cont...

Cu TPH Conc % Cu Cu TPH Conc % Cu Statistic Before (A): n After (B): t Total Pooled
15.69 27.48 29.58 26.58
19.00 29.78 33.43 26.17
41.64 31.25 33.39 27.98
19.39 29.83 34.09 32.02
13.63 26.10 46.38 32.54
17.46 26.92 41.31 31.83
23.25 29.28 36.74 30.93
16.55 29.07 47.47 32.14
25.53 29.42 49.15 32.46

cleaner circuit operation became less significant. The stage
recovery of copper in the cleaner circuit not only increased after
the column commissioning in September, 2006 (from average
93.9 per cent in August to 96.1 per cent in October) but also
became more consistent (from 3.8 per cent standard deviation in
August to 2.2 per cent in October). Operators find the columns
and the new cleaner circuit easier to control.
With the two column cells online, it is now possible to carry
out maintenance work on the second and third cleaner banks or
the first cleaner and cleaner scavenger banks without shutting
down the entire flotation circuit. The current circuit
configuration allows the first cleaner concentrate to be diverted
to the columns or the scavenger concentrate to the second cleaner
during partial cleaner circuit maintenance services.
FUTURE IMPROVEMENT PROGRAM
Commissioning of the two columns as rougher cleaners is the
first step of the current optimisation program for the flotation
circuit at Batu Hijau. Based on studies on ore characteristics and
current plant performance, additional circuit modifications are
being considered. These include:
1. Send the concentrate from the existing first two scavenger
cells, together with the rougher concentrate, directly to the
column cleaners. Copper minerals are approximately 92 per
cent liberated in the first two scavenger concentrates,
compared to 97 per cent liberated in the rougher concentrate.
2. Combine the column tailings with the scavenger
concentrate and regrind them together in the existing
regrind circuit. The average copper concentrate grades in
the column tailings and the scavenger concentrate are
8.5 per cent and 12.8 per cent, respectively.
3. Feed the combined scavenger concentrate and column
tailings after regrind to the second cleaner bank instead of
the first cleaner bank.
4. Discard the cleaner scavenger tailings directly to the final
tailings.
The aims of the above modifications are to maximise recovery
of the well liberated valuable particles in the rougher circuit and
avoid over-grinding these particles before the column cleaning
stage. Simultaneously, the circuit will provide sufficient grinding
to liberate the valuables from the composite particles in the
scavenger concentrate and the column cleaner tailings to improve
ore floatability and maximise both recovery and concentrate
grade.
A key consideration of the ongoing flotation circuit
modification at Batu Hijau is to treat different groups of particles
with different properties (ie particle size distribution, degree of
liberation and grade) separately in appropriate sections of the
circuit. It is anticipated that this will provide sufficient residence
time for relatively slow-floating, liberated fines in the cleaner
circuit while allowing multiple stages of cleaning. This will help
minimise entrainment of the fines to ensure a reasonably high
grade of final concentrate from the scavenger concentrate.
It may also be worthwhile to consider implementation of
additional fines flotation at the end of the cleaner circuit. Given
that already high stage recoveries are achieved in the cleaner, this
may not, however, be cost effective and more work is required to
prove this possibility. An alternative example of a possible
flotation circuit configuration to achieve this objective is given in
Figure 11.
Finally, it is worth mentioning that although cleaner scavenger
tailings often have a similar grade to plant fresh feed at Batu
Hijau, they are essentially two different types of ores. Under
normal plant operation, most of the valuables present as either
fines or composites in the cleaner scavenger tailings. These
particles most likely come from entrainment from the rougher
and scavenger circuits or are newly generated from regrinding.
They will be difficult to float again in the rougher/scavenger
circuit. Moreover, concentration of floatable gangues such as
pyrite is significantly higher in the cleaner scavenger tailings.
Recycle of cleaner scavenger tailings in this instance could result
in generation of unstable re-circulating load and have a
detrimental effect on rougher flotation. It is proposed, in future
circuit modifications, to allow provision to open the cleaner
scavenger tailings to final tailings to avoid these potential issues.
CONCLUSIONS
Two additional column cells were installed as single-stage
rougher cleaners at Batu Hijau in early September 2006. The
new cleaner circuit allows the relatively coarser, more liberated
rougher concentrate to be cleaned in a single stage separately
from the finer, lower grade scavenger concentrate. Plant results
show improvements in both overall copper recovery and final
concentrate copper grade since the column commissioning. The
measured 1.38 per cent improvement in the final concentrate
copper grade is proved to be statistically significant. Based on
the working carrying capacity determined in the operating
columns, there is a clear indication that the current cleaner
circuit will provide sufficient capacity to maintain recovery
during high mill throughput and feed grade periods.
In addition to metallurgical improvements, operability and
stability of the cleaner circuit have greatly improved after the
column commissioning. Moreover, the current cleaner circuit
configuration is also more flexible for maintenance planning.
With the additional capacity provided by the column cells,
there are opportunities for additional circuit modifications to
further improve metallurgical performance in the flotation plant.
ACKNOWLEDGEMENTS
The authors would like to acknowledge the involvement of Enzo
Artone, Eugenio Gatto and Dr Hideyuki Okamoto at various

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 101

X ZHENG, K McCAFFERY and R P HALOMOAN

Roughers 1-4 Scavenger 5-10

Final Tailings
Open Circuit
Closed Circuit
Cyclones 2nd Cleaners
Cyclones
Cleaner 1st Cleaners
Scavengers

Polishing
Mill
Regrind
Column Mill
3rd Cleaners
Cleaners

Final Concentrate

FIG 10 - New flotation circuit configuration proposal.

Roughers 1-4
Scavengers 5-10

Final Tailings

Column Open Circuit

Cleaners Cyclones Fines
Cleaner
Closed Circuit
Cyclones
Middlings
Cleaner

Middlings
Re-Cleaner
Fines
Re-Cleaner
Middlings
Regrind
Fines
Regrind

Final Concentrate

FIG 11 - Alternative future option for flotation circuit configuration.

stages of the project. The authors would also like to thank Nengah
Dwi Payana for his assistance during the column commissioning.
REFERENCES
Burger, B, Hatta, M, McGaffin, I and Gaffney, P, 2006. Batu Hijau –
Seven years of operation and continuous improvement, in
Proceedings SAG 2006 Conference, Vancouver, British Colombia,
Canada, 23 - 27 September.
Gatto, E, 2004. Flotation flowsheet options, PT Newmont Nusa Tenggara
internal report, 16 November.
Halomoan, R P, 2006. Laboratory batch flotation tests, grind-recovery
study, October 2005 – February 2006, PT Newmont Nusa Tenggara.
McCaffery, K, 2005. Cleaner circuit upgrade project, PT Newmont Nusa
Tenggara internal report, February.

102 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

Efficiency of the Detritor Mills at Zinifex Century Mine
M Gao1, L Reemeyer2, D P Obeng3 and R Holmes4
ABSTRACT
The Zinifex Century Mine employs a large number of Detritor mills for
fine and ultra-fine grinding to liberate silica and sphalerite at 80 per cent
passing 7.5 microns before final zinc flotation. A joint project between
CSIRO Minerals and the Zinifex Century Mine was carried out to
evaluate the Detritor mill performance. Plant surveys were conducted
around the Detritors in conjunction with laboratory tests. The survey data
showed that the Detritors in the ultra-fine milling circuit performed well.
They reduced a feed F80 sizing of 46 microns to a product P80 sizing of
7.5 microns at a gross energy consumption of 37.2 kWh/t.
Scale-up of ultra-fine milling Detritors was studied using a 0.55 kW
laboratory batch unit and a 355 kW full-scale Detritor by processing
Century zinc concentrate with 1 mm sand media. The results indicated
that the 0.55 kW laboratory mill underestimated the energy consumption
of the full-scale Detritor by about 20 per cent for a product P80 of seven
microns.
The performances of a horizontal IsaMill and a vertical Detritor were
compared as ultra-fine milling devices for processing Century zinc
concentrate. The experimental data demonstrated that, when using the
same grinding media, the energy efficiency of the two mills is similar
despite the difference in mill designs.
The Detritors used for regrind duty at the Century Mine performed
poorly when processing feed at 80 per cent passing 88 micron. The
energy consumption was high and the product size distribution was broad
with particles above 100 microns in the feed unbroken and particles
below ten microns overground. A coarser ceramic media of 3 mm was
1. MAusIMM, Senior Principal Research Scientist, CSIRO Minerals,
Technology Court, Pullenvale Qld 4069.
Email: Mingwei.Gao@csiro.au
2. MAusIMM, Group Metallurgist, Zinifex Limited, Level 29, 2
Southbank Boulevard, Melbourne Vic 3004.
Email: Laurie.Reemeyer@zinifex.com
3. GAusIMM, Senior Metallurgist, Zinifex Century Mine, Lawn Hill,
Mount Isa Qld 4825. Email: Daniel.Obeng@zinifex.com
4. FAusIMM(CP), Program Manager, Iron Ore Processing, CSIRO
Minerals, Technology Court, Pullenvale Qld 4069.
Email: Ralph.Holmes@csiro.au
Zn Rougher
Zn Scavenger
Zn Circuit Feed
Final Tail
Zn Cleaner 1
Six Regrind Detritors
FIG 1 - Century Mine zinc circuit with 21 Detritor mills.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
therefore tested in a laboratory Detritor to process the regrind Detritor
feed as an alternative to the 1 mm Colorado sand used at the Century
Mine. The ceramic media reduced the energy consumption significantly
and produced a narrower product size distribution, which is more
desirable for the downstream flotation process.
INTRODUCTION
The Zinifex Century Mine, located in north Queensland, is the
second largest zinc producer in the world and commenced
operations in November 1999. The concentrator consists of lead
and zinc beneficiation circuits with the lead scavenger tailings
feeding the zinc circuit. There are 21 Detritor mills in the zinc
circuit, six in the regrind circuit processing the zinc scavenger
concentrate and 15 in the ultra-fine milling circuit processing the
combined zinc rougher and zinc first cleaner concentrates
(Burgess, McGuire and Willoughby, 2001). The zinc circuit is
shown in Figure 1.
The Detritor mill originated from sand mills used in the
pigment industry (Lofthouse and Johns, 1998; Smith, 1999;
Lichter and Davey, 2002). It is a vertical stirred mill as shown in
Figure 2 with a mill chamber having a one to one ratio of height
to diameter. Long pins are located on a central shaft in the mill
that act as stirrers rotating at a tip speed of about 11 m/s. Detritor
mills rely on a set of screen panels with 300 micron apertures to
retain the fine sand media. The Zinifex Century Mine uses 1 mm
sand from Colorado Springs in the USA, which has high silica
purity and excellent sphericity and hardness.
EXPERIMENTAL
Plant surveys were conducted on 11 of the 15 Detritors in the
ultra-fine milling circuit and three of the six Detritors in the
regrind circuit. Feed and product samples for each mill were
taken for particle size analysis using a Malvern Laser Sizer and
for dry and wet slurry density measurement. Online operating
parameters, such as flow rate, slurry density and mill power
draw, were recorded for each mill.
Zn Cleaner 2-5
Final Tail
Final Zn Concentrate
Fifteen Ultra Fine
Milling Detritors
105
M GAO et al

The feed rate to No 16 Detritor in the ultra-fine milling circuit

was varied from 6.0 t/h to 18.5 t/h to determine the range in size
reduction through the mill. The results in Figure 4 show that the
Detritor was capable of producing a product P80 sizing as fine as
6.0 microns, but the Detritor achieved a much higher throughput
for a coarser product. For example, the mill throughput for a
coarse product P80 sizing of 16.0 microns was three times that for
a product P80 sizing of six microns.
100
Feed
18.5 t/h, P80 = 16.0 µm
80 15 .3 t/h, P80 = 12.6 µm
11.6 t/h, P80 = 9.8 µm

Percentage Passing, %
8.3 t/h, P80 = 7.5 µm
6.0 t/h, P80 = 6.0 µm
60

40

20

0
FIG 2 - Two of the Detritors installed at the Century Mine. 0.1 1 10 100

Particle Size, microns

The flow rate to one Detritor in the ultra-fine milling circuit
was varied from 6.0 t/h to 18.5 t/h and a similar test was repeated
for one regrind Detritor (7.8 t/h to 25.5 t/h) to establish
correlations between the specific energy and the product P80
sizing. Laboratory tests were also carried out on regrind Detritor
feed using a 0.55 kW batch Detritor, where the power draw of
the Detritor was measured with a three phase power metre wired
to the mill motor main power supply.
RESULTS AND ANALYSIS
Detritor performance in the ultra-fine milling
circuit
Figure 3 shows the feed and the product size distributions for 11
of the ultra-fine milling Detritors. The feed F80 sizing was 46
FIG 4 - Product size distributions for No 16 ultra-fine milling
Detritor at Century Mine for different throughputs.
Figure 5 shows the data that correlate specific energy with
product P80 sizing for ultra-fine milling. On a log-log scale, the
correlation is linear, which could be used to predict the
performance of the Detritor mill. It is clear that the specific
energy consumption of the mill was 20 kWh/t for a product P80
sizing of 13 microns. This specific energy increased to 42 kWh/t
for a P80 sizing of seven microns and 63 kWh/t for a P80 sizing of
five microns.
100
90
80
70
Net Energy Consumption, kWh/t

60
microns and the average product P80 sizing was 7.5 microns. 50 Feed F80 = 44 microns
The gross energy consumption of the mills on average was
40
37.2 kWh/t, which enabled each mill to treat 8.5 t/h of feed at a
normal operating power of 310 kW.
30

100
Mill 14 20
Mill 15
Mill 16
80 Mill 17
Mill 20
Percentage Passing, %

Mill 21
Mill 22 10
60 Mill 23
3 4 5 6 7 8 9 10 20 30
Mill 24
Mill 26 80% Passing Size of the Product (P80) in microns
Mill 27
40 Average feed
FIG 5 - Energy consumption versus product P80 sizing for an
ultra-fine milling Detritor at Century Mine.
20
In addition to the high energy consumption required for a fine
product, the mill discharge temperature increased dramatically as
0 the energy input increased. It is evident in Figure 6 that the mill
0.1 1 10 100 1000
discharge temperature reached 76°C at an energy input of
Particle Size, microns 50 kWh/t. This imposes a limit on further increases in energy
input. Alternatively, costly measures such as a cooling water
FIG 3 - Feed and product size distributions for the majority of the jacket or adding chilled water to the feed would have to be
ultra-fine milling Detritors at Century Mine. implemented.

106 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 EFFICIENCY OF THE DETRITOR MILLS AT ZINIFEX CENTURY MINE

70

500
60
Net Energy Comsumption, kWh/t

400
300

Net Energy Consumption, kWh/t

200
50 355 kW Detritor
100 0.55 kW Detritor
70
40 50
40
30
20
30

10
7
20
5
4
3
10 2
50 60 70 80
1 2 3 4 5 7 10 20 30 40 50 70 100
o 1
Mill Discharge Temperature, C

FIG 6 - Energy consumption versus mill discharge temperature for 80% Passing Size of the Product (P 80) in microns
an ultra-fine milling Detritor at Century Mine.
FIG 8 - Performance of laboratory and full-scale Detritors on
Century Mine zinc concentrate using 1 mm sand media.
Scale-up of Detritors A B
The scale-up of the Detritor mills was studied using a 0.55 kW
batch laboratory Detritor and a 355 kW full-scale mill, which are
shown in Figure 7. The tests with the two mills used the same
feed slurry and the same 1 mm sand media. The specific energy
was determined based on the electrical power reading recorded
for the mill motors.

A B

FIG 9 - Pilot scale horizontal IsaMill and full-scale vertical Detritor

used for comparing energy efficiency for processing Century Mine
zinc concentrate. (A) Four-litre IsaMill, (B) 355 kW Detritor.

100

1 mm sand media
80

FIG 7 - Laboratory (0.55 kW) and full-scale (355 kW) Detritor mills
Percentage Passing

60
used for scale-up studies. (A) 0.55 kW Detritor, (B) 355 kW Detritor.

40
The results in Figure 8 show that the 0.55 kW Detritor
consumed about 20 per cent less specific energy than the 355 kW
Detritor for producing a product P80 of seven micron. 20

36 kWh/t net, IsaMill

Comparison of a four-litre IsaMill and a 355 kW
Detritor
The efficiency of horizontal and vertical stirred mills was
compared by Parry, Klein and Lin (2006). The horizontal mill
was an IsaMill (Enderle, 1997; Johnson et al, 1998) using disc
stirrers, while the vertical mill was a Detritor mill using pin
stirrers. For the study reported here, a four-litre IsaMill and a
355 kW Detritor shown in Figure 9 were used to determine their
energy efficiency for processing Century Mine zinc concentrate
using 1 mm sand media. The results in Figure 10 show that the
two mills performed similarly when used to process the same
feed with the same grinding media, with the IsaMill possibly
producing a slightly sharper distribution at the coarse end of the
size distribution.
0 35 kWh/t net, Detritor
Feed
1 10 100
Particale Size, microns
FIG 10 - Comparison of grinding results for a four-litre IsaMill and
a 355 kW Detritor processing Century Mine zinc concentrate
using 1 mm sand media.
Detritor performance in the regrind milling circuit
The Detritors in the regrind circuit at the Century Mine process a
coarser feed at 80 per cent passing 88 microns compared with the
feed F80 of 44 microns for the ultra-fine milling circuit. Figure 11

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 107

 M GAO et al

80
70
below ten microns, and all particles above 80 microns in the feed
60 Regrind Mill , F80 = 88 µm
were broken. In addition, compared with 1 mm media, the results
50 Ultra Fine Mill, F80 = 44 µm
in Figure 15 show that the 3 mm media reduced the energy
consumption by more than 30 per cent for producing the regrind
Net Energy Consumption, kWh/t

40
product at 80 per cent passing 40 microns.
30
100

Feed
20 25.5 t/h, P80 = 46.8 µm
80 21.1 t/h, P80 = 37.5 µm
17.4 t/h, P80 = 29.5 µm

Percentage Passing, %
13.6 t/h, P80 = 18.3 µm
10 7.8 t/h, P80 = 9.4 µm
9 60
8
7
6
5 40
3 4 5 6 7 8 9 10 20 30 40 50 60

Product P80 , microns

20
FIG 11 - Effect of feed size on Century Mine Detritor performance
using 1 mm sand media.
0
0.1 1 10 100
shows the relationship between the specific energy and the
product P80 sizing for the Detritors in the two circuits. It is clear Particle Size, microns
that the feed particle size has a significant effect on Detritor
FIG 13 - Product size distributions for a regrind Detritor at
performance. The ultra-fine milling Detritor produced a product
Century Mine at different mill throughputs.
P80 of 13 microns using 20 kWh/t of net specific energy, whereas
the regrind Detritor reduced the feed to a P80 of only 22 microns 100
using the same energy consumption.
Feed to the regrind mills
Another deficiency with the regrind Detritor is the broad 3 mm Minemate ceramic
product size distribution produced as shown in Figure 12. During 80 1.0 mm Conditioned Colorado sand
the milling process, the fines below ten microns in the feed were
Percentage Pasasing

increased from eight per cent to 46 per cent, but the particles
above 100 microns remained unbroken. 60

100

Feed, F80 = 88 µm 40
Mill 6
80 Mill 7
Mill 9
Percentage Passing, %

20

60 Average:
Net Energy = 11.1 kWh/t
0
Feed F80 = 88 µm
1 10 100
Product P80 = 43 microns
40
Particle Size, microns

FIG 14 - Effect of media size on regrind Detritor product size

20 distribution for Century Mine regrind feed.

100
90
0 80
70 355 kW Detritor with 1 mm Colorado sand
Net Energy Consumption, kWh/t

0.1 1 10 100 1000

60 0.55 kW Detritor with 3 mm ceramic beads
Particle Size, microns 50 Feed, 80% passing 88 microns
40
FIG 12 - Product size distributions for three regrind Detritors 30
at Century Mine.
20

Figure 13 shows the regrind product size distributions for

different mill throughputs ranging from 7.8 t/h to 25.5 t/h. The
top five per cent of the product size distribution did not change as 10
9
the mill throughput reduced, even at a mill throughput as low as 8
7
7.8 t/h. Insufficient grinding of the coarse fraction in the feed 6
leads to insufficient mineral liberation and adversely affects 5
flotation efficiency. 4
To determine whether coarser media in the regrind Detritors at 5 6 7 8 9 10 20 30 40 50 60 70 80 90100
the Century Mine would improve milling efficiency, 3 mm 80% Passing Size of the Mill Discharge (P80 ) in Microns
ceramic media were tested in the 0.55 kW laboratory Detritor for
a feed with a particle size of 80 per cent passing 88 microns. The FIG 15 - Comparison of specific energy versus product P80 sizing
results are shown in Figure 14. It is clear that the 3 mm media for 1 mm sand and 3 mm ceramic grinding media for Century
produced a narrower product size distribution with less fines Mine regrind feed.

108 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 EFFICIENCY OF THE DETRITOR MILLS AT ZINIFEX CENTURY MINE
CONCLUSIONS
The plant surveys at the Zinifex Century Mine indicated that the
Detritors in the ultra-fine milling circuit were working well. The
mills produced a product P80 of 7.5 microns with a gross energy
consumption of 37.2 kWh/t. However, the Detritors in the regrind
circuit performed poorly. The mismatch of the medium size to
the feed size wasted up to 30 per cent of regrind power and
produced a broad product size distribution that adversely affected
flotation recovery and concentrate grade.
In an effort to improve the efficiency of the regrind Detritors at
the Century Mine, coarser 3 mm ceramic media were tested in a
laboratory Detritor as an alternative to the 1 mm media used in
the plant. The former significantly reduced the energy
consumption for processing regrind Detritor feed and produced a
narrower product size distribution.
Scale-up of the ultra-fine milling Detritors was investigated by
testing the same feed material in a 0.55 kW batch Detritor and a
355 kW full-scale mill using 1 mm sand media. The results
indicated that the 0.55 kW batch Detritor underestimated the
energy consumption of a full-scale Detritor by about 20 per cent
for a product P80 of seven microns.
ACKNOWLEDGEMENTS
Personnel at the Metallurgical Laboratory of the Zinifex Century
Mine and Ross Price from CSIRO Minerals are thanked for their
contributions to this study. Financial support from the Zinifex
Century Mine is also acknowledged.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
REFERENCES
Burgess, F, McGuire, I and Willoughby, R, 2001. Operation of sand mill
Detritors at Pasminco Operations, Fine Particle Processing and
Tailing Summit, Perth, 2 - 4 July.
Enderle, U, Woodall, P, Duffy, M S and Johnson, N W, 1997. Stirred mill
technology for regrinding McArthur river and Mount Isa zinc/lead
ores, in Proceedings XX International Mineral Congress, Aachen,
Germany, 21 - 27 September, pp 71-77.
Johnson, N W, Gao, M, Young, M F and Cronin, B, 1998. Application of
the IsaMill (a horizontal stirred mill) to the lead-zinc concentrator
(Mount Isa Mines Ltd) and the Mining Cycle, in Proceedings
AusIMM ’98 – The Mining Cycle, pp 291-298 (The Australasian
Institute of Mining and Metallurgy: Melbourne).
Lichter, J K H and Davey, G, 2002. Selection and sizing of ultra-fine and
stirred grinding mills, Mineral Processing Plant Design. Practice,
and Control (ed: A L Mular) (Society for Mining, Metallurgy and
Exploration: Littleton).
Lofthouse, C H and Johns, F, 1998. The ECC International Stirred Media
Mill and the Metals Industry, Comminution 8, Brisbane, 6 - 8
February.
Parry, J, Klein, B and Lin, D, 2006. Comparison of ultra fine grinding
technologies, in Proceedings XXIII International Mineral Processing
Congress, Istanbul, 3 - 8 September, p 177.
Smith, C, 1999. Svedala Stirred Media Detritor, IIR Conference Crushing
and Grinding ’99, Perth, 29 - 30 March.
109
Bolting Technology - Understanding and Utilising the Inherent
Mechanical Properties of Bolts/Nuts
B Docherty’

ABSTRACT
This presentation is designed to give a basic understanding of bolting technology and
terminology to the layperson as well as highlighting the importance of correctly utilising the
inherent mechanical properties of the bolts being used. Bolt breakage has always been an issue
associated with a milling operation and to a certain extent it has been acceptable, but in the
current business climate, where time is money and safety is paramount, it has become
increasingly unacceptable. Mill operators are continually coming under increased pressure to
improve mill availability and this is almost impossible to achieve if you are consistently
shutting down to address bolting issues. Attention to detail during the installation process can
greatly reduce the likelihood of having to shut down due to bolt breakage.

1. Regional Sales Manager, Donhad Pty Ltd, Locked Bag 7, Bassendean WA 6054. Email: Ben.Docherty@donhad.com.au

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 111

Advanced Mill Relining — A Key Operational Control (Not Simply
a ‘Must Do’ Maintenance Function)
J Russell1
ABSTRACT
RME has a vision for a new grinding plant operational paradigm made
possible by RME’s fast, safe Mill Relining System.
Some new liner designs can degrade SAG mill performance because of
mill volume loss, ball strike (rotational speed limitations) and other issues
often brought about by modifying liner design for maximum life and
without a backward glance at metallurgical consequences. Freshly relined
mills can initially produce as little as 80 per cent of their maximum
throughput, that maximum occurring toward the end of the liners’ wear
life.
RME’s proposition is to see new liners designed for maximum
commercial performance. This may mean more frequent mill relines but
with liners designed to yield optimum grinding, metallurgical and
profitability performance, not maximum life.
In other words, mill liner profiles and mill relining speed become key
operational controls and not simply ‘must do’ maintenance functions.
Taken to the extreme, liner profiles may prove capable of ‘tuning’ a
grinding plant for maximum profitability in any metal price environment.
Following the introduction, the first section of this paper provides an
update describing recent and significant developments within RME’s
range of Mill Relining Systems, the enabling technologies which permit
more frequent liner exchange without incurring additional downtime
imposts.
The second section outlines RME’s vision of a ‘liner profile versus
profitability’ commercial model, developed either intuitively or by using a
combination of new measurement and modelling technologies. These
new measurement and modelling tools provide the opportunity to predict
the optimum commercial outcome, off-line in the virtual world, before
committing to an expensive liner experiment.
INTRODUCTION
RME’s original goal, circa 1990, was to increase whole plant
annual utilisation by halving grinding mill relining times and, as
a consequence, whole plant shutdown times. This halving of
1. MAusIMM, Managing Director, Russell Mineral Equipment Pty Ltd,
141 - 143 Russell Street, Toowoomba Qld 4350.
Email: John.Russell@rmeAus.com
FIG 1 - A diagrammatic representation of the hypothetical difference between life optimised liners (left) and profit optimised liners (right).
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
reline times was to be achieved through mechanisation of the
mill relining processes which usually define each shutdown’s
critical path.
Mechanisation of the mill relining process, the focus of so
much of RME’s efforts over the last 16 years, possesses two key
performance indicators:
• relining speed, and
• relining crew safety.
RME’s worn-liner-removal, new-liner-placement Mill Relining
Systems, used proficiently, consistently deliver the fastest
and safest grinding mill relines in the world. Continuous
development work now provides the benefits of RME Mill
Relining System mechanisation for every mill operator, from the
smallest to the largest.
Looking beyond reline mechanisation, we find the liners
themselves. Liners have two key performance indicators as well:
• liner cost (cost/life relationship), and
• mill metallurgical performance.
If left in isolation, liner profile development to extend ‘liner
life’, with the aim of reducing annual liner cost, may have a
negative impact upon mill grinding performance in terms of
throughput (tonnes per day) and/or particle size reduction
(comminution). For example very thick (long lasting) liners
reduce mill volume and mill throughput. A 32’ diameter mill
performs as (say) a 31’ mill.
Any mill performance reduction will have revenue and
profitability consequences.
Given that RME’s Mill Relining Systems deliver consistent,
fast and safe relines, perhaps more relines per year with
performance maximised (not life maximised) liners may yield a
superior economic result, particularly in this high revenue
(record metal price) environment.
The contrast between ‘long life’ liners delivering a lesser
result, compared against ‘commercially optimised’ liners, is
shown diagrammatically in Figure 1.
113
J RUSSELL

Furthermore, it may be possible to ‘tune’ grinding mill liner Smaller mill relining systems for mills in the
profiles for maximum profitability in any current metal price 10’ - 20’/3 m - 6 m diameter range liners of
environment. 440 - 2200 lb/200 - 1000 kg
It should be noted that this commercial model may prove that
metallurgically optimised liners, changed more frequently, There exists a relatively large population of smaller mills in the
maximise profitability regardless of metal prices. If this is 10’ to 20’/3 m to 6 m diameter range. These mills typically
proved to be true, it would appear that too many years of very feature liners weighing less than 500 kg.
low metal prices have bludgeoned the world’s mining industry Smaller mills are usually relined slowly, manually and at
into a simplistic ‘cost minimisation’ mindset. considerable risk to the relining crew members. Access for
These better times may provide the opportunity to rediscover a mechanisation into the mill cavity can be difficult. Inside the
healthier performance and profit maximisation mindset, mill, reline crews struggle across a slippery ball charge carrying
displacing conceptually simple but erroneous, cost minimisation heavy steel or composite liners, supported on steel bars. This
habits and practices. scenario is accepted practice at many sites, simply because no
These concepts are developed under the ‘Liner Profiles to technical alternative has been available.
Maximise Profitability’ section. RME has addressed smaller mill relining system requirements
by developing a complete range of smaller mill ‘liner handlers’
RME MILL RELINING SYSTEMS including:
(THE ENABLING TECHNOLOGIES) • Millmast in-mill-erected liner handlers (refer to Figure 2);
An RME Mill Relining System consists of a ‘liner handler’ (a • small, conventional (crane type) liner handlers;
three axis liner-suspension mill liner handler or, more commonly, • small liner placement Mill Relining Machines.
a multi-axis, liner-placement Mill Relining Machine) plus at RME has also developed a smaller, pneumatic Thunderbolt
least one Thunderbolt recoilless hammer for worn liner removal. recoilless hammer for worn liner removal (refer to Figure 3).
Larger installations may also include powered feed chute
extraction and transportation resources.
Large, rich deposits such as Escondida and Freeport operate
large capacity mineral processing equipment and share the
obvious goals of maximising plant availability, utilisation,
recoveries and metal production. These characteristics combine
to justify highly mechanised mill relining equipment and
methods, which also solve associated occupational health and
safety issues.
In contrast, a junior miner, with a difficult underground
resource, may be limited by mine ore production, not the mineral
recovery plant throughput potential. Also, equipment is usually
of smaller capacity (often second hand) and capital constraints
may be considerable. Here, mill relining equipment selection
will probably be based on cost and meeting minimum legal and
ethical occupational health and safety requirements.
RME Mill Relining System solutions are now available for all
situations.

New grinding plant design and construction –

A focus on grinding mill relining
In any new grinding mill plant project, there are nine
parties which influence relining possibilities and outcomes and
consequentially whole plant availability outcomes:
• property owner;
• selected project manager/engineer;
FIG 2 - RME Millmast Powershift single boom Mill Liner Handler of
200 kg rated capacity developed for small to medium Ball mill
applications with higher ball charge levels such as this 12.5’/3.8 m
diameter Ball mill.

• grinding mill supplier;

• liner supplier;
• liner bolt supplier, operating personnel, responsibilities and
practices;
• maintenance personnel, responsibilities and practices;
• mill relining crew (perhaps contract); and
• mill relining technology supplier (RME).
The influence each of these parties has over every unique
grinding mill relining circumstance is described in detail in a
paper delivered to Crushing and Grinding Conference (Russell,
2002) and is recommended reading for all those involved in new
grinding plant design and construction.
One of the central purposes of this (Kalgoorlie) paper is to
provide a checklist for project management engineers, unfamiliar
with the relining process, in order to help avoid plant design FIG 3 - RME Thunderbolt 450 recoilless pneumatic hammer,
details which may impede future relining efficiency … for the delivering 450 joules per blow at one blow per second,
entire life of the project! continuously, powered from 100 psi/689 kPa plant air supply.

114 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 ADVANCED MILL RELINING — A KEY OPERATIONAL CONTROL (NOT SIMPLY A ‘MUST DO’ MAINTENANCE FUNCTION)
Larger mill relining systems for mills in the
16’ - 40’/5 m - 12 m diameter range liners
2200 - 16 500 lb/1000 - 7500 kg
Much of RME’s work has tended toward larger mill operators,
a market which demands high capacity, high performance
mill relining equipment. RME’s worn-liner-removal, new-liner-
placement Mill Relining Systems have been unique in halving the
time taken to reline large AG, SAG and Ball mills, worldwide.
Significantly, RME’s Mill Relining Systems have changed
plant management expectations regarding mill relining times,
resulting in shorter whole plant shutdowns and significantly
increased annual plant utilisation.
A typical, large-plant RME Mill Relining System consists of:
• powered feed chute extraction and transportation;
• three axis liner suspension mill liner handler or, much more
commonly;
• a seven or eight axis liner placement Mill Relining Machine;
and
• Thunderbolt recoilless hydraulic hammers.
RME’s large-capacity, liner-placement Mill Relining Machines
have provided the means to apply very large liners. At the time
FIG 4 - RME eight axis liner placement Mill Relining Machine of
6000 kg rated capacity. Permission to publish this photograph by
Newcrest Mining Limited is acknowledged.
FIG 5 - RME Thunderbolt 750 recoilless hydraulic hammer
in action.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
of writing (October 2006) RME has supplied 107 ‘liner handlers’
of which 25 are Mill Relining Machines of 3500 kg/7700 lb rated
capacity and over, including three machines of 4545 kg/10 000 lb
rated capacity, one 6000 kg/13 200 lb rated capacity and one of
7250 kg/15 950 lb rated capacity.
Large liners offer every advantage. RME’s Mill Relining
Machines can place (say) 5000 kg liners with the same speed as
1000 kg liners. Liner placement rates (m2/hr) are therefore
improved significantly with larger liners. Also, larger liner
placement is easy and efficient. Edge alignment of their vast
dimensions assures bolt hole alignment between liner and mill
shell for easy bolt insertion.
LINER PROFILES TO MAXIMISE PROFITABILITY
Mill liner functions
Mill liners have two functions:
1. The first function is to protect the mill’s structural shell
from destructive abrasion. Here, the key performance
indicator is cost of life, ie for how long will a set of liners
safely grind ore and protect the expensive mill shell from
damage and what is the cost of that protection. These cost
KPIs (dollars per tonne processed or dollars per month
protection) are easily measured and are therefore ‘attractive’
KPIs.
2. The second function is metallurgical performance. Liners
represent the point of engagement between the mill’s
installed power and the mineral ore. The way in which the
liner profiles engage the grinding media and mineral ore,
combined with the volume, geometry and rotational speed
of the mill, define the grinding process itself. Liners
therefore have this second and important task and that is to
help maximise the comminution process. Liners are
production variable since they wear and change shape.
Here, the liner’s KPI must be throughput and comminution
efficiency, combined for the purposes of discussion as
‘grinding performance’. This KPI is masked by a host of
other variables, especially swings in ore character, and the
data is ‘noisy’. This is a complex and therefore ‘unattractive’
KPI.
Reduced grinding performance after a reline –
why?
At some sites, RME personnel have become aware of a marked
drop-off in mill grinding performance immediately following
liner replacement and, in particular, shell liner replacement and
grate replacement.
Performance loss following grate replacement is easily
explained by open area. New grates have less open area than
worn grates and therefore throughput is reduced with new grates.
Some sites change every second grate to minimise this issue.
Performance drop-off following shell liner exchange is more
complex. In a high aspect SAG mill, around 70 per cent of the
‘lift’ comes from shell liners, while the remaining 30 per cent
comes from outer row end liner lifters. Interestingly, the latest
discrete element models clearly show extra lift from ‘corner
effects’ between end and shell liners.
Variables in shell liner design include:
• lifter height, which may vary over the mill length (new by
design and worn);
• lifter shape, particularly leading face angle;
• wear profile, a complex issue including ore characteristics,
mill speeds, liner material specification and casting/heat
treatment quality; and
• final profile (how thin do we risk?).
115
J RUSSELL
FIG 6 - A plot of liner weight versus year for RME’s 25 largest Mill Relining Machines (plus ‘Leinster’ the world’s first large-capacity Seven
Axis liner placement Mill Relining Machine (circa 1992).
The complexity of these variables has previously defied
detailed quantitative analysis. Liner development, on site, tends
to follow the simpler routes of ‘wear life’ maximisation, ie
responding to the ‘attractive’ KPIs.
This ‘wear life’ maximisation (and the associated annual ‘liner
cost’ minimisation and annual ‘utilisation maximisation’) thought
processes are common to both operating and maintenance
personnel.
Cost pressure to maximise liner life (minimise annual liner
costs) has a historical basis due to long periods of low
commodity prices. These commercial drivers have been
reinforced over the last few years due to the marked increase in
iron and alloying prices. Mill liners, with their high chrome and
molybdenum content, have been particularly affected.
Taken to the extreme, ‘cost reduction’ and ‘maximum
utilisation’ focii on milling operations can lead to one or more of
the following liner-related, grinding problems:
• reduced mill volume through large, thick (long lasting)
liners;
• reduced mill volume through ‘packing’ of ore material
between deep liner lifters;
• (packing can be loose, which falls out when the mill stops, or
cemented by clays, which remain once the mill stops);
• individual liners featuring long lives but with no liner life
being a multiple of another, eg grates – four months, shell
lifters – five months, f/e outer row – seven months, etc
(sensible, regular, planned maintenance becomes impossible);
• insufficient mill power draw at some time during liner life;
• ‘over-throwing’ of balls beyond toe of charge (particularly in
fixed speed mills) leading to ball and liner damage, slow and
dangerous relines (oxy-lancing) and, in the extreme, mill
shell damage (‘flowing’ liners which ‘spring’ the shell at
structural joints); and
• reduction in mill speed and power draw with new liners,
necessary to avoid ‘over-throwing’ of balls.
These effects, in combination, reduce mill performance with
new liners to as little as 80 per cent of maximum grinding
performance. A casual conversation with a customer recently
indicated 77 per cent of maximum capacity on start-up, that
maximum occurring at the end of shell liner life.
116 Fremantle, WA, 19 - 21 March 2007
The ‘long life liner’, ‘maximum annual mill utilisation’,
‘minimum annual liner cost’ mindsets are further reinforced by
historical fears of mill relining work. This fear is now unfounded
in practice because of the widespread adoption of RME’s fully
mechanised worn-liner-removal, new-liner-placement Mill
Relining Systems.
For maximum metallurgical performance and maximum
profitability, the appropriate question and associated KPI should
be: ‘what is the optimum liner design and liner exchange
program for maximum profitability?’
‘KPIs = ‘grinding performance’, ie tonnes processed, metal
recoveries, profitability’. These are the complex and unattractive
KPI’s mentioned earlier in this paper.
The key to answering this question, at minimum risk, may lie
in a combination of new technologies, now discussed.
NEW TECHNOLOGIES AND A ‘LINER PROFILE
VERSUS PROFITABILITY’ RELATIONSHIP
As discussed above, the inter-relationship between grinding
performance, liner cost, downtime cost, revenue and profitability
are complex.
Fortunately, a range of new technologies now exist which
should allow research organisations, independent suppliers or
motivated mine sites to develop a commercial model, which can:
• find the optimum relationship between grinding performance
(liner profile, liner life, relining time, mill throughput,
comminution effectiveness and recoveries); and
• commercial performance (liner cost, reline cost, downtime
cost, metal price, revenue and profitability).
This envisaged commercial model must yield a ‘liner profile
versus profitability’ relationship for a particular grinding plant.
Optimum liner profile may also prove to be a function of current
commodity price.
RME believes there is a ‘perfect’ liner profile for each grinding
plant and that taken to the extreme level of understanding (a very
accurate commercial model), the liners’ optimum profiles are
affected by the current commodity price.
The new technologies, and the prospect of combining these to
generate the ‘liner profile versus profitability’ relationship, are
especially exciting in these times of high commodity prices.
Ninth Mill Operators’ Conference
 ADVANCED MILL RELINING — A KEY OPERATIONAL CONTROL (NOT SIMPLY A ‘MUST DO’ MAINTENANCE FUNCTION)
Another (say) five per cent throughput, for little or no capital
cost, should prove attractive to most operations. The alignment
and combination of new and existing technologies are now
detailed.
The new technologies are:
• in situ liner imaging using either existing laser scanning
technology (as described below) or optical, spatial, 3D
imaging;
• discrete element modelling (DEM) of the grinding process
(perhaps the best models at this time combine DE and
empirical elements); and
• liner wear predictive models.
Established technologies, necessary to complete the ‘liner
profile to profitability model’ are:
• online particle size distribution measurement of SAG mill
feed,
• mill power draw and feed rate data (normal outputs from
plant control system), and
• conventional sampling of SAG mill discharge.
RME’s advanced worn-liner-removal, new-liner-placement
Mill Relining System performance is also critical to minimise
liner exchange downtime.
Liner imaging in situ
This is a recent but developed technology, currently used by
Alcoa of Australia (Alumina) in Western Australia (WA) and by
an increasing number of other Australian mines.
Scanalyse, Perth, Western Australia (a technology spin-off
from Curtin University in WA) has developed a mill liner
scanning technology including hardware and software.
The chosen German laser scanner is capable of securing over
seven million data points in three minutes with point location
accuracy of ±3 mm. Please refer to the raw data image Figure 7.
Analysing this raw data using proprietary software (Scanalyse)
yields very interesting information.
In its most basic form, this information can reveal areas of
highest wear, areas of lowest wear and, significantly, the residual
thickness of each liner.
The data ‘cloud’ which defines the exact liner surface profile,
in situ and in three dimensions, has another potent application in
the discrete element model of mill operations.
A side issue –mill shell ‘wash’ damage and
prevention
RME is encountering an increasing incidence of mill shell
wash where liners have been left in service too long and break
up. This exposes the relatively soft structural shell to the mill’s
process abrasion. Even small amounts of shell wash can
compromise liner-to-shell fit with increased chances of further
wash and liner break-up. Sufficient shell wash can undermine
the structural integrity of the mill shell.
In this time of insufficient, experienced, technical personnel,
this technology could avoid catastrophe. A mill shell collapse,
in operation, could easily cause a multiple fatality accident.
Economically, it would be a robust company to not suffer a
commercial ‘fatality’.
This residual liner thickness measurement technology could
form the basis of a preventative maintenance program to
minimise, if not eliminate, this risk.
Discrete element modelling (DEM)
Various research organisations and mill manufacturers are
developing DEM to better predict mill performance.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
FIG 7 - The data point ‘cloud’ defining the liner surface in situ.
(Courtesy of Scanalyse, Perth, WA).
RME understands that the better DEM also include empirical
factors. Perhaps DEM predicts accurately the behaviour of large
particles and balls toward the outside of the charge while the
empirical factors better describe the complex interior of the
charge.
At present, 3D DEM use idealised liners. Now there is an
opportunity to apply these 3D laser scans of mill’s liners at (say)
one month intervals from ‘new’ through to ‘worn out’ and provide
the foundation of a much more accurate DEM for any mill.
Liner wear predictive models
These 3D laser imaging techniques should provide a better
foundation for further developing liner wear predictive models.
RME’s vision –liner profile to profitability
commercial model
RME’s vision is as follows:
• Laser scan the liners of any mill, in situ, at (say) one month
intervals from ‘new liners’ through to ‘worn out’ liners.
• During this time, measure the power draw, mill feed rate, feed
size distribution (continuously, online) while sampling the mill
discharge product before and after the monthly laser scan.
• Input the laser scanning model into the mill’s discrete
element model (or DEM/empirical hybrid model) to achieve
a much more accurate model using real 3D mill liner profile
data.
• Adjust the DEM (or DEM/empirical) to match model
predictions to actual data (mill feed size distribution versus
power draw and mill discharge particle size distribution
sample results). Once correlated the final DEM can be
referred to as the ‘validated DEM’ for that mill’s grinding
performance.
• Using the validated DEM, experiment off-line, in the virtual
world, with new liner profiles with the goal of optimising
grinding performance.
• Using grinding performance, predicted liner life, liner cost,
relining downtime, etc develop the commercial model.
• Iterations of the combined technical (DEM) and the
commercial model will reveal a maximum profitability
optimum, as a function of liner profile and liner exchange
program.
117
J RUSSELL
RME’s contribution to optimising liner profiles
and liner exchange programs including future
relining technology developments
If RME’s vision for ‘profit maximised’ liner profiles, exchanged
more frequently, proves to be correct, the maximum benefit
(annual profitability) will be achieved by replacing ‘profit
maximised’ liners, probably more frequently, but with no
additional mill relining related downtime.
For medium to large SAG and AG mills, RME is targeting a 24
hour (feed off to feed on) shutdown every (say) three months,
necessary to exchange profit maximised shell liners.
The technologies necessary to achieve this level of
performance will be:
• powered feed chute extraction/transport/replacement
(existing technology);
• Thunderbolt StormPath – RME’s laser-line-targeted,
manually (not robotically) controlled, recoilless hammer
guidance system, each side of the mill shell (under
development) (refer to Figure 8);
• worn liner lifting tools (existing technology);
• single, large capacity, Mill Relining Machine of 6000 kg to
7500 kg capacity (existing technology); or
• twin Mill Relining Machine of +4000 kg capacity each (new
technology due for release mid 2008) (refer to Figure 10).
FIG 8 - Solidworks 3D image of RME’s Thunderbolt StormPath.
FIG 9 - RME’s (and the world’s) largest capacity Mill Relining
Machine featuring 7250 kg/16 000 lb liner capacity, liner elevator,
diesel engine, 4 × 4 drive and remote radio control supplied to PT
Freeport, Indonesia for 34’ and 38’ SAG mills.
118 Fremantle, WA, 19 - 21 March 2007
FIG 10 - RME’s Twin Machine Mill Relining System currently being
designed for Escondida’s Laguna Seca plant 38’ SAG mill.
The twin Mill Relining Machine solution provides the
opportunity for a most interesting workplace productivity
experiment using two competing teams, the Red team and the
Blue team, each supplied with separate but otherwise identical
resources.
The rules for this relining competition will be based on the
sport of soccer. A performance scoring system is being
developed and safety is ensured by ‘Safety Umpires’ with
‘yellow card’ and ‘red card’ warnings and send-off authorisation.
On this side of the world, both PT Freeport, Indonesia and
Newcrest Cadia will be pursuing minimum mill relining times
using their single, large capacity Mill Relining Machines of
7250 kg and 6000 kg respectively.
RME looks forward to working with all three customers to
maximise mill relining system performance.
CONCLUSION
RME has developed the world’s fastest and safest mill relining
method based on our unique worn-liner-removal, new-liner-
placement Mill Relining System. Relining ease and speed
allows more frequent liner exchange without incurring punitive
downtime. Very large capacity machine or twin machine systems
combined with Thunderbolt StormPath and other technologies
promise significant further gains in the near future.
The opportunity therefore exists to reduce mill throughput
losses due to new liner profiles which have been developed for
maximum life. A new operational paradigm is now possible
where ‘profit maximised liners’, exchanged more frequently,
deliver an optimum commercial result.
To minimise the risk of a costly experiment, RME believes
that the current DEM/empirical models of grinding mill
processes, combined with the new laser imaging technologies
and other existing technologies, can produce an accurate and
Validated DEM of grinding processes including the significant
effects of liner wear.
Once a reliable validated DEM is achieved, interesting
‘off-line’ virtual experiments can be undertaken to investigate the
effects of various liner profiles on mill throughput, comminution
efficiencies and profit.
The results of this work may indicate less aggressive liner
profiles which may require more frequent exchange.
Therefore, for each unique circumstance, RME believes there
is a perfect liner profile solution, certainly obscured through the
data ‘noise’ of a practical operation, but a solution which delivers
the best economic result for that operation.
Ninth Mill Operators’ Conference
 ADVANCED MILL RELINING — A KEY OPERATIONAL CONTROL (NOT SIMPLY A ‘MUST DO’ MAINTENANCE FUNCTION)
RME’s professional goal therefore, is to see the creation of a
grinding plant commercial model which can predict the optimum
mill liner design and liner exchange program necessary to
maximise annual throughput and comminution efficiency,
leading to maximum profitability.
Finding the grinding plant economic ‘sweet-spot’ (through
liner design and liner exchange program) will maximise whole
project economic performance.
Mill liner profiles and mill relining speed become key
operational and profitability controls and not simply ‘must do’
maintenance costs and functions.
RME’s fast, safe Mill Relining Systems have already changed
the relining time expectation paradigm by halving previous reline
durations. RME’s latest fast, safe Mill Relining Systems provide
the key to rapid, albeit more frequent, changes of grinding
performance optimised liners which maximise plant profitability.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
REFERENCES
Russell, J, 2002. New developments and practices in lifters and liners,
Paper presented to Crushing and Grinding Conference, Kalgoorlie,
WA. Available from: <http://www.rmeAus.com>.
Russell, J, 2005. Advanced grinding mill relining for process
metallurgists and management, paper presented to Randol Pacific
Conference, Perth, WA.
Russell, J, 2006a. Advanced grinding mill relining methods, in
Proceedings Metallurgical Plant Design and Operating Strategies
2006, pp 47-63 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
Russell, J, 2006b. Advanced grinding mill relining for process
metallurgists and management, paper presented to SAG Conference,
Vancouver, Canada.
119
The Effect of Aspect Ratio on the Grinding Efficiency of Open
and Closed Circuit AG/SAG Mills
S Morrell1

ABSTRACT proportion to length, eg if the mill length is doubled throughput

The aspect ratio (mill diameter/mill length) of autogenous and semi
autogenous mills varies widely and can be as high as three or as low as
0.5. The literature on the subject of what effect aspect ratio has on
grinding mill performance or energy efficiency is almost non-existent.
However, at the design stage of many grinding circuits the subject is often
vigorously debated amongst metallurgists. In this paper the performance
characteristics of mills with different aspect ratios are examined to
determine whether process performance characteristics are a function of
aspect ratio and whether high aspect ratio mills are any more or less
energy efficient than low aspect ratio units. Data from both pilot and
full-scale mills are used for this purpose in open and closed circuit
configurations.
does not double – it increases by a much smaller amount.
However, as a consequence of the specific energy increasing, the
lower aspect ratio provides a finer product as Figure 2 shows. The
shapes of the distributions from each aspect ratio do not appear to
differ appreciably from one another as Figure 3 indicates.
14
12
10
8

kWh/t
6

INTRODUCTION 4
mean
Autogenous (AG) and semi-autogenous (SAG) milling most 2 6 mm screen
3 mm screen
likely had their origins in the first decade of the 20th century 0
with the literature suggesting that simultaneous parallel 1 1.5 2 2.5 3
developments probably took place on two continents. Digre Aspect ratio (D/L)
(1969) cites experiments in 1905 that lead Jackson to pioneer
pebble milling of gold ores in South Africa whilst Harlowe
FIG 1 - Aspect ratio versus specific energy (data after Adam
Hardinge (1955) references work by his father in 1908 in which
he experimented with autogenous milling of magnetite in North and Hirte, 1973).
America. What subsequently sprang from these experiments was
not only the rise of AG/SAG milling to its current dominance in 2500
circuit design but also the development of two schools of thought mean
concerning aspect ratio. The early developments in South Africa 2000
6 mm screen
3 mm screen
spawned the low aspect ratio evolutionary branch, whilst in
P80 (microns)

North America design took the high aspect ratio route. 1500
Interestingly, despite the fact that 100 years have passed since the
beginnings of AG/SAG milling, there is almost no relevant 1000
literature on the subject of aspect ratio, particularly with
reference to throughput capacity, specific energy and energy 500
utilisation efficiency. This is despite the fact that the issue of
aspect ratio is one of the most hotly debated subjects during the 0
design stage of many comminution circuit engineering studies. 1 1.5 2 2.5 3
This paper attempts to make sense of what the differences are Aspect ratio (D/L)
between high and low aspect ratio mills from a purely capacity
/energy efficiency standpoint. FIG 2 - Aspect ratio versus P80 (data after Adam and Hirte, 1973).

LITERATURE
100
Arguably the single most important study on the subject of
aspect ratio was carried out by Adam and Hirte (1973). The
study involved the use of a 5.5 ft diameter pilot mill with a
minimum length of 2.2 ft extendable to 4 ft. This effectively
enabled tests to be carried out with aspect ratios (D/L) of 2.5 and
% passing

1.36 respectively. The mill was operated in closed circuit with a

pebble crusher and screen and was used in autogenous mode to
grind magnetite. The test results are summarised in Figures 1 and
2, which show plots of kWh/t and P80 against aspect ratio. Tests
were run with closing screen sizes of 3 mm and 6 mm and hence
the data for each screen are shown separately. Apart from the
short mill
change in aspect ratio nothing else was altered in the design or
long mill
operation of the mill. The results clearly show that as aspect ratio
10
decreases the specific energy increases (Figure 1). In terms of
0.01 0.1 1 10
capacity this means that the throughput does not increase in
size (mm)

1. Director, SMCC Pty Ltd, 29 Camborne Place, Chapel Hill Qld 4069. FIG 3 - Aspect ratio versus product size distribution (data after
Email: steve@smccx.com Adam and Hirte, 1973).

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 121

S MORRELL

The lower aspect ratio pilot mill therefore appears to trade
throughput for grind size. Hence, if full-scale mills behave in a
similar manner it can be inferred from these results that for a
given ore a high aspect ratio mill will provide a higher
throughput at a coarser grind than a low aspect ratio mill
operating with the same power draw, per cent ball load, per cent
filling, per cent of critical speed, grate configuration and open
area fraction. The data further indicate that from an energy
utilisation efficiency viewpoint there is no significant difference
between aspect ratios, as the operating work indices were the
same for both the high and low aspect ratio tests. Adam and Hirte
did not look at operating work indices but chose to analyse their
data in terms of the fines produced. They reached the same
conclusion, giving the following quote in their paper:
... (the) kWh per ton of minus 200 mesh material
produced is essentially the same in the long mill
as the short mill.
FULL-SCALE MILL DATA
The value of pilot tests such as those described in the previous
section is that aspect ratio can be studied whilst keeping all other
variables constant, therefore obtaining unequivocal results
concerning its influence. In particular, feed ore characteristics are
kept constant. With full-scale mills such control is not usually
possible and hence comparisons are made difficult by having to
use data from different sites with widely different operating
conditions on top of differing aspect ratios. The solution to this
problem revolves around finding a data reduction process that is
able to remove the influences of all relevant variables so that the
impact of aspect ratio can be studied in isolation. Such a data
reduction technique can be found in Equations 1, 2 and 3. These
equations are used by SMCC Pty Ltd to estimate the specific
energy of AG/SAG mills under all normal operating conditions,
eg ball load, mill speed, feed size and ore hardness and for all
normal aspect ratios and mill sizes (Morrell, 2006). The
equations are semi-empirical and are based on data from 43
different circuits and 70 different ore types. The range of
design/operating parameters covered by these data is shown in
Table 1.
T80 = 80 per cent passing size of AG/SAG mill circuit
DWi = drop-weight index
J = volume of balls (per cent)
φ = mill speed (per cent of critical)
f(Ar) = function of mill aspect ratio
a,b,c,d,e,f,g = constants
K = function whose value is dependent upon whether a
pebble crusher is in circuit
W = MiK x2 ( f (x 2 )
− x1
f (x 1 )
) (3)
where:
W = specific energy (kWh/t)
K = constant chosen to balance the units of the equation
Mi = index related to the breakage property of an ore (kWh/t)
x2 = 80 per cent passing size for the product
x1 = 80 per cent passing size for the feed
The predicted specific energies shown in Figure 4 were
calculated using Equations 1 - 3, in which there is a function
related to aspect ratio, though there are no parameters which
represent different energy efficiencies for particular mill designs
or operating conditions, ie it is assumed that the energy
efficiency is the same for all mills/circuits. The correlation is
very good, error analysis showing that there are no underlying
biases. In contrast to this result Figure 5 shows the correlation
using a similar equation to Equation 1 but with no allowance for
aspect ratio. The results are very different; the scatter being quite
pronounced in places, illustrating that aspect ratio has a
significant influence with respect to the specific energies of the
mills in the database. The typical response of this relationship is
shown in Figure 5 and can be seen to work in a similar manner to
the way that the Adam and Hirte data indicate occurs in pilot
mills.

45
TABLE 1
40
Range of circuits/ores.
35
Max Min
observed kWh/t

sg 4.63 2.5 30

DWi kWh/m3 14.2 1.7 25

Bond Wi kWh/t 26 9.05 20

F80 microns 176 000 19 400
15
P80 microns 427 20
10
Diameter m 12.02 3.94
Length m 9.5 1.65 5

Ball load % 25 0 0
0 5 10 15 20 25 30 35 40 45
Speed % 90 58 predicted kWh/t
Aspect ratio (D/L) 3.0 0.5
FIG 4 - Observed versus predicted specific energy with aspect

(
= K.F80 .DWi . 1 + c(1 − e ))
−1 ratio relationship.
− dJ
. φ .f(Ar )
a b e
S (1)

The data in Figure 5 also indicate that without an aspect ratio

g. S
T80 = f − b
(2) relationship the equation still works well in some cases. This is
DWi due to two reasons. The first is that the majority of mills in the
database have an aspect ratio of about two – a reflection of how
where: popular this mill design is. As so many mills in the database have
the same aspect ratio, an equation with no allowance for
S = specific energy at the pinion variation in aspect is obviously likely to work equally well in
F80 = 80 per cent passing size of the feed these cases as an equation that does have an aspect ratio term.

122 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 THE EFFECT OF ASPECT RATIO ON THE GRINDING EFFICIENCY OF OPEN AND CLOSED CIRCUIT AG/SAG MILLS

45 14 3.5

40 12 3

10 2.5
35

P80 (mm)
kWh/t
observed kWh/t

8 2
30
6 1.5
25
4 1
20
2 kWh/t 0.5
P 80
15
0 0
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
10
Aspect ratio (D/L)
5

0 FIG 6 - Typical response of Equations 1 and 2 to changes in

0 5 10 15 20 25 30 35 40 45 aspect ratio.
predicted kWh/t

45
FIG 5 - Observed versus predicted specific energy without aspect
ratio relationship. 40

35
The second reason is that some of the mills in the database are
30

single stage. The aspect ratio relationship is only relevant to
open circuit mills where throughput and grind size vary in
association with one another. In single stage mills grind size is
independent of aspect ratio as it is controlled by the action of
the classifier. Specific energy is also controlled in one sense by
the classifier, in that the P80 it produces dictates the energy
requirement, in conjunction with the throughput. Looked at
another way, in a single stage high aspect ratio mill its initial
tendency (ie when feed is first introduced into the mill circuit)
will be for it to give a higher throughput and coarser discharge.
However, due to the action of the classifier the coarser part of
the discharge will be returned to the mill causing the throughput
to be reduced slightly to accommodate the recycle load.
Eventually the throughput will stabilise once these actions
come to equilibrium. If the mill were low aspect ratio the initial
tendency would be for a lower throughput but finer discharge.
This time the classifier would cause less material to be recycled
back to the mill and hence corrective action in reducing the new
feed would be less than in the high aspect situation.
Equilibrium would eventually be reached at exactly the same
throughput and power draw, providing the final grind was the
same as that obtained in the high aspect ratio case. However, in
the low aspect ratio case it can be expected that the recycle load
would be lower. Overall energy efficiency would be the same,
as the Adam and Hirte data indicate should be the case. The only
situation where this similarity in overall performance may not
occur is where a very low aspect ratio mill was chosen in single
stage mode to grind to a relatively coarse grind. It is possible that
by the time the feed were discharged from the mill it would have
been overground to the point where the relatively coarse target
P80 could not be achieved by the classifier (it would only be able
to obtain a P80 finer than the target). The overall energy
efficiency would be no different; the mill would simply be
unable to obtain its desired target grind and have a higher
specific energy than the high aspect ratio alternative.
To illustrate the single stage situation further, the single stage
mill data were separated from the other data in the database.
A wide range of aspect ratios are included in this data set (from
0.5 - 2). Equations 1 - 3 were then used both with and without
an aspect ratio relationship to predict the specific energies of
each of the mills. These were then compared to the observed
values (Figure 7). As can be seen there is no significant
difference between the two approaches, further supporting the
assertion that the energy efficiency is not a function of aspect
ratio.
observed kWh/t
25
20
15
10
without aspect relationship
5 with aspect ratio relationship
0
0 5 10 15 20 25 30 35 40 45
predicted kWh/t
FIG 7 - Observed versus predicted specific energy for single stage
AG/SAG mills; with and without an aspect ratio term.
PILOT DATA
Although the Adam and Hirte (1973) test work used a variable
length pilot mill, such installations are very rare and in most
cases a standard 6 ft × 2 ft (D × L) mill is used for pilot studies.
Due to our much improved understanding of how AG/SAG mills
work and in particular how to characterise feed ore, pilot test
programs are not as common as they used to be and hence data
are becoming increasingly scarce. However, over the last six
years the author has collected data from nine pilot studies where
subsequently a full-scale plant has been built. Data were then
collected from these plants and compared with the pilot data. In
the context of this paper, the value of doing so is that the pilot
tests were carried out using 6 ft × 2 ft mills which have an aspect
ratio of three. Most full-scale mills tend to have aspect ratios of
two or lower and hence it is possible that in comparing the pilot
and full-scale data aspect ratio effects may become apparent.
As a first approach the raw pilot net kWh/t was plotted against
the raw full-scale kWh/t at the pinion (Figure 8). To arrive at this
latter figure power at the pinion was assumed to be 93.5 per cent
of the motor input power. The data indicate a definite correlation
though with the full-scale values being on average 21 per cent
larger than the pilot ones. Scatter in the data is also apparent and
is attributed to the following causes:
• the full-scale plant data did not always reflect exactly the
same ore as treated during the pilot study, though as far as
possible full-scale plant data were collected during periods
where the ore was as similar as possible;

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 123

S MORRELL

16 16

15 15
y = 1.2099x y = 1.0932x
Full scale kWh/t @ pinion

Full scale kWh/t @ pinion

14 14
13 13

12 12

11 11
10 10

9 9
8 8

7 7
6 6
6 8 10 12 14 16 6 8 10 12 14 16
Pilot net spec energy (kWh/t) Pilot net spec energy (kWh/t)

FIG 8 - Pilot net specific energy versus full-scale specific energy FIG 9 - Pilot net specific energy versus full-scale specific energy at
at pinion without aspect ratio correction. pinion with aspect ratio correction.

• conditions such as screen/trommel apertures, mill speed and 40

ball load were not always identical; and
• feed sizes differed.
The pilot data were then modified using the same aspect ratio
adjustment that is contained in Equation 1. The resultant plot is
shown in Figure 9 and indicates that the offset between the pilot
and full-scale results has reduced considerably to nine per cent. It
should be remembered that the pilot data relate to net conditions
and hence reflect the power used to grind the ore only, ie it is the
motor input power less motor, gearbox, drive train and bearing
losses. In contrast, the full-scale data relate to the estimated
power at the pinion drive and hence include the losses at the
pinion and due to bearing friction. These losses could
conceivably account for two to three per cent of total power,
leaving six to seven per cent as the true difference between the
pilot (corrected for aspect ratio) and full-scale specific energies.
Given that pilot tests are conducted under near-perfect conditions
with respect to control and feed consistency and the full-scale
data include the usual effects of variability, the six to seven per
cent difference is quite plausibly attributable to the influence of
control. This is not to say that the differences reflect higher
energy efficiency in the pilot mill, only that the kWh/t is lower
and hence the throughput is proportionately higher than at full
scale. The expectation is that the full-scale mill would have a
higher kWh/t than the pilot mill but would produce a slightly
finer grind. This is supported by the data in Figure 10, which
shows the operating work indices (as described by Morrell, 2006)
of the pilot and full-scale mills. As can be seen the full-scale
operating work indices are about three per cent higher than the
pilot ones. This difference is in line with earlier estimates of the
difference between the full-scale power at pinion and the pilot net
values, ie the difference is attributable to the differences in what
the power measurements in the pilot and full-scale mills relate to.
The implication of this conclusion are significant in that it
suggests that it is not appropriate when designing a full-scale
mill using pilot data to assume that the full-scale mill will have
the same specific energy as the pilot unit, or that it will be a
simple multiple of it regardless of the dimensions of the full-
scale mill. In particular, if a low aspect ratio mill is to be used
and the pilot tests were carried out in the usual high aspect ratio
6 ft × 2 ft mill the pilot specific energy may need to be inflated
by a considerable amount to account for the affects of geometry.
CONCLUSIONS
According to controlled pilot tests by Adam and Hirte (1973), as
mill length increases and aspect ratio therefore decreases the
specific energy of AG/SAG mills increases, all else being equal.
At the same time the product size becomes finer. The overall
energy efficiency, however, remains constant.
35
MOWi f scale pinion
30
25
y = 1.0292x
20
15
10
5
0
0 5 10 15 20 25 30 35 40
MOWi pilot net
FIG 10 - Pilot net operating work index versus full-scale operating
work index at pinion.
Analysis of a large database of industrial mills indicates that
these phenomena are also apparent with open circuit mills at full
scale. Hence a low aspect ratio mill operating with the same
power and operating conditions as a high aspect ratio machine
will have a lower throughput and produce a finer grind,
consistent with a similar energy utilisation efficient.
In single stage closed circuit operation, high and low aspect
ratio mills can be expected to have the same energy efficiency and,
providing they are operated under similar conditions of power
draw, ball load, speed, etc and grind to the same P80, will have
similar throughputs and specific energies. Recycle loads in the
high aspect ratio mill circuit are likely to be marginally higher.
Comparisons between pilot data using 6 ft × 2 ft mills and
full-scale mills with lower aspect ratios treating similar ores also
show signs of aspect ratio effects. These are apparent from the
differences between the specific energies of the pilot and
full-scale mills. By applying an aspect ratio adjustment to the
pilot data the resultant specific energies are similar to those
observed in associated full-scale mill circuits.
REFERENCES
Adam, H W and Hirte, D F, 1973. Autogenous grinding, the long and the
short of it, AIME Annual Meeting, Chicago, Illinois, SME of AIME.
Digre, M, 1969. Autogenous grinding in relation to abrasion conditions
and mineralogical factors, in Proceedings Auto Grinding Seminar,
Trondheim, A1.
Hardinge, H, 1955. Making rock grind itself, Eng and Min J, 156:84-90.
Morrell, S, 2006. Design of AG/SAG mill circuits using the SMC test, in
Proceedings International Conference Autogenous and Semi
Autogenous Grinding Technology, Vancouver, UBC.

