Professional Documents
Culture Documents
Proceedings
Conference 2007
19 - 21 March 2007
Fremantle, Western Australia
Published by
The Institute is not responsible as a body for the facts and opinions advanced
in any of its publications.
Flotation
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
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
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
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
Cyanide Measurement and Control for Complex Ores and P L Breuer and J A Rumball 249
Concentrates
-6
INTRODUCTION
-8
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
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
Ta
fic
m
er
ne
L
m
th
One classic example of embodied energy is the use of steel Energy usage in the total comminution system
grinding media. A typical ball mill installation treating a soft-
medium copper ore has the following features: The total mining and processing chain uses significant quantities
of energy. Figure 4 gives a breakdown of energy input at various
• throughput: 350 t/h, stages, excluding the impact of embodied energy.
• power consumption: 5.65 kWh/t, and In most instances the first comminution process is the
• media consumption: 125 kg/h. so-called chemical comminution that uses explosives. Drill and
In the case of steel grinding media the typical embodied blast operations are by nature a batch process that reduces
energy is estimated to be 32 MJ/kg (Hammond and Jones, 2006). material to a point where it can be handled. In many cases the
These values can be transformed into an equivalent kWh/t value size produced from blasting is matched to some of the largest
for the consumption of the media: material handling and transport equipment available, which
logistically has appeal, but in terms of energy this approach is
• media consumption per tonne of ore = 125/350 = 0.36 kg/t; open to question. The same thinking has led to increasingly large
• embodied energy per tonne of ore = 0.36 × 32 = 11.52 MJ/t, excavation and transport equipment. Figure 5 shows the trend in
or 11 520 kJ/t; and haul truck size over the last 50 years.
• embodied energy consumed through media wear = As comminution is all about breaking rock then the total
11 520/3600 = 3.2 kWh/t. journey from in situ must start to be considered. Integrated
Given that the power consumption for the installation is studies tend to focus on the energy balance between explosive
5.65 kWh/t, this means that the embodied energy in the media and mechanical comminution. Why should comminution not
adds another 56 per cent to the energy input for grinding. This is consider alternatives to explosives? Mechanical rock cutting or
a significant addition to the operational energy footprint of a continuous mining equipment has been widely researched and
comminution operation. pursued over the years, but seldom as part of an integrated
total design for comminution. There is no reason why such
RESPONSES TO NEW DRIVERS mechanical cutting devices should not be considered part of the
total picture. The main features of mechanical mining equipment
Given the new business drivers and constraints and the potential are:
impact of comminution on the total system, there needs to be
some changes to how comminution is considered. • type of ground engaging tool and cutting mechanism,
Comminution in its purest definition is size reduction and • vehicle to carry cutting device and apply force, 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
• product discharge mechanism.
final size for extraction or sale. In this sense the total design of a As with all comminution the aim is to exploit as much tensile
comminution system should actively be looking to: breakage as possible, thus avoiding wasted energy through
• reduce and optimise energy input through the replacement, or compressive crushing and shear zone development.
reduction in, the numbers of stages of comminution and The main cutting tools can be categorised as picks and discs,
associated treatment and handling; with picks used for soft rock and discs deployed in harder
material. There are many versions of the tools designed for
• reduce the amount of material processed; specific machine types and application.
• utilise equipment and processes that are suited to the Numerous studies recently have examined the use of
application of automation; and mechanical cutting devices to replace drill and blast. In
• minimise the size and footprint – physical and environmental. 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
100 10
50 8
6
0
1950 1960 1970 1980 1990 2000 4
Year 2
0
FIG 5 - Increase in haul truck payload since 1950 (development 1750 1800 1850 1900 1950 2000 2050
in truck payload capacity since 1950 (after Krause, 2001 and Year
Caterpillar, 2005)).
FIG 6 - Decline in mined copper grades.
usage should be considered as simply one of the important
factors. In open pit iron ore mining the breakdown of energy Part of the concept of a total systems approach for
input is approximately: comminution is to select and sort the ore from waste/gangue
and/or below-grade ore to allow only valuable material to be
• drill/blast: 10 - 15 per cent, processed. The aim is to produce improvements in the efficiency
• excavation/loading: 10 - 15 per cent, of the process and to reduce the unnecessary processing of waste.
This type of selectivity may potentially offer:
• haulage: 40 - 55 per cent,
• reductions in operating costs due to the decrease in tonnage
• processing: 10 - 15 per cent, and processed;
• other: 5 - 10 per cent. • recovery of previously uneconomical material from the
The situation highlights that it is drill/blast/load/haul which is existing waste stockpiles, which may lead to an extension to
the greatest consumer of energy, with haulage dominating. the life-of-mine;
Therefore, considering total system design of comminution • reduction in utilities such as water and power supply;
systems it would seem sensible to reduce in situ material to such
a size so that alternatives to trucks can be considered. Many • improved utilisation of ore reserves; and
operations already do this by primary and secondary crushing • upgrade of marginal ore.
down to 80 mm and then using overland conveyers. In trials of
surface, mechanical mining there is evidence to support the view In mining process streams, materials can be sorted at bulk flow
that they can generate -80 mm material directly from in situ rock or at individual particle level. The bulk sorting is aimed at
through a cutting process. The knock-on effect is dramatic, in identification of material properties ‘en-masse’. As such they are
that material can be conveyed and the requirement for drill, blast, used to discriminate in mass flow environments such as trucks,
load, truck haulage and even primary and secondary crushing is conveyor belts and load-haul-dump (LHD) buckets. Bulk sorting
removed. Obviously there are issues to be overcome, including is predominantly limited to low-value products, whereas
efficient methods for transferring from the surface miner to a individual particle sorting is traditionally restricted to high-value
conveyer based haulage system, but this should not be commodities such as diamonds.
insurmountable. Ore sorting is dependent on the detection of differences in
Another integrated comminution and transport approach is fundamental mineral properties. Sufficient difference must exist
represented by the use of mobile or in-pit crushing and in the fundamental properties to achieve an efficient ore sort. The
conveying systems. This is not a new solution, but the scale and properties that can be used as a basis for ore sorting include:
implementation of the systems has developed. One of the major • Photometric measurements – optical differentiation of waste
deployments of large scale mobile crushing and conveying is the from ore, based on the material being induced/fluoresced at a
overburden system at the BMA Goonyella Riverside operation. particular wavelength interval.
In this application a P&H 4100XPB electric shovel loads blasted • Thermal properties (conductivity and diffusivity). These
rock straight into a mobile crushing rig that utilises an MMD fundamental material properties form the basis for potential
1400 mineral sizer. Although the size reduction targeted is not selective heating sorting systems.
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 • Radiometric properties – based on determination of the
transported using standard overland conveyers, and distributed natural emission of radioactivity from an ore.
via spreaders (Anon, 2003). Currently this technology is • Optical properties – optical differentiation (brightness and
restricted to soft-medium ore types, but recent work in the colour) of waste from ore can be used as a basis for ore
application of sizers to medium-hard material types could see the sorting. Optical sorting could equally be included in
expansion of this approach. photometric sorting.
• Particle density and porosity (gravity separation).
Waste removal
• Electromagnetic properties – measurements of inductive
A major aspect in the use of energy in mining and processing is conductivity or galvanic resistivity can be used to
the extraction and subsequent treatment of waste material. Every discriminate between ore and waste.
• Other surface properties – including chemical breakdown size of comminution stages and the subsequent scale of disposal
properties or surface chemical properties (including friction). 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
Comminution and automation options higher flow rates per unit of equipment, or use comminution
Already comminution equipment uses more advanced equipment capable of a wider range of size reduction in a single
automation and control than other areas of the mining value unit.
chain. This is possible through online measurement, machine
health diagnosis and the use of plant-wide management systems. CONCLUSIONS
To make comminution equipment less personnel-intensive
further developments of sensors and condition monitoring are Comminution has always stood in its own right within
required, so that maintenance intervals can be better defined and processing. Over the years the unified approach to the mine
therefore more predictable. There also needs to be consideration through to product chain has meant that comminution has
of equipment size and maintainability. The move to ever- become a more integrated part of the process, but even this is not
increasing throughput has led to the development of enormous enough.
grinding mills which, although presenting maintenance challenges, Comminution plant can dramatically impact the stages up and
has dramatically reduced personnel numbers compared to previous downstream and in many instances it is these processes that
multi-stage crushing circuits.
relate closely to the modified business drivers now in place. Total
Modelling and simulation is not always seen as an enabler to system design is therefore an approach that needs to recognise
reduce human intervention in comminution, but the link is very the impact of comminution and how it can influence the overall
real. One of the major reasons for intervention is maintenance
process. The aim of the total system design being to minimise
and adjustment. Modelling and simulation allows for predictive
capability, thus it is possible to predict changes in equipment impact and implement designs/practices that optimise efficiency,
performance that are precursors to major maintenance events. particularly with regard to the realities of energy and human
resources.
There are numerous examples of such capabilities being
developed. The various mill monitoring devices being developed,
including acoustic and shell-mounted accelerometers are such ACKNOWLEDGEMENTS
devices. With these linked to models of the internal processes
such as discrete element modelling (DEM), changes in The author would like to thank David Tuppurainen of Rio Tinto
performance can be linked to the mill fundamentals and therefore OTX and Dave Bowman and Jahna Speedy from the Rio Tinto
a full picture of the mill condition can be achieved. Iron Ore Technology Division for their assistance.
Without tools to assist in predicting when maintenance or
adjustment is due, or overdue, the effects can be dramatic. REFERENCES
A concept not just confined to comminution, but with Ainley, J and Underwood, C, 2003. Australia’s teachers: Australia’s
significant ability to ease reliance on site-based operations future, in Review of Teaching and Teacher Education, Volume 3:
personnel, is the remote operations centre. Many mining Background Data Analysis, DEST, Canberra.
companies have these in various forms, but the recent trend has Anon, 2003. Overburden stripping at Goonyella Riverside, Australian
been to try and make these even more powerful. The main Journal of Mining, January/February:42.
enhancements to the traditional approach are that the centres will Batterham, R J, 2004. Comminution challenges for Rio Tinto, in
be: Proceedings Comminution Workshop 2004, 23 March 2004, pp i-v.
• truly remote from mine sites – more attractive to personnel; Caterpillar, 2005. Caterpillar achieves mining truck milestone as Decatur,
Illinois, plant readies, Number 5000, Release Number: 121PR04 –
• integrated multi-control facilities – process integrated with for North American Release [online]. Available from:
mining, maintenance, planning and geological functions; <http://www.cat.com/cda/components/fullArticle>.
• information management – access to background DTI, 2006. The top 800 UK and 1250 Global companies by R&D
information, maintenance manuals, procedures and historical investment – Commentary and analysis, 1:16-124.
databases; and Hammond, G and Jones, C, 2006. Inventory of Carbon and Energy –
Version 1.5, pp 4-6 (Department of Chemical Engineering,
• interactive – improved voice and multimedia abilities University of Bath: Bath).
allowing seamless interchange with site personnel. Hebblewhite, B, 2005. MTEC mining engineering education initiatives in
Australia, Presented to SME2005 Conference, Salt Lake City, Utah,
Minimise the size and footprint – physical and USA, 28 February 2005.
environmental Krause, G, 2001. Large heavy haulers: Is bigger better? Presented to
Learning Seminar No 6, University of Alberta: Edmonton.
Much of the ability to reduce the physical impact of Natural Resources Canada, 2005. Benchmarking the Energy Consumption
comminution plants is tied to the ability to treat reduced of Canadian Open-Pit Mines, p 20. Available from: <http://oee.
tonnages for the same or better generation of product. The two nrcan.gc.ca/publications/industrial/mining/open-pit/index.cfm?attr=2
main components impacted by gross throughput are the number/ 4>.
ABSTRACT • the key breakage mechanism in a SAG mill and the role of
This paper summarises some observations regarding the way SAG and mill lining systems, and
AG mills work based on the last ten years of pilot plant operation, • the roll of rock charge mass in determining throughput and
grinding circuit design and operational reviews for various projects in product size.
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: WHAT IS THE KEY MECHANISM FOR
• How do mill lifters and liners really work and have an effect on BREAKAGE IN A SAG MILL?
grinding?
• How does the mill charge composition impact on grinding Current understanding
performance?
The focus of the paper is on observations of mill behaviour in both Royston (2006) described the current understanding of how SAG
pilot plant trials and plant operation, particularly on Cadia ore. DEM, by mills operate at SAG2006. The following three bullet points
others, and JKSimMet modelling of SAG mill performance have greatly summarise the relevant aspects of his paper.
assisted in forming the ideas expressed in the paper and both are valuable
tools in developing our understanding of the comminution process.
• Objective of shell lifter design:
Put in the ‘extreme’ form the major observations and conclusions The two main process objectives in SAG mill
expressed in this paper are that: shell lifter design are to maximise the impact and
• the main purpose of mill lifters is to reduce mill liner wear and not to rate of delivery of grinding media at the toe of
promote rock breakage; the charge where most of the rock-breaking
• the often promoted theory that balls impacting the ‘toe of the charge’ action is assumed to occur and to avoid liner and
is a dominant form of breakage is incorrect; and ball damage.
• 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 • Relationship between mill speed and lifter wear:
circuit configuration. Increasing mill speed over the wear life of shell
lifters can maintain a sweet spot in mill
INTRODUCTION performance by continuing to direct ball impact at
the toe as both the lifter height falls and face
This paper is intended to be, in part, mildly controversial to angle increase with wear … It is worth noting that
promote discussion of how SAG mills work. The content is based highly worn shell lifters can still deliver adequate
on a combination of long-held views developed from pilot trials (though not optimal) milling performance.
by the author over the past 13 years, contemplation of comments
made, and data presented, at SAG 2006 and discussions held • The action of balls in the mill charge:
with mill operations and liner system designers. Within the mill charge, balls have to transfer
Some of the data presented dates from pilot trials conducted in energy into rock breakage via attrition and
the early 1990s whilst other observations relate to recent abrasion through charge motion in the mill … It
performance of large SAG mills, particularly Cadia’s 20 MW is now a common practice, to improve milling
SAG mill. rate, for SAG mills to be run with low charge
AG milling commenced in Australia with mills like the 32 ft levels at maximum ball charge levels. The
diameter Savage River mill in 1966. Since that time, the outcome is to increase ball-charge participation
introduction of balls (semi autogenous milling), initially at low by more turnover of the charge per rotation of
levels, has increased the throughput of mills fed with primary the mill and by increasing the ball-to-rock ratio
crushed ore. By the mid 1990s SAG mill ball loads had increased … With newer mills with higher load capacity,
to 20 per cent and SAG mill diameters were increasing to 40 ft high ball charge levels (say up to 18 per cent)
(Freeport, Cadia). A consequence of this was that the theory have been used. The objective again is to
proposed by liner and mill suppliers that the number of lifters increase ‘ballcharge participation’ through
should be equal to (2 × mill diameter – 2) and that 12 degree increasing the ball to rock ratio, while drawing
leading edges were appropriate, in practice led to ball on shell maximum power at the maximum allowable total
impacts at moderate to high mill speeds. At the same time, charge mass.
efforts to maximise mill throughput led to operation at low total
loads and further exacerbated the potential for mill liner damage Other studies on SAG mill performance have concluded that:
through ball impact. Trajectory calculation software focused on • the majority of the contact, shear, energy consumption and
getting the steel media to impact on the toe or the change – a
breakage occurs in the ‘kidney’ of the charge in a SAG mill
sensible approach to increasing mill liner life. Other outcomes
included reduced inter lifter packing and resulted in slightly (see red area in Figure 1, from Nordell, 2002);
increased mill volume and power draw (Karageorgos, 2001). • the importance of high energy ‘impact breakage’, where
Two main aspects of SAG mill performance are discussed in large rocks are broken into two or more parts by a single
this paper: impact event, varies with ore type and mill operating
condition; and
1. General Manager Technical Solutions, Ausenco Limited, 2404 • the majority of the energy consumed in a SAG mill is used in
Logan Road, Eight Mile Plains Qld 4113. producing fine (say <2 mm) particles through abrasion and
Email: greg.lane@ausenco.com low energy ‘impacts’.
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.
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
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).
This paper proposes that: Karageorgos, J, 2001. SAG milling at Fimiston, in Proceedings SAG2001
Conference, Mining and Mineral Process Engineering (University of
• mill liner/lifter profiles have very little impact on SAG mill British Columbia: Vancouver).
specific energy requirements for a given ore as long as key Lane, G and Dunne, R, 1998. Ore competency and the design of
criteria are met;2 FAG/SAG circuits, Paper presented to Comminution 98, Minerals
• under these circumstances, the main purpose of lifters is to Engineering Conference, Perth.
minimise impact and abrasive wear of the liner plates, and as Loveday, B, Morrison, R, Henry, G and Naidoo, U, 2006. An
a consequence; investigation of rock abrasion and breakage in a pilot-scale AG/SAG
mill, in Proceedings SAG2006 Conference, Mining and Mineral
• the concept of maximising the ‘energy efficiency of Process Engineering, pp III-379 (University of British Columbia:
breakage’ in SAG mills by maximising impact on the toe of Vancouver).
the charge is not of significant consequence. MacNevin, W, 1997. Kidston Gold Mines case study: Evolution of the
comminution circuit, in Proceedings Conference on Crushing and
ACKNOWLEDGEMENTS Grinding in the Mining Industry, Perth, May (IIR Conferences:
Sydney).
The author wishes to acknowledge the indirect input through Morrison, R, Loveday, B, Powell, M, Djordjevic, N and Cleary, P, 2006.
discussion of a number of people, including Steve Hart when at Applying discrete element modelling to different modes of breakage
Cadia and his colleagues, Paul Griffin and Larry Nordell, In in AG and SAG mills, in Proceedings SAG2006 Conference, Mining
addition, the participants in over a dozen SAG mill pilot and Mineral Process Engineering, pp III-407 (University of British
trials, including various personnel from client organisations, Columbia: Vancouver).
engineering companies, research organisations and laboratories Nordell, L, 2002. Personal communication.
over the past 15 years contributed to the learning experience. 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
REFERENCES SAG mill performance, in Proceedings SAG2001 Conference,
Clements, B, 2002. Personal communication. Mining and Mineral Process Engineering (University of British
Hart, S, Nordell, L and Faulkner, C, 2006. Development of a SAG mill Columbia: Vancouver).
shell liner design at Cadia using DEM modelling, in Proceedings Rajamani, R K, Latchireddi, S, Prathy, S K and Patra, T, 2005. Improving
SAG2006 Conference, Mining and Mineral Process Engineering energy efficiency via optimised charge motion and slurry flow in
(University of British Columbia: Vancouver). plant scale SAG mills, Annual report, December, DE-FC26-
Hollow, J and Herbst, J, 2006. Attempting to quantify improvements in 03NT41786 University of Utah.
SAG liner performance in a constantly changing ore environment, in Royston, D, 2006. Current developments in SAG mill liner design, in
Proceedings SAG2006 Conference, Mining and Mineral Process Proceedings SAG2006 Conference, Mining and Mineral Process
Engineering, pp I-359 (University of British Columbia: Vancouver). Engineering (University of British Columbia: Vancouver).
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. FIG 1 - Telfer location.
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 Treatment method
flotation trains with common concentrate handling and residue The new processing plant is divided into two parallel treatment
disposal facilities. The plant recovers coarse gold through a paths (trains) with the following unit processes: crushing,
gravity concentration circuit. Copper and fine gold are recovered grinding, gravity, copper flotation and pyrite rougher flotation.
through a series of different flotation modes; bulk, copper only, The plant has common copper concentrate handling, tailings
sequential and pyrite only flotation. disposal, pyrite cleaner flotation and pyrite concentrate leach
In the feasibility study each ore block was assigned a preferred facilities. The flow sheet for the new processing plant is
treatment route according to its rock type, copper domain and presented in Figure 2.
mineral grades. The following paper describes how the The two processing trains are capable of treating different ore
variability of the Telfer orebody dictates, in a master/slave blends with the treatment modes summarised in Table 1. The
relationship, the flotation circuit configuration required. preferred treatment method for each ore block depends primarily
on the copper (total and cyanide soluble) and sulfur grade of the
BACKGROUND ore and is selected to give the optimum economic outcome.
A gravity gold component is recovered regardless of treatment
The feasibility study evaluated the test work undertaken on method as the gravity circuit is situated in the grinding circuit.
different ore types and assigned gold and copper recoveries to
each ore block according to the following:
Geology and mineralogy
1. treatment method – processing mode,
The stratigraphic setting and copper domains for the principal
2. rock type – lithology, open pit ore source, Main Dome, are depicted in Figure 3 and
3. copper domain – weathering profile, and Figure 4, respectively.
The gold and copper deposits at Telfer are hosted within a
4. head grade – metal species. 1750 m thick sedimentary sequence comprising the Telfer
Formation, Malu Quartzite and Isdell Formation.
1. Senior Plant Metallurgist, Newcrest Telfer Operations, PO Box Several styles of mineralisation have been recognised, namely
6380, East Perth WA 6892. Email: mark.benson@newcrest.com.au narrow high-grade reefs, pod-like mineralised bodies, sheeted
vein-sets and large areas of low-grade stockwork mineralisation,
2. Previously: Telfer Gold Mine, PO Box 6380, East Perth WA 6892.
which form the majority of the sulfide resource. The
3. Manager Metallurgy, Newcrest Operations, Hyatt Business Centre, mineralisation is structurally controlled, but all styles display
Level 2, 20 Terrace Road, East Perth WA 6004. stratabound characteristics.
Email: simon.hille@newcrest.com.au Primary gold mineralisation is associated with pyrite-
4. Metallurgical Superintendent, Newcrest Telfer Operations, PO Box chalcopyrite and quartz-dolomite gangue. However, weathering
6380, East Perth WA 6892. Email: nick.jacob@newcrest.com.au has strongly modified the mineralisation, to depths ranging from
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
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 2. West Dome ore, a low-grade auriferous pyrite orebody
and primary ore types is irregular. Supergene minerals include containing lesser copper sulfides and higher pyrite values
gold with limonite/goethite, malachite and chrysocolla in the than the Main Dome orebody;
depleted zone, giving way to chalcocite, pyrite, digenite,
covellite, tenorite and cuprite at depth. 3. Middle Vale Reef (MVR) ores, comprising remnant reef
ore associated with higher grade massive auriferous copper
The Telfer Project ore sources consist of the following broad sulfide and pyrite mineralisation, with occasional
classifications: pyrrhotite, adjacent to the existing underground mine;
1. Main Dome ore, a low-grade auriferous copper sulfide and 4. M Series Reef (MRF) ore, comprising ore associated with
pyrite orebody providing the majority of the ore tonnage; the narrow vein reefs within the Malu Quartzite; and
5. Telfer Deeps underground mine ore, from the I Series Localised leaching and enrichment of gold and copper, and
development, with higher grade massive auriferous copper oxidation of sulfides with related occurrence of secondary copper
sulfide and pyrite mineralisation with occasional pyrrhotite. sulfides, occur in deeper positions along permeable reef horizons
and structures.
The weathering profile varies across the mine. Around Main Malu Quartzite Member (MQM), Footwall Sandstone Member
Dome, the redox boundary is generally 100 m to 200 m below (FSM), Malu Sandstone Member (MSM) and MRF ore account
the surface although the reef structures and impermeable for the majority of the Telfer open pit orebodies, with MQM
quartzite boundaries have a strong effect on the level of being the dominant lithology. In general, copper domain is
oxidation. strongly correlated with flotation performance and lithology is
Supergene enrichment and leaching of copper and, to a lesser associated with physical characteristics. The MVR and E reef
more localised extent, gold, occurs throughout the mine. Copper (REF) lithologies are typical of weathered reefs, with low
is, in general, far more mobile than the gold in the weathering competency, and have given poorer flotation response during test
profile, being leached from near-surface areas down to roughly work.