124 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

Liner Optimisation to Improve Availability of the Ridgeway
SAG Mill
M Weidenbach1 and P Griffin2

ABSTRACT
Mill relining is a major contributor to overall plant downtime. Reducing
the number of mill liner pieces removed from and replaced in the mill has
a direct impact on the overall reline time and in turn, mill availability. To
improve the mill availability of Ridgeway Operations’ 32 foot SAG mill,
a number of initiatives were implemented in terms of liner design. These
included the conversion of the shell from 40 to 30 rows, reducing the
number of pieces from 80 to 60 each shell reline and reducing the number
of pieces from 48 to 32 per feed end reline.
In addition, modifications were made to the discharge end of the mill
to extend the life of the outer pulp lifter and discharge cone, thereby
reducing the frequency of change out. The standard radial outer pulp lifter
was converted to an angled design. The discharge cone was converted to
the BullnoseTM design supplied by Bradken.
This paper discusses the design changes made and the expected
benefits to the Ridgeway Operations in terms of mill availability
improvements.

INTRODUCTION
FIG 1 - Ridgeway operations’ 32 foot SAG mill.
Newcrest’s Cadia Valley Operations (CVO) are located
approximately 25 km from the city of Orange in New South
Wales, Australia. The area has a long history of mining dating RIDGEWAY SAG LINERS PRE-OPTIMISATION
back to the mid 19th century. The Cadia open cut mine was Table 1 summarises the liner arrangement and reline time
developed in 1997 and processing operations commenced in June durations before the optimisation project began. The change out
1998. The Cadia concentrator has one of the world’s largest SAG time includes removal of bolts, knocking in of old liners,
mills (40 foot in diameter) with 20 MW of installed power, mucking out of liners, installing new liners, torquing up and
treating ore at a rate of 17 Mtpa. inching the mill for the next cycle.
The Ridgeway underground mine, discovered in November
1996 containing 50 Mt of reserves and averaging 2.4 g/t gold and OPTIMISATION OF SHELL LINERS
0.8 per cent copper, was commissioned in March 2002 (Hart
In a shell reline conversion of the shell from 40 rows to 30 rows
et al, 2003). The Ridgeway concentrator operates with a 32 foot
had the potential to reduce the number of liners involved from 80
SAG mill (Figure 1) with 6.5 MW of installed power, treating ore to 60. The existing design is shown in Figure 2.
at a rate of 5.6 Mtpa.
Over the last four years, to meet the demands of improving
reline efficiency and reducing downtime, significant liner
optimisation has occurred in the Ridgeway SAG mill.
The SAG Mill reline time usually drives the duration of major
plant shutdowns. Reducing downtime and improving mill
availability is a key element to improving cash flow and
profitability for CVO. Mill relining is a significant contributor to
overall downtime and decreasing the number of pieces removed
and replaced in the mill (whilst maintaining the wear profile)
reduces reline time and may increase the time between liner
change outs.
Relining equipment associated with the Ridgeway SAG mill
includes a Russell Mineral Equipment (RME) 3.5 tonne, 7 axis
liner handler and the use of RME Thunderbolt Liner Bolt
Removal Tools (750’s). Kaltech contracting perform the relines
on behalf of Cadia Valley Operations.
FIG 2 - Forty row shell liner design.

Increasing the width of the liners enabled the change to 30

1. MAusIMM, Metallurgical Superintendent, Newcrest Mining
Limited, Cadia Valley Operations, Private Mail Bag, South Orange
NSW 2800. Email: mark.weidenbach@newcrest.com.au
2. MAusIMM, Ore Treatment Manager, Newcrest Mining Limited,
Cadia Valley Operations, Private Mail Bag, South Orange NSW
2800. Email: paul.griffin@newcrest.com.au
rows (Figure 3) to be implemented with the existing drill hole
pattern on the mill shell. The increase in liner mass from
1.85 tonnes to 2.7 tonnes was well within the limitations of the
3.5 tonne liner handler. Total shell liner mass increased from
148 tonnes to 165 tonnes in line with a slight change in lifter
height and profile design.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 125

M WEIDENBACH AND P GRIFFIN

TABLE 1
Liner arrangement preoptimisation project.

Liner location Number of Weight Change out time per Wear life Annualised change
liners in set (tonnes) set (hours) (months) out time (hours)
Feed end
Inner feed end 16 1.5 4.53 3.8 14.3
Outer feed end 32 1.4 9.07 3.8 28.6
Shell
Shell liner (40 row) 80 1.85 22.1 5.3 50
Discharge end
Outer grates (radial) 32 1.1 8.16 7 14
Outer pulp lifter (radial) 32 1.5 20.4 8 30.6
Pulp discharger 8 3 12 12 12
Inner pulp lifter 16 1.9 10.2 45 2.7
Inner discharge end 16 1.2 4.53 14 3.9
Total 156.1

TABLE 2
Liner arrangement post conversion to 30 row shell.

Liner location Nunber of Weight Change out time per Wear life Annualised change
liners in set (tonnes) set (hours) (months) out time (hours)
Feed end
Inner feed end 16 1.5 4.53 3.8 14.3
Outer feed end 32 1.4 9.07 3.8 28.6
Shell
Shell liner (30 row) 60 2.7 16.6 5.3 37.6
Discharge end
Outer grates (radial) 32 1.1 8.16 7 14
Outer pulp lifter (radial) 32 1.5 20.4 8 30.6
Pulp discharger 8 3 12 12 12
Inner pulp lifter 16 1.9 10.2 45 2.7
Inner discharge end 16 1.2 4.53 14 3.9
Total 143.7

An additional benefit is that the number of bolts will be

FIG 3 - Thirty row shell design.
Reducing the number of liner pieces on the shell to 60 is
expected to provide a 5.5 hour reduction in reline time for a shell
reline. The annualised reduction in reline time is expected to be
12.4 hours (50 hours reducing to 37.6 hours) as shown by
comparing Table 2 with Table 1.
reduced from 360 to 240 with the reduction in pieces. Bolt stress
is not considered an issue based on a similar change that was
made in Cadia’s 40 foot SAG mill in 2003. Reducing the bolts by
a similar amount with a reduction in rows from 78 to 52 went
without incident.
Eight rows of trial liners were installed in April 2006 with a
complete changeover planned to occur in March 2007. The
optimisation to the 30 row profile enabled a design change to
occur from the traditional top hat design with a 25 degree face
angle (Figure 2) to the stepped angle lifter design shown in
Figure 3. This design consisted of two face angles consisting of
29.5 degrees at the base and 39.5 degrees at the top. To reduce
the risk of plate breakage with the increased lifter spacing, a
130 mm high mini lifter was incorporated into the design. The
change in face angle was aimed at allowing faster ramp up after
relines (reduced ball/liner impact with faster mill rotation)
creating the potential for greater throughput. Further, it allowed
additional height on the lifter (295 mm for the stepped angle
lifter design verses 246 mm for the top hat design) allowing for a
potential increase in wear life. The trial indicated wear rates did
improve and it is likely an increase in wear life from 23 to
27 weeks is possible. Associated benefits with a reduced degree
of packing are also expected.

126 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 LINER OPTIMISATION TO IMPROVE AVAILABILITY OF THE RIDGEWAY SAG MILL
OPTIMISATION OF FEED END LINERS
Wear on the feed end liners was occurring at the join between the
inner and outer liners as shown in Figure 4, in line with the
operating mill charge level.
High wear zone
FIG 4 - Feed end liner design post optimisation.
The result of this was both the inner and outer feed end liners
had to be changed out with only the areas of the liner either side
of the join worn.
It was proposed to extend the length of the outer liner and
shorten the inner liner as shown in Figure 5, to eliminate the need
to change the inner liner.
Wear
zone
FIG 5 - Feed end liner design post optimisation.
This change was implemented in May 2005 and resulted in the
shortened inner liner receiving minimal wear as it now sat out of
the charge. The number of pieces requiring change out on the
feed end during relines was reduced from 48 to 32 (only outer
liners changed). The annualised reduction in reline time was
14.3 hours as shown by comparing Table 3 with Table 1. An
additional benefit was reduction in the number of bolts on the
feed end from 128 to 112.
OPTIMISATION OF DISCHARGE END LINERS
To optimise the discharge end of the mill, innovation in design to
improve the life of the liners was required. The pulp discharger
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
was converted from standard chrome moly material to a
BullnoseTM rubber white iron composite (Figures 6 and 7), based
on trials in the 40 foot Cadia mill, with an expectation to double
the wear life. These were installed in September 2006.
White Iron
Rubber
FIG 6 - Bullnose discharge cone.
FIG 7 - Bullnose assembly.
The advantages of the BullnoseTM design were the use of white
iron in high wear areas, the reduction in weight in low wear areas
by the use of rubber, the reduction in support rings from two to
one compared to the original pulp discharger (Figure 8) and the
utilisation of a keyed installation system that enables simple
fitment of the final piece. The design incorporated a left- and
right-hand key BullnoseTM. This ensured that the key discharge
cone could be installed up to the mating parts without
interference or the requirement to remove adjoining inner pulp
lifters and inner discharge end liners (Faulkner, 2005). In
combination, these factors are expected to result in a reduced
number of liner change outs, less pieces and quicker change outs
as shown by comparing the predicted annualised BullnoseTM
change out time at 5.1 hours in Table 4 verses the original of 12
hours as shown in Table 1.
In addition to the conversion to the BullnoseTM, the outer pulp
lifter and grate were converted from a radial design to an angled
design in March 2005. The original radial design is shown in
Figure 9.
127
M WEIDENBACH AND P GRIFFIN

TABLE 3
Liner arrangement post modification to feed end liners.

Liner location Number of Weight Change out time Wear life Annualised change out
liners in set (tonnes) per set (hours) (months) time (hours)
Feed end
Outer feed end (extended) 32 2 9.07 3.8 28.6
Shell
Shell liner (30 row) 60 2.7 16.6 5.3 37.6
Discharge end
Outer grates (radial) 32 1.1 8.16 7 14
Outer pulp lifter (radial) 32 1.5 20.4 8 30.6
Pulp discharger 8 3 12 12 12
Inner pulp lifter 16 1.9 10.2 45 2.7
Inner discharge end 16 1.2 4.53 14 3.9
Total 129.4

TABLE 4
Liner arrangement post modification to discharge end liners.

Liner location Number of Weight Change out time Wear life Annualised change out
liners in set (tonnes) per set (hours) (months) time (hours)
Feed end
Outer feed end (extended) 32 2 9.07 3.8 28.6
Shell
Shell liner (30 row) 60 2.7 16.6 5.3 37.6
Discharge end
Outer grates (angled) 32 1.1 8.16 7 14
Outer pulp lifter (angled) 32 1.5 20.4 18 13.6
Pulp discharger (bullnose) 8 2.5 10.2 24 5.1
Inner pulp lifter 16 1.9 10.2 45 2.7
Inner discharge end 16 1.2 4.53 14 3.9
Total 105.5

FIG 8 - Original pulp discharger.

FIG 9 - Original radial design pulp lifter.

The radial pulp lifter suffered from high wear as a result of
backflow issues. With the design of the pulp lifter chambers
consisting of long/short/medium/short/long vanes in sequence,
each outer pulp lifter on the long vane was required to be
changed on an eight month cycle. Conversion to the angled
design in Figure 10 significantly reduced this issue by enabling
earlier pulp discharge, increasing the life of the pulp lifter from
eight to 18 months. The impact of this change on annualised
reline time was a 17 hour reduction (30.6 hours to 13.6 hours) as
shown by comparing Table 4 to Table 1.

128 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 LINER OPTIMISATION TO IMPROVE AVAILABILITY OF THE RIDGEWAY SAG MILL

CONCLUSION
Liner optimisation in the Ridgeway SAG mill has reduced the
annualised reline time from 156 hours (Table 1) to 105 hours
(Table 4) through a strategy of reducing the number of liners
required to be placed in and out of the mill and increasing the
wear life of critical liners in the mill that drove plant shutdowns.
The impact of this improvement has been a significant
contributor to increased revenue for Cadia Valley Operations.
Further potential benefits from increasing the shell change out
frequency from 23 to 27 weeks may add further to this reduction.

REFERENCES
Faulkner, C, 2005. Cadia bullnose discharge cone – Trip report, Internal
Report, Bradken Resources Limited.
Hart, S, Griffin, P, Cesnik, F, Gordon, D and Clements, B, 2003. Design
and commissioning of the Ridgeway concentrator, in Proceedings
Eighth Mill Operators’ Conference, pp 63-72 (The Australasian
Institute of Mining and Metallurgy: Melbourne).

FIG 10 - Angled pulp lifter and grate.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 129

Process Design of a Regrind Facility at the Leinster Nickel
Operations to Improve Concentrator Recovery
D R Seaman1, K E Barns2, P J Sharman1, S A Thomas3 and K J Stokes4
ABSTRACT
This paper presents the development of a regrind facility design at BHP
Billiton’s Leinster Nickel Operations (LNO) in Western Australia. The
recovery opportunity was identified from an analysis of plant operating
data together with scanning electron microscope (SEM) measurements
on relevant tailings streams from LNO. A laboratory procedure was
developed to determine the potential recovery improvement of installing a
regrind facility at LNO. The last step in formulating the process flow
sheet and defining the process design criteria was to construct and operate
a pilot plant, which incorporated four banks of flotation cells and an
Xstrata M20 Isa Mill. Pilot plant on/off trials were used to define the
expected metallurgical benefit. A prefeasibility engineering design and
cost estimate was completed. Following a successful internal business
evaluation, a feasibility study is currently underway.
INTRODUCTION
Leinster Nickel Operation is situated approximately 400 km
north of Kalgoorlie in Western Australia. The operation is owned
and operated by BHP Billiton Nickel West Pty Ltd. The
operation was purchased from the Agnew Mining Company in
1989 by WMC Resources, which was subsequently taken over by
BHP Billiton in 2005. A new concentrator was commissioned in
early 1993 with a nominal capacity of 2 Mtpa. This plant has
been optimised over the years and now has a nominal capacity of
2.8 Mtpa. The target P80 of the grinding circuit product is 60 µm.
1. Technical Delivery Group, BHP Billiton Nickel West Pty Ltd, 26
Dick Perry Avenue, Kensington WA 6151.
2. Xstrata Technology, Level 4, 307 Queen Street, Brisbane Qld 4000.
3. MAusIMM, Technical Delivery Group, BHP Billiton Nickel West
Pty Ltd, 26 Dick Perry Avenue, Kensington WA 6151.
4. Leinster Nickel Operations, BHP Billiton Nickel West Pty Ltd.
Fine
Fine Rougher
Scavenger
Plant Feed
Coarse Coarse
Rougher Scavenger
FIG 1 - LNO separate size flotation circuit.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
The majority of the concentrator feed comes from the
underground Perseverance Mine (PUG). Leinster produces a
nickel and iron sulfide concentrate containing pentlandite,
pyrrhotite and pyrite.
Leinster currently operates a separate size flotation circuit
following grinding and flash flotation (details of the flash
flotation circuit were published by Warder and McQuie, 2005).
The separate size flow sheet design is based on the successful
implementation of separate size flotation at BHP Billiton’s Mt
Keith Operation (Senior and Thomas, 2005) and is shown in
Figure 1.
Plant feed is cycloned, underflow and overflow streams report
to coarse and fine roughing and scavenging stages. Rougher and
scavenger concentrate streams are cleaned in a cleaner-cleaner
scavenger circuit as shown.
Coarse composite particles are concentrated in the coarse
scavenger and coarse cleaner scavenger flotation banks. The
development of this circuit has provided the ability to generate a
target stream for enhancing liberation through the installation of
a regrind circuit.
Regrinding of concentrate streams is an efficient use of
energy in the liberation of valuable mineral grains from
composite particles. Pease et al (2006) present the Xstrata Mt
Isa Pb/Zn concentrator flow sheet, which includes one stage of
lead concentrate regrinding and two stages of zinc concentrate
regrinding with the use of Isa Mill technology. Fine grinding
with inert media in a stirred mill is efficient and can have
benefit over the use of steel media in conventional ball mills.
Cesnik et al (2005) recently reported the development of a
regrind circuit upgrade at Newcrest’s Cadia Valley Operations.
A 355 kW Metso Stirred Media Detritor was installed to regrind
all their concentrate streams to less than 38 µm before final
cleaning.
Fine Cleaner
Fine Cleaner
Scavenger
Concentrate Tailings
Coarse Co Cleaner
Cleaner Scavenger
131
D R SEAMAN et al

FLOW SHEET DEVELOPMENT that was derived using pilot and full-scale survey data. This
proposed regrind flow sheet is shown in Figure 3.
Analysis of operational performance data at Leinster highlighted
Part of the flow sheet design process was to determine the
that coarse composite particles contributed significantly to nickel
optimum distribution of residence time between the coarse
losses in the coarse flotation tailings streams. These coarse
cleaner and the coarse cleaner scavenger cells. This was
particles are of too low grade to be recovered into the final
established from a sequence of laboratory, pilot and full-scale
concentrate stream.
trials at LNO. The coarse cleaner scavenger concentrate stream
Liberation characteristics of the two coarse tailing streams (proposed regrind mill feed stream) was open-circuited by
(Figure 2) revealed that there is potential to recover more of temporarily dumping this stream to the sump floor. Table 1
the unliberated nickel composites to the coarse scavenger shows the nickel units and mass flows in key streams for each
concentrate and coarse cleaner scavenger concentrate. The nickel survey. During the open-circuit survey, the recovery of low grade
units shown on the vertical axis represent the amount of nickel in nickel composites from both the coarse scavenger and coarse
these streams as a percentage of nickel in feed to LNO (monthly cleaner scavenger flotation banks was increased by increasing air
composite samples were used for this evaluation). and reagent flow rates and decreasing froth depths. Due to time
The coarse cleaner scavenger tail liberation analysis also shows and sump volume constraints the recovery was not maximised
that there is a large portion of unliberated nickel sulfides in this and it should be possible to further increase the recovery from
stream (greater than 50 per cent of nickel in this stream over CS2). both of these banks once a regrind circuit is in place to treat the
The nickel sulfide composites were not found to be preferentially additional cleaner scavenger concentrate stream.
associated with iron sulfides or non-sulfide gangue.
Currently, the recovery from the scavenger circuit is limited TABLE 1
by the ability of the cleaner circuit to maintain grade, and thus Open circuit trial summary.
there are potentially more nickel units available in the coarse
scavenger stream. Stream Mass Flow (t/h) Nickel units
A batch flotation test was conducted on the coarse cleaner Closed Open Closed Open
scavenger tail and greater than 90 per cent of the nickel in this Coarse cleaner 5.6 16.8 1.8 5.4
stream was recoverable by flotation. The concentrate produced, scavenger concentrate
even after a second stage of cleaning, was of insufficient grade to Coarse cleaner 11.6 15.3 1.4 1.0
report to final concentrate. This is because the composite nature scavenger tail
of the particles in this stream, which is supported by the SEM Coarse scavenger tail 182 166 8.3 6.8
analysis shown in Figure 2. More importantly, this flotation test
Coarse tails (total) 194 181 9.7 7.8
showed that the majority of nickel in the coarse cleaner
scavenger tail is floatable. Consequently the existing coarse
cleaner scavenger concentrate was chosen as a stream suitable Piloting
for regrinding. A similar batch flotation test conducted on the
Laboratory tests were conducted to further develop the flow
coarse scavenger tail also showed that a large portion of those
sheet and reagent requirements for increased recovery in the
nickel composites could be floated to the scavenger concentrate. coarse cleaner scavenger flotation cells. One of the key
A regrind flow sheet was developed based on concentrating parameters determined in the laboratory tests was how best to
coarse composites to the coarse cleaner scavenger concentrate, distribute residence time between the coarse cleaner and coarse
regrinding this stream and then sending the regrind discharge, cleaner scavenger banks. For optimal benefit, it was found that a
consisting of liberated material, to the fines circuit for upgrading. portion of lower grade material currently reporting to the cleaner
The location of where to direct regrind discharge to the fines concentrate should be reground in addition to the coarse cleaner
circuit was determined using a kinetic model of the LNO circuit scavenger concentrate.

Coarse Cleaner Scavenger Tail Coarse Scavenger Tail

1.00 2.00

0.90 1.80

0.80 1.60

non-sulphide non-sulphide
Ni distribution (Ni Units)

0.70
Ni distribution (Ni Units)

1.40
composite composite
FeS composite FeS composite
0.60 1.20

FeS liberated FeS liberated

0.50 1.00

NiS composite NiS composite

0.40 0.80

NiS liberated NiS liberated

0.30 0.60

0.20 0.40

0.10 0.20

0.00 0.00
+75 +53 +CS2 +CS5 +CS6 +75 +53 +CS2 +CS5 +CS6

FIG 2 - Mineral liberation analysis of the coarse tailings streams.

132 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 PROCESS DESIGN OF A REGRIND FACILITY AT THE LEINSTER NICKEL OPERATIONS

Following this laboratory development, a pilot plant was
constructed and operated to fine-tune the design and provide the
necessary process design criteria and mill sizing to enable a
prefeasibility engineering study. The piloting was also necessary
to generate an improved estimate of the potential recovery/grade
improvement that can be gained form installing a regrind circuit
at LNO. These were both necessary steps in developing a
business case for the project.
The pilot plant was constructed from four banks of flotation
cells and a regrind mill together with associated reagent dosing
and slurry transport pumps. The pilot plant was constructed in
such a way that it could replicate both the existing LNO coarse
cleaning circuit as well as a post-regrind flotation circuit (fine
cleaner).
Figure 4 shows a flow sheet of the pilot plant as operated with
the regrind mill in operation. The dotted lines in the figure show
alternate stream destinations.
Figure 4 also shows that the pilot plant was set up in such a
way that the individual cell concentrates from the coarse cleaner
bank could be diverted to either final concentrate or the coarse
cleaner scavenger concentrate stream. This was done to allow for
changes in the residence time distribution between cleaner and
cleaner scavenger duty in the pilot plant. Figures 5, 6 and 7 show
photographs of various components of the pilot plant.
Reagents were added to the pilot plant feed and mill discharge
streams in similar quantities to those used in the full scale plant.
Operation, surveying and mass balancing
The pilot plant was operated on a single shift (day) basis. For each
on/off trial, the pilot plant was allowed more than five residence
times to reach steady state before a survey was conducted.
For all surveys, timed samples of each individual concentrate
stream as well as combined streams and feed and tail streams
were collected, wet and dry weighed and assayed. This allowed
for greater confidence in the mass balancing of results. Each
sample consisted of a full stream cut to avoid any representative
sampling issues.

Fine
Fine Rougher
Scavenger

Fine Cleaner
Fine Cleaner
Scavenger

Plant Feed
Concentrate Tailings

Coarse Co Cleaner
Cleaner Scavenger

Coarse Coarse
Rougher Scavenger

Regrind Mill

FIG 3 - Proposed regrind circuit at LNO.

Plant Co Ro Con
Co Cleaner Scav Final Tail
Plant Co Sc Con

Co Cl Co Cl Scav

ISA M20 Mill

Fine Cleaner

Final Con

FIG 4 - Pilot plant flow sheet showing alternate configurations.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 133

D R SEAMAN et al

The pilot plant was operated as shown in Figure 4 with the

regrind mill in place (on) and then bypassed (off). The trials were
operated in this order to shorten the time taken to stabilise the
pilot plant after a change was made (rather than in a random
order). A P80 of 20 µm was targeted for the regrind mill
discharge. This target was chosen from the liberation
characteristics of the coarse cleaner scavenger tail (Figure 2).

Grade-recovery and selectivity relationships for a

selected survey
Figure 8 and Figure 9 show the metallurgical response of the
circuit with and without the regrinding stage. This test was
chosen from a series of on/off trials as the pilot plant operated
closest metallurgically to the full-scale plant on that day.

16

15

FIG 5 - Photograph showing one of the pilot flotation banks. 14

13

Nickel Grade
12

11

10

9
8

6
0 10 20 30 40 50 60 70 80 90 100
Nickel Recovery
Regrind Re-Clean (No Regrind) Plant

FIG 8 - Grade-recovery relationship with and without regrind.

35

30

25
MgO Recovery

FIG 6 - Photograph of the Xstrata Isa Mill M20 rig.

20

15

10

0
0 10 20 30 40 50 60 70 80 90 100
Nickel Recovery
Regrind Re-Clean (No Regrind) Plant

FIG 9 - MgO to nickel selectivity with and without regrind.

All the figures show cumulative grade and recovery values

FIG 7 - Photograph of the internals of the Isa Mill M20 rig.
The mass balancing was carried out by adjusting all collected
data to obtain the best mass balanced data set. Each data point
was weighted by a standard deviation to allow for differences in
scale of numbers and also for the accuracy in collection and
analysis of each sample. In all, the surveyed data provided robust
data for mass balancing.
down the bank of flotation cells. For both the regrind and no
regrind cases, the first three points (which have the same value)
are the cleaner concentrates reporting directly to final concentrate.
Figure 8 shows that the regrind circuit significantly improves
the grade-recovery relationship of the circuit. The pilot plant
with no regrind mill slightly underperformed the plant on that
day, but there is still a significant improvement of the pilot plant
with regrind over the full-scale plant. Figure 9 shows the
selectivity of nickel to MgO.
Figure 8 and Figure 9 show that the regrind mill improves
selectivity between nickel and MgO species. A similar
improvement was noted in the selectivity of nickel to iron

134 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 PROCESS DESIGN OF A REGRIND FACILITY AT THE LEINSTER NICKEL OPERATIONS

species. This implies that the improved liberation of the particles Mineralogy
in this coarse stream has improved separation between pentlandite
and both pyrrhotite and non-sulfide gangue species. Sized samples from a pilot plant survey (those presented in the
previous section) were analysed by a Mineral Liberation
Analyser (MLA) to determine their liberation characteristics.
Determining recovery improvement Figure 11 shows the liberation characteristics of the regrind feed
One of the required outcomes from the piloting test work was to and discharge streams in the pilot plant.
estimate the recovery benefit of installing a regrind circuit at These figures show a significant increase in the liberation of
LNO. The on/off trials were conducted successfully on five nickel sulfides, particularly with respect to composite particles
different days so that the results could be compared to the from the coarser fractions (>CS3). As a result these particles
full-scale plant over several days. Figure 10 shows a summary of have been liberated and moved into the finer fractions.
all five tests with corresponding points showing the plant Figure 12 shows the nickel grain size distribution derived from
performance on these days. the SEM data for the regrind feed and discharge streams.
The regrind feed stream has a P50 grain size of slightly less
than 20 µm, which corresponds to the P80 grind size selected for
18
re-grinding. The grain size distribution curve of the regrind
16
discharge stream is much steeper than that of the regrind feed.
14
Nickel Grade (%)

12
10 100
8
90
6
4 80
2
70
0
0 20 40 60 80 100 60
Nickel Recovery (%) % passing
50
Plant Regrind Reclean (no regrind)
40

FIG 10 - Grade-recovery relationship of all on/off trials. 30

20
In all cases, the pilot plant underperformed the full-scale plant 10
for the baseline (no regrind case) tests. It was found that the
0
water recovery in the pilot cells was greater than that in the
1 10 100
full-scale plants, resulting in an increased recovery of gangue
particles. Reducing the lip length on the cells assisted in reducing Grain Size µm
water recovery, but efforts to exactly replicate the full scale
Regrind Feed Regrind Discharge
coarse cleaner were unsuccessful. Despite this difference, when
the pilot plant operated with the regrind circuit, there was an
improvement in the grade-recovery relationship over both the FIG 12 - Nickel grain size distribution of regrind mill feed and
baseline run in the pilot plant and the full-scale plant itself. discharge streams.

Co Cl Scav Con -- RG Mill Feed RG Mill D/C

50 50

45 45

40 40
% Nickel of Pilot Plant Feed
% Nickel of Pilot Plant Feed

35 35

30
30
25
25
20
20
15
15
10
10
5
5
0
0 Total <CS6 CS6 CS5 CS4 CS3 CS2 CS1 +53-
Total <CS5 CS5 CS4 CS3 CS2 CS1 +53-75 75

NiS liberated NiS composite Pyrite Pyrrhotite NiS liberated NiS composite Pyrite Pyrrhotite

FIG 11 - Nickel distribution of regrind feed and product streams (relative to pilot plant feed).

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 135

D R SEAMAN et al
This effect is often quoted by inert grinding mill manufacturers
as a benefit over conventional grinding, which would not
typically increase the slope of this relationship, resulting in the
production of more slime material.
Further grinding of this tails material will not produce a
significant amount of recoverable nickel, but rather ultra-fine or
slimes material (due to there being few composite particles
remaining in the tails stream). Thus, it was concluded that a
regrind to 20 µm is adequate to enhance liberation.
Regrind mill sizing
Feed was pumped from the pilot plant concentrate launders or
the LNO concentrator concentrate hoppers to the Isa Mill feed
tank. The feed pump to the mill was turned on and once slurry
was observed discharging from the mill the mill was started. The
feed rate adjusted using the feed pump speed and the valve
recycling feed to the feed tank. Discharge from the mill was
either discharged into the floor drains or was pumped to the
‘fines’ cleaner flotation pilot cells. At the discharge tank a
calibrated ten litre bucket was used to measure slurry flow rate
through the mill. Discharge density from the mill was measured
using a Marcy scale.
While indicative power consumption plots were determined
when the mill was in circuit with the pilot plant, variations in
plant operating conditions made it difficult to produce Isa Mill
signature plots that were used to determine cumulative energy
consumption data.
To produce an Isa Mill signature, plot feed from the pilot plant
(or concentrator) was fed into the mill feed tank. The mill was
determined to be at operating density when the mill feed tank
was at the same density as mill discharge slurry. The feed tank
was continually topped up. When the mill was at operating
density, the product from the mill was retained in the discharge
tank. The mill discharge flow rate was measured and the density
monitored. The power draw and slurry discharge temperature
were monitored continuously. Samples for laser sizing were
taken at approximately midway through the run.
The mill was shut down upon filling the discharge tank.
Following this the mill feed tank was emptied and washed out
with water. The contents of the discharge tank were then pumped
into the Isa Mill feed tank. Once all the product had been
pumped into the feed tank, the discharge tank was hosed out and
the pumping line cleaned with water.
The mill feed pump was then restarted and the Isa Mill started.
The first 45 seconds of product from the mill was pumped onto
FIG 13 - Signature plot of pilot plant coarse cleaner scavenger.
136 Fremantle, WA, 19 - 21 March 2007
the ground to purge the mill of material remaining from the
previous run. Mill product was then pumped to the discharge
hopper as in the previous run with all parameters recorded and
density and sizing samples taken. The procedure was repeated
until the target grind size was achieved. The results from each
pass were used to construct the signature plot for the concentrate.
Throughout the Isa Mill operation Colorado sand was added
through the media addition port to maintain constant power draw.
Grindability
Several signature plots were constructed during the piloting
campaign. Of most interest is the one shown in Figure 13, which
was conducted on the pilot plant coarse cleaner scavenger
concentrate stream at two different densities (32 and 40 per cent
solids).
Of importance, Figure 13 shows that grinding efficiently in the
Isa Mill was not significantly affected when the pulp density of
the feed stream was reduced from 40 to 32 per cent solids.
During the open circuit trial, the pulp density of the coarse
cleaner scavenger concentrate was measured to determine
whether or not a high enough density was achievable to maintain
grinding efficiency in a potential regrind circuit. The density of
coarse cleaner scavenger concentrate during the open circuit trial
was 42 per cent solids.
Table 2 lists the signature plot equations and cumulative
energy consumption results.
TABLE 2
Isa mill energy requirements.
Test condition P80 signature plot Power required to
equation achieve a P80 of 20 µm
(kWh/t)
40% solids Y = 15 850x-2.4 35.0
32% solids Y = 25 100x-2.2 36.7
For design purposes, the higher energy consumption of
36.7 kWh/t was used to determine full-scale requirements.
Scale-up from pilot to full scale was completed on the basis of
energy requirement alone. The flow sheet development exercise
determined a nominal milling requirement of 20 t/h, grinding to a
P80 of 20 µm. Table 3 summarises the milling requirements
around these nominal design values.
Ninth Mill Operators’ Conference
 PROCESS DESIGN OF A REGRIND FACILITY AT THE LEINSTER NICKEL OPERATIONS

TABLE 3 ENGINEERING DESIGN AND FUTURE PLANS

Full scale mill power requirements. Based on the design criteria produced from the piloting exercise
P80 µm Specific Mill energy Mill energy at LNO, a prefeasibility study was commissioned by BHP
energy requirement requirement Billiton. Metplant Engineering Services Pty Ltd were successful
kWh/t kWh (20 t/h) kWh (25 t/h) in tendering for the prefeasibility study, which was completed in
18 46 920 1150 January 2006. The study delivered a ±15 per cent capital and
operating cost estimate for the installation of an Isa Mill M3000
20 37 730 920 at Leinster Nickel Operations. The cost estimates were used
22 30 600 750 together with the expected metallurgical benefits to generate a
business evaluation on the installation of a regrind facility. The
business case was successfully considered by BHP Billiton
Media consumption and type management, and it was decided to continue the project with
Inert grinding media, Colorado Sand, of size 8 × 12 mesh was a feasibility study, which commenced in October 2006 by
used in the pilot program. This sand was recommended for this Lycopodium Limited. This phase is planned for completion in
application by the mill manufacturers. Media addition was March 2007 after which it is expected that an order will be
monitored throughout the pilot program (total of over 500 kWh placed for the regrind mill.
milling), during which the mill consumed an average of
50 g/kWh. This media consumption rate is typical of other fine REFERENCES
grinding applications using similar quality media. Full-scale
media consumption is expected to vary according to media Cesnik, F, Hart, S, Dioses, J, Clements, B and Dunne, R, 2005.
quality fed to the mill. Low-grade concentrator regrind circuit upgrade at Newcrest’s
Cadia Valley Operations, in Proceedings Centenary of Flotation
Table 4 summarises the process design criteria of the proposed Symposium, pp 815-822 (The Australasian Institute of Mining and
regrind facility for Leinster Nickel Operations. Metallurgy: Melbourne).
Pease, J D, Curry, D C, Barns, K E, Young, M F and Rule, C, 2006.
TABLE 4 Transforming flow sheet design with inert grinding – the Isa Mill,
Canadian Mineral Processors Conference.
Summary of process design criteria for the proposed regrind
facility at Leinster Nickel Operations. Senior, G D and Thomas, S A, 2005. Development and implementation of
a new flow sheet for the flotation of a low grade nickel ore,
International Journal of Mineral Processing, 78:49-61.
Feed rate (solids) 20 t/h
Warder, J and McQuie, J, 2005. The role of flash flotation in reducing
Feed density 35% solids overgrinding of nickel at WMC’s Leinster Nickel Operation, in
Feed solids SG 3.6 Proceedings Centenary of Flotation Symposium, pp 931-936 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Feed P80 60 µm
Mill product target P80 20 µm
Mill power requirements 36.7 kWh/t
Media type Colorado Sand 8 × 12 mesh
Media consumption 50 g/kWh
Mill type Isa Mill M3000 (1 MW)

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 137

Advances in the MillMapper Technology for Grinding Mill Liner
Thickness Gauging
J Franke1 and D Lichti2
ABSTRACT
This paper presents recent advancements in the MillMapper software
technology, developed and used for thickness gauging of liners of
cylindrical grinding mills used for mineral comminution. MillMapper
uses calibrated laser scanner data captured inside a mill to determine
thickness at several million points on all liner surfaces with millimetre
accuracy. The paper is structured as follows. First, a review of the
technology and its numerous advantages over existing techniques and the
value it adds to operations are presented. These are followed by
discussion of recently-implemented innovations, including improved
visualisation and inspection tools, automatic ball charge volume
measurement and automatic mill liner volume/weight measurement.
Results from its application at several sites in Australia are presented
throughout.
INTRODUCTION
The need for thickness gauging of liners of cylindrical grinding
mills used for mineral comminution is well known. Liner failure
resulting from excessive wear can have expensive consequences.
Furthermore, mill performance and liner wear are known to be
correlated to lifter bar geometry (Napier-Munn et al, 1999;
Rajamanni et al, 2005). At present, the following techniques are
routinely used to gauge the thickness of mill liners:
1. visual inspection,
2. ultrasonic thickness gauging (for steel liners), and
3. physical inspection (eg nail inserted in rubber liners).
These methods all require an operator to enter the mill to
collect data, which requires that stringent safety procedures be
followed. Furthermore, all are fraught with inaccuracy. In the
case of ultrasonic measurements, for example, the curved liner
surfaces make it difficult to capture thickness measurements that
are orthogonal to the liner back, so the readings can easily be
biased. Furthermore, these methods typically only yield a few
dozen point measurements at unreferenced locations.
Profiling systems that measure mill linear geometry along
single profiles have recently emerged. Both mechanical and
laser-based systems have been developed (eg Chandromohan and
Powell, 2006) but these provide only very limited data coverage
and suffer in terms of accuracy from a number of systematic
error sources.
The MillMapper procedure is completely different from these
techniques. It adopts (to set specifications) state-of-the-art
terrestrial laser scanning technology and dedicated data
processing algorithms to create a detailed three-dimensional
thickness map – not just a few profiles – from millions of
accurate, referenced 3D points for the shell, feed and discharge
ends of a cylindrical mill. MillMapper is in use at a number of
Australian sites including those of major mining operators
ALCOA, BHP Billiton, Barrick, Newmont and KCGM. The
clear advantages and potential of MillMapper over other systems
have been recognised by others (eg Russell, 2006).
1. Scanalyse Pty Ltd, Suite 2, Enterprise Unit 5, 7A De Laeter Way,
Bentley WA 6102. Email: jochen.franke@scanalyse.com.au
2. Department of Spatial Sciences, Curtin University of Technology,
GPO Box U1987, Perth WA 6845. Email: d.lichti@curtin.edu.au
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
This paper first briefly presents laser scanning technology,
then describes the MillMapper procedure with a number of
examples, highlighting its many benefits. An update from the
SAG 2006 paper on MillMapper is then given (Franke, Lichti and
Stewart, 2006) by describing current and future developments that
are transforming MillMapper from a liner-wear assessment tool
to a production optimisation system.
TERRESTRIAL LASER SCANNING
A terrestrial laser scanner (TLS) is an active imaging system that
measures range to an object in a series of uniform increments of
arc, resulting in a three-dimensional point cloud of the object.
All surfaces with line-of-sight from the scanner are measured.
Cooperative targets such as prisms are not required since the
rangefinder is reflectorless; the only constraint is sufficient
surface reflectivity. Range is typically measured by time-of-flight
of a laser pulse or an amplitude-modulated, continuous wave
signal. Range accuracy varies between instruments and typically
ranges from a few millimetres to a few centimetres. Beam
deflection mechanisms include galvanometers, rotating mirrors
and motor-driven instrument head. The angular sampling
resolution is generally programmable with several options
available. The point cloud can comprise in excess of 45 million
points, depending on angular sampling interval and angular field
of view.
THICKNESS GAUGING BY MILLMAPPER
Data acquisition
Several methods of data acquisition exist and are pictured in
Figure 1. In the first, the laser scanner is mounted atop a
surveying tripod positioned roughly in the centre of the mill.
Alternatively, the scanner may be mounted on a beam and
inserted into the mill and operated remotely. This eliminates
operator set-up concerns and the need for time-consuming and
costly isolation procedures for vessel entry as required by other
inspection methods. In the third, the scanner is mounted on a
bracket and suspended from the mill manhole. Figures 2 and 3
show examples of laser scanner data captured inside a mill.
Data processing
Data capture of the inside of a mill with a laser scanner is a
straightforward process. The subsequent data processing is more
challenging and where the innovations of MillMapper lie. First,
it is necessary to correct the scanner measurements using
predetermined calibration parameters to remove inherent
systematic errors that degrade geometric data quality. See the
Quality Control and Assurance section for more information.
Next, MillMapper uses specially-developed data preparation and
registration algorithms to reference the laser scanner data to key
mill dimensions and calculate liner thickness.
Visualisation
MillMapper data can be displayed in natural greyscale
appearance, similar to a black and white photograph, so as to
enable milling professionals to visually inspect the liner surfaces
139
J FRANKE and D LICHTI

A B C D

FIG 1 - Laser scanner data acquisition options. (A) Remote data acquisition by inserting scanner on beam mounted on feed shoot trolley
during an inspection-only shutdown. (B) Standard tripod-mounted data acquisition during a reline shutdown. (C) Scanner mounted on
T-bracket during an inspection-only shutdown before insertion through manhole for remote data acquisition. (D) T-bar bracket viewed from
the outside of mill.

FIG 2 - 3D laser scan data of a mill shown in natural greyscale appearance. Top row from left to right: all liners, discharge liner, belly liner,
feed liner. Bottom row from left to right: two oblique views of discharge end; two oblique views of feed end.

FIG 3 - Detail views of 3D mill liner and ball charge scanner data in natural greyscale appearance.

140 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 ADVANCES IN THE MILLMAPPER TECHNOLOGY FOR GRINDING MILL LINER THICKNESS GAUGING

for their general condition or for damage. Figure 4 illustrates the
greyscale inspection capabilities of the software. The greyscale
display can also assist in interpreting the thickness data display
by providing a realistic visual liner-element cross-reference to
the thickness displays, which are explained further below. For
example, certain blue areas in the thickness display in Figure 5
are in fact lifter bars; red areas represent grate holes, etc.
The primary purpose and therefore functionality of MillMapper
is to display high density 3D mill liner thickness information at
the particular time the scanner data were acquired. MillMapper
achieves this by displaying liner thickness as a colour range
appropriate for the type of liner in question. Feed, belly
and discharge surfaces can be displayed individually or in
combination. A series of display tools demonstrated and explained
in the following figures optimise the use of this method as a
decision-making tool for milling engineers or management.
Milling operators or liner manufacturers may want to assess
overall liner wear distribution or shape as represented by
MillMapper thickness displays (Figures 5 - 9). Depending on the
actual liner age at the time of scanning this information may be
used to support liner design improvements. It is therefore
instructive to look at the isolated location of liner segments as
this provides a more specific indication of patterns of wear.
Thickness displays also enable milling engineers or management
to monitor critical liner portions if they have a history of failure,
for example.

FIG 4 - MillMapper natural greyscale 3D display facilitating inspection of mill liner condition or damage. Top left: discharge liner. Top
right: discharge liner detail. Bottom left: feed liner. Bottom right: feed liner detail – note missing throat liner plate exposing shell.

FIG 5 - MillMapper front view showing liner thickness represented by the colour scales. Left: feed liner – note the missing
throat liner segment exposing the underlying shell, which shows up as red, ie zero thickness. Right: discharge liner.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 141

J FRANKE and D LICHTI

FIG 6 - MillMapper thickness display variation functionality. The liner thickness colour display can be set so that each colour represents
between 1 mm and 50 mm thickness, thereby catering for the best possible visualisation of any type of liner. Top left: discharge liner
displayed with 50 mm colour intervals. Top right: discharge liner displayed with 20 mm colour intervals. Bottom left: feed liner displayed
with 50 mm colour intervals. Bottom right: feed liner displayed with 20 mm colour intervals.

FIG 7 - MillMapper perspective 3D view of belly and feed liner. FIG 8 - MillMapper orthographic 3D view of belly and feed liner.

Inspection of the circular belly liner shape can be challenging
in its traditional 3D display. MillMapper therefore features a
separate representation of belly liner thickness by unwrapping
the cylindrical liner surface onto a 2D plane (see Figure 9).
Reports
Aside from all graphical mill liner displays, MillMapper also
provides a range of numerical wear/thickness statistics. A global
‘one number’ indicator that can be used to quickly assess wear is
the percentage of points that lie below a critical threshold. For
example, 0.2 per cent of the two million points covering the belly
liner from one particular dataset, ie 495 points, are below a
30 mm threshold. Since the example liner has only worn through
half a life cycle, the majority of these points represent gaps
between liner segments, where the scanner detected the outer shell
of the mill. Milling engineers or managers may want to define a
reline criterion for their particular mills, such that whenever a
certain percentage of acquired data is output to be thinner than a
particular liner thickness a reline should be scheduled.

142 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 ADVANCES IN THE MILLMAPPER TECHNOLOGY FOR GRINDING MILL LINER THICKNESS GAUGING

FIG 9 - MillMapper belly liner unwrapping functionality. Top left: schematic showing the belly liner unwrapping procedure from ball charge
level. Top right: front view of 3D belly liner thickness plot. Middle left: unwrapped belly liner thickness plot at 20 mm colour intervals. Middle
right: unwrapped belly liner thickness plot at 50 mm colour intervals. Bottom left: unwrapped belly liner thickness plot detail. Bottom right:
front view of unwrapped belly liner thickness plot, ie cross-sectional view at 20 mm colour intervals.

Figure 10 shows a MillMapper histogram representing a liner ADVANTAGES

thickness and Figure 11 shows the related cumulative
distribution. Both these can be used to gain further insights into
how a liner wears. When comparing a histogram of an original
survey to that of a subsequent one, one or more of the peaks in
the latest histogram may shift to the left (which indicates wear)
much faster than other histogram portions, thereby possibly
indicating an ineffective lifter bar or similar feature.
MillMapper has the following advantages over competing
products:
1. Unlike visual inspection and ultrasonic thickness gauging,
MillMapper does not require physical human access into a
cylindrical mill, thus saving shutdown time and avoiding
the complexity of safety issues.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 143

J FRANKE and D LICHTI

FIG 10 - MillMapper liner thickness histogram for all scan points of an example belly liner. The first peak of the histogram indicates that the
average thickness of the lower liner plates was approximately 60 mm at the time of scanning. The peak at approximately 170 mm thickness
represents the average thickness of the highest lifter bars at the time of scanning.

FIG 11 - MillMapper liner thickness cumulative histogram for all scan points of an example belly liner. The curve shows that half of all liner
elements are thinner than approximately 80 mm. The right-hand edge of the curve indicates that there are no liner elements thicker than
approximately 175 mm.

2. Data collection takes less than five minutes, which is much
faster than any alternative method of inspection. Data capture
does not take one to 1.5 hours as erroneously reported by
Chandromohan and Powell (2006).
3. MillMapper samples liner thickness at several million points
within the mill (the entire visible mill surface can be
measured). Other techniques can sample liner thickness in, at
best, a few dozen manually-selected locations.
4. The relative location of all point samples is accurately known.
The other methods do not deliver this information.
5. The measurements are empirically more accurate than
those of the other techniques.
6. The MillMapper 3D digital mill liner wear model
is quality controlled through scientifically-proven
calibration procedures and rigorous statistical analyses.
7. Mill linear wear is visualised by a series of colour-
coded thickness maps and respective numerical 3D
thickness data for the shell, feed and discharge ends.
Other methods cannot provide this depth of
visualisation.

144 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 ADVANCES IN THE MILLMAPPER TECHNOLOGY FOR GRINDING MILL LINER THICKNESS GAUGING
8. The MillMapper 3D digital mill liner wear model enables
operators to correlate liner shapes with other pertinent mill
data such as energy consumption or milling efficiency at
respective critical liner life times (ie energy consumption
spikes or poor milling efficiency due to a particular liner
shape), and therefore facilitate liner design changes so as to
improve these other factors. This in turn has the potential to
facilitate significant cost savings.
QUALITY CONTROL AND ASSURANCE
It is important to remember that the measurements from any
instrument (ultrasonic probe, mechanical profiler or laser
scanner) are subject to both random and systematic errors.
Whilst the former can not be controlled and only modelled
stochastically, the latter are deterministic and can be modelled
and removed. Though laser scanners are complex instruments,
we have developed in-house methods, error models and software
for their calibration (Lichti and Franke, 2005; Lichti and Licht,
2006). This is an absolutely critical component to the thickness
gauging methodology described herein; without it laser scanners
would not likely be useful for mill liner thickness gauging. By
using our self-calibration procedure, the overall accuracy
improvement can be as high 80 per cent, allowing
millimetre-level positioning to ranges up to 10 m. Further details
can be found in Lichti and Franke (2005) and Lichti and Licht
(2006).
CURRENT AND FUTURE DEVELOPMENTS
The development of MillMapper is ongoing in response to
end-user needs. Originally conceived as a liner-wear assessment
tool, MillMapper is now being expanded to also serve as a
production optimisation system.
As reported by Russell (2006), mill throughput can be
significantly improved through optimisation of new liner design
in combination with shortened reline scheduling. Contrary to
what one might expect, a higher frequency of reline shutdowns
may not lead to higher combined loss of production, when
compared to longer reline scheduling if latest efficient reline
technologies are deployed (Weidenbach, Hart and Griffin, 2006)
and liner design is at optimum simplicity. Currently mine sites
can not reliably determine the most efficient shape of liners to
achieve optimum throughput without MillMapper because no
other comprehensive three-dimensional tool is available to
provide this information. MillMapper not only provides the wear
tracking for reliable shortened reline scheduling but also very
detailed liner shape information so as to facilitate the design of
new liners corresponding to the shape of a worn liner at its peak
throughput performance. Currently peak throughput performance
typically occurs towards the later stages of the liner life cycle,
not when the liner is new and, ideally, the mill should be most
productive.
Current MillMapper developments to facilitate optimum mill
operation include automatic ball charge volume measurement
and liner volume and therefore weight measurement so as to
provide expert control-system operators with the calibration data
for feed rate and ball charge replenishment settings during
milling. These MillMapper output features are derived from the
same laser scan data collected to determine liner thickness.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
Currently, operators determine ball charge volume based on crude
tape measurement surveys, which in many circumstances can not
provide adequate accuracy to achieve volume determination to the
nearest one per cent of actual volume, the accuracy desired to
achieve consistent breakage rates and product size.
Other recently-implemented innovations include improved
visualisation tools that allow users to interactively extract
cross-sections of areas of interest from the data. MillMapper will
also feature automatic discharge-end grate hole sizing, a
functionality which is expected to be available mid 2007. This
will enable sites that have a need for monitoring the expansion of
grate holes caused by wear so as to control product size to
determine change-out schedules based on grate hole sizes rather
than grate plate thickness.
SUMMARY
Originally conceived as a tool for thickness gauging of liners
inside cylindrical comminution mills, MillMapper is changing
into a production optimisation tool. It features a number of
advantages over existing thickness gauging methods, including
safety, speed, accuracy, complete data coverage and far superior
assessment functionality. More recent developments will
facilitate mill operation optimisation. Quality control is very high
and provided by rigorous instrument calibration procedures and
statistical analyses. As mentioned earlier, MillMapper is already
in use at a number of Australian sites including those of major
mining operators ALCOA, BHP Billiton, Barrick, Newmont, and
KCGM.
REFERENCES
Chandromohan, R and Powell, M S, 2006. A structured approach to
modelling SAG mill liner wear–monitoring wear, in Proceedings
SAG 2006, Vancouver, Canada, 23 - 27 September, Volume 3,
pp 133-148.
Franke, J, Lichti, D D and Stewart, M P, 2006. MillMapper: A new tool
for grinding mill thickness gauging, in Proceedings SAG 2006,
Vancouver, Canada, 23 - 27 September, Volume 3, pp 75-87.
Lichti, D D and Franke, J, 2005. Self-calibration of the iQsun 880 laser
scanner, in Proceedings Optical 3D Measurement Techniques VII,
Vienna, Austria, 3 - 5 October, pp 112-121.
Lichti, D D and Licht, M G, 2006. Experiences with terrestrial laser
scanner modeling and accuracy assessment, The International
Archives of the Photogrammetry, Remote Sensing and Spatial
Information Sciences, 36 (Part 5):155-160.
Napier-Munn, T J, Morrell, S, Morrison, R D and Kojovic, T, 1999.
Mineral Comminution Circuits: Their Operation and Optimisation,
pp 413 (Julius Kruttschnitt Mineral Research Centre, Indoroopilly).
Rajamani, R K, Mishra, B K, Latchrieddi, S, Patra, T N and Prathy, S K,
2005. On the dynamics of charge motion in grinding mills, Paper
presented to SME Annual Conference, Salt Lake City, 28 February -
2 March, 9 p.
Russell, J, 2006. Advanced grinding mill relining for process
metallurgists and management, in Proceedings SAG 2006,
Vancouver, Canada, 23 - 27 September, Volume 3, pp 11-22.
Weidenbach, M, Hart, S and Griffin, P, 2006. Application of a 6T liner
handler for Cadia Valley Operations’ 40’ SAG Mill, in Proceedings
SAG 2006, Vancouver, Canada, 23 - 27 September, Volume 3,
pp 59-74.
145
Practical Experience in the Design and Operation of Semi-
Autogenous Grinding (SAG) Mill Liners
D Royston1
ABSTRACT
Mill liners provide the wear-resistant surface within grinding mills, they
also impart motion to the mill charge, ie the grinding process, by
providing the key between the charge and mill shell, and, at the discharge
end, they remove the ground contents of the mill. As semi-autogenous
grinding (SAG) mills have grown in size the process aspects of liner
design and their impact on mill performance have become particularly
important. While feedback from practical experience is critical to good
mill liner design, increasingly sophisticated computer-based tools are
being developed to assist in mill liner design.
INTRODUCTION
SAG mill liner design, especially the impact of liner design on
the process aspects of mill operation, is critical to good SAG mill
operation. This paper draws on operating mill experience to
present a current understanding of SAG mill operation from the
perspective of both shell liner and grate/pulp lifter design. The
paper provides practical observations and interpretations of SAG
mill liner operation and performance: its goal is to identify and
describe the practical issues to consider in liner design for SAG
mills.
SHELL LIFTERS
Shell liner design and mill performance
The two main process objectives in SAG mill shell lifter design
are to maximise the impact and rate of delivery of grinding
media at the toe of the charge where most of the impact-induced
rock-breaking action is assumed to occur, and to avoid liner and
ball damage. Changing the face angles of shell lifters alters
grinding ball trajectories, and spacing between shell lifters
(McPherson a:b ratio) affects charge lifting rate hence mill
performance (McIvor, 1983; Powell and Nurick, 1996). Trajectory
and charge structure computer models incorporating these
concepts, together with practical operating experience and
engineering judgement are used in shell lifter design to meet the
above objectives. Advanced computing tools such as discrete
element modelling (DEM), now incorporating slurry effects
using smoothed particle hydrodynamics (SPH) (Cleary, Sinnott
and Morrison, 2006), contribute to mill liner design by providing
detailed illustrations and data on the interaction between liners
and mill charge motion.
Trajectory models generally track (for shell lifters of various
heights and face angles) the fate of the ‘lifter ball’, the ball that
sits against the lifter and plate (Royston, 2001). At constant mill
speed ball trajectories should degrade (ie balls fall more towards
the bulk charge than the toe) with time as lifters wear down.
Shell lifters with initially large face angles may start by directing
ball impacts at the toe and come ‘on-grind’ almost immediately
only to fall-off in performance. Any practical shell lifter design
has to perform over the full life of the shell lifter/liner. For fixed
speed mills, a common practice is to design shell lifters with
some (slight) overthrow when new, such that the liner ‘wears-in’
to come ‘on-grind’ early in the wear cycle with mill performance
1. MAusIMM, Principal, Royston Process Technology Pty Ltd, Unit 7,
435 Gregory Terrace (off Bathgate Street), Brisbane Qld 4000.
Email: royston@gil.com.au
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
falling away towards the end of shell lifter wear life; the aim is
maximise the wear-life of the shell lifter. It should be noted that
any special benefits from new initial shell lifter profiles may be
maintained only for a period, unless shell lifters are of substantial
size, as the shell lifter profile changes through wear.
Over-throwing is usually to be avoided. However, high levels
of overthrow (if the liners and balls can withstand the impacts)
results in rebounded balls entering the toe region through most of
the shell lifter wear life and with high levels of impact energy.
Hitting the shell just above the toe is the worst situation as the
ball-on-shell impact energies are highest, leading to shell and
ball damage with no compensating high-energy rebound into the
charge.
Increasing the spacing between the liners allows more balls to
fit in the ‘bucket’ (the space between the lifters), increasing
charge lifting rate and allowing more grinding media to be
thrown and delivered to the toe. This may result (for larger
buckets) in a dispersion of impacts for the thrown charge with the
loss of focus on impact at the toe. However, a positive outcome
for fixed speed mills is that some balls may continue to be
effective in hitting the toe region for longer through the lifter
wear cycle.
Mill speed can be increased along with a change in shell lifter
face angle to increase both impacts of balls at the toe and charge
turnover; these effects improve milling performance through
increased ‘ball-charge participation’. However, as mill speeds
increase above 78 - 80 per cent critical speed, pulp lifter
efficiencies could fall and affect overall mill performance.
Increasing mill speed progressively over the wear life of shell
lifters can maintain a ‘sweet spot’ in mill performance by
continuing to direct ball impacts at the toe as both the lifter
height falls and face angle increases with wear. The ‘sweet spot’
is usually tracked by the feedback of impact sound from
microphones mounted close to the mill. This approach coupled
with a focused impact, using smaller not larger bucket size, is a
promising current development in liner design and mill
operation, providing the potential for lower energy consumption,
increased throughput and start-up ‘on-grind’ with new lifters.
Current shell lifter designs commonly adopt large face angles
(typically 22° but up to say 30°, perhaps 35° with very large
lifters) with sufficient spacing between lifters to overcome
packing. An aspect of this is that liners are being designed and
supplied with substantial sections that can change the bucket
capacity significantly. On the latter point, there is some merit in
maintaining a ‘2D row’ shell drilling in new mills such that 2D,
(4/3)D or 1D liner spacing can be used: the fine tuning of
spacing between lifters and the related bucket capacity can then
be accommodated in the liner design. (‘2D’ means twice the
number of ‘feet’ in the mill diameter.)
For large mills using 125 - 140 mm feed ball size,
conventional new lifter-liner dimensions are around 300 -
350 mm overall height above the shell, around 100 mm plate
thickness and around 150 mm top width: detailed design depends
on individual mill circumstances. Increasing lifter height usually
leads to increased shell liner wear life, and changing the
direction of mill rotation regularly may also increase lifter life by
up to 20 per cent.
In traditional ‘HiLo’ liner systems, alternate rows of worn ‘Lo’
lifters are replaced with new ‘Hi’. This system appears to work
well in some smaller mills (say, around and below 24 ft),
147
D ROYSTON
especially where packing can be controlled to the level of the
‘Lo’ lifters. Traditional ‘HiLo’ replacement systems fail in larger
mills, especially in high impact environments where large
grinding balls are used and it is difficult to avoid breakage of
highly worn lifters. Usually such mills have been forced into
‘HiHi’ shell lifter systems to avoid liner breakage.
A recent development in liner design has been to introduce a
form of the traditional ‘HiLo’ lifter system where the ‘Hi’ lifters
are considerably larger than used in prior practice. The ‘Lo’ lifter
is kept to a height similar to the ‘Hi’ in a prior ‘HiHi’
arrangement (see Figure 1). The objectives here are to improve
wear life, to increase lifting rate, to continue to direct ball
impacts at the toe, and to change the shell liner wear (and
packing) distribution, while (at least) preserving the ball-impact
resistance of the original ‘HiHi’ lifter set. Such a ‘HiLo’ liner set
can be of substantial size with an asymmetric design that
requires uni-directional rotation of the mill.
FIG 1 - New HiLo lifter set (after Dunn, Fenwick and Royston 2006).
It is worth noting that highly worn shell lifters can still deliver
adequate (though not necessarily optimal) milling performance.
The common explanation for this is based on several factors:
• increasing mill speed (where possible) to compensate for
lowered shell lifter height;
• increased mill volume hence increased grinding media and
charge volume and interaction as liners wear down;
• increased spray of thrown media (‘Late’ and ‘Plate’ balls,
Royston, 2001) ensuring some balls continue to provide
effective impact hits in the toe region;
• with large balls in particular a significant part of the
comminution-milling process is by repeated ‘low energy’
impacts from the cascading charge;
• rigidities in the mill charge structure still provide enough lift
through the shoulder for some strong cascading action hence
breakage through multiple impacts; and
• for packed mills the effective height of the shell lifters can be
maintained through most of the shell lifter wear-life as the
packing thickness falls in proportion with shell lifter height.
Finally only a minor proportion of SAG mill power is used to
lift and throw the charge (ie to generate high-energy ball-rock
impacts at the charge toe) and this proportion falls as the liners
wear irrespective of mill speed. The major proportion of mill
power is used to turn the charge; especially as shell lifters wear
and lose height. It follows that repeated ball-rock short-range
low-energy impacts within the tumbling charge are a significant
proportion of the overall rock breakage process throughout the
life of the liner.
148 Fremantle, WA, 19 - 21 March 2007
Mill charge levels
Within the mill charge, balls have to transfer energy into rock
breakage via attrition and abrasion through charge motion in the
mill (ie the lifter-induced swirl at the toe, the non-thrown
cascading action and the relative motion of the rising charge).
It is now a common practice, to improve milling rate, for SAG
mills to be run with low charge levels at maximum ball charge
levels. The outcome is to increase ball-charge participation by
more turnover of the charge per rotation of the mill and by
increasing the ball-to-rock ratio. In the past, such mills had limits
on the total charge mass and/or motor capacity (and the resulting
low charge levels often resulted in increased liner damage, ball
and bolt breakage hence challenges to liner design).
With newer mills with higher load capacity, high ball charge
levels (say up to 18 per cent) have been used. The objective again
is to increase ‘ball-charge participation’ through increasing the
ball to rock ratio, while drawing maximum power at the
maximum allowable total charge mass.
In general, mills operating with low charge levels and a strong
focus on hitting the toe appear to be associated with harder
and/or large sized ore feed. Mills with high ball-charges operate
in effect as large ‘primary ball mills’ and appear to be associated
with smaller sized and/or softer ore feeds (and may also use large
shell lifter face angles and wide lifter spacing).
Wide-space and large-angle shell lifters
DEM provides detailed output on the effects of liner spacing and
angles on charge motion overall. In the early use of DEM, claims
were made that significant improvements in mill performance
should result from using wider-spacing and larger-lifter face
angles. However, practical experience in recent years teaches that
such changes may also lead to charge slippage, increase shell
liner wear and degrade overall mill and liner performance.
A review of larger SAG mills that had changed to
wider-spaced shell lifters and large face angles showed, in most
cases, the changes were driven principally by packing between
lifters or damage to liners and balls (Royston, 2004). As covered
above, wider lifter spacing can eliminate packing and larger face
angles can reduce damaging ball-on-mill impacts. With the
alleviation of these immediate issues, mill performance
improved. In addition, and almost inevitably with new and
expanding operations, other changes occurred in the circuit and
in the ore feed at the same time as changes to liner configuration.
All such changes would have affected mill performance,
particularly changes in ore hardness, which has a dominating
effect on mill performance. In that review it was difficult to
assign significant increases in mill performance just to changes
in lifter angle or spacing alone. Where mills did not suffer
packing/liner-breakage, changes in liner design (along the above
lines), at best, resulted in modest claims for increases in mill
performance.
Ore characteristics and shell lifter design
Changes in the hardness of ore fed to a SAG mill can cause
significant changes in mill throughput irrespective of shell lifter
design. In addition, changes in ore characteristics such as a
tendency towards packing can affect the efficiency of the shell
liners in a mill.
With harder ores, the milling rate may be maintained by
increasing grinding ball size. Liner design may then have to be
changed to provide shell liners (and other liners in the mill)
capable of withstanding the higher impact forces of the larger
ball size. This is not readily implemented (or reversed) hence the
need to plan ahead for changes in ore type. These issues are the
basis of a longstanding argument that SAG mills operate best
with ore feed of consistent characteristics.
Ninth Mill Operators’ Conference
 PRACTICAL EXPERIENCE IN THE DESIGN AND OPERATION OF SEMI-AUTOGENOUS GRINDING (SAG) MILL LINERS
Heavy packing, especially in larger mills, can reduce mill
charge lift and milling performance and increase lifter wear rates
significantly. An understanding of the packing characteristics of
the ore is a critical aspect of shell liner design. If for example the
ore supply is from a single source with constant but limited
packing characteristics it may be practicable simply to
accommodate some packing as part of the liner design; in lifter
and plate 2D shell liner designs this approach can add
significantly to plate life. If packing is severe then spacing the
liners to say 4/3D (with associated changes in face angle covered
above to avoid plate damage) may provide enough gap between
the shell lifters to prevent packing; then the shell lifter-liner
design has to be based on an impact environment where no
packing is present. Wider spacing to 1D may ensure no packing
can occur but at the risk of damage (through ball impact) to wide
exposed plates and of high liner wear rates due to charge
slippage.
END LINER AND GRATE STRUCTURES
Introduction
The overall charge level in the mill affects the wear patterns on
the end-liners (and the charge transfer effectiveness of the grate
through the degree of exposure of the grate openings to the
charge). Ball trajectories together with charge level (especially at
low levels) affect end liner design through the nature of ball
impacts on end liners and grates.
End liner design/life issues
An imaginary circular line that may be drawn on the rotating end
of the mill by the stationary ‘eye’ of the mill charge is referred to
here as the ‘eye-line’. The position of maximum wear on the feed
end (FE) lining is around this ‘eye-line’. To prevent (rapid)
abrasive wear of the end plates, lifter bars are used to deflect
charge from the plate.
The FE plate itself usually carries a central bar originally used
to stiffen separate plates and in integral FE lifter-liners, favoured
for larger mills, the bar forms part of the base for the lifting eyes.
The bar also acts to deflect charge and limit abrasive wear on
the plate. Recent trends include: increasing the size of this bar
to improve plate (and indirectly lifter) wear life, and for
uni-directional mills repositioning the bar better to protect the
high wear region immediately in front of the FE lifter.
With FE lifters the trend in recent years has been towards FE
lifters with angled leading faces and outer taper; these designs
aim to shed (ie avoid throwing) balls that could damage liners at
the head end, and to even out the wear along the FE lifter.
The change to large face angles on shell lifters has resulted in
a change towards radial-directed incoming trajectories of balls.
In these situations balls travel between (ie are not deflected by)
end lifters and hit the ends of exposed end-liners, resulting in
significant, often premature, damage.
Some feed end issues:
• Ensure good fit of FE parts with the conical mill head (and
mount on sound backing rubber); poor fit can lead to bolt
failure and plate-cracking.
• Limit exposure of parts to radial incoming ball impacts, ie
avoid exposed ends, protrusions and large bolt hole openings
that provide ball impact points that can lead (through
persistent impacts) to metal flow and/or fracture.
• For both the FE and DE, avoid mixing new with old parts in
ways that expose new parts to impact damage. Preferably
capture most of the wear on a limited number of parts and
change out all these parts together.
• Sequence change-out of FE lifters simultaneously with shell
lifters; this avoids high wear on old shell lifters.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
With the discharge end (DE), most large mills have adopted
cantilever grates usually with large open areas, inter-grate gaps
and a large proportion of pebble ports to promote pebble recycle
for pebble crushing. Nevertheless, there are mills with restricted
grate openings that aim to limit rock outflows in order to
promote a fine grind size, and other mills that use a
jet-return-trumpet system and have to limit in the opening size of
grate slots to allow steel media to be returned into the mill by the
water jet.
The general structural principles outlined above for FE liners
apply also to the DE, noting for grates and plates, the fit should
be a one-on-one match with the underlying pulp lifter (see later).
Grate performance as a liner is affected by:
• peening (metal flow that closes the grate openings) – check:
ball-grate impact situation, material of construction, slot
opening size, wear rates across the surface of the grate, need
for the affected openings;
• pegging, especially ball pegging – check: reverse taper
(nominally 5° relief angle each side), pebble ports, in-mill
ball-charge management, avoid recycling worn balls, ball
type and worn shape, rock fracture at pegging size;
• trailing-side outer-grate ball-impact damage – check: ball
throw, grate lifter throw, plate thickness, edge support,
casting integrity especially at the extremities of the grate,
web thickness, web support, wear induced strain relief rates
across plate;
• inner end grate damage – check: ball throw trajectory
especially if using large face-angle lifters near the head end,
local sources of plunging balls, protection provided by inner
plate and (worn?) lifters, metal flow between grate and next
inner plate, charge levels, casting integrity, web thickness,
wear-relief rates across plate, glacis (protective ramp) on
inner plate; and
• general plate damage – check thickness, thicker plates of
100+ mm are used in larger mills.
PULP LIFTERS
Introduction
Material is discharged from a SAG mill using pulp lifters and
pulp lifting effectiveness can affect the charge level in the mill,
hence grinding efficiency. Installed at the discharge end of the
mill, pulp lifters are a radial array of channels separated by
channel walls also known as vanes or septums. Each channel is
open to the mill at the outer end to allow material inflow
(through a grate) and at the inner end to allow discharge from the
mill through the mill trunnion (Napier-Munn et al, 1996).
Typically, the number of pulp lifters employed is the number of
feet in the mill diameter (a ‘1D array’).
Pulp lifters operate through a lifting and bailing action. Pulp
lifters fill with pulp (fine rock slurry with pebbles) flowing in
through grates. The pulp lifters, as they rotate with the mill, lift
the pulp until it flows towards the centre of the mill along a pulp
lifter channel. The pulp exits the mill via an inner ‘discharge
cone’ that diverts the pulp out of the mill through the trunnion
opening. Pebbles that fail to discharge fall back down the pulp
lifters, lower pulp-lifter efficiency and cause pulp-lifter wear.
Curved pulp lifters can improve pebble discharge and the wear
lives of pulp lifters.
Pulp already in the pulp lifter can flow back into the mill
through grate openings as the grates rotate out of the charge.
Control of this ‘pulp reverse flow’ is an important aspect of grate
and pulp lifter design.
Pulp lifter charge motion
The following description of the operation of conventional (and
curved pulp lifters) is based on video data of the discharge from
149
D ROYSTON

pulp lifters, wear patterns observed in pulp lifters, and a single

particle flow analysis (Denlay, Woods and Royston, 1997;
Royston, 2000, 2006). Curved and conventional pulp lifters have
been subject more recently to DEM flow analyses with similar
outcomes (Hart et al, 2001; Rajamani, Latchireddi and Mishra,
2003; Cleary, 2006).
The discharge for any pulp lifter has to be considered as two
components; one is a fluid-like flow of a fine rock slurry (referred
to here as ‘fluid pulp’) and the other a stream of larger rocks. As
the pulp lifter contents begin to move in the pulp lifter channels, it
can be assumed that the more fluid component separates from the
rocks that settle to the outer ‘base’ of the pulp lifter chamber. The
subsequent motion of the two components is different, and they
discharge at different points in the mill rotation.
The fluid pulp is the largest portion by volume of the charge in
the pulp lifter. It is positioned in the pulp lifter closer to the
centre of the mill, hence less subject to centrifugal forces (than
the rock component), and is free to adjust its level and position FIG 2 - Rocks in pulp lifter channel (after Hart et al, 2001).
(within the limits of levelling forces) in the pulp lifter. The solids
in the pulp lifter chamber by contrast are subject to friction 0.5

forces that restrain their movement in the pulp lifter.

0.4
Fluid pulp can flow readily to the centre of the mill; the
outcome in conventional pulp lifter is that most of the fluid pulp 0.3

is discharged around ‘11 to 2 o’clock’ in a clockwise rotation of

the mill. 0.2

The rock component starts motion towards the centre much

later, and the motion is more complex, than the fluid pulp 0.1

component. Straight-radial pulp lifters act, in effect, like shell

lifters with a ‘zero degree’ face angle and semi-infinite length. -0.5 -0.4 -0.3 -0.2 -0.1
0
0 0.1 0.2 0.3 0.4 0.5

As a result, rocks in the base of the pulp lifter are ‘pinned’ by -0.1

the force balance (between the outward centrifugal force and the
inward radial component of gravity) and friction until the base of -0.2

the pulp lifter passes through the shoulder of the mill charge.
After initial motion along the rising side channel wall (‘vane’) of -0.3

the pulp lifter, rocks disengage from that channel after the pulp
lifter passes ‘12 o’clock’ in the mill’s rotation and travel across -0.4

the pulp lifter cavity to contact the opposite channel wall of the
pulp lifter. After contact with the wall the rocks slide in along -0.5

that falling wall to discharge into the trunnion as the pulp lifter FIG 3 - Straight-radial pulp lifter – single particle track.
rotates towards and through the horizontal in the mill rotation.
Not all large rocks are discharged before the pulp lifter moves
below the point where the inertia in the rocks is insufficient for insufficient inward momentum and fail to discharge. They would
discharge. Any undischarged rocks ‘backflow’ down the pulp also impact on the lower channel wall causing abrasive wear at
lifter. that point, which is reflected in the wear patterns observed in
inner pulp lifters (see Figure 4). These tracks also indicate that
Backflow is a major source of wear in pulp lifters. The risk of
reducing channel wall length risks increased backflow. The
backflow is increased in pulp lifter systems where two or more
velocity data from these outputs indicate that the discharge
channels merge into one towards the centre of the mill. In such
channels themselves do not choke (ie fill with normal outflowing
systems late-arriving rocks fall from the upper channels into the
charge) to prevent outflow; choking appears to be the result of
‘lowest’ channel. This creates a burden of rock material in the
‘backflow rocks’ prefilling hence restricting pulp lifter capacity.
‘lowest’ channel that accelerates wear in that channel and
associated pulp lifter base. Hart et al (2001) provided an image 0.2

of a pile of rocks near the exit from the pulp lifter into the
trunnion at the point of backflow (see Figure 2). This rock 0.15

burden prefills and reduces the capacity of the affected pulp

lifters, limiting the overall lifting capacity of the pulp lifter 0.1

system. The risk of backflow increases with increase in mill

speed. For mills using straight-radial pulp lifters and pebble 0.05
recycle, as mill speed is increased loss of pulp lifting capacity
can be a factor limiting mill performance.
0

More recently, the issue of rock flow in pulp lifters, especially -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2

in the discharge cone, has been addressed in a single particle

-0.05
computer ‘pulp lifter motion tool’ that covers the full flow path,
including that in the discharge cone (Royston, 2006). Outputs for
a straight-radial pulp lifter illustrate the late arrival of a particle -0.1

(representing a single rock) at the discharge trunnion just as the

pulp lifter channel is about to rotate below the horizontal, -0.15

effectively eliminating discharge of rocks other than those

arriving with high inertia (see Figure 3). Particles from higher, -0.2

shorter channels fall to the lowest channel wall, and in these

examples the particle tracks indicate ‘the rocks’ would have FIG 4 - Detail of particle track in discharge cone – top channel entry.

150 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 PRACTICAL EXPERIENCE IN THE DESIGN AND OPERATION OF SEMI-AUTOGENOUS GRINDING (SAG) MILL LINERS

Grate – pulp lifter interaction

Grates allow the controlled passage of slurry and rocks from the
charge inside the mill into the pulp lifter chamber. For this
analysis the charge inside the mill can be considered to have two
parts. A dense part outside the ‘eye’ of the charge is connected to
and moves upwards with the mill shell lining; inside the ‘eye’
(towards the centre of the mill) there is a loose open structure of
charge falling towards the toe of the charge. The dense rising
charge structure is in close contact with and moves upwards with
the grate. This dense rising charge in the mill acts as a pump,
moving fluid pulp from the toe to the shoulder of the charge.
As long as grates are in contact with the rising mill charge,
grate openings that are covered by charge would pass fluid pulp
into the pulp lifter. In mills with no slurry pool, the position in
the mill rotation where most transfer takes place is around half
way between the toe and shoulder. Inside the ‘eye’ the loose open
structure of falling mill-charge presents little pulp to the grate. It
follows, that in order to be effective in receiving fluid pulp, grate
openings should be positioned from the ‘eye-line’ ‘outwards’. In
practice the outermost grate openings (ie those nearest the shell) FIG 6 - ‘Dynamic wash’ reverse flow.
appear to be the most effective for the transfer of pulp, hence the
importance of maximising open area in the outer part of the grate.
Pulp lifting capacity
Recent experience indicates too close spacing of narrow outer
slots (in an attempt to maximise outer open area) may result in a As noted above, pulp lifters operate by a bailing action. It
form of blinding restricting flow through the grate. In addition, follows that the size of the ‘bailing’ bucket, ie the ‘bottom’ outer
openings around the ‘eye’ region have a role in charge transfer base part of the pulp lifter, needs to be maximised. With an
especially with new grates and their (relative to worn) limited ‘L’-shaped pulp lifter, the head-side forms a right angle to the
open area and opening sizes that can restrict mill throughput. base (see Figure 7). For a given pulp lifter depth, the volume
For the efficient and effective performance of grates and pulp available at the bottom of an ‘L’-shaped pulp lifter is greater than
lifters it is important to recognise the impact of ‘pulp a pulp lifter with a base parallel to the mill shell at the level of
reverse-flow’ and the interaction between grate and pulp lifter the base of the grate. The ‘L’-shaped pulp lifter will maximise
requirements (Royston, 2000). levels of fill before a head develops (in the pulp lifter) to oppose
During the initial ‘static’ phase of pulp lifter fluid motion, as significantly inflow from the mill charge. Limiting bucket
the grate emerges from the mill charge, pulp is still held within capacity, especially for high throughput mills, leads to a need to
the pulp lifter (due to radial centrifugal force) but is not increase mill charge levels to produce the head required for flow
constrained axially. Reverse flow can occur through grate through the grate to fill the pulp lifter. Increasing mill charge
openings along a path parallel to the axis of the mill. In the parts pulp level can work against grinding performance, especially if it
of the grate at risk of allowing reverse flow, the distance from the results in pooling, ie excessive slurry at the toe of the charge,
outer end of the grate slot to the channel wall needs to be ultimately to the extent of flowing out of the feed end of the mill.
sufficient (say, 50 to 65 mm) to minimise this form of reverse
flow and to form an effective ‘launder’ directing flow along the
pulp lifter channel (see Figure 5). This criterion applies
particularly to openings around the central part of the grate.

FIG 5 - Control of ‘static’ reverse flow.

FIG 7 - ‘Tapered’ and ‘L’-shaped pulp lifter.

During pulp lifter-grate rotation over the top (‘vertex’) of the
mill, discharging pulp washes over the backs of inner grate slots.
During this, ‘the dynamic phase’ of reverse flow, any exposed
inner grate openings (especially inwards from the ‘eye’ of the
charge) would allow reverse flow of pulp back into the mill (see
Figure 6). This is an important reason to minimise openings
inwards of the ‘eye-line’.
Experimental data and practical experience appear to support
the need for adequate pulp lifter depths (front to back) to achieve
satisfactory lifting rates. Good depth also helps to meet the
‘launder height’ requirements to limit ‘static’ reverse flow. For
mills with no pulp recycle a useful guide is a pulp lifter depth

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 151

D ROYSTON

from front to back of around four per cent of mill diameter. For
mills with high recirculating loads, or high throughputs of say
soft ores, the pulp lifter depth requirement should be increased to
limit reverse flow.
In existing mills, a partial retrofit solution for increasing pulp
lifter depth is to increase the depth of the pulp lifter channel
towards the centre of the mill in a tapered pulp lifter (see
Figure 7). Another approach (for uni-direction mills) is to use
pulp lifters with pulp lifter channel walls (normally at a
right-angle to the head, Figure 5) angled downwards from the
grate towards the head (ie downwards to the right with respect to
Figure 5): this forms a trough to hold pulp on the head-side of
the pulp lifter and away from the grate during the lifting motion
of the pulp lifter.
While 1D pulp lifter arrays are typical, practical issues such as
the positioning of mill head-end joints can limit the number and
distribution of holes used to mount pulp lifters. With a lower
number of pulp lifters, the charge volume in each pulp lifter
chamber is increased (adding to the risk of reverse flow). A
lower number of pulp lifters also results in wider individual pulp
lifters and each takes longer to discharge and this increases the
risk of backflow. Increasing the number of pulp lifters adds to the
number of channel walls between pulp lifter chambers and, due
to the additional ‘dead’ volume of the walls, may lower the
overall pulp lifting capacity.
Particle track outputs (Royston, 2006) show different motions
in large compared with small mills where both have the same
outer curve radius (see Figures 8 and 9). With proportionally
longer straight to curved section in the larger mill, inward motion
may not be fully sustained once the charge reaches the straight
portion of the pulp lifter. At this junction, the charge is subject to
the same forces as would be present on a straight-radial pulp
lifter. The force balance may not strongly favour inward flow,
and the ‘rock’ could even be subject momentarily to slowing
forces. It follows that with the relatively longer straight portion
of the large mill, the charge would take longer (in the mills
rotation) to reach the centre. Note the comments above on
methods for increasing the curve radius, hence increasing the
proportion of curved to straight channel and improving pulp
lifter performance. Nevertheless, a proportion of straight (with
the curved) appears important; the inner straight portion is
required to direct efficiently to the centre any pulp lifter charge
that gets thrown to the other side of the pulp lifter during inward
charge motion after the pulp lifter rotates past the mill vertex. If
the curve extends to the centre, discharging rocks could have ‘to
climb the back’ of the curve towards the centre of the mill for
discharge.
0.5

0.4
CURVED PULP LIFTERS
Curved pulp lifters can discharge rocks much earlier in the mill 0.3

rotation, hence more effectively, and reduce significantly pulp

lifter wear due to backflow (compared with straight-radial). 0.2

However, they do require a commitment to uni-directional rotation

of the mill. Curved pulp lifters with various curved shapes and 0.1

forms of construction have long been used in grinding mills

(Gueguerian, 1982; Mokken, 1978; Taggart, 1947).
The retrofit ‘hockey-stick type’ curved outer pulp lifters and
grates, discussed here, have been installed in the 24 ft diameter
and 28 ft SAG mills at Northparkes Mines (NPM) (Dunn,
Fenwick and Royston, 2006) and in the 40 ft SAG mill at Cadia
(Hart et al, 2001). The principal objective in the first case (24 ft
SAG) was to improve solids flow in a pebble crushing circuit.
The first case showed the potential for long wear lives with
curved pulp lifters, probably a factor in the later two retrofit
decisions (Royston, 2006).
Owing to the curve, rocks in the base of the pulp sit at an angle
closer to their angle of friction (the slope angle at which a rock FIG 8 - Curved pulp lifter particle track – large outer curve.
will slide) compared with the flat sides of a straight-radial pulp
lifter. As the pulp lifter rotates, this allows the angle of friction to
be overcome earlier, initiating rock motion earlier than the 0.5

straight-radial counterpart. As a result of the earlier start, rocks

leave the pulp lifter earlier in its rotation. Early discharge can 0.4

reduce substantially the backflow of rocks; this is reflected in the

long wear lives of the curved pulp lifter examples above. 0.3

The degree of influence of the curve on the performance of the

0.2
pulp lifter depends upon the curved shape of the lifter, especially
near the mill periphery and, in the case of the retrofit type, the 0.1
proportion of curved to straight portions in the pulp lifter.
In the current retrofit type of curved pulp lifter designs, the 0
curve radius has been expressed, ie developed, around the -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5

available bolting over two previous outer ‘straight-radial’ pulp -0.1

lifter spaces. While this appears to be a good outcome for
smaller mills, for all 1D pulp lifter mills this approach results in -0.2
a similar curve radius irrespective of mill diameter (or the use of
an elongated less effective curve). The use of the same approach -0.3
in mills with fewer pulp lifters than ‘1D’, should allow a
(relatively) larger radius curve to be expressed within the -0.4
available bolting. Another approach for large mills would be to
select from the available bolt holes in the head over, say, three -0.5
prior outer ‘straight-radial’ pulp lifter spaces to provide a larger
curve radius. FIG 9 - Curved pulp lifter particle track – small outer curve.

152 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 PRACTICAL EXPERIENCE IN THE DESIGN AND OPERATION OF SEMI-AUTOGENOUS GRINDING (SAG) MILL LINERS
DISCHARGE CONE WEAR PATTERNS AND
DESIGN
In conventional straight-radial pulp lifter discharge cones, there
is significant wear of the channel walls that divide the pulp
lifters. This wear effect is higher on the longest channels in
discharge cones that merge two or more channels into one
towards the centre of the mill, especially where a lower fourth
channel receives all the charge from the upper three. Discharge
cone design with straight-radial pulp lifters is focused on wear of
the channel wall (vane). This may involve the use of
abrasion-resistant ‘white iron’ castings in discharge cones. In
composite metal-rubber parts the channel walls may incorporate
wear-resisting alloys. Multiple merged-channel diversion cones
need also to be designed to allow ‘indexing’, ie the rotation of
the cone on replacement. This allows the distribution of
‘backflow’ to unworn pulp lifter bases, spreading the backflow
wear amongst all pulp lifter chambers to improve the overall
wear life of straight-radial pulp lifters.
Curved pulp lifters produce a significantly different wear
pattern in the discharge cone compared with straight-radial pulp
lifters. Heavy wear is predominantly at the diversion curve of the
cone. Typically two flow-wear induced grooves develop on each
side of the channel wall: one is assumed to be associated with
fluid pulp flow discharge before the pulp lifter reaches the mill
vertex, the other with pulp and solids reaching the cone after the
pulp lifter has passed the mill vertex. In this situation, the design
effort is focused on the diversion and exit zones with the aim of
limiting replaceable wear parts (in metal or rubber) to the
diversion zone.
BOLTING
Typical bolt and nut combinations are in the metric system Class
8.8 bolts with Class 8 nuts (SAE Grades 5 <bolt and nut> are
similar). A higher Class (or Grade) provides the potential for
higher bolt tensions; however, with the higher tensile strength the
bolts can become vulnerable to stress raisers (eg from surface
damage in handling and installation) and fracture once initiated
can lead to rapid failure.
Good joint compression is critical to the long-term security of
bolts. If compression is lost across the joint, even though the bolt
may still be in tension, there can be movement across the joint.
The cyclical bending moment can lead quickly to bolt failure
(even in large diameter bolts) through fatigue cracking at a stress
raising point (say – one thread into the nut or where the threads
end on the bolt shank).
A certain amount of preload is required on the lining bolts to
ensure compression is maintained under varying liner loads.
Preload is a function of applied tightening torque to the nut and
bolt, and the torque is affected by the lubrication applied to the
bolt and nut. Bolt-supplier torque tables usually assume that the
bolts and nuts have clean non-corroded metal surfaces with some
light oil lubricant (sufficient to get consistent torque readings).
Dry or rusty bolts will be under tensioned at the usual torque
table settings. If special low-friction lubricants are used the
applied torques must be adjusted downwards to match, otherwise
bolts will be over-tensioned and this could lead to breakage.
The bolt torque may be applied initially by an air-driven
impact wrench ‘rattle-gun’, and completed by an hydraulic
torque wrench preferably by rotating the bolt without stopping
(ie tightening under constant dynamic friction) until the required
torque is reached. As new liners settle with use, good practice to
re-torque all bolts after a few hours of operation, the objective is
to obtain stable bolt tension.
The impact of grinding media directly on bolt-heads can lead
to bolt failure and for this reason bolts with deep heads (aimed to
provide reduced wear across the bolt head opening) have fallen
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
out of favour. Various schemes have been tried over the years to
prevent wear across bolt openings including inserts that were
held in the liner with the bolt head beneath and out of contact
with the insert (Denlay, Woods and Royston, 1997) and the
captive bolt systems described below. Current general practice is
to bury the bolt-head well inside the part so that the head
becomes exposed only at the point of maximum wear of the part.
The common wedge-seat bolts require good bolt alignment and
seating; spherical bolt-seats are a variation on the wedge seat,
aimed to provide good seating even with the bolt slightly out of
alignment.
Common sealing washer practice is to use a rubber sealing
washer captured in compression under a strong stiff steel washer
cap. Many proprietary seal designs are available aimed
particularly at sealing worn bolt holes. ‘Nylok’ type self-locking
nuts are popular but plain black nuts are also used.
Bolt removal from worn liners has been improved
considerably with the advent of recoil-less impact hammers, their
use has improved safety and reduced downtime. Screw-out bolts
are an alternative offered by some suppliers; in this design, bolts
with screwed-on heads are captured inside the liner. On liner
removal, the bolt shanks are screwed out through the shell
allowing the liner to fall in. This type of design has the advantage
of avoiding wear through the lifter bolt-hole region (a source of
weakness in conventionally bolted liners) as well as reducing
liner removal downtime. One disadvantage is that if the bolt
shanks will not come out, heavy liners with the bolting have to
be knocked-in together, which can be difficult.
LINER SIZE AND MATERIALS
There is now good experience in the use of double-wide and
double-long large liner sizes in large mills. There is pull and
push in this development. The pull is the need to reduce
downtimes and simplify change-outs by reducing the number of
liner movements in and out of the mill. The push is the
availability of large capacity liner handling machines and
associated bolt-removal impulse hammers (as well as other
equipment and procedures) aimed to mechanise and at the same
time to improve the safety of change-outs, especially of large
parts.
‘Double-wide’ parts are commonly used for end liners, grates
and shell liners, and foundries have directed efforts towards
supplying such parts. Important issues in moving to
‘double-wide’ parts are to ensure a full fit and for grates not to
overlap joints in the pulp lifters, ie one-on-one fit is required, or
otherwise there is a risk of breakage at or near joints. A potential
negative is that by removing joints between shell liners, there is
less capacity for impact strain relief through metal flow and this
has to be considered in design. A practical issue with
double-wide shell liners is the placement of the necessarily
substantial lifting eyes; if these fill the space between the lifters
they can provide an abrasive-wear path causing increased wear
on the lifters at that point.
Traditional lifter and plate designs continue to have their place
providing both opportunities to capture wear on smaller easily
removed parts, and at the extreme in very high impact
environments, providing opportunities for a liner design solution
through opportunities to direct high metal flow and for inter-part
movement to relieve and to counteract strain build-up (leading to
stress fracture).
‘Chrome-Molybdenum’ (Cr-Mo) alloy steel combines high
impact resistance and good wear life and is still the dominant
material of construction for SAG mill liners. ‘White Iron’
continues to be a material of choice for non-impact highly-
abrasive wear-zones (where Cr-Mo alloy steel is less effective).
Specialised alloy formulations around the Chrome-Moly type
materials have been promoted: these are based on creating
153
D ROYSTON
microstructures and through-hardness sometimes by careful heat
treatment and are claimed to provide improved wear resistance in
abrasion-prone parts of SAG mills. Other liner suppliers claim
harder high-carbon versions of their ‘standard’ Cr-Mo products
give similar outcomes; higher carbon appears to provide
improved abrasion-resistance at similar hardness levels (assumed
to be related to carbide content).
One interesting development is the bi-metallic liner using a
‘White-Iron’ insert that can give increased wear-life in
low-impact abrasion-prone locations such as end liners. In
addition, metal-faced rubber ‘Polymet’ type products with
designs resistant to damage through ball impacts are growing in
application in SAG mills.
REFERENCES
Cleary, P W, 2006. Advanced computer modeling for industrial granular
flows and particle breakage in mills, crushers and screens, in
Proceedings IIR Crushing and Grinding Conference, Townsville, 29
- 30 March.
Cleary, P W, Sinnott, M and Morrison, R D, 2006. Prediction of slurry
transport in SAG mills using SPH fluid flow in a dynamic DEM
based porous media, Paper presented to Comminution ’06, Perth, 15 -
17 March (submitted for publication in Mineral Engineering).
Denlay, D R, Woods, R J and Royston, D, 1997. Development of the ANI
Bradken Vortex Pulp Lifter and spherical head bolt mill liner fastener
system, in Proceedings Sixth Mill Operators’ Conference, pp 67-72
(The Australasian Institute of Mining and Metallurgy: Melbourne).
Dunn, R, Fenwick, K and Royston, D, 2006. Northparkes Mines SAG
mill operations, in Proceedings International Conference on
Autogenous and Semiautogenous Grinding Technology (SAG2001),
Vancouver, 23 - 27 September.
Gueguerian, R L, 1982. Discharge zone fastening assembly for grinding
mill, US Patent 4,365,763.
Hart, S, Valery, W, Clements, B, Reed, M, Song, M and Dunne, R, 2001.
Optimization of the Cadia Hill SAG mill circuit, in Proceedings
International Conference on Autogenous and Semiautogenous
Grinding Technology (SAG2001), pp 1:11-30, Vancouver,
30 September - 3 October.
154 Fremantle, WA, 19 - 21 March 2007
McIvor, R, 1983. Effects of speed and liner configuration on ball mill
performance, Mining Engineering, 35(6):617-622.
Mokken, A H, 1978. Progress in run-of-mine: (autogenous) milling as
introduced and subsequently developed in the gold mines of the
Union Corporation Group, in Proceedings 11th Commonwealth
Mining and Met Congress, Hong Kong.
Napier-Munn, T J, Morrell, S, Morrison, R D and Kojovic, T, 1996.
Mineral comminution circuits their operation and optimization, in
JKMRC Monograph Series in Mining and Mineral Processing,
Vol 2, Chapter 7 (Julius Kruttschnitt Mineral Research Centre:
Indooroopilly).
Powell, M S and Nurick, G N, 1996. A study of charge motion in rotary
mills, Part 1: Extension of the theory, Minerals Processing,
9(2):259-268.
Rajamani, R K, Latchireddi, S and Mishra, B K, 2003. Discrete element
simulation of ball and rock charge and slurry flow through grate and
pulp lifters, Preprint 03-108, SME Annual Meeting, Cincinnati, 24 -
26 February.
Royston, D, 2000. Grate-pulp lifter interaction in SAG/AG mills, in
Proceedings Seventh Mill Operators’ Conference, pp 63-67 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Royston, D, 2001. Interpretation of charge throw and impact using
multiple trajectory models, in Proceedings International Conference
on Autogenous and Semiautogenous Grinding Technology
(SAG2001), Vancouver, 30 September - 3 October, Vol 4,
pp 115-123.
Royston, D, 2004. Review of recent experience with large-angle, wide-
spaced shell lifter-liners, Mining Engineering, 56(12):73-76.
Royston, D, 2006. SAG mill pulp lifter design, discharge and backflow,
Mining Engineering, 58(9):57-62.
Taggart, A F, 1947. Handbook of Mineral Dressing Ores and Industrial
Materials, Chapter 5, pp 78-79 (Wiley).
Ninth Mill Operators’ Conference
Metallurgical Accounting in the Northparkes Concentrator —
A Case Study
W M Jansen1, R Morrison2 and R Dunn3
ABSTRACT
In many mineral processing operations, the reported production figures
for an accounting period may not make the best use of available data.
While most of the mathematical approaches, such as check-in/check-out,
mass balancing, used in effective metallurgical accounting strategies are
well understood, the sources and magnitude of measurement errors
generally are not. As part of the ‘AMIRA P754-Metal Accounting and
Reconciliation’ project, a study was undertaken at the Northparkes
copper/gold concentrator near Parkes, New South Wales to examine the
data used in their concentrator metallurgical accounting system.
This paper summarises the results of a detailed examination of the
mass measurement systems around the concentrator as well as the
primary sampling system used for grade estimation. The goal of this work
was to develop practical techniques to quantify the error (ie both accuracy
and precision) of the measured metal accounting information. With this
information, statistical confidence levels for this data can then be
calculated. If the errors are small, the effect of error in each piece of
measured data can then be propagated to other calculated production
figures such as recovery and feed grade using a statistical technique
called ‘the propagation of error’.
INTRODUCTION
As part of the AMIRA P754 – Metal Balancing and
Reconciliation study, evaluation of all sources of metallurgical
accounting data from the Northparkes copper-gold concentrator
was performed. These data sources can be divided into two
general areas:
1. tonnage estimation through mass measurement, and
2. grade estimation through sampling.
1. PhD Candidate, Julius Kruttschnitt Mineral Research Centre, Isles
Road, Indooroopilly Qld 4068. Email: w.jansen@uq.edu.au
2. MAusIMM, Chief Technologist, Julius Kruttschnitt Mineral
Research Centre, Isles Road, Indooroopilly Qld 4068.
Email: r.morrison@uq.edu.au
3. MAusIMM, Concentrator Superintendent, Northparkes Mines, PO
Box 995, Parkes NSW 2870. Email: rick.dunn@riotinto.com
FIG 1 - Metallurgical accounting data sources in the concentrator at Northparkes.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
This paper details some of the findings of the study. This study
is part of a larger project aimed at evaluation of the metallurgical
accounting strategy of the entire Northparkes operation from
mine haulage to concentrate load-out.
NORTHPARKES OPERATIONS
Overview
The Northparkes Mines (NPM) copper-gold operation is located
approximately 27 kilometres northwest of Parkes in Central West
New South Wales (NSW). Production presently comes from the
E26 underground mine, the first block cave mine in Australia,
with make-up concentrator feed coming from surface stockpiles
of ore from the now-defunct E22 and E27 open cut mines.
Mining commenced in 1995.
The NPM grinding circuit consists of a primary SAG mill with
oversize crusher followed by a secondary ball mill grinding
stage. The ore is then subjected to a two-stage flash flotation
process before being fed to a tertiary ball mill and on to the
flotation plant. The sulfide flotation process is comprised
of rougher-scavenger banks followed by Jameson cleaner,
cleaner-scavenger and Jameson re-cleaner steps. The combined
flash flotation and flotation plant concentrate is thickened,
filtered and stockpiled before train load-out to the port.
Concentrator metal accounting
Figure 1 shows an overview of the sources of metallurgical
accounting data in the Northparkes concentrator. The plant
consists of two modules that run in parallel. Each module has an
essentially identical flow sheet to the one shown in Figure 1. The
main difference between the two is that module 2 has roughly
twice the throughput capacity of module 1. Metallurgical
accounting information for each module comes from two mass
measurement sources as well as samples from four slurry streams
in the plant.
The first mass measurement point is a weightometer (belt
scale) measuring the ore fed to the concentrator from the primary
stockpiles, which are mainly fed by primary crushed ore from the
157
W M JANSEN, R MORRISON and R DUNN
underground mine. Some make-up feed for these stockpiles also
comes from surface stockpiles consisting of ore from surface
mines no longer in operation. The second mass measurement
point used for metallurgical accounting is a weightometer that
measures the mass of filtered concentrate from the combined
flash flotation and flotation plant concentrate streams. All of the
grade measurement samples are taken by the AMDEL on-stream
analysis system housed in the flotation plant. The system takes
real-time grade measurements of concentrator streams for
process control, but physical sample cuts are also taken, which
are analysed in the on-site laboratory to obtain metallurgical
accounting data.
Objectives
The objectives of this part of the Northparkes metallurgical
accounting study are:
1. to better understand the error in metallurgical accounting
data, and
2. to better quantify the statistical confidence in the
measurement and sampling processes in the concentrator.
In order to do this, it is necessary to examine each source of
metal accounting data and, if possible, develop techniques to
measure or better estimate the error associated with each.
There are two basic types of error that can affect the quality of
a measurement: systematic and random. These two errors can
affect measurement accuracy and precision. In the sense with
which it will be used in this document, precision is the measure
of how closely repeated measurements agree among themselves,
and it is commonly expressed in terms of standard deviation (σ)
and variance (σ 2 ). Systematic errors produce a measurement
bias. One definition of bias is the difference between the ‘true’
value and the average of a large number of replicate results.
Thus, accuracy is, in this paper, the combination of lack of bias
and high precision (Dodge, 2003).
Structure of the paper
This paper is the outcome of experiments and observations from
the aforementioned metallurgical accounting data sources, and
the results of this work are divided into three sections. The first
section details experiments performed on and observations made
of the two mass measurement systems in the concentrator:
the mill feed weightometers and the filtered concentrate
weightometers. The variation of wet mass measurements from
these sources is discussed as well as the variation in moisture
contents used to calculate dry tonnages from the wet mass
measurements.
The second section of the report takes a closer look at the
AMDEL grade measurement system. The results of an
experiment undertaken to quantify the variation in sample grades
from the system sampler is also presented.
The third section of the report uses information from the first
two sections of the report and proceeds step-by-step through
various metallurgical accounting calculations that take into
consideration the errors associated with measurements. Through
these calculations, a better appreciation of the effect of errors in
various measurements can be gained as well as quantification of
confidences associated with some of the reportable metallurgical
accounting figures. The technique of ‘propagation of error’
will also be introduced with application to metal accounting
calculations.
MASS MEASUREMENT
Feed and concentrate weightometers
The predominant method of estimating mass flow of conveyed
material in a mining and processing operation is by means of
158 Fremantle, WA, 19 - 21 March 2007
weightometer, or belt scale. It is a device attached to a conveyor
that measures the weight of material on the belt over a certain
length and the speed of the belt. Using these measurements the
throughput, usually in tonnes per hour, and the total integrated
flow, usually in tonnes, are determined.
The estimate of concentrator feed tonnage reclaimed from the
primary stockpiles comes from a weightometer on the SAG mill
feed conveyor belt to each module. Each weightometer supports
four weigh frame idlers and contains four load cells. Module 2
has a typical mass flow of about 400 t/h. The weightometer is
located just outside of the stockpile reclaim tunnels on a
horizontal section of the belt before the crushed SAG mill
recycle is added to the primary mill feed.
Concentrate production is measured by weightometers on two
belts that convey the filtered concentrate to the load-out
stockpiles. These weightometers support two weigh frame idlers
each and contain two load cells each with a typical flow of
20 t/h. The smaller and more uniform flow of the filtered
concentrate stream makes for favourable belt weighing
conditions. Therefore using a shorter weighing platform and
fewer load cells is normally sufficient for this application.
For all of the concentrator weightometers, a belt speed
tachometer is located on the tail pulley. The information
provided by the integrators includes the total tonnes across the
weightometer, the throughput of material in tonnes per hour and
the belt speed in metres per second.
Calibration
The weightometers are calibrated during plant shutdowns.
Calibration involves two steps:
• a zero calibration, and
• a span calibration.
To perform the zero calibration, the feed of ore to the belt is
stopped while the belt is still running. Then an electrician
performs the zero calibration procedure built into the integrator
software.
The span calibration procedure is nearly the same as the zero
span calibration. For the feed weightometers, a large metal
cylinder of a known mass is loaded onto the weigh frame while
the unloaded belt travels over the weightometer. The weight that
this static mass exerts on the weigh frame corresponds to a
typical load on the weigh frame during normal operation. This
way the span calibration is performed as close to real operating
condition as possible within the limitations of the system. There
are a number of other calibration options for belt scales that more
closely replicate real operating conditions, such as roller chains
running over the weightometer or material tests because these
more closely replicate belt tensions experienced during normal
operation. However, the required infrastructure for these tests
does not exist for these weightometers, and thus static weight
calibration must be used.
Span calibration of the filtered concentrate weightometers can
be done a number ways that are preferable to those available for
the feed weightometers. The weigh platform for these scales has
a protruding peg upon which one can put increasingly heavier
masses to obtain an incremental span calibration. However, the
conveyor infrastructure allows for a material test, which is an
improved calibration because it measures under actual belt
loading conditions instead of those simulated by a static weight.
The integrator software includes a module for performing a
material test, where the actual concentrate is conveyed over the
weightometer and into a container (usually the back of a large
truck) for a certain period of time. The tared container is then
weighed on a certified weigh bridge. This value is compared with
the tonnage read by the weightometer. Thus, the accuracy of the
weightometers can be assessed. This also satisfies the definition
of accuracy as ‘comparison with a known measurement’ since a
certified weigh scale is used for comparison.
Ninth Mill Operators’ Conference
 METALLURGICAL ACCOUNTING IN THE NORTHPARKES CONCENTRATOR — A CASE STUDY

A series of material load tests were performed for one of the
filtered concentrate weightometers both as a means of calibration
and as a check of its accuracy. Two runs were completed, and
Table 1 shows the results. These two runs equate to two
calibrations, after the first of which, the weigh scale data was
entered into the integrator.
TABLE 1
CV15 material load calibration results.
Weightometer Weigh scale Difference
reading (t) reading (t) (%)
Run 1 7.7 7.5 2.7
Run 2 8.4 8.43 0.4
It can be reasonably assumed that the procedure adjusts the
weightometer readings by 2.7 per cent after the first run.
Therefore the unadjusted second reading would likely be 3.1 per
cent, and hence the accuracy of the scale is about three per cent
off and the precision is almost 0.5 per cent. This equates to a bias
of about two to three per cent plus a standard deviation
(precision) of about 0.5 per cent. A better estimate of the
precision would require more data.
While performing a material run calibration is preferable to a
static load calibration, it is not always practical to do this due to
time restraints or the possibility of a concentrate stockpile
blocking truck access to the falling material. A parallel static
calibration was not performed on this occasion, so a comparison
between the material run calibration and static calibration
methods was not possible.
Mass flow measurement
A weightometer uses two pieces of data to calculate total
tonnage and throughput over a belt: the speed of the belt
measured by the tachometer and the weight exerted by the ore on
the weigh frame as measured by the load cells. In order to
examine the mass measurement figures given by these devices,
these two measurements must be first individually examined as
each has a characteristic variance and either may suffer from
some bias.
speed as a speed measurement device whose calibration has
drifted has the potential to impart a substantial bias to tonnage
measurements. This experiment can help identify bias in this
measurement, but can not provide any information about the
measurement variance. Routine simple checks such as this for all
site weightometers can help ensure minimal measurement bias
imparted from belt speed data.
Belt loading
The second measurement used to calculate the total tonnes and
throughput on a conveyor belt is the actual weight that the
material exerts on the weigh frame as it passes over the
weightometer. These weight measurements are taken many times
per second by the load cells supporting the weigh frame and
signals are converted to mass per length of belt. This value is
multiplied with the speed of the belt to obtain a throughput value,
and the throughput value versus time is integrated to give total
tonnes. There are a number of issues to consider that can affect
the belt loading value.
As mentioned, using a static mass to simulate weightometer
belt loading conditions does not accurately represent a belt
loaded with ore. Belt tension from a fully laden belt can have a
significant effect on the total mass and throughput readings
(Fristedt, 1986).
Dirt and dust build-up on weigh frames and/or the calibration
mass can gradually bias tonnage high by placing a constant extra
weight on the weigh frame. Figure 2 shows a small build-up of
concentrate on the weigh frame of one of the filtered concentrate
weightometers. To illustrate the potential effect, a build-up of
200 g of concentrate on a 2 m weigh frame and a belt speed of
0.3 m/s can impart a bias of about 78 tonnes high over a one
month period. Thus, 78 extra tonnes would be totalised and
reported. Keeping a weigh frame dust free is a simple
housekeeping issue that can prevent one possible source of bias.

Belt speed
A check of the error in the belt speed measurement given by the
tachometer is not part of the weightometer calibration procedure.
The current procedure assumes that this figure is correct, which
may not always be the case. Hence, a check of the belt speed
used during a feed weightometer calibration was performed. Two
marks were put on the belt 40 m apart, and the time required for
both marks to pass a reference point on the conveyor frame was
measured using a stopwatch. The results of the three runs
performed are shown in Table 2.
FIG 2 - Material build-up on CV15.

TABLE 2
Results of three runs of belt speed test versus There are a large number of smaller factors that can also
weightometer speeds. impact the functionality of weightometers, such as correct
installation, alignment, belt tracking, clean idlers and uniform
Distance b/w Measured Calculated Weightometer flow of material on a belt. Assessment of these factors and their
marks (m) time (s) speed (m/s) speed (m/s) impact on weightometer readings is typically the domain of
Run 1 40.04 38.93 1.029 1.03 measurement consultants. Despite the ability of measurement
Run 2 40.04 38.74 1.034 1.03 experts to effectively diagnose and solve problems that would
negatively impact weightometer measurements, these diagnoses
Run 3 40.04 37.57 1.066 1.07
are not often able to be quantified in terms of accuracy and
precision for a weightometer.
The results of this simple experiment appear to validate the The error between the calculated load and measured load
speed measurements made by the tachometer mechanism of the during a belt scale calibration procedure should not only be
feed belt weightometer. It is important to routinely check belt placed with the maintenance records but also with metallurgical

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 159

W M JANSEN, R MORRISON and R DUNN
accounting records. This type of information can help to provide
an estimate of the active bias in the mass measurement systems
and can be considered when trying to explain some production
discrepancies between measurements. However, caution should
be exercised as it can be difficult to tell over what period this
bias, or portion of it, has been active.
Most of these means to combat the introduction of systematic
errors into the mass flow data via belt loading measurements can
be routinely implemented by maintenance technicians or
operators for more reliable metallurgical accounting mass data.
The majority of the error in a belt scale has been found to be
normally due to conveyor design and resulting belt effects and
user practices (Colijn, 1983).
Mass measurement variance
Ideally, one would like to be able to measure not only bias in the
belt loading and the belt speed, but also the precision for each
measurement. The key to an improved metallurgical accounting
system is to develop a greater understanding of the sources and
magnitudes of measurement errors both random and systematic.
Since belt speed and belt loading are the two raw measurements
used in calculating the wet feed mass to the plant, one must
understand the variation of measurements in a statistical sense
before a confidence can be quantified for this figure.
In reality, it is difficult to experimentally quantify the variance
of these measurements. Additional infrastructure would be
required for a meaningful experiment. Furthermore, the raw
measurements taken by the weightometers are processed in a
number of ways before being output. At the moment, the
variance for equipment such as this is estimated by measurement
experts, though the basis of the estimate is often unclear.
‘Accuracy’ is often claimed as a percentage of full scale without
reference to what confidence limits might apply.
Moisture content
The quantity of ore calculated from the measured speed and
loading of the conveyor belts is the wet tonnage. As most sites,
including Northparkes, report their production values in dry
metric tonnes (dmt), a moisture content factor is used to convert
the wet tonnage measurements from the weightometers to dry
tonnages. Thus, accurate determination and reporting of the mass
of conveyed material relies on an accurate assessment of the
moisture content of the material.
Feed moisture content accuracy
Because the concentrator feed has a coarse size distribution with
particles up to 200 mm+, representative sampling is a challenge
due to the large required sample sized. Currently, a large belt cut
sample (3 - 5 m) is periodically taken from the feed belt, and the
entire sample is weighed wet and dry in the tared drums in which
the sample is taken. Accurate moisture determination using this
method is difficult due to sample handling.
Because of the importance of using an accurate moisture
content value for the feed, an alternative moisture sampling
procedure is being investigated. Most of the moisture in the
coarse ore feed is present in the finest fraction, and the moisture
is usually a function of the size distribution of the material,
which can be more accurately determined. Thus more samples of
finer material can be collected, weighed wet and dry and sized,
and the moisture value can be routinely adjusted to suit the
conditions of the ore.
Feed moisture content variance
Although reasonable figures for the feed moisture content can be
determined, there is still no information regarding the variation
in this moisture value for the ore feed. It would be quite difficult
160 Fremantle, WA, 19 - 21 March 2007
and labour intensive to determine a relative standard deviation or
variance for a feed moisture and even more difficult to do this
task with any regularity as the ore changes. This would require a
large number of belt cuts from the same lot of ore with precise
determinations of moisture in each sample. Even with a
streamlined moisture content determination procedure, this
process would be difficult to do regularly. First, one must
consider whether or not it is worth doing.
The vast majority of variance in the dry feed mass figure
actually comes from the wet feed mass measurement from the
weightometer. For typical Northparkes tonnage data using a feed
moisture of two per cent, the moisture variance accounts for less
than 0.05 per cent of the variance of the dry tonnage figure.
Thus, regardless of the variance of a measured moisture factor,
this number is relatively insignificant. However, the variance of
the measurement is not to be confused with the accuracy of the
measurement as described earlier. A poor assumption will always
lead to a bias in the ‘mass processed’ estimate.
Concentrate moisture content variance
The filtered concentrate stream has different characteristics from
the feed stream, which make it easier to measure in a number of
ways. Taking representative samples for moisture is much easier
for this stream due to its much finer size distribution. Because of
this, sample handling is less complicated and loss of moisture is
easier to prevent. Northparkes take an hourly moisture cross-belt
grab sample from the filtered concentrate belt and add it to a
shiftly composite sample in a sealed container.
A simple experiment was conducted to quantify the variance
of such moisture content measurements used in the calculation
of dry concentrate mass. Ten cross-belt grab samples of filtered
concentrate were taken from each of the two filtered
concentrate belts in quick succession and placed individually in
sealed bags. Once the samples were taken, the wet and dry
masses were weighed in the lab using the current moisture
analysis procedure.
The goal of this experiment is to examine the variation in
moisture content between the ten samples for each belt in order
to quantify the sampling variance that one can expect for a given
moisture sample. The underlying assumption of this experiment
is that the concentrate sampled from each belt was from the same
batch and had the same moisture content. The results of the
experiment are shown in Table 3.
TABLE 3
Moisture contents and statistics for filtered concentrate samples.
CV14 CV15
Mean 7.67 Mean 8.61
Rel SD 1.34% Rel SD 1.05%
Variance 0.0106 Variance 0.0082
The results from the table show relatively little variation
between the moisture samples taken, suggesting that sampling
accuracy and moisture analysis process are highly repeatable (ie
they have good precision). This experiment does not, however,
conclude anything about moisture variations during a shift which
might impact the calculated dry concentrate mass. Thus it may
be useful to further examine typical variations in moisture over a
shift in order to determine optimal sampling frequency, and this
investigation is continuing at Northparkes.
These variances are relatively small considering their impact
on the final dry tonnes calculation. As with the feed stream, a
much greater contribution of overall dry tonnes variance comes
from the wet mass measurement from the weightometers.
Ninth Mill Operators’ Conference
 METALLURGICAL ACCOUNTING IN THE NORTHPARKES CONCENTRATOR — A CASE STUDY
GRADE MEASUREMENT
AMDEL sampling system
Grades for metallurgical accounting streams in the concentrator
are obtained from composite sample cuts of the sample stream
feeding an AMDEL on-stream analyser. The on-stream data from
the AMDEL analyser is used specifically for process control,
whereas the data used for daily and monthly metallurgical
accounting comes from lab analysis of the composite samples.
These are physical samples taken at pre-set time intervals by an
automatic cutter mechanism through the sample slurry streams.
The slurry streams sampled for process control and/or
metallurgical accounting purposes and their corresponding
sample cut frequencies are listed in Table 4.
TABLE 4
Data usage and cut frequencies for AMDEL on-stream sampler.
Slurry stream Process Met Cut frequency (min)
control accounting Module 1 Module 2
Flotation plant feed X X 30 30
Flash clotation X X 60 20
concentrate
Final concentrate X X 40 15
(combined)
Final tailings X X 10 40
Rougher tailings X - - primary samplers is representative of the slurry. This is highly
Cleaner scavenger X - - dependent on proper implementation and maintenance of the
feed
The slurry that is sent through the ADMEL analyser comes
from primary sampling mechanisms installed on the main slurry
lines in the flotation plant. Most of these are pressure pipe
samplers. These samplers depend on a turbulent feed stream in
order to minimise bias in the obtained sample, and this is
achieved by installation and regular replacement of turbulence
bars in the main slurry line just prior to the sample tap.
Recommended installation for this type of sampler is in a vertical
rising line with at least ten diameters of straight pipe before the
sampler (Outokumpu, 2006).
Sampling experimental procedure
An experiment was devised to examine the variance of sample
grades between samples taken immediately after one another for
each stream. The goals of this experiment are to observe the
sample cut mass variation and to examine the variation in grade
of successive samples. Finally, determining and modelling a
relationship between sample grade and expected variance, if one
exists, would be an important contribution to an improved metal
accounting strategy at Northparkes.
The procedure for the AMDEL analysis experiment consisted
of the following steps:
1. collect, label and obtain tare weights for enough buckets to
obtain seven samples of each metallurgical accounting
slurry stream from modules 1 and 2 as well as cleaner
scavenger feed and rougher tails streams from module 2;
2. during ‘typical’ operating conditions, take a number of
manual cuts from the AMDEL machine for the first sample
for the first stream sampled so that there is enough sample
for preparation and duplicate assay analysis;
3. as quickly as possible, take the same number of cuts for the
next five samples in succession for that first stream;
4. for the seventh sample of each stream, take twice as many
cuts as taken in each of the first six samples;
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
5. seal buckets to prevent loss of moisture in samples;
6. repeat steps two through five for the remaining streams to
be sampled;
7. obtain wet mass + bucket of each sample and subtract
bucket tare to obtain wet mass of sample;
8. filter and dry each sample in oven until dry and obtain dry
mass of sample by subtracting filter paper weights from dry
mass + filter paper mass;
9. for the seven samples for each stream, prepare sample and
obtain subsample for analysis using same procedure as a
normal metaccounting composite sample;
10. submit each subsample for duplicate analysis for both
copper and gold; and
11. obtain lab assay results and begin analysis.
The overall variance of a typical daily AMDEL composite
sample is comprised of many component variances such as: the
variance from the sampling procedure, the variance from sample
handling and preparation, and the variance from the assay
procedure. For this experiment, time and labour constraints
allowed quantification only of the overall variance and the
variance from the assay procedure by using duplicate analysis.
For metallurgical accounting, however, only the overall variance
is necessary to quantify the overall measurement precision.
It should be noted that two basic assumptions are used in this
experiment. Firstly, it is assumed that the sample tapped from the
slurry pressure pipe sampler systems. Secondly, it is assumed
that the successive samples taken are from the same lot of
material and that the small amount of time between taking
samples does not affect the material being sampled. In other
words, the assumption is that the short-term variation of the
slurry properties over the sampling period is negligible in this
experiment. An interleaved sampling experiment would not
require this second assumption by factoring out the variance
potentially introduced by the short time difference between
samples.
Sample data analysis
The underlying goal of this experiment is to determine an overall
variance that can be expected for samples taken from the AMDEL
equipment for daily metallurgical accounting composite samples.
Thus, while there are many uses in quantifying the variance
components of the overall variance, the overall variance is of
greatest importance in a larger metallurgical accounting sense. The
total variance of a measurement is required for calculating a
statistical confidence interval for that measurement.
Samples of the rougher tailings and cleaner scavenger feed
streams were also taken to assess their potential of reduction of
variance of calculated results, even though they are not currently
used for metallurgical accounting. These samples were also
collected and analysed in order to obtain a generalised error
model.
Sample cut masses
The current AMDEL system at Northparkes takes sample cuts at
predetermined time intervals that make up 24-hour composite
samples. If some cuts are much larger than others, they will
represent a larger portion of the composite sample than they
should. This can add a bias to the composite sample and, if large
enough, could misrepresent the production grades for that day.
Thus, the variation in cut masses of the AMDEL samples was
first explored to quantify this effect on the daily sample grades.
161
W M JANSEN, R MORRISON and R DUNN
From the wet and dry masses of the samples, the average mass
per cut of the automated sampler can be calculated. Because each
sample in this experiment consisted of more than one cut of the
sample stream, the average mass per cut had to be calculated by
dividing the total dry mass by the number of cuts in that sample.
By doing this instead of taking one cut per sample, some of the
variability between individual cuts is inevitably smoothed out.
However the results are illustrative of the magnitude of
variability in cut mass that can be expected from each stream.
Table 5 shows the mean dry mass per sample cut and the cut
mass relative standard deviation for each sampled stream.
TABLE 5
Standard deviations of AMDEL sample cuts.
Module 1 Module 2
Avg dry Rel std dev Avg dry Rel std dev
mass per cut (%) mass per cut (%)
(g) (g)
Float feed 49 8.35 70 5.62
Final con 125 8.15 56 4.25
Final tails 45 3.98 108 3.42
Rougher T - - 123 5.93
Cl Sc feed - - 18 7.89
Fl Fl con 265 7.32 107 39.17
The table reveals a difference in the consistency of cuts
between sampled streams with the lowest standard deviation in
the final tailings streams and the highest in the module 2 flash
flotation concentrate (due most likely to surging). While relative
SDs of eight per cent and greater may seem quite high at first,
the effect is negligible as there are many cuts taken over the
24-hour period in which the composite sample is made.
Spreadsheet simulation reveals that even with the cut mass
variations shown in Table 5 and very high short-term grade
fluctuations throughout a day, the effect on the daily composite
grade is minimal.
Sample variation models
This study is primarily concerned with the use of overall
variance in a metallurgical accounting and reporting context. The
variance for each stream of interest can be calculated from the
lab assay results from the AMDEL experiment. This information
can show the variance that may be expected for certain grades of
material, but it is even more useful to be able to predict the
expected variance at any grade of material coming through the
system. If there was a relationship determined between the grade
and the variance that could be expressed in a single variation
model, this model could then be used to predict the variance for
any grade. Thus, the next stage of the analysis is to determine if a
relationship exists, and, if so, the best way to model this
relationship for general metallurgical accounting usage.
Two common ways of expressing measurement variation are
using standard deviation or variance. In order to develop a
variation model from the experimental data, the standard
deviations of the assay values for copper and gold were
calculated for each stream for both modules and then plotted in a
graph to determine if a relationship in the data exists.
Application of a single term error model to data such as this
must be carefully considered. For measurements of very low
concentrations (such as very low grades), a constant minimum
error is often found. This means that the error, measured in this
case as standard deviation, does not grow continuously smaller as
the measured grade of material decreases, but rather levels off to
some minimum value. This can be very important to consider if
162 Fremantle, WA, 19 - 21 March 2007
this error is significant. For high grade materials, measurement
error increases with the assay value and will typically flatten out
to a maximum value eventually. These ‘end behaviours’ of an
error model often necessitate the use of two-term error models,
such as those seen commonly in analytical chemistry. These use
an additive term for low-concentration measurements and a
proportional term for high concentration measurements (Rocke
and Lorenzato, 1995).
A two-term model was considered, but in this case, it was
deemed unnecessary because the standard deviation for the assay
data at very low assays was always a very small number
approximating zero. The additive error term usually applied to
account for the minimum error at low concentrations was taken
to be zero in this analysis. Thus, a one term proportional error
model is left.
The type of model selected satisfies the need for a single term
proportional model whose error levels off to a maximum value at
high grades. It is also the model that has the best fit regression
for the data. The data points in Figure 3 show the copper grade
versus the corresponding standard deviations for the feed, final
concentrate and final tailings for each module. The model shown
on the figure is a least-squares regression for a power equation.
(Note the use of logarithmic scales for the axes.)
FIG 3 - Copper error model showing assay versus SD.
The equation describing this regression line is an error model
which relates generically the grade of the sampled stream of
interest to the variation that one can expect at that grade using
the current sampling and assaying procedure. For the copper
error model shown in Figure 3, the mathematical model is of the
form:
SDabs = 0.0181 ∗ ACU
0 . 9332
(1)
where:
SDabs is the absolute standard deviation of the assay value
ACU is the copper grade in per cent copper (for example, a
33 per cent copper assay is entered as 33 and not 0.33)
For this model, an R-squared value of 0.948 was achieved;
R-squared is a statistical measure of how closely the regression
line approximates the actual data points where 1.0 would be a
perfect fit of the data.
To better illustrate the behaviour of this model Figure 4 was
developed, which shows both the same absolute model as shown
in Figure 3 (solid line) as well as the calculated relative standard
deviation (dashed line). The secondary y-axis on the right shows
the scale for the relative standard deviation model. Note the
x-axis showing copper assay is no longer in log scale so as to
better illustrate the end behaviour of the models.
Ninth Mill Operators’ Conference
 METALLURGICAL ACCOUNTING IN THE NORTHPARKES CONCENTRATOR — A CASE STUDY
It can be noted from Figure 4 that the absolute standard
deviation increases sharply until about ten per cent copper, after
which it begins to level off. There is very little increase in the
absolute SD at copper grades above 30 per cent. The relative SD
curve shows a high relative standard deviation for low grades
(feed, tailings) that decreases as grade increases until finally
levelling off at higher grades. The chosen one-term absolute SD
model and accompanying relative SD model require no additive
constant-error term in this case, and they also satisfy the need for
the higher grade material assays to level-out to a maximum
standard deviation.
FIG 4 - Copper error model showing assay versus absolute SD
and relative SD.
A similar analysis procedure used with the copper data was
applied to the gold assay data for each stream. The grade versus
standard deviation was again plotted, and a regression was
performed as shown in Figure 5.
FIG 5 - Gold error model showing assay versus SD.
For the gold data, the model describing this general regression
is:
SDabs = 0.0354 ∗ AAU
0.8328
(2)
where:
SDabs is the absolute standard deviation of the assay value
AAU is the gold grade in g/t Au
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
FIG 6 - Gold error model showing assay versus absolute SD
and relative SD.
The R-squared value of this regression was 0.913. The relative
SD is shown along with the absolute SD in Figure 6. Similar
behaviours are seen in this figure as with the copper model
shown before.
One source of improvement for the gold error model would be
to expand this model to higher ranges of gold assays through
future sampling and variance experiments. While copper assays
at Northparkes rarely extend above 40 per cent copper, the gold
assays are frequently above the 12.5 g/t maximum encountered
during this sampling experiment.
The next step in this study is to incorporate this information in
an updated Northparkes metallurgical accounting system. The
standard deviation estimates provided by these error models
would be used to estimate standard deviation/variance for daily
AMDEL composite samples. The standard deviations would then
be used to calculate and assign statistical confidence levels to the
measured assay values that are either directly reported or used to
calculate other reportable figures. These models can also be
updated by periodic experimentation as mine life progresses.
Using raw measurements and their associated variances to
calculate other values (eg recovery) with associated confidences
can be done using a technique called propagation of error, which
is the focus of the next section.
METALLURICAL ACCOUNTING CALCULATIONS
Introduction
The concentrator metallurgical accounting strategy at Northparkes
relies upon a number of sampling and measurement sources.
Some of the reportable figures for an accounting period are
directly measured, but many are derived from direct
measurements. In order to derive reportable figures for a period,
such as recovery or back-calculated feed grade, measurement
data is fed into various equations to calculate these. Thus it is
useful to first identify the sources of the measurement data and
the equations currently used to derive reportable figures. Then
calculations will be shown that incorporate measurement
error information from this study. Finally, there will be a
demonstration of how variance in individual measurements
propagates through metal accounting calculations to give overall
errors and confidence limits for derived figures.
Reportable metallurgical accounting figures
In order to examine the metallurgical accounting calculations, it
is often helpful to start at the end and work backwards. First, it is
necessary to examine the metallurgical accounting results that
163
W M JANSEN, R MORRISON and R DUNN

must be reported, and then outline the calculations used to arrive

at those figures. Table 6 shows an excerpt from an example
(
F = Fwet − Fwet ∗ Qfeed ) (5)

monthly production report for Northparkes including some of the

relevant reportable figures. C = C wet − (C wet ∗ Qcon ) (6)

where:
TABLE 6
Excerpt from example Northparkes production report. Fwet is the wet feed mass
Qfeed is the feed moisture content
Parameter Unit Module 1 Module 2 Total
SAG mill feed dmt 129 831 202 631 332 462 Cwet is the wet concentrate mass
Feed grade Cu % 0.83 1.08 0.98 Qcon is the concentrate moisture content
Au g/t 0.52 0.37 0.43 Equations 3 and 4 show a mass balance for the overall mass of
Concentrate Cu % 32.89 36.17 35.02 material entering and leaving the concentrator, but a component
grade Au g/t 17.26 9.81 12.42 balance can also be performed, which equates the amount of
copper or gold entering the concentrator with the amount of
Recovery Cu % 91.12 91.49 91.37
copper or gold leaving, based on the assay grades of the material.
Au g/t 76.76 71.60 74.03 This is expressed in Equation 7:

This metallurgical accounting and reconciliation study is Ff = Cc + Tt (7)

concerned primarily with monthly accounting periods. At
Northparkes there is not enough data available over a daily where:
period to perform accurate accounting due to: f is the grade of copper or gold (grade is in per cent Cu or g/t
1. the presence of a flash flotation system, and Au) of the feed
2. the lag between the production of concentrate and the c is the concentrate grade
concentrate filtration and mass measurement. t is the tailings grade
Because this lag becomes negligible over a monthly Because the actual feed grade of material is not directly
accounting period and the monthly figures are more important measured due to difficulty of representative sampling and the
for corporate reporting, examination of the monthly accounting presence of a flash flotation circuit, it must be back-calculated
figures is the focus of this study. Including hold-ups in the from other measured values and the back-calculated tailings
balance could reduce this accounting period to a few hours. tonnage (from Equation 4). Rearranging Equation 7, this is
shown in Equation 8:
Metallurgical accounting equations
Cc + Tt
Now that the figures required for metallurgical accounting f = (8)
reporting have been outlined, the next step working backwards is F
to examine the equations used to calculate these figures. The
current Northparkes metallurgical accounting system uses the The next reportable figure to calculate is the plant copper or
two product formula based on a simple mass balance, which gold recovery based on the measurements and back-calculated
states that what goes into the concentrator is equal to what comes figures from the two-product formula. Recovery, R, is simply the
out. This is expressed mathematically in Equation 3 as: proportion of material of interest that was recovered to the
concentrate divided by the total amount of material of interest
F =C + T (3) fed to the plant, and this is expressed as a percentage through
Equation 9:
where:
 Cc 
F is the dry feed tonnage R =   .100 (9)
 Ff 
C is the dry concentrate tonnage
T is the dry tailings tonnage Rearranging and substituting values from the various forms of
The implicit assumption in this equation is that hold-up in the the two-product formula will yield a recovery calculation based
plant is negligible compared with the flow through the plant for only on the grades of the material, expressed in Equation 10:
the operating period. If this is not true, a term for the change in
content (closing stock – opening stock) must be added. As c( f − t)
R= .100 (10)
Northparkes only measure the tonnage of feed coming into the f (c − t)
plant and the tonnage of filtered concentrate leaving the plant,
the tailings tonnage must be back-calculated from these The ‘contained metal’ in the concentrate figure is the
two figures, which is expressed with a simple rearranging of remaining reportable figure to be calculated. This is determined
Equation 3 to Equation 4: simply by multiplying the concentrate dry tonnes, C, by the
concentrate grade, c, as shown in Equation 11:
T = F −C (4)
M = C ∗c (11)
Because the above equations are based on the dry tonnages of
material, the F and C terms in the equations must be calculated where:
since only wet tonnages are measured by the feed and
concentrate weightometers. These wet tonnages must be M is the contained metal in the concentrate, which is either
multiplied by a measured moisture content value for each stream tonnes of copper or grams of gold, depending on the assay,
as shown in Equations 5 and 6: c, used

164 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 METALLURGICAL ACCOUNTING IN THE NORTHPARKES CONCENTRATOR — A CASE STUDY

Metallurgical accounting data sources TABLE 9

The final step in working backwards to examine the concentrator List of metallurgical accounting variance and variance sources.
metallurgical accounting system at Northparkes is to list the
sources of data used in calculating reportable figures. This Variance Variable Source
measured data was featured in the metallurgical accounting Wet feed mass var σ 2 Fwet Estimated
equations above, but is listed more concisely in Table 7, which
shows each reportable figure required along with the data Feed moisture content var σ 2 Qfeed (negligible)
required to obtain each figure. Wet concentrate mass var σ 2 Cwet Estimated
Concentrate moisture content var σ 2 Qcon Measured (negligible)
TABLE 7 Concentrate grade var σ 2c Exp error models
Monthly metaccounting reportable figures and data required Tailings grade var σ 2t Exp error models
for their calculation.