60 m to 80 m below the surface, where there is a strong zone of
supergene enrichment. The subhorizontal supergene copper THE MASTER/SLAVE RELATIONSHIP
‘blanket’ is generally near or above the oxide/partially oxidised/ As previously mentioned, the ore fed to the Telfer concentrator
sulfide redox boundary. can be classified into four groups according to the preferred
treatment mode: bulk, sequential, copper only and pyrite only. The copper only mode circuit configuration incorporates flash
The type of ore (rock type and copper domain) being treated flotation with dedicated flash cleaner, copper rougher, copper
dictates the flotation circuit configuration required, or in process scavenger (pyrite rougher) and copper cleaner/recleaner circuits.
control language, is the master control element. Processing of ore The range of Cu:S ratios requires selective flotation conditions,
in the incorrect treatment mode, either due to operational error or where the amount of pyrite reporting to the final concentrate is
by misclassification of ore, can have detrimental effects on gold managed through chemical depression. This is achieved by a
and copper recovery and copper concentrate grade. In this way combination of a selective thionocarbamate copper collector
the flotation circuit and the mineral products are slaves to the (RTD11A) and lime/cyanide addition.
mineralogy of the ore blend being fed while accommodating the
RTD11A is preferentially added to the flash flotation feed with
market constraints for the products.
the copper rougher circuit fed with the grinding circuit cyclone
overflow. Additional RTD11A can be added to the head of the
Bulk flotation copper rougher and scavengers as required. The cyanide is dosed
The simplest of the flotation modes at Telfer, bulk flotation is in the flash cleaners and copper cleaner/recleaners to depress
defined as a non-selective copper flotation process where all pyrite and maximise the copper concentrate grade.
sulfide species are floated and a bulk copper concentrate is Copper only flotation mode results in an economic copper
produced. It is typically characterised by Cu:S ratios >0.7, concentrate containing both Cu and Au as the main valuable
generally being high in chalcocite and low in pyrite. There can minerals. If the ore entering the plant has a lower Cu:S ratio of
be a variety of other copper species present (covellite, digenite, <0.4 and is low in CNSCu then it is possible to exploit this
chalcopyrite, bornite, native copper, malachite, cuprite) but will difference by recovering greater amounts of the pyrite by
typically produce an economic concentrate of >20 per cent Cu. reducing the strength of the cyanide depression. This may lead to
The majority of the bulk ore comes from the upper open pit a change in the circuit configuration. Typically CPS is not
region, which is dominated by the supergene structures. utilised in copper only mode as it leads to large amounts of pyrite
The bulk mode circuit configuration incorporates flash activation and the flotation circuit may struggle to achieve the
flotation with dedicated flash cleaner, copper rougher, copper target concentrate grade. So even within the selected processing
scavenger (pyrite rougher) and copper cleaner/recleaner circuits. modes the specific mineralogy dictates the performance of the
The high Cu:S ratio allows for the non-selective flotation circuit in a master/slave relationship.
conditions. These conditions include a non-selective sulfide
collector in potassium amyl xanthate (PAX), lime addition to Sequential flotation
adjust pH to 9.0 and sodium sulfide (Na2S) to utilise controlled
potential sulfidisation (CPS). The most difficult flotation mode that Telfer has to offer,
PAX is preferentially added to the mill discharge to allow sequential flotation is defined as a selective copper and pyrite
conditioning time, servicing both the flash flotation, located in (gold) flotation process where the copper sulfide species are
the grinding circuit recirculating load, and the copper rougher floated to produce a copper concentrate while the pyrite
circuit, which is fed with the grinding circuit cyclone overflow. component of the ore is depressed using high pH and cyanide.
Additional PAX can be added to the head of the flash and copper The pyrite is then activated and floated to recover gold through a
roughers as required. Na2S for CPS is dosed in the feed to the separate processing circuit consisting of cyanide leaching and
pyrite roughers and copper cleaner conditioning tank to cyanide recovery. Sequential ores are typically characterised
maximise the recovery of weakly sulfidic minerals, copper by Cu:S ratios <0.4, with a mixture of copper sulfides,
carbonates and copper oxides. predominately chalcopyrite. They also contain high proportions
Bulk flotation mode results in maximum recovery of gold and of pyrite bearing gold and are typically low in CNSCu. The
copper whilst maintaining the target concentrate grade. However, majority of the Sequential ore comes from the lower open pit and
if the ore entering the plant has a lower Cu:S ratio of <0.7, it Telfer Deeps, which is dominated by the Malu and Isdell
becomes very difficult to achieve a saleable copper concentrate, stockwork and reef structures, is primary in nature with minimal
which may lead to a change in the circuit configuration. When weathering.
the ore blend is high in copper oxides and copper carbonates, Different conditions are required to recover the copper and
whilst still indicating a high copper to sulfur ratio, the flotation pyrite concentrate separately. The copper selective collector,
circuit may struggle to achieve the target concentrate grade. This RTD11A, is added to the mill discharge, head of the flash
is where the CPS is of greatest value if applied correctly but can rougher, head of the copper rougher and the copper cleaner
lead to a dilution of the final concentrate grade. So even within circuits. Utilising RTD11A in combination with lime/cyanide
the selected processing modes the specific mineralogy dictates addition to the flash cleaners and copper cleaners allows a
the performance of the circuit in a master/slave relationship. saleable grade copper concentrate to be produced. PAX is not
used in the flash flotation and copper rougher circuits to
Copper only minimise pyrite recovery to the copper concentrate. The copper
rougher tail feeds the pyrite rougher circuit where CPS and PAX
One of the more challenging flotation modes at Telfer, copper addition ensures maximum recovery of pyrite and gold to the
only flotation is defined as a selective copper flotation process pyrite cleaner circuit. To reduce the copper content of the pyrite
where only copper sulfide species are floated to produce a copper cleaner concentrate, a two-stage cyclone desliming circuit is
concentrate. The pyrite component of the ore is not economic to used. The deslimed pyrite concentrate feeds the pyrite leach
recover for further processing. It is typically characterised by
circuit to recover gold.
Cu:S ratios in the range between >0.4 and <0.7 generally being a
mixture of copper sulfides, oxide and carbonates with moderate Sequential flotation has similar issues to copper only treatment
to high proportions of pyrite. These copper only ores are in that the amount of cyanide required for depression is
generally characterised by very high proportions of cyanide dependent on the proportion of pyrite to the copper minerals in
soluble copper (CNSCu) which, due to their slow floating nature, the feed. High quantities of CNSCu or a high CNSCu:Cu ratio in
would report to the pyrite concentrate in unfeasible quantity. the feed can result in slow floating copper minerals exiting in the
Another feature of this processing mode is that pyrite depression copper rougher tail and being activated in the pyrite rougher
is necessary to produce an economic concentrate of >20 per cent circuit. Recovery of copper minerals into the pyrite concentrate
Cu. The majority of the copper only ore comes from the results in excessive cyanide consumption in the leach circuit. If
transitional area of the upper open pit, which is dominated by the the ore to be fed contains a CNSCu:Cu ratio greater than
supergene structures which have been highly weathered. 0.5 and/or a CNSCu grade above 400 pmm, it is likely that the
copper reporting to the pyrite concentrate will be too high to addition, the feed could vary dramatically over short periods of
feasibly treat in the leach circuit. The ore may then require bulk time requiring significant changes to the flotation circuit to
or copper only treatment (depending on Cu:S grade). Hence the maintain froth stability, pull rates, concentrate grade and
processing team needs to decide which of the minerals is to recovery.
dominate the economic decision (master) and adjust the circuit Some ore types contained high levels of oxidised clay-type
accordingly (slave). minerals from the outer siltstone member (OSM) section of the
open pit. With such ores, it was found that the slurry was
Pyrite only excessively viscous at the design flotation density of 35 per cent
solids, resulting in poor air dispersion through the pulp, boiling
One of the challenging flotation modes at Telfer, pyrite only and froth instability. As a result the float feed density target had
flotation is defined as a pyrite (gold) flotation process where to be dropped to 30 per cent and at times as low as 28 per cent
there is insufficient copper sulfide species to produce a copper
solids, requiring significantly higher addition of water through
concentrate. The pyrite is floated into an economic concentrate
the grinding circuit. At times the higher water additions resulted
to recover gold through a separate processing circuit. It is
in lower throughputs due to limited supply of make-up water.
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 Flash flotation commissioning
typically very low in cyanide soluble copper. The majority of the
On plant start-up the flash flotation circuit was not yet available.
pyrite only ore comes from the lower open pit and Telfer Deeps,
which has been depleted in copper minerals. Like sequential ore As previously mentioned, the flash flotation treats a portion of
the lithology is dominated by the Malu and Isdell stockwork and the recirculating load (cyclone underflow) in the grinding circuit.
reef structures and is primary in nature with minimal weathering. This removes easily floatable minerals at the earliest possible
opportunity preventing over-grinding and therefore reducing
Pyrite only mode recovers only a pyrite concentrate, and
therefore a more simplified reagent regime can be employed. reagent consumption and improving roughing circuit performance.
However, the main concentrate production needs to come from Approximately 40 per cent of all copper concentrate produced is
the pyrite roughers to enable recovery through the pyrite cleaners from the flash flotation circuit.
to the pyrite leach circuit. This requires the majority of the The flash flotation circuit consists of an Outokumpu SK1200
flotation circuit to be under-utilised as there is no copper rougher cell with the concentrate reporting to two OK6.5TC
concentrate produced. Both CPS and PAX are applied to the flash cleaners producing a final concentrate.
conditioning tanks at the head of the pyrite rougher circuit to Without the flash flotation circuit, the roughing and
ensure maximum recovery of pyrite and gold to the pyrite cleaning/recleaning circuits became overloaded in terms of froth
cleaner circuit. mineral loadings. Increased frother dosages were required to
Pyrite only flotation has some similar issues that exist in maintain froth stability in the roughing circuit and throughput
sequential flotation in that a careful watch needs to be was restricted to avoid high tails metal losses.
maintained on the amount of CNSCu which reports to the pyrite
concentrate, due to the impact that the copper species will have Once commissioned, the high levels of metal production from
on the cyanide consumption in the pyrite leach. Again the the flash flotation circuit initially caused some significant
processing team needs to decide which of the minerals is to materials handling issues. The concentrate recovered included
dominate the economic decision (master) and the circuit to adjust significant levels of native copper, which combined with almost
accordingly (slave). 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
COMMISSIONING EXPERIENCE
concentrate over a long and indirect route. A redesign of the
pumping system was necessary, including upgrading the pumps,
Bulk flotation commissioning redesigning the transfer pipework and incorporation water
First ore was processed through Train 2 on 23 November 2004. injection points. Current operation relies heavily on water
At this time there was very limited instrumentation and addition and process control to manage the high specific gravity
construction of Train 1, and parts of Train 2 were continuing. material over the long pipe run.
Some of the critical elements for flotation operation and
monitoring were not available or not adequately commissioned Suboptimal process control tuning and
and calibrated. Train 1 was commissioned on the 18 January availability of instrumentation
2005.
The commissioning of a new concentrator is usually a As a result of significant instability in the grinding circuit, the
challenging task and at Telfer this was certainly the case for a flotation circuits experienced large flow surges. This would cause
number of reasons. flotation cell pulping and overflows due to the ineffectiveness of
the float level control to deal with such large surges.
Complex, highly variable ore mineralogy The proportional, integral and derivative (PID) settings for the
various control loops within the Telfer concentrator were initially
Bulk flotation ores have many different copper domains and rock set at default levels for slurry commissioning. These settings
classifications in addition to grade variations. Copper mineralogy were changed ‘on the fly’ as the circuit quickly ramped up to
varies from oxides, carbonates, partially oxidised copper sulfides design throughputs with over 3000 m3/h of slurry entering the
to readily floatable copper sulfide minerals such as chalcocite, flotation circuit. Control of flow control valves for water, air and
bornite, covellite and chalcopyrite. This mineralogy variability is reagent addition and level control on hoppers required rapid
typical of a supergene geological structure, which some refer to tuning to minimise the environmental impact of slurry spillage
as a ‘mineral soup’. The variable copper grades and mineralogy and to allow some semblance of steady operation. Bedding down
result in different flotation characteristics and therefore variable of the control loops during the commissioning of Train 2 was of
flotation operating targets, such as pulp densities and reagent dose an extremely high priority, not only to ensure environmental
rates. impact (spillage outside of bunded areas) was minimised whilst
During commissioning the effect of the different copper recovery targets and concentrate specifications were met, but
domains and rock types on flotation was not well understood. In also due to the fact that Train 1 was still under construction.
Construction activities were being conducted within close Reagent addition and CPS optimisation
proximity of hoppers and slurry lines, with loss of control having
potential negative ramifications on industrial relations. Early Reagent addition rates from the feasibility study test work
learnings during the commissioning of Train 2 were immediately provided a starting point for dose rates but were quickly reduced
applied to Train 1 before commissioning of this circuit two to gain better control of froth stability without adversely
months later. affecting concentrate grade and recovery. This is common when
scaling up bench top test work flotation reagent regimes to plant
Some of the critical elements for flotation operation and scale operation.
monitoring were not available or not adequately commissioned
and calibrated. One such piece of equipment now considered Initial addition rates for MIBC exceeded 30 g/t to ensure
critical to the flotation process at Telfer is the multi-stream adequate froth stability to the flotation circuits. Alternative
analyser (MSA). The MSA analyses a number of critical streams frothers to MIBC were trialled, resulting in the switch to an
throughout the circuit via an X-ray fluorescence elemental alcohol-glycol blend frother, DSF-004. Stronger frothers were
analyser. The outputs of estimated grades for copper, iron, per trailed which provided improved flash flotation stability;
cent solids and inferred sulfur can then be used to monitor stream however, a compromise was made to reduce adverse overfrothing
grades in order to optimise the process. in the cleaning circuit. Frother addition rates, DSF004, have
Poor reliability in the pumps feeding the analyser resulted in since been reduced to 3 - 10 g/t.
this important control system being unavailable for much of Telfer commenced operation utilising PAX for the non-
the first year of operation. During commissioning and the selective flotation of sulfide minerals into a bulk copper
subsequent period until the MSA was completed, flotation concentrate. Initial addition rates were based on feasibility, at
control was maintained by collecting spot samples of critical times exceeding 150 g/t. The addition point of the PAX was
streams for assay every four hours. The assaying had on average found to have a significant effect on froth stability. Addition of
a six-hour turn around time. Often, by the time an issue was PAX at the head of the flash and copper rougher circuits revealed
identified in the circuit, it had already been half a shift before detrimental effects on froth stability due to a lack of
any corrective action could take place, resulting in lower than conditioning. Addition of PAX to the mill discharge became the
desired recoveries and/or concentrate grades. preferred option providing conditioning time prior to both the
Resolving the pump issues and the opportunity to calibrate the copper rougher and the flash circuit, PAX addition was refined to
streams has now allowed the MSA to be used as an instantaneous 30 - 50 g/t by being staged throughout the circuit, though a
measure of the flotation circuit performance, resulting in much change in the ore being delivered to the plant meant treatment in
tighter control over pull rates and the addition of reagents. bulk mode, with exclusive use of PAX as the collector, was short
The digital froth camera system initially installed at Telfer lived.
proved to be unreliable due to deficiencies in the hardware Treatment of high sulfide feed and ore with a high Cu:S ratio
architecture utilised and the negative effects of sunlight. This required the introduction of RTD11A, a copper selective
system was not employed during commissioning or in the first thionocarbamate collector. Operating in copper only mode
year of operation. The system required an upgrade and has now became the norm in the early stages of operation, with RTD11A
been incorporated into Telfer’s process control. The digital froth added to the mill discharge, as well as the heads of both the
camera system is an important control method used to monitor copper and flash roughing circuits. The nature of RTD11A meant
concentrate pull rates ensuring they are adequate through the
little conditioning time was required and no issues were found
roughing circuits and manipulate the cell levels as required.
with froth stability.
Controlled potential sulfidisation (CPS) utilises the addition of
Froth stability issues sodium sulfide (Na2S) or sodium hydrogen sulfide (NaHS) in
The initial commissioning period was punctuated by several order to reactivate partially oxidised copper sulfide or carbonate
froth stability issues. The main causes were both mineralogy and minerals. Coating of the mineral surface with fresh sulfide ions
chemically driven. The mineralogy issues related more to the allowed sulfide collector absorption and flotation recovery. The
host rock rather than the type of sulfides being recovered. The CPS at Telfer took considerable effort to understand and
highly weathered outer siltstone member (OSM) ore, which hosts optimise.
the E reef structure, caused the most significant froth issues. As
Feasibility study copper scavenger CPS was controlled to a
previously mentioned the viscous slurry affected the formation of
the froth, whilst the naturally hydrophobic gangue minerals slurry potential of -200 mV Eh. Operating CPS to this Eh set
produce a highly stable froth once formed. point in the plant proved to be suboptimal. Instead, operating to a
sulfide ion potential (Es) of -500 mV was determined to be
Telfer commenced operation utilising methyl isobutyl carbinol
optimal for the flotation of malachite, the predominant oxidised
(MIBC), a weak alcohol-based frother. As Train 2 commenced
operation without the flash flotation circuit, increased mineral copper mineral present in the Telfer orebody. Once this was
loadings in the copper rougher circuit resulted. The mineral understood the measurement of the slurry potential was
loading, along with the viscous nature of the OSM ore, required converted from standard oxidation reduction potential (Eh) to
high addition rates of MIBC to stabilise the froth. that of the sulfide ion only (Es) through the use of a sulfide ion
The Telfer circuit was then found to be very sensitive to selective electrode.
frother addition. Increasing frother to the rougher circuit resulted Depression of pyrite is critical at Telfer, both to maintain a
in ‘over-frothing’ in the cleaner circuit making the cells ‘race’ saleable copper concentrate, and to ensure the gold bearing
with uncontrollable mass pull rates. The large proportion of pyrite reports to the leaching circuit via the pyrite roughers. Lime
gangue entrainment in the froth reduced the grade, and the was initially the only reagent available for this. Lime was added
persistent nature of the froth reduced the efficiency of the cleaner both as quick lime to the concentrator feed belts and milk of lime
and recleaner concentrate pumps. Recovery was then impacted to the flotation circuits. The lime was used to modify pH to 9 to
due to lower achievable mass pull. 10 in the roughing circuit and 10 to 11 in the cleaning circuit.
To resolve frothing issues, close attention was required to ore Minor success in depressing the pyrite resulted; however, the
blending the limit the proportion of OSM ore. As covered below, nature of the ore treated and the delay in the construction of the
the frother strength and type was altered to provide greater permanent cyanide dosing circuit required the installation of a
control and the addition points optimised. The addition rates temporary cyanide dosing system. The temporary system dosed
were drastically reduced and a higher level of process control high cyanide addition rates to the flash and copper rougher
was adapted. circuits to depress pyrite. Once the permanent system was
available, preferential cyanide dosage to the cleaner circuits • Reagent conditioning time is often excluded in final design
drastically decreased reagent consumption whilst significantly in a quest to reduce capital cost; however, when dealing with
improving control over pyrite depression. This optimisation is a a complex chemical flotation process it can be invaluable to
change from design but has enabled a higher level of control and both fast optimisation and process control.
improved performance.
Sequential
Sequential commissioning
• Excessive cyanide depression of pyrite results in poor
Limited amount of sequential ore feed was available from the re-activation and recovery;
mine in the upper levels of the open pit delaying sequential
flotation commissioning by 14 months. Sequential flotation
• modifying addition of depressant into the cleaning block
allows for tighter control, reduced reagent consumption and
commissioning occurred in January of 2006, coinciding with
enhanced performance;
the production ramp-up of the redeveloped Telfer Deeps
underground operation. • pyrite re-activation was enhanced at pH 8.0 - 7.0, through the
In the initial sequential trial, it proved very difficult to float the addition of acid;
pyrite after successful pyrite depression through the addition of • on fresh sulfide ores, a slightly negative pulp Eh (-80 to
cyanide and copper flotation. The reason for this was not known -120 mV) in the pyrite conditioning discharge has provided
at the time, but was suspected to be a re-activation issue, related the best flotation results;
to either pH or Eh. Test work showed: • 50 - 60 g/t PAX addition is optimal to recover pyrite but not
• excessive cyanide depression results in poor re-activation; leave residual PAX in process water that may float pyrite in
the flash and copper roughers;
• pyrite re-activation was enhanced at pH 8.0 - 7.0, through the
addition of acid; • treatment of oxidised ores in sequential mode results in
excessive levels of oxidised copper minerals in the pyrite
• on fresh sulfide ores, a slightly negative pulp Eh (-80 to concentrate, as they require CPS to activate, resulting in
-120 mV) in the pyrite conditioning discharge has provided uneconomic downstream cyanide leaching;
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
• residual reagents in the process water (recycled) were the (copper and pyrite); and
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.
• 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. CURRENT OPERATING PHILOSOPHY
On fresh sulfide ores, typical grades of pyrite thickener feed Bulk
have been 0.5 per cent Cu and 35 per cent S. Sequential
treatment of oxidised ores (high CNSCu content) result in pyrite At the current point in the project life, there is not a great deal of
concentrates with a high copper content (>3 per cent) as the bulk ores coming from the open pit. Bulk treatment is the easiest
oxidised copper minerals do not float in the copper circuit, but and most favourable treatment mode and when bulk ores are
are activated by CPS and report to the pyrite concentrate. treated, the flotation conditions are geared for maximum
To reduce the copper content of the pyrite cyanide leach feed, recovery with PAX collector addition at higher dose rates and
a two-stage cyclone deslime circuit is used. Analysis of copper CPS to recover all floatable minerals. The high Cu:S ratio allows
minerals in pyrite concentrate by size shows these minerals are good concentrate grades at these less selective flotation
predominantly in the slimes size fraction. Optimisation of conditions.
cyclone cut points for most effective copper rejection is ongoing.
Copper only
KEY LEARNINGS – COMMISSIONING Copper only mode is the least favourable in terms of metal
recovery. Due to concentrate grade constraints, it is not possible
Bulk to recover all the pyrite into the copper concentrate, and due to
the presence of oxidised cyanide soluble copper minerals, it is
• Ore variability has a significant effect on flotation operating
not possible to produce a separate economically treatable pyrite
parameters; ore blending to maintain steady feed
concentrate. Therefore, there are gold losses associated with the
characteristics is critical.
pyrite that reports to tailings.
• MSA is critical for optimal operation of the flotation process. A number of strategies have been employed to minimise such
• Froth stability needs to be maintained through all ore blends losses, including adding cyanide depressant to the cleaners only
and different addition strategies need to be applied to to maximise free gold recovery and the use of stronger PAX
accommodate both ore variability and flotation unit collector in the scavengers to increase pyrite recovery if
performance. concentrate copper grade allows.
• Feasibility study reagent additions were much higher than
required in the processing plant. Bench scale test work Sequential
regimes need to be viewed with a high degree of scepticism To ensure good copper concentrate grades and a clean pyrite
and an open-minded approach to the initial operating reagent concentrate, it is essential to get a good split between the copper
regime should be adopted. and pyrite concentrates. This is controlled by correct reagent
• Reagent addition rates have a huge effect on froth stability in additions, pull rates and ore classification and blending.
addition to flotation performance. Close control of reagent The decision whether an ore is to be treated in copper only or
addition is required at all times. sequential mode is an economic one – if the pyrite concentrate
produced is economic to leach, the ore is treated in sequential • The importance of ore blending based on mineralogy and
mode. In addition to the economics, it is necessary to control flotation response, not necessarily solely on economic metal
copper metal in the pyrite concentrate fed to the leach circuit to grades.
ensure the SART copper and cyanide recovery process is not
• The role that process control has in stabilising and optimising
overloaded, this is managed by ore blending and correct
the process. Some consideration should be given to
classification of sequential ore. allocation of a separate process control team, which is
dedicated to this task alone during commissioning.
CONCLUSION • The need to approach all bench scale reagent regimes with an
The more complex the mineralogy the more processing options open mind when applied to the processing plant and rely
are necessary to accommodate the variability while still achieving more heavily on experienced operators and visual indicators
an economic result. The Telfer ore processing circuit is unique in to conduct coarse reagent adjustments.
this way for a copper-gold deposit where the minerals dictate, in • Reagent conditioning time can have a major influence on the
a master/slave relationship, the processing configuration of the effectiveness and quantity of the reagent necessary.
flotation circuit. • Close attention needs to be paid to the water balance and the
The commissioning of the new Telfer concentrator, while impact of recycled water on the chemical balance within the
challenging, has been an extremely successful ramp-up of a large circuit.
dual train concentrator, exceeding nameplate capacity within the
six weeks of operation which is now consistently processing at ACKNOWLEDGEMENTS
rates in excess of 20 Mtpa. The flotation performance for both
The authors would like to thank Newcrest Mining Limited and
copper and gold has also tracked closely to the feasibility study
the management at Telfer Gold Mine for permission to publish
estimates, which are primarily based on fixed tailings grade this paper. The authors would also like to recognise the entire ore
models. processing team for their commitment and perseverance through
The key learnings from the commissioning that the the challenging times of commissioning to renew Telfer and
metallurgical team would like to share are: optimise a world-class ore processing plant.