Reportable figure Variable Units Required To quantify the confidence levels in the measured data
measured data themselves using their variances, mean and sample size,
SAG mill feed F dmt Fwet, Qfeed Equation 12 can be used:
Feed grade f Cu %, Au g/t Cwet, Qcon, Fwet, zs
Qfeed, c, t µ=x± (12)
n
Concentrate grade c Cu %, Au g/t Directly measured
Recovery R %, Cu and Au Cwet, Qcon, Fwet,
where:
Qfeed, c, t
Concentrate filtered C dmt Cwet, Qcon µ is the estimate for the value
Contained metal M Cu (t), Au (g) Cwet, Qcon, c x is the calculated mean value of the data
(filtered)
s is the calculated standard deviation
Concentrate Qcon % Directly measured
moisture n is the sample size
z is defined as the number of standard errors from the mean,
The table shows that all of the concentrator monthly which depends on the confidence level to which one would
metallurgical accounting reportable figures are calculated from like to calculate
six basic measurements taken: two wet mass measurements, two The value for z used to calculate 95 per cent confidence limits
moisture contents and two sampled grades. Each of these six is 1.96.
measurements will have an inherent measurement error For example, if 40 samples were taken of copper concentrate
associated with it, typically expressed as a standard deviation or with a calculated mean value of 36.5 per cent Cu and an absolute
variance. Table 8 summarises each measurement required for standard deviation of 0.55 per cent Cu, the 95 per cent
obtaining reportable figures and the respective measurement
confidence levels would be 1.96*0.55/(√40) = 0.17 per cent Cu.
sources.
This can be expressed by writing that the estimated copper grade
for the sampled lot is 36.5 per cent ± 0.17 per cent at a
TABLE 8 95 per cent confidence level. Statistically, this means that one
List of metallurgical accounting data and data sources. can state that the true value is between 36.33 per cent Cu and
36.67 per cent Cu with only a one in 20 chance of being wrong.
Data Variable Source of Data This above calculation is applicable for measured quantities
Wet feed mass Fwet CV10 and CV03 feed where a variance or standard deviation can be experimentally
weightometers determined or estimated. However, to calculate a confidence
Feed moisture content Qfeed Experimentally determined interval for figures derived from measured data, one can obtain a
factor (not routine) variance or standard deviation for the derived figure using a
technique called ‘propagation of error.’ Using this technique, the
Wet concentrate mass Cwet CV14 and CV15 con
weightometers individual errors in each piece of data used for calculating the
reportable figure contribute to the estimate of error in the derived
Concentrate moisture Qcon Routine hourly grab samples figure. This process is outlined briefly in the following section.
content
Concentrate grade c AMDEL daily composite Error propagation through metallurgical
samples
accounting calculations
Tailings grade t AMDEL daily composite
samples Having an estimated or measured variance for all measured data
allows one to assign confidences to each measured value as well
as each calculated value by propagating the error through the
Assuming bias-free measurements, the variance in each of calculations. The propagation of error through any function can
these measurements is required to quantify the confidence one be determined using the law of partial differentiation. This
has not only in the actual measured value but also in any figures technique is actually an approximation for when the relative
that are calculated from measured data. The quantification and errors are small, as in the errors in the measurements in question
estimation of these variances has been one of the central focuses here, and it is based on the Taylor series expansion of the error of
of this study, and the sources of estimations for these variances in a multivariable function. The technique is often referred to as
the concentrator are shown in Table 9. ‘propagation of variance’ when errors are replaced variances. For
Once the measurement variances have been quantified or example, if z = f(x, y), then one can calculate the variance in z
estimated, this information can be used to quantify the statistical based on the variances in x and y using the following
confidence levels in each of the measured and calculated figures. Equation 13:

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 165

W M JANSEN, R MORRISON and R DUNN

2
 ∂f  involves summation and weighted averaging. The variance in the
2
 ∂f 
σ 2z =   σ 2x +   σ 2y (13) aggregated values must also propagate through these calculations
 ∂x   ∂y  to the monthly totals and averages as well.
For example, when reporting the monthly feed tonnage, the
where: daily feed tonnages for the month are added for a total monthly
σ 2x is the variance in x tonnage. The variances of each daily tonnage value must be
propagated (in this case they must be added) in order to obtain
σ 2y is the variance in y the overall variance for the monthly tonnage value. For a
monthly concentrate grade, a weighted average based on the
(∂f / ∂x ) is the partial derivative of f with respect to x daily concentrate tonnages and daily concentrate grades is
(∂f / ∂y) is the partial derivative of f with respect to y calculated, and the variances of each of these values for each day
must propagate through the weighted average calculation,
This relationship is true when the random variables x and y are according to the principles outlined above.
independent. From this point on variance will be denoted with An Excel spreadsheet has been developed to automate the
the variable V instead of σ 2 . propagation of variance calculations for these variances and
For calculation of the dry feed and concentrate masses from relative standard deviations, which also calculates the statistical
wet masses and moisture contents, Equations 14 and 15 show the confidence intervals for each figure for reporting. The
variance propagation calculations for dry feed mass variance and spreadsheet also produces a monthly metallurgical/production
dry concentrate mass variance, respectively: accounting report showing the values and their confidence
intervals. An example excerpt of a report from this spreadsheet is
VF = (1 − Qfeed ) VFwet + Fwet
2 2
VQfeed (14) in Table 10.
This new report shows the value reported along with the
VC = (1 − Qcon ) VCwet + C wet absolute confidence intervals (eg tonnes plus or minus tonnes, or
2 2
VQcom (15)
per cent Cu plus or minus per cent Cu) as well as relative
confidence intervals. The production accounting statistics have
The propagation of error through the two-product formula also been separated from the metallurgical accounting statistics.
calculation of feed assay from mass measurements of feed and A close inspection of the absolute and relative confidence
concentrate and their variances and from grade measurements of intervals for the recovery values between modules 1 and 2 in
concentrate and tails and their variances is a well-understood Figure 8 reveals a discrepancy. The data used to generate this
process. The formula for calculating the variance, Vf, of the report comes from early 2005, when module 1 was still treating
calculated feed assay based on these measurements and variances oxide ore, which had generally poor recoveries, high concentrate
is shown in Equation 16: moisture values and high tailings grades (compared to module 2).
The higher tailings grade of module 1 is directly responsible for
 (c − t)  C (t − c)
2 2 2 2
 C C  the higher module 1 recovery SD since, according to the error
Vf 1 −  Vt +   Vc +   V +  V (16) models shown in Figures 3 - 6, this higher grade translates to a
 F  F  F  C  F2  F
higher absolute SD. Thus, the high absolute tailings grade SD
propagates through the calculation of recovery and results in a
Calculating the variance for the recovery, VR, using high recovery SD. Tailings grade SD is the most dominant
propagation of variance follows the same pattern as for the last contributor to the recovery SD in the propagation of error
few calculations and is shown in Equation 17: calculations (from a recovery calculated only from grades),
whereas it has very little contribution to the back-calculated feed

VR =
1  c 2t 2
+
( f − t)2 t 2 V + c 2 (c − f )2 V  grade SD. It must be pointed out, however, that the models were
 V t (17) developed using sulfide ores and may not be appropriate for use
f 2 (c − t)  f (c − t)2 (c − t)2 
2 2 f c
with old oxide data.
Knowledge of the confidence levels in each measurement and
Using the above calculations, variances (and standard calculated value allows the site to pinpoint areas for
deviations) can be obtained for all of the calculated figures based improvement in reportable figure precisions and allows one to
on the measured data and their variances. understand the level of confidence that can be placed in each
For monthly metallurgical accounting, some reportable figures of these reported values. Thus, not only are there technical
are obtained from an aggregation of daily values, which typically benefits to this information but improved corporate governance

TABLE 10
Production report including uncertainty values.

Metallurgical accounting statistics – ore processing (February 2005)

Parameter Unit Module 1† Module 2†
Value ± Absolute Relative Value ± Absolute Relative
SAG mill feed dmt 155 505 ±2915 ±1.9% 238 479 ±4655 ±2.0%
Feed grade % Cu 1.17% ±0.02% ±2.0% 1.01% ±0.02% ±2.4%
g/t Au 0.57 ±0.012 ±2.1% 0.43 ±0.009 ±2.2%
Concentrate % Cu 36.36% ±0.56% ±1.6% 35.33% ±0.54% ±1.5%
grade g/t Au 19.41 ±0.33 ±1.7% 11.71 ±0.20 ±1.7%
Recovery (Cu) % 49.89% ±1.82% ±3.6% 90.89% ±0.36% ±0.4%
(Au) % 54.05% ±1.80% ±3.3% 70.75% ±1.22% ±1.7%

† All values reported at a 95 per cent confidence level (±2σ).

166 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 METALLURGICAL ACCOUNTING IN THE NORTHPARKES CONCENTRATOR — A CASE STUDY

compliance comes as well. Another interesting bit of information

that can be obtained through a study on error propagation is the
sensitivity of the error in the final reportable figures to changes
in the individual measurement errors of measurements used to
derive the figure.

Sensitivity analysis of derived figures

The error in derived figures comes from propagation of the errors
of the measurements used in their calculation. Both calculated
feed grade and recovery require six measurements for their
calculation, and the error in each of these measurements will
affect to some extent the overall error in a derived figure. Some
errors will have a larger impact than others, and to investigate the
impact of each measurement and its error on the feed and
recovery figures, a sensitivity analysis was performed.
To perform this sensitivity analysis, each of the individual
component measurement SDs was varied over a range of values
while keeping the other five component measurement SDs
FIG 8 - Effect of measurement SDs on calculated copper
constant. The effect on the calculated feed grade SD and
recovery SD.
recovery SD for both copper and gold was recorded for a range
of values for each scenario. This analysis was done for all six
measurements to observe their individual impacts on the final
feed and recovery standard deviations. This sensitivity analysis TABLE 11
was performed on a set of example monthly production data Sensitivity analysis rel SD values held constant.
from Northparkes (February, 2005). Thus, the measurement SDs
were altered for all 28 days of the month for each measured Measurement SD Value held constant
quantity individually and the effect on the monthly feed grade Cu Au
and recovery SD values is shown. Table 11 shows the relative SD
Concentrate grade SD 1.4% 2.4%
values used for the measurements held constant during the
analysis. These ‘baseline’ SD values represent estimates of SDs Tailings grade SD 2.4% 4.9%
for these measurements based on estimations and investigations Wet feed tonnage SD 5% 5%
in this report. Wet con tonnage SD 2.5% 2.5%
Figures 7 and 8 illustrate the effect of altering the individual Feed moisture SD 5% 5%
measurement SDs on the copper feed grade SD and copper
recovery SD, respectively. The x-axis represents the measurement Con moisture SD 1.2% 1.2%
SDs for the altered individual measurements and the y-axis
shows the corresponding feed grade or recovery relative SD. and recovery and would therefore both appear as near horizontal
Thus, each line on the graph represents an individual sensitivity lines on the graphs. Both copper and gold showed the same
analysis as the remaining five measurement SDs are held trends for the various measurement SDs.
constant according to Table 11.
The graphs show that the wet feed tonnes SD and wet
concentrate tonnes SD have the greatest effect on the calculated
feed grade SD. The greatest reduction in feed grade SD would
come from reduction of the wet feed tonnes SD. Conversely,
reduction of the tailings grade measurement SD (and the two
moisture SDs) would have the least effect on the feed grade SD.
Put simply, if one would like to improve the confidence
associated with the calculated feed grade, improvement in
weightometer precision is most important followed by tailings
sampling for grade.

REFERENCES
Colijn, H, 1983. Weighing and Proportioning of Bulk Solids, Chapter 7,
pp 207-219 (Trans Tech Publications: Clausthal-Zellerfeld).
Dodge, Y, 2003. The Oxford Dictionary of Statistical Terms, sixth edition
(Oxford University Press: New York).
Fristedt, K, 1986. Belt Weighing Errors – From where do they origin,
Bulk Solids Handling, 6(5):963-968.
Outokumpu, 2006. Slurry sampling for online analysis and accounting,
FIG 7 - Effect of measurement SDs on calculated copper feed [online]. Available from: <http://www.outokumputechnology.com/
grade SD. 35365.epibrw> [Accessed: 11 September 2006].
Rocke, D M and Lorezato, S, 1995. A two-component model for
measurement error in analytical chemistry, Technometrics,
Only four of the six measurement sensitivity runs are shown 37(2):176-184 (American Statistical Association and American
on each graph. This is because the feed and concentrate moisture Society for Quality Control).
measurement SDs have a negligible effect on the SDs of the feed

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 167

Managing Mineral Processing Plants in Australia — Tips and
Pitfalls
R S Francis1 and G D Cooper2
ABSTRACT
There are no manuals or handbooks for budding mill superintendents or
process plant managers. Most who obtain this level of responsibility have
gained their training from watching the good and bad of their
predecessors and the successful ones have themselves graduated from the
school of hard knocks. This paper attempts to provide some helpful
guidelines and direction for those who aspire to better manage a mineral
processing facility. The authors have 20 years combined experience in
senior management roles at nine different mineral processing operations
in Australia and overseas, covering gold, base metals and mineral sands.
This summary of tips and pitfalls will hopefully benefit future leaders in
mineral processing and metallurgy.
INTRODUCTION
This paper does not cover the technical and design aspects and
issues of managing mineral processing plants, but does make
reference to some technical functions that should be carried out.
Nor does it cover every aspect of the job or that of a particular
commodity, as it targets what are, hopefully, the main issues that
should be addressed by a person in that position of responsibility.
The focus is, however, on the issues that take up the most time –
people.
The title Plant Manager has been used as the generic term for a
variety of similar roles in Australian mineral processing plants,
such as Mill Superintendent, Plant Superintendent, Manager –
Ore Processing, Metallurgy Manager, etc.
The title Site Manager has been used as the generic term for
roles that have titles such as Resident Manager, General Manager
and Mine Manager.
Twelve key areas have been selected that the most senior
mineral processing official on site should pay close attention to.
These are by no means all the areas required to perform the
function to the utmost capability, but cover the most important
areas. The degree of importance and the order of priority will
vary from person to person, from site to site and from company
to company.
Attention to environmental, safety and occupational health
issues is always of paramount importance in every operation. It is
assumed, therefore, that this is integrated with each of the 12
selected areas.
CRITICAL ACTIONS
Assemble a good team
One cannot overemphasise the importance of having a strong,
skilled and cooperative team. In addition, particularly for fly-in,
fly-out (FIFO) operations it is critical to have a supportive and
capable second-in-charge (2IC) to run the operation whilst you
are on rest and recuperation (R and R). This person should have a
common and consistent approach to the operations as you do.
1. FAusIMM, Group Metallurgist, Consolidated Minerals Ltd, 28
Ventnor Avenue, West Perth WA 6005.
Email: rfrancis@consminerals.com.au
2. MAusIMM, Senior Project Metallurgist, Sinclair Knight Merz, 6th
Floor, 16 St Georges Terrace, Perth WA 6000.
Email: gcooper@skm.com.au
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
It is critical that a comprehensive handover process is put in place
to ensure consistency and clear understanding of the requirements.
Upon taking up a new position, you usually inherit an existing
team, and you will have to make the most of it. Opportunities
will arise to make changes when there is a turnover of people.
Our experience has shown that the best make up of teams is with
a balanced mixture of personalities and characteristics (as
measured, for example, by the use of a tool such as the Myer
Briggs Type Indicator (MBTI)) and where a healthy level of
conflict or tension is present. In addition, there should be a
balanced mixture of experience and skill levels with attention
given to career succession planning (ie do not overload with all
seniors and no juniors or vice versa). Ideally the members of the
team should have a good mix of ‘doers’ who are going to make
things happen and ‘thinkers’ who are going to come up with new
ideas.
Identify the ‘knockers’ and ‘shockers’, ie those who are
negative at all times, and those who are poor performers doing
the absolute minimum to get by. These employees can obstruct
and hinder progress sometimes in an underhanded way. You will
need to address the issues with them to either modify their
behaviour or be prepared to take a more drastic step. Sometimes
you will come across the ultimate challenge of the employee
who retains the right to be miserable at all times. These are the
most difficult to manage (or manage out), because they usually
do a reasonable job.
Quite often, for a variety of reasons there is an unhealthy
rivalry between the operating shift crews. This can lead to a
severe drop off in morale and contribute to additional production
losses. Some tactics that have been successfully used to deal with
such a situation include:
• Swapping around the shift supervisors onto other crews.
Often crews work together for a long period, get to know
each other’s strengths and weaknesses, but can become
complacent. Usually, the person being changed around has
some initial objections, but after a few shift rotations is quite
supportive of the change.
• Split the shift crew. Usually, the whole shift has shift
changeover at the same time. Therefore, they never cross
over with other crews. By splitting the crews so that about
half the crew change over a couple of days before or after the
rest of the crew, a more healthy relationship is developed
with the other overlapping crews. This can also assist in
improving operator coverage at shift change.
• Form a crew that covers day and night shift. At a FIFO
operation, changes were made to the rosters from a 2/1 to a
9/5 (Francis and Hille, 2002). Instead of having four separate
crews, only two larger crews were created that covered day
and night shift for the time on site. In the next work rotation,
it is possible to work alongside someone who was on the
night shift while you were on the day shift.
If you are fortunate enough to work in a sufficiently large and
progressive company, and there are suitable technical challenges,
take the opportunity to introduce graduates into the department
on a regular basis. This brings in new blood to the team and
replenishes the junior professionals. Be sure to support them by
having suitable mentors made freely available for them.
169
R S FRANCIS and G D COOPER
People-planning is critical; failure to do so could mean large
amounts of unnecessary and urgent extra work for you.
The most effective team must be prepared to believe in you
and follow your lead. Make sure that you support them at every
turn in public and, if you have to counsel them, do it in private.
Try to make the job fun and have a sense of humour – it helps.
Master tip
If you have a reasonably solid team, the answer to the following
two questions should be in the affirmative:
• Can you go on vacation and leave things in the hands of a
competent 2IC knowing that everything will be in a
reasonable shape upon your return?
• If a person in a key role, such as a foreman, decides to
suddenly resign, do you have personnel ready and competent
to step up to that position?
Staff retention
It is a challenge to do so, but it is critical to build up a competent
team. What is more difficult is keeping the team together for as
long as possible and replacing those that leave with equally
competent team members.
Numerous employee turnover surveys (Brereton, Beach and
Cliff, 2003) have been conducted in Australian mines over the
past few years and the results indicate a wide range of turnover
from five to 30 per cent, with some sites experiencing rates of
over 60 per cent. In general, because of the various career
development opportunities that arise, it can be expected that
most metallurgical professionals will stay in a position for
approximately two to four years. The surveys also indicated that
the turnover was less for mill and maintenance personnel. In
general, expect a retention time of between three and ten years.
Support the staff by providing good tangible and intangible
working conditions. Ensure that the basic tangible conditions are
covered, such as fair pay, appropriate crib room and changing
facilities, but don’t go overboard by providing everything that is
asked for by the operating crews – it will only lead to difficulties
later on as they will have become spoilt. Ensure that your staff
have the tools, support and information to do a good job
(Dunn-Lampe, 2005). Increasingly important are the intangible
benefits (Beams, 2005), including areas such as planned career
development, supervisor support and flexible work practices,
which are often more important than simply the tangible issues.
In a survey conducted by The AusIMM on its members in late
2005, the skills shortage was seen as the single most important
issue currently facing the mining industry (Anon, 2006). The
second most important issue was managing staff turnover. All
this makes a greater challenge to maintain a knowledgeable,
skilled and experienced workforce.
Master tip
There are a number of reasons why people leave and it is not
usually simply about money, but more due to a lack of
challenging work and a lack of supervisor support. Remember
that people join companies and leave bosses.
Provide regular performance reviews/feedback
Different companies have different standards in this area, and it
is critical that you nevertheless take time out for ad hoc
one-on-one chats with all your staff. Get to know something
interesting about each person, such as whether they are into
horse racing, yachting, pigeon racing, etc and then brush up a
little on the subject so that you can continue the conversation at
another time.
170 Fremantle, WA, 19 - 21 March 2007
Allocate a set time for discussion with your subordinate and
don’t change this time for other meetings. If you do change the
time, it sends the message that you give low priority to this
person. Whilst in discussion, focus your attention on that person
for the agreed period and don’t interrupt it to answer the phone
or receive other visitors.
Be constructive and honest with feedback during performance
reviews. If you are required to give the person bad news, give
them good news first, then bad news and close on a positive
note. It won’t be an effective performance review if you open
proceedings negatively. Above all, remain positive – most
situations are retrievable.
Try to get the employee to tell you what type of roles and
tasks they want to do. Use your influence to make it happen, put
their name forward for key positions for permanent roles, task
forces, projects or studies. Actively support them to achieve
their goals.
Good managers maintain an ‘open door’ philosophy – it allows
free communications at all levels. Those who do not, are
considered unapproachable by their staff.
Master tip
Take time to ‘walk the talk’, find out what your subordinates are
talking about. You may have said or done something that has
been misinterpreted, causing concern amongst the workforce.
Be aware of the big picture
Early on in a new position, learn as much as possible about the
company you work for. Be able to identify the key players
(usually on the company web page) and know and understand
your company’s strategic direction and corporate vision. Be sure
to attend all corporate visitor presentations. Feed this information
onto all of your people – most are usually interested. Openly
display interest in and support the big picture with those who are
in your department, and don’t criticise the company – you are
seen as being its representative.
The upside of knowing about the big picture corporate
direction is that it should be a guide for you in your own
direction and that of your department and sets the scene for the
type of people you need to employ, budget inputs and capital
requests.
Metallurgists usually come into contact with many other
professionals on site and in the corporate offices. It is not quite
the case with most other professionals, eg accountants don’t
usually deal with electrical or mechanical engineers and it is rare
for mining engineers to work with instrument and control
engineers. This means that you should have a reasonable
understanding of what the others do, so you have a head start on
viewing the big picture!
Practice at becoming an active listener to your subordinates,
peers and superiors.
Be aware of the overall cost breakdown for the site, and in
particular have a sound understanding of the costs in your area.
It is almost guaranteed that some of your subordinates will be
brighter and more inventive than you, as may be some of your
peers, and perhaps even your boss! Your credibility becomes
tainted when your manager comes up with an idea you refuse to
use and in the end they were proved to be correct.
Master tip
Take care not to get bogged down by the technical issues. It may
seem like fun and it is after all what you know best and feel most
comfortable with. However, you are a manager and your job is to
lead your team and influence others.
Ninth Mill Operators’ Conference
 MANAGING MINERAL PROCESSING PLANTS IN AUSTRALIA — TIPS AND PITFALLS
Acting department manager roles
There are several situations where a reliable and capable 2IC will
be acting in a role of the department manager. These include:
• Whilst on R and R at a FIFO operation. It is common for a
senior metallurgist to act as department manager, and an
individual may end up performing this role for significant
periods of time. The person in the acting role has all the
responsibility but no permanent authority. For those in such a
position, use this as an opportunity to practise the role. Leave
critical decisions for when the manager returns, ring the
manager or, better still, resolve with the site manager. It is
best to establish how to do this between the manager and 2IC
early on in the proceedings.
• Whilst you are on leave or off-site for business reasons.
These are usually one-off occasions and it is advisable to
have a competent 2IC to manage the department in your
absence. Give written guidelines to them on what issues you
want to be contacted about and how you want to be
contacted.
• Acting in the role for several months. Sometimes, the 2IC
could be in the role for extended periods whilst a
replacement is being sought for the head of department. This
is a tough situation as there are all the responsibilities, but
there is usually limited authority. Decisions cannot be
deferred and problems must be tackled. At the beginning of
this period, it is suggested that the 2IC has an open
discussion with the site manager to establish the expectations
of the position, including future handover to the new
incumbent. It is advisable that suitable mentors or advisors,
either inside or outside the organisation, are available to
provide advice on technical and personnel matters.
Master tip
Ensure you have a trustworthy 2IC. There have been times when
the person acting in your position has a closer relationship with
the site manager than you do. In this situation damage may be
done to your credibility by a ‘white-anting’ 2IC who undermines
your authority without you being there to defend yourself.
Managing upwards
You will experience a wide range of different styles of site
managers, but most will be mining engineers. Broadly, there are
three types of managers:
• Those that never visit the plant and leave you alone totally.
• Those who constantly visit the mill, talk to operators and
fitters and think they are metallurgists. These can make your
life a misery.
• Those that have confidence in their staff and mostly leave
you alone, but want to be informed immediately when
something significant happens.
Spend time with your manager as early as possible in your
appointment to get to know what type you have; what level of
problem they want to know about; and how often they want to
see you. Don’t pop in to their office three times a day if they are
happy with a once a week catch up. Ensure they find out
problems from you (and not from some other source), and what
type of communication method they prefer, be it phone,
face-to-face, email or written memorandum). On a FIFO site you
will probably have more than one manager, so you have to put
time and effort into understanding how each one likes to manage
and be communicated with.
Sometimes you will have the privilege yourself to act as the
site manager, which can be tough if things are not going well in
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
your plant. You are filling the shoes of your superior and you
have to maintain the status quo. It will be career limiting if you
decide to make wholesale changes and you probably won’t be
asked to act as site manager again. Be briefed beforehand on the
site manager’s expectations and make sure you fulfil them or
have watertight reasons if you do not.
You will be regularly required to provide an explanation of the
plant’s performance. You need to get all your facts straight
without stretching the truth or blaming other departments. Be
sure to check out the facts and don’t be caught by the site
manager during one of their walk-arounds. Be careful of all the
excuses that you come up with for poor performance – they are
usually remembered, and you will have difficulties if you cannot
verify them.
Master tip
Don’t ever upset the site manager’s secretary or personal
assistant – they can be an extremely helpful person during key
times.
Statutory and other responsibilities
As the most senior metallurgist on site you will need to have a
clear understanding of your responsibilities and what can get you
into serious trouble – this goes beyond the duty of care. Ensure
that early in your tenure you find out and take appropriate action
regarding the following:
• Know what inspections and checks have to be performed.
Always keep these up to date. Be seen doing the inspections
yourself and involve key people. Do as much as possible
yourself, but maintain a balance and don’t delegate it all.
• Keep up to date with changes in relevant legislation, in
particular the Mines Act and Regulations. Be aware of what
is in the legislation, because site managers are usually
conversant with them.
• Know what can get you terminated, what can put you in
court or, worse still, what can put you in gaol.
• Always have a clear, simple and logical understanding of
why the plant is not performing to expectations and be
able to explain concisely to others who are not trained
metallurgists. This is especially important for presentations
and discussions with executive management.
All operations have been designed and developed to generate a
profit. Ensure that your costs are under control and that you
clearly understand what the cost drivers are and how they are
influenced. If you have a cost over-run in a particular area, be
sure to get it under control as quickly as possible.
One area where you have primary responsibility is for the
metallurgical accounting. It is critical to ensure that your
metallurgical accounting is maintained to an auditable standard.
Failure to do so can lead to serious difficulties, particularly if
there is inventory missing and unaccounted for. This can lead to
long explanations of where the company’s assets are and whether
you are in control or not.
There was an example where 900 ounces of gold could not be
accounted for at the end of a month’s operation despite numerous
checks and measurements. Embarrassingly, a month later the
material was found in the circuit. That system was out of control
and it took almost a year to fine tune a robust and accurate
system to prevent those kinds of issues occurring again. A key
measure of business unit performance is cash flow and ‘missing’
inventory can have a major adverse effect on this measure. Spend
time setting up a system that is fail-safe, using auditable
accounting methods to check for any discrepancies.
171
R S FRANCIS and G D COOPER
Often you will be called upon to conduct a tour of your plant
with visitors ranging from analysts to mine inspectors. It is
worthwhile making sure that housekeeping is to an acceptable
standard. Your plant is above ground, it is easily seen by all –
keep it in a reasonably good state of cleanliness, but don’t go
overboard. Keeping the plant clean will also assist in the
operational crews having pride in their workplace.
Setting an example – pride and passion
There are a number of areas where you should set an example as
people will be constantly watching your behaviour. Some people
will watch for weaknesses and seek to exploit them.
• Ensure that you set a high visible personal standard in safety.
If you see something being done unsafely and don’t do
something about it you will be seen as condoning the action
and setting an unhealthy precedent for an unsafe operating
practice. Ensure that you enthusiastically support any
company safety initiatives.
• Maintain a reasonably appropriate dress standard and be
neatly groomed. Always wear appropriate safety gear – never
be caught out being below the accepted standard. It will be
impossible to address breaches if you don’t follow company
guidelines.
• Be aware of pedestal behaviour, where junior employees
consider you as a god and hang off every word you say. You
are just the same as everyone else, except that you have
different responsibilities. In these situations you may fall into
the trap where a thoughtless ‘off the cuff’ comment can be
misquoted as being fact.
• Do not become involved in spreading rumours or by the
inappropriate use of email and internet – it could become
awkward in counselling individuals.
• Do not speak ill of any company employees, especially your
seniors and peers – you may think it, but don’t say it. Treat
others the way you would want them to treat you.
• Show impartiality with your team. Don’t have favourites.
Remain somewhat aloof. It will be easier to delegate tasks
and counsel employees by doing so.
• Be a mentor to others around you. Provide mentors from
other areas for those in your department, and don’t forget to
have a mentor for yourself. Ideally, a mentor should not be
your immediate superior.
• Be punctual for appointments, particularly if you are chairing
the meeting or where there are many company senior
managers in attendance.
• Get dirty if needed, particularly when there is a larger than
normal spillage and ‘all hands on deck’ are required. It will
be appreciated if you perform this task every now and again
to help out, but don’t make it a habit.
• Start and finish work at reasonable hours, as people will
notice and make comments if you are consistently late for
work or leave consistently early. They don’t know or care
how many other hours you are working.
• Be prepared to work occasionally after hours if necessary,
such as on weekends or in the evenings. Take a stroll through
the plant and talk to some of the operating and maintenance
crew. They will appreciate you taking the time to do so.
• You are responsible for the costs, equipment and people in
your plant. Demonstrate that responsibility and when
appropriate, admit fault – do not abdicate the responsibility
onto someone else.
• Be prepared to show your plant to dignitaries, directors,
analysts and other visitors. Be proud of your plant, show
passion in the work that you are involved in and be able to
172 Fremantle, WA, 19 - 21 March 2007
explain the process in simple terms. Non-metallurgists think
that metallurgy is a black art – don’t let them leave site still
believing that.
• You need to act appropriately at official and social company
functions outside of work hours. Inappropriate behaviour can
lead to bad publicity which could undermine your authority
at work.
• Do not abuse the privilege of free lunches or dinners,
company cars, phones, or credit cards. People watch your
every move and in most companies there will be a code of
conduct issue.
• Senior managers hate surprises. Be sure to communicate bad
news at an early stage and keep them informed on the
progress.
Master tip
Never let alcohol affect your performance or behaviour. Never
arrive at work with a hangover and definitely not with a blood
alcohol level above the company’s limit.
There will be times when it becomes all too much and you
need to take stock of your situation as to what is causing the
stress. Don’t sweat the small stuff – it is only a job. However, it is
a reasonably high pressure job and if you are not having fun, get
out and try something different.
Run effective meetings
There has been much written and said about running and
managing meetings. They are an important part of the job and
they take up a lot of busy people’s time. Some key guidelines to
help make them more productive include:
• Ensure that the meetings in your department have agendas –
don’t waste people’s time, or they won’t come to the next
meeting.
• Ensure action items have names and due dates assigned to
deal with them. Follow up to ensure that the items have been
actioned.
• Production meetings can be tedious and boring, especially
for non-metallurgists when going into specific details on
metallurgical issues. Make the meetings ‘to the point’.
• Do not attend or have meetings for meeting sake (just
because we’ve always had one is not a good enough reason)
– challenge the status quo.
• A good meeting is a quick meeting – don’t let them drag on
for hours.
Master tip
Every two to three months organise a meeting with the shift
supervisors. The logistics can be complicated, but getting these
key production leaders in the same room for a few hours without
production pressures is invaluable. From the shift supervisor’s
point of view, they hardly ever have time to get to know their
counterparts on the other shifts. Getting them together away from
normal production or operating pressures can bring a team
together to work cohesively. It also allows the supervisors to
offload a number of issues which don’t often get discussed. At
this time there is also an opportunity to get some key people such
as the site manager or the senior geologist to talk with them in a
workshop environment.
Challenge the status quo
A good plant manager will constantly challenge others in their
area on the status quo. You will often hear: ‘we’ve always done it
that way! Why should it change?’ Many people like to be in their
Ninth Mill Operators’ Conference
 MANAGING MINERAL PROCESSING PLANTS IN AUSTRALIA — TIPS AND PITFALLS
comfort zone and don’t want to make changes or ‘rock the boat’.
Unfortunately, life does not remain the same; there is always
continual change and room for continuous improvement.
This can be illustrated by the following exercise. At a
workshop attended by one of the authors several years ago, each
group had to compete to build a tower out of paper and sellotape
to suspend a fifty cent piece as quickly as possible. Fortunately,
one member of the group had partaken in a similar exercise and
quickly designed and built the required structure. As our team
came first, we shared the design and construction characteristics,
before repeating the competition. Again we came first, but if we
had built it in the same time as in the first round we would have
been in eighth place. As the most senior metallurgical
professional on site, it is your duty to seek out new ways of
improving performance and getting the most out of the ore, the
equipment and the people.
Be prepared to make reasonable experimental changes to the
plant, and if it does not go according to plan, admit it and correct
it. Do not make wholesale changes from which you cannot
recover.
If you want to implement a process change, be sure to actively
engage the operating personnel, particularly the shift supervisors,
to see how they would go about making that change. Listen
carefully to their answer as they will probably provide the most
effective method. More importantly, as they gain ownership, they
will do their utmost to ensure that the process change does not
fail. Ensure you have your shift supervisors’ support when
making major changes. If they do not support you, there will be
problems with the operating crews in accepting them.
At a gold operation several years ago, there was a requirement
to conduct a trial to retreat some tailings material over a several
week period. Presenting the crews with this challenge and asking
for the most effective approach resulted in unexpected
enthusiastic support which provided a more than successful trial.
Do not make changes for changes sake. Sometimes you will be
pressured by your site manager to make some quick
extraordinary changes. Think through carefully before you act to
minimise the negative impact. Be alert for the controlling site
manager who thinks they are a plant manager as they sometimes
like to make wholesale changes in your absence.
Identify people that support change and work with them.
Involve these people at the early stages. If they have buy-in they
will help to make the changes and incremental improvements
will occur.
Master tip
Do not make a whole series of process changes to conditions in
the plant late on Friday afternoon or just before you fly out and
then leave the crews with no one to contact for the weekend or R
and R break. This leaves the crews with little or no support and
could undermine your authority and respect.
Working with other departments
You are in a unique position where your section is on the
receiving end of the ore to be processed and the plant has to turn
out a product that usually has to meet tight specifications. This
can sometimes be seen as the ‘raw end of the pineapple’,
particularly if there is a poor rapport between the mineral
processing, geology and mining engineering departments. On the
contrary, you are in a most powerful situation, where, if you
know the ore and how it behaves, you will end up with a great
deal of influence and control. You need a supportive site manager
for this to occur.
Unfortunately, it is rare for the mining and geological
professionals to take more than a passing interest in what takes
place in a processing plant. Do not focus on the details of the
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
plant operation only – you need to be aware of the big picture.
You can do nothing about the product specifications as they are
controlled by sales contracts. However, it is your responsibility to
produce the product within these specifications to ensure that it is
saleable. You won’t get much mileage from complaining about
the ore that you are fed, particularly if you have nothing to back
it up with and given that most site managers are mining
engineers, the mining department will almost always get the
sympathetic ear. You need to be in a position where you can
influence your geology and mining colleagues so that the
treatment plant receives ores that are of acceptable quality and of
a consistent nature. Mine to mill is a good start, but the big
picture is really mine to market. Be sure that the material you are
given is capable of making a product that is always within
specifications. Take time to understand the mineralogy in both
the valuable product and the gangue.
Depending on the nature of the organisational structure, the
maintenance function can be part of the plant manager’s
responsibility. This can be quite daunting to the newer plant
managers, as you sort out the balance between production and
maintenance. In a complicated processing operation the
maintenance section will probably report directly to the site
manager. This means that the plant manager has to deal
effectively with the maintenance manager in order to fulfil both
the production and maintenance requirements.
Master tip
Spend time gaining a better understanding of what is happening
in the mining and geology departments. Make regular trips to the
pit or underground and have a clear understanding where the ore
is and will be coming from. Bring these departments closer to the
processing plant by showing them what makes it work better and
show them the impact of having to treat inferior quality ores.
The future – Generations X and Y
Future trends will be different from what has been experienced in
the past. Many professionals now have double degrees, usually
with an affinity for the commercial side. Research has indicated
(Stuart, 2005) that few of the younger people are prepared to
consider working in the minerals industry in either a professional
or trade role and many have no desire to work in regional or
remote areas of Australia. It disturbingly indicates that they will
most likely not stay as site metallurgists for too long and will
want to relocate to the city to pursue other interests.
Traits of Generation Y (born between 1980 and 1994) are said
to be (Kotler, 2003; Stuart, 2005):
• technological savvy,
• high expectations of the future,
• an obsession with achievement,
• emphasis on immediate gratification,
• greater acceptance of racial and ethnical differences,
• are at school now or have left in the last ten years,
• network and communicate madly,
• gain knowledge on the need-to-know basis, and
• are concerned about the environment (but few are studying
the earth sciences).
Younger generations have a tendency to focus on lifestyle
issues rather than career development. This could make it
difficult to recruit potential employees in outback residential or
FIFO roles as many of the younger generation are not keen to
move away from the cities and FIFO arrangements may have a
limited shelf life. This shortage of the younger generation
combined with older workers leaving the workforce to retire
further exacerbates the skills shortage.
173
R S FRANCIS and G D COOPER
In the past, loyalty played a big part in maintaining a stable
workforce. Younger employees are now looking towards the
short term and are not necessarily looking for a long career with
a single company or corporation.
Lifestyle issues may become the primary focus where life
balance is more important than bank balance. Pay is not
necessarily the prime motivator, but lifestyle is, and that does not
include long hours, stress, health concerns and broken
relationships. The generation X (born between 1965 and 1979)
group puts the quality of personal life ahead of work life (Kotler,
2003).
Our thinking has to change and accommodate the younger
generation’s ways of doing things. For the Generation Y
employees, the trend is away from the ‘one size fits all’
approach. Open communications and a healthy employer-
employee relation is fundamental to the attraction and retention
of employees (Beams, 2005).
A major challenge will be how to successfully pass on the
skills and expertise that the older generations have gained over
many years of sometimes bitter experience to the younger
generations.
CONCLUSION
There are three important things to look after in the position as
the plant manager. They are: the people, the plant and the ore. If
you look after the people, the other two will be looked after. The
position of plant manager is one where you are the Jack and
Master of all trades – you have to be capable and competent in
many areas.
174 Fremantle, WA, 19 - 21 March 2007
ACKNOWLEDGEMENTS
The authors wish to thank their respective employers for
allowing the paper to be published, and for the input that was
received from wives and colleagues, in particular from Allan
Quadrio.
REFERENCES
Anon, 2006. Survey of minerals industry professionals – Executive
summary, The AusIMM Bulletin, 2:8-9.
Beams, C, 2005. Workforce risk management and the looming labour
shortage, The AusIMM Bulletin, 5:50-51.
Brereton, D, Beach, R and Cliff, D, 2003. Employment turnover as a
sustainability issue (revised version), in Proceedings 2003
Sustainable Development Conference, pp 1-18 (Minerals Council of
Australia: Brisbane).
Dunn-Lampe, M, 2005. Engaging staff in a disconnected world, Chifley
Business School GRAD Newsletter, December-January(4).
Francis, R and Hille, S, 2002. Setting best practice: FIFO at the Mt Keith
concentrator, in Proceedings Metallurgical Plant Design and
Operating Practices, pp 129-141 (The Australasian Institute of
Mining and Metallurgy: Melbourne).
Kotler, P, 2003. Marketing Management, 11th edition, p 291 (Prentice
Hall: New Jersey).
McCrindle, M, 2003. The ABC of XYZ: Generational diversity at work
[online], McCrindle Research Pty Ltd. Available from:
<http://www.quayappointments.com.au/email/040213/images/genera
tional_diversity_at_work.pdf > [Accessed 20 June 2006].
Stuart, D, 2005. Engaging Generation Y with the minerals industry, The
AusIMM Bulletin, 4:69-72.
Ninth Mill Operators’ Conference
Process Design Aspects at Bendigo Mining’s Gold Operations
E J McLean1, T E Hayward2, D H English3 and S J Smith4

ABSTRACT • reefs occur as groups or ‘clusters’ in cross-section suggesting

Gold mineralisation at Bendigo is exclusively associated with quartz
reefs, commonly found within or near anticlinal fold hinges. The gold is
typically coarse grained (0.5 mm to 5 mm), free milling and displays a
complex, nuggetty distribution within the veins.
This ore is highly amenable to gravity recovery processes. As the
Bendigo operation is not permitted to treat whole ore by a cyanide
leach/CIL circuit, the flow sheet is more complex than conventional gold
recovery circuits. Primary gold recovery is from a gravity circuit closing
the grinding circuit to a bulk gravity concentrate and from a flotation
circuit treating cyclone overflow to a bulk float concentrate. The plant
capacity is relatively low by current gold industry standards at 600 000 tpa.
Re-establishing gold production at Bendigo has presented several
challenging, and sometimes competing, issues in process design and
operation. The mine, process plant, associated services and infrastructure
are at Kangaroo Flat, a low density, semi-rural zone in the southern
urbanisation of the Bendigo metropolitan area.
One of the main processing challenges during the design was the
competing comminution requirement to avoid over-grinding, which
would flatten or smear the characteristic nugget gold and hinder effective
gravity gold recovery, whilst also providing sufficiently well liberated
particles that would respond well to flotation.
Important, non-metallurgical challenges were to minimise impacts on
the local community from noise, dust, air emissions, visual impact, water
egress and off-site traffic.
an interrelated structural control;
• reef clusters form subhorizontal ribbons when viewed in
three dimensions; and
• mineralised ribbons repeat at regular intervals at depth below
the historical workings.
The gold mineralisation in the system is exclusively associated
with quartz veining and faulting, commonly found within or near
anticlinal fold hinges. The gold is typically coarse to very coarse
grained (0.5 mm to 5 mm) and displays a complex, nuggety
distribution within the veins.
Within an envelope some 16 km long by 4 km wide, 15 major
lines of reef (anticlines) accounted for the bulk of the historical
hard rock gold production on the Bendigo goldfield. An inferred
resource of 11 million ounces of gold has been identified by
BML beneath the historic workings, to a depth of 1500 m, on
five of the 15 major lines of reef. This resource forms the basis of
the Bendigo Gold Project.
An aerial view of the project showing the main locations in the
Bendigo area is shown in Figure 1.

INTRODUCTION
Gold was discovered at Bendigo, about 200 km north of
Melbourne, in 1851. In a little over 100 years since then,
approximately 22 million ounces of gold were produced from the
Bendigo goldfields. About 18 million ounces reportedly came
from mining of 40 million tonnes of narrow vein quartz reef ore
from underground mining at an average depth of 700 m. In
places, mining extended to a depth of around 1400 m.
Bendigo Mining Limited (BML) has been active in the
Bendigo area since the mid 1980s, carrying out exploration,
developing the resource and consolidating exploration leases. By
late 2000 BML amalgamated the separate ELs into a single
tenement, EL3327, to give the company control of the entire
Bendigo goldfield and some 300 km2 of surrounding countryside.
Detailed studies by BML on the geology and past production,
confirmed by recent exploration, describes the gold
mineralisation at Bendigo by the ribbon geological model. The
model indicates that:
• gold bearing quartz reefs occur in close proximity to the
anticline axes;
• reefs have one long dimension, which is subparallel to the
anticline crest (ie subhorizontal);

1. MAusIMM, Manager Process Queensland Operations, Ausenco

Limited, 2404 Logan Road, Eight Mile Plains Qld 4113.
Email: eddie.mclean@ausenco.com
2. Senior Metallurgist, Ausenco Limited, 2404 Logan Road, Eight Mile
Plains Qld 4113. Email: thomas.hayward@ausenco.com
3. Project Metallurgist, Bendigo Mining Limited, PO Box 2113,
Bendigo Mail Centre Vic 3554.
Email: denglish@bendigomining.com.au
4. MAusIMM, Principal Metallurgist – Director, Aurifex Pty Ltd,
25 Barrett Street, Wembley WA 6014.
Email: stuart.smith@aurifex.com.au FIG 1 - Aerial photograph with project features.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 175

E J McLEAN et al
The Swan decline is located at the Kangaroo Flat portal site,
the southern, and shallowest known part of the reef system,
to access the orebody for exploration, sampling and mine
development purposes. This intersects the Deborah and
Sheepshead lines of reef, which will be mined and treated in the
initial phase.
Permitting and processing
In accordance with the requirements of the 1998 Environmental
Effects Statement (EES) and the 2004 Supplementary
Environmental Effects Statement (SEES), surface storage of the
tailings from the processing facility at the Carshalton site are
permitted as follows:
• Gravity tailings produced from gravity processing and which
have no chemical treatment.
• Flotation tailings produced from flotation and which have
had contact with flotation chemicals.
• Disposal of separate cycloned overflow and underflow
products from flotation tailings.
• Gravity/flotation tailings stored in purpose-designed facility
for possible future recovery and reuse as by-products.
• Tailings from float concentrates which have been treated
with flotation chemicals and the residue from these
concentrates which have been treated with cyanide. Tailings
from the leaching of float concentrates will be detoxified
prior to storage. The weak acid dissociable (WAD) cyanide
level will be less than 50 mg/L, in accordance with the
International Cyanide Management Code, at the point of
discharge into the facility.
• Float concentrate leach residue stored in permanent (lined)
facility. Float concentrate, and the resultant leach residue,
represents a small fraction of the total plant tailings, thereby
reducing chemical use and storage.
• Leachate, if generated, will not be released to the
groundwater.
Consequently, a cyanide leach and carbon-in-pulp or
Merrill-Crowe type operation on whole ore treatment were not
applicable process routes. Preliminary metallurgical testing done
on gravity and whole ore leaching showed comparable recoveries
to gravity/float/intense leach of float concentrate. The latter
process route was considered to offer environmental advantages.
Concentrate processing at the Carshalton site, which includes
cyanide leaching (of gravity and float concentrates),
electrowinning, smelting and leach residue detoxification was
approved.
Project key stages
The Feasibility Study, managed by AMC Consultants Pty Ltd,
was completed in 1Q04. BML awarded an EPCM contract to
Ausenco Services (Ausenco) in 3Q04 for the design and
construction of a 300 000 tpa process plant, with allowances for
future expansion to 600 000 tpa, and associated services to treat
the underground ore. Following an interim engineering and cost
review in 1Q05, BML increased the treatment plant capacity to
600 000 tpa. Mobilisation to site occurred in May 2005 and first
gold was poured in July 2006.
PROCESS CHARACTERISTICS OF
BENDIGO ORES
Metallurgical testing
Extensive metallurgical testing for ore characterisation and
variability assessment was carried out at bench scale on more
than 90 bulk samples, and by pilot testing at the New Moon
176 Fremantle, WA, 19 - 21 March 2007
Processing Plant (NMPP) on about 35 000 t of development ore.
The test work included the following major activities:
• comminution testing and physical characterisation of ore
hardness and competency,
• extensive gravity testing,
• flotation response,
• recovery trends with grind size,
• diagnostic size-by-size analyses on residues,
• leach kinetics on flotation concentrates,
• settling and thickening characterisation,
• cyanide detoxification, and
• tailings dewatering.
The majority of the test work for the Feasibility Study was
carried out between 2001 and 2003. A further series of test work
was carried out in 2004 to provide additional information for
flow sheet optimisation and process design purposes.
Key outcomes and conclusions from the metallurgical test
work and pilot plant trials were:
• Pilot plant trials treated ore with high head grades, 8 -
14 g/t Au, consistent with typical feed grades expected from
the mine during the initial years of operation.
• Bulk samples with grades from 3 - 20 g/t Au provided an
assessment of gravity recovery for the range of head grades
likely from the mine.
• The ore is considered moderately hard from work index
testing on five samples. These produced a range of values of
13 to 20 kWh/t for the rod mill work indices and 13 to
17 kWh/t for the ball mill work indices.
• The same samples showed a range of abrasion indices from
0.20 to 0.45.
• Liberation of much of the free (nugget) gold occurs at coarse
sizes around 1 to 2 mm.
• Gold and particle size distribution analyses show that metal
value in fractions coarser than 1 to 2 mm can be variable, but
is typically 50 per cent of the total value in the head. This is a
higher proportion of metal value in these coarser sizes
compared to most other mines in Australia (Johansen, 2005).
• Gravity gold recovery from bulk samples is variable,
particularly in lower grades, and generally in the range of 60
to 90 per cent for the higher grades tested. A summary of the
results of recovery by grade is shown in Figure 2. The
proportion of feed to gravity concentrates, corresponding to
these recoveries, is generally 0.3 to 0.6 per cent by weight.
• Gravity gold recovery from the pilot trials on parcels of ore
from higher grade reef areas are shown in Table 1.
TABLE 1
Gravity recovery, New Moon Processing Plant.
Stream Description S3-101 S3-Backs Garrards
Head Au g/t (calc) 9.50 8.71 14.0
Au g/t (assay) (9.74) (8.05) (14.2)
Conc Recovery % Au 77.6 74.9 88.2
Conc Weight % 0.5 0.8 1.8
• Recoveries are consistent with those achieved for the bulk
samples, but generally with a higher weight proportion from
feed to gravity concentrates.
• Flotation gold recovery from the pilot trials treating screened
gravity tails (minus 300 µm) gave gold recoveries (from float
feed) in the range 76 to 89 per cent.
Ninth Mill Operators’ Conference

100
90
80
70
Recovery, % Au 60
50
40
30
20
10
0
3.0 4.0 5.0
The Bendigo ores tested are notable for their relatively high
grades, excellent gravity gold recovery response, extremely coarse
liberation size (by mining industry standards) and good response
to flotation. The ores are relatively ‘clean’ metallurgically with
minor amounts of arsenic (from arsenopyrite) as the only
significant deleterious element in the concentrate.
Flow sheet development
As evident from the mineralisation characteristics and the testing
program, the main design consideration for the process plant is to
maximise gold recovery in the gravity circuit in a cost-effective
manner. Maximising gravity gold recovery is expected to
maximise overall gold recovery as coarse, well liberated gold
that is not recovered by the gravity recovery system goes to the
flotation circuit and is likely to be lost to flotation tailings. A
further consideration for overall gold recovery is to ensure
adequate liberation of gold is achieved as unliberated particles of
gold or pyrite/gold may also be lost to flotation tailings.
To maximise gold recovery, the flow sheet was developed with
the following processing strategies:
• fine crushing,
• maximum deportment of crushed ore to bulk gravity (coarse)
gold recovery prior to ball milling,
• upgrading of bulk gravity concentrate to a product suitable
for smelting,
• intensive leaching of tail from gravity concentrate upgrade,
• grinding and liberation of non-gravity recoverable pyrite/
gold for subsequent flotation,
• flotation of a pyrite/gold concentrate,
• carbon-in-leach recovery of the gold from the flotation
concentrate, and
• conventional gold recovery system from loaded carbon.
DESIGN CONSIDERATIONS FOR THE
PROCESS PLANT
Comminution circuit selection
Analysis of the test work indicated that gravity gold recovery is
strongly dependent on the proportion of -4 mm product that
could be generated by a crushing circuit.
Ninth Mill Operators’ Conference
PROCESS DESIGN ASPECTS AT BENDIGO MINING’S GOLD OPERATIONS
6.0 7.0 8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0 16.0 17.0 18.0
Head grade, g/t Au
FIG 2 - Bulk samples gravity recovery.
Three and four stage crushing
A flow sheet based on conventional three-stage crushing to
nominally -10 to 12 mm would not achieve the desired product
size. A further concern was that optimum plant metallurgical
performance in the ball mill and gravity recovery circuit would
not be achieved due to flattening, over-grinding or smearing of
gold particles in the ball mill. This results in lock-up of gold in
the milling circuit and renders gravity and classification
processing less efficient as the flattened particles present a large
surface area to flowing water films.
A four-stage crushing circuit would produce a finer product
closer to the crush size required, but with an increasingly
complex circuit that required a large footprint within the limited
area available for the processing plant in the site as permitted.
Circuits of this type treating high silica ores typically have high
operating costs and increased maintenance demands to maintain
operating availability. The finer crush size may reduce, but not
fully alleviate the flattening and smearing issues in the ball
mill.
Two stage crush and HPGR
A high pressure grinding rolls (HPGR) system was considered
for the tertiary comminution stage following a conventional two-
stage crushing circuit. Batch testing of about 1300 kg of ore was
carried out at a manufacturer’s laboratory. Unit performance
parameters on power and wear rates were obtained. The HPGR
manufacturer testing showed an approximate ten to 15 per cent
increase in the proportion of -2 mm product (‘natural’ nugget
size) with the HPGR in open circuit, and even higher with the
HPGR in closed circuit when compared to the product
distribution predicted from tertiary crushing.
As plant testing with the HPGR during the plant trials at
NMPP was not able to be carried out, a series of comparative
option studies were carried out to assess the potential benefit of
increased recovery using a HPGR with various crushing
arrangements. With the HPGR in the circuit, a simple payback in
less than three years required an increase in overall gold recovery
of one per cent (at assumed gold price US$400/oz).
The decision to proceed with the HPGR in circuit was made
on the basis of technical, operating, capital cost, operating cost,
lead time, manufacturer support and performance warranty
considerations (Johansen et al, 2005).
Fremantle, WA, 19 - 21 March 2007 177
E J McLEAN et al

Grinding mill selection TABLE 2

A ball mill and a rod mill were considered for the grinding NMPP gravity tailing gold distribution.
duty. A ball mill was selected for this circuit on cost, operational
and environmental considerations. The main issues against use of Gravity tail Units S3-101 S3-Backs Garrards
a rod mill were: Grade tail Au g/t (calc) 2.14 2.20 1.68
• indicative capital cost of rod mill is at least 25 per cent more Au g/t (assay) (2.43) (1.65) (1.86)
expensive than ball mill for the same power; Tail size P80 µm 365 350 280
• lower operating availability expected due to more frequent Dist’n o’all % Au in tail 22 25 12
planned stoppages to load rods to mill; +600 µm % Au in size 8 9 17
• increased noise is generated from steel rod/rod and steel -600 +106 µm % Au in size 21 21 17
rod/shell impacts in rod mill; -106 µm % Au in size 71 70 66
• smearing of gold onto steel liners and rods would be Head Total % Au 100 100 100
considerably higher than in rubber lined ball mill; and
• the ball mill produces a finer size distribution than rod mills, was not included as it introduces another process stage, adds cost
particularly for particles less than about 0.1 mm where most
and complexity to the circuit, and places additional demand on
of the liberated sulfides (for flotation) are expected to occur.
the plant footprint.
The distribution of grades by size in gravity tails (see Figure 3)
Coarse gold gravity recovery shows that grades in the +600 µm fractions are significantly
The main consideration for the gravity circuit was to provide an higher (by five to ten times) than grades in their neighbouring
opportunity for liberated particles in fine crushed ore to be size fractions. This indicates incomplete liberation and that
recovered ahead of grinding in the ball mill. This provides potential recoverable gold exists in these coarser fractions.
earliest ‘escape’ for the coarser gold particles and limits their
exposure to flattening and smearing in the grinding mill. 6.0
Strategies for gravity gold recovery include: 5.5
• Feed size to the gravity/grinding circuit is nominally 4 mm 5.0
4.5
by closing the HPGR circuit with a screen ahead of grinding.
Grade Au g/t

4.0
• Gold in gold trap, which recovers liberated coarse gold at the 3.5
feed box to the screen closing the HPGR circuit, and is piped 3.0
directly to the gold room. 2.5
• Primary gravity recovery in jigs which receive screened 2.0
undersize (new feed). These also receive ball mill discharge 1.5
as a circulating load. 1.0
0.5
• Primary jig concentrate is upgraded in centrifugal bowl 0.0
concentrators. This concentrate goes to the gold room for
-106+75

-425+300

-600+425

-850+600
-75

-150+106

-212+150

-300+212

-1000+850

+1000
further processing thus in two stages, amenable gold particles
can report to the gold room.
• Gold particles in jig tails are recovered by centrifugal bowl S3-101

concentrator. These scavenged gold particles then report to

the gold room.
• Fine particles in the gravity tailings that meet the closing
cyclone classification size report to cyclone overflow and go
to the thickener ahead of flotation.
• Internally, the oversize from the gravity scalping screen
returns to the ball mill for grinding.
• Cyclones close the grinding circuit with underflow returning
to the ball mill. Four opportunities are therefore provided for
gold recovery before the new feed goes to the grinding mill.
Liberation size and grinding
The grinding circuit at the NMPP comprised a ball mill and
primary gravity recovery by a jig in closed circuit with a sieve
bend screen. Screen undersize was treated in a secondary jig and
scavenger centrifugal bowl. Analyses of gravity tails from three
parcels of ore with similar head grades to the range of grades
expected in the plant are summarised in Table 2. The majority of
the gold in the gravity tails is in the finer size fractions that are
amenable to good float performance.
Although about 20 per cent of gold in the tails is in the -600
+106 µm fractions, grades in these fractions (0.5 - 0.8 g/t Au) are
similar to, or just above final float tail grade. Additional
pre-screening and rejection of a coarse fraction ahead of flotation
S3-BACKS Size (microns)
GARRARDS
FIG 3 - NMPP gravity tails gold grade by size.
Selection criteria for the grinding circuit product were:
• mill circuit product size of 100 per cent passing 600 µm;
corresponding P80 is 212 µm;
• grinding circuit closed with hydrocyclones; and
• avoid fine screening due to propensity of screen cloth
apertures to peg and blind from angular ground particles.
Design considerations for other process areas
Flotation circuit
Bench and pilot test work showed bulk recovery of sulfides and
free gold particles were achievable in a rougher flotation circuit,
to a concentrate typically two to three per cent by weight of float
feed. A key design consideration was the ability of the float cells
to handle coarse, or ‘sands’ type, material, similar to that in unit
float cell applications treating cyclone underflow in a grinding
circuit. Coarser particles sizes, greater than about 150 µm were
not targeted for flotation due to their low grade and poor float
characteristics.

178 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 PROCESS DESIGN ASPECTS AT BENDIGO MINING’S GOLD OPERATIONS
A sample cut from final float tailings, representing about three
per cent by weight, was provided to detect further potential
gravity recoverable by treating this bleed stream in a centrifugal
bowl concentrator.
Disposal of float tailings
Disposal of tailings and management of the tailings area
considered several objectives:
• able to recover a coarse sands product if, in the future, a
market was identified and specifications were confirmed;
• recover as much of the water (and residual flotation
chemicals) contained in the flotation tailings slurry directly
back to the process water circuit;
• reduce process water accumulation in the pond in the tailings
area; and
• enhance deposition characteristics of the slimes, which
would improve the settled density of these particles.
The float tailings are cycloned and the overflow thickened.
Thickened slimes from the cyclone overflow are pumped to
designated areas in the tailings storage facility. The cyclone
underflow, or potential sands product, is pumped to separate
areas in the tailings storage facility.
Carbon-in-leach (CIL) circuit
A moderate preg-robbing effect was identified in the test work.
Activated carbon is present in all tanks in the seven stage CIL
circuit.
Carbon stripping and regeneration
Gold is recovered from the loaded carbon in a conventional
pressure Zadra elution circuit. Stripped carbon is regenerated in
batches in a horizontal kiln. The regeneration system treats
100 per cent of the carbon inventory from stripping to remove
organic fouling from flotation reagents. This ensures carbon
activity is returned to ‘near-new’ condition suitable for return to
the last CIL tank. An after-burner treating the regeneration kiln
exhaust gases is included to remove odours from burning of
organics.
Disposal of CIL tailings
The SEES requirement was that CIL tailings are detoxified
before discharge into the storage facilities and that the cyanide
content is less than 50 mg/L CN-WAD at this point. BML set an
operating target of less than 20 mg/L CN-WAD from the
commencement of operations.
Bench scale, semi-continuous testing was carried out on
samples of float concentrate from NMPP to confirm the cyanide
destruction method selected and determine process parameters.
The copper catalysed, air/SO2 method was selected based on its
proven performance, robust operating system and cost-effective
plant. The detox test results were analysed and design criteria
selected to achieve sufficient residence time in the reactor,
adequate dosing capability for reagents, and supply of oxygen,
which would enable the BML target to be achieved.
Due to the large air demands and the requirement to place the
detoxification facility in a building, the system was designed to
utilise oxygen injection to replace the more typical use of low
pressure air for the reaction demands.
Scrubbing exhaust fumes from gold room
The final gravity concentrate, from the various upgrading stages
in the gold room, contains remnant heavy minerals comprising
mainly pyrite, but with varying amounts of minor arsenopyrite.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
The final concentrate, loaded into trays, is placed in the calcine
oven and dried overnight before smelting.
In the smelting furnace and, in some conditions in the calcine
oven, arsenopyrite can oxidise to form arsenic trioxide (arsenic
trioxide has a boiling point of 465°C but sublimes at 193°C).
Smelt furnace fumes along with fumes from the calcine oven flue
and electrowinning cells are ducted to the gold room scrubber.
The wet scrubber system uses a weak sodium hydroxide solution
circulated through the tower to remove contaminants in the gold
room area as well as from the sodium metabisulfite and lime
mixing tanks.
A test work program was carried out to provide information
on the amount of arsenic oxidation expected. A range of
arsenopyrite contents in concentrate, degrees of oxidation and
rates of oxidation were assessed to determine conservative
arsenic trioxide concentration criteria for scrubber design.
Further, initial calcination temperatures are limited to 250°C to
allow monitoring and evaluation of smelting methods and calcine
conditions.
Noise attenuation
Noise targets are a high profile community issue for this plant
and targets have been set for maximum allowable noise levels at
various times of the day and weekend. These targets must be
achieved to meet the conditions of the work plan and license.
A noise study for the project was undertaken for the SEES.
Further noise modelling work was carried out during the plant
design stage to provide recommendations for building design and
noise attenuation practices.
Features of the design include the bulk of the crushing circuit
located below ground level in an insulated building and the use
of insulated buildings to cover most remaining plant. Conveyors
have been covered and sound skirts installed to mitigate noise.
PROCESS AND PLANT DESCRIPTION
General
The main activities for processing gold ore at BML’s operations
are:
• crushing;
• grinding and gravity concentration;
• gravity concentrate upgrade in gold room;
• flotation;
• cyanidation of gravity concentrate upgrade tailing and
flotation concentrate;
• acid washing and elution of loaded carbon;
• regeneration of eluted carbon;
• detoxification of cyanidation leach residue slurry,
• electrowinning and smelting to doré bullion;
• de-sliming of the flotation tailings to produce a sand product;
and
• reagent mixing, storage and distribution.
Overview
The overall process logic is shown in the simplified process flow
sheet in Figure 4 and Figure 5.
The plant design relies on gravity methods (gold trap, jigs
and centrifugal concentrators) to recover the majority (up to
70 per cent is expected) of the gold in the ore. The recoverable
remainder is collected by froth flotation. The gold production
routes are:
179
E J McLEAN et al
FIG 4 - Simplified process flow sheet, one of two.
180 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference
 PROCESS DESIGN ASPECTS AT BENDIGO MINING’S GOLD OPERATIONS
FIG 5 - Simplified process flow sheet, two of two.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 181
E J McLEAN et al
• Gravity concentrate is upgraded in the gold room using
shaking tables and a centrifugal bowl concentrator to produce
a final product suitable for calcination and smelting.
• Gold room tails (from shaking tables) are subject to intensive
cyanidation. The soluble gold is electrowon from the leach
solution. Cathode clean-up is smelted in the gold room.
• Flotation concentrates are treated using the CIL process.
Gold is eluted from loaded carbon and is electrowon from the
pregnant eluate. Cathode clean-up is smelted in the gold
room.
Process plant
A brief description of the processing stages is provided below.
A plan view of the process plant general arrangement is shown in
Figure 6.
Crushing and HPGR
Ore is delivered by truck from the underground mining operation
to a run-of-mine (ROM) storage pad. A front-end loader transfers
ore from the ROM pad stockpile to the two stage crushing plant.
A conveyor delivers crushed ore to the fine ore bin. Crushed ore
reclaimed from the fine ore bin is fed to the high pressure
grinding rolls (HPGR) operating in closed circuit with the ball
mill screen.
Extremely coarse gravity recoverable gold nuggets are
recovered in the ball mill screen gold trap in the screen feed box.
Gravity and grinding
The ball mill screen undersize is pumped to the primary jigs. Jig
tails gravitate to the centrifugal concentrator via a scalping
screen. Spinners upgrade primary jig concentrates.
Tails from the centrifugal concentrator are pumped to the mill
cyclone cluster. The cyclone underflow gravitates to the ball mill
feed chute. Ground ore slurry discharging from the ball mill
combines with mill screen undersize in the mill pump hopper.
The slurry is pumped back to the gravity gold recovery circuit.
Thickening and flotation
The (grinding circuit) cyclone overflow slurry gravitates to the
float feed thickener to remove excess water. Thickener underflow
is pumped to the conditioning tank where dilution water to
control flotation feed slurry density, activator, collector and
frother are added. The overflow of the conditioning tank feeds
the flotation cells. Flotation is in a single bank of rougher cells in
a 1/2/2 arrangement. Flotation concentrate slurry is pumped to
the concentrate thickener.
Tails storage facility
Float tailing slurry is pumped to the sand cyclone cluster. Fines
are removed in the cyclone overflow, which gravitates to the
flotation tail thickener to remove excess water.
The cyclone underflow (sands) slurry is diluted to a controlled
pumpable density and pumped to a section of the tailings storage
facility. Thickener underflow is pumped to the flotation tailings
storage facility. Supernatant water is decanted from the settled
solids on the TSF and pumped back to the process water tank.
Gold room gravity recovery
Concentrate from the mill screen gold trap discharges into a
kibble in the gold room. Coarse concentrates from the gold trap
are hand sorted to recover coarse gold nuggets. After sorting,
coarse concentrates are crushed using the concentrate roll
crusher and fed to the gold room gravity spinner.
182 Fremantle, WA, 19 - 21 March 2007
Spinner and centrifugal concentrator concentrates from the
gravity/grinding circuit are periodically flushed to the gravity
concentrate surge tank and pumped to the gravity concentrate
tank in the gold room. Accumulated spinner and centrifugal
concentrator concentrates together with gold room gravity
spinner tails are fed to the gold room fine screen. Oversize
particles in the concentrates are fed to the concentrate roll
crusher and then to the gold room gravity spinner. Spinner tails
are returned to the gold room fine screen.
Fine concentrates (undersize) are fed to the concentrate jig. Jig
concentrates are upgraded using two shaking tables in series,
whereby the final concentrates from the tables and the gold room
spinner are calcined prior to direct smelting with fluxes into doré
bullion.
Intensive leach and gold recovery
Jig tail and table tails from the gold room are leached in a
proprietary system (intensive leach reactor, or ILR), where gold
dissolves under aggressive reaction conditions in an enclosed
rotating drum. Cyanide and sodium hydroxide solutions are
added to the ILR. Clarified pregnant solution containing
solubilised gold (electrolyte) is circulated through an
electrowinning cell. Gold is deposited on the cathodes. The
barren eluate is recycled to the ILR circuit.
At the completion of the electrowin cycle, the loaded cathodes
are washed by high pressure water sprays and the washings
drained from the cell and the sludge filtered. The filter cake is
dried prior to smelting with fluxes into doré bullion.
Flotation concentrate leaching
The flotation concentrate slurry is thickened to remove excess
water and combined with tails slurry from the ILR. The
combined slurry is screened to remove trash, then mixed with
lime slurry for pH control and pumped to the CIL circuit.
Sodium cyanide is added to the slurry as it enters the first
agitated CIL tank. Oxygen, to aid dissolution, is added to the CIL
tanks down-shaft in the agitator in each tank.
Dissolved gold is adsorbed onto activated carbon granules.
Carbon is advanced counter-current to the pulp flow by air lifts.
Air swept screens retain carbon in each tank.
Detoxification
Leached slurry, discharged from the agitated CIL tanks, is
treated in the cyanide detoxification reactor using a modified
copper catalysed, air/SO2 process. This reduces the (weak acid
dissociable) cyanide concentration to below permitted limits. The
detoxified CIL tailings slurry is pumped to a separate lined
impoundment.
Loaded carbon gold recovery
Loaded carbon from the first CIL tank is recovered by air lift and
screened. Water sprays on the screen assist in removing slurry
from the carbon. The separated loaded carbon gravitates to the
acid wash column. After acid treatment and water rinsing to
remove the residual acid, the carbon is transferred to the elution
column.
Gold is stripped (eluted) from the loaded carbon into a gold
bearing solution (eluate) using a hot (130°C) caustic cyanide
solution. The partially cooled eluate is circulated through a
dedicated electrowinning cell. At the completion of the
electrowin cycle, the loaded cathodes are washed by high
pressure water sprays and the washings drained from the cell and
the sludge filtered. The filter cake is dried prior to smelting with
fluxes into doré bullion.
Ninth Mill Operators’ Conference
 PROCESS DESIGN ASPECTS AT BENDIGO MINING’S GOLD OPERATIONS
FIG 6 - Plan view process plant general arrangement.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 183
E J McLEAN et al
Regeneration
Barren, or stripped, carbon is processed through a regeneration
kiln which operates at 750°C to restore (reactivate) the gold
adsorption capacity of the carbon. The reactivated carbon is
water quenched, screened and returned to the last CIL tank.
Reagents
The reagent area includes facilities for the offloading, mixing,
storage and distribution of flotation collector, activator, frother,
flocculant, hydrated lime, sodium hydroxide, sodium cyanide,
sodium metabisulfite and hydrochloric acid. Reagents used may
vary from time to time depending on performance, cost and
availability.
184 Fremantle, WA, 19 - 21 March 2007
ACKNOWLEDGEMENTS
Bendigo Mining Limited’s assistance in preparation of this paper
and permission to publish is acknowledged. The input of BML
and its consultants, for aspects of process design and project
development in treating these ‘nuggety’ quartz reef ores and
meeting challenging site specific conditions, is also
acknowledged.
REFERENCES
Johansen, G, 2005. A solution to grade estimation in a high nugget
environment – the Bendigo experience, The AusIMM Bulletin,
1:67-73.
Johansen, G, English, D, Lane, G, Hayward, T and Gardula, A, 2005.
Bendigo utilises high pressure grinding rolls to maximise gravity
recovery, The AusIMM Bulletin, 2:54-58.
Ninth Mill Operators’ Conference
Beaconsfield Gold Mine — Ironing Out the Bugs
R Holder1
ABSTRACT
The Beaconsfield gold mine processing plant is a complex operation and
utilises crushing, grinding, gravity gold concentration, sulfide flotation,
bacterial oxidation, cyanidation and zinc cementation to recover gold
from the partially refractory ore. This paper discusses the challenges,
developments and advancements throughout the processing plant since
start up, which have culminated in a 25 per cent increase in the mill
throughput over the nameplate design and a concomitant gold recovery
increase from 75 to 92 per cent. The improvements have been attained via
advancements in mechanical integrity, metallurgical review, process
control and operator training, resulting in the current operation producing
over 630 000 ounces of gold to date.
INTRODUCTION/HISTORY
Originally discovered in mid 1877, the Beaconsfield gold mine
in northern Tasmania started its life as an open pit. Subsequent
underground development saw mining activities follow the
Tasmania reef down to 454 metres, at which point it was closed
as a result of the increasing refractory nature of the ore, pumping
difficulties and labour shortages due to the First World War. Ore
production during the life of the mine was 1.085 Mt at a
recovered grade of 24.5 g/t gold to yield 854 570 oz.
Beaconsfield’s second era of gold production occurred in late
1999. The current operation is a joint venture between Allstate
Explorations NL as manager, and Beaconsfield Gold NL. The
late start-up and deficient engineering design hampered
production and gold recovery. This led to the joint venture
partners having significant financial difficulties, which in turn led
to the appointment of an administrator to Allstate Explorations
NL on 8 June 2001, and a receiver/manager for Beaconsfield
Gold NL in late June 2001.
Geology
The orebody mineralogy consists of a quartz/ankerite/sulfide
vein hosted within sandstone, siltstone and limestone. The
sulfides are predominantly pyrite, arsenopyrite and minor
chalcopyrite with lesser galena and sphalerite. Gold occurs in
close association with arsenical pyrite, being refractory in part,
and as free gold.
Process summary
The Beaconsfield Mine JV treatment plant is designed to treat
sulfide ore at a rate of 200 000 tpa to produce approximately
100 000 ounces of gold per annum. The mined ore is partially
refractory and requires oxidation of the sulfides to allow the gold
to be leached using cyanide. Two gold products are produced; a
gravity dore and a leach dore post bacterial oxidation and
cyanide leach circuits.
Ore from the underground mine consisting of 2.5 per cent
sulfide sulfur and 10 - 20 g/t gold is transported to the plant by
semi-trailer, crushed to less than 14 mm and stored in a fine ore
bin silo. This crushed ore is then milled at a rate of 25 t/h in a
closed circuit ball mill to 80 per cent passing 170 microns. The
gravity gold circuit consists of a Gekko in-line pressure jig (IPJ)
in parallel with a 30” Knelson concentrator. The IPJ concentrate
1. FAusIMM, Mill Superintendent, Allstate Explorations NL, PO Box
58, Beaconsfield Tas 7270. Email: RichardH@allstateexp.com.au
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
is upgraded further via a 20” Knelson concentrator. The
combined 20” and 30” Knelson concentrates are upgraded over a
Gemini GT1000 table, prior to being calcined and smelted.
Gold bearing pyrite and arsenopyrite minerals are recovered in
a rougher/scavenger/cleaner flotation circuit incorporating flash
flotation. The flotation concentrate is reground to 22 microns
before being oxidised using bacteria in the bacterial oxidation
circuit. The oxidised gold bearing concentrate is washed of iron
and arsenic soluble salts and acid. The resulting acidic waste
stream is neutralised with flotation tailings and limestone and
pumped to the tailings dam. The washed oxidised gold bearing
slurry is leached with cyanide and then filtered to recover the
dissolved gold in the pregnant solution stream. Finally, the gold
in solution is recovered using zinc cementation precipitation
process (Merrill Crowe) before smelting into leach dore bars.
The barren solution from the Merrill Crowe circuit is detoxified
using hydrogen peroxide to achieve a 2 ppm weak acid
dissociable (WAD) cyanide level within the tailings dam.
Flotation tailings are classified to produce a coarse fraction,
which can be mixed with cement and used for underground
backfill when required. The remainder of the flotation tailings
are thickened and pumped to the tailings dam.
Refer to Figure 1 for the process flow sheet.
GRAVITY CIRCUIT
The original mineralogical test work on Beaconsfield core
samples indicated that the gravity gold recovery ranged from
1.4 per cent to 66.4 per cent, with an average recovery of
51.2 per cent. The initial gravity gold recovery achieved by the
‘as built’ processing plant averaged 27 per cent.
The original gravity circuit as installed is shown in Figure 2.
The circuit consisted of pumping 50 t/h of ball mill discharge to
a Gekko in-line pressure jig (IPJ1500), with the concentrate
being upgraded via a 12” Knelson concentrator. The 12” Knelson
concentrate was finally tabled over a small Gemini GT250 table.
Tailings from the jig, 12” Knelson and table reported back to the
mill discharge hopper. A 20” Knelson was installed on the
flotation tailings as an ‘insurance measure’, with the intent to
recover any free gold misreporting to the final tailings.
Several gravity gold surveys were undertaken and indicated
the following:
• Free gold was being recovered to the flash float concentrate,
which then reported to the bacterial oxidation (BacOx)
circuit bypassing the gravity circuit. Gravity recoverable gold
(GRG) tests on the BacOx feed indicated that up to 15 per
cent of the gold present was recoverable via a gravity device.
• The tabling of gravity concentrate on day shift resulted in
spikes in the float tailings due to inefficient operation of the
table, and an inability to produce a ‘clean’ concentrate for
smelting.
• Knelson dump times were over one hour to accommodate the
batch tabling operation and concentrate storage limitations.
The optimal cycle time was significantly lower.
• Gold recovery by the jig was dropping off below 212 µm.
• The jig bed was blinding with grinding ball shards, reducing
the efficiency and gold recovery.
• Treating the flotation tailings via the 20” Knelson was
ineffective.
185
R HOLDER
Ore from Mine
Crushing Grinding
Gravity Circuits
Flash Flotation
Gold Dore
Bacterial Oxidation
Residue Filtration
Cyanide Leaching
Filtration and Washing
Zinc Precipitation
FIG 1 - Beaconsfield process flow sheet.
Based on the findings of surveys, several circuit and
operational changes were implemented and include:
• A 30” Knelson was installed in parallel to the in-line
pressure jig, and was also fed from the mill discharge hopper.
The tailings from the 30” Knelson then reported to the flash
float cell thereby limiting the misreporting of gravity
recoverable gold to the flotation concentrate.
• The mill discharge hopper was partitioned, such that all
ground product was treated through a gravity device prior to
classification or flotation.
• The gold room table was enlarged to a Gemini GT1000 (after
an unsuccessful attempt with a Wilfley table) and
subsequently automated such that the middlings stream was
pumped back to the table feed. Tabling occurred 24 hours per
day without the need for an operator, resulting in improved
efficiency and security. An intensive cyanide leach reactor
was explored to replace the table; however, the potential risk
associated with a subsequent toxic effect to the ensuing
bacterial oxidation circuit and long payback period ruled this
out.
• The gold room table tailings stream was redirected to the
flotation concentrate regrind circuit and not back to the ball
mill, as it contained sulfides and fine gold that was subject to
further treatment, reducing the potential of loss to the
flotation tailings.
• The in-line pressure jig was modified by cutting down the
internal segmented ring from 100 mm to 30 mm to reduce
the potential blinding of the ragging by shards of steel from
the grinding media. The jig ragging and screen aperture was
also enlarged to increase the yield from 2 t/h to 6 t/h.
• The 20” Knelson was moved from the float tails to treat the
increased jig concentrate mass flow.
• The 12” Knelson, previously treating the jig concentrate, was
moved to the regrind circuit to recover fine gold that reported
to concentrate via flotation in the unit cell, conventional
flotation circuit or from the gold room table tailings. The 12”
Knelson concentrator speed was increased to develop 120 ‘G’
186 Fremantle, WA, 19 - 21 March 2007
Classification
Flotation
Regrind Mill
Backfill Plant
Mine
Neutralisation
Tails Thickener
Leach Residue
Tails Dam
Cyanide Destruction
force as opposed to the standard 60 ‘G’ in order to increase
the recovery of fine (<10 µm) gold.
• Following the automation of the Gemini table, the Knelson
concentrators were set up on an 18 minute dump cycle to
maximise recovery and minimise any displacement of gold
by the coarse sulfides. Due to the competing trade-off
between short cycle times and increased ‘off-line’ periods,
the 30” Knelson was fitted with a ‘dynamic braking module’,
increasing the utilisation.
The current gravity circuit flow sheet is shown in Figure 3.
Figure 4 trends the mill feed grade versus gravity gold
recovery as various process improvements were installed and
commissioned. The trend highlights that gravity gold recovery
increased with feed grade through 2003 - 2004. However, the
gravity gold recovery has remained elevated at approximately
50 per cent since early 2005 when feed grades treated were
similar to those treated in the first few years of the operation,
prior to circuit changes, when gravity recovery averaged 27 per
cent. Gravity recoverable gold (GRG) tests on the BacOx feed
has indicated <0.2 per cent of the gold was recoverable to a
gravity concentrator validating the success of the installed
circuits and operational strategies.
FLOTATION CIRCUIT
The original flotation circuit was designed to recover the gold
bearing sulfides with performance guarantees to produce a
concentrate grade of 27 per cent sulfide, and a flotation tailings
<0.3 g/t gold. Actual performance achieved the target sulfur
grade, but gold loss to tailings was initially 1 g/t gold.
The original flotation circuit consisted of a flash flotation
cell in the primary grinding circuit in combination with a
conventional flotation circuit treating cyclone overflow post
gravity gold recovery. The conventional float circuit consists
of a bank of rougher/scavengers with the rougher/scavenger
concentrate being cleaned in cleaner cells to achieve the desired
sulfide concentrate grade for optimal bacterial activity. Reagents
used are copper sulfate for arsenopyrite activation, sodium
isobutyl xanthate and interfroth IF106 frother.
Ninth Mill Operators’ Conference
 BEACONSFIELD GOLD MINE — IRONING OUT THE BUGS

Ore from Mine (100%)

Ball Mill
Mill Discharge Inline Jig 12” Knelson

Overflow
Classification Gemini Table
Underflow

Unit Cell Flotation Gravity Gold

(28%)

Concentrate 20” Knelson

Regrind

Float Tails (6%)

BacOx (66%)

FIG 2 - Original gravity gold circuit and gold balance (treating 200 per cent of new feed tonnage).

Ore from Mine (100%)

Ball Mill Mill Discharge Inline Jig 20” Knelson

Classification 30” Knelson Gemini Table

Unit Cell Gravity Gold

(53%)

Flotation Concentrate 12” Knelson

Regrind

Float Tails BacOx (43.4%)

(3.6%)

FIG 3 - Current gravity gold circuit and gold balance (treating 400 per cent of new feed tonnage).

60 25
Gemini table upgrade
Table automation

50 30" Knelson to
milling circuit 20
Feed grade (g/t)

40
% Gravity

15

30 Spirals installed on
BacOx residue

Regrind milling trials 10

20
Feed Grade
Coarsened grind size.
P80 75 to 120 µm, then 12” Knelson installed % Gravity Gold
170 µm in regrind circuit Recovered
10 5
Mar-99 Dec-99 Aug-00 Apr-01 Dec-01 Sep-02 May-03 Jan-04 Sep-04 May-05 Feb-06 Oct-06
Date

FIG 4 - Effect of circuit modifications on gravity gold recovery.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 187

R HOLDER
Surveys of the flotation circuit and subsequent mineralogical
analysis indicated that:
• The majority (>50 per cent or 0.5 g/t) of the gold and sulfide
loss in the flotation tailings was <10 µm. Gold and sulfur
recovery in the -10 µm fraction was low at 53 per cent (refer
Figure 5).
• The upgrade ratio of the gold and sulfides was poor in the
<10 µm fraction and exacerbated by entrainment and
subsequently poor selectivity.
• Mineralogical analysis of the tailings highlighted that fine
gold and sulfides were encapsulated within ankerite
(carbonaceous gangue mineral), and there were lean
composites in the scavenger concentrate that required further
liberation to permit recovery to final concentrate (refer
Figure 6).
• Froth structure was weak due to the large froth surface area to
cell volume ratio for the quantity of mineral to be recovered.
• Control of the circuit was by manual sample collection,
LECO analysis for sulfur followed by manual adjustment to
the air to achieve the target concentrate grades. This four
hour turn around process was almost futile based on the
40 minute circuit residence time and cyclic feed grades.
Based on the plant surveys and observations, the following
changes were made to the circuit:
• coarsening of the grind size from 75 µm to 170 µm,
70
60
50
% distribution
40
30
20
10
0 0
-10 10
Tailings mass %
Flotation Gold recovery
FIG 5 - Flotation size by size recovery and gold/sulfide distribution in the flotation tailings.
188
• installation of in stream analysis system for inferring a sulfur
grade,
• cycloning and regrinding the coarse or higher specific value
minerals recovered to the scavenger concentrate prior to
refloating,
• installation of froth crowders, and
• fully automating the circuit operation.
Coarsening of the primary grind size
The original comminution circuit was designed to achieve a
cyclone overflow P80 of 75 µm. This figure was derived from
laboratory test work. Comparison of the size distributions of the
plant cyclone overflow with that of metallurgical laboratory test
samples revealed that despite having the same P80 value, this was
only one point on the distribution curve that was totally different.
As can be seen in Figure 7, the size distribution achieved in the
laboratory, using a rod mill was very different to the plant data.
The laboratory rod mill product contained approximately
47.5 per cent <38 µm and 14 per cent <10 µm whereas the plant
product contained 58 per cent <38 µm, and 30 per cent <10 µm.
To reduce the fines generation prior to flotation, several trials
were undertaken and included, manipulation of the crushed ore
feed size to the ball mill, ball charge, milling density and cyclone
inclination. All modifications had a minor benefit due to
limitations by the cyclones. Ultimately the four operating 10”
100
80
Recovery
60
40
20
21 38 53 75 106 150
Size (µm)
Tailings Au distribution Tailings S distribution
Flotation Sulfur Recovery
FIG 6 - Flotation tailings mineralogy.
Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

100
P80 correct @ 75 µm, but
large difference in % -10 µm
Cumulative % Passing
80
60
40
20
0
10
FIG 7 - Laboratory versus plant grind size distribution.
Warman Cavex cyclones were replaced by a single 15” cyclone.
There was an immediate improvement with a reduction in the
circulating load from 400 per cent to 250 per cent and the P80
increased from 75 µm to 170 µm.
The progressive coarsening of the grind size resulted in the
following changes:
• The P80 increased from 75 to 170 µm, and the mass per cent
<10 µm reduced from 30 per cent to 14 per cent.
• Flotation selectivity improved, resulting in lower gold and
sulfide loss to tailings.
• Mill throughput increased from 25 t/h to 38 t/h for the same
power draw (ie 35 per cent reduction in power per tonne
milled). The higher throughput rate has permitted
campaigning of the ore to provide opportunity for
maintenance and improved labour utilisation.
• Grinding media consumption, pump wear rates reduced by
50 per cent.
In stream analysis
In stream analysis (ISA) was reviewed and installed at
Beaconsfield for the following reasons:
• To maximise gold and sulfide recovery at the target sulfur
grade with varying feed grades.
• To provide consistent feed sulfur content to the bacterial
oxidation circuit to maintain ‘steady state’ operation.
• To reduce operator judgement/intervention.
• To instantly highlight process problems (reagent blockages,
etc).
• To reduce assaying costs in the laboratory (one or two
calibration check samples per day versus 18 to 24 process grab
samples). The contract laboratory was reduced from 24 hour
coverage to 12 hour coverage saving over $100 000 pa.
An AMDEL in stream analyser was installed to measure the
iron and density of the cleaner concentrate and flash float
concentrate. From the iron and density count rates, an inferred
sulfide grade was determined and utilised as the process variable
for the flotation circuit automation.
Regrinding the scavenger concentrate
The mineralogical analysis of the flotation circuit highlighted
that there were lean composites being recovered to the scavenger
Ninth Mill Operators’ Conference
BEACONSFIELD GOLD MINE — IRONING OUT THE BUGS
Feasability (lab)
Plant (Dec 00)
Plant (Oct 05)
100 1000
Size log (µm)
concentrate but failing to be recovered to the final concentrate
due to the low flotation kinetics. The lean composites ultimately
reported to the flotation tailings. This situation was exacerbated
by the progressive coarsening of the primary grind size. To
improve the probability of recovery of the value minerals within
the lean composites to the final concentrate, the scavenger
concentrate was cycloned, with the coarse and/or higher SG
minerals reporting back to the primary ball mill feed. The regrind
cyclone overflow reported back to the rougher feed.
Froth crowders
Froth volume/height adjustment by crowders permits increasing
or decreasing the retention time of solids in the froth as a
separate measure of controlling selectivity. Flotation cells need
to be selected for the optimal surface area, carrying rate and
concentration ratio for the specific task and particle size range.
The surface area per unit volume of the installed rougher and
scavenger cells at Beaconsfield for the required mineral recovery
was high at 0.66 m2/m3, resulting in shallow froth depths and cell
pulping due to lack of froth stability. Transverse froth crowders
were subsequently installed lowering the surface area per unit
volume in the roughers by 55 per cent and the scavengers by
80 per cent. The use of the crowders resulted in greater flotation
selectivity, circuit stability and a 40 per cent reduction in frother
usage.
Flotation circuit automation
Following the installation of the ISA, the float circuit was fully
automated to maintain a constant concentrate sulfur grade
irrespective of changes to the feed grade. Cascade control loops
adjusting air addition (primary control) and level (secondary
control) were set up to:
• Maintain the target unit cell concentrate sulfur grade.
• Maintain the target cleaner concentrate sulfur grade.
• Manipulate the rate of rougher concentrate recovery as the
feed sulfur grade varied. This was achieved by using the
pinch valve positioning on the cleaner tailings as the process
variable to control the rougher pull rates.
• Maintain a constant scavenger cell pull rate, using scavenger
regrind cyclone pressure as the feedback loop.
Refer to Figure 8 for the current flotation circuit and control
logic.
Fremantle, WA, 19 - 21 March 2007 189
R HOLDER

Cyclone
overflow Air

Cleaner air rate adjusted to

maintain target sulfur
Cleaner grade base on ISA PV

In stream
Cleaner
Rougher air rate adjusted to maintain analyser
concentrate
constant flow through cleaners to
accommodate feed grade variations
Air
Air

Rougher
Tails
Scavenger
Feed
Tank

P
Scavenger air rate adjusted
to maintain target regrind
To primary mill cyclone pressure

FIG 8 - Flotation circuit automation.

The result of the ISA installation and subsequent automation
of the circuit has:
• stabilised the feed grade to the BacOx circuit;
• provided greater control to lower the final sulfur concentrate
grade and to maximise gold recovery without detriment to
the bacterial activity;
• reduced the annual contract laboratory cost; and
• reduced operator dependence, which was part of the
justification to reduce manpower by 20 per cent.
The developments within the gravity and flotation circuits
have resulted in a 40 per cent reduction in gold loss to tailings,
equating to approximately $2.6 million dollars per annum.
BACTERIAL OXIDATION REVIEW
The Bacterial oxidation circuit installed at Beaconsfield uses the
Mintek-Bactech technology (BacOx), as opposed to the other
commercial provider being Billiton Process Research (BIOX).
The BacOx process is used to oxidise the sulfides (pyrite and
arsenopyrite) and render the encapsulated gold amenable for
dissolution by cyanide. Without oxidation of the sulfides, the gold
recovery via direct cyanidation ranged from five to 60 per cent
during the feasibility testing.
oxidises the sulfides. The mechanism is considered indirect
since the bacteria do not need to attach to any solid surface.
The residence time of a continuous bacterial oxidation process
is dependent on two factors:
• time required for the bacterial population to double, and
• time needed to oxidise the material of interest.
Due to the time requirement for the bacterial population to
double, the circuit was designed with three primary reactors all
in parallel where 75 - 85 per cent of the oxidation occurs, with a
residence time of approximately three days. The slurry from the
three primary reactors is transferred to a secondary, tertiary and
finally a quaternary reactor in series prior to exiting the circuit,
as shown in Figure 9. If the residence time in the primary
reactors was reduced there is a potential to ‘wash out’ the
bacteria, ie the bacteria are being displaced from the reactor by
new feed at a faster rate than the bacteria can double.
1
Oxidised
Product
2 4 5 6
Feed
y

Bacterial oxidation is similar to chemical leach processes and

ar

Tank
nd

ry

ar

involves electrochemical reaction between an oxidant and the

co

rtia

rn
Se

ate

sulfide minerals. The primary oxidant is the oxygen in the air and 3
Te

Qu

the secondary oxidant is iron (III). The bacterial culture acts as a

catalyst and gains its energy from the electron transfer that
occurs in the oxidation-reduction reactions. Primary Reactors

The bacteria utilised are mesophilic iron and sulfur oxidising
cultures. Two major mechanisms have been postulated (Ralph
and Wensheng, 1993) to describe the oxidation of pyrite and
arsenopyrite by the bacteria:
• Direct attachment to the mineral surface, oxidising the
sulfide for an energy source.
• Indirect mechanism where the bacteria catalyse the
conversion of ferrous to ferric iron in solution, which then
FIG 9 - BacOx reactor process flow.
The bacterial oxidation plant was to be designed to reach a
level of 95 per cent oxidation processing 2.85 t/h of concentrate
containing 27 per cent sulfide. The plant, as designed and
installed, struggled to achieve 80 per cent oxidation at 1.6 t/h
(ie 47 per cent of target). The reasons for the failure to achieve
the oxidation performance at design throughput were:

190 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 BEACONSFIELD GOLD MINE — IRONING OUT THE BUGS

• The design of the agitation system was defective, with the
agitator blades being positioned more than halfway up the
tank, resulting in the bottom third of the tank filling with
solids, thereby reducing the residence time for the bacterial
numbers to double.
• Failure of the gearbox output shafts, internal gears, blades
and agitator shafts occurred. Subsequent engineering review
via strain gauges installed on the reactor blades and shafts
indicated that the impeller loads acting on the reactor shafts
were underestimated by up to eight times. The main cause
of the additional load on the agitators was due to an
inadequacy of tank baffling needed to control the swirling
action in the tank. Refer to Figure 10 for a photo of a failed
agitator blade.
• Installation of spirals on the BacOx residue to recover
partially oxidised (short circuited) sulfides and liberated
gold, which were subsequently returned to the regrind/
gravity circuit prior to BacOx.
• Installation of a weir overflow system between reactors as
opposed to the requirement for manually operated airlifts for
level control, thereby stabilising residence time in each
reactor.
The current optimal conditions and controls for sulfide
oxidation at Beaconsfield are shown in Table 1.
TABLE 1
Optimal conditions and controls for the Beaconsfield
oxidation circuit.

Optimal conditions Beaconsfield control

pH in range of 1.2 - 1.4 Target flotation concentrate grade
Agitator shaft of 25 per cent sulfide sulfur for
acid balance.
Presence of carbon dioxide to Ankerite (CaCO3.FeCO3) is a non-
promote cellular growth of sulfide gangue diluent in the
bacteria flotation concentrate and the
source of carbon dioxide
Air sparge ring
Dissolved oxygen level to be Blowers plus back-up diesel
greater than 2 ppm in the reactors compressor to maintain some
Blade torn from oxygen to at least one reactor in
hub and folded event of a power outage
back Nutrients to sustain growth Addition of potassium, nitrogen
and phosphorous (fertiliser)
Temperature range of 43 - 47°C Cooling towers and heat
Cooling coils (the oxidation reaction is exchanger tubes in the reactors
exothermic and the Beaconsfield
FIG 10 - BacOx reactor. mesophile bacteria die at 49°C)
Less than 40 g/L iron + arsenic in Operate reactors at 14 per cent
solution (there is a retardation of solids density to limit ionic
The rectification work was extensive, and required: bacterial activity above 40 g/L Fe concentration of the substrate
• modifications to agitator shafts, + As)
Five days residence time Tank volume (2300 m3 to treat
• strengthening every agitator blade with sandwich plates and 3 t/h of flotation concentrate)
bigger bolts,
No or low levels of toxins Control of water quality used in
• structural strengthening of the agitator support beams, process
• installation of four additional tank baffles to control swirling
action within the tank,
• installation of pedestal bearings under the gearboxes to Bacterial oxidation residue
dissipate the loads from the agitator into the structure rather During the oxidation of the sulfides, there is a weight loss of
than through the gearbox, approximately 45 per cent. Mineral breakdown results in a
• lowering of the agitator and air spare rings by 1.8 metres, particle size reduction of the product to a P80 of approximately
and 15 µm. Following the oxidation of the sulfides, the oxidised
product is thickened in an EIMCO Ultrasep thickener, with the
• upgraded the air blowers to accommodate the increased head acidic overflow containing the solubilised iron and arsenic
pressure produced by lowering the air sparge ring. stabilised and neutralised using flotation tailings and powdered
Following the rectification of the engineering defects and limestone.
achieving steady state operation, the throughput and oxidation Gold loss to the residue thickener overflow was visually
levels did rise, however, not to the design levels. Further circuit evident during the initial stages of operation, and no
developments were implemented to exceed the original contingency was in place to recover the gold losses in this
throughput rates at the desired oxidation level of 95 per cent. The stream estimated at over $1 000 000 pa. The loss was attributed
developments included: to a small differential in the thickener rise rates and the solids
• Using four primary reactors in parallel as opposed to three to settling rate. A range of chemicals were trialled with limited
stabilise the operating environment for the bacteria. success, and ultimately a second thickener was installed,
immediately halving the gold loss. Further analysis of the
• Lowering the P80 of BacOx feed from 38 µm to 22 µm to system resulted in an automated sand filter being installed to
increase the surface area for reaction. This change had the treat the thickener overflows, reducing the loss by a further
greatest effect on the circuit performance. 50 per cent.
• Judicious manipulation of nutrients, sulfur grade, temperature The thickened BacOx residue is washed over a vacuum belt
and pulp density. filter to remove cyanide consumers prior to cyanidation.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 191

R HOLDER
LEACH CIRCUIT
The cyanide leaching circuit is different to most other operations.
The circuit does not use carbon for recovery of the solubilised
gold, but consists of six cyanidation tanks, followed by belt
filtration and cake washing to recover the high grade gold liquor.
Once the gold is recovered into a clean solution, the gold is
precipitated using zinc cementation (Merrill Crowe). The
rationale for the non-conventional circuit was that the Merrill
Crowe was purported to be more suited to the low volume, high-
grade solutions. In comparison, conventional carbon in leach
(CIL) would have necessitated large volumes of carbon
movement in the relatively small leach reactors.
One of the major problems with biological oxidation
pretreatment is the excessive consumption of cyanide during the
subsequent cyanidation step, due to the production of thiocyanate
(SCN-). Thiocyanate is formed as a result of side-reactions of
cyanide with reactive sulfur species generated during biological
oxidation.
The initial leach performance was severely impacted by the
low levels of oxidation resulting in competition for cyanide and
complexing with intermediate sulfur species. Concentrations of
up to 10 000 ppm thiocyanate were encountered, as well as high
levels of weak acid dissociable (WAD) cyanide complexes due to
the presence of cyanide soluble copper in the ore.
The conventional silver nitrate, rhodamine titration for free
cyanide determination was hampered by the colour of the
solution post bacterial oxidation, as well as the equilibrium
between the orders of cuprous cyanide. During titration with
silver nitrate in the presence of copper cyanide complexes the
equilibrium between Cu(CN)32- and Cu(CN)43- changes. The shift
in equilibrium results in the reduction in the order of cuprous
cyanide and the release of CN-, which is falsely titrated as ‘free’
cyanide:
Cu(CN)32- → CN- + Cu(CN)22-
Variations in operator titrations for free cyanide were up to
1500 ppm on the same sample. Operator adjustment of the
cyanide levels resulted in wild swings from 1000 ppm to
10 000 ppm in cyanide concentration where the target levels
were around 2500 ppm. The copper and cyanide level
fluctuations impacted on the subsequent detoxification circuit.
The circuit improved via the following advancements:
• increased levels of oxidation precyanide leaching, thereby
reducing the competition for cyanide by the intermediate
sulfur species;
• alkaline preaeration of the slurry prior to cyanidation to
convert residual sulfur to sulfate, reducing SCN- generation;
• multiple point dosing of cyanide through the circuit to
control within a fine range; and
• installation of two cyanide analysers to monitor the cyanide
concentration via potentiometric determination, with
subsequent automatic dosing of cyanide to the individual
tanks.
ZINC CEMENTATION (MERRILL CROWE)
The zinc cementation process for the recovery of gold and silver
from cyanide solutions was introduced in 1890 and was
eventually replaced by CIP from 1949. Zinc cementation is
generally preferred over CIP for treating high silver:gold ratios,
as well as treating high solution tenors as experienced at
Beaconsfield.
Like cyanidation, cementation is a heterogeneous redox
process. Zinc is dissolved at the anodic areas of the surface and
the electrons released serve to reduce the aurous ions at the
cathodic areas. The overall chemical reaction for the cementation
of gold by zinc is:
192 Fremantle, WA, 19 - 21 March 2007
Zn + Au(CN)2- + H2O + 2CN- → Au + Zn(CN)42- + OH- + 0.5H2
The Merrill Crowe precipitation process is adversely affected
by a range of impurity elements including copper cyanide,
sulfate and thiocyanate, all of which were in abundance
following bacterial oxidation. Initial gold recovery performance
was poor, and theoretically (Marsden and House, 1992) the
process should not have worked with the levels of impurities
generated post bacterial oxidation.
Perseverance and trial and error has resulted in the
cementation efficiency improving from a low of 60 per cent to
99.5 per cent via:
• a greater understanding of the solution chemistry,
• alkaline preaeration in the leach circuit reducing SCN-
generation,
• operating the pH at 12 - 12.5, and
• manipulating the free cyanide level set-points based on the
copper concentration.
CYANIDE DETOXIFICATION
The site environmental conditions are to attain a 2 ppm WAD
cyanide discharge level from our tailings dam. This was achieved
using Caro’s acid (H2SO5) for cyanide detoxification. Caro’s acid
is an oxidant formed by the reaction between hydrogen peroxide
with sulfuric acid:
H2O2 + H2SO4 → H2SO5 + H2O
Caro’s acid oxidises cyanide to cyanate:
H2SO5 + CN- → H2SO4 + CNO-
The cyanate hydrolyses to form carbonate and ammonium
ions:
CNO- + 2H2O → CO32- + NH4+
Caro’s acid also oxidises weakly bound metal cyanides (WAD
cyanide), resulting in the metals being precipitated as
hydroxides.
The initial detoxification process was problematic due to
erratic solution chemistry, high copper WAD cyanide levels, and
high cyanide levels. The operation of the circuit was streamlined
and costs reduced by:
• increased levels of oxidation through BacOx circuit,
• installation of spirals on the BacOx residue to scalp any
partially oxidised sulfides for re-treatment,
• alkaline preaeration of the leach feed to reduce thiocyanate
generation,
• automation of cyanide monitoring and addition through the
circuit, and
• use of hydrogen peroxide as opposed to Caro’s acid due to
the presence of copper catalysing the cyanide oxidation.
CYANIDE MANAGEMENT
Allstate Explorations became a signatory to the International
Cyanide Management Code in April 2006.
The site management have worked in collaboration with
ORICA to develop, trial and commission a ‘mini-sparge’ cyanide
system, making the handling and use of one tonne boxes
obsolete. The mini-sparge installation has resulted in the risk
rating to site personnel and the environment reducing to less than
two per cent of the original value compared to the use of one
tonne boxes of cyanide.
Ninth Mill Operators’ Conference
 BEACONSFIELD GOLD MINE — IRONING OUT THE BUGS

100

95

% Recovery 90

85

80

75

70

65

60
Fe 01

Fe 04
Fe 99

Fe 00

Fe 02

Fe 03

Fe 05
Ma 0 0

Ma 0 3

Ma 0 4

06
Ma 01

Ma 02

Ma 05
No 00

No 03
No 01

No 02

No 04

No 05
Au 01

Au 05
Au 00

Au 02

Au 03

Au 04
v-

v-
v-

v-

v-
v-

v-
b-

b-

b-
b-

b-

b-

b-
y-
y-

y-

y-

y-

y-
g-
g-

g-

g-

g-

g-
No

FIG 11 - Beaconsfield gold recovery over time (adjusted for gold contribution from tailings re-treatment).

SAFETY AND PERSONNEL CONCLUSION

Significant challenges were encountered in the first 18 months of
operation, including:
• commissioning a new processing plant with major design and
installation issues incorporating a failed bacterial oxidation
circuit and no surge capacity between unit processes;
• new workforce, some with no mineral processing experience;
• administrator appointed, resulting in job insecurity, high
turnover of personnel and inability to recruit experienced
personnel;
• management focus on tomorrow and next week rather than
next year;
• low morale, inability to concede any additional benefits to
employees based on financial situation; and
• above industry average LTIFR.
The management team and employees worked through the
issues and over time developed a stable workforce. Operators
have undertaken behavioural based safety training and are
trained to Certificate 3 in National Metalliferrous Mining
Package (MNM30305).
Figure 11 shows the climb in recovery over time as the operation
overcame many impediments. The metallurgical, operational and
personal developments has resulted in an improved safety record,
gold recovery increasing from 75 to 92 per cent, along with a
25 per cent increase in throughput and a 20 per cent reduction in
man hours required to operate the plant.
ACKNOWLEDGEMENTS
The author wishes to acknowledge Allstate Explorations NL for
permission to publish this paper and the contributions made by
all present and past personnel in advancing the plant through
many trials and tribulations.
REFERENCES
Hedley, N and Tabachnik, H, 1968. Chemistry of Cyanidation (American
Cyanamid Company: Wayne).
Marsden, J and House, I, 1992. The Chemistry of Gold Extraction, pp 597
(Ellis Horwood Limited: London).
Ralph, D and Wensheng, Z, 1992. The biological oxidation of
arsenopyrite ore, in Proceedings Biomine ’93 International
Conference and Workshop Applications of Biotechnology to the
Metals Industry, pp 7.1-7.9 (Australian Mineral Foundation:
Australia).

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 193

A Review of Processing at the Cannington Ag/Pb/Zn Mine
R Alford1 and G Clarke2

ABSTRACT
Commissioning of the Cannington processing plant began in October
1997. The nameplate capacity (of 1.55 million tonnes of ore per annum)
was achieved relatively quickly. Since then (like many mines) the
operation has strived to increase throughput and improve valuable metal
recoveries, to offset declining head grades and lift the value of the
operation. The processing plant now has the capability of treating over
3.2 million tonnes of ore per annum and in financial year 2005/06 the
combined concentrate production exceeded 600 000 wet tonnes,
approximately 80 per cent above concentrate output in the first full year
of operation (at the design feed rate). This improved performance was
achieved by making changes and additions to each of the major unit
processes – primary grinding, flotation, leaching, filtering and paste fill
preparation. The changes have included: adding a fines flotation circuit,
expanding the grinding circuit, adding a pebble crusher, switching from
continuous to batch leaching, building paste conveyors and
decommissioning the column flotation cleaning circuits. This paper
reviews these changes, outlines the current operating practices and
discusses the challenges or opportunities that remain.
INTRODUCTION
The Cannington Ag/Pb/Zn mine is located in NW Queensland
and is named after a nearby pastoral station. Apart from the
mine, there is little infrastructure in the immediate area and the
nearest provincial city, Mount Isa, is over 300 km by road to the
north-west. As a result, the mine is a fly-in, fly-out operation and
most of the employees reside in Townsville, approximately
800 km to the north-east of the mine.
The key features of the ore are the high silver and fluorine
grades. Obviously the high silver grade is a big positive for the
operation and a significant portion of the mine revenue is derived
from the silver. Fluorine is a negative and the recovery of
fluorine to the concentrates must be minimised; the processing
flow sheet includes extra or specialised stages to reject fluorine
from the products.
At Cannington, the ore is extracted from the underground mine
using an open stoping method. The ore is trucked and hoisted
to the surface stockpile. The processing flow sheet includes
grinding, flotation, fluorine leaching, dewatering and paste fill
preparation. The saleable products from the processing plant are
a Pb concentrate and a Zn concentrate. The majority of the Ag
reports to the Pb concentrate. The concentrates are trucked to a
rail siding at Yurbi (near Cloncurry) then railed to Townsville,
for shipment to smelters in Australia and overseas.
This paper reviews each of the processing steps, describing the
current practices and highlighting the changes made over the
nine years of operation since commissioning.
zones – the Southern Zone and the Northern Zone. These contain
similar mineralisations and are separated by a fault. Ore was
sourced only from the deeper, larger and higher grade Southern
Zone in the early years of the mine. In the last three years ore
production has been augmented by mining from the Northern
Zone. Prior to mining, the mean grades of the Southern Zone
were 12.4 per cent Pb, 4.9 per cent Zn and 582 g/t Ag, and the
mean grades of the Northern Zone were 8.9 per cent Pb, 3.0 per
cent Zn and 371 g/t Ag.
The ore has been divided into ten mineralisation types, on the
basis of the Pb/Zn ratio and the gangue (either Fe-rich or
siliceous). Four mineralisation types dominate mine production
and efforts are made to blend the ore, underground and on the
surface, to ensure that the Pb and Zn feed grades to the plant are
stable. Blending is also used to control the amount of talc and
arsenic in the feed blend, as high amounts of these components
can have an adverse effect on concentrate quality.
As an example of the Cannington mill feed ore blend, the
mean feed grades for calendar year 2005 (CY 05) are presented
in Tables 1 and 2 (for selected metals and minerals). The gangue
minerals are largely non-sulfides (or NSG) and these include
fluorite, talc, quartz and magnetite. Fluorine is present in fluorite,
talc and some other minor minerals. The ore generally contains a
relatively small component of gangue sulfides, the most common
is pyrrhotite. Arsenic in the ore is present as arsenopyrite and
lollingite. The silver mineralogy is relatively complex. Silver is
TABLE 1
CY 05 mill feed metal assays.
Element Assay
Ag (ppm) 505
Pb (%) 10.2
Zn (%) 3.34
Fe (%) 14.8
Sb (%) 0.046
As (%) 0.106
SiO2 (%) 41.0
CaO (%) 8.02
MgO (%) 1.47
TABLE 2
CY 05 mill feed mineral distribution.

Mineral Distribution (%)

OREBODY DESCRIPTION
Galena 11.7
Prior to the start of mining, the resource was defined as being Sphalerite 5.9
43.8 million tonnes, containing 11.6 per cent Pb, 4.4 per cent Zn
Pyrrhotite 2.2
and 538 g/t Ag (Walters and Bailey, 1998). There have been no
significant extensions to the resource since mining commenced, Pyrite 0.8
although exploration continues. The resource is divided into two Other sulfides 0.3
Fluorite 5.9
1. MAusIMM, Senior Technical Metallurgist – Concentrator, BHP Talc 1.5
Billiton, Olympic Dam Expansion Project, GPO Box 1777, Adelaide
Quartz 29.5
SA 5001. Email: Robert.A.Alford@BHPBilliton.com Formerly:
Senior Research Metallurgist, BHP Billiton, Cannington Mine. Magnetite 7.3
2. MAusIMM, Principal Metallurgist, BHP Billiton, Cannington Mine. Other NSG 34.9

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 195

R ALFORD and G CLARKE
present in solid solution with all the sulfides and in distinct silver
minerals. The silver minerals include acanthite, dyscrasite,
freibergite, native silver and pyrargyrite. The most common (in
CY 05) was freibergite.
HISTORY
The deposit was discovered in 1990 as a result of drill testing of
magnetic targets by BHP Minerals (Walters and Bailey, 1998).
Mine development proceeded swiftly and the first concentrate
was produced in 1997.
Commissioning the processing plant was completed quickly
by standards of complex base metal operations, however not
without encountering problems (Leung et al, 2000). To expedite
the production of concentrate and efficiently utilise resources,
the plant was commissioned in stages. The primary grinding
circuit was commissioned in late September 1997. The first
saleable Pb concentrate was produced on 25 October 1997. The
plant was fully commissioned by late December 1997. The
primary grinding circuit operated for two weeks at the design
throughput in November 1997, but ore availability prevented
continuous operation at this rate until mid 1998. Ag recoveries
exceeded the design target virtually from day one, which was an
important financial milestone for the operation. Pb and Zn
recoveries matched or exceeded design targets within six months.
The target recoveries were 80 per cent, 85 per cent and 75 per
cent for Ag, Pb and Zn, respectively.
One significant hurdle encountered during the commissioning
phase was difficulty rejecting fluorine from the concentrates. The
main cause of the problem was that the feed blend contained
higher than expected amounts of naturally floating talc, which
contains fluorine. The fluorine leaching circuit is effective for
removing only the fluorite from the concentrate. To remove talc,
part of the flotation circuit was reconfigured to act as a
preflotation stage. This step significantly reduced the fluorine
content of the Pb concentrate; however, fluorine rejection was
still unsatisfactory.
In the first months of production, a satisfactory product could
only be made by targeting a high Pb grade (resulting in recovery
loss). Residual fluorine in the Pb concentrate (at the target grade)
was due to fluorine bearing silicates locked with galena that
could not be removed by leaching. After discussions with the
operation’s customers, the fluorine specification was raised,
which allowed improved Pb recovery.
Since commissioning (like many mines) the operation has
strived to raise throughput and improve valuable metal
recoveries, to increase the value of the mine and offset declines
in feed grades. Over the years, steady progress has been made.
The annual feed and concentrate wet tonnages are presented in
Table 3 (note, typical moisture contents are three per cent, seven
per cent and nine per cent for the feed, Pb concentrate and Zn
concentrate, respectively). In financial year 2005/06 (FY 05/06)
the feed rate exceeded the design of 1 550 000 wet tonnes by
approximately 100 per cent. Between September 2005 and
February 2006 the average feed rate was 450 t/h (wet), not
including periods when the plant was shut down. This gives an
indication of the capability of the processing plant in the future.
In FY 05/06 the combined concentrate production exceeded
600 000 wet tonnes, approximately 80 per cent above
concentrate output in the first full year of operation (at the design
feed rate). The relative gain in concentrate production is less than
the feed rate gain due to a decline in the feed grades.
In FY 06/07 the ore tonnage to be processed is expected to be
below FY 05/06, due to a halt in ore production from the
Southern Zone of mine from May to November 2006. The halt
was required to complete an underground rehabilitation program
to improve ground conditions and ensure employee safety (see
news release, Anon, 2006). Beyond FY 06/07 annual ore
production is expected to return to normal levels.
196 Fremantle, WA, 19 - 21 March 2007
TABLE 3
Annual feed and concentrate tonnages.
Financial year Feed (t, wet) Pb conc (t, wet) Zn conc (t, wet)
FY 97/98† 1 023 130 163 988 61 944
FY 98/99 1 426 823 220 173 96 515
FY 99/00† 1 632 732 294 584 136 134
FY 00/01 1 837 464 316 779 165 351
FY 01/02 2 238 715 376 438 155 416
FY 02/03 2 382 983 377 992 169 053
FY 03/04 2 689 262 406 593 146 641
FY 04/05 3 183 628 457 317 141 089
FY 05/06 3 114 153 430 458 182 674
† Annualised.
The steady increase in mill throughput was achieved by
completing numerous capital projects. Many of these were
relatively minor de-bottlenecking projects, but two significant
plant expansions have been completed, in 2001 and 2003. In
addition to the plant expansions in 2001 and 2003, a significant
capital project was completed in 2001 to increase recovery of
fine Pb and Ag particles.
In the original design, the autogenous grinding mill (AG mill)
and the concentrate filters were deliberately oversized to provide
operational flexibility and mitigate the start-up risks (Leung
et al, 2000). The marginal cost of increasing the size of these
items was relatively minor.
In mid 2001 mine output was nominally increased from 1.5 to
1.8 million tonnes per annum, by expanding the mine into the
upper levels of the Southern Zone footwall (Smith, 2002).
Modifications to the paste plant and an expansion of the grinding
circuit were required to treat the extra ore. The paste plant was
modified by installing a novel 350 m conveyor system to allow
delivery of paste to the new part of the mine. The conveyor and
associated equipment was commissioned in December 2001. The
grinding circuit was expanded by adding a pebble crusher. A
temporary pebble crusher was installed in September 2001 and
used until November 2002. A permanent pebble crusher was
commissioned in early 2003.
The processing plant was expanded or upgraded in the second
half of CY 03 to allow sustainable treatment of 2.4 million
tonnes of ore per annum (some details of the expansion have
been reported earlier, see Okely and Fargher, 2004). The 2003
project was associated with an expansion of the mine to allow
ore production from the Northern Zone. Just prior to the
expansion the processing plant was operating near the target rate
(of 2.4 million tonnes per annum) and most of the capital works
were justified on the basis of improved metal recoveries and/or
improved concentrate quality. In early 2004 additional capital
was approved to expand the processing plant to nominally
2.8 million tonnes per annum. These additional works were
relatively minor de-bottlenecking exercises.
Work completed during the 2003 expansion included upgrades
to the grinding and Pb regrind circuits, installation of a new
larger zinc cleaner circuit, expansion of the Pb fines flotation
circuit, de-commissioning of the Pb flotation columns, a new
paste conveyor, an expansion of the Pb concentrate filter and an
upgrade to the reagent system. The major components were
commissioned in October 2003.
The fine Pb or split float circuit was commissioned in July
2001, to increase fine Ag and Pb recovery (Torrisi and Smith,
2003). The split float circuit project comprised installation of
classification stages to separate the fine Pb from the feed, and
installation of rougher and cleaner flotation cells to separately
upgrade the fine material. Overall Ag and Pb recovery (to the Pb
concentrate) was improved by installing this circuit.
Ninth Mill Operators’ Conference
 A REVIEW OF PROCESSING AT THE CANNINGTON AG/PB/ZN MINE

PROCESSING DETAILS
Grinding
The current grinding circuit consists of two stages, an AG mill
followed by a Vertimill or tower mill; see Figure 1. This
configuration has been in operation since October 2003, when
the Vertimill and associated equipment was installed. The design
and commissioning of the Vertimill has been described by
Jankovic, Valery and Clarke (2006), and selected equipment
details are presented in Appendix 1.
The grinding circuit feed is stockpiled on the run-of-mine pad
(or ROM pad). Typically 100 000 to 300 000 tonnes of ore is
stored on the ROM pad, sorted into three ore types – hoisted,
hauled and high-grade zinc ore. The hoisted ore is crushed
underground and trucked from the shaft or hoist to the ROM pad
(a distance of approximately 1.6 km). The hoisted ore is
predominantly Southern Zone material. The hauled ore is trucked
to and crushed on the surface, and largely originates from the
Northern zone. The required blend of ore is stacked onto the
feeders using a front end loader. The three reciprocating plate
feeders are located in a tunnel below the ROM pad.
During the 2003 expansion the AG mill circuit cyclone pack
was relocated to the Vertimill circuit. A new pack containing the
larger diameter Cavex 400 cyclones was installed in closed
circuit with the AG mill. The expansion also included the
installation of a large surge or conditioning tank between the
grinding and flotation circuits (colloquially known as the ‘JA’
tank). The tank has a residence time of approximately one hour.
It is used to smooth the feed to the flotation circuit and mitigate
the effect of short grinding circuit stoppages.
Since commissioning the new grinding configuration, the feed
rate has been gradually stepped up from the pre-expansion rate of
about 300 t/h (wet), to over 450 t/h (wet). The increase has been
achieved by increasing the fraction of pebble ports, improving
the utilisation of the pebble crusher and installing an expert
control system (the Metso OCS). The downside of the
throughput increase has been an increase in the flotation feed
size, from the target 80 per cent passing size (P80) of 90 microns
to up to 170 microns (see Figure 2, which shows the average
monthly data since October 2003).
The increase in flotation feed size has resulted in a drop in Pb
and Ag recovery; the correlation between Ag recovery and P80 is
shown in Figure 3. A 40 micron increase in feed size results in
approximately a two per cent drop in Ag recovery and a one per
cent drop in Pb recovery. No correlation between Zn recovery
and feed size has been observed. The recovery drop is thought to
be largely due to a decrease in the liberation of the valuable
minerals in the feed, rather than due to a drop in flotation
residence time. Galena liberation in the feed is correlated to P80,
and finer grinding of plant feed samples in the laboratory leads to
higher Ag and Pb recoveries. No correlation between Ag mineral
liberation and P80 has been observed. The increase in mine
revenue due to the throughput increase has more than
compensated for the recovery loss.

Primary Cyclones

Pebble Secondary Cyclones

crusher

Feed

Vertimill
AG mill

FIG 1 - Grinding circuit flow sheet.

180.0 92.0
170.0
160.0
90.0

150.0
88.0
P 80 (microns)

Ag Rec. (%)

140.0
130.0 86.0
120.0
84.0
110.0
100.0 82.0
90.0
80.0 80.0
320 340 360 380 400 420 440 460 90 100 110 120 130 140 150 160 170
Feed rate (t/h, wet) P80 (microns)

FIG 2 - Feed rate versus flotation feed size. FIG 3 - Ag recovery versus flotation feed size.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 197

R ALFORD and G CLARKE

Flotation minerals (typical reagent dosages are given in 2). The collector
3418A is used to maximise Ag recovery and represents about 20
The flotation circuit comprises four sections – the prefloat per cent of the total collector dosage to the Pb circuits. Lime and
circuit, the fine Pb or split float circuit, the main Pb circuit and sodium metabisulfite (MBS) are used to reject Zn from the Pb
the Zn circuit (see Figure 4). The concentrates from the fine Pb circuit. Lime is added to the recycle water to maintain the pH in
and main Pb circuit are combined. A simplified flotation circuit the water between nine and 9.5. MBS is added to Pb cleaning
mass balance is given in Table 4. The float cells are circuits. When the prefloat circuit is offline, SEX is added to the
predominantly the trough style Outokumpu cells of varying secondary grinding circuit.
volumes (see Appendix 1 for further details). The exception is
When the circuit was expanded in 2003, the old Zn regrind
the split float circuit which uses tank cells. The original circuit
mill and cleaner circuit was reconfigured for Pb cleaning,
included flotation columns for final Pb and Zn cleaning, but
creating two identical and parallel Pb cleaning circuits. A new
these are no longer used.
Zn regrind and cleaning circuit was installed. The pre-expansion
Pb cleaning circuit comprised a stage of conventional cleaning
Prefloat Fines and a stage of column cleaning. At the time of the expansion it
Feed was thought that the columns may be required to further upgrade
the Pb concentrate, but this has not been the case and the
columns have not been used since October 2003.
Tail In the Zn circuit a small amount of copper sulfate (CuSO4) is
added to activate the sphalerite, and SEX is used as the collector.
Pb Con After the 2003 expansion a drop in Zn recovery was observed.
This was linked to changes in feed mineralogy and to the
introduction of grinding media to the grinding circuit. In 2005
the use of (relatively) high chrome media in the secondary mill
was trialled, to limit the formation of iron oxy-hydroxy species
Pb Circuit Zn Circuit on the surfaces of the valuable minerals. These iron oxy-hydroxy
species are hydrophilic and reduce the floatability of the minerals
Tail
(see Greet, Kinal and Steiner, 2005). Zn recovery improved
during the trial period, but a slight drop in Ag and Pb recovery
was observed. A permanent switch to high chrome media has
been made. The net change in revenue is slightly positive. In
addition, media consumption has been reduced.

Leaching
Cannington operates two leach circuits to remove calcium
fluoride (CaF2) from the Pb and Zn concentrates. Aluminium
Pb Con Zn Con
sulfate (AlSO4) is added to leach the calcium fluoride. Greater
FIG 4 - Flotation circuit. than 90 per cent of the calcium fluoride is dissolved under the
following conditions: the Al:F molar ratio is greater or equal to
0.5, the pH less than or equal to 3.5, and the pulp temperature is
TABLE 4 greater than or equal to 50 degrees Celsius. Sulfuric acid is used
Simplified flotation circuit mass balance. to modify the pH. The general expression for leaching of calcium
fluoride is:
Steam Grades Recovery (%) 2− 2−
CaF2 + 2Al 3 + + 3SO4 ⇔ CaSO4 + 2AlF 2 + + 2SO4
Ag (ppm) Pb (%) Zn (%) Ag Pb Zn
Feed 500 10.0 3.5 This novel (and patented) process was developed specifically
for the Cannington operation by BHP researchers.
Pb conc 3400 70.0 4.2 86.5 88.9 15.2
Each leach circuit contains two tanks that are operated in a
Zn conc 310 4.2 49.0 3.1 2.1 70.8 batch mode. Reagents are added to the thickened concentrate
Tail 63 1.09 0.60 while one tank is being filled. While this occurs, leached
concentrate is drawn from the parallel tank and filtered. The
original leach circuits contained several tanks in series, but
The prefloat circuit comprises a rougher and one stage of
excessive corrosion of these tanks necessitated the construction
cleaning. The prefloat circuit is operated only when the feed
of a new circuit or a major overhaul of the original circuit. The
contains high amounts of talc. In recent years the feed has
contained relatively little talc and the circuit has been used simplest and cheapest solution was to build the new batch
intermittently since early 2004. Some preparation and planning circuits (Leung et al, 2000).
would be required to effectively use the prefloat circuit again.
When the prefloat is used, only the frother methyl isobutyl Dewatering
methanol (MIBC) is added to the rougher. The cleaner operates High rate thickeners are used to dewater the concentrates and
as a talc reverse float. Carboxyl methyl cellulose is used to the tailing. The thickened concentrates report to the leaching
depress the talc and the Cytec collector 3418A is added to circuits. The thickened tailing is directed to either the paste plant
recover silver from the prefloat rougher concentrate. The prefloat or the tailings dam.
cleaner tail is directed to the final tail. The reverse cleaning stage Horizontal, recessed plate pressure filters are used to filter the
was implemented in CY 2000 to minimise Ag losses. leached concentrates and produce material suitable for
In broad terms the reagent operating strategies for the split transportation to the smelters. As part of the 2003 expansion
float and main Pb circuit are similar. The Cytec collector 3418A program, the number of plates in the Pb concentrate filter was
and sodium ethyl xanthate (SEX) are used to float the Ag and Pb increased from 96 to 120.

198 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 A REVIEW OF PROCESSING AT THE CANNINGTON AG/PB/ZN MINE

Paste fill preparation
The paste plant treats the thickened flotation tailing (65 per cent
solids) using a batch process, to produce fill for the underground
mine (75 to 80 per cent solids). The tailing is first filtered using
two parallel vacuum disc filters. The filter cake is then combined
with cement and water in a spiral mixer, to the required yield
stress. The cement addition is typically three to five per cent of
the solids content. The paste is distributed to the underground
stopes using a combination of holding hoppers, screw and belt
conveyors, positive displacement pumps, fully cased boreholes
and distribution pipes.
In FY 05/06 the paste plant produced 1 900 000 tonnes of
paste using approximately 58 per cent of tailing. In this year
paste production (and tailings recovery) was slightly less than the
mine requirement. In broad terms the goal is to recover two
thirds of the tailing to paste. Although progress has been made in
the recent years, the paste plant remains a bottleneck in the
overall mine and mill operation. Improvements have been made
by reconfiguring the plant to allow paste to be directed to
more than one bore line at a time, by improving mine to mill
communications and by increasing paste plant utilisation.
FUTURE DIRECTIONS AND CONCLUDING
REMARKS
The operation seeks to improve in three ways:
1. reduce the flotation feed size (and increase the liberation of
the Ag and Pb minerals);
2. reduce the cross contamination of the concentrates (by Pb
and Zn), that is, the loss of Zn to the Pb concentrate and
visa versa; and
3. increase the capability of the paste plant.
At present the Cannington metallurgy group is working on a
number of projects aimed at reducing the flotation feed size or
more precisely, increasing the liberation of the valuable minerals
in the flotation feed. These include: fine tuning of the grinding
circuit expert system, introduction of flash flotation to the
grinding circuit and bleeding low grade pebbles from the circuit.
One way to increase paste plant utilisation (and make minor
improvements in flotation feed size) is by increasing the utilisation
of the processing plant. There is scope to improve the utilisation of
the processing plant and the maintenance group have implemented
a number of projects to lift utilisation. Paste plant production has
been limited by availability of tailing in the past.
Liberation analyses of monthly plant products show that
significant amounts of liberated sphalerite report to the Pb
concentrate. Typically 70 per cent of the sphalerite in the Pb
concentrate is liberated. Similarly much of the galena that reports
to the Zn concentrate is liberated. There is an opportunity to
improve the mine’s revenue by improving Pb/Zn selectivity.
The improvement in mine output has been significant in the
last nine years. While a similar increase in output is unlikely in
the future, there remain many ways to improve performance.
ACKNOWLEDGEMENT
The authors wish to thank BHP Billiton for permission to publish
this paper.
REFERENCES
Anonymous, 2006. BHP Billiton news release to the Australian Stock
Exchange, 26 April.
Greet, C, Kinal, J and Steiner, P, 2005. Grinding media – Its effect on
pulp chemistry and flotation behaviour – Fact or fiction?, in
Proceedings Centenary of Flotation Symposium, pp 967- 972 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Jankovic, A, Valery, W and Clarke, G, 2006. Design and implementation
of an AVC grinding circuit at BHP Billiton Cannington, SAG 2006
Conference.
Leung, K, Andreatidis, J, Duffy, M and Torrisi, C, 2000. Commissioning
the Cannington project – Creating success, in Proceedings Seventh
Mill Operators’ Conference, pp 123-126 (The Australasian Institute
of Mining and Metallurgy: Melbourne).
Okely, A and Fargher, D, 2004. Cannington growth project – A partnering
approach to design and implementation, in Proceedings
Metallurgical Plant Design and Operating Strategies, pp 623-636
(The Australasian Institute of Mining and Metallurgy: Melbourne).
Smith, R, 2002. Conveying of paste backfill – Focusing on zero harm of
the environment at Cannington, in Proceedings Green Processing
Conference, pp 271-277 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
Torrisi, C and Smith, R, 2003. Bringing split float to reality – The
Cannington Experience, in Proceedings Eighth Mill Operators’
Conference, pp 73-80 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
Walters, S and Bailey, A, 1998. Geology and mineralization of the
Cannington Ag-Pb-Zn deposit: An example of Broken Hill-type
mineralization in the eastern succession, Mount Isa inlier, Australia,
Economic Geology, pp 307-329.

APPENDIX A1.1
(Selected) equipment details.

Equipment Details
Feeders Krupp reciprocating plate feeders, three units
AG mill Svedala 5.8 MW (maximum operating power approximately 5.4 MW), 8.5 m diameter × 4.5 m length.
AG mill grates 50% pebble port grates (65 mm wide) and 50% slotted grates (15 mm slots with one 75 mm pebble port on each grate)
Trommel panels 305 mm × 610 mm panels, (generally) 50% 12 × 15 mm and 50% 15 × 30 mm apertures
Pebble crusher Metso HP300
Primary cyclones Warman, 8x Cavex 400, 80 mm spigots, 140 mm vortex finders
Secondary mill Metso Vertimill 1500 (maximum operating power)
Secondary cyclones Warman Cavex 250, 19 cyclones, 65 mm spigots, 90 mm vortex finders
Split flotation primary Warman, 30x Cavex 150, 35 mm spigots, 50 mm vortex finders
cyclones
Split flotation secondary Mozley, 50 mm
cyclones (currently not in
use)
Prefloat rougher Outokumpu OK38, three cells

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 199

R ALFORD and G CLARKE

APPENDIX A1.1 (cont...)

(Selected) equipment details.

Equipment Details
Prefloat cleaner Outokumpu OK3, three cells
Split float roughers Outokumpu, three TC100 followed by four OK16 cells
Split float cleaners Outokumpu, three stages, first two TC40, second and third one TC40 each
Pb roughers Outokumpu, OK38, 11 cells
Pb cleaners Outokumpu, two parallel circuits, each three stages, first stage four OK16 cells, second stage four OK16 cells, third stage
three OK16 cells
Zn roughers Outokumpu, OK38, 11 cells
Zn cleaners Outokumpu, three stages, first stage four OK38 cells, second stage four OK38 cells, third stage three OK38 cells
Regrind mills Metso Vertimill 800 (maximum operating power approximately 600 kW), three mills – Pb A, Pb B and Zn
Regrind cyclones Krebs U6, 27 mm spigots and 38 mm vortex finders. Pb A: 10x, Pb B: 10x and Zn: 11x
Pb concentrate thickener Outokumpu Supaflo high rate thickener, 20 m diameter
Zn concentrate thickener Outokumpu Supaflo high rate thickener, 15 m diameter
Tailing thickener Outokumpu Supaflo high rate thickener, 30 m diameter
Pb filter Industrial Process Machinery, Lasta MC G (120/120 chambers)
Zn filter Industrial Process Machinery, Lasta MC Advantage (60/60 chambers)

APPENDIX A2.1
Typical reagent dosages.

Section Reagent Dosage (g/t)

Total circuit MIBC 15
Lime 300
Acid 3200
Al2O3 700
Pb circuit SEX 150
3418A 30
MBS 150
Zn circuit CuSO4 70
SEX 60

200 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

Design of a Large-Scale Concentrator for Treatment of a Copper
Skarn Orebody
J Glatthaar1, G Lane2, M Phillips3 and T Hayward4

ABSTRACT Kham copper-gold project is located in the south east corner of

In December 2004 Phu Bia Mining Ltd commissioned Ausenco
the Phu Bia contract area in Laos (Figure 1) and will be operated
International Limited to conduct a feasibility study for its Phu Kham by PBM.
copper-gold project located in Laos.
The feasibility study for a mining and processing operation treating
12 Mtpa of ore averaging 0.56 per cent copper and 0.25 g/t gold to
produce an anticipated annual average 200 000 t of concentrate
containing 25 per cent copper was completed in early 2006.
The plant design incorporates a single processing line and includes a
gyratory crusher, overland conveyor, coarse ore stockpile, SAG and ball
mill grinding circuit, flotation circuit including regrind and three stages of
cleaning, concentrate dewatering and concentrate load-out facilities.
The Phu Kham copper-gold mineralisation comprises two principal
styles: stockwork-disseminated and banded-massive sulfide within skarn
limits. Primary copper in both mineralisation styles occurs as
chalcopyrite and secondary copper sulfide minerals. Skarn includes the
copper and gold mineralisation contained in pyrite, silicate, magnetite
and haematite skarns, and in quartz-pyrite altered sediments.
Both the stockwork-disseminated and banded-massive sulfide ores
contain relatively high proportions of pyrite relative to the copper and
gold mineralisation. Thus, the flotation metallurgy of the Phu Kham
orebody is characterised by the need to reject a high proportion of the
pyrite during differential flotation to yield a concentrate grade of
approximately 25 per cent copper. The rejection of pyrite is accompanied
by the rejection of some copper and gold and can result in modest
metallurgical recovery. Consequently, the greater the pyrite content, the
lower the expected copper and gold recoveries.
The relative pyrite content is expressed on the basis of sulfur:copper
ratio. For Phu Kham ores, the sulfur:copper ratios are in the range eight to
14 on a blended annual basis, but can be greater than 20 in the skarn
components. Higher grades of copper and gold are also present in the
skarn mineralisation.
The design concepts for the concentrator were influenced by the need
to provide a low capital cost approach for the project due to the relatively
low grade of the ore and the modest metallurgical recovery.
The concepts used to minimise capital cost whilst retaining operating
and maintenance flexibility are discussed, together with the metallurgical FIG 1 - Location plan of the Phu Kham copper-gold deposit.
challenges imposed by the high pyrite mineralisation.