The collision efficiency is mainly controlled by the bulk RCellEnt = Rw (ENT) (7)
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 TMRT =
(
V 1− εg ) (8)
and density, bubble size and velocity all affect the value of the Q
collision efficiency. For fine particles (<12 µm), the collision
efficiency often limits their flotation rate, giving rise to reduced where:
recovery at plant scale. Following collision, the particle must
attach to the bubble. This process is controlled by the pulp and V is the cell effective volume
surface chemistry of the system that, together, ultimately control εg is the gas hold-up
the surface hydrophobicity. The attachment efficiency is defined
as the fraction of colliding particles that reside on the bubble for Q is the volumetric flow rate of slurry in the cell feed
a time greater than the induction time (Dobby and Finch, 1987; This paper focuses on two comprehensive surveys (Survey 1
Dai, Fornasiero and Ralston, 1999). Surface hydrophobicity may and Survey 2) in the C3 roughers in which cell-by-cell and
be quantified by the contact angle. In the case of particle size-by-size metallurgical, hydrodynamic, and pulp/surface
attachment to the bubble, the receding contact angle is most chemistry information were obtained. The two surveys were
relevant. In this current paper, Time-of-Flight Secondary Ion conducted at significantly different feed tonnage giving rise to
Mass Spectrometry (ToF-SIMS) is used to estimate the copper significantly different feed particle size distributions and pulp
sulfide mineral receding contact angle, based on an existing residence times. Survey 2 was used to calibrate the model to the
correlation between contact angle and adsorbed collector size-by-size flotation data, while Survey 1 was used to
intensity determined by ToF-SIMS for single mineral samples of corroborate the model parameter values. A series of simulations
chalcopyrite (Duan, Fornasiero and Ralston, 2003). The stability are then performed with the aim of quantifying the change in
efficiency (Es) is controlled both by hydrodynamics and surface copper recovery, on both a size-by-size and unsized basis, with
hydrophobicity. The probability of bubble-particle aggregate changes in values of the hydrodynamic parameters and surface
stability depends on the attachment force between the bubble and properties of the copper sulfide mineral. This was carried out in
the particle in relation to the external forces in the cell an effort to identify the most promising route to increase overall
environment. Schulze (1977) suggested that the average mean copper recovery in roughing.
vessel energy dissipation is more important in determining the
mean stability of the aggregates, rather than local energy EXPERIMENTAL
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 Plant survey conditions
stability efficiency becomes important for particles greater than Two plant surveys were conducted at significantly different feed
150 µm. The collision frequency between bubbles and particles, tonnages and particle size distribution (Table 1). The C3
z, in a turbulent environment is highly dependent on the local concentrator consists of three parallel rougher banks, consisting
energy dissipation in the cell, bubble and particle size and pulp of four stages per bank, and 3 × 85 m3 Wemco cells per stage.
viscosity, and is effectively independent of buoyancy-related The feed tonnage during the surveys in Table 1 is for a single
forces (eg Abrahamson, 1975). However, in the quiescent zones bank (A) of cells, the bank feed tonnage being calculated from
of a flotation cell, bubbles and particles will rise and settle due to the C3 SAG mill feed rate, less transfer to C4, and assuming
buoyancy. These body forces are also included in the collision equal splitting of the slurry between the three rougher banks.
frequency calculation used here (Bloom and Heindel, 2002). Equal splitting was confirmed by tracer tests on the slurry splitter
Finally, the recovery of particles from the collection zone to system. For Survey 2 at the higher feed tonnage, the pulp density
the launder is described. Using the collection zone rate constant, of the rougher bank feed increased, due to reduced water
kc, the particle recovery from the collection zone for a perfectly additions in grinding at the higher feed tonnage. The increased
mixed cell can be calculated as follows. The mixing regime in pulp density in rougher feed is reflected also in the rougher
the cell must, of course, be verified. In the case of the PTFI tailing pulp density, discussed further below. The increased pulp
rougher cells a tracer test was conducted to demonstrate that the density has consequences on coarse copper recovery as discussed
liquid phase was perfectly mixed. further below. Lime addition to the SAG and ball mill feeds
maintained the pH at 9.5 in rougher feed for both surveys 3. the -38 µm fraction was cyclosized, and the cyclosizer
(Table 1). The primary copper collector, Cytec 7249 (a mixture discharge collected in a centrifuge to produce a fraction
of mono- and di-(isobutyl) thiophosphate collector), is added to +3.1 - 12.9 µm on copper sulfide (Fraction 7 in Table 2),
the ball mill feeds at an addition rate determined from the copper and the cyclosizer cone products were collected (Fractions
units in the rougher feed. The secondary copper collector, 5 and 6 in Table 2); and
sodium isobutyl xanthate (SIBX), is stage added to each rougher
stage at an addition rate determined from the copper units in the 4. the +38 µm fraction was dry sieved at 300, 150, 75, and
rougher tailing (Table 1). The alkyl group chain length (R = 4) of 38 µm.
the collectors controls the maximum receding contact angle The final size ranges on copper sulfide, are shown in Table 2,
attainable on the copper sulfide mineral surface (Sutherland and
as well as the distribution of copper sulfide in the feed to the
Wark, 1955), the consequences of which are discussed further
rougher bank for Surveys 1 and 2. In the analysis of the data,
below. The frother is OTX 140 (Table 1).
some size fractions have been grouped as very coarse, coarse,
intermediate, fine and ultra-fine size fractions (Table 2). Size
Metallurgical sampling and analysis distributions within the size fractions were confirmed by a light
The metallurgical surveys consisted of measuring the mass flow scattering technique (Malvern Mastersizer) on separate samples
rate of concentrate for individual cells and cell groupings. The of chalcopyrite. Table 2 also shows the particle size point used
cell groupings are discussed further below. Further, samples of for plotting of the size-by-size data for copper sulfide and for the
the rougher bank feed, timed individual cell concentrates, model calculations. The particle size point is based on the
combined bank concentrate, intermediate stage tailings, and volume average of the lower and upper limits of the particle size
rougher bank tailing (duplicate samples) were also collected. The range (Table 2).
samples were weighed, filtered, dried and weighed again to The program Banks for Excel was used to mass balance the
calculate the pulp per cent solids. Head samples were analysed streams on a size-by-size basis, and to determine the flows of
for copper, iron, sulfur, silver and gold. The metal assays were minerals and water, recovery of minerals and water, pulp specific
converted to copper sulfide, pyrite and non-sulfide gangue (NSG) gravity, residence time per cell, and the undistributed rate
assays assuming that all the copper was in a mixture of
constant, k, of each mineral, per size range, per cell grouping.
chalcopyrite (90 per cent) and chalcocite (ten per cent) (based on
The individual flow rate of cell concentrates were uniformly
Mineral Liberation Analysis (MLA) of products), sulfur not
resident in copper sulfide was contained in pyrite (FeS2), and that scaled to achieve agreement between the total flow rate of
the remaining sample mass was as NSG. concentrate and the concentrate flow rate determined from the
copper metal balance over the whole bank and the bank feed
Subsamples were also sized using a specific procedure (see
tonnage.
Table 2):
Entrainment is controlled by water recovery (Rw) and the
1. subsamples were wet screened at 38 µm; entrainment factor (ENT). The water recovery values are
2. the -38 µm fractions were passed through a 30.2 mm high designated per flotation cell in the model and are the same values
pressure (90 psi) pre-cyclone to produce an overflow measured in the plant surveys. The entrainment factor values are
fraction -6.8 µm on copper sulfide (Fraction 8 in Table 2); calculated using parameters necessary in an empirical model
TABLE 1
Plant survey conditions.
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
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.
(Savassi et al, 1998), (δ) the drainage parameter, and (ξ) the 2.0
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
the bubble rise velocity is likely to be much lower due to the high commensurate increased pulp density in the rougher feed and
bulk viscosity of the slurry and the adsorption of surfactants and tailing streams (Table 4). Consequently, increased residence time
contaminants. In this present study, the effect of bulk viscosity in Survey 2 cannot be a factor to account for the copper recovery
only on the bubble rise velocity is taken into account. Assuming decrease in this survey. There is very close agreement for both
that the bubbles behave as low density, hard spheres the bubble surveys between the feed head copper grade and recalculated
rise velocity using the well-known drag coefficient method may head copper grade, based on the grades and flows of the
be used (Lapple and Sheperd, 1940). Here CdReb2 is calculated, individual concentrates and rougher tailing streams (Table 4).
where Cd is the drag coefficient and Reb is the bubble Reynolds
number. The calculated CdReb2 is 5.43 and, using the standard Hydrodynamic results
correlation, Rep may be determined as 0.21. Using the equation The mean bubble diameter was consistently between 1.3 to
for the bubble Reynolds number (Lapple and Sheperd, 1940), the 1.7 mm across the rougher bank (Figure 1). The superficial gas
bubble rise velocity is calculated as approximately 1 cm s-1 rate generally increased down the rougher bank from
(Table 3). approximately 0.7 cm/s to 1.2 cm/s. These values are in the
A residence time distribution test of the rougher bank was normal range for cells of this type (Power, Franzidis and
performed using a saturated NaCl solution as a tracer. The tracer Manlapig, 2000). It is important to note that for the calculation of
test showed that the flotation cells in the rougher bank behaved the fundamental model, discussed further below, that the rate
as perfect mixers and thus the recovery calculation shown in constant and recovery was calculated for each cell separately,
Equation 4 could be used. While the measured residence time is using the bubble diameter and superficial gas rate separately
not a necessary input variable to the model (the residence time is measured for each cell. It is also important to note that the
calculated based on the feed flow rate, the cell volume and the occurrence of slower floating components of copper sulfide
gas hold-up), it does provide a useful check for some of the mineral in the latter stages of flotation (cells 7 to 12), as noted
model calculations. The measured residence time per cell (in further below, can not be explained by adverse changes in either
Stage 1) determined from the tracer test was 3.0 minutes, while bubble diameter or superficial gas rate down the bank of rougher
the calculated residence time for Surveys 1 and 2 was 2.6 and 3.2 cells.
minutes respectively in Cell 1 (Table 4).
Metallurgical results
Surface analysis
The flotation results will be considered firstly on an unsized
ToF-SIMS was used to estimate the receding contact angle on basis (Figure 2). In increasing the feed tonnage in Survey 2, it is
copper sulfide mineral phases. The ToF-SIMS spectra were clear that the proportion of slow floating components increased.
obtained using a PHI Model 21000 TRIF II system equipped The reasons for the change in floatability are likely to be related
with a gallium liquid metal ion gun in pulse mode. Imaging of to one or a combination of changes in:
the sample involved mapping for positive and negative ions of • feed particle size distribution,
interest. Twenty-five copper sulfide particles were analysed in
the concentrate and tailing to obtain statistical information on the • liberation class of the copper sulfide mineral in the size
amount of hydrophobic and hydrophilic species on the surface. fractions,
• gas dispersion down the bank of cells, and
RESULTS • hydrophobicity of the copper sulfide mineral surface.
It is not the purpose here to quantify these differences, though
Metallurgical reconciliation it is possible to do this with the aid of the fundamental flotation
The increased feed tonnage in Survey 2 caused a decrease in model. The principal contribution comes from the change in feed
copper recovery, across all size fractions, of approximately particle size distribution. The decrease in copper recovery with
5.1 per cent (Table 4). In spite of the increased feed tonnage the increased feed particle size (Table 4) is consistent with
in Survey 2, there was increased mean pulp residence time in historical correlations of copper rougher recovery with the per
Survey 2 due to the reduced water additions in grinding, and 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
% in Size Range
80 Cell 6
% Passing
Fundamental Model Survey 2
Recovery (%)
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.
100
Residence Time (min)
Cell 12
0 5 10 15 20 25 30 35 40 45 (A)
80
Cell 3
40
0.10
Cell 2
20
Cell 1
0
(A) Survey 1 Cell 12
1.00 (B)
80
Cell 6
60
0.10 Cell 3
40
Cell 2
20
Cell 1
(B) Survey 2
0.01 0
+300 m 1 10 100 1000
--75 +12.9 m Particle Size ( m)
- 300 +150 m
--12.9 m
-150 +75 m FIG 5 - Copper mineral recovery as a function of particle size and
cumulative number of cells for Survey 2. The continuous lines are
FIG 4 - Fraction remaining (log scale) as a function of cumulative for the fundamental model fit for (A) copper sulfide bearing
mean pulp residence time for (A) Survey 1 and (B) Survey 2, and particles having a contact angle of 78° in all particle size fractions
for various size fractions +300, +150 - 300, +75 - 150, +12.9 - 75, and (B) copper sulfide bearing particles having a variable contact
-12.9 microns. The continuous lines are for a floatability angle in each size fraction as shown in Table 7.
component model with two components with parameters shown
in Table 5.
minutes in Survey 2) (Figure 2). However, increasing their
flotation rate may have a small impact upon their recovery as
further below, using the fundamental model. At this stage, it can discussed further below.
be stated that the copper sulfide minerals in size fractions below
150 µm should behave uniformly with respect to particle contact In the case of particles greater than 60 µm, there is evidence
angle, assuming that mineral locking only contributes to changes for both a reducing flotation rate and the occurrence of a
in particle contact angle with particle size. For size fractions non-floating component, particularly for the very coarse size
greater than 150 µm, reduced liberation of the copper sulfide fraction (Figure 4 and 5). The occurrence of a non-floating
mineral in composite particles will reduce the overall contact component of copper sulfide mineral was confirmed in
angle of the copper sulfide bearing particle. Changes in recovery laboratory batch flotation tests on the rougher tailing during the
between the surveys as noted in Figure 4 are also apparent in the survey period, and is due to locking of the copper sulfide mineral
floatability class and rate constant values in Table 5. In within hydrophilic gangue mineral phases in both the coarse and
increasing the pulp density from Survey 1 to Survey 2, there is an very coarse size fractions. The significance of copper sulfide
increase in the rate of flotation of the coarse and very coarse size mineral locking is commented upon further below, but it does
fractions, and a decrease in the rate of flotation of the highlight the importance of precisely defining the size-by-
intermediate and fine size fractions. liberation characteristics of the feed, a point that is discussed
The recovery of the copper sulfide with respect to particle further below. This particular flotation plant shows an apparent
size exhibits typical behaviour, with its apparent flotation rate high rate of recovery of the copper sulfide mineral. To gain more
decreasing for particle sizes below and above approximately confidence in these results, the flotation rate was benchmarked
60 µm (Figure 5a), the particle size point for the fourth size against flotation rates measured in other large flotation plants,
fraction defined in Table 2. In practice, and in general, the treating mainly chalcopyrite and with a similar collector type
decreasing flotation rate of a hydrophobic value mineral with (Table 6). Table 6 shows that the flotation rate of the copper
decreasing particle size below 60 µm does not become a problem sulfide mineral encountered in this study is not unusually high.
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 Surface analysis to estimate copper sulfide
is usually sufficient residence time, under the prevailing mineral contact angle
hydrodynamic conditions, to recover hydrophobic particles in the
size range of +12 - 75 µm. There is greater than 95 per cent ToF-SIMS measurements suggested a very hydrophobic surface
recovery for this size range at six cells (19.3 minutes in Survey for fully liberated particles of the copper sulfide mineral with an
2), one half the installed flotation capacity (Figure 2). Even for adsorbed collector/SO3- ratio greater than 0.01, and therefore an
copper sulfide mineral below 12 µm, their reduced flotation rate inferred contact angle of 70° or greater (Duan, Fornasiero and
does not greatly impact upon their recovery at 12 cells (42.7 Ralston, 2003). The contact angle of the value mineral was taken
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 Number of cells in Volume per cell (m3) Total pulp residence
Rate constant (min-1) Recovery across cell group time in cell group
group (%) (min)
to be 78° (Table 3), a value characteristic of the particular for these particle size ranges (-75 µm). In terms of the physical
collector hydrocarbon chain length (R = 4) used in the plant meaning, it is reasonably suggested that, by the action of the
(Sutherland and Wark, 1955). For contact angles greater than impellor, all particles are directed through the impeller region
70°, both the attachment (Ea) and stability (Es) efficiencies are where particle collection by bubbles takes place at a rate
very close to unity for particles sizes less than 75 µm, indicating characteristic of this local energy dissipation. For the strongly
that the flotation rate will be limited by the collision rate for this hydrophobic (and liberated) copper sulfide mineral particles, less
particle size range, rather than particle hydrophobicity, for very than 75 µm, the particles apparently remain attached to the
high contact angles. The significance of this statement is bubbles (Es = 1) and are transported to the top of the flotation
discussed further below. Other studies have also shown that cell. It is important to note that the flotation rate and recovery of
chalcopyrite has a limiting contact angle close to 78° at high fine copper sulfide (-12 µm) is limited by collision efficiency,
adsorption densities of collector of the same chain length as used which is controlled through the cell hydrodynamics, rather than
in this present study (Walker, Walters and Richardson, 1986). surface hydrophobicity (Ea = Es = 1).
Attention will now turn to copper sulfide mineral contained in
Model fit to Survey 2 particles greater than 75 µm. In the case of copper sulfide
mineral-bearing particles in the coarse (+150 - 300 µm) and very
Attention will now turn to the fitting of the fundamental model, coarse (+300 µm) size fractions, assuming a single class of
outlined in the introduction, to the flotation data of Survey 2. mineral phase (ie a single contact angle of 78°) did not allow
Specifically, the collision efficiency, Ec, is calculated from the close agreement of the model with the experimental data
generalised Sutherland Equation 20 in Pyke, Fornasiero and (Figure 5a). Specifically, the model significantly overestimated
Ralston (2003). The attachment efficiency, Ea, is calculated using the recovery of copper sulfide mineral in these particular size
Equations 22 - 26 in Pyke, Fornasiero and Ralston (2003). The fractions. To this point, the model did not allow the occurrence of
particle-bubble stability efficiency, Es, is calculated using a non-floating component of copper sulfide mineral. It is
Equations 28 - 34 in Pyke, Fornasiero and Ralston (2003). The reasonably assumed that this is related to the occurrence of
bubble-particle collision frequency is calculated using Equation copper sulfide mineral in composite particles in these coarser
26 in Bloom and Heindel (2003), which takes account of body size fractions (Table 2). An approach to adequately model this
forces and where the bubble and particle velocity distribution
flotation behaviour was to assume a distribution of copper sulfide
variances are calculated using Equations 17a and 17b in Bloom
mineral liberation values in each size fraction of the rougher
and Heindel (2003).
flotation feed. Three different liberation classes were chosen:
The calculated recovery values of the fundamental flotation Class 1 and 2 were both particles containing partially or fully
model are shown as continuous lines in Figure 5, assuming that liberated copper sulfide mineral, and Class 3 were hydrophilic
the copper sulfide bearing particles have a contact angle of 78° particles containing completely unliberated copper sulfide
across all particle size fractions. It is clear that for particles mineral (contact angle equals zero) for which the attachment
greater than 75 µm the model overestimates their recovery. This
efficiency (Ea) is zero (Table 7). Class 3 gives rise to a
is certainly related to the reduced level of liberation of copper
non-floating component of copper sulfide. The contact angles of
sulfide mineral in these coarser size fractions, as previously
the composite particles were calculated using the Cassie equation
discussed. Inferred liberation levels are discussed further below
which adequately model the flotation data for particle sizes (Cassie and Baxter, 1944), assuming the contact angle of the
greater than 75 µm. copper sulfide and gangue mineral phases were 78° and 0°,
respectively, in all cases (Table 7).
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 θ c = cos−1 (f1 cosθ 1 = f2 cosθ 2 + ....+ fn cos θ n ) (10)
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 where:
were much less than the experimental values at the same cell f is the area fraction of copper sulfide mineral in the
number. As previously noted, the copper sulfide minerals over composite particle
this size range are strongly hydrophobic, and calculations of
attachment (Ea) and stability efficiencies show that both these c is composite
parameters are very close to unity over this particle size range
1,2,...n represent the mineral components of the particle
(-75 µm). Therefore, the discrepancy in recovery between the
model calculation using these parameter values, and experiment The percentage of partially liberated and completely
could only be accounted by the collision frequency. Key unliberated copper sulfide mineral increases with increasing
parameters that determine the collision frequency are energy particle size above 75 µm (Table 7). For these coarse unliberated
dissipation and bubble rise velocity. It was only when the particles, detachment from bubbles (Es decreases) considerably
calculated local energy dissipation (11 W/kg) and bubble rise reduces their flotation rate due to both their large particle size
velocities (1 cm/s) were used that a close agreement was found and low overall hydrophobicity. Their flotation rate and recovery
is lower due to reduced bubble-particle aggregate stability across in the feed. There is close agreement between the fundamental
the flotation cell volume. The presence of composite particles model and Survey 1 on an unsized basis for most cells (Figure 2,
was important only for the coarse and very coarse size fractions Survey 1). A possible reason for the discrepancy in cells 2 and 3
(Table 7). Incorporating different levels of liberation allowed a may be related to the lower pulp density in Survey 1, decreasing
very close fit of the model to the experimental data (Figure 5b). the pulp viscosity, and increasing the bubble particle collision
The very close agreement between the experimental data and the frequency and rate constant for copper sulfide particles less than
model is also apparent on an unsized basis (Figure 2, Survey 2). 75 µm.
TABLE 7
Inferred liberation distribution of the copper sulfide mineral in the rougher feed (Survey 2).
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).
100
80
Base Case Energy Dissipation = 11 W/kg
60 Case 1 - Energy Dissipation = 13.8 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
20
(A) Copper Sulfide Contact Angle
0
80
For Cases 1 to 6, the relevant hydrodynamic parameter was small extent (Figure 6a). On an unsized basis, decreasing energy
adjusted in each cell down the bank. For Cases 7 and 8, the same dissipation in Case 2 increases copper sulfide mineral recovery
hydrodynamic values were used in the additional cells as in the by three per cent (Table 8). It is noted that decreasing the energy
final cells of the existing rougher bank. It is noted that changes in dissipation will have a positive effect on coarse copper sulfide
recovery as discussed below are based on the model calculations. mineral recovery by increasing the stability efficiency, while fine
For the PTFI system, it is most beneficial to affect positive copper sulfide mineral recovery is reduced due to the decrease in
changes in coarse copper sulfide mineral recovery, over fine collision frequency, these particles’ flotation rate being principally
value mineral recovery, due to the preponderance of coarse controlled by the collision efficiency and bubble-particle collision
copper sulfide bearing particles in the feed (Table 2, Figure 3). frequency. Similar observations are made in relation to the effect
As an example, decreasing the energy dissipation in Case 2 of changes in bubble size, where an increase in bubble size in
increases coarse and very coarse sulfide mineral recovery, while Case 3 increased coarse copper sulfide mineral recovery at the
only reducing ultra-fine copper sulfide mineral recovery to a expense of ultra-fine copper sulfide mineral recovery (Figure 6b).
100
Recovery (%) 80
Base Case
60 o
Case 9 - CA = 85 , Energy Dissipation = 8.8 W/kg
Bubble Diameter = 1.63 mm
o
40 Case 10 - CA = 85 , Energy Dissipation = 8.8 W/kg
Bubble Diameter = 1.0 mm
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 basis (Table 8). Increasing the contact angle of the copper sulfide
efficiency of coarse particle size fractions, but decreases the increases the contact angle on the composite particles overall and
collision efficiency of fine particle size fractions. reduces their tendency to detach from bubbles. Under these
Increasing the collector chain length also benefits coarse and modified conditions it is possible to reduce the bubble size,
very coarse value mineral recovery (Figure 7a). Increasing the which otherwise would have decreased the bubble-particle
hydrophobicity of the copper sulfide mineral surface will not stability efficiency.
increase the recovery of copper sulfide mineral contained within
completely unliberated classes (Table 7). However, it will DISCUSSION
increase the recovery of copper sulfide mineral in partially
liberated composites. Increasing the copper sulfide mineral The use of a fundamentally derived flotation model allows a
surface contact angle to 85° in Case 5, increases the copper comprehensive view of the rougher bank performance on a
sulfide mineral recovery by 3.5 per cent on an unsized basis size-by-size basis. The model lends itself for benchmarking plant
(Table 8), an increase of similar magnitude to a decrease in flotation data. For example, reasons for the low flotation rate of
energy dissipation in Case 2, or an increase in bubble size in particles may be discerned (eg particle surface hydrophobicity
Case 3, alone. It is noted that the change in contact angle of the versus cell hydrodynamics). The relative contributions to a low
copper sulfide mineral surface from 78° to 85° is characteristic of flotation rate may be quantified, and the effect of common
an increase in collector hydrocarbon chain length (to R = 5) variables investigated to determine the most profitable avenue to
(Sutherland and Wark, 1955). Increasing the contact angle on the increase flotation rate. In the case of PTFI, the principal form of
exposed copper sulfide mineral increases the contact angle on the copper loss is as copper sulfide mineral partially locked in
composite particle overall, and increases the stability efficiency composite particles contained within coarse and very coarse size
of coarse composite particles. Increasing the contact angle on fractions of the rougher tailing. These particles have apparently
fine copper sulfide has little effect on their recovery, as the only low exposure of the copper sulfide mineral at the particle
attachment efficiency is already at its maximum value. It is also surface and hence the hydrophobicity of the overall composite
worthwhile pointing out that, for the PTFI system, simply particle is low. For example, the contact angle for coarse (+150 -
increasing the available residence time by increasing the number 300 µm) copper sulfide bearing composites is only 30°,
of cells, does not appreciably increase copper recovery on either compared with the contact angle of the exposed copper sulfide
a sized (Figure 7b) or an unsized basis (Table 8). Changes in the mineral (78°) (Table 7). The low contact angle adversely affects
operating conditions of the cells or the hydrophobicity of the their flotation rate and recovery due to detachment of the particle
copper sulfide mineral need to be pursued to increase copper from the bubble in the turbulent environment, giving rise to low
recovery. stability efficiency.
Increasing the contact angle of the copper sulfide mineral The simulations suggest that it is desirable to firstly target
allows also the bubble size to be reduced in order to pursue increases in copper sulfide mineral hydrophobicity by increasing
increased ultra-fine copper sulfide mineral recovery. The most the collector chain length, and following this step, increase fine
significant, single, process variable to increase ultra-fine copper copper sulfide mineral recovery by reducing the bubble diameter.
sulfide mineral recovery is to reduce the bubble size in Case 4, as The first step will increase coarse composite recovery, which is
previously noted (Figure 6b). Neither increasing the energy of prime importance to overall copper recovery. The effect of
dissipation nor increasing the copper sulfide mineral contact copper sulfide mineral exposure (ie mineral liberation class) in
angle alone is as effective. However, decreasing the bubble size the composite particles on their recovery is shown in Figure 9.
alone decreases coarse and very coarse copper sulfide mineral In the case of the coarse size fraction, there is virtually zero
recovery due to reduced stability efficiency (Figure 6b). By recovery of composite particles with a copper sulfide mineral
increasing the contact angle of the copper sulfide mineral, it exposure less than ~15 per cent, for 78° contact angle of the
becomes possible to target decreases in bubble size to affect copper sulfide mineral (Figure 9). The critical surface exposure
increases in ultra-fine copper sulfide mineral recovery, without increases to ~30 per cent for the very coarse size fraction, also at
adversely affecting coarse copper sulfide mineral recovery (Case 78° contact angle of the copper sulfide mineral (Figure 9).