The deposit was discovered by a Normandy-Anglo American

INTRODUCTION joint venture with exploration between 1994 and 1997. During
this period the deposit was drill delineated to an approximate
Ownership and development depth of 300 m below surface by 64 diamond drill holes.
The Phu Kham copper-gold project is owned by Pan Australian In 2003, Phu Bia Mining Ltd (PBM) relogged the entire
Resources Limited through a Lao-registered company, Phu Bia previous drill core and compiled a geological model for the
Mining Limited (PBM). PBM has a Mineral Exploration and deposit. This work defined the deposit as the distal segment of a
Production Agreement (MEPA) with the Government of Laos. copper-gold porphyry system. An initial mineral resource
This agreement permits the company to explore for, develop and estimate was completed in 2002 using the 1994-97 drill hole
mine precious and base metals, within specific time frames, database. PBM has completed an in-fill drill program in the
within the 2595 km2 Phu Bia contract area in Laos. The Phu central zone of the deposit. A revised mineral resource and
mining reserve based on this latest drilling was completed in late
2005.
1. FAusIMM(CP), Concentrator Manager, Phu Bia Mining Limited.
Email: jong@panaustralian.com.au In December 2004, PBM commissioned Ausenco International
Limited (Ausenco) to complete a feasibility study for a mining
2. General Manager Technical Solutions, Ausenco Limited, 2404 and processing operation treating between 9 Mtpa and 12.5 Mtpa
Logan Road, Eight Mile Plains Qld 4113. of ore at a head grade of 0.6 per cent copper and 0.4 g/t gold to
Email: greg.lane@ausenco.com
produce an anticipated 250 000 tpa of copper-gold concentrate
3. Lead Process Engineer, Ausenco Limited. containing 23 - 25 per cent copper and approximately 8 g/t gold.
Email: marius.phillips@ausenco.com The concentrate was planned to be transported to markets
4. Senior Process Engineer, Ausenco Limited. predominantly in Thailand, China, India and Japan via road
Email: thomas.hayward@ausenco.com transport, ports and shipping.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 201

J GLATTHAAR et al
Ausenco managed the feasibility study with input from third
party consultants and engineers in the design of mining,
geotechnology, metallurgical test work, tailings storage facility,
high voltage power supply, environmental assessment and
concentrate marketing.
In 2006, PBM commissioned Ausenco to commence detailed
design, procurement and construction of the project. The two
13 MW grinding mills were purchased in June 2006 with the
plant scheduled for commissioning in the first quarter of 2008.
PROCESS DESIGN
The process design program consisted of metallurgical and
mineralogical studies to define the ore processing characteristics
followed by a concept definition phase where concepts that
would optimise capital cost were developed to suit the location
of the project.
The metallurgical performance was a function of the
mineralogy. The Phu Kham copper-gold mineralisation
comprises two principal styles: stockwork-disseminated and
banded-massive skarn. Primary copper in both mineralisation
styles occurs as predominantly chalcopyrite in the primary zones
and chalcocite, digenite and other secondary sulfides in the
transition ore zone. Skarn ores included the copper and gold
mineralisation contained in pyrite, silicate, magnetite and
haematite skarns, and in quartz-pyrite altered sediments. Both
the stockwork-disseminated and banded-massive skarn ores
contain relatively high proportions of pyrite relative to the copper
and gold mineralisation.
Upper zone transition ores were relatively soft with altered
copper mineralisation and complex associations with pyrite. This
resulted in higher predicted throughputs, poorer flotation
recoveries and relatively low concentrate copper grades
considering the dominant chalcocite mineralogy.
The primary ore zones were comprised of more competent ore,
resulting in lower grinding circuit throughputs and better copper
and gold recoveries in flotation.
The flow sheet selection was typical of a large copper
concentrator except for the inclusion of a third stage of cleaning
and the use of sand mill based regrind.
FIG 2 - Schematic section of Phu Kham orebody.
202 Fremantle, WA, 19 - 21 March 2007
The flow sheet is comprised of:
• ore delivery to a primary gyratory crushing;
• primary crushed ore stockpiling;
• apron feeder reclaim;
• a SABC grinding circuit;
• rougher and three stage cleaning flotation circuit;
• rougher concentrate regrind;
• concentrate thickening, filtration and load-out;
• direct flow of unthickened tailings to the tailing storage
facility (TSF); and
• recovery of process water from the tailings dam.
The plant layout was constrained by the topography. Plant site
selection involved an iterative process with the plant location
moving several times due to ground condition, proximity to local
communities, proximity to open pit mine and environmental
control considerations.
The final location allows gravity flow of tailings and all plant
run-off to the tailings dam. Thus, the environmental impact of the
project is constrained to the mine and adjacent tailings dams
areas.
OREBODY DESCRIPTION
Geology and mineralogy pertaining to metallurgy
In a simplistic sense the host rock is an altered porphyry
overlaying redbeds. The orebody consists of two zones, thrust
and block zones (Figure 2). Within these zones the ore
is comprised of stockwork-disseminated sulfide and banded-
massive sulfide mineralisation. Overlaying the rock types and
mineralisation events is secondary alteration that has resulted in:
• a copper deficient oxide layer containing gold that is
currently being heap leached (Phu Bia project),
• a mixed ore zone of oxide and transition sulfide
mineralisation and host rock (not included as ore),
Ninth Mill Operators’ Conference
 DESIGN OF A LARGE-SCALE CONCENTRATOR FOR TREATMENT OF A COPPER SKARN OREBODY

• a transition ore zone containing dominant chalcocite
(denoted ‘c’ ore),
• a lower transition zone with mixed chalcocite and
chalcopyrite (denoted ‘t’ ore), and
• a primary ore zone containing dominant chalcopyrite
(denoted ‘p’ ore).
The supergene zone is relatively small and for the purposes of
the study was incorporated in the transitional zone.
Pyrite is the dominant sulfide mineral in both the transition
and primary ore zones resulting in typical sulfur:copper grade
ratio within the economic ore zone of between ten and 12.
The term ‘skarn’ is used interchangeably to describe particular
types of mineralogy within the banded/massive mineralisation
category. The skarn ore exists as a mixture of mostly pyrite and
silica based skarns with some magnetite skarn and quartz-pyrite
altered sediments.
Disseminated stockwork mineralisation was described as
stockwork-disseminated (abbreviated to ‘SWD’) and suffixed as
either upper transition, lower transition or primary alteration
(abbreviated to ‘c’, ‘t’ or ‘p’, respectively).
Banded/massive mineralisation was described by a
differentiated ‘skarn’ type, eg sediment (abbreviated as SDM),
pyrite (abbreviated as ‘PSK’), magnetite (abbreviated as ‘MSK’)
and silicate (abbreviated as ‘SSK’). Likewise, each differentiated
skarn type is suffixed according to alteration type.
A number of rock type based and production period based
samples were assessed using either optical or automated
mineralogical analysis methods. The mineralogical characteristics
of these samples suggested that problems would be encountered
in the design of a treatment process. The copper sulfide to pyrite
mass ratios for these samples, which ranged from 1:5 to 1:14,
were appreciably greater than those displayed for most other
porphyry skarn deposits worldwide.
Secondary enriched copper sulfide minerals, chalcocite and
covellite, indicated altered mineralisation within the deposit.
Since chalcocite and covellite occurred in close proximity to
pyrite, often as rims, some activation and subsequent flotation of
pyrite is likely. The principal objective of process design was to
ensure that adequate flotation differential is maintained between
copper sulfides and pyrite in the rougher flotation stage.
The low copper sulfide to pyrite mass ratio in the samples
suggested that establishing and maintaining a sufficient
differential flotation rate between pyrite and the copper sulfides
to achieve high concentrate grades would be difficult at high
copper sulfide recovery. This indicated the need for strong
suppression of pyrite in the cleaner stages and some selectivity in
the rougher flotation stage.
Based on mineralogical investigations, a flotation feed sizing
P80 in the range of 100 to 120 µm resulted in the copper sulfide
minerals being adequately liberated from the non-sulfide gangue
host for rougher flotation. Of the unliberated copper sulfides, the
majority are locked in structurally simple binary composites with
non-sulfide gangue. A typical particle of this class contained
about 20 per cent by weight copper sulfides and the majority
should be readily captured in the rougher flotation circuit.
Assuming a rougher flotation feed sizing of about P80 106 µm,
the sizing of the rougher concentrate was expected to be
approximately P80 75 µm. These rougher concentrates require
regrinding. A regrind target of P80 of less than 38 µm is required
ahead of cleaner flotation.
Head samples based on mine production and mineralogical
classes were ground to a P80 of 106 µm and the MLA extended
liberation analysis technique used to assess the liberation of five
size fractions:
• >106 µm, 20 per cent of mass;
• <106 >38 µm, 36 per cent of mass;
• <38 >C5, 24 per cent of mass; and
• <C5, 19 per cent of mass.
The size fraction assays indicated a bias of copper deportment
to the finer fractions and non-copper sulfides (pyrite) to the <106
>38 µm fraction.
Chalcopyrite was the dominant copper mineral with minor
chalcocite, bornite and tennantite in primary ore. Transition ore
was dominated by chalcocite, with minor bornite and tennantite.
Approximately 15 per cent of the mineral suite was pyrite.
Chalcopyrite liberation in the <38 µm fractions was
83 per cent reducing to 20 per cent in the >106 µm fraction.
Bornite, chalcocite and tennantite showed poorer liberation,
indicating more difficult metallurgy in treating the transition
ores.
Principal non-Cu binary associations of chalcopyrite were with
non-sulfides (estimated at 70 to 80 per cent) and pyrite (20 to
30 per cent). Chalcocite showed a higher tendency to be
associated with pyrite in composites, again presumably due to
copper mineral alteration in the transition zone.
In summary, the MLA assessments indicated that:
• transition zone ore with higher chalcocite may have higher
Cu mineral and pyrite association and, thus, result in more
difficult flotation separation; and
• supported the optical mineralogy conclusion that a regrind to
approximately 38 µm will provide sufficient liberation to
achieve a 23 per cent Cu concentrate grade and reasonable
recovery.
Mine schedule
The mineralogy (based on ore type code) of the feed to the plant
as a function of ore schedule is summarised in Table 1.
PROCESSING REQUIREMENTS
Key test work outcomes
The principal challenge with the Phu Kham ore was the high
pyrite content and resulting high sulfur:copper grade ratio and
consequential difficulties in separating the sulfide minerals.
Pyrite must be rejected from concentrate to yield a target
concentrate grade of >23 per cent copper. The rejection of pyrite

TABLE 1
Ore type based on mine schedule.

Year 1 2 3 4 5 6 7 8 9 10 11 12
SWDc (%) 51 41 39 31 22 19 7 1 0 0 0 0
SWDt (%) 25 26 35 42 41 39 38 38 21 7 8 2
SWDp (%) 17 11 10 12 20 28 43 45 59 69 84 62
SKNc (%) 4 7 6 4 4 2 0 0 0 0 0 0
SKNt (%) 2 11 9 7 9 7 5 8 9 2 2 3
SKNp (%) 1 3 2 3 4 5 7 7 11 22 6 33

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 203

J GLATTHAAR et al

is typically associated with the rejection of some copper and gold
and can result in modest metallurgy with some pyrite skarn ores.
The feed sulfur:copper ratio for Phu Kham is in the range eight
to 14 on a blended annual basis, but may be >20 in some
locations. The greater the pyrite content, the lower the expected
copper and gold recoveries.
The metallurgical test work program for the feasibility study
was completed in five phases. Samples for the test work program
were selected from within the Central Zone of the orebody by
PBM geologists based on general criteria agreed by Ausenco and
PBM.
Phase 1 included an initial assessment of comminution and
flotation characteristics using samples selected in the basis of ore
types and preliminary oxidation profiles.
Phase 2 of the test work program was principally comminution
test work, but flotation test work was added during the study to
allow a preliminary assessment of the impact of core oxidation.
• Phase 2A was comprised of comminution test work on fresh
core from selected areas and ore types, thought to be the
most competent. These samples were to set the lower
throughput limits of comminution circuit capacity and were
sourced from new holes that twinned existing holes to enable
the use of whole core.
• Phase 2B was comprised of comminution test work on
samples selected to represent period composites split into
skarn and stockwork ore types. The samples were
composited from the general core inventory by PBM
geologists to represent the mining periods based on an
interim mine schedule.
• Phase 2C test work was conducted to compare the flotation
response of fresh core with that of the PBM core inventory
(used for the remainder of the test work program). The core
drilled for the Phase 2A samples was used to prepare fresh
(unoxidised) core samples, and the core from the existing
twin hole used to represent ‘oxidised’ core.
Phase 3 of the test work program was undertaken using
composite samples representing mining periods defined by the
interim mine schedule. The samples were composited based on
the volume defined by the pit outlines and the available core by
PMB geologists. The Phase 3 program was entirely centred on
flotation metallurgy and the preparation of samples for
engineering design test work.
Phase 4 included all test work conducted for engineering
design on flotation circuit products (ie regrind mill specific
energy) and the feed material (eg materials handling test work).
During the feasibility study it became apparent that core
oxidation was impacting on metallurgy. To evaluate the
metallurgy of fresh core, a fifth phase of metallurgy was added to
the test work program. The Phase 5 program focused on the
metallurgy of the transition zone material that was most
impacted by ex situ core oxidation.
Comminution test work outcomes
The ore characteristics for comminution circuit design were
defined as a function of ore type and mining period. In addition,
samples selected to represent the highest competency and major
ore types (at the middle and base of the proposed pit) were
assessed. The results are summarised in Table 2, Table 3 and
Table 4.
The Phase 1 samples (Table 2) were heavily weighted towards
primary ore and did not reflect the less competent material in the
mine schedule. Table 3 summarises the comminution data from
the SSKp and SWDp ore types in the primary ore and indicates
significant variation in the competency of SSKp ore, presumably
associated with rock fracture and/or the level of silicification.
Table 4 summarises the data selected on the basis of the
interim mine schedule. This reinforced the expected lower
competency of the ores to be mined early in project life and
resulted in decision to design the concentrator for a maximum
throughput of 14 Mtpa to allow for higher throughput and cash
flow during the early payback period.

TABLE 2
Summary of Phase 1 comminution results.

Sample BWI (kWh/t) RWI (kWh/t) Ai SMC test parameters

A b SG
SWDtm/SWDth 12.8 12.0 0.19 105 2.70
SWDpm/SWDph 15.5 14.1 0.44 70 2.94
SDMt 10.7 8.8 0.07 163 2.89
SDMp 14.6 16.3 0.29 53.1 3.23
PSKp 11.0 14.4 0.40 69.0 3.83
MSKp 11.8 13.9 0.29 72.8 4.21
SSKp 14.4 17.4 0.40 44.2 3.36
Mixed skarn 12.4 14.7 0.14 57.2 3.05

TABLE 3
Summary of Phase 2A comminution test results.

Sample BWI UCS (Mpa) RWI (kWh/t) Ai JK Drop Weight test parameters
(106 µm screen)
(kWh/t)
A b SG
S1-SSKp 12.1 4.2 10.2 0.12 108.7 2.71
S2-SWDp 16.3 7.1 15.4 0.13 55.9 2.70
S3-SSKp 14.2 82.7 17.4 0.42 34.1 3.12
S4-SWDp 16.2 11.4 15.4 0.04 65.5 2.74

204 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 DESIGN OF A LARGE-SCALE CONCENTRATOR FOR TREATMENT OF A COPPER SKARN OREBODY

TABLE 4
Summary of Phase 2B comminution test results.

Sample BWI (106 µm BWI (150 µm RWI (kWh/t) Ai SMC test parameters
screen) (kWh/t) screen) (kWh/t) A b SG
Stockwork transition Year 1 12.0 10.0 9.84 0.107 154 2.55
Skarn transition Year 1 11.0 10.4 8.80 0.023 111 2.64
Stockwork transition Year 2/3 12.4 10.6 10.3 0.049 151 2.73
Skarn transition Year 2/3 12.0 11.9 10.8 0.134 135 3.16
Stockwork primary Year 2/3 14.7 13.0 13.1 0.092 130 2.68
Skarn primary Year 2/3 16.6 15.9 13.3 0.182 101 2.95
Stockwork transition Year 4/7 14.8 n/a 12.3 0.109 115 2.98
Skarn transition Year 4/7 13.7 n/a 12.0 0.142 103 3.11
Stockwork primary Year 4/7 13.7 n/a 12.9 0.110 135 2.83
Skarn primary Year 4/7 12.5 n/a 10.0 0.177 161 2.90
Stockwork primary Year 8/11 15.3 n/a 16.7 0.244 63.6 2.83
Skarn primary Year 8/11 12.2 n/a 12.5 0.167 86.0 3.35

Figure 3 illustrates the trend between the Bond rod mill work Further test work was performed using further modifications to
index and the A × b parameters derived from the JK Drop Weight the reagent regime (higher pH in cleaners and modified reagent
tests and the SMC tests. addition) to improve pyrite depression. The higher cleaner pH
showed improved cleaner pyrite depression for oxidised core. For
20
primary ore samples, a 1 to 2 kg/t reduction in lime consumption
between old and new core under the high lime/xanthate scheme
18 was confirmed and the DTC collector reagent regime gave better
16 recoveries at lower operating costs due to reduced collector and
lime consumptions. The modified DTC-based flotation regime
14 with pyrite depression in the cleaner stage at moderate pH, and a
weaker frother appeared optimum.
RWI (kWh/t)

12
Table 5 summarises the metallurgy from some of the locked
10
cycle tests carried out on Phase 5 composite samples. The
8 estimated equilibrium results were calculated based on the
redistribution of the middlings streams. The tests with less than
6 Phase 1 98 per cent of metal reporting to products had a middling load
4 Phase 2A that was increasing throughout the test. The higher the transition
Phase 2B ore component, the higher the middling load indicating greater
2
deviation from the equilibrium condition.
0 Typically, the metallurgy of the Phu Kham ore improved with
0 50 100 150 200
decreasing sulfur to copper ratio and increasing primary ore
Axb content. Thus, primary ore containing low pyrite content yields
high copper recoveries, whilst skarn ores with high pyrite content
FIG 3 - Comparison of Phases 1, 2A and 2B Bond rod work in the transition zone yield low copper (and gold) recovery.
index and A × b data. Figure 4 illustrates the relationship between gold and sulfur
recovery (sulfur recovery indicating pyrite recovery) for the
selective and bulk batch flotation test approaches. There was not a
Optimum grind size significant difference between the two approaches in this regard.
The optimum grind size was determined for selected Phase 1
samples and all Phase 3 samples. The results were consistent and 100
indicated an optimum laboratory grind P80 of 106 µm. Some 90
transitional ore will benefit from a grinding circuit P80 as fine as
80
75 µm. The transitional ore is typically relatively soft and the
70
Au Recovery (%)

grinding circuit has some capacity to grind finer to realise some

of the potential advantage of a finer grind on the softer ores. 60
50

Flotation data 40
30
Flotation test work was, in the main, completed on core that had 20
been stored in the Phu Kham site core shed for periods of
10
between three and 18 months. Core sample oxidation for the
transition zone ore resulted in lower copper recoveries. A revised 0
0 20 40 60 80 100
reagent regime based on a dithionocarbamate (DTC) collector
S Recovery (%)
improved copper recovery when compared with the initial high
pH/xanthate reagent regime. However, difficulties were
experienced in achieving the same copper grades and recoveries FIG 4 - Relationship between gold and sulfur recovery in
to concentrate with old core when compared with ‘new’ core. open circuit.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 205

J GLATTHAAR et al

TABLE 5
Summary of some of the Phu Kham locked cycle test work.

Phase Sample Calculated head grade Per cent metal Copper concentrate
reporting to products predicted metallurgy
Cu (%) Au (g/t) S (%) S/Cu Cu Au Cu grade Cu rec Au rec
(%) (%) (%)
5 Chalcocite ore 0.64 0.38 8.48 13.3 70 79 22.7 58.6 37.3
5 Upper transition ore 0.37 0.26 2.91 7.9 96 96 24.0 57.5 21.3
5 Lower transition ore 0.73 0.31 7.46 10.2 92 95 26.3 80.6 39.1
5 Primary ore 0.92 0.41 8.03 8.7 100 95 23.4 85.6 56.8

Other test work
Metallurgical test work was also completed on the following:
• ore samples for materials handling and mass flow design, and
• rougher concentrate to determine regrind specific energy.
No test work was conducted on concentrate dewatering or
tailings thickening. The concentrate dewatering test work was
not conducted due to the absence of pilot plant derived
concentrates and the size distribution dependence of dewatering
circuit performance. Tailings thickening test work was not done
due to the absence of a tailings thickener in the flow sheet.
Tailings deposition and environmental test work (acid mine
drainage) was completed by the tailings dam and environmental
engineers engaged by PBM.
The overland conveyor delivers ore from the primary crusher
to the crushed ore stockpile. The conveyor is a single run
approximately 1.2 km long and 1200 mm wide. A 1350 kW
drive system runs the conveyor at 3.5 m/s for a maximum rate of
2500 t/h. The drive system consists of three drives and
accompanying gearboxes.
The conveyor is covered to limit water runback to the crushing
station during the wet season.
The coarse ore stockpile was designed as a conical structure
with two reclaim apron feeders. The cheaper option of the
installation of a single belt feeder and two vibrating feeders was
considered and rejected due to issues associated with belt
feeder design and vibrating feeder operation on potentially
‘sticky’ ore.

PROCESS DEVELOPMENT Grinding

The discussion is based on the following process plant unit
operations:
• primary crushing,
• overland conveyor and stockpile,
• grinding,
• flotation and regrind,
• concentrate dewatering and handling,
• reagents,
• tailings disposal, and
• utilities and services.
The overriding principal in the design of the plant was to
provide a fit-for-purpose low-cost process plant suitable for the
treatment of the Phu Kham ore. The in-pit value of the reserve
demanded that the design of the plant be cost effective. The
designs of some facilities reflect cost optimisation and, where
appropriate, the limitations of the process facilities are noted in
the following discussion.
A cost minimisation approach was only adopted where the
design inputs were clearly defined by test work or supporting
studies.
The flow sheet is presented in summary form in Figure 5. The
process plant site is illustrated in Figure 6.
Primary crushing, conveying and stockpile
A large ROM stockpile was originally included in the plant
layout. However, space restrictions resulted in the ROM
stockpile being relocated to inside the pit and the primary
crusher ROM pad has limited storage capacity.
The primary gyratory crusher size was optimised and selected
to take double dumps from Caterpillar 785 trucks, or their
equivalent. This allowed the elevation and cost of the crushing
station to be optimised. A 54 × 75 gyratory crusher is able to
handle 1200 mm top size at approximately 2000 t/h.
In summary, the following process was used for each
comminution test work data set.
• The total grinding circuit comminution energy was
determined using the Bond equation, a circuit efficiency
factor and the Bond rod and ball mill work indices for all ore
types. The circuit efficiency factor was ratioed to the DWI
data. The Bond ball mill work indices for a 150 µm closing
screen were calculated (where not measured) using a
correlation between the BWI data:
BWI(150 µm) = BWI(106 µm) × 0.9165
• The SAG mill specific energy was determined using an
Ausenco proprietary correlation between SAG mill specific
energy and the JK Drop Weight and SMC parameters.
• The ball mill specific energy was determined by difference
between the total circuit specific energy and the SAG mill
specific energy.
The resulting SABC circuit mill specific energy values are
summarised in Table 6.
A 13 MW variable speed SAG mill and a 13 MW ball mill
were selected for the project. Both mills have a twin pinion drive
system.
Figure 7 illustrates the mill throughput limitations based on the
selected mill sizes by year.
The grinding circuit was designed to be sufficiently flexible to
cope with the majority of the ore types identified in the
comminution test work program.
Cyclone underflow can be split between the SAG and ball
mills. This allows any deficiency in ball mill power, and the
resultant coarse grind, to be corrected through the use of excess
SAG mill capacity (see Century, Ridgeway and Telfer circuits for
example).
Pebble crushing is not required initially and installation of the
pebble crushers may be deferred until Year 3 of operation.

206 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

Ninth Mill Operators’ Conference

Future
Pebble Crushers
Primary
Crusher SAG Mill

DESIGN OF A LARGE-SCALE CONCENTRATOR FOR TREATMENT OF A COPPER SKARN OREBODY

Coarse Ore Stockpile

Cyclone Process
Cluster Water

Ball Mill
Fremantle, WA, 19 - 21 March 2007

Protection
Screen
Cyclones Feed
Pump
Rougher
Conditioner
Rougher Flotation
Concentrate
Filter
Screen

Regrind
Cyclones

ISAMill Cleaner 1 Scavenger

Trash
Screen Concentrate
Cleaner 2 Cleaner 3 Thickener
Cleaner
Screen
Conditioner
Concentrate
Storage

FIG 5 - Simplified process flow sheet.

207
J GLATTHAAR et al

Flotation and regrind • over copper selective roughing due to the potential for copper
and gold associations with pyrite that require the rougher
The flotation circuit was originally conceived as a bulk sulfide
concentrate to be reground to provide sufficient liberation
flotation circuit over 20 per cent mass recovery to rougher
and optimum recovery at target copper concentrate grade,
concentrate. This was modified to moderate (12 per cent) mass
and
recovery to rougher-scavenger concentrate with subsequent
rejection of non-copper sulfides and non-sulfide gangue in the • over bulk sulfide flotation (>20 per cent mass recovery) due
cleaning circuit. This selection was made: to difficulties in rejecting pyrite recovered to rougher
concentrate in the cleaning circuit in laboratory test work.
The compromise approach adopted used a copper and gold
mineral selective collector in roughing at moderate pH with no
additional pyrite depression and the collector dosage trimmed to
yield rougher pyrite recoveries that allowed acceptable final
concentrate quality to be achieved in cleaning using sodium
cyanide and lime as pyrite depressants.
Flash flotation was considered for Phu Kham. Flash flotation
is used in many copper/gold flotation circuits to minimise the
‘overgrinding’ of free gold in the grinding/classification circuit.
The occurrence of ‘free’ gold in Phu Kham ores appears to be
limited as flotation gold assay balances in test work were very
consistent. Flash flotation was not included in the flow sheet on
that basis. A similar approach was adopted in the design of
Ernest Henry and flash flotation has not been retrofitted to that
operation.
The OK200TC rougher flotation cells were selected, with
OK70TC cells in the first cleaner and cleaner scavenger circuit
and OK20TC cells in the second and third cleaners.
FIG 6 - Plant site in preparation, May 2006. A non-ferrous media regrind technology was selected for Phu
Kham. A single 2600 kW M10000 ISAMill® was selected on the
basis of simplicity and additional capacity, over the 1700 kW
3000 required, allowing increased regrind flexibility and potentially
higher copper and gold recoveries with transition ore types.
2500 Non-ferrous regrind technology was selected due to the
importance of rejecting pyrite in the cleaner circuit and the
2000 potential activation of pyrite when iron based media is used.
Throughput, t/h

The three-stage cleaning circuit was adopted to maximise

1500 copper concentrate grade. The circuit is a conventional open
circuit cleaner scavenger flow sheet (Figure 5).
1000
SAG Limited Concentrate dewatering and handling
500 Ball Limited
Shared Power Limited The concentrate thickener and filter were sized based on the
0
12 Mtpa feed rate using an assumed unit area thickening rate
0 2 4 6 8 10 12 based on experience with similar concentrates and a 0.7 per cent
Year
Cu head grade. Design concentrate production rates of 39 t/h for
the thickener, and 41 t/h for the concentrate filter were used. This
compares with maximum annual averages of approximately
FIG 7 - Mill throughput limitations for a SABC circuit with two 31 t/h based on 91.3 per cent utilisation for the thickener and
13 MW mills. 33 t/h based on 85 per cent utilisation for the filter.

TABLE 6
Summary of SABC circuit grinding mill specific energy values by year.

Year Specific energy (kWh/t) Ore type

SAG mill Ball mill Total SWDt (%) SWDp (%) SKNt (%) SKNp (%)
1 4.37 6.22 10.59 78.0 4.0 17.0 1.0
2 4.52 6.60 11.12 79.0 2.0 18.0 1.0
3 4.49 6.69 11.18 82.0 5.0 12.0 1.0
4 5.28 8.63 13.92 72.0 12.0 12.0 4.0
5 5.27 8.58 13.85 65.0 19.0 14.0 2.0
6 5.07 8.36 13.43 51.0 37.0 5.0 7.0
7 5.06 8.26 13.32 43.0 42.0 8.0 7.0
8 6.45 9.11 15.56 30.0 53.0 9.0 8.0
9 7.14 9.27 16.41 10.0 65.0 5.0 25.0
10 6.93 9.26 16.20 7.0 78.0 2.0 13.0
11 6.95 9.31 16.26 5.0 80.0 4.0 11.0

208 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 DESIGN OF A LARGE-SCALE CONCENTRATOR FOR TREATMENT OF A COPPER SKARN OREBODY
An IPM pressure filter was selected and discharges filter cake
to a stockpile below the filter. The filter was sized by the vendor
based on experience and test work on a concentrate from a
comparable operation.
The covered storage facility for filter cake has been minimised
to reduce the capital cost associated with the concentrate storage
structure. Space has been made available for outside concentrate
storage should availability of transport be limited from time to
time. The outside storage facility will require stockpiles of
concentrate to be sheeted temporarily to protect the stockpile
from rain and reduce dust emissions.
Reagents
The reagent’s make up and storage facilities included lime slaking
based on powdered quick lime supply, a sodium cyanide mixing
and storage system and DTC and frother, methyl isobutyl
carbinol (MIBC), storage tanks.
Tailings dewatering and disposal
Both, rougher scavenger and cleaner scavenger, tailings report to
the tailings collection hopper and directly to the tailings dam
without prior thickening.
An improvement in water quality is expected by the use of
recovered process water from the tailings dam. This is
considered important in maintaining high selectivity against
pyrite in the flotation circuit.
Sampling
The sampling system is comprised of multi-fin, multi stage static
samplers for the primary metal accounting streams. Secondary
streams, the data from which is required to define the flotation
circuit mass balance, are sampled using static dual fin launder
samplers. Tertiary streams, the data from which is required for
process monitoring, will be sampled using pressure pipe
samplers.
All the samples are pumped to a Courier on-stream-analyser
for continuous analysis and monitoring of the plant performance.
The Courier return samples are gravitated back to the process
point of origin.
The design and layout currently makes space allowance for the
future installation of a particle size analyser on the classification
cyclone overflow.
DESIGN AND LAYOUT
General
The plant site is located on a ridge (Figure 6) and slopes from the
coarse ore stockpile to the tailings discharge. The sloping site
presents some advantages due to the use of gravity flow and
reduced pumping requirements. However, the earthworks are
relatively complex to set the various relative levels and optimum
slopes required by the layout.
Primary crushing
The primary crusher is located adjacent to the pit exit to
minimise mine haulage costs. The crusher has two dump stations
and a drive in sump to allow clean out of the dump hopper. The
crusher is an open face design (eg Cadia) rather than the
in-ground variety (eg Telfer, Ernest Henry).
The crusher discharges directly to a heavy duty sacrificial
conveyor that delivers ore to the overland conveyor. An apron
feeder was not included to reduce capital cost.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
Ore conveying
The overland conveyor is designed to process a peak flow of
2750 t/h and conveys crushed ore from the primary crusher to the
crushed ore stockpile. The overland conveyor is fitted with a belt
condition monitoring system and weightometer for mass flow
measurement of ore transferred to the stockpile.
Stockpile and reclaim
The crushed ore stockpile is located just over a ridge. This allows
the stacking conveyor to project horizontally to the stacking
position.
The reclaim system consists of two apron feeders each capable
of delivering the full feed rate. The apron feeders are located in
close proximity to maximise the live capacity of the stockpile.
Whilst the reclaim chamber is constructed of concrete, ‘Armco’
style conveyor and emergency escape tunnels are provided. Dust
control is via a large mobile exhaust fan adjacent the emergency
exit tunnel.
Grinding
The layout of two twin pinion drive mills with outboard
discharge end motors is relatively complex if footprint and layout
are to be optimised. Access to the grinding floor by mobile
equipment is provided by a ramp and cutting the mill feed end
into the sloping terrain. This reduces the storage area required for
liner/lifters.
The cyclone feed pumps’ location evaluation study took some
time to complete. Numerous options were considered with all
options complicated by the location of the motor plinths or the
motors and the need to split the cyclone underflow between the
SAG and ball mills. The final solution was to put the line
overhead between the adjacent SAG and ball mill drives. Access
to the cyclone feed pumps, motors and lines and mill motors and
gear boxes is by mobile crane.
Pebble crushing
The pebble crushing circuit is yet to be detailed. The conveyors
may be installed for mill start-up to allow the trommel oversize
to be recycled and the current pebble jet return system to be
removed.
Flotation plant
The limited available area necessitated a very compact flotation
footprint. As the concentrator is being constructed on top of a
hill, the naturally steep topography was utilised to minimise the
elevation of the flotation circuit thus saving capital.
The individual flotation cells are elevated on low steel
structures that enable gravity flow of tailings away from the plant
area without the use of tailings discharge pumps.
The compact nature of the flotation circuit resulted in
restricted mobile maintenance access to the flotation circuit. As
a result a tower crane has been allowed for to provide
maintenance crane access. The selected tower crane has a 55 m
boom length with a maximum lifting capacity of 3.6 tonne when
the boom is fully extended.
Regrind
The selection of the regrind circuit was an iterative process with
a number of various regrind technologies being considered. The
feasibility test work indicated that a specific energy requirement
of between 8 kWh/t and 10 kWh/t was sufficient to produce a
product size with a P80 of 38 µm.
209
J GLATTHAAR et al
Originally it was envisaged that five Metso 350 kW SMDs
would be utilised in the Phu Kham regrind circuit. The limited
available footprint and the potential for better mineral liberation,
grade and recovery associated with finer grinding culminated in
the selection of a single 2600 kW M10000 ISAMill for the
regrind circuit.
The proposed regrind circuit will operate in open circuit with
rougher concentrate being submitted to the regrind stage, prior to
cleaner flotation. Cleaner scavenger concentrate can also be
pumped to the regrind circuit to assist with middlings liberation.
The regrind circuit is located adjacent to the flotation circuit
and maintenance access is provided by the tower crane.
Concentrate handling
The concentrate handling area is located adjacent to the flotation
circuit and is approximately four metres lower than the flotation
circuit. This area is comprised of a high rate thickener and
vertical plate pressure filter. Concentrate is stockpiled in a
storage shed and loaded to concentrate trucks by front-end
loader.
The concentrate thickener is fitted with a scum recovery
system. Froth scum is returned to the thickener feed. Thickener
overflow is collected in a separate thickener overflow tank and
pumped to the SAG mill discharge for use as mill dilution water.
APPENDIX 1
FIG A1.1 - Phu Kham process plant model as at October 2006.
210 Fremantle, WA, 19 - 21 March 2007
Maintenance access to the concentrate handling area is
provided by the site tower crane.
Tailings disposal and water management
The lack of real estate, the minimum capital approach to the
project, and the opportunity to improve the quality of the process
water resulted in the exclusion of a tailings thickener. In addition,
the positive water balance and monsoon seasons dictated that
harvesting and managing process water could be conducted at
the tailings storage facility.
The plant has two tailings streams, rougher tails and cleaner
scavenger tails. Both the rougher tails and cleaner scavenger tails
are individually sampled, prior to being combined in the final
tails collection hopper.
The fall from the plant to the tailings storage facility (TSF) is
approximately 150 m. The process plant tail streams gravitate to
the TSF via a series of break tanks in order to control the slurry
velocity and vent line pressure to atmosphere.
The tailings reclaim water is harvested using pontoon pumps,
which transfer the reclaim water to a decant water transfer tank.
As with the tailings disposal system, for the decant water
return system there is approximately 150 m of head from the
surface of the TSF to the plant. In order to minimise the pumping
head required, the decant water is returned to the process plant
via a series of water transfer stations.
Ninth Mill Operators’ Conference
Plant Expansions and Process Development at LionOre’s Lake
Johnston Operations
I Mitchell’, D Zoetbrood2, P Henderson2,S O’Neil12 and A Adem2

ABSTRACT
Treatment of nickel ore at Lake Johnston commenced in late 2001 at a rate of 250000 tpa.
Being the first production mine operation for LionOre Australia, and at that stage lacking a cash
flow, the process plant was assembled from predominantly second-hand plant but with a view
of expansion plans in the future.
Progressive expansions in 2004 to 500 000 tpa and again in 2006 to 1.5 Mtpa have occurred
to increase production, counteract falling grades and accommodate changing ore types.
The transition from a high-grade, low tonnage operation treating massive style ore to a low-
grade, high tonnage operation treating disseminated style ore has not been without its trials and
tribulations.
This presentation details the progressive exercise of installation of additional equipment and
reconfiguration of existing circuits and the impact made on productivity and metallurgical
success.
Details of the underlying driving forces, competing goals, compromises and successes
associated with the plant expansions and process development are outlined for the five-year
period over which the plant throughput has increased sixfold.

1 . MAuslMM, Manager - Metallurgy, Lionore Australia (Nickel) Ltd, Lake Johnston Operations, PO Box 906, West Perth WA 6872.
Email: ianmitchell @lionore.com.au
2. LionOre Australia (Nickel) Ltd, Lake Johnston Operations, PO Box 906, West Perth WA 6872.

Ninth Mill Operators’Conference Fremantle, WA, 19 - 21 March 2007 21 1

Maintaining the Benefit — How to Ensure Mine to Mill Continues
to Work for You
A Dance1, W Valery2, A Jankovic3, D La Rosa4 and S Esen5
ABSTRACT
Metso Minerals Process Technology Asia-Pacific (MMPT-AP) has been
working with many mining companies worldwide performing ‘Mine to
Mill’ or Process Integration and Optimisation (PIO) studies. MMPT-AP
has developed a proven methodology to improve the efficiency of the
mine-mill interface and gain maximum benefit. The PIO methodology
involves rock characterisation, mathematical modelling and simulation to
generate a list of operational and control changes for both the mine and
concentrator.
Following the methodology, the list of recommendations is developed
relatively easily; the more challenging part depends on the company.
Success of a PIO project relies on the ability of an organisation to
implement change – both operationally and culturally: the human
element.
When a PIO project is started, there is initial interest and commitment
from the company. But how can this be maintained? How does an
operation know if the benefit continues? MMPT-AP is now developing
long-term relationships with clients to ensure that the increased
production and improved efficiency continues. Direct and indirect
measurements are used to determine the key factors involved and to
understand the cause/effect relationships. Simulation tools also can be
used to compare current with recommended practices – so when material
properties vary over time, the benefits can still be quantified.
PIO is about empowerment: by measuring and understanding what
parameters affect concentrator performance, the site personnel now have
the knowledge to make a difference. Rather than simply reporting
variable or poor concentrator performance, operations can predict when
production will be affected and put controls in place to prevent it from
happening. This requires support and commitment from the company or
‘buy in’. Ensuring this buy in continues is the challenge facing every
company involved in PIO work – but the benefits are well worth it.
This paper will review MMPT-AP’s experiences in implementing PIO
projects in the long term: from both the consulting and operational
viewpoints. MMPT-AP’s involvement with a number of companies will
be discussed as examples of what works and what does not.
INTRODUCTION
Believe it or not, the term ‘Mine to Mill’ has been used in the
mining industry for over a decade with many operations dabbling
in the concept and resulting in some successes and many failures.
The idea is a simple but effective one: how can the various stages
of the mining process be aware of possible upstream or
downstream issues and work together for greater efficiency? This
typically relates to the mine-mill interface or how the mine can
produce a more suitable, higher value or higher quality
concentrator feed.
1. Manager – Process Integration and Optimisation, Metso Minerals
Process Technology Asia-Pacific, Unit 1, 8 - 10 Chapman Place,
Eagle Farm Qld 4009. Email: adrian.dance@metso.com
2. FAusIMM, General Manager, Metso Minerals Process Technology
Asia-Pacific.
3. MAusIMM, Manager – Development and Process Engineering,
Metso Minerals Process Technology Asia-Pacific.
4. Manager – Process Control and Information Engineering, Metso
Minerals Process Technology Asia-Pacific.
5. Crushing Process Technology Engineer, Metso Minerals Process
Technology Asia-Pacific.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
Of course, what higher quality mill feed means will vary from
operation to operation. In some cases it is finer fragmentation, in
others it is well blended for grade and lacking in contaminants
and it can even indicate that certain ore types are, in fact, not
profitable and should be considered mineralised waste.
Yet for such a simple concept, the application of Mine to Mill
has proven to be very difficult. The staff members of Metso
Minerals Process Technology Asia-Pacific (MMPT-AP) have
been involved in a significant portion of these efforts either as
MMPT-AP projects or in our former roles as operations and
research engineers. What have we learned over the past decade?
That the issues remain human ones and the engineering
challenges are readily overcome.
MMPT-AP has developed a proven methodology for applying
what we call Process Integration and Optimisation or PIO. PIO
reflects the fact that optimising concentrator feed goes beyond
run-of-mine (ROM) fragmentation and considers all aspects of
improving mill performance from throughput, recovery and final
concentrate grade to lower operating costs.
However, almost all of our clients are focusing their attention
on increased throughput, so the examples included in this paper
reflect that interest.
PIO METHODOLOGY
The methodology involves a number of steps: benchmarking,
rock characterisation, measurements, modelling/simulation and
where required, material tracking. A PIO project is normally
comprised of a number of site visits spaced over a few months.
The first site visit is to establish current operating practice,
initiate rock characterisation and collect measurements of blast
fragmentation and mill performance. This is followed by
modelling and simulation studies to determine how to best
exploit the hidden inefficiencies. These recommendations are
then followed by further site visits to implement the changes,
monitor the results and ensure the improvements are maintained
over time.
Benchmarking and process audits
The first step of a PIO project is to benchmark the current
practices by auditing the operation and control of the blasting,
crushing, grinding and flotation processes.
The quality of blast pattern implementation is assessed and the
resulting ROM fragmentation measured using image analysis.
The crushing, grinding and flotation circuits are surveyed and
process control strategies reviewed. All of these measurements
allow mathematical models to be developed for the complete
process chain. These models are later used to simulate the impact
of operational changes in the mine or concentrator on the entire
process.
Rock characterisation
Once the current operating performance has been measured
under one set of conditions, the effect of changing rock
properties can be quantified. This involves rock characterisation
or defining domains of similar properties. In this paper, this will
be discussed in terms of blast fragmentation. The process could
equally be applied to other quality parameters such as flotation
performance, leach recovery, lump to fines ratio, etc.
215
A DANCE et al
The MMPT-AP methodology for rock characterisation utilises
simple and inexpensive measurements that can be performed by
trained site personnel. Quite often the measurements are already
being collected by the operation. The advantage of simple
measurements is the amount of data that can be collected in a
very short time frame, as the samples do not require shipping to
an outside laboratory. When attempting to characterise an entire
orebody, the density of data is very important.
For rock characterisation, MMPT-AP use measurements of
rock strength (point load index, PLI and/or UCS) and rock
structure (rock quality designation, RQD and/or fracture
frequency). Both PLI and RQD measurements can be taken on
drill core and point load tests can also be performed on irregular
shaped samples of material.
The PLI value can be correlated to unconfined compressive
strength (UCS) as well as the JKMRC drop weight test
parameters A and b. The drop weight parameters are necessary in
order to model the crushing and grinding circuits. Therefore, the
use of the point load index allows sites to characterise their rock
properties quickly and easily while still making use of the
sophisticated grinding models that are available.
The rock structure is represented by the RQD value that
indicates the fracture frequency present in the drill core. This
measurement is routinely taken at operations for geotechnical
purposes but has been shown to be very useful in blast
fragmentation modelling in the absence of detailed rock mass
structure mapping.
Once the PLI and RQD data are available, the range of rock
properties are mapped out and domains are defined (see
Figure 1). Within each domain, the material will behave similarly
under the same blast conditions while all of the domains cover
the complete range of rock properties that are present.
The domain structure shown in Figure 1 follows the existing
ore type characterisation used by the site but expands further into
areas of structure (coarse, medium and fine) and strength (soft,
medium and hard). The ranges of strength and structure used are
based on the variability of the orebody. The more variable the
PLI and RQD values measured in the orebody, the greater
definition required for domains. In the example given in
Figure 1, the RQD values were divided into ranges of zero to
30 per cent, 30 - 60 per cent and >60 per cent while the PLI
values were divided into 0 - 3 MPa, 3 - 6 MPa and >6 MPa.
Once the domains have been defined, different blasting
practices, crushing and milling operational strategies are
established. Through modelling and simulation studies, the
impact of blending different domains can be reviewed. Most
PLI 0 - 3
RQD 0 - 30
Volcanic RQD 30 - 60
RQD >60
RQD 0 - 30
Diorite
RQD 30 - 60
RQD >60
RQD 0 - 30
Intermediate
RQD 30 - 60
Tonalite
RQD >60
RQD 0 - 30
Young
RQD 30 - 60
Tonalite
RQD >60
FIG 1 - Example of blasting domains (Dance et al, 2006).
216 Fremantle, WA, 19 - 21 March 2007
importantly, as the rock properties have now been well
characterised and the processes modelled, the variable nature of
the material can now be compensated for.
Modelling and simulation
The measurements collected while at site are combined with the
rock characterisation domains to model the complete process
chain. MMPT-AP use these data to develop site-specific models
of blast fragmentation, crushing, grinding and flotation. This
allows customised blast patterns to be developed that optimise
both crushing and grinding performance. For each domain, blast
designs are recommended to generate the optimal fragmentation
size for downstream processes. This may involve an increase or
decrease in energy level (or powder factor) depending on the
rock characteristics of each domain.
The objective is to minimise the overall cost for the entire
process by distributing the energy required sensibly and
effectively where it is best applied. Near-field vibration
measurements and models are used to confirm that pit wall
stability issues are considered in the blast designs.
In addition, the crushing and grinding models allow the impact
of operational and control strategies to be investigated. For
example, what is the best closed-side-setting to operate my
primary crusher at in terms of production and product size? What
target load should I use in my SAG mill when processing this
domain? What is the tendency for this material to be SAG mill,
ball mill or recycle crusher limited?
All of these questions can be evaluated using the model of all
the stages of comminution (blasting, crushing and grinding).
BENEFITS OF PIO TO AN OPERATION
It is quite extraordinary how the mining industry has allowed
uncertainty and unpredictability to be an accepted part of normal
operation. It is still common for the concentrator to be unaware
of changes in material properties and how they influence mill
performance. In addition, a lack of accurate material tracking so
that prediction of when these changes will happen is very
unlikely.
This perhaps could be understood before the application of
technologies such as online image analysis, GPS equipment
tracking and sophisticated geological modelling software.
However, these tools are now readily available and can answer so
many outstanding questions: ‘Why is the mill tonnage down?’,
‘How long will it last?’ and more importantly, ‘What can be done
to prevent it in the future?’
PLI 3 - 6 PLI >6
V-FS V-FH
V-CS
V-MM V-MH
V-CM V-CH
D-FS D-FM
D-CS D-CM D-CH
IT-MM
IT-MH
IT-CH
YT-CH
Ninth Mill Operators’ Conference
 MAINTAINING THE BENEFIT — HOW TO ENSURE MINE TO MILL CONTINUES TO WORK FOR YOU
Time and time again, the explanation of poor performance is
related to ‘bad ore’ without any measurements to back it up. We
as ‘metal manufacturers’ operate in a manner that other
manufacturing sectors could never do – and yet, we still manage
to make a profit (particularly in times of high metal prices).
Imagine the increase in efficiency possible through the
understanding of what material parameters are important to
concentrator performance and the accurate measurement,
tracking and control of those parameters. This is achievable by
every operation worldwide.
Performance improvements
Needless to say, the performance improvements experienced as a
result of PIO projects have been significant; typically, five to
20 per cent increases in throughput through controlled blasting
and crushing practices. In fact, these improvements can occur
without any increase in operating cost but instead, through the
understanding of where to add energy and where to reduce
energy. In one case, the concentrator experienced a throughput
increase of 33 per cent at no added cost.
When finer ROM fragmentation is required through higher-
energy and controlled blasting, quite often the benefits solely in
the mine outweigh any added blasting costs. That is, higher
equipment availability, better digging rates, lower haulage costs,
improved crusher production … the list goes on. With the current
mining and crushing methods, finer material tends to be easier
and cheaper to handle.
Of course, the benefits downstream in the concentrator are
magnified through consistent, high quality feed that leads
to a stable and more predictable performance. In the PIO
methodology discussion above, the definition of blasting
domains allows different blasting and crushing strategies to be
developed. The reason is that SAG and AG mills are no different
from the rod and ball mill circuits operated in the past – all
grinding circuits benefit from consistent and prepared feed
material. After compensating for the harder/softer and blockier/
finer nature of the material, the end product through blasting and
crushing is more consistent – leading to steady and predictable
concentrator performance.
What is the value to an operation to say with confidence: ‘We
know what the mill tonnage will be tomorrow, next week, next
year and until the end of mine life as we can control it.’?
Introduction of measurements
Before concentrator performance can be controlled, the
important material properties affecting mill performance must
first be defined and more importantly, measured. In this paper,
the discussion will focus on mill throughput and be limited to the
60
50
Weight % in Fraction
40
30
20
10
0 0
FIG 2 - Effect of feed size on SAG mill tonnage (Dance, 2001).
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
effects of material hardness and size. It is very important to
isolate the two as they can often be confused. In addition,
measurement of hardness and size must be done objectively and
if possible, automatically. At one operation, the changes in mill
tonnage observed were related to rock hardness – high tonnage
was ‘soft’ ore and low tonnage was ‘hard’ ore; and of course,
there was nothing that could be done about that. After installing
an online image analysis system and measuring rock hardness
independently, it was determined that hardness had almost no
effect – the changes in mill tonnage were due almost entirely to
feed size. An example of how feed size and the amount of ‘fines’
affected mill tonnage is shown in Figure 2 over a 24-hour period.
It can be seen in Figure 2 that feed size controlled mill
behaviour – more fines, higher tonnage. There was almost no
effect of hardness for this operation. As far as the mill was
concerned, if it was the same size, it processed at the same rate.
Actually measuring material properties – online and
automatically – led to an understanding of what was actually
important. In addition, it provided greater predictability. It was
found that X per cent fines would always lead to Y throughput.
In other words, prepare the feed material and the mill
performance was entirely predictable.
This is very empowering. Instead of a situation where
concentrator performance was changing due to an abstract
hardness factor, it was changing due to material size – that is
entirely under the control of the operation. In other words,
Mother Nature makes the orebody hard or soft, but blasting and
crushing practices make the material big or small. (Stockpile
design and operation can moderate these effects somewhat.) By
knowing how feed size influenced mill performance, it was then
possible to regulate the feed size to achieve the highest possible
mill throughput – every day of the year.
Using MMPT-AP’s PIO methodology detailed above, it is also
possible to predict mill performance until the end of mine life
based on drill core measurements of PLI and RQD with models
of blast fragmentation, crushing and milling circuits linked
together. Such predictions of mill tonnage can be entered into
every ore block of the mine planning software to create a
geometallurgical model.
Methods to monitor and control
As described above, measurements are the key to understanding
what material properties are important to concentrator
performance and in what priority. It is typical that rock
hardness and feed size are the two key parameters affecting
concentrator throughput; however, other factors such as clay
content and specific gravity may also have an effect.
1800
1600
1400
Mill Tonnage (t/h)
1200
% Fines
1000 % Medium
% Coarse
Mill t/h
800
600
400
200
217
A DANCE et al
Material hardness can be measured in a variety of ways. For
blast domain definition, MMPT-AP relies on point load index
and the ability to correlate the PLI values with drop weight test
parameters. This is very useful for simulation as the models
require estimates of rock strength and breakage characteristics.
For relative comparison, other methods can be used such as
drilling penetration rates and blastability indices. What is
important is that the values used are reproducible and not
subjective (that is, potentially prejudiced or biased).
ROM fragmentation size measurements are now possible using
commercial image analysis systems that have been around for
the past ten years. The strengths and weaknesses of these
systems have been well documented and, provided the user
understands them, the systems provide reliable and accurate
data. (Figure 2 is a good example of applied image analysis
measurements.)
As most operations blend different ore sources in their mill
feed, an understanding of where the material originates is
absolutely essential. This allows the actual concentrator
performance on this material to be reported back into the
geological block model for comparison with any estimates or
prediction of similar material nearby in the future. Material
tracking is a necessity for geometallurgical modelling.
There are currently two methods to track material movements
from the mine to the concentrator: model-based and sacrificial
instruments.
The first method of tracking material involves the development
of a software program to record the movements of material from
the open pit or underground to the intermediate or long-term
stockpiles, through the crusher and coarse ore piles and into the
concentrator. Each stockpile can be represented by simple perfect
mixing models or if necessary, more sophisticated three-
dimensional models.
The models allow the effect of material mixing and delays to
be incorporated and provide a reasonably accurate estimate of
mill feed. Such a system provides much greater definition or
detail on changes in concentrator feed and can be updated as
frequently as every 15 minutes. A daily summary will not
provide such a degree of detail.
Another method for tracking material movements being
employed by MMPT-AP are ore block markers called SmartTags.
MMPT-AP has developed passive radio frequency (RF)
POST BLAST
[ID,x,y,z]
RECEPTOR
SAG MILL FEED
FIG 3 - Use of radio frequency tags to track material movements (Dance et al, 2006).
218 Fremantle, WA, 19 - 21 March 2007
transponders for use in blasted material monitoring. These RF
tags are small, robust and inexpensive and can be dropped into
the blasthole stemming column or placed on the muck pile
surface post-blast (see Figure 3).
The tags are not powered but are detected by antennas placed
over conveyor belts. Each tag has a unique identifying number
that the antenna transmits to a remote computer for recording
along with the date/time. By noting the initial position of each
tag (ie blasthole ID), an estimate of the origin of the material
being processed can be made. By tracking the actual material
itself, concerns about estimating stockpile volumes, mixing and
retention times can be avoided.
Throughput forecasting and geometallurgical
modelling
The measurement and understanding of how rock hardness and
size affect mill performance is needed to optimise ROM material
properties on a daily basis. Operations personnel can adjust the
blast pattern, explosive distribution and primary crusher operation
in order to achieve a consistent feed size to the mill. If necessary,
harder and softer ore types can be blended or campaigned
separately under different grinding circuit operating conditions.
The essential feature of this scenario is the ability to predict
how the material will perform. That includes the situation when
the material was not treated correctly – that is, when tailored
blasting and crushing practices were not followed (as can happen
at times).
The ability to predict concentrator performance under different
operating conditions and well into the future allows different
mine plans to be investigated. For example, what will be the
demand on trucks and shovels if an operation decides to reduce
the blast energy? What will happen over time as the material
properties change? This is the aim of throughput forecasting and
geometallurgical modelling.
After defining representative blasting domains, MMPT-AP
simulate the impact of different rock properties and operating
conditions using the customised models of blasting, crushing and
grinding. Mapping the orebody using rock strength and structure
measurements allows the mill throughput to be simulated as far
into the orebody as the drill core has penetrated. An example of
this is shown in Figure 4.
POST CRUSHER
DETECTOR
[ID, time]
DETECTOR
[ID, time)
Ninth Mill Operators’ Conference
 MAINTAINING THE BENEFIT — HOW TO ENSURE MINE TO MILL CONTINUES TO WORK FOR YOU

In Figure 4, point load index and RQD values are plotted
against mine elevation. It became very clear that in this case,
rock strength and structure would vary quickly between 800 and
650 m in elevation and then stabilise. In the early stages of the
operation, blasting energy could be minimised but in order to
sustain expected mill throughputs, would need to increase as the
mine deepened. In other words, the initial blast patterns would
not adequately fragment the material to maintain mill tonnage
and adjustments would need to be made. In addition, simulations
indicated that additional mill power would need to be installed in
order to keep the final grind size from coarsening.
The data presented in Figure 4 – along with a detailed mine
plan and blast conditions – can be used to generate a trend of
expected mill throughput and final grind size. This allows
120
100
different mine plans and blasting scenarios to be simulated
quickly and easily.
Figure 5 shows an example of throughput forecasting where
daily concentrator tonnages over a three and a half year period
are plotted as blue diamonds. For this operation, throughput
varied from below 4000 tph to over 7000 tph due to a wide range
of ore strengths and structures mapped by PLI and RQD values.
MMPT-AP developed a throughput model based on rock
strength and structure as well as blasting conditions. (In this case,
ore grade was also used to account for hardness variations not
picked up by the PLI results.) Material tracking was necessary in
order to estimate the daily distribution of mill feed from the
different ore sources as well as assign PLI and RQD values to
each source.
12
RQD
RQD
Is50
10

Point Load Index, Is50 (MPa)

80 8
RQD(%)

60 6

40 4

20 2

0 0
0 100 200 300 400 500 600 700 800 900
Mine Elevation (m)

FIG 4 - Rock strength and structure with mine elevation (Dance et al, 2006).

FIG 5 - Daily throughput predictions including operational envelope.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 219

A DANCE et al
The model predictions are shown in Figure 5 as an upper
and lower ‘operational envelope’. The operational envelope
represents small changes in the input parameters that would
result in higher or lower expected mill tonnage. For example, for
the upper boundary, the rock properties were softened and the
blasting energy increased within an expected variation. The
reverse was done for the lower boundary. Within these expected
variations for normal operation, the predicted concentrator
tonnage varied by 800 tph. Over the three and a half year period,
the average relative error for the model was 0.84 per cent or
±46 tph.
HOW TO ENSURE SUCCESS
Over the past decade, MMPT-AP personnel have been involved
in many Mine to Mill or PIO projects – some successful and
some not. In every case, issues were identified and solutions
presented to the operation that represented significant
improvements in throughput or efficiency. Following a
prescribed methodology, the engineering solutions fall out and
become clear very quickly. Whether or not the operation decides
to implement the recommendations is typically the first hurdle
as, at times, the solutions are a step-change from current
operating practices. The second challenge is sustained
implementation over time. When MMPT-AP has been involved
in successful projects, a number of common traits were in place:
management support and the presence of a site champion,
assistance from an outside agency and ownership of the situation
through changes in key performance indicators.
Site champion and management support
The selection of a site champion is a bit of a misnomer – a true
champion cannot be selected, they must take it upon themselves
to support the project wholeheartedly despite the complaints and
resistance of others. They cannot be told to be the champion.
They must also possess a combination of technical skills (to be
able to explain the benefits) and persuasive skills (in order to
negotiate with the parties involved). It is all about dealing with
cultural change and how the key personnel handle it. The
champion must also have sufficient seniority (or ready access to
seniority) so that they cannot be ignored. An easier option is for
managers to voice their total support of the work and agree
completely with the decisions of the champion. For the work
environment is not a democracy and ultimately whatever the boss
says, goes.
Initial expectations are that if the message is made clear and
the benefits demonstrated, people will decide themselves to
adopt the new practices and support the project. Unfortunately,
this is rarely the case and the voice of senior management
ultimately needs to ring resoundedly in everyone’s ears that
voluntary support is not an option. The economic benefits of PIO
projects can gain the acceptance of senior personnel as they see
more of the ‘big picture’ than others do. However, commitment
or ‘buy in’ from workers on the grinding floor is unlikely unless
financial incentives are put in place (see below).
Outside assistance
Despite the greatest intentions, PIO efforts rarely succeed
without some level of outside assistance or agency. This is
because generally in-house projects do not attract enough
attention from upper management. The advantage of dealing
with a consultancy group experienced in this field is the
confidence their personnel bring with them to site. They have all
done this before and can quickly identify where the benefits can
be made and put systems in place to measure the gains.
You can almost measure the growing level of interest while the
group are at site – and feel the collective sigh of relief when the
consultants depart to generate the report. This is where the
220 Fremantle, WA, 19 - 21 March 2007
champion comes into play: ‘What can we do straight away based
on the comments and conclusions made by the consultants?’ and
‘Why are we waiting for the report? Why don’t we act
immediately?’
In order to sustain a level of interest at site, regular
maintenance visits need to be scheduled by the outside group.
This is not only to ensure the benefits continue, but also to
motivate people despite the belief that ‘The ore has changed
since you were here last’.
Key performance indicators
To reinforce the changes required, key performance indicators
(KPIs) – or how senior staff are measured – need to be aligned
with the project’s objectives. That is, KPIs for the mine
personnel need to include the quality parameters identified at the
start of the project. This is where measurement of these
properties is so important. For example, if finer fragmentation of
certain ore types is found to be vital for concentrator
performance, online measurement of ROM fragmentation size
and the monitoring of these measurements can be used to set up a
KPI and a success rate. Of course, it is equally important when
setting up this KPI to quantify what is expected from the mine
and not deviate over time (or ‘move the goal posts’). It cannot be
described as ‘good’ fragmentation but rather ‘a P80 of no greater
than…’ as measured by an automated system.
An interesting concept is establishing a contract between the
mine and the concentrator as done between the concentrator and
their customer: a smelter or refinery. In this contract, the
concentrator must clearly specify what the quality and quantity
requirements are and, in the event of not meeting these
specifications, the penalties involved. It would certainly change
the economics of a mine if it had to sell its product to the
concentrator rather than simply send whatever is available –
worth considering no matter how radical it may sound.
The mine needs to truly believe that the important parameters
for downstream performance are under their control and that they
can satisfy the agreed-to KPI and perhaps even exceed it. If they
do not consider it important, then the project is likely to fail.
Financial incentives
However crass this may sound, if an operation is to benefit
through change then it is sensible to let the employees share in
the wealth created. In general, there is no greater motivator than
money. This can be as simple as establishing bonuses based on
KPIs related to mill feed quality (once again, well defined). This
will raise the importance of such KPIs as there is a financial
incentive.
MMPT-AP has observed the approach of a general workforce
gainsharing program and unfortunately, it was not successful.
When someone does not truly believe they can make a
difference, it is hard to motivate them to implement change.
Work ethic is one thing, but implementing change in the face of
peer pressure is another; particularly in a unionised environment.
However, monetary bonuses targeting the key players in the mine
and concentrator are a persuasive tool.
Regular audits/benchmarking
In all the PIO projects that MMPT-AP has been involved with,
most implement the necessary changes and immediate successes
are gained, a few continue over time but it is rare that the benefits
are maintained in the long term. To address this, MMPT-AP is
now strongly recommending that regular site visits be agreed to
at the outset of a project.
One of the reasons for the long-term difficulties is continuity.
A champion involved with a successful PIO project can be
rewarded with a promotion or a position at another operation.
Ninth Mill Operators’ Conference
 MAINTAINING THE BENEFIT — HOW TO ENSURE MINE TO MILL CONTINUES TO WORK FOR YOU

Alternatively, they may simply decide to move on. What
continuity is in place to ensure the benefits are sustained? The
outside consultancy – with their standard operating protocol –
can provide that continuity. MMPT-AP are working more and
more with operations establishing support contracts with a
regular number of site visits per year. Perhaps this will increase
the success rate in the long-term of PIO projects.
During these site visits, process audits and benchmarking will
reveal if the systems put in place (ie blast implementation,
crusher operation) are still working. Has the material changed
enough in its characteristics to warrant further changes? The
modelling/simulation tools can be used again to verify if the
benefits have degraded or if opportunities lie elsewhere. Finally,
it revives interest in the project and maintains communication
between all parties.
8000
7000
Concentrator Tonnage (t h)
/
As an example, the data presented in Figures 6 and 7 cover the
period from 2004 to mid-2006 for an operation that was involved
in a PIO project in late 2003. In both figures, the daily
concentrator tonnage, average mill feed rock strength (RQD
value) and blasting powder factor (in kg/t) are shown. In late
2003, it was identified that rock strength played a significant role
in blast fragmentation and to achieve a more consistent ROM
size, blasting patterns should be adjusted to match the RQD
value. In Figure 6 for 2004, it is clear that powder factor
followed RQD very closely and, as a consequence, mill
throughput was reasonably steady.
Therefore, a PIO project was successfully implemented with
blast patterns modified to suit rock conditions and a consistent
mill feed size resulting in a steady and higher mill tonnage.
1.4
Mill Tonnage
Powder Factor
RQD
1.2

6000

Powder Factor (kg/t), RQD

1.0

5000
0.8

4000

0.6
3000

0.4
2000

0.2
1000

0 0.0
Apr-04

May-04

Aug-04
Dec-03

Jan-04

Mar-04

Jun-04

Jul-04

Sep-04

Nov-04

Dec-04

Jan-05
Feb-04

Oct-04

FIG 6 - Mill tonnage, powder factor and RQD values – 2004.