10, Figure 8). Case 10 gave the highest increase in copper sulfide Increasing the copper sulfide mineral contact angle from 78° to
mineral recovery of approximately four per cent on an unsized 85°, allows composite particles, which contain only ~10 per cent
and ~20 per cent copper sulfide mineral exposure to be recovered mineral contact angle to 85°, decreases in bubble diameter can be
in the coarse and very coarse size fractions, respectively. This is sought to increase ultra-fine copper recovery without adversely
because there is a small increase in the contact angle of the affecting coarse composite recovery (Figure 10). Increasing the
composite particle overall (Figure 9), and an increase in their contact angle of the copper sulfide mineral has little affect on
stability efficiency. ultra-fine copper recovery, which is principally controlled by cell
Reducing the bubble diameter will increase fine copper sulfide hydrodynamics through the collision efficiency and collision
recovery by increasing the collision efficiency, and is the most frequency terms (Figure 10).
effective method in doing so as previously noted (Figure 6b). It
should also be stressed that adding more cells using the existing CONCLUSIONS
cell hydrodynamic conditions is not an effective means to
increase overall recovery. One method to decrease the bubble To increase overall copper recovery in roughing at PTFI, at the
diameter is to increase the frother concentration to the critical current flotation feed size, it is necessary to address copper
coalescence concentration (Grau, Laskowksi and Heiskanen, recovery in coarse, low-grade composite particles. The most
2005). The effect of bubble diameter on model calculations of effective means to increase coarse composite particle recovery is
the recovery of ultra-fine (-6.8 µm) and coarse (+150 - 300 µm) to increase the contact angle of the exposed copper sulfide
composite (0_30 per cent liberation class) copper sulfide bearing mineral in the composite. Surface analysis suggests that the
particles is shown in Figure 10. Decreasing the bubble diameter contact angle of the copper sulfide mineral is at its maximum
adversely affects coarse composite recovery at a copper sulfide value characteristic of the alkyl chain length of the collector
mineral contact angle of 78° (Figure 10), due to reduced stability used. Therefore, it is only possible to increase the contact angle
efficiency, while it increases ultra-fine copper recovery due to by increasing the chain length of the collector, increasing the
increased collision efficiency. By increasing the copper sulfide contact angle of the composite particle overall. A further benefit
Recovery (%) / Contact Angle (o)
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
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 (%)
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.
of increasing the contact angle on the coarse composite particles Grau, R A, Laskowski, J S and Heiskanen, K, 2005. Effect of frothers on
is that hydrodynamic conditions can be pursued which may bubble size, International Journal of Mineral Processing, 76(4):
225-233.
target increased fine copper recovery, but which would normally
compromise coarse copper recovery. The most effective means to Koh, P T L, 1984. Compartmental modelling of stirred tank for
flocculation requiring a minimum critical shear rate, Chem Eng Sci,
increase fine copper recovery is to decrease the bubble size. By 39:1759-1764.
increasing the contact angle on the copper sulfide mineral alone, Lapple, C E and Shepherd, C B, 1940. Calculation of particle trajectories,
the overall copper recovery can be increased from 89.9 per cent Ind Eng Chem, 32:605-617.
to 93.4 per cent at the coarse grind (20 per cent +220 µm). By Mular, M and Mosher, J, 2006. A pre-production review of PT Freeport
increasing both the contact angle on the copper sulfide mineral Indonesia’s high pressure grinding roll project, in Proceedings
and reducing the bubble size the overall copper recovery can be SAG2006 (Canadian Institute of Mining and Metallurgy: Montreal).
increased from 89.9 per cent to 94.0 per cent at the coarse grind. 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
ACKNOWLEDGEMENTS Institute of Mining and Metallurgy: Melbourne).
The authors gratefully acknowledge the sponsors of the P260D Pyke, B, Fornasiero, D and Ralston, J, 2003. Bubble particle
(Mineral Flotation) project and specifically PT Freeport heterocoagulation under turbulent conditions, Journal of Colloid
Interface Science, 265:141-151.
Indonesia for permission to publish this paper. The authors
would also like to thank Ari Partanen and Arif Priyambodo for Sam, A, Gomez, C O and Finch, J A, 1996. Axial velocity profiles of
single bubbles in water/frother solutions, International Journal of
their contribution and support. The authors would also like to Mineral Processing, 47(3-4):177-196.
thank Edy Sanwani and the P9M project for the gas dispersion Sanwani, E, 2003. Gas dispersion and hydrodynamic characteristics of
data, William Skinner for the ToF-SIMS data, and Sergio Vianna, selected Wemco 85 m3 flotation cells at PT Freeport Indonesia Mine,
George Tsatouhas and Tsvetomir Georgiev for their assistance on AMIRA P9M project report, JKMRC, Brisbane.
site. 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.
REFERENCES
Schubert, H, 1989. The role of turbulence in mineral processing unit
Abrahamson, J, 1975. Collision rates of small particles in vigorously operations, in Challenges in Mineral Processing (eds: K Sastry and
turbulent fluid, Chemical Engineering Science, 30:1371-1379. M C Fuerstenau), pp 272-289 (Society for Mining, Metallurgy and
Bloom, F and Heindel, T J, 2002. On the structure of collision and Exploration: Littleton).
detachment frequencies in flotation models, Chemical Engineering Schulze, H J, 1977. New theoretical and experimental investigations on
Science, 57:2467-2473. stability of bubble/particle aggregates in flotation: A theory on the
Camp, T R and Stein, P C, 1943. Velocity gradients and internal work in upper particle size of floatability, International Journal of Mineral
fluid motion, J Boston Soc Civ Eng, 30:219-237. Processing, 4:241-259.
Cassie, A B D and Baxter, S, 1944. Wettability of porous surfaces, Trans Schulze, H J, 1993. Coagulation and Flocculation (ed: B Dobias),
Faraday Soc, 40:547-551. Chapter 7 (Dekker: New York).
Dai, Z, Fornasiero, D and Ralston, J, 1999. Particle-bubble attachment in Sutherland, K L and Wark, I W, 1955. Principles of Flotation (The
mineral flotation, Journal of Colloid and Interface Science, 217(1): Australasian Institute of Mining and Metallurgy: Melbourne).
70-76. Tucker, J P, Deglon, D A, Franzidis, J-P, Harris, M C and O’Connor, C T,
Dai, Z, Fornasiero, D and Ralston, J, 2000. Particle-bubble collision 1994. An evaluation of a direct method of bubble size distribution
models – A review, Advances in Colloid and Interface Science, measurement in a laboratory batch flotation cell, Minerals
85(2-3):231-256. Engineering, 17(5-6):667-680.
Dobby, G S and Finch, J A, 1987. Particle size dependence in flotation Vera, M A, Franzidis, J-P and Manlapig, E V, 1999. Simultaneous
derived from a fundamental model of the capture process, determination of collection zone rate constant and froth zone
International Journal of Mineral Processing, 21(3-4):241-260. recovery in a mechanical flotation environment, Minerals Engineering,
Duan, J, Fornasiero, D and Ralston, J, 2003. Calculation of the flotation 12:1163.
rate constant of chalcopyrite particles in an ore, International Walker, G W, Walters, C P and Richardson, P E, 1986. Hydrophobic
Journal of Mineral Processing, 72(1-4):227-237. effects of sulfur and xanthate on metal and mineral surfaces,
Grano, S, Weedon, D and Akroyd, T, 2005. Freeport plant surveys – C3 International Journal of Mineral Processing, 18:119-137.
concentrator sized data, IWRI confidential report to PT Freeport
Indonesia, March.
ABSTRACT It was envisaged that, at the time of submitting this paper, the
OCS© rule base and models would be fully commissioned, and
As part of a broad program to optimise the performance of their grinding
circuit, BHP Billiton Cannington have installed an OCS© Expert system that a substantial amount of performance data would be available
together with three VisioRock™ online particle size analysis cameras. to demonstrate the effectiveness of the system. Limitations in the
The main purpose of the system is to stabilise grind size and feed to the amount of ore able to be supplied to the crushing and grinding
flotation circuit. circuits by the mine, and delays in the commissioning of the
This particular application of OCS© is the first installation to incorporate particle size indicator (PSI) have unfortunately delayed the final
Metso Minerals Process Technology-Asia Pacific’s (MMPT-AP) Dynamic deployment of the control strategy. Nonetheless, significant
SAG/AG Mill Model, which is able to predict numerous mill parameters performance data from the initial stages of the strategy
that were previously unable to be measured with the existing plant development have been collected and these are presented in this
instrumentation. Integrated into the logic of the control algorithm are paper. Descriptions of the Dynamic Model and how it will be
online vision based, particle size measurements of the AG mill feed and used in the final control strategy are also detailed, along with
recycle crusher product. comparisons of model performance with logged process data.
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. THE CANNINGTON CIRCUIT
The Cannington silver, lead and zinc deposit, located in
INTRODUCTION north-west Queensland, was discovered in June 1990. At the
BHP Billiton’s Cannington operation is the largest and lowest present time, the Cannington ore resource is 42 million tonnes
cost silver and lead producer in the world. A great part of its with 551 ppm of silver, 12.1 per cent lead and 4.7 per cent zinc.
success has been due to the continuous innovation and Ore is extracted from underground operations via both open
optimisation undertaken in the nine years since commissioning. stoping and bench mining methods, with access via a decline and
This has seen the initial annual throughput to the concentrator production from a hoisting shaft. The Cannington processing
increase from 1.5 Mtpa in 1997 to a record 3.1 Mtpa in 2005. plant was commissioned in October 1997 with an initial
As part of ongoing optimisation projects at Cannington, production rate of 1.5 Mtpa (180 t/h).
personnel recognised that Advanced Expert Control of the The 2002 expansion study investigated the addition of a pebble
crushing and grinding circuit would allow further performance crusher to the grinding circuit as well as the introduction of
increases of the crushing and grinding circuits. These increases pebble ports to the AG mill. By utilising the pebble crusher for
would be unable to be gained by running the circuits under partial or full pebble crushing, circuit throughput was able to be
manual control. To this end, Metso Minerals Process Technology increased to 320 t/h. The pebble crusher also allowed another
– Asia Pacific (MMPT-AP) in Brisbane and Metso Minerals Cisa control option for stabilising the circuit throughput stable due to
(MMC) from Orleans, France, were contracted to design and variations in feed ore hardness and size. A permanent Metso
commission an expert control strategy based on their OCS©, HP300 pebble crusher was installed after the study. With
(Optimising Control System), and VisioRock™, online image increased pebble crusher capacity and pebble ports in the
analysis systems. The control strategy utilises online particle size discharge grates of the AG mill, throughputs in excess of 350 t/h
measurements from the VisioRock™ system, expert rules based were able to be achieved. However, due to the subsequent coarse
on both fuzzy and crisp logic as well as model based control flotation feed, these throughputs could not be sustained without
using the Valery Dynamic SAG/AG mill model. This has been recovery losses. In order to reduce the flotation circuit feed size
ported as a Dynamic Linked Library (DLL) and is being used by and circuit product variations, a grinding circuit expansion
the soft sensor module of OCS©. The principal goal of the
project was then undertaken.
control strategy is to maximise and stabilise mill throughput
while maintaining the flotation feed grind size to a value such During the design and engineering stage of the expansion
that recovery is not negatively affected. 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
1. Manager – Development and Process Engineering, Metso Minerals in closed circuit with primary cyclones followed by a secondary
Process Technology Asia-Pacific (MMPT-AP), Unit 1, 8 - 10 stage of grinding consisting of one Metso VTM 1500 WB
Chapman Place, Eagle Farm Qld 4009. (1.1 MW) Vertimill® in closed circuit with secondary cyclones.
Email: David.LaRosa@metso.com
Grinding circuit simulation results (Jankovic, Valery and Shi,
2. Process Control Metallurgist, Cannington Mine, Base Metals, BHP 2001; Jankovic and David, 2002) indicated that sustained
Billiton, PO Box 5847, Townsville Qld 4810. throughputs of 350 t/h with a final P80 of 95 microns could be
achieved with the introduction of both pebble crushing and the
3. MAusIMM, Principal Metallurgist, Cannington Mine, Base Metals,
BHP Billiton, PO Box 5847, Townsville Qld 4810. VTM 1500 WB Vertimill®.
with the metallurgist in mind. This enables simple configuration • A Neural Network module.
of the control strategy, reduces the software footprint and
subsequent hardware requirements, and increases the speed of • A Vision Module, which can act in the following different
code execution. modes:
Typically OCS© supplements conventional distributed control • VisioFroth for froth flotation plants,
system (DCS) or programmable logic controller (PLC) controls. • VisioRock for SAG and AG milling or crushing and
OCS© reads process information from the PLC or DCS network, screening plants,
and uses various techniques to determine new set points,
generally several times per minute. The new set points are then • VisioPellet for pelletising plants, and
automatically applied at the PLC or DCS level, without • VisioTruck to measure run-of-mine (ROM) fragmentation
necessitating any action from the plant operators. The set points at the primary crusher.
calculated by OCS© continuously pursue strategic objectives. For
example, they could maintain a required product quality while • A Tag Statistics module, which generates online statistical
targeting a technical and economical objective. This typically information about the process (averages, histograms, etc),
results in an increase in capacity, product quality and value, as and may be used for evaluation of plant performance.
well as a possible decrease in production costs. OCS© communicates with the process it is controlling via
The OCS© software structure makes the application both OPC, ODBC, DDE or DAO communications and there are many
powerful and easily maintained. The implementation third party drivers on the market that allow it to communicate
methodology and development of the control strategy always with PLCs via ModBus, ModBus Plus and other popular
involves metallurgists, operators and other plant personnel, and protocols.
therefore promotes ownership, results in realistic control
strategies and allows excellent acceptance of the system. 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
OCS© overview allow a complete picture of the process status. Set points written
OCS© software hosts, in a single software application, a number back to the PLC network will include the AG mill fresh feed and
of modules necessary for optimally controlling a mineral water, pebble crusher feed, primary and secondary cyclone
processing plant. An OCS© application may involve some or all pressure and Vertimill® feed water. This may be extended as the
of the following modules: control strategy matures.
• The Fuzzy Expert module: a real time expert system, with an
inference engine, built-in crisp and fuzzy logic reasoning and VisioRock™ overview
a knowledge base, this is the core of any OCS© application.
Online image analysis has been routinely used in mineral
• A Process Display module: easily configured in order to processing plants to monitor the particle size of SAG mill feed
provide one or several user interface(s), such as flow and crusher feed for a number of years (La Rosa et al, 2001).
diagrams, animated process mimics and data entry forms.
It was a natural progression to include a particle size analysis
The user may graphically develop new interfaces and
manipulate them using simple Visual Basic code. module with OCS© and for the application of ore particles on
conveyor belts, generally feeding grinding mills or as crusher
• A Soft Sensor module with one or more adaptive predictive product, this module is called VisioRock™ (Guyot et al, 2004).
models. This is a robust and mature technology and Metso Minerals
• An Optimiser module, with a powerful constrained Process Technology groups have installed over 40 cameras at
sequential quadratic programming (SQP) algorithm. more than ten sites globally.
VisioRock™ hardware consists of high quality analogue or to wear in the machine and change as large adjustments were
network cameras focused onto an appropriately lit area of the made to the machine during its re-commissioning. This is
conveyor belt. Areas of direct sunlight must be excluded from valuable objective data about the crusher performance.
the field of view of the camera as this can cause errors in the
software delineations.
THE USE OF DYNAMIC MODELS IN OCS©
The computation steps to convert a single colour image to a
user-definable size distribution are complex and include filtering,
PROCESS OPTIMISATION
contrast enhancement, segmentation, background rejection, fines OCS© soft sensors typically make use of so-called deep
identification, calculation of the 2D particle size distribution knowledge about the physics or the chemistry of the process. A
followed by a 2D to 3D particle conversion. Additional analyses soft sensor uses such knowledge to infer numerous
such as colour, texture and shape per size class are also non-measurable process variables from few measurements. The
calculated concurrently. The VisioRock™ module can process core of an OCS© soft sensor within the modelling module is a
over two high resolution images per second or lower resolution dynamic model of the process. The model of the process is
images at between 15 and 30 frames per second. Figure 2 shows embedded in an adaptive tuning algorithm called an Extended
a screen shot of the VisioRock™ screen processing the Kalman Filter (EKF) that dynamically estimates values of
Cannington pebble crusher product. The raw video image with process parameters. The model, coupled with its adaptive tuning
delineation contours are shown on the left while the dynamic algorithm is a ‘learning’ model.
particle size distribution and histogram are shown to the right. In some cases, the dynamic model may be linked to an
At Cannington there are three VisioRock™ cameras, one on optimiser specific to the application. At any time, the state of the
the underground jaw crusher product, another on the AG mill process is known from the ‘learning’ model and the optimiser
feed conveyor before the crushed pebbles join the feed and can use the predictive capabilities of the model to infer optimised
another on the pebble crusher product. The jaw and pebble set points to achieve the best possible performance in the near
crusher product cameras are used to monitor product ‘quality’. future.
Alarms will be generated when out of specification product is The combination of soft sensors, modelling modules and the
detected, as these can cause disturbances in the grinding circuit expert system in OCS© makes it possible to benefit from the
due to coarser than expected product being processed. The AG ability of the modelling modules to generate valuable
mill feed particle size distribution will be used as an input to the information about the state of the process and to predict
dynamic model. optimised set points. Another major advantage of this approach
As an example of how crusher product size can change over is that the expert system can take into account numerous
time, Figure 3 shows the P80 for the pebble crusher after a constraints and facts that are not handled by the mathematics of
rebuild. The P80 values can be seen to drift upwards in response the models to ensure safe optimised control.
The OCS© optimiser continuously looks for the best path to the process. The model makes a prediction of the output values
peak performance using the mathematical models of the soft using the input values and the model equations and parameters
sensor module. It takes objectives and constraints into account, that describe the physical process.
and computes theoretical optimum model-based set points using The output is the point in OCS© where actual measurements
an SQP (sequential quadratic programming) algorithm. (called observations) are compared to predicted measurements.
Conceptually, these set points can be applied directly, although in The Extended Kalman Filter looks at the difference and adjusts
most cases they are delivered to the fuzzy expert module, which the parameters depending on the relative confidence of the model
makes realistic use of the model-based outputs. The structure of and the observation. New estimates of the model states and/or
the OCS© soft sensor module is shown in Figure 4. parameters are then calculated and produce updated estimates of
The dynamic model of a process is built up from separate the outputs. If we are more certain that the observations are
models of the components or objects in the process. For example accurate than the model, the model parameters are allowed to
a grinding circuit model may consist of models for feeders, mills, vary more to make the estimates fit the observations. Conversely,
pumps and cyclones. A process model describes the relationship if we have more confidence in the model than the observations,
between inputs and outputs of a piece of equipment or section of the parameters will be changed less. Setting the relative
©
FIG 4 - OCS soft sensor structure (David La Rosa).
confidence of the model and observations is a critical step in start-up the state of the initial charge distribution in the mill is
tuning the model. When OCS© runs, it combines all the estimated and quickly converges to arrive at a best fit for the
component models into one EKF, which constitutes a model flow mass balance around the mill.
sheet of the entire process. Once the mass balance is achieved, gross power and its
components; kinetic, potential, cone and no-load powers are
The Valery Dynamic AG/SAG mill model calculated. The Valery Dynamic Model uses Morrell’s power
calculations (Morrell, 1996) that estimate the charge profile,
Central to the final control strategy at Cannington is the Valery shoulder and toe angles of the charge, charge radius and slurry
Dynamic SAG/AG mill model (Valery, 1997; Valery and Morell, pool formation. This information can be used for visual sanity
1995; Valery and Morrell, 1997; Morrell et al, 2001). checks in the operator display. It also uses Powell’s outermost
The model incorporates the following major features: charge trajectory (Powell) calculation to provide valuable
• a method to calculate the energy absorbed by rock particles feedback on impact of grinding media for the purposes of liner
in the mill, protection. The slurry flow out of the mill is estimated with an
empirical formulation based on the mass balance around the mill
• more realistic equations for mass transport, and the grate characteristics.
• online prediction of power draw and outermost charge Breakage and discharge mechanisms are the two main
trajectory, components of the dynamic model. Breakage mechanisms are
• a relationship between mill charge composition and breakage described by the breakage rate of each size fraction and its
rates based on projected surface areas of the media and target appearance function. For SAG mills, approximations of ball
particles, and charge at any time are required to arrive at best estimates of
breakage rates. Impact energy levels of every size fraction are a
• the effect of slurry content and hold-up on mill behaviour. function of drop height in the mill and this is calculated from the
This is a population balance model extended with empirical estimated charge profile. Mill speed is also an important
formulations to calculate power and slurry flow. The model also component of breakage rates.
estimates the mill charge distribution at every OCS© cycle, in The discharge mechanism describes the flow of water and size
this case every 30 seconds. The mass balance of each size fractions finer than the grate size out of the mill and relies on the
fraction is calculated from the fresh feed, existing charge in that maximum discharge rate of the finest size fraction. The
size fraction, material exchange between size fractions as a result maximum discharge rate is calculated from the slurry flow
of breakage (through abrasion and impact breakage events) and estimation.
product discharge. The model requires mass inputs at each cycle, Actual and calculated power are used in the Extended Kalman
eg new feed rate (from weightometer measurements), feed size Filter to correct breakage rates and load size distributions. Figure 5
distribution (from VisioRock™) and water flow rates. At model shows a flow diagram of the Valery Dynamic Model.
FIG 5 - Outline of the dynamic AG/SAG mill model algorithm (David La Rosa).
Soft sensor implementation in OCS© primary grinding circuit allows more feed to be processed by the
© mill without approaching overload conditions. In the initial
The OCS HMI (human machine interface) of the Cannington stages of the project, focus was placed on controlling the pebble
grinding circuit includes the two grinding stages described crusher to maximise AG mill feed rate. Figure 8 shows a
earlier and is shown in Figure 6. For convenience, the primary
schematic of the control logic.
and secondary grinding stage circuits are modelled in two
separate soft sensor modules. The soft sensor modules are used In this strategy the operator enters a bearing pressure set point.
to predict process state information that are not measured online In the short term, this is proportional to the load in the mill.
and which are used in the OCS© expert system advanced control OCS© then adjusts the feed to the pebble crusher using fuzzy
logic to achieve the final control objectives. These include logic sets to determine how far away and how fast to change the
product size distributions, ore hardness parameters, mill per cent pebble crusher amps set point to meet this criteria. In this way
solids, etc. the pebble crusher is protected from overload situations by
The process model configuration is done entirely through a controlling the amps set point instead of the feed rate.
graphical interface as shown in Figure 7. The user is able to The simplified OCS© rules for adjusting the pebble crusher
select component models from a library of population balance amps setpoint and feed rate to the mill, are:
models developed in VB.net by MMC and then, by connecting • if bearing pressure is high THEN increase pebble crusher
the various inputs and outputs of the component models, builds amps setpoint,
the process flow sheet.
• if bearing pressure is low THEN decrease pebble crusher
OCS© also includes tools to develop custom models in VB.net.
amps setpoint,
For instance, the AG mill model used at Cannington is a model
customised by MMC and based on Valery’s dynamic model that • if bearing pressure is at setpoint THEN no change to pebble
was ported from the original Fortran code by MMPT-AP. crusher amps setpoint,
Each large coloured block shown in Figure 7 corresponds to a • if bearing pressure is very high THEN decrease fresh feed
component model. The component models used for the rate to the AG mill, and
Cannington grinding circuit are Feeder, AG mill, Vertimill®,
• if bearing pressure is very low THEN increase fresh feed rate
Screen, Sump with pump, Crusher and Cyclone.
to the AG mill.
The sensors, or measurements, correspond to the white blocks
in Figure 7. There are two types of measured variables, measured The pebble crusher control rules respond as follows:
inputs such as: • if pebble crusher amps are high THEN decrease pebble
• feed belt scale; crusher feed setpoint,
• mill inlet water; • if pebble crusher amps are low THEN increase pebble
crusher feed setpoint, and
• sump dilution water;
• if pebble crusher amps are at setpoint THEN no change to
• number of cyclones open; pebble crusher feed setpoint.
• VisioRock™ D20, D50, D80, D100 for the fresh feed; The definitions of high, low, very high, etc are all fuzzy and
• crusher feed; the rules also take into account the rate of change of the process
parameter they are describing. The magnitude of subsequent
and measured outputs (or observations) such as:
setpoint changes will vary according to these criteria. Decrease
• VisioRock™ D20, D50, D80, D100 for the crushed pebbles; and increase changes are similarly described.
• mill power; This strategy, while quite simple, and requiring less than 20
rules, allowed the AG mill to be taken to areas of operation that
• mill bearing pressure;
were not considered stable under manual control. Table 1 details
• mill recycled pebbles; and the statistics of mill throughput before and after OCS© control of
• cyclone feed flows, densities and pressures. the pebble crusher.