8000 1.4
Tonnage
Powder Factor
7000 RQD 1.2
Concentrator Tonnage (t / h)

Powder Factor (kg/t), RQD

6000
1.0

5000
0.8

4000

0.6
3000

0.4
2000

0.2
1000

0 0.0
Apr-05

Aug-05

Nov-05

Apr-06
Dec-04

Jan-05

May-05

Sep-05

Dec-05

Jan-06

May-06
Mar-05

Jun-05

Jul-05

Oct-05

Feb-06

Mar-06

Jun-06

FIG 7 - Mill tonnage, powder factor and RQD values – 2005 and 2006.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 221

A DANCE et al
The data in Figure 7 shows that in late October-early November
2005, the rock structure became consistently blockier and steadily
worsened to the end of 2005 and into the middle of 2006. As the
material became more difficult to fragment effectively using the
same blast patterns, concentrator throughput was affected.
This is the point where regular audits are useful. With the
material becoming more difficult to process, what is the
decision? Should further changes to blasting and crushing
practices be trialled or not? With measurements of how much the
material properties are changing and how long the conditions
will continue (based on drill core data), a decision can be made
that makes economic sense for the entire operation. For the
personnel involved, where are the KPIs? Where are the
individual financial incentives to investigate this problem and to
implement further changes?
The ultimate objective of any Mine to Mill exercise is for the
mine and concentrator to communicate with one another and
better appreciate each other’s problems and requirements. Also,
to use the economics of the entire operation as the basis for any
decisions involving operating practices. With the effect of
reducing powder factor for this blockier material demonstrated
on the concentrator throughput, there might be scope for
alternative blasting designs and/or crushing practices.
SUMMARY
The concept of Mine to Mill has been applied in the mining
industry now for over a decade with some successes and many
failures. MMPT-AP has been involved with many of these
projects and developed a proven methodology called process
integration and optimisation or PIO. The methodology involves
benchmarking, rock characterisation, measurements, modelling/
simulation and where required, material tracking. The rock
222 Fremantle, WA, 19 - 21 March 2007
characterisation step defines blasting domains and allows
different blasting and crushing strategies to be developed.
PIO is about empowerment: by measuring and understanding
what parameters affect concentrator performance, site personnel
now have the knowledge to make a difference. Following a
prescribed methodology, the engineering solutions are quickly
identified while the human issues are more difficult to overcome.
When MMPT-AP has been involved in successful projects, a
number of common traits were in place: management support
and the presence of a site champion, assistance from an outside
agency and ownership of the situation through changes in key
performance indicators.
The ultimate objective of any Mine to Mill exercise is for the
mine and concentrator to communicate with one another and
better appreciate each other’s problems and requirements. Also,
to use the economics of the entire operation as the basis for any
decisions involving operating practices.
REFERENCES
Dance, A, 2001. The importance of primary crushing in mill feed size
optimisation, in Proceedings SAG 2001, Vancouver, Canada.
Dance, A, Valery, W, Jankovic, A, La Rosa, D and Esen, S, 2006. Higher
productivity through cooperative effort: A method of revealing and
correcting hidden operating inefficiencies, in Proceedings SAG 2006,
Vancouver, Canada.
Renner, D, La Rosa, D, DeKlerk, W, Valery, W, Sampson, P, Noi, S B and
Jankovic, A, 2006. AngloGold Ashanti Iduapriem mining and
milling process integration and optimisation, in Proceedings SAG
2006, Vancouver, Canada.
Tondo, L A, Valery, W, Peroni, R, La Rosa, D, Silva, A, Jankovic, A and
Colacioppo, J, 2006. Kinross’ Rio Paracatu Mineração (RPM)
mining and milling optimisation of the existing and new SAG mill
circuit, in Proceedings SAG 2006, Vancouver, Canada.
Ninth Mill Operators’ Conference
Improving Fine Copper and Gold Flotation Recovery — A Plant
Evaluation
T Rivett1, G Wood2 and B Lumsden3
ABSTRACT
A statistical ‘randomised block design’ full plant trial of magnetic
conditioning of flotation feed was undertaken at Rio Tinto’s Northparkes
Mine in New South Wales. Magnetic conditioning of flotation feed has
been shown to improve the flotation of <38 µm paramagnetic minerals
consistent with selectively aggregating these paramagnetic minerals. The
process was evaluated using shift composite samples that were analysed
for copper and gold. Because the process targets <38 µm minerals, all
samples were sized and the size by size recoveries were calculated for
each sample. The size by size analysis targeted the effect on fine mineral,
reduced the plant noise and shortened the test period required to get
statistically significant results. An increase of gold and copper recovery
was found in the <20 µm fraction to a high level of statistical
significance. For the >20 <38 µm size range the copper recovery
increased to a high level of significance, whereas the increase in gold
recovery in this fraction was to a lower statistical significance. There was
no statistical improvement in the >38 µm fraction for either metal.
Approximately half the copper and gold losses at Northparkes are in the
<20 µm size fraction. While the sulfide copper minerals chalcopyrite and
bornite are known to be paramagnetic; gold is not paramagnetic. Gold’s
response to magnetic conditioning, however, is consistent with a number
of literature references where, depending on the gold’s mineralogical
disposition, gold has been shown to respond to magnetic treatment.
INTRODUCTION
The operational efficiency of all mineral separation processes is
within certain limits. These limits can be reduced and the
efficiency increased by the introduction of new technology that
enhances the mineral separation process. These enhancements
may be expanding the parameters over which the process can be
applied, reducing the cost of the process or by improving the
efficiency of separation within existing parameters. It is not
enough for the operation of a commercial plant that the
efficiency of the mineral separation technology can be enhanced
in a laboratory, or only be shown to do so in a laboratory. For a
new technology to be practical the efficiency must be enhanced
in a way that is applicable to commercial mineral processing
operations. Applicability to commercial mineral processing
operations means that the technology can be practically applied
to existing operations, that the improvement can be clearly
demonstrated in the operation and that it is economically and
operationally beneficial.
Furthermore, a mill operator must be able to evaluate a new
technology in a commercial plant with all the normal background
variability associated with hour to hour operations of the plant
but not related to the test (commonly termed noise), that occurs
in such a plant and be able to measure or see that the benefit is
real. Given the efficiency of modern mineral separation
technologies it is unrealistic to expect that a new technology is
going to give a benefit so large that it is immediately obvious.
Rather the benefit is going to be relatively small and at the
margins of the operation and only apparent to the operator after a
1. Process Control Engineer, Northparkes Mines, PO Box 995, Parkes
NSW 2870. Email: thomas.rivett@riotinto.com
2. Plant Metallurgist, Northparkes Mines, PO Box 995, Parkes NSW
2870. Email: glenn.wood@riotinto.com
3. Centre for Multiphase Processes, University of Newcastle,
University Drive, Callaghan NSW 2308.
Email: BarryLumsden@compuserve.com
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
period of statistical testing. Nevertheless, such a benefit, while
only measured in single digit percentage points will be of
immense economic benefit and valued at many millions of
dollars of revenue, easily justifying the effort of the evaluation.
Recovering fine sulfide minerals by flotation
Froth flotation of minerals is a technology that has been used
commercially for around 100 years. Over the 100 years,
significant resources have been expended on new technologies to
improve its efficiency. Among these new technologies have been
chemically based technologies, mechanically based technologies
and technologies devoted to the operational control of the
process. These new technologies have improved the efficiency of
froth flotation significantly.
However, one of the major limitations in froth flotation
remains the efficient, selective recovery of minerals less than
10 µm in size. (Trahar, 1981; Trahar and Warren, 1976). These
<10 µm valuable minerals continue to present a challenge to
operators of froth flotation plants. Improving the efficiency of
separation of these <10 µm particles offer the operator of a froth
flotation plant the opportunity to enhance the economic viability
of their operation. Moreover, all froth flotation plants, even those
that grind to a coarse size range inevitably find that a large
component of their losses are these <10 µm particles.
It is generally accepted that the poor recovery and efficiency of
separation of <10 µm sulfide particles is only due to their size. It
is not chemical but purely physical, and due to their poor
collision efficiency with the air bubble (Trahar, 1981).
Many and various strategies and technologies have been
investigated over the years to improve the recovery of fine sulfide
mineral. These efforts have had some success. Fuerstenau,
Chander and Abouzeid (1979) outline in detail some of the
different strategies and technologies that have been investigated
to improve the recovery of fine mineral. The strategies and
technologies that have been investigated have been both
chemical and mechanical. One method that has been investigated
has been to aggregate the fine mineral. Sivomohan (1990) details
some of the aggregation processes investigated, which have
included mechanical methods such as high intensity conditioning
(shear flocculation) and chemical aggregation methods such as
those using flocculants or salts.
Some of the important problems that fine particle aggregation
methods have had to overcome to be utilised successfully in
industrial plants is; their lack of selectivity, their large capital and
operating costs and the fragility of the aggregates in industrial
flotation plants. Failure of an aggregation method to overcome
any one of these barriers results in the method being unsuitable
in an industrial plant.
Of the aggregation methods in the literature, none appear to
have received widespread acceptance in commercial flotation
operations.
Magnetic aggregation of fine paramagnetic
minerals
That magnetic particles will aggregate when magnetised has
been known for decades. However, it was only with the
development of very high strength magnetic fields that the
223
T RIVETT, G WOOD and B LUMSDEN
investigation of aggregating paramagnetic particles occurred.
Svoboda (1987) gives a very good outline of the factors effecting
magnetic aggregation of paramagnetic particles. From a practical
perspective most of the literature outlines laboratory experiments
or theoretical modelling of the process, and do not generally
apply the process to the valuable paramagnetic sulfide particles
that are recovered by flotation. Nor does the extensive literature
apply the process to flotation. However, the theoretical and
laboratory basis is equally applicable to sulfide as well as
non-sulfide minerals.
Engelhardt, Ellis and Lumsden (2005) have clearly
demonstrated that magnetic aggregation of fine paramagnetic
particles can be used to increase the flotation recovery of the fine
paramagnetic minerals chalcopyrite and sphalerite. Their paper
summarises the literature that outlines the theoretical basis of
aggregation of fine paramagnetic particles.
Gold mineralogy
Gold, particularly when found in base metals ores is often
recovered by froth flotation. Gold mineralogy, whether in base
metal ores or other ores, can be complex. Though mostly
occurring as gold metal, rather than a gold mineral, it can occur
as free native gold or be associated with different minerals in the
ore, depending on the nature of the ore formation. The difficulty
in being precise about gold mineralogy is that in most flotation
ore the gold is in such low concentrations that seeing the gold is
difficult, therefore understanding much about its occurrence and
associated mineralogy is also difficult (Henley, 1975).
Table 1 shows that gold is not paramagnetic but it is
diamagnetic. This means that with magnetic fields of industrial
strength gold is not magnetised. However, gold in its distribution
within the ore may be associated with paramagnetic minerals
such as chalcopyrite or marmatite. The literature contains a
number of studies where magnetic separation methods have been
used to concentrate and recover gold (Corrans and Levin, 1979;
Corrans, 1980; Corrans and Dunne, 1985; Ciesla, 2003). Gold
that can be recovered by magnetic separation is gold that occurs
in minerals with a strong magnetic susceptibility, because the
concentration of paramagnetic minerals is only moderately
successful by magnetic separation.
TABLE 1
Magnetic susceptibility of minerals and gold.
Mineral Reported magnetic susceptibility
(Svoboda, 1987) (M3kg-1 10-9)
Chalcopyrite 1595.9
Bornite 100.5
Marmatite 38 - 5900
Cassiterite 2136.3
Pyrite 1-5
Quartz -5.7
Gold -0.15
Laboratory scale test work undertaken by Corrans and Levin
(1979), with wet high intensity magnetic separation (WHIMS)
on a range of samples from ore to leach or flotation tailings gave
gold recoveries in the magnetic fraction of up to 95 per cent.
Other researchers (Corrans, 1980; Watson, Rassi and Bahaj,
1983; Corrans et al, 1984; Corrans and Dunne, 1985; Conway
and Dunne, 1986; Svoboda, Corrans and Spitze, 1986; Svoboda,
Lazer and Te Riele, 1987) have published laboratory and plant
test work on the recovery of gold from Witwatersrand ore and
residue by WHIMS. Ciesla (2003) recovered more than
70 per cent of the gold from copper flotation tailings by high
gradient magnetic separation (HGMS).
224 Fremantle, WA, 19 - 21 March 2007
The mineralogy of the gold that resulted in it having a high
magnetic susceptibility is not clearly demonstrated. Low levels
of gold make mineralogical analysis problematic. However,
explanations by the above authors range from: gold in sulfide
minerals, gold in silicates, gold in oxide minerals, gold
containing iron or gold with iron or iron oxide coatings.
Martinez (1988a, 1988b) applies a coating of colloidal iron or
iron hydroxide to ores and tailings to magnetise the gold
particles so that they can be recovered by magnetic separation.
Gold may also be closely associated with sulfide minerals such
as chalcopyrite, galena, bornite, pyrite and arsenopyrite (Henley,
1975). Some of these minerals are paramagnetic, so that gold
may be concentrated magnetically if the associated sulfide can be
concentrated magnetically.
Bheemalingeswara (1995) has shown that the coating of gold
particles with iron oxide/iron hydroxide gave the particles
sufficient magnetic susceptibility to be concentrated in the
laboratory by magnetic separation.
In summary, gold’s disposition in the ore as well as in the
mineral flotation slurry will determine whether the gold is
amenable to magnetic methods. Other researchers have shown
that gold can be associated with paramagnetic minerals, or with
magnetic minerals, or has iron impurities within the gold matrix,
or is coated with a magnetisable iron coating. Gold in any of
these forms has been shown to respond to magnetic methods. In
the flotation process where paramagnetic minerals such as
chalcopyrite are being recovered then increasing the recovery of
fine chalcopyrite would be expected to increase the recovery of
the fine gold associated with the chalcopyrite. As well, in the
flotation slurry there is fine metallic iron and iron hydroxides,
generated from the grinding process and the normal alkaline
flotation conditions, so that it is possible that the free gold or
even gold on the surface of a silicate particle may absorb iron or
iron hydroxides and so be able to be magnetised and aggregated.
The gold in the ore at Northparkes mine is mostly found
disseminated as fine particles, most particles are less then 5 µm.
The particles are generally associated with the copper bearing
sulfide minerals; however, there is a significant percentage found
along the grain boundaries of the silicate host rock.
Randomised block design trials
Napier Munn (1995) outlines in detail some paired methods for
testing process variables with specific reference to mineral
processing operations. The paper gives an excellent outline of
randomised block design trials. As Napier-Munn identifies, the
main advantage of randomised block design methodology is that:
it allows the significance of the main effects and
their interactions to be rigorously tested (are
they real or aren’t they?) whilst eliminating and
indeed separately testing for, the effect of the
interfering factor (here time).
Randomised block design uses the analysis of variance
(ANOVA) statistical procedure to evaluate the data.
Napier-Munn (1995) identifies that the key to evaluation in the
minerals processing industry is to determine whether a real
difference is occurring above the daily variations of the noise of
the plant. This is the critical part of plant evaluation. The process
change may be effective in the laboratory and even effective in
the plant but being able to ‘see’ its effectiveness and determine
that the effect is real is the crucial technique.
One way to improve the ability to ‘see’ a change in the plant is
to reduce the background noise. Since magnetic aggregation only
affects the recovery of fine mineral, with great insight the
Northparkes mine site personnel further modified the normal
design of the experiment so that the flotation performance of the
Ninth Mill Operators’ Conference
 IMPROVING FINE COPPER AND GOLD FLOTATION RECOVERY — A PLANT EVALUATION
fine mineral could be determined separately from the coarse
mineral. All samples were sized at 20 µm and 38 µm. As the
results show this significantly reduced the variance and allowed
the difference to be ‘seen’ much more readily. Particularly in the
case of determining gold recovery this was effective. Coarse gold
in a sample is notoriously difficult to accurately measure and so
removing this coarse gold noise from the testing was vital. This
sizing step was a significant change in methodology to normal
trials and may have more widespread implications.
Northparkes mine and operation
The Northparkes copper-gold mine is located in the central west
of New South Wales and has been operating since 1993. The
operation is a joint venture between Rio Tinto and the Sumitomo
Group.
A diagram of the Northparkes Module 2 flow sheet is given in
Figure 1. The processing plant comprises two parallel modules,
Module 1 with a design capacity of 245 t/h and Module 2 with a
design capacity of 423 t/h. Ore is crushed and then ground in a
SAG mill and two stages of ball milling to a flotation feed size
(P80) of 95 - 110 µm. Flash flotation is used in the grinding
circuit to recover the free floating course material; the grinding
is followed by conventional flotation that consists of rougher
scavenger banks with NaHS, frother, xanthate and thiono-
carbamate as chemical additives. The rougher concentrate is then
cleaned by Jameson cells to produce the final copper gold
concentrate.
Concentrate is thickened and filtered and sold by contract and
on the spot market for smelting. Tailings after thickening report
to the tailings dam.
Copper mineralogy is a mixture of chalcopyrite and bornite.
There is little associated pyrite.
Testing methodology
Testing was carried out on Module 2. The flotation module has
two large existing conditioning tanks that were used for NaSH
conditioning when high oxide ore was treated. The ProFlote®
magnetic conditioning devices were installed in the second of
these conditioners where the slurry residence time is 2.5 minutes.
The ProFlote® conditioning devices consist of high strength high
gradient rare earth permanent magnets arrayed within a stainless
steel assembly. The strong magnetic field of about 5000 gauss
contacts the slurry. At frequent and regular intervals the magnets
are automatically removed from the slurry to allow for the
removal of ferromagnetic material that may blind the magnets.
A randomised block design program was used. The first day
after turning the magnetic conditioners ‘on’ or ‘off’ was an
equilibration day and the second day was a test day. The nature
of randomised block design testing is that paired samples are
taken. Randomising and pairing is established at the beginning of
the testing and locked in for the trial period. If during one of the
pairs there is a major plant shutdown or disturbance for a
prolonged period of time, or sampling or laboratory issues occur
then both data points in the pair are discarded. Once the testing
was standardised and underway and a few teething problems
solved about 70 per cent of the data was considered useful. Most
of the discarded data related to an automatic sampler problem, or
one of the ball mills being down for maintenance. The more
intensive sample preparation in the laboratory did not present a
problem after the teething period.
The plant automatic samplers sample the final concentrate,
final tail, plant feed and flash float tail (flotation feed). This
presented a problem in the testing because the final concentrate
includes the flash concentrate. For the test work either the final
concentrate could be used or the concentrate excluding the flash
concentrate sampled manually. Because the flash concentrate has
a very low quantity of fines, and because the manual sample is
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
inherently inferior to an automatic sample, it was decided to use
the automatic final concentrate sample. Automatic samplers took
normal shift composite samples and these samples were assayed
as well as subsamples being split for size analysis. Samples were
sized at 20 µm and 38 µm with the +38 µm sizes being combined.
Plant operation was monitored so that on days where a mill
was shut down or some other major change occurred samples
were discarded. Note that the block design makes it highly
unlikely for the results to be biased by the flash float concentrate
being included in the concentrate sample. Changes in flash float
performance will only increase the noise in the data.
RESULTS AND DISCUSSION
The results from the randomised block testing are summarised in
Table 2 for copper and Table 3 for gold.
There is no statistical difference between the feed grades,
concentrate grade or copper and gold distribution in the different
size fractions.
The analysis of variance of the recovery for the different
fractions for copper and gold are given in Table 4 for copper and
Table 5 for gold.
Copper results
The copper recovery increase in the <20 µm fraction and the 20 -
38 µm fraction has been shown by ANOVA analysis to be, in
Napier-Munn’s (1995) terms, real. The best estimate of the
increase in copper recovery is the difference in the means, which
is 2.08 per cent for the <20 µm fraction and 0.98 per cent for the
20 - 38 µm fraction. The results, increased recovery in the finer
fractions and no increase in recovery in the coarser fractions, are
consistent with a mechanism of magnetic aggregation of fine
paramagnetic minerals improving their flotation. The results are
consistent with those of Engelhardt, Ellis and Lumsden (2005).
Although the grind size is not particularly fine at Northparkes
it is interesting to note that about 70 per cent of the copper is less
than 38 µm. It is also worth noting that sizing the sample allowed
for a significant reduction in the noise with the standard
deviation of the paired differences being much lower in the
<38 µm fractions compared to the >38 µm fractions. This meant
that the statistical significance in the <38 µm fractions was much
higher than for the total sample, shortening the trial and giving
greater confidence in the results.
Gold results
The gold recovery increase in <20 µm fraction has been shown
by ANOVA analysis to be ‘real’ with the best estimate of the
increase being the difference in the means, which is 3.51 per cent.
Again, the increase in the gold recovery was for the finest
fraction with no real increase in the coarser fractions. This is
consistent with an aggregation mechanism. Of particular note is
the substantial increase in the standard deviation for the coarser
size fractions. Despite as large or larger differences in the means
the results cannot be said to be real because the standard
deviations for the courser size fractions are so large. Given that
about 50 per cent of the gold in the feed is in the >20 µm
fraction, the very large standard deviations in gold recovery for
these fractions means that testing to demonstrate a real difference
in recovery over the whole sample (without sizing) would have
required a very long trial.
There is not sufficient data, nor sufficient gold in this very
low-grade gold ore to determine on which gold the ProFlote is
impacting. Whether magnetic aggregation is impacting on the
fine gold associated with copper sulfide minerals; bornite and
chalcopyrite; or with fine free gold that has a magnetite or
iron-hydroxide coating or whether it is some other paramagnetic
or ferromagnetic mineral it is impossible to say.
225
 T RIVETT, G WOOD and B LUMSDEN
226

TO FLOAT CELLS
PP089/90
PP115
(Frother)
(NaHS)
CV12 BL01-03
Metal PP110 PP112 Blowers
Detector (Collector) (Collector)
CV21 AG21
Oversize PP103
Crusher AG22 (Promoter)
Ore
Ore CR03 Weightometer
Stockpile
Stockpile 2 Ball Charger 1st Roughers
2nd Roughers 1st Scavengers 2nd Scavengers
BC02 (Dorr Oliver) CV450
Optional CY06 A-H (Dorr Oliver) CV451 (Dorr Oliver) (Dorr Oliver)
FT15-16 CV452 CV453
15" Krebbs FT17-18 FT19-20 FT21-22

Apron TK36
Feeder Conditioner 1 TK37
AF01
Conditioner 2
Feeders SV06 Trommel ML06
FE06-09 Magnet CV11 PP091
Svedala TL06
(MA02) (Frother)
(2400x6100) Optional

CV10 Weightometer

NC
ML03 NO
(8500x4300)
From
To Flotation Circuit PP080
Grinding
(8/6 E-AH)
Circuit PP079
PP222 AG30/31 (12/10 E-M)
(12/10 F-AH) HP25
CY04 A-O HP41
HP13 (15" CAVEX)
CY03 A-D TK57/58 HP12
(26" CAVEX) PP075 Cleaner-Scavenger Feed
(8/6 E--AH) Conditioner
PP064/65 CV330 CV329 FT27
FT45
(12/10 F-AH) Ball Charger (8 Downcomers)
BC03
Flash Float
Splitter
Rougher Cleaner
(Skimair) PP223 SUMP Cleaner-Scavengers
(65VD-GPS) (Dorr Oliver)
CV454
Jameson FT23-26
Flash Float To HP41 Cell
CV334 Cleaner
MODULE 2 RECLAIM ML04
(5500x9400)
(Outokumpu 5)
CV457 FT28
Trommel
(321) & GRINDING TL04
FT46
MODULE 2 Optional
(3 Downcomers)

Recleaner

(330) CIRCUIT CV332 FLOTATION (340)

Jameson
Cell

SCATS HP11
Optional
CIRCUIT HP26
PP076
(6/4 D-AH)
CV459
Fremantle, WA, 19 - 21 March 2007

PP081
To Con Flash Float Con (6/4 D-AH)
To Primary Hopper Pump

PP068 SUMP
(65VD-GPS) HP15
PP082
(4RV-AF) To Tailings
Thickener

PP066/67 To Concentrate Thickener

(14/12 FF-AH)

From Mod 2 Tails

PP105 Pump
Mod 1 Concentrate (Floc)
Thickener O/F From Mod 1

MODULE 2 CONCENTRATE From Concentrate Pump

Tails Pump

THICKENING (355) & PP134 SUMP

(65 VDS-GP)
Tailings Thickener
FILTRATION (360) CIRCUIT Mod 2 Concentrate
Thickener
TH01
TH03

PP146 SUMP
(65 VD-SPR)
PP141 SUMP
(Johnson)

PP129 PP128
To Process (8/6 EE-AH) (8/6 EE-AH)
Water Tank

From Retreat
TK43 Thickener O/F
PP118/119
Oxalic Acid Storage OA PP177 PP138
NA Nitric Acid Storage AG28 (SP100 & 3/2 C-AH)
TH02 Fresh
Process Dilution
PP131
(8/6 EE-AH)
PP130
(8/6 EE-AH)
PP120/121 Retreat Water
& Hoses
(SP100 & Thickener
BLAGDON)
PP133
Mod 2 Disc Filter PP215 (8/6 EE-AH) PP132
(Outokumpu CC-30) To Mod 1 Concentrate (SP40) (8/6 EE-AH)
FL01 Thickener
TK52
SV07 Concentrate StorageTank
Schenik Screen TK45
Gland TK54
PP161/162 Water Process 355-DM01 335-DM02
Feed Water Tailings Dam #1 Tailings Dam #2
Tank Tank
PP125/126/127
(14/12 F-S)
CV16 CV14

To Gland

Weightometer
Water System

PP052/53
THICKENING (355) &
PP042/43
TAILINGS STORAGE
Ninth Mill Operators’ Conference

DM04 DM05 DM03

FACILITY (335)
Process Water Dams Return Water Dam

Concentrate Concentrate
Stockpile 2 Stockpile 1

Drawing Name: "Module 2 Sulphide Process Flowsheet"

Checked by: OPD Date: 06/04/05
Module 2 Sulfide Process Flow Sheet Compiled by: Tom Pethybridge
Date of Previous Version: 09/12/04

FIG 1 - Northparkes module 2 flow sheet.

 IMPROVING FINE COPPER AND GOLD FLOTATION RECOVERY — A PLANT EVALUATION

TABLE 2
Mean copper results for ‘on’ and ‘off’ days.

Fraction Feed (%Cu) Conc grade (%Cu) Tail (%Cu) Recovery (%) Distribution (% of Cu in
feed in fraction)
<20 µm ON 1.098 43.15 0.119 89.34 52.3
<20 µm OFF 1.075 43.27 0.139 87.26 53.7
>20 <38 µm ON 1.199 45.41 0.033 97.28 18.7
>20 <38 µm OFF 1.134 45.36 0.043 96.30 18.1
>38 µm ON 0.395 28.67 0.107 73.06 29.0
>38 µm OFF 0.392 29.00 0.111 71.86 28.2

TABLE 3
Mean gold results for ‘on’ and ‘off’ days.

Fraction Feed (ppm Au) Conc grade (ppm Au) Tail (ppm Au) Recovery (%) Distribution (% of Au in
feed in fraction)
<20 µm ON 0.48 17.09 0.07 86.15 51.1
<20 µm OFF 0.47 17.02 0.08 82.86 53.2
>20 <38 µm ON 0.32 9.99 0.11 66.66 11.8
>20 <38 µm OFF 0.30 9.80 0.11 62.54 11.4
>38 µm ON 0.21 5.79 0.15 29.47 37.1
>38 µm OFF 0.22 6.16 0.16 26.33 35.4

TABLE 4
ANOVA analysis of copper results.

Fraction % Difference Std dev Degrees of freedom ‘F’ Significance

<20 µm 2.08 3.0 13.1 6.78 >97,5
>20 <38 µm 0.98 1.1 13.1 11.84 >99
>38 µm 1.20 4.6 13.1 0.94 N.S

TABLE 5
ANOVA analysis of gold results.

Fraction % Difference Std dev Degrees of freedom ‘F’ Significance

<20 µm 3.51 3.8 12.1 11.37 >99
>20 <38 µm 3.43 9.0 12.1 1.88 N.S
>38 µm 3.17 18.9 12.1 1.49 N.S

Engelhardt, Ellis and Lumsden (2005) showed that magnetic
aggregation could increase the recovery of silver, a metal that
like gold often occurs in elemental form in ores and is also
diamagnetic. These gold results are then consistent with the
results presented by Engelhardt, Ellis and Lumsden (2005).
Economic impact
The economic impact of these improvements in recovery to
Northparkes mine are substantial. This technology is now part
of the Northparkes flow sheet, and testing on Module 1 will
commence in 2007.
CONCLUSION
Magnetic aggregation has been shown to improve the efficiency
of separation of flotation by improving the recovery in the fine
fractions. Both copper and gold recovery in the fine fractions has
increased. These increases are real and beneficial. The results are
consistent with a magnetic aggregation mechanism.
Because magnetic aggregation targets the fine fractions these
fractions were separately assayed and the improvement was
shown to occur in these fractions alone. Targeting the fine
fractions in carrying out the trial reduced the plant noise and
allowed the improvement to be ‘seen’ more quickly.
The randomised block design method has also been shown to
be effective in ‘seeing’ these improvements.
Perhaps one of the major innovations of this work is the use of
sized samples to reduce the plant noise. While this has been
specifically applied to a process that targets fine mineral, the
methodology may also be applicable to other processes that don’t
necessarily target fine mineral. For instance, in testing a process
change that the laboratory has shown to improve the process
across all sizes it might be beneficial in the plant to measure the
effect on the total sample as well as specifically the fine fraction.
Particularly in the separation of gold, measuring the effect on
the fine fraction where the variance is much smaller might be an
indicator of what is happening across all sizes, especially if the
laboratory demonstrates this is an appropriate conclusion. Or it
may be sufficient to the plant that the process improves the fine

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 227

T RIVETT, G WOOD and B LUMSDEN
gold recovery alone. Whichever, looking at changes in size
fraction recovery could be a preferred or additional method of
testing that allows the operator to see small ‘real’ differences
more readily, than with a non-sizing testing method.
ACKNOWLEDGEMENTS
Barry Lumsden would like to thank Professor Jameson, Director
of the Centre for Multiphase Processes at the University of
Newcastle for his assistance with Barry’s research and Rick
Dunn, formerly of Northparkes mine, for his assistance in the test
work. Barry Lumsden would like to thank AusIndustry for its
financial support of his research and Northparkes mine for
allowing this work to be presented.
REFERENCES
Bheemalingeswara, K, 1995. Possible effects of iron oxide coating in the
recovery of particulate gold from stream sediments, Journal of
Geochemical Exploration, 52(3):373-380.
Ciesla, A, 2003. Use of a strong magnetic field to recover weakly
magnetic grains dispersed in some mineral wastes, Prace Naukowe
Instytutu Podstaw Elektrotechniki I Elektrotechnologi Politechniki
Wroclawskiej, 38:262-292.
Conway, T and Dunne, R, 1986. The influence on gold recovery of
de-sliming before wet high-intensity magnetic separation, in
Proceedings International Conference on Gold, Volume 2: Extractive
Metallurgy of Gold, pp 419-428 (South African Institute of Mining
and Metallurgy: Johannesburg).
Corrans, I, 1980. A development in the application of wet high-intensity
separation, in Proceedings International Symposium Fine Part
Process (ed: P Somasundaran), Vol 2, pp 1294-1309 (The American
Institute of Mining, Metallurgical and Petroleum Engineers, Inc:
New York).
Corrans, I and Dunne, R, 1985. Optimisation of the recovery of gold
and uranium from Witwatersrand residues, in Proceedings XV
International Mineral Processing Congress, Vol III, pp 497-509
(Bureau de Recherches Geologiques et Minieres).
Corrans, I, Gilbert, W, Liddell, K and Dunne, R, 1984. The performance
of an industrial wet high-intensity magnetic separator for the
recovery of gold and uranium, J S Afr Inst Min Metall, 84(3):57-63.
228 Fremantle, WA, 19 - 21 March 2007
Corrans, I and Levin, J, 1979. Wet high-intensity magnetic separation for
the concentration of Witwatersrand gold-uranium ores and residues,
J S Afr Inst Min Metall, 79(2):210-228.
Engelhardt, D, Ellis, K and Lumsden, B, 2005. Improving fine sulfide
mineral recovery – Plant evaluation of a new technology, in
Proceedings Centenary of Flotation Symposium, pp 829-834 (The
Australasian Institute of Mining and Metallurgy: Melbourne).
Fuerstenau, D, Chander, S and Abouzeid, A, 1979. The recovery of fine
particles by physical separation methods, in Beneficiation of Mineral
Fines Problems and Research Needs (eds: P Somasundaran and N
Arbiter), pp 3-61 (The American Institute of Mining, Metallurgical
and Petroleum Engineers, Inc: New York).
Henley, B, 1975. Gold-ore mineralogy and its relation to metallurgical
treatment, Minerals Sci Engng, 7(4):289-312.
Martinez, E, 1988a. Recovery and magnetic concentration of gold, South
African Patent No ZA 8708828.
Martinez, E, 1988b. Process for concentration of gold and uranium
magnetically, US Patent No US 4726895.
Napier-Munn, T, 1995. Detecting performance improvements in trials
with time-varying mineral processes, Mineral Processing, 8(8):
843-858.
Sivamohan, R, 1990. The problem of recovering very fine particles in
mineral processing – A review, International Journal of Mineral
Processing, 28:247-288.
Svoboda, J, 1987. Magnetic Methods for the Treatment of Minerals
(Elsevier: Amsterdam).
Svoboda, J, Corrans, I and Spitze, M, 1986. The effect of pH on the
recovery of uranium and gold by high-gradient magnetic separation,
International Journal of Mineral Processing, 17(1-2):83-98.
Svoboda, J, Lazer, M and Te Riele, W, 1987. Selective magnetic
separation of uranium and gold, IEEE Transactions on Magnetics,
MAG-23(1):283-293.
Trahar, W, 1981. A rational interpretation of the role of particle size in
flotation, International Journal of Mineral Processing, 8:289-327.
Trahar, W and Warren, L, 1976. The floatability of very fine particles – A
review, International Journal of Mineral Processing, 3:103-131.
Watson, J, Rassi, D and Bahaj, A, 1983. The recovery of gold
and uranium from gold ore leached residues by HGMS, IEEE
Transactions on Magnetics, MAG-19(5):2136-2138.
Ninth Mill Operators’ Conference
Reducing WAD Cyanide at the Discharge Spigot
J A Rumball1, S D Munro2 and M L Habner3
ABSTRACT
In a gold CIP/CIL circuit, adding new circuit water to a final tails
thickener and then directing the thickener overflow to the head of the
circuit significantly reduces the cyanide concentration at the TSF
discharge spigot. The new water is used to dilute the final tails.
The dilution and recycle of the tails in this manner may have the added
benefit of reducing reagent consumption, in some cases to the point
where the reagent savings justify the expenditure and Cyanide Code
compliance is achieved at no extra cost.
Potential circuit configurations and water recycles can be modelled by
splicing a flowsheeting and thermodynamics package, such that changes
to the composition of the water can be reduced to chemical speciation
with a Gibbs free energy minimisation enabling a complete understanding
of the circuit water chemistry.
INTRODUCTION
The water chemistry in any circuit can be complex. Gold circuits
are no exception. Cyanide can complex with, or react with, a
variety of species. The degree of these reactions will define how
much of the cyanide is available for gold leaching and titration.
When lime is added it may precipitate magnesium and/or sulfate
from the process water and raise the pH. The pH in turn defines
how much cyanide is available for gold leaching and titration.
Such a complex series of interactions can only be followed with
a robust model.
Fortunately, many solution chemistry reactions are fast and
may be modelled using a Gibbs free energy minimisation;
however, many are not and require a detailed understanding of
the system under investigation. For cyanide some reactions are at
equilibrium whilst others are kinetically limited.
Because of the recognised complexities, there is significant
merit in modelling the water chemistry in a gold circuit. A
detailed model that takes into account thermodynamic
equilibrium, many of the recognised kinetic constraints and is
‘calibrated’ against plant data, can provide a valuable insight and
enable a more thorough optimisation than could otherwise be
achieved.
1. MAusIMM, Gold Program Manager, CSIRO Minerals, PO Box 7229,
Karawara WA 6152. Email: john.rumball@csiro.au
2. Previously: Research Scientist – Gold Program, CSIRO Minerals.
Currently: Senior Processing Engineer, Rio Tinto Operational and
Technical Excellence. Email: scott.munro2@riotinto.com
3. MAusIMM, Senior Metallurgist, Kalgoorlie Consolidated Gold
Mines Pty Ltd. Email: mhabner@kalgold.com.au
FIG 1 - Basic gold CIP flow sheet.
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
This paper discusses the construction of a gold circuit water
chemistry model and the use of that model to demonstrate that in
some circumstances it is possible to meet our environmental
obligations whilst reducing operating costs at the same time.
EXPERIMENTAL PROCEDURES
The platform
The platform upon which we have built our model is SysCAD
9.0. SysCAD is a modular sequential flowsheeting package
similar to MetSIM, Aspen and a host of others. SysCAD is
relatively cheap to purchase and maintain, has an open
architecture and a COM interface allowing communication with
other packages such as Excel and the Outokumpu thermodynamic
database HSC Chemistry 5.11. It should be noted that
Outokumpu have recently introduced their own SIM flow sheet
module (HSC Chemistry 6.0), which is capable of achieving the
same functionality; however, significantly greater programming
complexity is required to achieve a result.
Our basic gold circuit model includes a grinding circuit
(mixing only), leach circuit, adsorption circuit, tails thickener
and tails dam (see Figure 1). Complexity was later added by
including a float circuit and float tails thickener.
Within each unit, the dominant chemical reactions were
modelled. The model also took into account reagent addition,
recycle and a Gibbs free energy minimisation to calculate the
stable species at each stage.
Leaching
Because this study is focused on water chemistry, gold leaching
chemistry is of less importance, and is not included in these
simulations. Instead a number of other reactions are considered
to emulate the complicated changes that occur. Cyanide is
allowed to react with metals in solution, but the reaction with
oxygen to form cyanate or with sulfide to form thiocyanate is
kinetically constrained. Without this constraint, all the cyanide
would react through to cyanate or thiocyanate due to the
thermodynamic instability of cyanide.
Similarly, although the ore used in this simulation contains a
considerable quantity of haematite, a kinetic constraint is placed
on its dissolution to prevent it all being dissolved by cyanide.
A pH fall through the circuit is simulated by slowly dissolving
pyrite according to:
4FeS2 + 15O2 + 14H2O = 4Fe(OH)3 + 8SO42- + 16H+
229
J A RUMBALL, S D MUNRO and M L HABNER
The amount of pyrite dissolving is limited by calibrating the
model against plant data.
Cyanide addition is controlled to a titratable cyanide set-point
with a control loop. Titratable cyanide is calculated from the CN-
+ one cyanide from Cu(CN)43-, but excludes HCN(aq). WAD
cyanide is calculated by combining the CN-, HCN(aq), and
dissociable copper cyanides.
Lime is added as a slurry containing 22 per cent CaO to a pH
set-point, with the pH calculation taking into account the
buffering effect of magnesium in hyper-saline water.
The control of reagents to a set-point is a key feature of the
model, allowing reagent consumption to be calculated under
different model scenarios.
Adsorption
The reactions occurring in the adsorption circuit are largely a
continuation of those occurring in the leach circuit. Although the
model has the capability of following gold, it is not significant in
the context of this study. The exception to this is the gold in
solution tails. The recycle of solution gold can significantly
reduce solution gold losses and play an important role in the
process economics. For the purpose of this exercise the gold in
solution tails has been set at 0.025 mg/L.
One reaction that is important in the adsorption circuit is the
adsorption of copper on carbon. At this stage the model includes
copper adsorption rates and equilibrium loading profiles and so
copper in the adsorption circuit is well modelled. What is not
modelled is the behaviour of copper in the elution circuit and its
recycle.
TSF
The physical and chemical processes occurring in a tailings dam
are difficult to understand with certainty due to the myriad of
environmental and process interactions. Diurnal and seasonal
fluctuations have a large impact, changing beach angles,
changing water hold-up and understanding the settling processes
are all important in a TSF model. For the purpose of this work,
plant data indicating a water recycle of 31 per cent was used.
This leaves a settled mud density of 68 per cent, a value within
the range indicated by Newson and Fahey’s (2001) study of core
samples taken from gold tailings dams.
An excess of carbon dioxide at atmospheric concentration was
added to the TSF, lowering the pH of the tails decant water to
7.2. This matched closely the pH of plant decant water.
HCN(g) loss was proportional to the HCN(aq) concentration
after the pH had fallen due to CO2 adsorption and calibrated
against the decant cyanide concentration.
The Mg(OH)2 precipitated in the circuit was assumed to settle
to the floor of the tailings dam and not interact with the
atmospheric CO2 saturated decant water. This resulted in the
decant water having a similar magnesium concentration to the
adsorption circuit tailings, a conclusion that matched measured
plant data.
Detox
In some of the models a Detox circuit was included. Detox was
with peroxide on a clear solution achieving 30 per cent peroxide
utilisation.
Due to the complexity of the model, not all the details are
included here but in all cases the model was calibrated against
plant data to ensure that the base case conditions matched those
prevailing in an operating plant. The model does, however,
remain weak when extrapolating to a significantly different set of
operating conditions and it is only with further experience that
the robustness of the model will improve in this respect.
230 Fremantle, WA, 19 - 21 March 2007
RESULTS AND DISCUSSION
Traditionally, new circuit water is added to the grind. One way of
reducing the WAD cyanide at the discharge spigot is to add new
water to the tails and increase the level of recycle. The WAD
cyanide concentration at the discharge spigot is thus reduced
through dilution. Obviously, this is only likely to be feasible with
a tailings thickener.
The thickener overflow water is recycled to the front of the
circuit. This water obviously contains cyanide, so not only is the
concentration of cyanide discharged to the tails spigot reduced,
but the total amount of cyanide discharged is reduced. Recycling
the cyanide reduces the amount of new cyanide that needs to
be added and further compounds the reduction in cyanide
discharged to the spigot.
An added advantage is that the water that would otherwise be
discharged to the tails spigot is low in magnesium. Through
dilution, some of this water is now recycled resulting in a lower
lime demand.
Obviously the flow of streams, their chemistry and the impact
of recycles becomes very complex very quickly which is why we
have used models to develop a quantifiable understanding of the
processes occurring.
The model also serves to remind us that there may be
operational issues associated with the proposed changes in
circuit configuration. The introduction of cyanide to the grind
and perhaps more importantly to the classification circuit, where
HCN(g) volatilisation may become a safety issue, are important
considerations to take into account.
Quantifying the consequences of the dilution strategy
demonstrates that the benefits can be significant. Under the
conditions modelled, the combined effect of dilution plus recycle
reduces the WAD cyanide at the discharge spigot from 124 to
61 mg/L and the amount of cyanide discharged from 45 to
18 kg/h (see Figure 2).
Not only are there environmental benefits; it is also calculated
that adding new water to the tails thickener may reduce operating
costs from $1.63 to $1.23/t of ore treated (see Figure 3). Solution
gold lost to tails has been counted as an operating ‘cost’ such that
the benefits of recycling solution gold to the head of the leach
circuit and providing a second opportunity to capture this gold
become clearly apparent.
The model also enables the sensitivity of the circuit to
operational changes to be assessed. For example, increasing the
density of the tails thickener underflow from 65 to 70 per cent
may not be beneficial. Although the intuitive conclusion is that
the increased density should lead to a further reduction in WAD
cyanide at the discharge spigot, the model suggests that water
balance requirements limit the amount of new bore water that
can be added in this scenario and thus the expected benefits (see
Figure 4).
The model can also be modified to consider more complex
circuits. A major complexity that has been considered is the
introduction of a flotation circuit. Flotation complicates the
picture substantially in that recycle of cyanide to a float circuit is
generally not acceptable. To overcome this, a Detox circuit may
be considered and/or separate water circuits may be an option.
In the ‘two separate water circuits’ scenario the tails thickener
overflow is used to dilute the feed thickener underflow (see
Figure 5). This scenario is limited by the amount of new water
that can be added to the tails thickener. In fact in this scenario,
insufficient water can be added to the tails thickener to
compensate for water losses in the TSF and consequently
additional water is required to top up the process water tank. The
environmental benefits and cost savings are substantial, but
reduced when compared to those available in a circuit without
flotation (see Figure 6).
Ninth Mill Operators’ Conference
 REDUCING WAD CYANIDE AT THE DISCHARGE SPIGOT

140 50
WAD cyanide concentration
WAD cyanide discharged 45
120
40

WAD cyanide discharged (kg/hr)

WAD cyanide concentration (mg/L)
100 35

30
80
25
60
20

40 15

10
20
5

0 0
Base case Water to tails thickener

FIG 2 - WAD cyanide concentration and flow to the TSF discharge spigot.

1.80 Solution gold lost

Cyanide cost
1.60 Lime cost

1.40
Cost ($/t of ore treated)

1.20

1.00

0.80

0.60

0.40

0.20

0.00
Base case Water to tails thickener

FIG 3 - Lime and cyanide operating costs combined with cost of gold lost in solution.

140 50 1.80 Solution gold lost

WAD cyanide concentration
Cyanide cost
WAD cyanide discharged 45 1.60
120 Lime cost
40
WAD cyanide discharged (kg/hr)
WAD cyanide concentration (mg/L)

1.40
100 35
1.20
Cost ($/t of ore treated)

30
80 1.00
25
60 0.80
20
0.60
40 15
0.40
10
20
5 0.20

0 0 0.00
Base case Water to tails thickener 70 % underflow density Base case Water to tails thickener 70 % underflow density

FIG 4 - Environmental and economic consequences of increasing the tails thickener underflow density from 65 to 70 per cent solids.

In the ‘overlapping water circuits’ scenario the tails thickener loss of some of the solution gold to the leach feed thickener
overflow is directed to the leach feed thickener feed (see overflow, from which it is assumed to be lost (see Figure 8).
Figure 7). The overflow from the leach feed thickener is then Another potential disadvantage to this circuit is that gold
bled back to the front of the circuit. This circuit demonstrates leaching may commence in the leach feed thickener and this
benefits in reduced WAD to the discharge spigot, versus the solution gold will be ‘discarded’ with the leach feed thickener
base case, reduced cyanide consumption, but suffers from the overflow.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 231

J A RUMBALL, S D MUNRO and M L HABNER

FIG 5 - Flow sheet showing two separate water circuits.

Detox costs
140 50 1.80
WAD cyanide concentration Solution gold lost
WAD cyanide discharged 45 1.60
Cyanide cost
120 Lime cost
40
WAD cyanide discharged (kg/hr)
WAD cyanide concentration (mg/L)

1.40
100 35
1.20
Cost ($/t of ore treated)

30
80 1.00
25
60 0.80
20
0.60
40 15

10 0.40
20
5 0.20

0 0 0.00
Base case Two separate water circuits Base case Two separate water circuits

FIG 6 - Environmental and economic consequences of operating two separate water circuits.

FIG 7 - Flow sheet showing two overlapping water circuits.

232 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 REDUCING WAD CYANIDE AT THE DISCHARGE SPIGOT

Detox costs
140 50 1.80
WAD cyanide concentration Solution gold lost
WAD cyanide discharged Cyanide cost
45 1.60
120 Lime cost
40

WAD cyanide discharged (kg/hr)

WAD cyanide concentration (mg/L)

1.40

100 35
1.20

Cost ($/t of ore treated)

30
80 1.00
25
60 0.80
20
0.60
40 15

10 0.40

20
5 0.20

0 0 0.00
Base case Two separate water circuits Two overlapping water circuits Base case Two separate water circuits Two overlapping water circuits

FIG 8 - Environmental and economic consequences of operating two overlapping water circuits.

CONCLUSIONS The complexity of the cyanide-pH-lime interactions, together

with recycle streams and the necessity of maintaining a water
By adding new water to a tails thickener rather than the front of a balance means that scenarios of this type are best studied in
gold CIP/CIL circuit it may be possible to reduce the WAD flowsheeting packages that have the ability to calculate
cyanide at the discharge spigot to meet Cyanide Code limits. The the equilibrium chemistry and accommodate user specified
new water dilutes the tails as well as recycling some of the limitations to equilibrium conditions such that the flow sheet
cyanide so that less new cyanide is required to achieve accurately reflects how the plant will operate. The CSIRO
concentration set-points in the leach. Minerals Gold team within the Parker Centre has such a model.
In Western Australia the benefits are magnified in that
substantial savings in lime consumption are also possible with ACKNOWLEDGEMENT
the recycle of water that has already had the magnesium
The authors would like to thank KCGM for their support of some
precipitated.
of the work presented in this paper.
Adopting the same strategy in circuits that include flotation
complicates the picture due to the detrimental impact of cyanide
on the floats. This can be overcome and substantial benefits may REFERENCE
still be achieved by operating two separate water circuits and/or Newson, T A and Fahey, M, 2001. Site investigations of saline gold
introducing a Detox circuit. tailings storages in Western Australia, The AusIMM Proceedings,
306(2).

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 233

Evaluating an Alternative to the Gravity Recoverable Gold (GRG)
Test
G K N Subasinghe1
ABSTRACT
Knelson concentrators are commonly used to recover fine liberated gold
in gold processing plants. The decision to employ a gravity gold circuit
consisting of centrifugal devices such as Knelsons is primarily taken on
the results of GRG tests, among other considerations.
While the GRG test is considered a material characterisation test,
which is useful in the design stages, its use in evaluating the performance
of an existing unit is questionable. The problem is exacerbated by the fact
that it is time consuming and costly.
This paper examines the validity of the GRG test for performance
evaluation and discusses alternative test work that can be performed to
evaluate the performance characteristics of an operating machine.
Principles that are useful in selecting optimal operating parameters, such
as optimal fluidising water flow rate for a given ore/machine combination,
are also discussed and illustrated with laboratory and plant data.
INTRODUCTION
Recovery of gold in gold processing plants involves the
liberation of gold and gold carriers by comminution followed by
separation processes such as cyanidation and/or flotation
techniques. The grinding characteristics of gold (Banisi and
Laplante, 1991) in the comminution circuit is such that the
liberated gold does not break into finer fragments due to its
malleable properties and gets concentrated in the classifier
underflow. It has been reported that the circulating load of gold
may be as high as 8000 per cent, while that of the ore is about
250 per cent. This provides an opportunity to recover the
liberated gold particles at a coarser size by means of a gravity
concentration technique. When a gravity recovery unit is
introduced into the circuit, the overall recovery of gold has been
known to increase appreciably. Other benefits of recovering gold
by gravity include increases in net smelter returns for the gravity
recovered gold, reduced gold losses in tailings due to over
grinding, avoidance of leaching coarse gold particles that require
higher residence times which restricts capacity, reduced reagent
costs, reduced carbon stripping and reactivation costs associated
with recovery of leached gold and reduced tailings disposal and
environmental issues to deal with resulting from decreased
cyanide consumption. It is also an alternative processing method
for preg-robbing ores.
It is well known that traditional gravity concentration
techniques are less effective on fine particles. The advent of
Knelson concentrators (KC) in the early eighties provided a
machine capable of recovering finer particles, resulting in gravity
recovery becoming an integral part of most gold processing
plants around the world and contributing to higher gold
recoveries. KCs exert a high centrifugal force on the particles
that enter the rotating bowl similar to that of a Knudson bowl, but
with the advantage that the particle bed formed within the bowl
is made to expand by the introduction of the fluidising water
through orifices on the outer periphery of the bowl, which
increases the recovery of heavy mineral particles.
Being a gravity concentrating device, KCs are more effective
when the gold is liberated at a coarser size. When the gold is not
liberated at a sufficiently coarse size or when it is locked up in
1. MAusIMM, Senior Lecturer in Mining Processing, Department of
Mineral Science and Extractive Metallurgy, Murdoch University,
Perth WA 6000. Email: n.subasinghe@murdoch.edu.au
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
carrier minerals such as sulfides, KCs are less effective, as the
amount of material that can be trapped inside the grooves of the
bowl is small. In such cases where higher mass recovery is
required, other gravity units such as in-line pressure jigs (IPJ) are
typically preferred.
Laplante, Woodcock and Huang (2000) devised a laboratory
procedure to determine the amount of gold in an ore that can be
recovered by centrifugal gravity methods and termed it gravity
recoverable gold (GRG). It is known as the GRG test and has
found wide acceptance in the mineral industry as a measure of
the amenability of an ore to gravity gold recovery. This paper
discusses the merits and drawbacks of the test and discusses an
alternative procedure to measure and improve the performance of
KCs used in gravity gold circuits.
THE GRAVITY RECOVERABLE GOLD
(GRG) TEST
The GRG test (Laplante, Woodcock and Huang, 2000) is
performed on a sample of ore prepared and tested under a
stipulated set of conditions in a three inch laboratory Knelson
concentrator (LKC). The sample size ranges from about 60 kg to
100 kg depending on the type of ore and associated gangue
densities. The test comprises three recovery stages. The feed to
the first stage is crushed and rod milled to 100 per cent passing
850 microns. The feed to the second stage is a sample of about
27 kg from the tailings of the first stage, ground to about 45 -
60 per cent passing 75 microns. The third stage treats about
24 kg from the tails of the second stage ground to about 80 per
cent passing 75 microns. The ore and fluidising water flow rates
used in the first stage are about 1200 g/min and 8 L/min
respectively, while those of the third stage are 400 g/min and
5 L/min. The corresponding feed rates for stage two are
intermediate between the rates used in stages one and three.
At each stage, all of the concentrates and about 600 g of the
tails are screened from 20 - 600 microns and each fraction is
assayed for gold. The tailings are wet screened first and fractions
coarser than 105 microns are pulverised prior to fire assaying.
The five coarsest size classes of stage one concentrate are
upgraded by hydrosising and subjected to microscopic
examination. The aim of the microscopic examination is to
identify the extent of non-GRG content present in this
concentrate due to insufficient liberation.
The stage recoveries are calculated based on the assays and
weights of the concentrates and the tailings of the third stage.
The cumulative amount of gold in the three concentrates is
expressed as a percentage of the total gold in the sample and is
considered the GRG content in the ore.
Features of the GRG test
The GRG test indicates the cumulative gold recovery obtained
when the feed is sequentially passed through a LKC in three
stages at progressively decreasing feed sizes. This is similar to
the classical release analysis procedure used in flotation. The
GRG test results indicate the maximum possible recovery under
the stipulated operating conditions in the LKC, which also
reflects the degree of liberation and breakage characteristics of
the ore.
235
G K N SUBASINGHE
Generally, the liberation size is indicated by the stage with the
highest gold recovery. A higher gold recovery in stage one
indicates a coarser liberation size and that at stage three indicates
a finer liberation size. Figure 1 shows typical cumulative
recovery curves resulting from GRG tests on different ore types.
The position and curvature of the plots indicate not only the
liberation size of the gold but also the ease with which it can be
recovered from the associated gangue particles. The shape of the
curve also offers insights into the possible recovery methods. A
GRG content less than 30 per cent indicates that gravity
concentration may not be an effective option; however, for
Cu-Au ores flash flotation may be a suitable option. If the GRG
content is above about 60 per cent and the shape of the curve is
more convex, which indicates a coarser liberation size, then a
high yield device such as an in-line pressure jig may be an
option. For intermediate GRG contents gravity concentrating
units such as Knelson or Falcon concentrators may be more
effective.
FIG 1 - GRG versus particle size for different ore types
(Laplante, Woodcock and Huang, 2000).
The GRG data has been used to evaluate gravity gold circuits
and determine the benefits of employing gravity concentration
devices such as Knelson or Falcon concentrators, as well as
evaluating alternative circuit configurations. This involves the
evaluation of recovery effort, which essentially is the product of
the process efficiency (viz GRG recovery) and the fraction of the
circulating load treated by gravity (Laplante and Xiao, 2001).
Drawbacks of the GRG test
The main drawback of the concept of GRG stems from its
definition. Gravity concentration devices achieve mineral
separation by applying forces on particles. Thus, particles
recovered in one machine may not necessarily be recovered in
another due to the differences in the extent of the forces applied,
the way they are applied and particularly if the feed size is
changed. According to the GRG concept, the percentage of gold
recovered in a three inch LKC from a stage-ground feed material
is taken as the GRG content. However, the feed material fed to an
industrial scale KC is of a different size and the separating forces
exerted by the machine on the particles are different to those of a
LKC. Hence the GRG content must be different in different
machines. The proponents of the GRG test believe that about two
thirds of the GRG recovered in a LKC would be recovered in an
industrial scale unit. However, there are industrial KC units of
various sizes that inherently recover different amounts of gold.
Thus, the concept of GRG must at least in part be machine
dependent.
The second drawback of the GRG test is that there is some
amount of subjectivity associated with the test itself. The
fluidising water flow rates in the three stages are not strictly
stipulated and can be varied according to the size of the feed
material (Guerney, Laplante and O’Leary, 2003). This would
mean that the GRG content would vary, particularly if the ore
hardness and the breakage characteristics of the ore are such that
it produces more fines during the staged grinding step. The GRG
236 Fremantle, WA, 19 - 21 March 2007
content of the softer ores are therefore more susceptible to
changes in the fluidising water flow rates used by individual
operators.
Another ambiguity arises in estimating the non-GRG fraction
of the coarse particles recovered in a LKC. It is subjective and
requires a highly skilled operator for the assessment.
While the GRG test results may indicate the amenability of an
ore to centrifugal gravity concentration, it would not be useful in
tuning an operating KC for maximum gold recovery due to the
differences in the operating conditions of the LKC and the plant
KC. In the light of variability in ore hardness characteristics
and/or grind size changes in the comminution circuit, the
characteristics of the feed particles that enter the KC will vary,
resulting in varying gold recoveries. The extent to which these
changes occur in a KC cannot, however, be extracted by standard
GRG tests.
ALTERNATIVE APPROACH
In assessing the efficiency of any mineral separation process, it is
essential in the author’s opinion that the material characteristics
or the susceptibility of the ore to separation be separated from
the machine characteristics (Kelly and Spottiswood, 1982).
Material characteristics are generally obtained by subjecting a
sample of the feed material, as is, to a near perfect separator of
similar type to that used in the separation. For example, to assess
the feed to a screening operation, a sample of the ore is subjected
to laboratory sieving, which is considered highly efficient.
Similarly, feed to gravity separators may be assessed by their
behaviour when subjected to heavy liquid separation. In some
instances in the absence of a near perfect machine, a carefully
stipulated test procedure may be used for assessment, such as,
the release analysis procedure for flotation. The separation
achieved in these tests is generally presented as separability
curves and they indirectly represent the degree of liberation of
the feed material. In this context, the GRG test is such a
procedure and hence may be considered a material
characterisation test.
On the other hand, the machine characteristics are obtained by
subjecting fully liberated material or artificially reconstituted
feed mixtures to the machine. The separation data is often
presented as performance curves, which are commonly known as
partition curves or Tromp curves. In assessing the performance
of an existing machine it is the machine characteristics that need
to be established first.
Coulter and Subasinghe (2004) proposed a method of
quantifying separation data of a Knelson concentrator by
analysing the forces acting on a particle within a KC. They
proposed a performance criterion (X), defined as the ratio of the
drag force (Fd) on a particle due to fluidising water to the
combined effect of centrifugal (Fc) and Bagnold (FB) forces,
which are functions of particle size, bowl rotational speed,
particle density and solids concentration. The complete
derivation may be found elsewhere (Coulter and Subasinghe,
2004).
Figure 2 shows a schematic view of how heavy mineral
concentration takes place within a KC. The feed material, most
commonly cyclone underflow or ball mill discharge, enters
through a central feed tube as a slurry. The slurry descends onto
the base plate at the bottom of the rotating conical bowl and is
thrown outwards as initially there is no centripetal force acting
on the particles to maintain a circular path. As a result, some of
the particles get trapped within the grooves of the bowl forming a
concentrate bed while some get carried upward into the tailings
stream by the rising current of water. Injection of water through
fluidisation ports located in the ribs of the bowl prevents
compaction of the concentrate bed. This creates a fluidised bed
(Harris, 1984), which acts as a concentrating chamber for heavier
gold particles under enhanced gravitational forces.
Ninth Mill Operators’ Conference
 EVALUATING AN ALTERNATIVE TO THE GRAVITY RECOVERABLE GOLD (GRG) TEST

FIG 2 - Schematic view and forces acting on a particle within the KC concentrate bowl.

For a particle that gets trapped in the bowl and traverses a Volume retained versus new performance
circular path, a dynamic equilibrium would be established criterion for uniform material
between the forces as described by Equation 1.
Volume Retained (cm 3 )
100

Fc = FB + Fd (1)

where: 10

the centripetal force Fc = mrω2 (2)

1 1
the fluid drag force Fd = C D A ρ f Vf2 (3)
2 1.00E-07 1.00E-06 1.00E-05 1.00E-04 1.00E-03

2 Performance Criterion
 dV 
and the Bagnold force FB ∝ d 2 , ρ p , λ ,   (4) 596 Quartz 253 Quartz 421Quartz 98 Magnetite
 dy  98 Quartz 777 Quartz 106 Quartz

The imbalance of the forces under non-equilibrium conditions FIG 3 - Volume of material retained as a function of the
may be quantified by a parameter X, defined as: performance criterion.

Fd material collected in a LKC when the feed material is comprised

X=
Fc∗
where:
Fc* is the net force due to Bagnold forces and the centripetal
force
This implies that each particle that enters the bowl experiences
the net effect of the above forces, quantified by X, which
determines the probability of retention of that particle within the
bowl. Particles with smaller X values have a high probability of
retaining in the bowl while those with high X values will be
mostly rejected and report to the tailings. This approach enables
one to determine a performance curve for the machine and use it
to determine the optimum fluidising water flow rate to suit the
nature of the gangue and gold particles present in the feed, ie
their size, density and size distribution. It also helps to gain
insights into the fluidised state of the bed under different
operating conditions.
OPTIMISATION OF KNELSON CONCENTRATOR
PERFORMANCE
Particle recovery
Coulter and Subasinghe (2004) showed that the amount of
concentrate recovered in a KC depends on the performance
criterion X described above. Figure 3 shows the amount of
(5)
entirely of uniformly sized single mineral particles of various
sizes.
At low X values, the particle bed is more compacted against
the outer perimeter of the bowl while at larger values of X the
bed transforms to a more fluidised state and recovers less
material. The transition point is characterised by X*.
If the Y axis in Figure 3 is normalised by the amount of
material fed, then it represents the probability that a particle
would remain within the bowl and hence it may be considered
the performance curve of the machine. When a mixture of
particles of different sizes and/or densities is fed, the amount of
each material recovered in the bowl may be described by a model
of the form:
  X  n
Vcal = Voi fi exp  −    (6)
 X  
 ∗ 
where:
fi is the fraction of mineral species i in the feed
V0i is the amount of mineral i recovered when it is fed alone
Figure 4 shows the observed versus calculated volumes of
material collected in the concentrate of each individual mineral
component in the mixtures using Equation 6. The data covered a
wide range of material such as sand of different size fractions,

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 237

 G K N SUBASINGHE

Calculated volume versus observed volume TABLE 1

piece-wise defined model
Separation data of a mixture at various operating conditions.
Calculated Volume (cm 3)

40

30 Probability of retention
Run No Magnetite Silica 1 Silica 2 Silica 3
20 1 0.05 0.013 0.142

10 2 0.225 0.066 0.153

3 0.275 0.082 0.142
0 4 0.025 0.003 0.003
0 10 20 30 40 5 0.075 0.029 0.174
Observed Volume (cm 3 )
6 0.175 0.058 0.161
Single Size Mix Silica 7 0.0375 0.132 0.108
Density Mix Silica Density Mix Magnetite
Size and Density Mix Silica Size and Density Mix Magnetite 8 0.125 0.145 0.079
9 0.275 0.129 0.092
FIG 4 - Calculated versus observed values for all data from
10 0.018 0.016 0.013
Equation 6.
11 0.025 0.103 0.061
magnetite, tungsten of different sizes and in various proportions. 12 0.15 0.174 0.071
More details of the results may be found elsewhere (Coulter and
Subasinghe, 2004; Sargent and Subasinghe, 2006). The data
fluidising water flow rate, bowl rotational speed, size distribution
correlation indicates the adequacy of the model.
and the densities of the particles comprising the feed, a dynamic
Table 1 shows the observed probability of retention of each equilibrium is established within the bowl. The gold particles
component when mixtures of sand and magnetite particles were
that enter the bowl then have to segregate through the established
separated under different fluidising water flow rates and bowl
rotation speeds. dynamic bed while displacing some of the gangue particles. This
segregation process of heavy particles penetrating the bed
Figure 5 shows this data plotted as probability of retention comprising lighter gangue has been shown to be a first order
versus X values for each component. It can be observed that the
process (Sargent and Subasinghe, 2006) and that the heavy
recovery of heavy mineral (magnetite) is highest for runs 2, 3
and 9. At low X, the bed is too compact and only the heavy mineral accumulated in the bowl reaches a saturation point. The
mineral particles that are small enough can segregate and enter optimal cycle time would be dependent on the rate constant of
the bed. At high X, the bed is more fluidised and the material this segregation process, which in turn depends on the grade of
collected is less. It may be postulated that the recovery of heavy the heavy mineral in the feed, bowl capacity and the segregation
mineral is dependent on the nature of voids present in the kinetics which are functions of the feed composition and
dynamic bed, which in turn is dependent on the size and size operating variables.
distribution of the feed and operating conditions. Thus the Figure 6 shows the saturation volume of heavy mineral
optimal fluidising water flow rate, which maximises the gold collected as a function of the heavy mineral content in the feed
recovery from a given feed, could only be determined by and Figure 7 shows the variation of the rate constant of
conducting test work on a sample of the original plant feed and segregation as a function of the performance criterion X.
not on its ground product as is the case with the GRG test.
Based on the concepts discussed above, a computer model to
Cycle time optimise the KC operating parameters such as fluidising water
flow rate and cycle time, which takes into account the feed
When the feed enters an operating KC, the concentrating characteristics and operating parameters, is being developed to
chamber in the bowl is filled with gangue. Depending on the maximise gold recovery.

Probability of Retention Vs X
Mixture: silica of sizes 98, 421 and 596 micron and magnetite of size 98 microns
1.00E-06 1.00E-05 1.00E-04 1.00E-03 1.00E-02 run1
1 run2
highest recoveries of magnetite
run3
run4
0.1 run5
run6
run7
run8
0.01
run9
run10
run11
0.001 run12

FIG 5 - Separation of a mixture of particles in a LKC under various operating conditions.