The sensors are linked to model inputs (measured inputs) or to
model outputs (measured outputs). The other links in the TABLE 1
graphical model configuration correspond to the ore streams OCS© pebble crusher control results.
between component models. For instance, the stream definition
for the AG mill circuit includes 30 size classes, the energy input, Average Standard Sample period
water and total solids flows. throughput deviation (days)
(t/h) (t/h)
Once the process model is configured, the EKF configuration
and tuning phases start. These two phases are under development Before OCS© 430 63 30
at the present time for the Cannington project. OCS© provides After OCS© 463 38 40
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 It was shown that having expert system control on the pebble
consists of selecting the model parameters that will be crushing circuit significantly decreased the variability in mill
continuously adjusted online while the tuning consists of throughput. What was not expected, however, was the magnitude
defining the range and errors (weights or relative confidence) for of the throughput increase that was achieved with this strategy
each adjusted parameter and measured output (observation). (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
INITIAL PEBBLE CRUSHER CONTROL also reduced while OCS© was in control. It is expected that the
final grind size was coarser under these conditions; however, no
STRATEGY
data was collected to quantify this. Even if conservative
The performance of the Cannington AG mill is extremely assumptions are made about recovery losses due to an increased
sensitive to the amount of scats crushed in the pebble crushing final grind size, this simple strategy amounted to a significant
circuit. The removal of these critically sized particles from the positive benefit to the operation.
ONGOING DEVELOPMENT OF THE FINAL Such simulation results have given personnel involved in the
CONTROL STRATEGY project confidence that the model and optimiser can be used to
further enhance the control strategies at Cannington.
As mentioned in the introduction, it was envisaged that the full The new strategy will add control of the primary and
control strategy would be commissioned by the time this paper secondary cyclone pressure setpoints based on sump level,
was written. Ore limitations from the mine have meant that there modelled per cent solids and pump speed as well as water
is little pressure on the crushing and grinding circuits for addition to the AG mill, Vertimill® and sumps. Most importantly,
production and it would be difficult to rigorously test the OCS© the balance between throughput and grind size will be
system over a wide range of operating conditions. Additionally, continuously monitored and adjusted when necessary. This work
as one of the KPIs of the control system is the final grind size, is ongoing and the results will be reported when available.
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
CONCLUSIONS
have spent time since the last site visit replaying captured The installation of the VisioRock™ and OCS© expert control
historical data through the OCS© projects and tuning the models. systems at BHPB Cannington have allowed the operation to
For example, Figure 9 shows a comparison between measured achieve another step change in plant performance. Significant
bearing pressure and estimated mill load for the Cannington AG stable throughput increases have been proven to be achievable,
mill over a 24-hour period. even with a simple control strategy. AG mill overload events
FIG 9 - Comparisons between measured and estimated AG mill load variables (David La Rosa).
have been reduced, minimising rehandle costs associated with La Rosa, D, Girdner, K, Valery, W and Abramson, S, 2001. Recent
the cleaning up of scats dumped on these occasions. applications of the Split-Online Image Analysis System, in
Consequently, pebble crusher utilisation has also increased. Proceedings Southern Hemisphere Meeting on Mineral Technology
2001, 1:15-19.
The dynamic model will allow process variables to be Morrell, S, 1996. Power draw of wet tumbling mills and its relationship
estimated and used where previously not possible and it is to charge dynamics, IMM Trans, C38-C62.
believed that further performance gains will be realised by the Morrell, S and Valery Jnr, W, 2001. Influence of feed size on AG/SAG
operation once the final model-based strategy is implemented. mill performance, in Proceedings SAG2001 – SAG Mill Circuit,
International Conference on Autogenous and Semiautogenous
REFERENCES Grinding Technology, 1.
Morrell, S, Valery, W, Banini, G and Latchireddi, S, 2001. Developments
Clarke, G and La Rosa, D, 2006. Changes and upgrades at Cannington, in in AG/SAG mill modelling, in Proceedings SAG2001 – SAG
Proceedings IIR Crushing and Grinding Conference 2006 (Institute Mill Circuit, International Conference on Autogenous and
for International Research). Semiautogenous Grinding Technology, 1.
Guyot, O, Monredon, T, La Rosa, D and Broussaud, A, 2004. VisioRock, Powell, M S, 1991. The effect of liner design on the motion of the outer
an integrated vision technology for advanced control of comminution grinding elements in a rotary mill, International Journal of Mineral
circuits, in Proceedings IIR Crushing and Grinding Conference 2004 Processing, 31:163-193.
(Institute for International Research). Valery, W, 1997. A model for dynamic and steady-state simulation of
Jankovic, A and David, D, 2002. Further simulations for Cannington autogenous and semi-autogenous mills, PhD thesis, JKMRC,
grinding circuit expansion, JKTech report, Job number 02212. University of Queensland, Brisbane.
Jankovic, A and Man, Y, 2003. Technical support during commissioning Valery, W and Morrell, S, 1995. The development of a dynamic model for
and optimisation of the new Cannington comminution circuit, Metso autogenous and semi-autogenous grinding, Minerals Engineering,
Minerals Process Technology Asia-Pacific report, Job number 32-03. 8(11):1285-1297.
Jankovic, A and Valery, W, 2004. Optimisation of the new Cannington Valery, W and Morrell, S, 1997. A dynamic and steady-state model of
comminution circuit, Metso Minerals Process Technology autogenous and semi-autogenous mills, in Proceedings International
Asia-Pacific report, Job number 24-04. Conference Comminution 1998.
Jankovic, A, Valery, W and Clarke, G, 2006. Design and implementation Valery, W and Morrell, S, 2001. JKSagCharge – A tool to estimate
of an AVC grinding circuit at BHP Billiton Cannington, in AG/SAG mill charge volume and position online, in Proceedings
Proceedings SAG2006 – Circuit Design, International Conference on Comminution 2001.
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.
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
H
INTRODUCTION
Base metal flotation concentrates are primarily sourced from
ores containing metal sulfides. Some deposits contain oxide N
metal minerals and or partially oxidised sulfides. Metal - K
+
recoveries from ores containing supergene and oxide O
mineralisation are usually affected by losses of these minerals to
tails, due to the poor efficiency of conventional sulfidisation O
techniques and recovery with conventional sulfide collectors.
The availability of a flotation collector which floats tarnished FIG 1 - Typical AM28 hydroxamate structure.
sulfides and oxide metal minerals represents a considerable
advance in the upgrading of supergene and oxide base metal
deposits. At Radio Hill, AM28 is used in conjunction with traditional
sulfide collectors to allow the simultaneous flotation of oxide and
A review of the application of chelating reagents as flotation sulfide copper minerals.
collectors by Hanson and Fuerstenau (1991) reported on the
potential of hydroxamates for the upgrading of oxidised base AM28 replaces conventional sulfidisation for oxidised mineral
metal ores and minerals. To this end, Ausmelt Limited has flotation and is a non-hazardous product. Extensive toxicological
developed and patented a range of alkyl hydroxamates (Hughes investigations on AM28 and its predecessor AM2, have
2004, 2006) as flotation collector reagents for use with oxidised been carried out by NICNAS (National Industrial Chemicals
and supergene base metal ores and tailings. 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
1. Principal Geochemical Consultant, Ausmelt Limited, 12 Kitchen manufacture and distribute AM2 (NICNAS Code is NA/917)
Road, Dandenong Vic 3175. Email: Terry.Hughes@ausmelt.com.au and AM28. Both AM2 (approximately 50:50 C8:C10) and
2. Metallurgical Manager, Ausmelt Limited, 12 Kitchen Road, AM28 (approximately 95 per cent C8) are classified as a ‘non-
Dandenong Vic 3175. Email: Kenneth.Lee@ausmelt.com.au dangerous goods’.
The formation and structure of AM28 is discussed in detail
3. AAusIMM, Manager, Ausmelt Minmet Metallurgical Laboratory,
Unit 3, 8 Yandina Road, West Gosford NSW 2250. in Hughes (2006). Although fatty hydroxamic acids and
Email: Graeme.Sheldon@tpg.com.au hydroxamate reagents have been the subject of many research
and patent publications, little information is provided on the
4. Senior Metallurgist, Fox Radio Hill Pty Ltd, PO Box 1629, Karratha structure and mode of the mineral surface attachment of the
WA 6714. Email: Jerehmy.Bygrave@foxradiohill.com.au reagents. Great care is taken during the manufacture of AM28 to
5. MAusIMM, Production Manager, Fox Radio Hill Pty Ltd, PO Box minimise the content of free fatty acid (FFA), which will react
1629, Karratha WA 6714. Email: Laurie.Mann@foxradiohill.com.au with soluble calcium ions in the pulp to form a soap-based froth.
The successful implementation of the AM28 manufacturing intermittently since 1911, with the most recent operations
process depends on the molar proportions of the feedstock ceasing in 1976. Total production from this deposit was 12 000 t
reagents, the control of reaction temperature and the sequence of of direct shipping ore grading 27.4 per cent copper.
reaction additions. The final product contains 45 - 50 per cent The Whundo copper-zinc deposit is located at the contact
alkyl hydroxamate, which is a 50:50 mixture of the acid and the between a mafic dominated volcanic sequence and a felsic
potassium salt. volcanoclastic sequence within the Whundo Group. The deposit
The novel structure of AM28 combined with the saturated C8 is interpreted as a VMS type deposit associated with the
hydrophobic tail makes the reagent highly effective as a selective development of a back arc extensional environment during the
collector of oxidised metal minerals in the froth flotation Archean. The deposit has been gently folded. Uplift and erosion
process. In practice the C8 tail promotes bubble attachment with during the Archean and more recent times has developed a small
minimal over-frothing from clays and fines and the hydroxamate copper supergene zone above the Eastern end of the deposit and
moiety selectively attaches to the oxidised metal surface (eg a more significant zinc-copper supergene zone over the Western
Cu(II)) by chelation. 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
The application of AM28 to copper flotation dipping fold, with minor remobilisation of massive sulfide into
Most copper deposits have a mineralogy and geochemistry which local shears.
changes significantly with depth. In a typical Cu ore deposit, the The copper minerals present in the supergene ores are
Cu and associated Au or Mo are often significantly enhanced by approximately equal amounts of digenite and chalcocite with
oxidative supergene enrichment above the primary chalcopyrite lesser amounts of chalcopyrite and covellite. Significant copper
zone. Cu minerals above this zone are progressively more oxide mineralisation has also been identified and includes
oxidised until at or near the surface, fully oxidised Cu species cuprite, with lesser malachite and azurite.
such as malachite and azurite occur. Cu minerals in the primary The digenite is predominantly present as fine scale veinlets of
and secondary enrichment zone (eg chalcocite, covellite, bornite, coarse pyrite and marcasite whose radial growth habit suggests
chalcopyrite and pyrite) can be recovered by conventional former pyrrhotite. The chalcocite occurs as fine veins through
flotation reagents. However, many sulfide minerals, such as coarse pyrite, but on a coarser scale than the digenite. It also forms
chalcocite (Kuhn, 1968) are readily surface oxidised and may coarse composites with chalcopyrite and non-invaded pyrite. The
benefit from the addition of AM28 to the flotation reagent chalcopyrite occurs with the chalcocite as above, and the covellite
regime. once rimming chalcopyrite. The main gangue mineral composites
In the upper zone of oxidised enrichment oxides, carbonates, are with sulfides in the coarser grains in quartz. There is talc and
sulfates and hydroxides and metallic Cu can be found. In this chlorite present that tend to be discrete fines.
zone, which is often stockpiled for future treatment, the use of
AM28 will allow marketable Cu concentrates to be recovered THE RADIO HILL FLOTATION CIRCUIT
using the existing mine plant flotation systems.
The flotation circuit at Radio Hill consists of a Rougher cell
RADIO HILL MINE, AUSTRALIA (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
Fox Resources currently operates the Radio Hill project, which is two cleaner cells (10.35 m3) that are in series to upgrade the
located 35 km South of Karratha in the West Pilbara region of rougher and scavenger concentrate from 12 per cent Cu to a
Western Australia. The Radio Hill treatment plant is capable of 27.4 per cent Cu final concentrate. The tail from the cleaner cells
producing copper, nickel and zinc concentrates. The plant is is combined with the scavenger tail and sent to tailings. Figure 2
currently configured to treat supergene copper ore sourced from shows a flow diagram of the circuit. Concentrate from the rougher
the West Whundo open pit. cell can also be routed straight to final concentrate bypassing the
The Whundo and Yannery mining leases cover an area of cleaning circuit. The retention time through the circuit is 16.0
approximately 900 ha and are located 10 km south of the Radio minutes (13.2 minutes in the rougher cell and 2.8 minutes in the
Hill treatment plant. Copper was mined from the Whundo deposit scavenger cell) and 10.2 minutes in both cleaner cells.
Feed
Cyclone
Cleaner Cells
Final Concentrate
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)
FIG 3 - In-stream analysis trends of flotation feed, final concentrate and final tail.
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.
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
D32 (mm)
size is one of the keys to understanding flotation recovery. Surprisingly, Bateman TC 50 Impala
these relationships are poorly understood, largely due to the difficulties 1.5 Metso RCS 30 Impala
associated with gas dispersion measurements in industrial environments.
The McGill gas dispersion sensors (bubble size, gas rate, gas holdup) are Outokumpu 28U Raglan
1
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, 0.5
which investigated the effect on bubble size of frother type and
concentration, gas rate, temperature and rotor speed. A Denver laboratory 0
cell was used to investigate the effect of elevation above sea level. Frother
0 0.5 1 1.5 2 2.5 3
concentration proved to have the predominant influence on bubble size,
and the effect of frother type can be normalised as a function of the Jg (cm/s)
critical coalescence concentration (CCC). Measurements made at many
plant sites worldwide are compared to the model developed using the FIG 1 - The D32 versus Jg relationship from earlier plant studies.
Metso RCSTM unit.
D32 = DO + CJg
n
(1)
INTRODUCTION
Shortly after the industrial advent of froth flotation at the Broken where:
Hill operation just over 100 years ago, T A Rickard (Rickard, Jg is the superficial gas velocity
1916) observed:
Do is the bubble size at Jg = 0
… we know that the key to the flotation process is
to be found, not in the oil, the acid, or the n is a constant
apparatus, but in the bubbles. C is a parameter dependent on the system chemistry and
Today, with the advent of new measurement technologies and bubble generation mechanism
computational techniques, the industry is finally getting around This relationship and the notion of a finite, creation bubble
to better defining the role of bubbles in flotation. Part of the size (Do) at Jg = 0 that appears to be similar (0.5 - 0.6 mm) for all
problem has been the lack of robust sensors capable of making the flotation plant systems investigated was intriguing and called
sufficiently precise measurements in a reasonable time frame in for further study in a more controlled environment outside the
the industrial environment. The recent development of gas plant.
dispersion sensors by the group at McGill University (Gomez The current investigation was therefore launched to more fully
and Finch, 2002) has helped to accelerate the study of the role investigate the relationship between bubble size and the main
bubble size plays in flotation at operating plants (Pyecha et al, operating variables of industrial flotation systems, viz frother
2006; Hernandez-Aguilar, Thorpe and Martin, 2006; Coleman, concentration and type, air rate, impeller speed (power intensity),
Urtubia and Alexander, 2006). pulp temperature and gas density. The latter two are often not
The impetus behind the work reported here was a series of deliberately controlled but could be important as many plants
earlier studies (Nesset et al, 2005; Nesset et al, 2006) in which operate in remote regions at extreme temperatures or at altitudes
the McGill gas dispersion sensors were used to measure bubble were the air density is considerably less than at sea level. The
size, superficial gas velocity and gas holdup in a suite of method selected was to employ an available pilot sized tank-cell
industrial flotation plants. The findings from those studies, supplied by Metso Minerals (a fully-instrumented 0.8 m3 pilot
shown in Figure 1, suggested that the Sauter mean diameter, the unit with a Svedala mechanism) that was sufficiently large to
D32, was well represented by the following relationship: 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.
1. PhD Candidate, McGill University, Department of Mining, Metals
and Materials Engineering, M H Wong Building, 3610 University PREVIOUS RESEARCH
Street, Room 2480, Montreal QC H3A 2B2, Canada.
Email: nessetech@videotron.ca
Characterising the bubble size distribution
2. Professor, McGill University, Department of Mining, Metals and
Materials Engineering, M H Wong Building, 3610 University Street, A population of bubbles in an aerated and agitated vessel will
Room 2600, Montreal QC H3A 2B2, Canada. have a (number) frequency distribution of bubble sizes that may
Email: jim.finch@mcgill.ca exhibit any of a variety of shapes; normal, lognormal or bimodal
3. Senior Research Associate, McGill University, Department of (Bailey, Gomez and Finch, 2005; Hernandez-Aguilar and Finch,
Mining, Metals and Materials Engineering, M H Wong Building, 2005) being the most common. As evident from Figure 3, the
3610 University Street, Room 2580, Montreal QC H3A 2B2, Canada. surface area and volume distributions (by bubble size) can vary
Email: cesar.gomez@mcgill.ca significantly from the number distribution for a typical
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
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
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)
The exponent n was found to be 0.25 for most porous sparger Gorain, Franzidis and Manlapig (1999) recognised that frother
types, while the C parameter depends primarily on frother concentration plays the largest role in bubble size, although it
concentration and also corrects for the effective active area of the does not appear in the equation. They concluded that given the
sparger. At the University of Cape Town a photometric device for good correlation of the data fitting, the frother concentration
measuring bubble size by extracting slurry samples through a must have been sufficiently high and the variation sufficiently
narrow capillary tube (the UCT bubble size analyser, Randall et little in these plants so as not to greatly influence the measured
al, 1989) was used to measure bubble size in a laboratory column bubble size. Although not specifically designed as a model of
as a function of different variables including air rate, temperature, bubble size, the re-arranged Gorain relationship is the only one
viscosity, particle size and pulp density (O’Connor, Randall and that predicts D32 for air rate and key variables in mechanical
Goodall, 1990). The data were fitted to power-law relationships flotation machines. Note that P80 can be considered a measure of
similar to Equation 10. The exponent, n, for bubble size versus pulp viscosity and S relates directly to power intensity.
gas rate was 0.4 and 0.47 respectively for air-water and The importance of considering frother in the bubble generation
air-water-solids, and n = 0.26 for the effect of viscosity in the process is evident from the observation of Harris and has been
air-water system. These tests were in the absence of frother. In demonstrated and commented on by several researchers recently
another series of tests, the authors note (same reference) the (Cho and Laskowski; 2002; Laskowski; 2003; Gélinas and Finch,
overwhelming effect of frother on bubble size, and that various 2005; Nesset et al, 2006). The notion that frother addition
xanthate collectors, alone or in combination with frother, had no reduces surface tension and hence the bubble size is an erroneous
additional effect on bubble size. one that still prevails (Gupta and Yan, 2006). One current
Several workers attempted to model the hydrodynamic mechanism considers the effect of frother (and electrolyte) on
performance of particular laboratory flotation machines ordering the local water structure, which increases resistance to
including Sawant et al (1981), Nonaka, Inoue and Imaizumi water drainage and hence retards coalescence (Finch, Gélinas
(1982), O’Connor, Randall and Goodall (1990), and Al Taweel and Moyo, 2006).
and Cheng (1995). They all found that the power intensity had a The review of the literature has pointed out that there is
measurable relationship with either gas holdup or the measured currently no specific model that incorporates the key
bubble size (O’Connor). It is suspected that these small manipulated and process variables for the prediction of bubble
laboratory machines do not scale well to the larger machines size in forced-air, mechanical flotation machines. At best, there
since measurements are made in very close proximity (a few are models from aerated stirred reactor design that can give an
centimetres) to the impeller/stator region, with its high indication; however, these systems differ in some important
recirculation rate of both fluid and gas back into the high shear aspects. In addition to the absence of stators, they operate at
region, compared to larger industrial units where the fluid flows different levels of gas rate and power intensity and (typically)
and bubbles leave the impeller/stator region. Hernandez-Aguilar, without the addition of surfactants. This work attempts to address
Gomez and Finch (2002) have shown the sensitivity between this shortcoming.
sampling location and measured bubble size distribution in a
Denver laboratory machine.
EQUIPMENT AND EXPERIMENTAL PROCEDURE
The first attempt at modelling the multiple effects of process
and flotation machine variables in industrial mechanical The Metso Minerals 0.8 m3 RCSTM pilot flotation cell was
forced-air machines using direct measurement of bubble size (the selected as it provided for power measurement and variable
UCT method) in plant environments was performed by Gorain, speed drive as well as being sufficiently large to accommodate
Franzidis and Manlapig (1997). Their work related the surface mounting of the McGill gas dispersion sensors. The installed unit
area flux (Sb) to the variables tested on a pilot flotation machine is shown in Figure 4, including the inset that shows the immersed
with interchangeable rotor/stator parts. The resulting model gas rate and gas holdup sensors. The arrangement is schematically
showed (Gorain, Franzidis and Manlapig, 1999): illustrated in Figure 5. Note the positioning of all sensors at the
same cross-sectional reference plane. The testing protocol calls
Sb ∝ Sb (Q / A) As P80
c d e
(11) 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
where:
conditions of the bubble flux as it is about to exit the liquid
b, c, d, e are fitted constants (pulp) phase. All three sensors were positioned symmetrically on
the radius (15.7 cm from the wall representing 73 per cent of the
S is the peripheral impeller speed distance from the centre of the cell) at the same elevation
Q is the volumetric air flow rate
A is the cell cross-sectional area GAS HOLD -UP
As is the aspect ratio of impeller diameter to height BUBBLE VIEWER
GAS RATE
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) FIG 4 - Metso pilot cell showing position of McGill bubble viewer
S As P80 and, inset, the gas rate (Jg) and gas hold-up (εg) sensors.
Air flow
TABLE 1
FCV
RPM Five frothers used in the laboratory testing.
Torque Frother Description Supplier Mol weight
Bubble size
P Pentanol Simple alcohol Fisher 88
Froth phase
Gas MIBC Methyl isobutyl carbinol Dow 102
velocity D-250 Polypropylene glycol Dow 235 - 265
X-sectional area
alkyl ether
reference plane
Gas holdup F-140 Mixture aldehydes Flottec Mixture of C8 -
Water phase and ketones C22, typical
200 - 250
•frother F-150 Polypropylene glycol Flottec 410 - 440
Impeller Stator isobutyl carbinol (MIBC), while D-250 was the standard frother
for the tests on impeller speed, water temperature and gas
FIG 5 - Schematic of the Metso pilot cell, manipulated variables density. The frother tests were conducted at two levels of
and sensor arrangement. superficial gas velocity, Jg = 0.5 and 1 cm/s. Frother additions
were incrementally and sequentially made in order to increase
(50 cm) from the cell bottom, a location that had been confirmed the concentration levels tested. For each new addition level, ten
through test work as giving consistent Jg and Db values. minutes of agitation in the cell without air preceded the
measurements. This was found to be critical in order to ensure
The details of the instruments and measurement techniques complete mixing as the MIBC and F-140, in particular, did not
have been well described elsewhere (Gomez and Finch, 2002; dissolve readily.
Hernandez-Aguilar, 2005; Gomez et al, 2006) and will be
reviewed only briefly here. The measurement of Jg relies on The tests for varying impeller speed were designed to establish
sensing volumetric change in a closed tube through a pressure the effect of power intensity on bubble size. It was confirmed
sensor thus giving a slope of pressure versus time. It was through testing that power (P) increase on the Metso cell did
established through testing that 15 repeat measurements of slope indeed follow the expected cubic power law relationship as a
were sufficient to yield steady mean and variance values for Jg function of impeller speed (N) (ie P ∝ N3, non-aerated and
(relative standard deviation 2.4 per cent). Bubble size distribution aerated). A doubling of impeller speed was possible on the unit,
is measured using a sampling tube and visualisation chamber thereby covering impeller tip speeds from 4.6 to 9.2 m/s,
with backlighting from which a video camera captures bubble providing an almost eight-fold increase in power intensity.
images. Typically, 10 000 individual bubble images are Typical impeller operating tip speeds on industrial machines are
processed by the commercial software (Northern EclipseTM, from 5 - 7 m/s. The impeller speed tests were run at two levels of
Empix Imaging) and deposited into an Excel file in order to Jg, 0.5 and 1 cm/s.
generate the distributions. The gas holdup (ε g) sensor is Varying gas density tests were included in an effort to simulate
comprised of two tubes that measure conductivity. One tube is elevation since many plants are located well above 3000 m, some
open and reports conductivity for the aerated liquid while the approaching 5000 m in Bolivia and Peru. Mixing air and helium
second tube is partly restricted at the bottom and reports (He) proved a convenient means of obtaining gas mixtures with
conductivity for gas-free liquid; the ratio of the readings permits the desired density representing different altitudes. Individual
the void fraction to be calculated. Both Jg and ε g measurement mass flow meters and controllers for air and He were used to
data were captured using commercial software (iFixTM, GE feed the mixture through an inline mixer before being introduced
Fanuc Automation). to the Denver flotation machine. Mixtures between zero per cent
The Metso cell has a nominal capacity of 800 litres; however, and 100 per cent He were tested; however, only a mixture of 60
a standard filling volume of 700 litres was established. Montreal per cent He is required to simulate an altitude of 5000 m, where
tap water was used and, except for the temperature tests, the cell air density is about half that at sea level. Only data for Jg values
was filled the day before a test to allow the water to equilibrate to of 0.25 and 0.3 cm/s were used, as the ability of the Denver
room temperature, typically about 16 - 18°C. Air was controlled machine to fully disperse the gas at higher rates (above 0.5 cm/s)
via a rotameter valve (rough adjustment) on the supply line and was suspect. Other test conditions were D-250 at 5 ppm and
fine adjusted via the Jg measurements to achieve the desired set impeller speed of 1600 rpm (5.9 m/s tip speed).
point. Bubble size, Jg and ε g measurements were taken The tests to establish the effect of water temperature served a
simultaneously. The tests conducted to establish the effects of dual purpose as the range of water temperature (3 - 41°C)
gas density were run on a standard Denver laboratory flotation introduced a correspondingly significant change in the (dynamic)
machine with mass flow meters and controllers on the air and water viscosity (1619 to 641 µPa-s). The changes in both water
helium lines feeding the cell. It should be noted that laboratory density (999.9 to 991.8 kg/m3) and surface tension (.07522 to
tests were conducted on the two-phase water-gas system only, .06943 N/m) over this range are very small in comparison with
while the reported plant results are for three phase water-solids- viscosity changes, so the test became, de-facto, an investigation
air. of the effect of viscosity over a temperature range that is
Table 1 lists the five frother types and suppliers. These representative of what flotation plants in many parts of the world
represent four of the major chemical groups of frothers. Pentanol experience. These tests were run at a Jg of 1 cm/s (at 5 ppm of
was added to the list of four commercially available products to D-250) and temperature was allowed to either increase (if
extend the range of molecular weight and carbon-chain length. starting below room temperature) or decrease (if starting above
(Plant surveys had also indicated that short chain alcohols can room temperature). Hot water was used for the warmer tests and
enter as impurities in commercial xanthate supplies (Gélinas and the naturally cold Montreal tap water of April/May was used for
Finch, 2005)). The standard frothers used for the tests the colder tests. In total, four series of tests were conducted and
investigating gas rate were Dowfroth-250 (D-250) and methyl the results combined.
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)
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.
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)
D-250 0
10.0
CCC0
4 6 8 10 12 14
CCC50
CCC75
IMPELLER TIP SPEED (m/s)
CCC90
D32 (mm )
CCC99 FIG 9 - D32 versus impeller rotational tip speed for increasing
1.0 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
0.1 Gas density
0 0.5 1 1.5
Jg (cm/s) The approach that varying the gas density can be used to
simulate elevation relies on the assumption that the type of gas
FIG 8 - Jg versus D32 for MIBC and D-250, fitted model curves (He mixed with air in this case) will not affect the bubble
and experimental data. 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,
the data points, the Do values (the notional D32 creation size for 1990). The data presented in Figure 11 showing D32 versus the
Jg = 0) are not a constant as postulated in the earlier work, but gas density (ρg, ie air-helium mixture) relative to 100 per cent air
decrease with increasing frother concentration. Similarly, the (ρo) at standard conditions (0.1 MPa, 25°C) was fitted to an
parameter C is also dependent on the frother concentration. The equation of the form:
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
[ ]
n
20 1.6
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 (%)
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 )
Liquid temperature
Comparison with plant data
Figure 12 shows both the standard curve of (dynamic) water
viscosity versus temperature (secondary y-axis) for 1 - 50°C, and
Bubble size, D32 and D10
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 The opportunity presents itself for comparing the fitted models
relating D32 to the water viscosity (ηl) for a given temperature for D32 versus superficial gas velocity and frother concentration,
relative to the viscosity at 20°C (ηo). The fitting yielded a value with data collected from the plant surveys. The MIBC model
of n = 0.777 for the following equation: curves for Jg (Equation 1 having n = 0.5) for frother levels
relative to CCC of 0, 50, 75, 90 and 99 per cent are compared to 100 CCC0
data from six plants in Figure 13. It is seen that the plant data CCC50
exhibit similar shaped curves to the laboratory generated models, 80 CCC75
with all data falling within the boundaries of maximum and CCC90
Sb (1/s)
minimum bubble size (CCC0 and CCC99). The model and the 60 CCC99
plant data presented are for forced air mechanical flotation Impala Bateman
40 Impala Metso
machines, but even when data for self-aerating machines are
Raglan
included in the plot (not shown) the fit within the model Lac des Iles
boundaries is equally good. This agreement suggests that the 20 WMC-LNO
model based on a two-phase test is applicable to the three-phase Escondida-LC
industrial data. 0
0 0.5 1 1.5 2
10.0 Jg (cm/s)
CCC0
CCC50
FIG 14 - Frother addition model (MIBC) versus plant data for D32
CCC75
CCC90
versus Sb. Model curves are indicated for frother levels of 0, 50,
D32 (mm)
CCC99
75, 95 and 99 per cent of CCC.
1.0 Impala Bateman
Impala Metso
Raglan DISCUSSION
Lac des Iles
WMC-LNO The need for a competent bubble size model for flotation
Escondida-LC machines goes beyond simply having an equation for predicting
0.1 bubble size. In today’s economic climate the need to rapidly
0 0.5 1 1.5 2 bring performance of new plants up to design, and to continually
Jg (cm/s) improve the performance of existing plants, is a matter of
survival. Robust models are required in all aspects from
FIG 13 - Frother addition model (MIBC) versus plant data for D32 equipment and process design, to plant optimisation, to
versus Jg. Model curves are indicated for frother levels of 0, 50, troubleshooting, benchmarking and seeking opportunities for
75, 95 and 99 per cent of CCC. 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
Figure 13 can be used to indicate where a plant is operating tanks without stators, and while there are a few basic models that
with respect to frother dosage. It was not possible to obtain have been developed for laboratory flotation machines, the need
actual plant measurements of frother concentration during the to build an understanding of how both chemistry and machine
surveys; however, as a baseline the data for Raglan and WMC variables affect bubble size on full-size machines was clearly
can be assumed to approach the 100 per cent frother CCC level evident. This work has taken some initial steps, starting with the
since their pulps are highly saline solutions, between 25 and two-phase water-air system. That the plant data all falls within
120 g/L salt. It has been demonstrated that electrolyte solutions the range of the model (Figures 13 and 14) suggests that the
exceeding an ionic strength of approximately 0.4 (equivalent to effects of solids may be of secondary importance to the primary
23 g/L NaCl) act in an equivalent manner to frother variables of frother concentration (relative to the CCC) and
concentration at 100 per cent CCC (Quinn, Gomez and Finch, superficial gas velocity (Jg). Some effect of solids on bubble size
2006). It can be seen from Figure 13 that the Raglan and WMC can be anticipated, either through the primary process of creation
data fall very close to the CCC99 model line. It is noted that all or secondary processes such as coalescence, but it also seems
of the data fall at concentrations at or above the CCC75 curve, possible that a portion of this effect could be the way in which
suggesting that none of the plants are operating at dramatically particles, especially fines, alter the fluid viscosity. This work has
reduced frother concentrations, but that there may be room for demonstrated that increasing water viscosity has a major impact
some improvement for several. The effect of the pulp on D32 (Equation 15) so the way in which solids influence
temperature at the various plants has not been incorporated into viscosity may impact bubble size as well. Extending the test
the data of Figure 13, and will be an obvious factor as range for viscosity effect on bubble size beyond that achievable
demonstrated earlier in this paper. This will account for some of by simply changing water temperature would clearly be of
the differences between plants, but again, likely to be a interest.
secondary effect as most plant pulps tested fall into the 20 - The strong influence of chemistry in determining bubble size
35°C range. 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
Bubble surface area flux, Sb which are closely linked to frother concentration. It would seem,
The data presented in Figure 13 can be readily converted to Sb, therefore, that the chemistry of the system is establishing both
the bubble surface area flux, a parameter that links directly to the initial (creation size, Do) bubble size as well as the rate of
bubble size change with increasing gas rate. This may be
flotation kinetics (Equation 5). This is shown in Figure 14.
somewhat of a departure from current views that see frother as
Again, the agreement with plant data is striking. The maximum inhibiting coalescence after initial bubble creation, but not the
(CCC99) and minimum (CCC0) boundaries shown in Figure 14 creation size itself (Alves et al, 2002; Laskowski, 2003). Figure 8
serve to represent the true limits of Sb for these types of flotation shows the strong effect of frother concentration on the initial
machines, since bubble size cannot be any larger than CCC0 nor bubble size created. In further support of this notion that the
effectively smaller than CCC99. Figure 14 is therefore seen as a frother is strongly influencing the initial bubble size, is the
useful tool for assessing the performance of a plant and for finding that increasing power intensity by a factor of eight
indicating the opportunity for improving flotation kinetics (Figure 9) did not change the bubble size distribution in any
through improving the Sb. Clearly there is some room for significant way. There is no doubt that such a marked increase in
improvement of Sb at several of the operations. power intensity has increased turbulence and shear but it has had
Cho, Y S and Laskowski, J S, 2002. Effect of flotation frothers on bubble Hinze, J O, 1955. Fundamentals of the hydrodynamic mechanism of
size and foam stability, Int J Miner Process, 64:69-80. splitting in dispersion processes, AIChE J, 1(3):289-295.
Coleman, R G, Urtubia, H E and Alexander, D J, 2006. A comparison of Kelly, E G and Spottiswood, D J, 1982. Introduction to Mineral
BHP-Billiton’s Minera Escondida concentrators, in Proceedings 38th Processing, pp 27-29 (J Wiley and Sons).
Annual Meeting of the Canadian Mineral Processors (division of Laskowski, J S, 2003. Fundamental properties of flotation frothers, in
CIM), pp 349-370. Proceedings 22nd International Mineral Processing Congress, Cape
Dahlke, R, Gomez, C O and Finch, J A, 2005. Determining operating Town, pp 788-797.
range of a flotation cell from gas holdup vs gas rate, Minerals Machon, V, Pacek, A W and Nienow, A W, 1997. Bubble sizes in
Engineering, 18(9):977-980. electrolyte and alcohol solutions in a turbulent stirred vessel, Trans
Deglon, D A, Egya-Mensah, D and Franzidis, J P, 2000. Review of Instrum Chem Eng, 75(A):339-348.
hydrodynamics and gas dispersion in flotation cells on South African Martinez-Bazan, C, Montanes, J L and Lasheras, J C, 2000. Bubble size
platinum concentrators, Minerals Engineering, 13(3):235-244. distribution resulting from the break-up of an air cavity injected into
Dobby, G S and Finch, J A, 1986. Particle collection in columns – Gas a turbulent water jet, Physics of Fluids, 12(1):145-148.
rate and bubble size effects, Can Metall Quart, 25(1):9-13. Nesset, J E, Gomez, C O, Finch, J A, Hernandez-Aguilar, J R and DiFeo,
Fallenius, K, 1976. Outokumpu flotation machines, in AM Gaudin A, 2005. The use of gas dispersion measurements to improve
Memorial Volume (ed: M C Fuerstenau), Vol 2, pp 838-861 (The flotation performance, in Proceedings 37th Annual Meeting of the
Society for Mining, Metallurgy and Exploration: Littleton). Canadian Mineral Processors (division of CIM), pp 401-421.
Finch, J A and Dobby, G S, 1990. Column Flotation, 180 p (Pergamon Nesset, J E, Hernandez-Aguilar, J R, Acuña, C, Gomez, C O and Finch,
Press). J A, 2006. Some gas dispersion characteristics of mechanical
Finch, J A, Gélinas, S and Moyo, P, 2006. Frother-related research at flotation machines, Minerals Engineering, 19(6-8):807-815.
McGill University, Minerals Engineering, 19(6-8):726-733. Nonaka, M, Inoue, T and Imaizumi, T, 1982. A microhydrodynamic
Gélinas, S and Finch, J A, 2005. Colorimetric determination of common flotation model and its application to the flotation process, in
industrial frothers, Minerals Engineering, 18(2):263-266. Proceedings 14th International Mineral processing Congress,
Gomez, C O, Acuña, C A, Girgin, E H and Finch, J A, 2006. Bubble Toronto, Vol III, pp 9.1-0.19.
distributions in laboratory flotation machines: Measurement and O’Connor, C T, Randall, E W and Goodall, C M, 1990. Measurement of
evolution with increasing gas velocity, in Proceedings International the effects of physical and chemical variables on bubble size, Int J
Mineral Processing Congress, Istanbul, pp 579-586. Min Pro, 28:139-149.
Gomez, C O and Finch, J A, 2002. Gas dispersion measurements in Parthasarathy, R, Jameson, G J and Ahmed, N M, 1991. Bubble break-up
flotation machines, CIM Bulletin, 95(1066):73-78. in stirred vessels predicting the Sauter mean diameter, Trans I Chem
Gorain, B K, Franzidis, J P and Manlapig, E V, 1997. Studies on impeller E, 69(A):295-301.
type, impeller speed, and air flow rate in an industrial scale flotation Pyecha, J, Sims, S, Lacouture, B, Hope, G and Stradling, A, 2006.
cell – Part 4: Effects of bubble surface area flux on flotation kinetics, Valuation of a Microcell™ sparger in the Red Dog column flotation
Minerals Engineering, 10(4):367-379. cells, Minerals Engineering, 19(6-8):748-757.
Gorain, B K, Franzidis, J P and Manlapig, E V, 1999. The empirical Quinn, J, Gomez, C O and Finch, J A, 2006. Effect of salt type and
prediction of bubble surface area flux in mechanical flotation cells, concentration on gas dispersion properties, in Interfacial Phenomena
Minerals Engineering, 12(3):309-322. in Fine Particle Technology, 6th UBC-McGill-UA International
Gupta, A and Yan, D S, 2006. Mineral Processing Design and Symposium on Fundamentals of Mineral Processing (eds: Z Xu and
Operations: An Introduction, p 559 (Elsevier: Amsterdam). Q Liu), pp 417-428 (Canadian Institute of Mining and Metallurgy:
Montreal).
Harris, C C, 1976. Flotation machines, in AM Gaudin Memorial Volume,
(ed: MC Fuerstenau), Vol 2, pp 753-815 (The Society for Mining, Randall, E W, Goodall, C M, Fairlamb, P M, Dold, P L and O’Connor,
Metallurgy and Exploration: Littleton). C T, 1989. A method for measuring the sizes of bubbles in two and
three-phase systems, Journal of Physics E: Scientific Instruments,
Heiskanen, K, 2000. On the relationship between flotation rate and 22:827-833.
bubble surface area flux, Minerals Engineering, 13(2):141-149.
Rickard, T A, 1916. The Flotation Process, 364 p (Mining and Scientific
Hernandez, H, Gomez, C O and Finch, J A, 2003. Gas dispersion Press: San Francisco).
and de-inking in a flotation column, Minerals Engineering,
16(6):739-744. Sawant, S B, Joshi, J B, Pangarkar, V G and Mhaskar, R D, 1981. Mass
transfer and hydrodynamic characteristics of the Denver type
Hernandez-Aguilar, J R and Finch, J A, 2005. Validation of bubble sizes flotation cells, Chem Eng J, 21:11-19.
obtained with incoherent imaging on a sloped viewing window,
Chemical Engineering Science, 60:3323-3336. Sridhar, T and Potter, O E, 1980, Interfacial areas in gas-liquid stirred
vessels, Chem Eng Sci, 35:683-691.
Hernandez-Aguilar, J R, Gomez, C O and Finch, J A, 2002. A technique
for the direct measurement of bubble size distribution in industrial Wilkinson, P M and van Dierendonck, L, 1990. Pressure and gas density
flotation cells, in Proceedings 34th Annual Meeting of the Canadian effects on bubble break-up and gas hold-up in bubble columns, Chem
Mineral Processors (division of CIM), pp 389-402. Eng Sci, 45:2309-2315.
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.
CIRCUIT 1 CIRCUIT 2
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
A bottom-driven JKMRC flotation cell with a volume of five of cell feed, concentrate, tail, a sample of the pulp below the
litres was used for all of the batch flotation tests, which were pulp-froth interface and from the froth surface. Froth recovery
performed under identical conditions. No reagents were added to was measured once in each FCTR cell for each circuit. The froth
the samples before the tests were conducted. The cell was recovery samples were taken at the same time as the first survey
operated using a shallow froth depth to minimise the effect of the around the FCTR.
froth phase, and a froth recovery of 100 per cent was assumed in
all subsequent calculations. Concentrate was collected in
FCTR model validation
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 Circuit configurations
bucket and the mass of each tray and the tails sample were For the predictive validation tests, data were collected around a
measured. further five FCTR circuits. The results from three of the five
validation circuits will be discussed in this paper. The three
Gas dispersion measurements FCTR model validation circuits discussed in this paper are
Gas dispersion characteristics (air hold-up, superficial gas shown in Figure 3.
velocity and bubble size) were measured to characterise the Circuit 5 consists of a bank of six rougher cells, a bank of four
operation of the FCTR cells and to allow the estimation of cleaner cells and a bank of two recleaner cells. The concentrates
bubble surface area flux values for each operating cell. from the cleaner cells are pumped to the recleaner bank and the
The air hold-up was determined by isolating each cell recleaners produce the final concentrate. The tails from the
individually and measuring the difference in pulp level between rougher, cleaner and recleaner banks are discarded and there are
the cell operating normally and not operating. This was done by no recycle streams.
simultaneously stopping the impeller, shutting off the air to the Circuit 6 is similar to Circuit 3 and consists of a bank of six
cell, stopping the feed to the cell, and closing the tails valve. The rougher cells, a bank of four cleaner cells and a bank of two
superficial gas velocity was calculated as the air flow rate in the recleaner cells. The concentrates from the cleaner cells are
cell divided by the cell cross-sectional area. The bubble size was pumped to the recleaner bank and the recleaners produce the
measured using the UCT bubble size analyser (Tucker et al, final concentrate. The recleaner bank tails are recycled to the
1994). Bubble size measurements were performed in roughers 1, head of the cleaner bank. The cleaner tails are recycled to the
2, 5 and 6 and cleaners 1 and 6. 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
Froth recovery measurements flotation. The tails from the rougher bank are again discarded.
Froth recovery was measured using the mass balance method Circuit 7 consists of a bank of five rougher cells, a bank of six
(Alexander, Franzidis and Manlapig, 2003). This requires samples cleaner cells and a single recleaner cell. In this circuit, the last of
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 From the estimated floatability characteristics it was found
circuits using a recleaner bank, the effective volume of each that the sulfide minerals were best represented as consisting of
recleaner cell is 35 litres (70 litres total recleaner capacity for two floating fractions (fast and slow) and a non-floating fraction,
two cells). In this case, as a larger rougher cell is used as a and the non-sulfide gangue was best represented as consisting of
recleaner cell, the effective volume of the recleaner cell is a slow floating fraction and a non-floating fraction. Pentlandite
125 litres. Therefore, the recleaning capacity in this circuit is consisted mainly of fast-floating material and a small proportion
almost doubled. The residence time in the cleaner circuit is also of slow-floating and non-floating material. Pyrrhotite consisted
increased as the two cells used previously for recleaning mainly of slow-floating material. Chalcopyrite consisted mainly
(Circuits 5 and 6) are used as additional cleaners. of fast-floating material. Non-sulfide gangue contained a small
In all the validation circuits depressant (guar) was added to the proportion of slow-floating material but was predominantly
cleaner and recleaner feed streams to minimise unwanted talc non-floating.
flotation. Where data from two, three or four calibration circuits were
used to derive the ore floatability characteristics, it was assumed
Metallurgical surveys that the feed ore floatability characteristics were the same for
each circuit. A single set of floatability characteristics was
Metallurgical samples were taken around each of the FCTR successfully fitted to the combined survey and batch flotation
validation circuits to determine the circuit performance and to test data from each circuit. No significant differences in the
establish the mass balance for each circuit. Three repeat floatability characteristics were found with an increasing number
metallurgical surveys were performed for Circuit 5 and two of calibration circuits.
repeat surveys for Circuits 6 and 7. The samples were collected
and assayed using the same procedure as that used in the model MODEL VALIDATION RESULTS
calibration surveys.
The validation was performed using the feed floatability
MODEL CALIBRATION RESULTS characteristics determined from the various calibration models
discussed above, in conjunction with estimates of the model
The surveys around the calibration circuits were first mass parameters. JKSimFloat V6.1 was used to simulate the
balanced. The ore floatability characteristics for each calibration performance of each validation circuit. A number of different
circuit were estimated using the mass balanced survey data in methods were used to predict the performance of the validation
conjunction with the laboratory batch flotation test results and circuits. The prediction results for two of the methods will be
the measured model parameters (the degree of entrainment was discussed in this paper. In prediction method 1, the water flow to
calculated using the technique described by Savassi (1998) using the concentrate in each cell was predicted using an empirical
the recovery to the concentrate of a liberated non-floating tracer water recovery model based on the flow rate of solids to the
present in the ore). A number of flotation models were calibrated concentrate. In prediction method 2, experimental water flows
using data from one, two, three or four calibration circuits. were used.
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
FIG 5 - Comparison of experimental and predicted mineral grades for predictions producing the highest (A) and lowest (B) average
relative errors (Circuit 5).
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.
100 100
High NiS Low NiS
90 90
High FeS Low FeS
80 80
High CuS Low CuS
70 70
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 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 • Non-sulfide gangue recovery is under-predicted for the
and within the experimental standard deviation for most model that gives the highest average relative errors and
streams in the circuit. accurate for the model the gives the lowest average relative
• The prediction of pyrrhotite recovery is poor (gradient = 0.44) errors. The non-sulfide gangue grade is over-predicted for the
model producing the highest average relative errors (gradient
for the model producing the highest average relative errors
= 1.07). The over-prediction in non-sulfide gangue grade is
with the majority of points outside the experimental standard
linked with the under-prediction of the floatable minerals
deviation. For the model producing the lowest average relative
grades.
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 Comparison of models and prediction methods
average relative errors, the poor prediction of pyrrhotite Table 4 shows the average relative errors based on the number of
recovery may be related to the addition of guar as discussed calibration circuits and prediction method used for all models
previously, or to the under-prediction of water recovery. tested.
• The pyrrhotite grade is under-predicted for the model The prediction results show that the average relative error
producing the highest average relative errors. decreases as the number of calibration circuits increases.
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 (%)
60 60
50 50
40 40
FIG 8 - Comparison of experimental and predicted mineral recoveries for predictions producing the highest (A) and lowest (B) average
relative errors (Circuit 7).
100 100
90 90
80 80
70 70
Predicted Grade (%)
50 50
40 40
FIG 9 - Comparison of experimental and predicted mineral grades for predictions producing the highest (A) and lowest (B) average
relative errors (Circuit 7).
Flotation model parameters These results have shown that the predictive validation is
expected to be accurate, even for a circuit with several recycle
The accurate estimation or measurement of the flotation model streams, as long as the operating conditions of the circuit to be
parameters (bubble surface area flux, froth recovery, residence predicted are within the range of the operating conditions under
time, degree of entrainment and water recovery) is critical for the which the model was calibrated. Future research should be
accurate prediction of circuit performance. undertaken to investigate how far accurate predictions can be
This work showed a significant difference between the made outside the range of operating conditions. For now it is
predictions based on the two different methods of estimating the recommended that the calibration circuit contain flotation banks
water flow. As expected, the predictions using experimental that treat material of a similar type to that which must be
water flows were significantly more accurate in most circuits. predicted (eg cleaning or recleaning stages).
Predictions based on experimental water recoveries with a single
calibration circuit were within three per cent of, or more accurate ACKNOWLEDGEMENTS
than, the predictions based on the empirical water recovery
model with four calibration circuits. With an improved water The authors would like to thank Dorr-Oliver Eimco and GL&V
recovery model, fewer calibration circuits would be required to for making the FCTR available to the AMIRA P9 project. The
achieve the same prediction accuracy. authors would like to thank the staff at Kambalda Nickel
Operation for their assistance with the test work at KNO. The
Complexity of the model calibration circuit authors would also like to thank Dan Alexander, David Seaman
and Brigitte Seaman from the JKMRC for their assistance in the
This work showed that, of the four calibration circuits, the most operation of the FCTR at KNO. Finally, the authors would like to
accurate prediction of the performance of the three validation acknowledge the AMIRA P9 project and its sponsors for the
circuits was achieved using either Circuit 2 (two-stage cleaning
funding which made this test work possible.
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 REFERENCES
streams, was unable to predict circuit performance accurately. As Alexander, D J, Bilney, T and Schwarz, S, 2005. Flotation performance
all validation circuits were two-stage cleaning circuits, vital improvement at Placer Dome Kanowna Belle Gold Mine, in
operating information was missing from the circuit based only on Proceedings 37th Annual Canadian Mineral Processors Conference,
a single stage of cleaning. pp 171-201 (The Canadian Institute of Mining, Metallurgy and
The design of the model calibration circuit is therefore critical. Petroleum: Montreal).
It must consist of flotation banks that treat material of a similar Alexander, D J, Franzidis, J-P and Manlapig, E V, 2003. Froth recovery
type to that which must be predicted (eg cleaning or recleaning measurement in plant scale flotation cells, Minerals Engineering,
stages). 16:1197-1203.
Furthermore, the complexity of the circuit will affect the Alexander, D J and Morrison, R D, 1998. Rapid estimation of floatability
components in industrial flotation plants, Minerals Engineering,
stability of circuit operation. A highly complex circuit with many 11:133-143.
recycle streams may be difficult to control hence the model
Alexander, D J, Runge, K C, Franzidis, J-P and Manlapig, E V, 2000. The
calibration measurements may not be representative of the actual application of multi-component floatability models to full-scale
operation of the circuit. flotation circuits, in Proceedings Seventh Mill Operators’
Conference, pp 167-177 (The Australasian Institute of Mining and
Number of model calibration circuits Metallurgy: Melbourne).