238 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 EVALUATING AN ALTERNATIVE TO THE GRAVITY RECOVERABLE GOLD (GRG) TEST

Saturation volume versus feed grade CONCLUSIONS

10.0
Saturation Volume (cm3 )

It has been shown that the GRG test is a material characterisation

8.0 test which provides information on the degree of liberation of the
ore, the influence of breakage characteristics on liberation and
6.0 amenability of ore to centrifugal gravity concentration, which are
useful at the design stage. However, performance optimisation of
4.0 existing operating units requires an approach based on machine
characteristics. Such a method that takes into account the forces
2.0 acting on particles has been developed and also shown how it can
be used to evaluate optimal fluidising water flow rates of a
0.0 Knelson concentrator depending on feed characteristics and other
0.0 1.0 2.0 3.0 4.0 5.0 6.0 operating parameters.
Feed Grade (wt. % Magnetite)
REFERENCES
8 L/min 12 L/min 16 L/min
Banisi, S and Laplante, A R, 1991. The behaviour of gold in Hemlo
Mines Ltd grinding circuit, CIM Bulletin, 84(955):72-78.
FIG 6 - Saturation volume as a function of magnetite content.
Coulter, T and Subasinghe, G K N, 2004. A mechanistic approach to
modelling Knelson concentrators, Minerals Engineering, 18:9-17.
First order rate constant versus performance Guerney, P, Laplante, A R and O’Leary, S, 2003. Gravity recoverable
criterion for magnetite gold and the mineral liberation analyser, in Proceedings 35th Annual
0.8 Meeting of Canada Mineral Processing, Ottawa, January,
pp 401-416.
Capture Rate k (min -1 )

0.7
0.6
Harris, D, 1984.The Knelson concentrator – Applications in Australia, in
Proceedings Gold-Mining, Metallurgy and Geology Conference,
0.5 pp 101-106 (The Australasian Institute of Mining and Metallurgy:
0.4 Melbourne).
0.3 Kelly, E G and Spottiswood, D J, 1995. Introduction to Mineral
Processing (John Wiley and Sons).
0.2
Laplante, A R, Woodcock, F C and Huang, L, 2000. A laboratory
0.1
procedure to characterise gravity recoverable gold, Trans of Society
0 for Mining, Metallurgy and Exploration Inc, 308:53-59.
0.E+00 1.E-05 2.E-05 3.E-05 4.E-05 5.E-05 6.E-05 7.E-05 8.E-05 9.E-05 1.E-04 Laplante, A R and Xiao, Z, 2001. Optimizing gravity recovery: The role
Performance Criterion of the recovery effort, in Proceedings 33rd Annual Meeting of
Mineral Processing, Ottawa, January, pp 371-388.
1wt. % Magnetite 3wt. % Magnetite 5wt. % Magnetite
Sargent, D and Subasinghe, G K N, 2006. Selecting optimal operating
conditions for Knelson concentrators, in Proceedings Metallurgical
FIG 7 - The first order capture rate constant as a function of the Plant Design and Operating Strategies 2006, pp 274-289 (The
performance criterion. Australasian Institute of Mining and Metallurgy: Melbourne).

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 239

The Importance of Geometallurgical Analysis in Plant Study,
Design and Operational Phases
D David1
ABSTRACT
If an orebody has been drilled to achieve some level of regulatory
compliance, then a number of associated geological databases will exist,
representing 10 000 to 100 000 m plus of core intervals. The databases
usually contain data about assays, lithology, mineralisation and
geophysical attributes. They will also include location information such
as collar positions and downhole surveys. Obviously this data is collected
by geologists for, predominantly, geologists and mining engineers.
However, from a process design viewpoint process engineers usually
ignore the geology where possible and rely on a small number of
metallurgical test samples. Typically these test samples are sourced from
convenient and available cores or from a small number of specially
drilled metallurgical holes. One of the worst instances known to the
author was the design of a SAG mill using only two SAG specific test
results (one for each of the major ore types) while, geologically, the
orebody was defined by more than 30 000 m of drilling.
This paper sets out a philosophy to maximise the use of the geological
data, both for process design and in preparation for the commissioning
phase of projects. Geological data is used in the identification of
preferred locations for metallurgical test samples, for checking the
metallurgical validity of the geological ore-type definitions, where
necessary developing new ore-type definitions (geometallurgical
domains) and developing mathematical relationships for the estimation of
important metallurgical factors across the geological database. If the
estimation of factors (for example ore competence) is considered reliable
enough then there is also the possibility of feeding calculated data back
into the block modelling process to allow the generation of
metallurgically relevant outputs in the mine plans.
INTRODUCTION
The use of geological data for optimisation of operating plants
has become a significant part of the modern process mindset.
In recent times this type of activity has been performed under
the banner ‘Mine to Mill’, ‘Mine to Port’, ‘Mine to Metal’ or
‘Mine to Market’. The underlying principle is to use spatial
metallurgical information in a feed-forward mode to drive
production planning, mine planning, blast design, blending
strategies and plant set-up.
Unfortunately the definition of how such an analysis is to be
carried out, together with how the results should be implemented
as a routine part of day-to-day site operations, is less clear.
Typically, the implementation of detailed ‘Mine to Mill’
optimisation strategies (such as tailoring blasting parameters to
rock mass properties on a blast by blast basis) is put into the
too-hard basket unless there is a site champion (with the requisite
authority, understanding and energy) driving the process. There
are some encouraging success stories in the field of Mine to Mill
with recent examples provided by Burger et al (2006) and Dance
et al (2006).
The reasons for less-than-successful implementation are many
and varied. At some sites, the outcomes of the Mine to Mill
investigations have shown only marginal potential for process
improvement. At other sites the data collection, data analysis and
process control systems are not sophisticated enough to facilitate
implementation. At other sites the potential benefits from
implementation may be substantial but are smaller than the
1. MAusIMM, Process Consultant, GRD Minproc Ltd, Level 14, 140
St Georges Terrace, Perth WA 6000.
Email: dean.david@minproc.com.au
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
benefits to be gained by addressing fundamental design or
operational problems.
However, at sites where benefits are highly likely or have been
demonstrated by testing, the most significant impediments to
implementation of a workable system are:
• a lack of useful orebody domain definitions,
• a lack of relevant metallurgical test data, and
• a less than adequate understanding of critical issues in the
disciplines outside of the native discipline of the site
champion.
These three issues are usually obvious on an established
operating site. However, they are almost always present at the
beginning of process design. By addressing these issues in the
early life of the orebody it is possible to entrench constructive
mine-to-mill thinking in the design philosophy.
AIM OF PROCESS DESIGN
When designing a plant the owner has a number of objectives.
These include:
• minimise capital cost,
• minimise operating cost,
• minimise schedule time to production,
• have a robust and effective process,
• repay borrowings/generate shareholder returns as soon as
possible, and
• make a profit.
When designing the same plant the engineer also has a number
of objectives, not all of them common with the owner:
• optimise exposure to contractual risk,
• minimise technical risk,
• maintain reputation,
• meet the client’s expectations, and
• make a profit.
The main asset (and the only asset in a greenfields project) is
the orebody, as defined by the physical cores and the library of
core data. It is the responsibility of the process designer to
understand that asset and propose the best investment strategy for
the owner.
TYPICAL PROCESS DESIGN
Experience in the analysis of numerous operational and design
data sets, from both greenfields and operating sites, has shown
that the actual influence of the geological data upon process
design is generally limited (or missing), and often misdirected.
Typical approaches to conducting a metallurgical test program
for process design follow a well trod path:
• Determine the major ore types, usually by geological
definition such as lithology.
• Source samples ‘representative’ of the ore types, typically
by drilling metallurgical holes at strategic locations.
Metallurgical sample holes are typically larger diameter
241
D DAVID
(63 mm PQ and above) compared to exploration holes (often
48 mm HQ, or below). The aim of drilling metallurgical
holes is to generate significant sample mass to generate ore
type composites and allow an extensive test program to be
carried out on a single sample.
• If pilot plant trials are to be conducted, source a large sample
of ore (20 to 100 t).
• For a ‘comprehensive’ pilot test program, source tonnage size
samples of each major ore type.
• Conduct characterisation testing to define the properties of
the major geological ore types.
• Possibly conduct tests on ore blends representative of mining
time periods.
• Conduct a variability test program making sure that all ore
types are represented a number of times.
The data from this test program is then analysed to estimate
critical design factors including SAG and ball mill specific
energies, crusher design parameters, flotation recoveries and
leach recoveries. These values are used to set process design
criteria and then to size process equipment.
There are a number of significant issues inherent in this
approach, all of which can detrimentally impact on the project.
There is a high probability that the ore types determined on the
basis of geological classifications have little or no metallurgical
relevance. This leads to a lack of metallurgical definition
between ore types because the ranges of the metallurgical
properties for each lithology overlap and they are highly variable
within an ore type. Acceptance of the domains and associated
average test results as definitive regularly creates confusion and
myths about ore type properties that mislead designers and,
ultimately, operators.
The sourcing of specific ‘metallurgical’ samples, by drilling a
few large diameter holes, limits the spatial distribution of what is
considered the most definitive set of test results available for
design. This is a hit-and-miss approach based on the twin
assumptions that the ore type definitions are meaningful and that
the selected locations for the holes will provide ‘typical’
examples of those ore types.
In focusing on ‘metallurgical’ samples, the much larger
resource of well-distributed samples in the core library is often
ignored. The ability to utilise these samples in a meaningful
manner is strongly influenced by the choice of characterisation
test methods and the available processing options for the type of
ore (ie flotation for base metal sulfides or leaching for a
free-milling gold ore).
Core samples extracted for metallurgical purposes often
bypass the normal geological characterisation steps because the
overriding aim is to use 100 per cent of the core for test work.
There may also be other factors, such as immediate sample
refrigeration to minimise possible oxidation, that prevent
meaningful characterisation. It is then largely a matter of
assumption that the samples do actually represent typical
examples of the major ore types. By the time this assumption is
checked the samples have usually been transported large
distances and been through a number of processing steps. If the
assumption proves incorrect, then the ability to extract additional
sample is likely to be limited by drill rig availability and drill
target priority. Even more problematic, if ore type domain
assumptions prove incorrect, then there is a similar likelihood
that the second set of samples will also be unrepresentative.
Tonnage-scale samples extracted for pilot testing are, for many
reasons, likely to be unrepresentative. Because large masses are
required for pilot testing it is necessary to either open up the
orebody or conduct a program of drilling large diameter holes.
For hard rock deposits, a drilling program for pilot work would
typically consist of core in the range 83 mm to 150 mm in
242 Fremantle, WA, 19 - 21 March 2007
diameter. For shallower and softer ores (eg nickel laterites) pilot
samples are often collected using a 1 m diameter Caldwell drill.
The largest tonnage pilot samples are usually extracted by
opening up the orebody by trench, trial shaft, winze or adit
mining. The choices of locations for accessing the orebody are
limited by geology, topography, timing and cost. Typically, bulk
pilot samples are representative of the conveniently available ore,
rather than representative of the orebody or an ore type. Of the
available methods for extracting pilot plant feed samples, the
most representative is large diameter drilling at a relatively large
number of locations.
If samples of individual ore types are extracted for pilot
testing, the same representation problems described above apply.
However, the problems are magnified because each ore type
makes up only a proportion of the orebody and some ore types
may be more accessible than others.
Characterisation of ore types by test work is typically
restricted to what are seen as the essentials. If large diameter
cores are extracted specifically for drop weight testing (DWT),
for example, a decision is often made, especially on smaller
projects, that there is no need to conduct variability testing in
respect to SAG mill competence. In the case of SAG milling, this
means that the mill design will have limited (or no) variability
information. As the SAG mill is one of the highest capital cost
items in the circuit, and a likely throughput bottleneck, this is a
short-sighted approach.
The preparation of time-period blends is often done with the
available samples or is done mathematically using the
‘representative’ ore-type data. Again this approach is
questionable because it relies upon the assumption that the
available or tested samples of the ore types are representative.
Unfortunately, even if the available ore type samples are
‘typical’, the chances that the ore type is consistent throughout
the orebody are low. Often there are significant trends in
breakage properties, even within well defined ore types, that are
related to factors independent of the geological divisions.
DATA COLLECTION DRIVERS
All of the above situations, resulting from typical design
practice, are regularly encountered by the author. Unfortunately
such situations are often discovered after commissioning and
during an investigation into the reasons for a circuit failing to
meet performance expectations. Because fixing the problems
during operations is common enough to be considered normal,
the fundamental question that needs to be addressed is: ‘What
are the reasons that owners and process engineers are satisfied
with such risky practices being applied during the design phase
of a project?’
The extraction of metallurgical test samples and the generation
of results both suffer from a number of inconvenient timing
issues. The problems are mostly related to the reasons the data is
collected in the first place.
One of the biggest barriers to conducting an appropriate
metallurgical test program is the fact that, in the eyes of the
owners, the value in the orebody has already been established to
the satisfaction of the market and the regulators by complying
with the appropriate reporting code. Compliance with these
codes (such as JORC) requires that an extensive and expensive
drilling program be conducted. In the value establishment phase
the owner is rightly focused on the reporting of resources and ore
reserves. A positive market response to the geological data places
a present value on the resource, a value that incorporates
estimates of capital and operating cost. However, as the purpose
of getting the market to respond is to raise funds for a feasibility
study (or establish value for a sale), the cost estimates must
always be dominated by assumption.
Ninth Mill Operators’ Conference
 THE IMPORTANCE OF GEOMETALLURGICAL ANALYSIS IN PLANT STUDY, DESIGN AND OPERATIONAL PHASES
By the time a process study team has a chance to look at the
orebody there are already a large number of process assumptions
in place. There is also a timeline for development (often
unrealistic) and marketing may also be underway for
concentrates. There is no denying that all these factors are
essential for a successful project. However, if there is no
allowance for adequate process development in the timeline, then
the project risk increases significantly. In some instances,
metallurgical test work is seen as a necessary evil that simply
delays the time to plant commissioning.
As there are no regulatory codes governing process
development reporting, the owner returns to a mode where
standard market rules of disclosure apply. This results in the
process development stage being treated quite differently to the
exploration phase by both the owner and the market.
Another key driver for data collection is the recognised
complexity of the orebody or the process. If the orebody has
obvious variability issues or contains potential problem elements,
often it is recognised in the exploration phase and the exploration
budget is expanded accordingly. The additional exploration
requirements often alert the owners to seek expert process advice
early.
A further driver is the complexity of the known process. For
example, in the case of low grade nickel laterites the complexity
of the PAL (pressure acid leach) process is well recognised and a
significant budget and time frame are usually allocated for
process development.
Conversely, if the process is recognised as simple (gold CIL,
copper flotation, etc) then a relatively small process development
budget is set and the development time frame is compressed. It is
these projects that often face the greatest risk if the process
development allowance (both budget and time) is inadequate.
Additional test work that causes a schedule delay is always
bad news. The later in a project, the worse it is received. The
process engineer must either live with the imposed schedule or
attempt to extend the schedule by a number of months so that
adequate test work, or even pilot testing, can occur. The latter
approach is rarely successful and the outcome depends mainly
upon the magnitude of the identified risk in the context of the
project. The outcome is also strongly dependent on how
successfully the process engineer has communicated the issues to
the owners.
There are many factors that have a high probability of forcing
the job of the process engineers to fit into an inadequate time
frame and budget. Opposing this, the process engineer must
consider contractual and reputation risk exposure for the
engineering company should the circuit fail to meet expectations.
DOMAINS
The fundamental building block of geometallurgical analysis is
the domain. For the purposes of this paper, a domain is an
exclusive classification of ore by a common property or set of
properties.
Lithological domains
The simplest domain set is lithologically based, typically
assigned by the geologists early in the exploration phase. The
common link is the rock type assignments. Typically, some
lithologies will also be described as ore and some as waste,
providing a rudimentary metallurgical basis to the domain
allocations.
In many projects this is where the classifications start and end.
The description of the domains by lithology lasts for the life of
the project regardless of the appropriateness for the process
engineer.
In some instances lithological domaining is eventually
recognised as being inadequate, but the solution offered is worse
than the initial problem. The domains that are found to have
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
‘similar’ metallurgical properties are simply grouped together
giving fewer equally uninformative domains.
The definition of ‘similar’ in the context of domain assignment
is often misleading. The metallurgical similarities between
lithological domains often exist because each of the lithologies
exhibits a similar, but broad, range of metallurgical properties.
The root cause of the problem is usually that the lithological
domains are not appropriate and an unrelated set of
metallurgically relevant domains can be found within each of the
major lithologies.
To simply group the lithologies together is to ignore the fact
that the lithological domains may be meaningless in a
metallurgical sense and an inappropriate starting point. The
ultimate result of this grouping process is to discard the
geological ‘baby’ with the geological ‘bathwater’, potentially
ignoring the majority of the vast database of geological
information. Having declared the lithological domains redundant
the next unfortunate step is to create a metallurgical database,
independent of the geology. Typically this involves large
numbers of specialist tests designed to map the metallurgy of the
deposit.
This is not geometallurgy. It is essentially rendering the
geological database to being a source of assays to complement
the new and expensive metallurgical database.
Rather than improving the relationship between process
engineers and geologists, this process has the strong potential to
build a new barrier. A new set of metallurgical terms is
introduced that must then be imposed on the geologists and
miners that, unfortunately, have no geological relevance and little
mining relevance.
Geometallurgical domains
To assign geological domains requires a geologist. Assigning
metallurgical domains requires a metallurgist. To assign geo-
metallurgical domains, by extension, requires a geometallurgist.
A geometallurgist, in the majority of cases, should start out
being a metallurgist or process engineer. The aim is to assign
domains that are metallurgically relevant while utilising the
geology. This is best done by someone who first understands the
process intimately and, secondly, has the capacity and experience
to understand the geology, especially the process implications of
the geological data.
The first step in assigning geometallurgical domains is having
an understanding of what is important to the process. Each process
has its own characteristics. A few of these are listed in Table 1.
TABLE 1
Important process issues to understand on an orebody-basis
for a number of circuit types.
Circuit type Critical properties
SAG mill circuit • Competence
• Grindability
Stage crush and mill circuit • Crusher work index
• Grindability
Iron ore (haematite) • Lump/fine ratio
• Impurity levels (S, P, etc)
Copper sulfide ore • Copper mineralogy
• Alteration
• Impurity levels (F, As, etc)
Gold ore • Gravity recoverable gold
• Cyanide recoverable gold
• Refractory gold
• Copper grade
• Preg-robbing properties
Nickel laterite • Beneficiation characteristics
• Acid consumption
• Slurry viscosity
243
D DAVID
Few of these properties are measured directly in the geological
analysis of the core intervals. However, a large number of these
properties have the potential to be inferred by using one or more
of the pieces of information within the geological database.
The second step is to select a number of samples from
throughout the orebody that represent a reasonable spatial
distribution and cover the lithological domains. Priority must
also be given to testing ore that is to be treated early in the life of
the operation. To select the test samples it is essential to gain an
understanding of the most likely mine plan together with
alternative plans that may gain ascendancy during the study
phase of the project.
The number of samples to be selected will be determined by
the complexity and size of the orebody. The number should be
high enough to allow a wide spread of likely ores (and some
probable dilution waste) to be tested. It is much better to test
localised samples than to generate composites and test these. The
localised samples can be directly related to the geology and will
have a higher natural variability than composites. A minimum of
20 samples should be metallurgically tested as this should
provide a basis for statistical analysis and pattern analysis.
The selected samples must then be either extracted from the
orebody or sourced from within the available core library. Given
that the project schedule is usually very high priority, the focus
should be on sourcing samples already extracted from the
orebody as core. In addition, it is essential to ensure that the
samples have been subjected to the full suite of geological
analyses. Note that RC drill samples are generally unsuitable for
this test program unless the critical properties are independent of
comminution characteristics and are unaffected by a potential
loss of fines during extraction.
The samples are then metallurgically tested to determine the
critical parameters. Critical properties are typically those that
determine throughput, final product rates and saleability of
product. Increasingly, some environmental properties can also be
considered critical in some situations. Critical properties relate
directly to key process design criteria. The most appropriate
metallurgical tests are standard procedures that provide
fundamental characteristics rather than flow sheet specific
procedures that may be subject to unintended variability. For
example, it is better to conduct rougher only flotation tests that
determine floatable and non-floatable ore components than to do
a complex rougher/cleaner procedure that measures the response
of the ore to the test conditions regardless of characteristics such
as feed grade or mineralogy.
After the metallurgical testing is completed and the results are
analysed the process engineer will have available for analysis a
set of geological data and an associated set metallurgical data.
The most important investigative work from a geometallurgical
viewpoint follows this milestone.
Geometallurgical investigation
The geometallurgical investigation phase is an examination of
data relationships. The first step is to look for correlations
between the geological data sets and the related critical
metallurgical test results. This can be done either visually, using
graphs, or using statistical tools such as analysis of variance
(ANOVA) tables. Usually it is better to look at the data
graphically because this allows the early detection of non-linear
relationships, it allows outliers to be identified and investigated
and it allows for the detection of incorrect or assumed data.
To determine if the existing geological domains are appropriate
the metallurgical results are grouped into their lithologies and
examined for clusters and patterns. If the critical properties
overlap in a number of the geological domains then it is highly
probable that the geological domains are metallurgically
meaningless.
244 Fremantle, WA, 19 - 21 March 2007
From this point on, the analysis can take a number of
pathways. If the lithological domains do have metallurgical
relevance then lithology may be sufficient for geometallurgical
domain definition. Unless the relationships between metallurgical
parameters and lithology are highly correlated it is prudent to
continue to search for additional correlations, such as with grade,
alteration, etc. If there are a number of other geological data sets
that appear to influence the metallurgy then the relevance of
these can be examined using techniques such as multiple linear
regression and non-linear data fitting.
If the lithology has no relationship to the properties, and the
metallurgical properties are variable beyond the limits of
experimental error, then the search for relationships must be
wide ranging. Rather than adopt a trial and error approach there
are some logical starting points:
• Ignore geological data sets that show little or no variability as
they are unlikely to relate to a variable metallurgical property.
• Unless there is a logical reason or a probable correlation it is
usually safe to ignore trace elements.
• Analyse the metallurgical data set for clusters and determine
if any geological data may be a predictor for those clusters. If
distinct clusters are found for the metallurgical properties
then it is likely that a basis for geometallurgical domaining
has been found.
Important geological factors to incorporate in the investigation
include:
• major valuable elements;
• related elements such as sulfur;
• major non-valuable elements (typically Si, Ca, Mg, Al and
Fe);
• ratios between elements (eg Cu:S, Mg:Si, Zn/Fe, Fe:MgO);
• alteration reporting – note that alteration reporting may be
difficult to interpret, especially if there are related alteration
types that should be combined for metallurgical purposes;
• core fracture data such as RQD (the percentage of core in a
given length that is in pieces greater than 100 mm in length,
a potential pointer to competence), fracture frequency and
point load test results; and
• other potentially useful data such as colour, Mohs hardness,
depth, northing, easting, etc.
From an understanding of the physics of the process,
determine all the factors likely to contribute to the metallurgical
property in question and begin with these, both in isolation and
using multiple regression techniques.
Look for unusual non-linear data dependencies that may mask
an important relationship. For example, a relationship may be
valid below a certain point but becomes scattered or constant
above that point. Recovery by flotation is one property that is
typically low at low feed grades, increases to a maximum at a
higher feed grade and then stays at the maximum at still higher
feed grades. Assuming it exists, a graphical analysis of flotation
test recovery versus core grade would reveal this relationship and
the correct response may be to apply a constant recovery to cores
above a threshold grade. Below that threshold a mathematical
relationship between core assay and recovery can be applied.
Look for relationships within geological groups, other than the
lithologies, such as alteration types. These relationships may be
distinctly different between geological groups and may need to
be described by a number of independent equations which, for
example, are invoked depending on the alteration type.
Hopefully at the end of the investigation one or more
predictive relationships with correlation coefficients (R2) greater
than about 0.75 have been found. Alternatively, clustered
metallurgical properties have been related to geological
predictors, which may or may not be numerical.
Ninth Mill Operators’ Conference
 THE IMPORTANCE OF GEOMETALLURGICAL ANALYSIS IN PLANT STUDY, DESIGN AND OPERATIONAL PHASES
The correlation coefficient is a useful value for screening
possible predictive relationships. It is most appropriate when
comparing alternatives for the prediction of a common
parameter. It is less useful when comparing the prediction of one
parameter with another.
Judgement of the prediction outcome is best done by
evaluating the standard error, or the standard deviation (SD) and
then calculating the relative error or relative SD. Relative error or
SD both indicate the inaccuracy that can be expected when using
the predictive method.
At this point the data relationships are mostly theoretical and
they must be verified. It is essential that the relationships are
checked to ensure that they are valid beyond the data set from
where they were developed. It is also essential to check the effect
of extrapolating the relationship beyond the limits of the base
data ranges.
Verifying geometallurgical relationships
Potential geometallurgical relationships are relatively easy to
verify, mainly because of the large size of the geological
database.
However, the first verification step is to check the standard
error for your prediction within the metallurgical database. If the
standard error is large compared to the accuracy you require (for
example if your prediction method for Bond ball mill work index
has a standard error of 5 kWh/t when it is desirable to predict
within 1 kWh/t) then a prediction basis has not been established.
If the standard prediction error within the data set is acceptable
then the predictive method should be applied across the whole, or
part of, the geological database.
As the geological database is not a single database but a
collection of related databases, it is first necessary to arrange the
data into an analysable form. The problem encountered for
spreadsheet users is that values such as assays are typically
measured every metre, SG may be once every 20 m, RQD is
likely to be by the core tray and lithology is likely to be by
lithological boundaries. To facilitate analysis of large geological
databases it is necessary to normalise the data set to a common
interval basis. The most appropriate basis is the assay basis,
usually intervals of 1 or 2 m. Geological analysis software is
generally able to generate normalised databases in text or Excel
format directly from the geological data sets.
Once all the geological data is in one place it is important to
clean up the database. Unfortunately, as it is not possible to
predict all the problems and errors that may exist in such a
database, the clean-up process is something that is difficult to
automate. The database must be examined for blank records,
missing critical data, duplicated data, inconsistent data (ie a mix
of ppm and per cent values in the same column) and many other
varieties of error.
Use a feature such as conditional formatting, data sorting and
data filtering to quickly identify data points that lie outside
the possible measurement range (eg assays >100 per cent,
SG values >5). Once a problem is identified, action can be taken
to either delete or repair the record. Repair within the
geometallurgical database is only possible in obvious cases, such
as where ppm and per cent are mixed as mentioned above. Any
instances where repairs have been made should be documented
and the owner of the geological database informed. Ideally, the
owner of the geological database will correct the errors at source
and re-issue the database.
Once a clean database has been developed it is then possible to
conduct predictive verification and domain allocation.
To set up a predictive verification of the models it is necessary
to apply the predictive technique across the entire database. This
may require the creation of numerous calculated fields. Once the
predicted properties are in place it is best to use relatively
sophisticated techniques, such as pivot tabling in Microsoft
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
Excel, to extract, group, count and average data within potential
geometallurgical domain definitions.
If a chosen system of geometallurgical domains is to be
metallurgically useful then it is necessary that each of the
domains exhibit distinct clusters of metallurgical properties. If
the clusters are diffuse and overlap then the chosen domains are
not likely to be any better metallurgically than the original
lithological domains.
If distinct and valid data clusters are not found then the
predictive technique is not verified and the analysis must either
be repeated or abandoned.
On the assumption that a reasonable prediction method has
been established, and that a reasonable basis exists for
geometallurgical domaining, it is then possible to utilise this in
the design and mine planning process. The advantage of
following this methodology is that the existing geological
database becomes a metallurgical resource and there is also
potential to incorporate some of the metallurgical findings back
into both the geological modelling and the mine planning
processes. In addition, the process engineer has gained an
excellent picture of the orebody geology and the geologist has
probably learned a large amount about what is important for the
future process plant.
APPLYING GEOMETALLURGY
The engineering application of the geometallurgical findings can
be the difference between success and failure or the difference
between a good result and an excellent result.
Study phases
The study phases of a project are the most crucial as they
determine the form of the final design. Investigations during the
study phases are, in the context of the project, inexpensive. It is
of interest to note how it is relatively easy to justify funds to fix
an operational problem when it could have been solved for a
fraction of the price during the study phases.
Options study
An options study compares, at a high level, the various options
available for extraction and processing of the orebody and the
options for marketing the products. At the options study level it
is not necessarily useful to have too much information because
this can slow the evaluation process and mask the high level
issues. However, in the options phase preparation must be made
for the next study phases, the prefeasibility and the definitive
feasibility studies. However, money for an extensive metallurgical
test program may not be available at this point. Funding is
generally dependent upon the outcome of each study phase.
During the options study phase it is important to be alert to
early indications of disconnects between the geology and the
metallurgy. For example, if the lithologies seem to have similar
metallurgical properties or if ores defined as the same ore type
have vastly different properties then geometallurgical analysis is
probably essential.
A key outcome from an options study phase should be a set of
preliminary investigations designed to test geometallurgically
relevant possibilities. This is likely to include extracting more
samples from each of the lithologies and extracting some key
samples from significant zones that appear to be different to
those already tested.
The outcome from the options study phase should be a short
list (preferably only two alternatives) of economically attractive
options.
Prefeasibility study (PFS)
The prefeasibility study is intended to arrive at a preferred option
out of the shortlist and to begin flow sheet definition. The level
245
D DAVID
of detail required is related directly to the accuracy of the
estimate required. Typically the prefeasibility estimate accuracy
is in the region of ±25 - 30 per cent.
A geometallurgical view of the orebody provides the engineer
with additional process selection drivers not necessarily
considered in a more superficial design procedure. For example,
when choosing between single and two-stage milling options a
detailed understanding of the ore types, their variability and the
stability needs of the downstream process are essential to ensure
the decision is technically correct. In situations where the grade,
mineralogy and hardness variability of the ore is significant but
the downstream process is robust over a wide range of operating
conditions then a single stage mill is the most likely
recommendation. However, when the process requires stability
and accurate control over particle size then a two-stage mill is
the likely recommendation, together with an ore blending
strategy.
Because the required geometallurgical testing burden is small
compared to some other methodologies, the majority of the
required test program can be performed during prefeasibility. If
correctly planned during the options study phase, and if focused
on available samples rather than new core, the total test program
can be completed in two to three months. In many cases, the
resulting data set will be adequate not only for the prefeasibility
study, but also for the definitive feasibility study. If the data set is
subsequently found not to be definitive, the results will provide
an excellent basis for planning the definitive test work program.
It is recommended that by the end of the prefeasibility study the
chosen domains for the orebody are geometallurgically relevant. It
is also recommended that the plan for full confirmation and
characterisation of the geometallurgical domain selections are
included as recommendations in the prefeasibility report.
Definitive feasibility study (DFS)
The definitive study phase is often also called the Bankable
Feasibility Study. The accuracy of such a study is, at a
maximum, ±15 per cent. Considering the financial uncertainty
can be as high as ±10 per cent (due to interest rates, exchange
rate fluctuations and equipment/materials/labour prices) there is
little, if any room for technical uncertainty. Ideally, at the
commencement of a DFS the basic flow sheet has been finalised
and only one option is ‘left on the drawing board’. However, it is
typical at the commencement of a DFS for trade-off studies to be
conducted to determine the most appropriate equipment types for
the selected unit process. Examples of trade-offs are Jaw versus
Gyratory primary crushing and conventional versus high rate
versus paste thickening.
At the definitive study phase the client is acting on board
approval to proceed, usually with the intention of transitioning
into funding approval for the project as a whole. The date for
first production is probably set and a tight time frame is imposed
for the study, usually aiming for a funding decision at a particular
board meeting. From the engineer’s perspective the client
expectation is that the test work carried out in the previous study
stages, together with any additional test work already approved,
is going to be sufficient for design. Again, the engineer must
determine if this is the case. If more test work is essential and the
potential exists for extending the study time frame then the
engineer has some significant justification work ahead.
The key issue in a DFS is having definitive data for design.
Declaring data to be definitive is firstly the job of the design
engineer. The engineer must be satisfied that design decisions are
based on fact and able to withstand scrutiny. Scrutineers include
the client representatives, the vendor’s technical teams, the
Bank’s resource analysts and the Board’s consultants. If there are
performance warranties then scrutiny within the engineering
company should be the toughest of all to pass.
246 Fremantle, WA, 19 - 21 March 2007
A definitive study must establish the real value of the resource.
This requires a well developed mine plan with an associated ore
delivery schedule. To establish a cash flow prediction for the life
of the operation based on the mine plan it is essential that the ore
delivery schedule informs the process engineers of variability in
the critical process parameters. Realistic cash flow projections
are only possible with an understanding of the period by period
variations in head grade, throughput rate, operating cost per
tonne and product quality.
It is also necessary to estimate the ‘noise’ that should be
expected for each of the critical process parameters on an
operational time scale. In the DFS stage, the shortest predictive
time frame within an ore delivery schedule is usually one month.
Late in the production schedule the time scale is usually one
year. Worse for judging variability, a mine plan is usually
constrained, in part, by the need to smooth head grade and
product output. It is not appropriate to assume that the flexibility
required of the plant design can be determined from the month
by month variability of the mine plan. Operational flexibility
must be based on the expected shift by shift variability.
The issue is then to find a data source that does have the
possibility of representing the variability of the critical process
parameters. The most obvious source is the library of core data
and a second, related, source is the block model.
The core library represents metre by metre variability within
the orebody. This is too detailed to use directly as a
representation of shift by shift variability. The core library data is
best examined as a set of location-based groups, an example of
which is already utilised in the mining world as the block model.
If available, the block model data can be used but a few checks
are required first. The block size should represent between ten
and 30 hours of production, a mass that is relevant to the
operational time scale. The blocks should also contain grouped
(usually Krieged) estimates of the critical process parameters, or
grouped data that can be used to calculate the critical parameters
for the block. The blocks should also be relevant in a
geometallurgical sense. If these conditions are met then an
examination of the variability in the block data can provide a
meaningful basis for process design.
The most difficult part of this step is to ensure that the relevant
data is calculated on a block basis. To do this an understanding is
required of the linkage between process response and geological
data and the geological model must be able to provide the critical
process parameters. Without a geometallurgical analysis none of
this is possible.
In one recent example, which will not be identified, a plant
was designed for an average grade in the region of 1.5 per cent
and an upper expected feed grade of about 2.3 per cent. During
start-up the plant was required to deal with grades fluctuating
between a low of about 0.5 per cent and a high of eight to nine
per cent. The basis of design was estimated from the mine plan
variability. However, the certainty of receiving a much wider
range of feed grades was painfully obvious for anyone that
wished to examine the core library.
At the end of the DFS the process engineer must understand
the critical process parameters in terms of their operational-
timescale variability and in terms of their relationship to the
mine delivery schedule. The geometallurgical domains should be
well established in all aspects of the project thinking.
Design phase
The design phase of a plant is when uncertainty is extremely
unhelpful. If design questions and assumptions persist at this
stage, then the previous stages were probably carried out
inappropriately. Unfortunately, many projects suffer because
myths and assumptions have not been properly tested before it is
too late. These myths sometimes underpin key design decisions
and if they are disproved late in the project development cycle
Ninth Mill Operators’ Conference
 THE IMPORTANCE OF GEOMETALLURGICAL ANALYSIS IN PLANT STUDY, DESIGN AND OPERATIONAL PHASES
they can be damaging to both schedule and reputation. It is this
last point that most often results in myths persisting, even when
proven beyond doubt to be false.
The process engineer that inherits a project at the design phase
is in a difficult position. Not having been responsible for the
study phases, the engineer has not had the opportunity to select
samples for test work or test assumptions and myths. The
engineer must, during bid preparation, examine the data made
available by the client and dig deep enough to be satisfied that
the detailed design path is clear. However, if serious problems
are detected during the evaluation phase that demand more test
work then the engineer is in a dilemma. Depending on the client,
the message may or may not want to be heard. As the engineer is
primarily in the business of building plants the focus may be
directed at winning the job first and sorting out the problems
later. The decision will be based on risk exposure for the
engineer and the contractual basis upon which the job is carried
out. Ultimately, it is likely the client will pay the price, either if
some additional test work and an associated schedule delay or
through poor performance at the very time the client should be
making money from their asset.
Had the engineer been responsible for the studies, assuming
they were carried out competently, then there should be far less
technical and schedule risk in the transition to project. It is
recommended that as a minimum, the definitive study phases and
then move into the design and construct phase. It is strongly
suggested that the engineering firm be involved from the earliest
stages of the project so that geometallurgical thinking (together
with its extensive design experience) is incorporated from the
start. Unravelling the geometallurgical issues at the definitive
stage and beyond is difficult and expensive. Building
geometallurgical thinking into the project from its inception will
provide a strong foundation for team-based thinking across the
owner’s and engineer’s teams.
Operational phase
The operational phase is where the appropriateness of the
engineer’s designs, especially those parts based on assumption
and unproven myth, are exposed for all to see. During
commissioning it is expected that mechanical issues will arise,
minor design problems will need rectifying and operation will
not be smooth. However, it is not appropriate that problems be
caused by the plant being fed ore that it didn’t expect to see.
Examples include:
• Incompetent oxide ore being fed to a SAG mill that has been
set up at commissioning for competent fresh ore, resulting in
damage to mill liners.
• Overloading of the flotation circuit, especially the cleaning
stages, due to unrecognised variability in the mill feed. The
result is usually a significant loss of recovery or a significant
reduction in plant throughput (if throughput turn down is
possible with the selected equipment).
• The generation of unsaleable products due to mineralogical
problems or the presence of deleterious elements.
All these outcomes are predictable and examples of instances
like these are not uncommon.
As the engineer is usually responsible for commissioning it is
in its best interest to understand, as fully as possible, what is to
be expected during the commissioning phase. As the client wants
to be earning cash as rapidly as possible it is also in their best
interests to allow the engineer to do the work that is necessary to
eliminate any of these surprises. Even a week of commissioning
delay is likely to result in losses that dwarf the expenditure
necessary to detect these losses ahead of time.
Post commissioning the owner will finally have a mine plan
that is correct, at least for a couple of weeks into the future. This
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
is unlikely to be the same as the mine plan available for the
definitive study and will change regularly on a short-term tactical
basis. Without a means of predicting the performance of the ore
that has actually been scheduled, often ready for treatment within
the week, the plant is destined to operate in reactive mode. With
a meaningful method for the prediction of critical process
properties the plant can be a proactive participant in the short-,
medium- and long-term production planning processes, rather
than a victim.
Many operations have found, usually a few years after
commissioning, that the ore they are supposed to receive never
comes. The revelation then dawns that the ore types they have
relied on since the earliest days of design do not possess
predictable sets of process properties and are more often
misleading than useful. It is at this point that many operations
employ someone to conduct a geometallurgical analysis.
Addressing geometallurgical issues at an operating site with no
historically effective geometallurgical practices is a much more
difficult problem than building geometallurgy into the design
process and is beyond the scope of this paper.
CONCLUSIONS
The implementation of geometallurgical analysis should result in
close cooperation and a common language being developed
between the geologists, mining engineers, process engineers and
project managers.
The earlier that geometallurgical analysis is implemented in a
project’s life the more value it will add to the project.
During the study phases, geometallurgical analysis should
progress in line with the accuracy of estimation required and
with the associated level of funding that is available. At options
study level the metallurgical appropriateness of the geological
view of the orebody must be established. At the prefeasibility
stage the geometallurgical domains should be defined and the
definitive test program should be outlined. At the definitive stage
the domains should be entrenched in the project, the domain
properties defined and the variability of the properties
established.
At plant commissioning the ore domains the plant receives
should not be a surprise and the critical metallurgical properties
should be adequately defined.
During operations the owner’s team should be able to plan
production on a reliable basis so that what the plant actually
receives as feed in the following weeks, months and years is
what they planned.
ACKNOWLEDGEMENTS
The author acknowledges the permission of GRD Minproc to
publish this paper and the assistance of the GRD Minproc
Mining and Geology section, with a special mention to Annick
Manfrino, for helping to make geometallurgical analysis a reality
within our organisation. Most importantly acknowledgement is
due to many colleagues over the years, metallurgists, miners and
geologists, with a desire to talk a common language so that we
can work together to maximise the value of our orebodies.
REFERENCES
Burger, B, Hatta, H, McGaffin, I and Gaffney, P, 2006. Batu Hijau –
Seven years of operations and continuous improvements, in
International Autogenous and Semiautogenous Grinding Technology,
2006 (eds: J Allan et al), Vol 1, pp 120-132 (Department of Mining
Engineering, University of British Columbia).
Dance, A, Valery, W, Jankovic, A and La Rosa, D, 2006. Higher
productivity through cooperative effort: A method of revealing and
correcting hidden operating inefficiencies, in International
Autogenous and Semiautogenous Grinding Technology, 2006 (eds: J
Allan, et al), Vol 1, pp 120-132 (Department of Mining Engineering,
University of British Columbia).
247
Cyanide Measurement and Control for Complex Ores and
Concentrates
P L Breuer1 and J A Rumball2
ABSTRACT
Measuring cyanide in the leach liquors associated with high copper
and/or sulfide ores is a non-trivial exercise. It is well known that copper
complexes with cyanide and reduces the amount of cyanide available for
gold leaching. What is not so well known is that the rhodanine end point
overestimates the amount of cyanide available for leaching in the
presence of copper.
Sulfide ions also interfere with a rhodanine end point titration,
although this is less of a problem as the black silver sulfide precipitate
formed clearly indicates to the operator that something is wrong.
Thiosulfate, which forms readily from the alkaline oxidation of
sulfides, causes an overestimation of the cyanide available for leaching
due to the formation of Ag(S2O3)-.
The net result of these observations is that rhodanine is rarely of value
as an end point indicator in complex solutions. It generally overestimates
the amount of cyanide available, providing plant metallurgists with a false
sense of security.
Potentiometric titrations are generally much more capable of
accurately measuring cyanide in complex solutions. The interfering effect
of copper can be resolved and it is even possible to estimate the copper
concentration using the inflexions associated with the titration of CN- +
Cu(CN)43- versus Cu(CN)32-.
Using the potentiometric method sulfide ions can be determined before
the cyanide measurement is made due to the formation of AgS2, although
the electrode will foul with repeated measurements. Thiosulfate can also
be distinguished from cyanide with a potentiometric titration.
Rhodanine and potentiometric titrations are both affected by the pH of
the liquor. At pH 9.2 half the cyanide will be present as CN- and half as
HCN(aq) with only the CN- being titratable. This does not hold in the
presence of a buffer active around pH 9. In the presence of sufficient
buffer, both the CN- and HCN(aq) are titrated. Ca(OH)+ may buffer the pH
when large amounts of lime have been added.
INTRODUCTION
Cyanide measurement is an important control parameter in the
recovery of gold from ores using cyanidation. Commonly, a
silver nitrate titration using a rhodanine indicator end point is
employed on mine sites to measure the cyanide concentration of
filtered samples. A potentiometric determination of the end point
is also possible and this has become the basis for online
measurement techniques employed to automate cyanide control
in some gold plants.
The processing of more complex ores and concentrates using
cyanidation has generated a new range of issues and challenges
in the measurement and control of the cyanide concentration in
these processes. The major issues identified are interference from
high copper concentrations and the presence of sulfide ions. In
some gold plants economics also constrain operation to pH <10.
This not only creates OH&S issues with regards to the presence
of HCN(g), but can impact on the cyanide measurement depending
on the method used. Understanding the relationship between the
various cyanide measurement values and the gold dissolution
1. Research Scientist – Gold Program, Parker Centre (CSIRO
Minerals), PO Box 7229, Karawara WA 6152.
Email: paul.breuer@csiro.au
2. MAusIMM, Gold Program Manager, Parker Centre (CSIRO
Minerals), PO Box 7229, Karawara WA 6152.
Email: john.rumball@csiro.au
Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007
process is also very important in ascertaining the appropriate set
point for the process. These aspects are discussed in this paper.
EXPERIMENTAL PROCEDURES
Unless stated otherwise, solutions were prepared from AR grade
reagents and deionised water.
Rhodanine titrations were conducted with a 5 ml sample, two
to three drops of indicator and titrated directly with 0.05 M silver
nitrate using a Metrohm 665 Dosimat (for constant addition rate)
until the end point was observed. Potentiometric titrations were
conducted using a Metrohm 716 Titrino auto-titrator with 0.01 M
silver nitrate solution. Samples (1 - 5 ml) were diluted with
deionised water to achieve sufficient volume for the
measurement set-up. All titrations were repeated in at least
triplicate and average values are reported.
An amperometric method developed by CSIRO Minerals was
also utilised. This method measures the silver dissolution rate at
a fixed potential (-0.05 V versus the standard hydrogen electrode,
SHE) in filtered solutions or slurries.
Gold leach rates and electrochemistry were measured in air
saturated solutions using a rotating electrochemical quartz
crystal microbalance (REQCM), which is described elsewhere
(Jeffrey, 1998).
RESULTS AND DISCUSSION
Cyanide analysis techniques
Cyanide is traditionally measured by titrating with silver ions
with the end point determined using a rhodanine indicator. In the
case of coloured solutions potassium iodide has been used as an
alternate indicator. A refinement to this method is to measure the
potential of a silver wire immersed in the titration cup and use
the potential inflection at approximately -100 mV (SCE) as the
end point.
Ionic cyanide in solution may also be measured
amperometrically (Heath, Rumball and Bailey, 1999) and with a
variety of other techniques. Table 1 presents the most common
methods used and the species measured by each technique.
For gold leaching it is the first three methods that are of
interest as these potentially give a measurement of the cyanide
available to leach gold. It should also be noted that on-site
TABLE 1
Species measured by various cyanide analysis techniques.
Analysis technique Species measured
Silver nitrate titration – rhodanine or CN-
KI indicator
Silver nitrate titration – potentiometric CN-, S2- and estimate of Cu
Amperometric CN- and estimate of Cu
Flow injection/ligand exchange/ WAD CN, total CN
UV digestion
Distillation WAD CN, total CN
IC and HPLC Metal cyanides, CNO-,
SCN-, S2-, S2O32-
249
P L BREUER and J A RUMBALL

cyanide measurements are typically reported as mg/L NaCN, The amperometric technique gave results different from both
whereas samples submitted to analytical laboratories are often the calculated ionic cyanide and the titration technique
reported as mg/L CN-. (Figure 2). As this technique is measuring the silver oxidation
rate, the ionic cyanide concentration at the electrode surface
Influence of pH decreases as ionic cyanide complexes with the oxidised silver.
The buffering effect of the carbonate results in some dissociation
of HCN(aq), which provides more ionic cyanide for silver
Cyanide measurement oxidation. However, diffusion of ions to and from the surface
into the bulk solution minimises the dissociation of HCN(aq) at
In gold plants where the economics constrain the leach pH to
the electrode surface. Clearly, carbonate has some positive effect
below ten, both CN- and HCN(aq) exist. The CN-/HCN(aq) on the silver oxidation rate. However, it is dependent on the
distribution as a function of pH is described by Equation 1 (pKa diffusion rate and thus does not correlate with either the cyanide
= 9.2). Knowing the total cyanide ([CN]T), the ionic cyanide measured by titration or the ionic cyanide concentration.
concentration can be calculated using Equation 2 and is
graphically shown in Figure 1. 100

100 80

CN (mg/L NaCN)
80 60
CN (mg/L NaCN)

60 40
Theoretical ionic cyanide

-
20
Titration
40
Theoretical ionic cyanide Amperometric
-

20
Titration 0
Amperometric 8.6 8.8 9 9.2 9.4 9.6 9.8 10 10.2 10.4

0 pH
8.6 8.8 9 9.2 9.4 9.6 9.8 10 10.2 10.4
FIG 2 - Comparative measurement of cyanide in buffered solutions
pH
(0.1 M KCl, 93 mg/L NaCN, 250 mg/L CO32-).
FIG 1 - Cyanide measurement as function of pH in unbuffered
solutions (0.1 M KCl, 93 mg/L NaCN). Gold leaching
So which cyanide measurement is important for gold leaching?
[CN ] -

= 10 (pH − pKa )
(1)
The gold leaching process is dependent on both the ionic cyanide
[ HCN ] (aq )
and oxygen concentrations. A graphical representation of the
gold leach rate as a function of ionic cyanide concentration is
shown in Figure 3. For a given oxygen concentration, the gold
[CN ] = [CN] 10 ( −
T
pH − pKa )
(
/ 1 + 10 (pH − pKa ) ) (2) leach rate is cyanide diffusion limited at ionic cyanide
concentrations less than X and oxygen diffusion limited at ionic
cyanide concentrations greater than X, where X is known as the
The cyanide measurement by silver nitrate titration critical cyanide concentration.
(colorimetric or potentiometric end points) follows the ionic
cyanide concentration as a function of pH, which indicates
HCN(aq) is not measured under these conditions. The Cyanide limiting rate
amperometric method also gives cyanide measurements that
follow the ionic cyanide concentration. This clearly indicates that
the oxidation of silver (or gold) is dependent on the ionic cyanide
Au leach rate

concentration. If the sample pH is increased above 11 before the

cyanide measurement, then CN- + HCN(aq) ([CN]T) is measured. Oxygen limiting rate
For buffered solutions, the cyanide measured by titration is
higher than the calculated ionic cyanide (Figure 2). This is
because buffering the pH according to Equations 3 or 4 allows
further HCN(aq) dissociation without a pH fall that would
otherwise stabilise the HCN(aq). In Figure 2, the titration
measured the total CN- + HCN(aq) because the buffer
concentration was greater than the HCN(aq) concentration. At
buffer concentrations less than the HCN(aq) concentration, only X
the equivalent moles of HCN(aq) to buffer will be measured by CN- concentration
titration, in addition to the CN-.
FIG 3 - Schematic of oxygen and cyanide limiting rates of
CO23 − + H + ⇔ HCO3− (3) gold dissolution.

Ca(OH) + H + ⇔ CA2 + + H2O

+
(4) With a decrease in pH, the ionic cyanide concentration
decreases, and the gold leach rate can change from oxygen
diffusion limited to cyanide diffusion limited. Under cyanide
HCN (aq ) ⇔ H + + CN − (5) diffusion limiting conditions it would thus be expected that the
gold leach rate would decrease with decreasing pH relative to the
Ag + + 2CN − ⇒ Ag(CN)2
−
(6) ionic cyanide concentration. This is shown not to be the case in

250 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 CYANIDE MEASUREMENT AND CONTROL FOR COMPLEX ORES AND CONCENTRATES

Figure 4. The reason for the higher than expected gold leach rate TABLE 2
at pH’s <10 was further investigated by conducting electro- Speciation of solution containing 22 500 mg/L NaCN +
chemical studies using a REQCM. The increase in gold leach 8000 mg/L Cu+ (pH 11).
rate was found to be attributed to the generation of OH- ions
from oxygen reduction concurrently occurring on the gold Species mol/L mg/L (mg/L (mg/L Cu)
surface. The OH- ions react with HCN(aq) which releases more concentration NaCN)
ionic cyanide.
[Cu+] 2.16E-26 0.0 0.0
4.0E-05 [CN-] 2.53E-02 659.1 1241.4
Au leach rate (mol m s )
-1

3.5E-05 70
[CuCN(s)] 0.00E+00 0.0 0.0 0.0
-2

CN- (mg/L NaCN)

3.0E-05 60
[HCN(aq)] 1.30E-04 3.5 6.4
2.5E-05 50
[Cu(CN)2-] 1.38E-05 1.6 1.4 0.9
2.0E-05 40
[Cu(CN)32-] 6.99E-02 9896.7 10 275.7 4441.4
1.5E-05 30
Au leach rate [Cu(CN)43-] 5.60E-02 9384.4 10 975.1 3557.8
1.0E-05 20
Calculated ionic cyanide Titratable 8.13E-02 3985.2
5.0E-06 10 cyanide
2 x ionic cyanide
0.0E+00 0 Total copper 1.26E-01 8000.0
8 8.5 9 9.5 10 10.5 11
pH

FIG 4 - Effect of pH on gold leach rate measured using a REQCM 22500 mg/L NaCN + 8000 mg/L+ Cu
(0.1 M KCl, 71 mg/L NaCN, air saturated). 7000 9000

6000 8000
Two possible oxygen reduction reactions can take place
(Equations 7 and 8). For both oxygen reduction reactions, one 5000 7000
[NaCN] (mg/L)

mole of hydroxide is produced per electron consumed. Thus, for

[Cu] (mg/L)
4000 6000
each mole of Au oxidised one mole of OH- ions is generated
from oxygen reduction and hence up to one mole of CN- can be 3000 5000
released from HCN(aq). Hence, gold leach rates up to double
Rhodanine
(based on the solution ionic cyanide concentration) are 2000 4000
Potentiometric
theoretically possible where the gold leach rate is cyanide
1000 Potentiometric Cu 3000
diffusion limited. This theoretical maximum gold leach rate is
shown as a dotted line in Figure 4 (2 × ionic cyanide, with a 0 2000
maximum of CN- + HCN(aq)), and is only slightly higher than the 0 1 2 3 4 5 6 7
measured gold leach rates shown in this figure.
Rate of AgNO3 addition (mL/min)
− −
O2 + 2H2O + 2e ⇒ H2O2 + 2OH (7)
FIG 5 - Comparison of rhodanine and potentiometric titrations for a
O2 + 4H2O + 4e − ⇒ 4OH− (8) solution containing 22 500 mg/L NaCN + 8000 mg/L Cu+.

Having established the effect of pH on the gold leach rate, the

effect of carbonate was determined under cyanide limiting 200 20
mV
conditions. The gold leach rate measured at pH 9 with 1 g/L
diff mV
CO32- present was found to be 2.86 × 10-5 mol m-2 s-1, which is 0 Ag+ + 2Cu(CN)32- Ag(CN)2- + 2Cu(CN)2-
15
13 per cent faster than without CO32- (see Figure 4). Thus, at +
Ag + 2Cu(CN)4 3-

pH’s below 9.2 the presence of carbonate may enhance the gold -
2Cu(CN)32-
diff mV
-200 Ag(CN)2 +
mV

leach rate depending on the diffusion rates prevailing. 10

Ag+ + 2CN- Ag(CN)2-

-400
High copper concentrations
5
-600
Cyanide measurement
The thermodynamic speciation of a solution containing -800 0
0 1 2 3 4 5 6 7 8 9 10 11
22 500 mg/L NaCN + 8000 mg/L Cu+ is presented in Table 2.
The titratable cyanide in this table is a calculated value which mLs of 0.01M AgNO3
includes the free CN- and one CN- from Cu(CN)43-. The cyanide
determination by titration with potentiometric end point for this FIG 6 - Potentiometric silver nitrate titration of a solution containing
solution is shown in Figure 5 to agree closely with this value. An 22 500 mg/L NaCN + 8000 mg/L Cu+.
added advantage of the potentiometric titration is that an estimate
of the copper concentration can be made from the difference
between the potentiometric differential peaks for the reaction of The rhodanine titration on the other hand significantly
silver with Cu(CN)32- (Figure 6). The copper estimates shown in overestimates the cyanide concentration (standard deviation
Figure 5 are slightly higher than the actual copper concentration, of the repeat measurements was 200 mg/L). The cyanide
which appears to be associated with the slight underestimation of concentration determined by rhodanine also increases
the titratable cyanide concentration. It has previously been significantly with increasing silver nitrate addition rate. It is
reported that the amperometric technique can also provide an suggested that the high copper concentration interferes with the
estimate of the copper concentration (Dai, Jeffrey and Breuer, rate of Ag+ association with the rhodanine affecting the end point
2005). determination. The mechanism for this is unclear at present.

Ninth Mill Operators’ Conference Fremantle, WA, 19 - 21 March 2007 251

P L BREUER and J A RUMBALL

Decreasing the addition rate reduced the overestimation;
however, the end point becomes very difficult to interpret below
2 ml/min due to the very slow colour change. A ten-fold dilution
was found to exacerbate the overestimation with the apparent
cyanide concentration increasing by more than 15 per cent.
A possible explanation is that the silver complexes faster with
cyanide from the Cu(CN)32- complex than with rhodanine and
the equilibration (Equation 9) with the indicator complex
(Rh-Ag+) is slow and also influenced by the concentration.
Rh − Ag + + 2Cu(CN)3 ⇔ Rh + Ag(CN)2 + 2Cu(CN)2
2− − −
(9)
Comparisons between the titration values and the
thermodynamic titratable cyanide for two different cyanide
concentrations with copper present are shown in Figure 7.
Clearly, the rhodanine titration significantly overestimates the
available cyanide for gold leaching when copper is present, the
magnitude of which appears to be related to the copper
concentration. The potentiometric and amperometric techniques
on the other hand closely measure the available cyanide for gold
leaching.
6000
determination), and PbOH+ and Pb2+ become soluble, which
interferes with the rhodanine indicator (gives the salmon pink
end point colour). This can be avoided by maintaining the sample
pH >11 or by the addition of a concentrated lead nitrate solution
prepared in 1 M NaOH. Addition of solid PbO or PbCO3 could
also be used.
The results of measurements made using a concentrated lead
nitrate solution prepared in 1 M NaOH are shown in Figure 8.
The slightly lower cyanide titration values after lead addition
could be attributed to the low solubility of lead cyanide or the
oxidation of some sulfide to form thiocyanate. This is supported
by the potentiometric titration value determined without lead
addition. This determination is possible as silver sulfide was
found to form preferentially to silver cyanide and thus it is
titrated before the cyanide (Figure 9). Hence, the sulfide does not
need to be removed for the potentiometric titration. This has the
added benefit that the sulfide concentration can also be
determined concurrently using this method. However, online
potentiometric titration set-ups and the amperometric method are
affected by sulfide ions; online potentiometric titrations
generally only detect one end point and hence would include
sulfide ions in the cyanide determination, and the build-up of
silver sulfide precipitate effects subsequent measurements for
both techniques unless removed.

5000 4000 mg/L NaCN + 5 mM Na 2S

Titrated [NaCN] (mg/L)

4000
4000
3900
3000
[NaCN] (mg/L)

3800
2000
Rhodanine Rhodanine
Potentiometric 3700
1000 Potentiometric

3600
0
0 1000 2000 3000 4000 5000
Thermodynamic titratable [NaCN] (mg/L) 3500
0 50 100 150 200 250
PbNO3 (in 1 M NaOH) addition (% of req)
FIG 7 - Comparison of thermodynamic and titration values for
cyanide with 8000 mg/L Cu+ present.
FIG 8 - Effect of lead nitrate (in 1 M NaOH) pretreatment on
cyanide titration of solution containing 5 mM S2- + 4000 mg/L
Gold leaching NaCN.
The leach rate of gold in cyanide solutions containing copper has
been studied previously (Breuer, Jeffrey and Dai, 2005). This
work showed that gold does leach in the presence of Cu(CN)32-; 200 90
however, the leach rate is significantly slower than in the 100 mV 80
presence of CN- and Cu(CN)43-. Hence, the potentiometrically 0 diff mV 70
titratable cyanide measurement is the cyanide value of -100
60
importance for gold leaching. -200
diff mV

50
mV

-300
Ag+ + 2CN- Ag(CN)2- 40
Sulfide ions -400
2Ag+ + S 2-
30
-500
Ag 2S 20
-600
Cyanide measurement
-700 10
Direct rhodanine titration of a sample containing sulfide is not -800 0
possible as silver is precipitated as silver sulfide, which masks 0 2 4 6 8 10 12 14
the rhodanine indicator colour. mLs of 0.01 M AgNO3
The addition of lead ions and filtering before titration is a
standard method for removing the interference of sulfide ions. 2-
FIG 9 - Potentiometric titration of solution containing 9 mM S and
Most importantly, it should be noted that the form in which the 4000 mg/L NaCN.
lead is added can affect the results. For example, if lead nitrate or
a concentrated lead nitrate solution is added to remove the Gold leaching
interference of sulfide ions before conducting a cyanide titration,
the pH decreases due to the excess lead precipitating as PbO. The leach rate of gold in cyanide solutions containing sulfide
This results in some HCN(aq) forming, which is not measured ions has been studied previously (Jeffrey and Breuer, 2000). This
in the case of the potentiometric titration (low cyanide work showed that gold dissolution is very slow in the presence of

252 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

 CYANIDE MEASUREMENT AND CONTROL FOR COMPLEX ORES AND CONCENTRATES
sulfide ions and the presence of lead can assist in overcoming the
effect of sulfide ions. Recent laboratory test work confirmed that
gold dissolution does not occur to any appreciable extent when
sulfide ions are present in the leach (Figure 10). Once sufficient
oxygen was added (at the 24 hour mark) to oxidise the sulfide
ions, improved gold recovery was observed. Sulfide ions are not
typically found in gold leach solutions unless the ore or
concentrate contains appreciable reactive sulfides. For the latter,
being able to quantify sulfide ions is important for controlling
reagent addition to optimise gold recovery.
Thiocyanate and thiosulfate ions
The interference of thiocyanate ions on the rhodanine and
potentiometric titration cyanide measurements was investigated
with a solution containing 4000 mg/L NaCN and 100 mM KSCN
(5.8 g/L SCN). No interference of thiocyanate was identified in
the determination of cyanide by either titration method.
The potential for interference of thiocyanate on the concurrent
cyanide and copper determinations using the potentiometric
titration was investigated with solutions containing 22 500 mg/L
NaCN, 2000 or 8000 mg/L Cu+ and 100 mM KSCN. In
comparison to when no thiocyanate was present, a small increase
was found in the cyanide determination which was reflected by a
similar decrease in the copper determination. This is possibly due
to a complexation equilibrium of thiocyanate with copper (cyanide
complexes much stronger than thiocyanate), which displaces a
small amount of CN-. Thus, the cyanide determination most likely
reflects the ionic cyanide available for gold leaching. However,
the copper estimation is slightly low and at high thiocyanate
concentrations this can be as much as 15 per cent.
Thiosulfate ions have also been found in cyanide leach solutions
where reactive sulfides are present in the ore. Thiosulfate was
found to interfere with the cyanide determination using the
rhodanine titration method, resulting in an over-determination of
the cyanide concentration. AgCN(s) was found to precipitate
before the change in the rhodanine indicator colour was
observed, which correlated with silver having formed AgCN(s)
and Ag(S2O3)-.
The potentiometric method was effective in distinguishing
between the end points of the silver reactions with cyanide and
thiosulfate, with accurate determination of the ionic cyanide
concentration possible. However, in copper cyanide solutions
containing thiosulfate an estimate of the copper concentration
9000
8000 Cu
Fe
7000
Sulfide ions
Cu, Fe, S2- (mg/L)
6000 Au recovery
5000
4000
3000
2000
1000
0 0
0 5 10
FIG 10 - Gold leaching from a concentrate containing reactive sulfides.
Ninth Mill Operators’ Conference
using the potentiometric method is not possible due to the
concurrent titration of the third cyanide associated with copper
and thiosulfate. Thiosulfate does not interfere with the
amperometric cyanide measurement when -50 mV is utilised.
However, interference and contribution to silver oxidation by
thiosulfate does occur at the higher potentials used to estimate
the copper concentration.
CONCLUSIONS
Potentiometric titrations are recommended in preference to
rhodanine titrations when measuring the cyanide concentration in
circuits treating sulfide or copper rich ores and concentrates.
The potentiometric titration may also be tailored to yield
sulfide and copper ion concentration as part of the cyanide
measurement procedure.
Using default set-up procedures, the online potentiometric
titration measurement of cyanide will be affected by sulfide ions
and a silver sulfide precipitate will eventually foul the electrode.
The amperometric technique potentially offers comparable
capabilities to the potentiometric titration, though careful set-up
of the amperometric sensor is required for these complex leach
solutions.
Gold leaching is hindered by the presence of sulfide ions and
hence sufficient air or oxygen addition is required in the leach to
eliminate or minimise the presence of sulfide ions. The gold
leach rate does not always follow the ionic cyanide or any of the
cyanide measurements from the various techniques with pH.
Above pH 9.2 the gold leach rate is constant, and only decreases
with decreasing pH below 9.2 when cyanide is limited.
ACKNOWLEDGEMENT
The authors would like to thank Newcrest Mining Ltd for their
financial support for a portion of the work presented in this
paper.
REFERENCES
Breuer, P L, Jeffrey, M I and Dai, X, 2005. Leaching and recovery of
copper during the cyanidation of copper containing gold ores, in
Treatment of Gold Ores (eds: G Deschenes, D Hodouin and
L Lorenzen), pp 279-291 (The Canadian Institute of Mining,
Metallurgy and Petroleum: Montreal).
Oxygen addition started
60
50
Au recovery (%)
40
30
20
10
15 20 25 30 35 40 45 50
Time (hr)
Fremantle, WA, 19 - 21 March 2007 253
P L BREUER and J A RUMBALL

Dai, X, Jeffrey, M I and Breuer, P L, 2005. The development of a
flow injection analysis method for the quantification of free
cyanide and copper cyanide complexes in gold leaching solutions,
Hydrometallurgy, 76(1-2):87-96.
Heath, A R, Rumball, J A and Bailey, S, 1999. An amperometric method
for measuring cyanide in CIP/CIL circuits, Minerals Engineering,
12(11):1313-1326.
Jeffrey, M I, 1998. A kinetic and electrochemical study of the dissolution
of gold in aerated cyanide solutions: The role of solid and solution
phase purity, PhD thesis, Curtin University of Technology, Bentley,
Werstern Australia.
Jeffrey, M I and Breuer, P L, 2000. The cyanide leaching of gold
in solutions containing sulfide, Minerals Engineering, 13(10-11):
1097-1106.

254 Fremantle, WA, 19 - 21 March 2007 Ninth Mill Operators’ Conference

New Technology and Approaches in Reagent Development and
Applications in the Processing of Base and Precious Metals and
Industrial Minerals
D R Nagaraj'

ABSTRACT
Over the past decade or so, our efforts have focused on developing the rationale and concepts
for new approaches and technologies in reagent development and optimisation in the processing
of base and precious metals and industrial minerals. These efforts have resulted in new reagents
for base and precious metals recovery in flotation, and the new nanoparticle-based technology
in industrial minerals processing. Another major outcome is in the development of a
comprehensive program called Flotation Matrix 100 to facilitate the implementation of a
holistic approach to reagent selection and optimisation in flotation. This is a much needed
alternative to the traditional approach to reagent optimization, which is mostly reductionistic
and incremental. The critical elements of Flotation Matrix 100 include a web-based, searchable
knowledge base for flotation reagents and their applications; tools to quantify plant needs;
expert system reagent selection software; and protocols and statistical tools for laboratory and
plant best practices needed for successful implementation in the plant. An overview will be
given using examples to highlight the application of the new reagents and technology.

1. Cytec Industries Inc, 1937 West Main Street, Stamford CT 06877, USA. Email: d.r.nagaraj@cytec.com

Ninth Mill Operators' Conference Frernantle, WA, 19 - 21 March 2007 255