Alexander, D J and Wigley, P, 2003. Flotation circuit analysis at WMC
The accuracy of the prediction of circuit performance was found
Ltd Olympic Dam Operation, in Proceedings Eighth Mill Operators’
to improve with the number of circuits used in the model Conference, pp 41-52 (The Australasian Institute of Mining and
calibration. This could be due to an increase in the amount of Metallurgy: Melbourne).
data used in the model calibration and to the inclusion of a wide Coleman, R G, Esvelt, M, Franzidis, J P and Manlapig, E V, 2003. The
selection of data from circuits of increasing complexity. Floatability Characterisation Test Rig as a pilot plant for flotation
However, this improvement in the accuracy of prediction must be testing – An operator’s viewpoint, in Proceedings Eighth Mill
weighed up against the increased test work (in terms of cost, Operators’ Conference, pp 81-86 (The Australasian Institute of
analysis time, etc) that must be performed to calibrate more than Mining and Metallurgy: Melbourne).
a single flotation circuit. Coleman, R G, Franzidis, J P and Manlapig, E V, 2004. Using the
Flotation Characterisation Test Rig for industrial flotation plant
CONCLUSIONS design, in Proceedings Metallurgical Plant Design and Operating
Strategies 2004, pp 135-149 (The Australasian Institute of Mining
The AMIRA P9 flotation model and modelling methodology has and Metallurgy: Melbourne).
been successfully validated using the FCTR. Overall, the Coleman, R G, Urtubia, H E and Alexander, D J, 2006. A comparison of
predictions of the circuit performance were accurate and within BHP-Billiton’s Minera Escondida flotation concentrators, in
experimental standard deviations for most streams in the circuits. Proceedings 38th Annual Canadian Mineral Processors Conference,
The average relative error at each data point from all predictions pp 349-370 (The Canadian Institute of Mining, Metallurgy and
ranged from five per cent to 45 per cent. The predictions of the Petroleum: Montreal).
key variables of pentlandite and chalcopyrite recovery were Harris, M C, Runge, K C, Whiten, W J and Morrison, R D, 2002.
accurate. JKSimFloat as a practical tool for flotation process design and
optimisation, in Proceedings SME Mineral Processing Plant Design,
The prediction of pyrrhotite recovery and grade produced the Practice and Control Conference, pp 461-478 (The Society for
largest errors. If the prediction of pyrrhotite performance was Mining, Metallurgy and Exploration: Littleton).
excluded from the analysis, the average relative error at each data Rahal, K R, Franzidis, J-P and Manlapig, E V, 2000a. Flotation plant
point from all predictions ranged from four per cent to 18 per modelling and simulation using the Floatability Characterisation Test
cent. The prediction of pyrrhotite recovery appeared to be Rig (FCTR), in Proceedings MINPREX 2000, pp 339-344 (The
dependent on the addition of guar to the cleaner and recleaner Australasian Institute of Mining and Metallurgy: Melbourne).
feed streams of the circuit to be predicted. When the guar Rahal, K R, Franzidis, J-P and Manlapig, E V, 2000b. The application of
addition rates were significantly different to the range of addition the Floatability Characterisation Test Rig (FCTR) in flotation
rates used in the calibration circuits, the prediction of pyrrhotite research, in Proceedings Seventh Mill Operators’ Conference, pp
recovery was inaccurate. These errors were reduced when 339-344 (The Australasian Institute of Mining and Metallurgy:
experimental water recovery values were used. Melbourne).
Rahal, K R, Nelson, M G, Lelinski, D, Gritton, K S, Manlapig, E V and Schwarz, S, Alexander, D J, Whiten, W J, Franzidis, J-P and Harris, M C,
Franzidis, J-P, 2000. Design and operation of a Floatability 2006. JKSimFloat v6: Improving flotation circuit performance and
Characterisation Test Rig, in Proceedings SME Annual Conference understanding, in Proceedings XXIII International Minerals
(The Society for Mining, Metallurgy and Exploration: Littleton). Processing Congress, Vol III, pp 1717-1722 (International Minerals
Savassi, O N, 1998. Direct estimation of the degree of entrainment and Processing Congress: Istanbul).
the froth recovery of attached particles in industrial flotation cells, Schwarz, S and Kilgariff, B, 2005. Flotation optimisation studies of the
PhD thesis, University of Queensland, Brisbane. Perilya Broken Hill lead circuit, in Proceedings Centenary of
Schwarz, S, Alexander, D J and Coleman R G, 2006. Development of a Flotation Symposium, pp 913-922 (The Australasian Institute of
hybrid floatability model, in Proceedings XXIII International Mining and Metallurgy: Melbourne).
Minerals Processing Congress, Vol III, pp 1860-1865 (International Tucker, J P, Deglon, D A, Franzidis, J-P, Harris, M C and O’Connor, C T,
Minerals Processing Congress: Istanbul). 1994. An evaluation of a direct method of bubble size measurement
in a laboratory batch flotation cell, Minerals Engineering,
7(5/6):667-680.
ABSTRACT and bubble collapse rates. These systems are being installed in
many operations around the world. The predominant control
Metso Minerals Cisa have developed Visiofroth™ – a system which
calculates parameters of the froth surface using vision from remote philosophy employed by these systems involves manipulation of
cameras mounted about individual flotation cells. The parameters froth velocity to maintain metallurgical targets (eg grade or
calculated from this system can be utilised within process control recovery).
strategies to achieve stabilisation or optimisation of the flotation process. There is scope for better utilisation of the parameters measured
As part of a JKMRC AMIRA P9 field campaign, froth vision was by Visiofroth™. Improved understanding of:
collected from various cells operated at different air rates and froth depths
within Rio Tinto’s Northparkes copper roughing and scavenging circuit. 1. the relationship between the measured parameters and the
At the same time, samples were collected to determine the cell metallurgical performance of the cell; and
metallurgical performance at each operating condition. 2. how the various operational parameters (eg air rate, froth
These froth images have been processed using the Visiofroth™ system depth, frother concentration, reagent dosage rate) impact on
with froth velocity, bubble size, froth collapse rate and various froth
these measured parameters,
colour parameters calculated. These parameters have been compared to
the measured concentrate flow and concentrate grade produced at each will lead to more effective process control and optimisation.
condition. This analysis shows that the concentrate tonnage flow is In this paper, Visiofroth™ is used to analyse froth images
strongly correlated with froth velocity whereas the concentrate grade is
best correlated using froth velocity in combination with a bubble stability
collected from rougher and scavenger cells operated at different
parameter which affects the bubble size distribution on the top surface of air rates and froth depths. Parameters derived from this analysis
the froth. are evaluated to determine if they can be related to the measured
This paper summarises the results from this analysis as well as an grade and recovery produced at each operating condition.
overview of the Visiofroth™ system and the parameters that are
measured by the software.
THE VISIOFROTH™ SYSTEM
INTRODUCTION Visiofroth™ is an image analysis system for online measurement
of flotation froth properties. Visiofroth™ camera and lighting
Traditionally operators have used the visual appearance of the assemblies are typically mounted directly above the flotation cell
froth to make decisions about when and what changes need to be just inside the cell lip. The camera signals are relayed back via
made to optimise flotation circuit performance. With the
ethernet cable and optical fibre to a central processing computer,
development of inexpensive and efficient camera technology and
faster computer processing speeds, it is now possible to capture typically located in the plant control room.
and measure various characteristics of the froth in a timely This central processing computer uses the image stream from
manner. These measurements can be used within a process each camera to calculate various froth specific parameters.
control strategy to stabilise and optimise flotation cell operation. Currently the system calculates the froth velocity, froth bubble
These systems have the advantage that they can be ‘on the job’ size distribution, various colour and brightness indices, froth
all of the time and provide quantitative rather than vague collapse rate and image stability. Froth texture matching
qualitative measures of the froth appearance. algorithms and various statistical parameter calculations (eg one
Visiofroth™ is a froth image analysis system that has been hour average, rate of change) are also provided within the
developed by Metso Minerals Cisa. It can calculate various Visiofroth™ software.
parameters associated with the froth phase, including the froth Within the control room, the flotation circuit operators are
velocity, various froth colour parameters, bubble size distribution 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
1. Manager – Flotation Process Technology, Metso Minerals Process
Technology Group – Asia Pacific, Metso Minerals, Unit 1, 8 - 10
the software, which enables the operator to view and assess the
Chapman Place, Eagle Farm Qld 4009. changes in the froth parameters as a function of time.
Email: kym.runge@metso.com As Visiofroth™ is a module of OCS©, Metso Mineral Cisa’s
2. Graduate Process Technology Engineer, Metso Minerals Process expert system, further processing of the froth parameters is easily
Technology Group – Asia Pacific, Metso Minerals, Unit 1, 8 - 10 accomplished. Froth parameters can be used, often in
Chapman Place, Eagle Farm Qld 4009. conjunction with other sensor information (eg on stream
Email: jaclyn.mcmaster@metso.com analysis), to calculate new operational set points, which can be
3. Process Control Engineer, Metso Minerals Process Technology
outputted directly to the plant’s DCS or PLC. The usual control
Group – Asia Pacific, Metso Minerals, Unit 1, 8 - 10 Chapman strategy currently employed is to maximise froth velocity (and
Place, Eagle Farm Qld 4009. Email: michael.wortley@metso.com thus cell recovery) through air and froth depth adjustment whilst
not falling below concentrate grade targets or exceeding other
4. Manager – Process Control and Information Engineering, Metso
Minerals Process Technology Group – Asia Pacific, Metso Minerals,
process constraints. The system can also generate automatic
Unit 1, 8 - 10 Chapman Place, Eagle Farm Qld 4009. alarms associated with abnormal froth parameters (eg excessive
Email: david.larosa@metso.com froth velocity).
5. Product Development Manager, Metso Minerals Cisa Process In the following sections, the Visiofroth™ parameters that are
Technology Group, 3 Avenue Claude Guillemin, 45060 Orléans to be investigated within the study presented in this paper will be
Cedex 2, France. Email: olivier.guyot@metso-cisa.com described and the method of calculation will be outlined.
Froth velocity Visiofroth™ analyses a defined area of the image and calculates
the parameters associated with three different colour models: the
The speed at which the froth moves towards the launder is a RGB colour model, the HSV colour model and the LAB colour
parameter often used by operators to assess the mass pull and model. The average colour descriptors of the image are reported
recovery off a flotation cell. Visiofroth™ measures the velocity as well as the proportion of pixels within a subset of the colour
in both the x and y direction, as well as the resultant velocity descriptors.
over a specified region of froth. A modified fourier transform
comparative technique is used to assess the displacement in both The RGB colour model is based on the Cartesian coordinate
the x and y dimensions of consecutive images. This displacement system where colour is defined by its position in the RGB colour
coupled with the frame collection rate is used to calculate a cube. The origin is black, with the extremities of the three axes
velocity. The reported velocity is an average calculated from being red, green and blue. The point furthest from black, passing
processing up to 30 frames per second over a typical ten second through the cubic centre, is white. The remaining three points are
cycle period. The algorithm produces a robust velocity estimate magenta, cyan and yellow. A colour is defined by the values of
even in poor lighting conditions. red, green and blue it possesses.
The HSV colour model provides an alternative method of
defining colour through its analysis of hue, saturation and
Bubble size distribution intensity – parameters more akin to a human’s perception of
The texture or bubble size distribution is often used by operators colour. Hue is related to a pixel’s true colour (eg green, blue or
as a barometer of concentrate quality or as an indicator of the orange) and is represented as an angular displacement from the
need for a reagent change. Visiofroth™ uses state-of-the-art red axis. Saturation is a measure of how much white light is
watershed techniques to delineate the bubbles within the image. present in a hue and ranges between zero and 100 per cent. For
The equivalent circle diameter of these bubbles is then calculated example, red is a fully saturated colour whereas pink (red and
and used to assign each bubble to a bubble size distribution class. white) is less saturated. Finally, value is a measure of a colour’s
This bubble size count in each class is reported within the system intensity or brightness – which is also reported as a value
and can be used to calculate a number of different bubble size between zero and 100 per cent. A value of zero for any colour is
distribution parameters (eg percentage of bubbles in a size class, white whereas a value of 100 of any colour is black.
average bubble size). Like the HSV colour model, the LAB colour model attempts
Watershed techniques are suitable for bubble delineation of to decouple colour from its intensity. It is also designed to
froth images because the pixels that delineate the border of associate similar looking colours closer to each other within the
bubbles are usually darker than those in its centre. For these colour space. The luminance (L) is a value representing the
types of segmentation problems, watershed is a technique that intensity of the colour and is reported as a value between zero
produces stable segmentation with continuous segmentation and 100 per cent, extending from black to white. The Lab ‘a’
boundaries (Gonzalez and Woods, 2002) and is thus being used parameter is an axis spanning from 100 per cent red to 100 per
in a number of froth vision systems (Sweet et al, 2000; Hargrave cent green and the Lab ‘b’ parameter is an axis spanning from
and Hall, 1997). Visiofroth™ also enables adjustment of the 100 per cent yellow to 100 per cent blue.
parameters associated with the watershed algorithm to enable
optimisation of the delineation of bubbles in the image. Froths Bubble collapse rate
with a wide size distribution will, however, exhibit some areas of
over-segmentation and/or some areas of under-segmentation, The rate at which bubbles coalesce within and on the surface of
regardless of the parameters chosen. The number of the froth has been shown to have a strong impact on the recovery
pixels/bubble (which will be related to the zoom setting on the and drainage within the froth phase (Zheng, Franzidis and
camera) has been found to affect the result of the segmentation Manlapig, 2004; Hatfield, 2006). It has been shown to be
algorithm. strongly affected by the frother concentration, hydrophobicity of
The segmented image and the bubble size distribution are the attached mineral particles and bubble loading (Hatfield,
displayed, pictorially, within the software. Different colours can be 2006).
assigned to the bubbles of different size on these images. Results Visiofroth™ reports a measure potentially associated with the
of a Visiofroth™ bubble segmentation is shown in Figure 1. 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
Colour and brightness descriptors
the ‘per cent change in bubble surface area per frame pair’.
Operators often visually assess the concentrate grade produced This parameter is related to the size and presence of bubbles
from a cell by the colour of the froth. There are a variety of measured by the system. It will also be affected by velocity as
different colour models/descriptors that can be used to quantify the bubble size histogram will potentially change as bubbles
the ‘colour’ of a pixel or the colour distribution across an image. move out of the field of view.
A B C
FIG 1 - Results of the Visiofroth™ bubble segmentation algorithm: (A) original image, (B) segmented image and (C) colour differentiation
of bubble size.
Rougher Scavenger
Concentrate Concentrate
FIG 2 - Northparkes rougher and scavenger flotation circuit.
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.
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
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
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.
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)
These relationships, however, all exhibited considerable FIG 12 - Comparison of the average bubble size with the
scatter, an example of which is shown in Figure 11, and therefore concentrate and top of froth grades measured in the different cells
would only be useful for a rough estimate of the grade being during the study.
produced from a cell.
Although the froth was illuminated during image collection Correlations between grade and collapse rate
and black plastic was used to minimise ambient light
interference, there were noticeable differences in the brightness The best correlation, however, was obtained by comparing top of
of the images during the course of a day of measurement. All froth grade with Visiofroth™’s bubble collapse rate parameter
colour models suffer from the fact that their colour parameters (Figure 13). This result is similar to that found by Morar et al
are affected by the degree of luminescence. It is interesting to (2006) who analysed images from the Northparkes rougher using
note that the HSV model and Lab model, which attempt to Smartfroth software. In their analysis they found the correlation
differentiate between colour and brightness, were better peak, which they use as a measure of bubble collapse rate, was
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 (%)
14.0
Collapse Rate
In this Northparkes case study, solids and water flow from a Hadler, K, Barbian, N and Cilliers, J J, 2006. The relationship between
particular cell on a particular day was related to the measured froth stability and flotation performance down a bank of cells, in
froth velocity. Thus it is appropriate to use the velocity measured Proceedings XXIII International Mineral Processing Congress,
Volume 1, pp 516-523, Istanbul, Turkey, 3 - 8 September.
by Visiofroth™ as a means of increasing or decreasing recovery
to achieve a particular processing target. These types of control Hargrave, J M and Hall, S T, 1997. Diagnosis of concentrate grade and
mass flowrate in tin flotation from colour and surface texture
loops have the advantage that they can respond to a change more analysis, Minerals Engineering, 10(6):613-621.
quickly than one dependent on on-stream analysis alone.
Hatfield, D P, 2006. Factors affecting froth surface descriptors and the
The grade of attached particles on the surface of the froth was implications for froth structure and flotation performance, PhD thesis
linearly correlated with the bubble collapse rate. This (unpublished), University of Cape Town, Cape Town, South Africa.
relationship included data collected from the different cells Hatfield, D P, Bradshaw, D and de Jager, G, 2003. The relationship
operated at different air rates and froth depths on different days between concentrate yield and descriptors from a machine vision
of measurement and thus over the period of the field visit proved system in a platinum flotation application, in Proceedings XXII
relatively robust. Overall concentrate grade was best predicted International Mineral Processing Congress, pp 929-936, Cape Town,
using an equation incorporating the collapse rate and velocity South Africa, 29 September - 3 October.
parameters. These correlations, potentially, allow the calculation Holtham, P N and Nguyen, K K, 2002. On-line analysis of froth surface
of concentrate purity targets for use in a process control in coal and mineral flotation using JKFrothCam, International
Journal of Mineral Processing, 64(2-3):163-180.
optimisation strategy.
Kewe, T, 2003. Advances in flotation process control at Porgera Gold
This study has verified the findings of Morar et al (2006) who Mine using JKFrothCam imaging technology, in Proceedings Eighth
showed that it is better, when investigating froth image data, to Mill Operators’ Conference, pp 87-92 (The Australasian Institute of
consider the froth as made up of two distinctly different types of Mining and Metallurgy: Melbourne).
particles – those attached to the bubbles and those entrained in Morar, S H, Forbes, G, Heinrich, G S, Bradshaw, D J, King, D, Adair,
the plateau borders within the froth phase. Delineation of the B J I and Esdaile, L, 2005. The use of a colour parameter in a
problem in this way, allows correlations to be developed which machine vision system, Smartfroth, to evaluate copper flotation
would otherwise, potentially, have been hidden. performance at Rio Tinto’s Kennecott Utah Copper Concentrate, in
Proceedings Centenary of Flotation Symposium, pp 147-151 (The
Bubble collapse rate in the froth was found to be solely a Australasian Institute of Mining and Metallurgy: Melbourne).
function of the grade of attached particles and therefore would Morar, S H, Hatfield, D P, Barbian, N, Bradshaw, D J, Cilliers, J J and
seem independent of particle froth loading. This result may have Triffett, B, 2006. A comparison of flotation froth stability
important implications as we move towards improved measurements and their use in the prediction of concentrate grade, in
understanding of the froth phase and how we should manipulate Proceedings XXIII International Mineral Processing Congress,
variables to achieve optimal performance. 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
ACKNOWLEDGEMENTS AMIRA P9M flotation report 02/09 (unpublished).
The authors would like to acknowledge Eddy Sanwani, David Sweet, C G, Bradshaw, D J, Cilliers, J J L, Wright, B A, de Jager, G and
Seaman, Cesar Gomez, Jorge Torrealba, Brigette Comley, Marco Francis, J J, 2000. The extraction of valuable minerals from mined
Vera, Michael Rosenfield and Esther Soden who were involved ore, Smartfroth Adams & Adams Patent Attorneys, Pretoria, A&A
in the JKMRC campaign from which the data used in this paper Ref: V13676 (1-11) 1999, Full Patent October 2000.
was collected. We’d also like to acknowledge the support of the van Olst, M, Brown, N, Bourke, P and Ronkainen, S, 2000. Improving
Northparkes metallurgists and technical assistants who provided flotation plant performance at Cadia by controlling and optimising
assistance with the test work. Northparkes is also thanked for the rate of froth recovery using Outokumpu Frothmaster™, in
Proceedings Seventh Mill Operators’ Conference, pp 127-135 (The
giving permission for this data set to be analysed by Metso
Australasian Institute of Mining and Metallurgy: Melbourne).
Minerals and for reviewing the analysis results.
Vera, M A, 2002. A touch of froth: How bubble–particle aggregates take
the strain, PhD thesis (unpublished), University of Queensland,
REFERENCES Brisbane, Australia.
Bonfazi, G, Serranti, S, Volpe, F and Zuco, R, 2001. Characterisation of Zheng, X, Franzidis, J-P and Manlapig, E V, 2004. Modelling of froth
flotation froth colour and structure by machine vision, Computers transportation in industrial flotation cells. Part I: Development of
and Geosciences, 27:1111-1117. froth transportation models for attached particles, Minerals
Engineering, 17:981-988.
Gonzalez, R C and Woods, R E, 2002. Digital Image Processing, Second
edition (Prentice Hall, Incorporated: New Jersey).
single level sensor and a pair of level control valves located on commissioning, stabilisation and optimisation. The duration of
the discharge of the second tank in series. Each cell has an each of these phases will depend on the complexity of the
individual air injection control valve and froth analyser (camera) process and the available technical resources to drive the
to measure the froth behaviour on top of the flotation cell. improvement.
The copper rougher feed and copper rougher tails streams are The Telfer metallurgical team had outlined a simple
sampled and analysed by multi-stream analyser (MSA) to approach in line with the classical philosophy during the
determine metal grades. pre-commissioning phase of the project, which maximised the
use of available personnel. These steps for flotation were
The copper cleaners and scavenger cleaners outlined as:
The objective of the copper cleaners is to upgrade the copper • Commission – ensure all instruments are functioning and
rougher concentrate to generate a product sufficient for the critical control functions operate. This was accomplished
recleaning circuit. The objective of the scavenger cleaners is to through a series of factory acceptance testing (FAT) and site
minimise losses of valuable mineral once it has reported to the acceptance testing (SAT).
cleaning block.
The concentrate from the copper roughers is pumped to the • Stabilisation – commenced with simple field loop tuning of
copper cleaner circuit. Each cleaner and scavenger cleaner cell the circuit. This was followed by circuit characterisation,
has individual level and air injection control. The copper rougher which determined key upstream and downstream influences
concentrate and the cleaner scavenger tail concentrate and tails on major components.
are sampled and analysed by the MSA for metal grades. • Optimisation – firstly develop a feed forward responsive
control to process disturbance, followed by application of the
The copper recleaners froth cameras and lastly incorporation of the MSA to control
product quality.
The purpose of the recleaners is to bring the grade of the copper
concentrate to the required market specification. The recleaners After the commissioning of both processing trains had been
consist of three cells linked together sharing one level sensor. successfully completed and the stabilisation process begun,
The air injection control has one flow controller and three Manta Controls were invited to work with the Telfer
manual valves allowing the operator to manually adjust the split metallurgical team to continue the stabilisation and optimisation
of air between the three cells depending on the desired pull-rate phases of the process control philosophy.
and copper grade being produced. The recleaner concentrate is The Manta Controls hierarchy of controls identifies all the
also sampled through the MSA for metal grades. elements required to achieve the objectives of the optimisation.
The hierarchy of control is similar to a pyramid; if one of the
The pyrite roughers lower level elements is not available the final objective may not
The objective of the pyrite roughers is to maximise the recovery be achieved. (see Figure 3).
of pyrite concentrate from the ore. Their design is the same as This version of control hierarchy was natural fit with Telfer
the copper roughers. The concentrate from Train one’s pyrite Process control philosophy and adapted as a logical way for the
roughers goes to the pyrite cleaning circuit, which is not two parties to effectively interface.
discussed in this paper. Train two’s concentrate is feed to the
copper cleaners. LEVEL 1 – FINAL CONTROL ELEMENTS
PROCESS CONTROL PHILOSOPHY – TELFER The first step at Level 1 is to audit all available instrumentation
operation, electrical and electronic performance to ensure that it
When commissioning a new plant, classical process control is performing within specification. This is time consuming but
philosophy is made up of several key phases, these being; essential before any higher level control can be performed.
Copper Copper
recovery grade
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
This was then followed by critical circuit parameter To resolve this issue a feed forward signal has been added
assessment, which involved system step testing and field from the first cell’s level controller to the second cell’s level
response measurements. Based on this work the first element at controller and so on through the whole rougher circuit. In this
Level 1 that needed to be addressed was the response of dual dart way any disturbance on the first cell will not have any impact on
valves on each flotation cell. It was determined that their any of the subsequent cells, as the amplitude of the upstream
performance can be improved as follows. disturbance is adjusted in the downstream level controller.
The improvements in the level stability are shown in Figure 4
Dual valve control – split range and Figure 5. The trends are of the levels of two adjacent
There is a small amount of dead band in the positioners, making flotation cells during a normal start-up procedure. Both cell’s set
it difficult for the dart valves to be positioned at better than points are changed by the same amount, but the second cell
one per cent. To minimise the effect of dead band, the first valve experiences a greater disturbance caused by the first cell’s
opens from zero to 100 per cent as the control signal changes change in throughput. After the stabilisation, split range and
from zero per cent to 50 per cent and the second valve opens valve linearisation were implemented the cells experienced a
from zero per cent to 100 per cent as the control signal changes smaller disturbance even though the step change to the set point
from 50 per cent to 100 per cent. This method halves the effect was 50 mm greater than the previous trend.
of dead band.
Cell flow coordination – froth production rate
Valve linearisation
The cell flow coordination can only be applied to flotation cells
The trends of the level controllers displayed a non-linear that are equipped with a means to measure froth rate being
response due to the shape of the dart and the seat. This was very ‘pulled’ (produced) by the flotation cells. This is currently
pronounced at outputs of 40 per cent to 60 per cent where the configured on the rougher circuit.
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 Froth analyser (camera) data
control elements throughout each controller’s range. The froth analyser provides a lot of information that can be used
The combination of both dual valve split ranging and valve for control and includes:
linearisation has enabled the flotation circuit to perform in a
• bubble X and Y velocity and vector;
consistent stabilised fashion with effective response to level
deviations. • bubble colour made up of three signals – red, green, blue
(RGB);
LEVEL 2 – REGULATORY CONTROL B • statistical bubble size information;
• bubble load;
Cell stabilisation
• bubble stability; and
If the level control valve in an upstream cell opens up, it allows
more slurry into the downstream cell, thus increasing that cell’s • picture brightness.
level, which in turn opens the second cell’s level control valve. The quality of the information was assessed to identify what
This means a disturbance in the first cell will be seen in each information could be useful for control and six weeks of data
cell thereafter. Depending on the individual cell tuning the was sufficient to perform the assessment. By establishing
disturbance, more often than not, is amplified. This was correlations between plant performance and the froth analysers’
particularly noticeable during plant start-ups where the operator data, the most suitable information for control purposes was
would use manual control to prevent overflows. identified.
500
Cell 1 SV mm
Cell 1 PV mm
Cell 2 SV mm
400 Cell 2 PV mm
300
200
100
2/06/05 11:50 2/06/05 11:59 2/06/05 12:08 2/06/05 12:17 2/06/05 12:26
500
Cell 1 SV mm
Cell 1 PV mm
Cell 2 SV mm
400 Cell 2 PV mm
300
200
100
17/11/06 21:20 17/11/06 21:29 17/11/06 21:38 17/11/06 21:47 17/11/06 21:56
The bubble velocity in the X axis proved to be a reliable roughers, but not the grade of the concentrate produced. If more
measurement and was representative of the flotation cells’ concentrate is being pulled, the hopper level rises and the pump
performance. The bubble velocity provides an indicative figure speed increases to maintain the level. This is not an exact
of the mass going over the lip of the flotation cells. A weakness measurement, but can be used to prevent the roughers pulling
of the bubble velocity measurement is that it does not take into more concentrate than the pump can cope with and ensure that
account the amount of gangue that is entrained in the froth the roughers are not pulling too little concentrate. However, the
reporting to the concentrate launder. weakness of this control is that the grade of the product is not
The froth from all the copper rougher cells is fed into a single known even though consistent mass is achieved.
hopper and then pumped to the copper cleaner circuit. The The hopper pump speed is used as a master signal to tell all the
hopper pump speed is used to control the hopper’s level and also bubble velocity controllers to pull a little harder and reduce the
provides a good indication of the mass of concentrate being froth depth if the pump speed is below set point, or slow down
pulled off the roughers. By combining the bubble velocity from the total pull rate by increasing the froth depth if the pump speed
each cell with the pump speed it is possible to get an indication is above set point.
of concentrate mass production.
LEVEL 3 – REGULATORY CONTROL A
Pull rate control
Disregarding the flotation reagent effects within the cell for the Cell flow coordination
sake of simplification, there is a direct correlation between Each pair of rougher cells shares a level controller. Without
decreasing the froth depth (raising the level of the slurry) and the operator intervention, in most cases, only one cell per bank ends
bubble velocity of the froth going over the lip of the flotation up pulling concentrate. To ensure each cell in the bank is pulling
cells. Therefore the shallower the froth depth the faster the pull the same quantity of concentrate the cells must be balanced. The
rate. To control the pull rate of each cell, the bubble velocity is only way to do this is to balance the pull rate between each cell
used to determine the required level. using air addition.
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 Balancing the twin cells
bubble velocity over the flotation cell’s lip will vary depending
on the load of the bubbles and the amount of gangue that is One cell can end up pulling a lot harder than the other for a
entrained with the bubbles. The copper rougher hopper pump number of reasons:
speed gives a good indication of the mass being pulled by the • the air injection balance between the two cells,
• the percentage mineral in the first cell will be slightly higher Pulp chemistry pH control
than the second cell thus forming a more stable froth bed,
The pH is modified in the Telfer circuit to assist in the selective
• the flow rate of slurry through the cells tends to raise the flotation of copper sulfides from pyrite and to reduce the effects
level in the first cell with respect to the level in the second of oxidised species on the pulp chemistry. The pH is controlled
cell, and via two controllers, one that adds quick lime to the feed of the
• relative hydraulic head differentials between paired cells grinding circuit and one that adds milk of lime to the copper
with upstream and downstream effects. 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
Using the difference in bubble velocity between the two cells
alone cannot achieve a steady balance. Understanding why the to respond to the long-term changes (step addition) in pH. These
two cells became unbalanced enabled the implementation of one loops were decoupled to prevent the loops oscillating and the
part of the Manta Froth Cube that uses multiple inputs and one quick lime addition was linked to the mill feed rate. By
output as listed in Table 1. decoupling the two control loops the pH control improved during
a mill start-up as shown in Figure 6 and Figure 7. The figures
This part of the Froth Cube balances the two cells effectively
indicate the relationship between the mill feed rate and the
and ensures both cells continually pull concentrate.
copper roughers’ pH.
In Figure 7 the mill start-up was not as smooth, yet the copper
TABLE 1 roughers’ pH was more stable.
Twinned cells balancing control elements.
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
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
pH control showed the copper roughers tails mean reading improved from
0.078 per cent Cu to 0.073 per cent Cu and the standard deviation
The stabilisation of the pH has also contributed to better improved from 0.017 per cent Cu to 0.007 per cent Cu.
recovery.
The copper rougher concentrate grades shown in Figure 12 and
Figure 13 indicate that the standard deviation in copper rougher
Time to start-up
concentrate has also decreased.
The time taken to get the rougher cells to recover effectively After optimisation the concentrate grades are more consistent.
during start-up has also been reduced. The concentrate mass being pulled from the roughers has also
Figure 9 shows start-up of the roughers after being optimised. increased. Data taken over a 30 day period before and after the
The flotation froth depths upon start-up are at 400 mm, then drop optimisation indicated the concentrate mass recovered by the
to 350 mm and are controlled from there on. The flotation depths copper roughers increased by 23 per cent. The results indicate
reach their optimum operating levels within two hours of start-up; that with the control hierarchy in place, the copper roughers
however, concentrate is produced within 30 minutes of re-start. operate with a higher level of repeatability in their performance.
By looking at the rougher concentrate hopper pump speed, we The higher mass of concentrate delivered to the cleaner circuit
can see that with the automated start-up sequence control has ensured that the cleaning circuit is operating at maximum
(Figure 9) significantly more production is realised than in the capacity.
manual mode (Figure 8) of operation. During the optimised
start-up the flotation froth depths are continually being adjusted
Key benefits
to maximise the mass pulled from the roughers. This appears
more unstable than the manual start-up but is achieving the The added benefits of the stabilisation have been:
objective of maintaining the hopper pump speed around 75 per
cent. The time taken to reach the optimum level has been • the flotation circuits are easier to operate;
reduced by at least 50 per cent with the optimised control. • the operators can spend less time monitoring the flotation
MSA samples of the concentrate and tails grades of Train 2 cells enabling them to spend more time on grade reagent
copper roughers were analysed over a three-day period. These control, thereby enhancing production further;
results were plotted on a histogram. Figure 10 and Figure 11 show • more predictable flotation circuit performance; and
the copper rougher tail grade results are more consistent after the
optimisation. The statistical analysis of the before and after data • 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
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
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.
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
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.
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
Final Concentrate
FIG 2 - Schematic representation of the previous flotation circuit flow sheet.
The circuit consists of five parallel banks of roughers/ The cleaner circuit provides multiple stages of cleaning to
scavengers. Each bank has ten Wemco 130 m3 flotation cells minimise entrainment and maximise final concentrate grade,
with the first two cells acting as roughers and the remaining eight especially for the scavenger concentrate. Froth washing is
cells as scavengers. The residence time in the plant applied to the final stage of cleaning to further reduce entrained
rougher/scavenger bank is normally greater than 25 minutes, particles in the concentrate.
with a rougher residence time of approximately five minutes. The final tailings of the cleaner circuit are returned to the head
The entire rougher concentrate stream is reground in a tower of the rougher circuit. The recycle provides an opportunity to
mill (known as the polishing mill at Batu Hijau) before being fed recover valuables in the cleaner tailings when there is a flow
to the cleaner circuit. There is no classifier in the polishing mill surge through the cleaner circuit and/or the cleaner circuit is not
circuit. The polishing mill product P80 can vary between stable. Open circuiting this tailings stream can be an option with
70 microns and 90 microns depending on the feed size (80 - high pyrite ores.
120 microns), stream flow rate (150 - 300 m3/h) and density (30 - The initial design targets for copper recovery in the roughing/
40 per cent solids). scavenging stage and the cleaning stage were 91 per cent and
A separate regrinding circuit consisting of two tower mills and 98 per cent, respectively. The overall plant recovery of copper
a hydrocyclone cluster (known as the regrind mills at Batu Hijau) from fresh feed was designed to be 89 per cent.
is set up for grinding the scavenger concentrate. Scavenger
concentrate reports to the hydrocyclones so that only the coarse INITIAL FLOTATION CIRCUIT PERFORMANCE
fraction from the hydrocyclone underflow is reground. The
closed circuit arrangement is designed to produce a final Figure 3 shows copper recovery in the individual rougher,
grinding product P80 of about 45 microns. scavenger and cleaner circuits as well as the overall plant
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
100%
TABLE 3
95% Summary of column design criteria (all data referring
Overall plant recovery
90%
to one column only).
Final Tailings
Feed Cleaner
Regrind 1st Cleaner
Scavenger
Cyclones
Polishing Rougher
Mill Cleaner
Regrind
Columns 3rd Cleaner
Mills
2nd Cleaner
In Figure 8, there appears to be an improvement in the overall It has also been observed in the plant that the two additional
plant copper recovery after the column commissioning. However, columns tend to stabilise the cleaner circuit. The impact caused
statistical verification of the improvement has not been achieved by disturbances from the rougher and scavenger circuits on the
for this limited data set due to the relatively large data dispersion.
100
In Figure 9, the improvement in final concentrate copper grade
Overall Plant Copper Recovery (%)
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.
35
35
30 30
25 25
20 20
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.
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
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 circuit while allowing multiple stages of cleaning. This will help
recovery of copper in the cleaner circuit not only increased after minimise entrainment of the fines to ensure a reasonably high
the column commissioning in September, 2006 (from average grade of final concentrate from the scavenger concentrate.
93.9 per cent in August to 96.1 per cent in October) but also It may also be worthwhile to consider implementation of
became more consistent (from 3.8 per cent standard deviation in additional fines flotation at the end of the cleaner circuit. Given
August to 2.2 per cent in October). Operators find the columns that already high stage recoveries are achieved in the cleaner, this
and the new cleaner circuit easier to control. may not, however, be cost effective and more work is required to
With the two column cells online, it is now possible to carry prove this possibility. An alternative example of a possible
out maintenance work on the second and third cleaner banks or flotation circuit configuration to achieve this objective is given in
the first cleaner and cleaner scavenger banks without shutting Figure 11.
down the entire flotation circuit. The current circuit
configuration allows the first cleaner concentrate to be diverted Finally, it is worth mentioning that although cleaner scavenger
to the columns or the scavenger concentrate to the second cleaner tailings often have a similar grade to plant fresh feed at Batu
during partial cleaner circuit maintenance services. 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
FUTURE IMPROVEMENT PROGRAM particles most likely come from entrainment from the rougher
Commissioning of the two columns as rougher cleaners is the and scavenger circuits or are newly generated from regrinding.
first step of the current optimisation program for the flotation They will be difficult to float again in the rougher/scavenger
circuit at Batu Hijau. Based on studies on ore characteristics and circuit. Moreover, concentration of floatable gangues such as
current plant performance, additional circuit modifications are pyrite is significantly higher in the cleaner scavenger tailings.
being considered. These include: Recycle of cleaner scavenger tailings in this instance could result
in generation of unstable re-circulating load and have a
1. Send the concentrate from the existing first two scavenger detrimental effect on rougher flotation. It is proposed, in future
cells, together with the rougher concentrate, directly to the circuit modifications, to allow provision to open the cleaner
column cleaners. Copper minerals are approximately 92 per scavenger tailings to final tailings to avoid these potential issues.
cent liberated in the first two scavenger concentrates,
compared to 97 per cent liberated in the rougher concentrate. CONCLUSIONS
2. Combine the column tailings with the scavenger Two additional column cells were installed as single-stage
concentrate and regrind them together in the existing rougher cleaners at Batu Hijau in early September 2006. The
regrind circuit. The average copper concentrate grades in new cleaner circuit allows the relatively coarser, more liberated
the column tailings and the scavenger concentrate are rougher concentrate to be cleaned in a single stage separately
8.5 per cent and 12.8 per cent, respectively. from the finer, lower grade scavenger concentrate. Plant results
3. Feed the combined scavenger concentrate and column show improvements in both overall copper recovery and final
tailings after regrind to the second cleaner bank instead of concentrate copper grade since the column commissioning. The
the first cleaner bank. measured 1.38 per cent improvement in the final concentrate
copper grade is proved to be statistically significant. Based on
4. Discard the cleaner scavenger tailings directly to the final the working carrying capacity determined in the operating
tailings. columns, there is a clear indication that the current cleaner
The aims of the above modifications are to maximise recovery circuit will provide sufficient capacity to maintain recovery
of the well liberated valuable particles in the rougher circuit and during high mill throughput and feed grade periods.
avoid over-grinding these particles before the column cleaning In addition to metallurgical improvements, operability and
stage. Simultaneously, the circuit will provide sufficient grinding stability of the cleaner circuit have greatly improved after the
to liberate the valuables from the composite particles in the column commissioning. Moreover, the current cleaner circuit
scavenger concentrate and the column cleaner tailings to improve configuration is also more flexible for maintenance planning.
ore floatability and maximise both recovery and concentrate With the additional capacity provided by the column cells,
grade. there are opportunities for additional circuit modifications to
A key consideration of the ongoing flotation circuit further improve metallurgical performance in the flotation plant.
modification at Batu Hijau is to treat different groups of particles
with different properties (ie particle size distribution, degree of ACKNOWLEDGEMENTS
liberation and grade) separately in appropriate sections of the
circuit. It is anticipated that this will provide sufficient residence The authors would like to acknowledge the involvement of Enzo
time for relatively slow-floating, liberated fines in the cleaner Artone, Eugenio Gatto and Dr Hideyuki Okamoto at various
Final Tailings
Open Circuit
Closed Circuit
Cyclones 2nd Cleaners
Cyclones
Cleaner 1st Cleaners
Scavengers
Polishing
Mill
Regrind
Column Mill
3rd Cleaners
Cleaners
Final Concentrate
Roughers 1-4
Scavengers 5-10
Final Tailings
Middlings
Re-Cleaner
Fines
Re-Cleaner
Middlings
Regrind
Fines
Regrind
Final Concentrate
stages of the project. The authors would also like to thank Nengah Gatto, E, 2004. Flotation flowsheet options, PT Newmont Nusa Tenggara
Dwi Payana for his assistance during the column commissioning. internal report, 16 November.
Halomoan, R P, 2006. Laboratory batch flotation tests, grind-recovery
study, October 2005 – February 2006, PT Newmont Nusa Tenggara.
REFERENCES McCaffery, K, 2005. Cleaner circuit upgrade project, PT Newmont Nusa
Burger, B, Hatta, M, McGaffin, I and Gaffney, P, 2006. Batu Hijau – Tenggara internal report, February.
Seven years of operation and continuous improvement, in
Proceedings SAG 2006 Conference, Vancouver, British Colombia,
Canada, 23 - 27 September.
Zn Rougher
Zn Scavenger
Zn Circuit Feed
Final Tail
Zn Cleaner 1
Zn Cleaner 2-5
Final Tail
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
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.
70
500
60
Net Energy Comsumption, kWh/t
400
300
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
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
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
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
100
90
0 80
70 355 kW Detritor with 1 mm Colorado sand
Net Energy Consumption, kWh/t
CONCLUSIONS REFERENCES
The plant surveys at the Zinifex Century Mine indicated that the Burgess, F, McGuire, I and Willoughby, R, 2001. Operation of sand mill
Detritors in the ultra-fine milling circuit were working well. The Detritors at Pasminco Operations, Fine Particle Processing and
mills produced a product P80 of 7.5 microns with a gross energy Tailing Summit, Perth, 2 - 4 July.
consumption of 37.2 kWh/t. However, the Detritors in the regrind Enderle, U, Woodall, P, Duffy, M S and Johnson, N W, 1997. Stirred mill
circuit performed poorly. The mismatch of the medium size to technology for regrinding McArthur river and Mount Isa zinc/lead
the feed size wasted up to 30 per cent of regrind power and ores, in Proceedings XX International Mineral Congress, Aachen,
produced a broad product size distribution that adversely affected Germany, 21 - 27 September, pp 71-77.
flotation recovery and concentrate grade. 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
In an effort to improve the efficiency of the regrind Detritors at (Mount Isa Mines Ltd) and the Mining Cycle, in Proceedings
the Century Mine, coarser 3 mm ceramic media were tested in a AusIMM ’98 – The Mining Cycle, pp 291-298 (The Australasian
laboratory Detritor as an alternative to the 1 mm media used in Institute of Mining and Metallurgy: Melbourne).
the plant. The former significantly reduced the energy Lichter, J K H and Davey, G, 2002. Selection and sizing of ultra-fine and
consumption for processing regrind Detritor feed and produced a stirred grinding mills, Mineral Processing Plant Design. Practice,
narrower product size distribution. and Control (ed: A L Mular) (Society for Mining, Metallurgy and
Scale-up of the ultra-fine milling Detritors was investigated by Exploration: Littleton).
testing the same feed material in a 0.55 kW batch Detritor and a Lofthouse, C H and Johns, F, 1998. The ECC International Stirred Media
355 kW full-scale mill using 1 mm sand media. The results Mill and the Metals Industry, Comminution 8, Brisbane, 6 - 8
indicated that the 0.55 kW batch Detritor underestimated the February.
energy consumption of a full-scale Detritor by about 20 per cent Parry, J, Klein, B and Lin, D, 2006. Comparison of ultra fine grinding
for a product P80 of seven microns. technologies, in Proceedings XXIII International Mineral Processing
Congress, Istanbul, 3 - 8 September, p 177.
Smith, C, 1999. Svedala Stirred Media Detritor, IIR Conference Crushing
ACKNOWLEDGEMENTS and Grinding ’99, Perth, 29 - 30 March.
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.
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
FIG 1 - A diagrammatic representation of the hypothetical difference between life optimised liners (left) and profit optimised liners (right).
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.
Larger mill relining systems for mills in the of writing (October 2006) RME has supplied 107 ‘liner handlers’
16’ - 40’/5 m - 12 m diameter range liners of which 25 are Mill Relining Machines of 3500 kg/7700 lb rated
2200 - 16 500 lb/1000 - 7500 kg 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
Much of RME’s work has tended toward larger mill operators, 7250 kg/15 950 lb rated capacity.
a market which demands high capacity, high performance Large liners offer every advantage. RME’s Mill Relining
mill relining equipment. RME’s worn-liner-removal, new-liner- Machines can place (say) 5000 kg liners with the same speed as
placement Mill Relining Systems have been unique in halving the 1000 kg liners. Liner placement rates (m2/hr) are therefore
time taken to reline large AG, SAG and Ball mills, worldwide. improved significantly with larger liners. Also, larger liner
Significantly, RME’s Mill Relining Systems have changed placement is easy and efficient. Edge alignment of their vast
plant management expectations regarding mill relining times, dimensions assures bolt hole alignment between liner and mill
resulting in shorter whole plant shutdowns and significantly shell for easy bolt insertion.
increased annual plant utilisation.
A typical, large-plant RME Mill Relining System consists of: LINER PROFILES TO MAXIMISE PROFITABILITY
• powered feed chute extraction and transportation;
Mill liner functions
• three axis liner suspension mill liner handler or, much more
commonly; Mill liners have two functions:
• a seven or eight axis liner placement Mill Relining Machine; 1. The first function is to protect the mill’s structural shell
and from destructive abrasion. Here, the key performance
• Thunderbolt recoilless hydraulic hammers. indicator is cost of life, ie for how long will a set of liners
safely grind ore and protect the expensive mill shell from
RME’s large-capacity, liner-placement Mill Relining Machines damage and what is the cost of that protection. These cost
have provided the means to apply very large liners. At the time 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.
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 The ‘long life liner’, ‘maximum annual mill utilisation’,
detailed quantitative analysis. Liner development, on site, tends ‘minimum annual liner cost’ mindsets are further reinforced by
to follow the simpler routes of ‘wear life’ maximisation, ie historical fears of mill relining work. This fear is now unfounded
responding to the ‘attractive’ KPIs. in practice because of the widespread adoption of RME’s fully
This ‘wear life’ maximisation (and the associated annual ‘liner mechanised worn-liner-removal, new-liner-placement Mill
cost’ minimisation and annual ‘utilisation maximisation’) thought Relining Systems.
processes are common to both operating and maintenance For maximum metallurgical performance and maximum
personnel. profitability, the appropriate question and associated KPI should
Cost pressure to maximise liner life (minimise annual liner be: ‘what is the optimum liner design and liner exchange
costs) has a historical basis due to long periods of low program for maximum profitability?’
commodity prices. These commercial drivers have been ‘KPIs = ‘grinding performance’, ie tonnes processed, metal
reinforced over the last few years due to the marked increase in recoveries, profitability’. These are the complex and unattractive
iron and alloying prices. Mill liners, with their high chrome and KPI’s mentioned earlier in this paper.
molybdenum content, have been particularly affected. The key to answering this question, at minimum risk, may lie
Taken to the extreme, ‘cost reduction’ and ‘maximum in a combination of new technologies, now discussed.
utilisation’ focii on milling operations can lead to one or more of
the following liner-related, grinding problems:
NEW TECHNOLOGIES AND A ‘LINER PROFILE
• reduced mill volume through large, thick (long lasting) VERSUS PROFITABILITY’ RELATIONSHIP
liners;
As discussed above, the inter-relationship between grinding
• reduced mill volume through ‘packing’ of ore material performance, liner cost, downtime cost, revenue and profitability
between deep liner lifters; are complex.
• (packing can be loose, which falls out when the mill stops, or Fortunately, a range of new technologies now exist which
cemented by clays, which remain once the mill stops); should allow research organisations, independent suppliers or
• individual liners featuring long lives but with no liner life motivated mine sites to develop a commercial model, which can:
being a multiple of another, eg grates – four months, shell • find the optimum relationship between grinding performance
lifters – five months, f/e outer row – seven months, etc (liner profile, liner life, relining time, mill throughput,
(sensible, regular, planned maintenance becomes impossible); comminution effectiveness and recoveries); and
• insufficient mill power draw at some time during liner life; • commercial performance (liner cost, reline cost, downtime
• ‘over-throwing’ of balls beyond toe of charge (particularly in cost, metal price, revenue and profitability).
fixed speed mills) leading to ball and liner damage, slow and This envisaged commercial model must yield a ‘liner profile
dangerous relines (oxy-lancing) and, in the extreme, mill versus profitability’ relationship for a particular grinding plant.
shell damage (‘flowing’ liners which ‘spring’ the shell at Optimum liner profile may also prove to be a function of current
structural joints); and commodity price.
• reduction in mill speed and power draw with new liners, RME believes there is a ‘perfect’ liner profile for each grinding
necessary to avoid ‘over-throwing’ of balls. plant and that taken to the extreme level of understanding (a very
These effects, in combination, reduce mill performance with accurate commercial model), the liners’ optimum profiles are
new liners to as little as 80 per cent of maximum grinding affected by the current commodity price.
performance. A casual conversation with a customer recently The new technologies, and the prospect of combining these to
indicated 77 per cent of maximum capacity on start-up, that generate the ‘liner profile versus profitability’ relationship, are
maximum occurring at the end of shell liner life. especially exciting in these times of high commodity prices.
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
FIG 8 - Solidworks 3D image of RME’s Thunderbolt StormPath. 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.
FIG 9 - RME’s (and the world’s) largest capacity Mill Relining Therefore, for each unique circumstance, RME believes there
Machine featuring 7250 kg/16 000 lb liner capacity, liner elevator, is a perfect liner profile solution, certainly obscured through the
diesel engine, 4 × 4 drive and remote radio control supplied to PT data ‘noise’ of a practical operation, but a solution which delivers
Freeport, Indonesia for 34’ and 38’ SAG mills. the best economic result for that operation.
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. 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).
The lower aspect ratio pilot mill therefore appears to trade T80 = 80 per cent passing size of AG/SAG mill circuit
throughput for grind size. Hence, if full-scale mills behave in a
similar manner it can be inferred from these results that for a DWi = drop-weight index
given ore a high aspect ratio mill will provide a higher J = volume of balls (per cent)
throughput at a coarser grind than a low aspect ratio mill
operating with the same power draw, per cent ball load, per cent φ = mill speed (per cent of critical)
filling, per cent of critical speed, grate configuration and open f(Ar) = function of mill aspect ratio
area fraction. The data further indicate that from an energy
utilisation efficiency viewpoint there is no significant difference a,b,c,d,e,f,g = constants
between aspect ratios, as the operating work indices were the
K = function whose value is dependent upon whether a
same for both the high and low aspect ratio tests. Adam and Hirte
pebble crusher is in circuit
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: W = MiK x2 ( f (x 2 )
− x1
f (x 1 )
) (3)
... (the) kWh per ton of minus 200 mesh material
produced is essentially the same in the long mill where:
as the short mill. W = specific energy (kWh/t)
K = constant chosen to balance the units of the equation
FULL-SCALE MILL DATA
Mi = index related to the breakage property of an ore (kWh/t)
The value of pilot tests such as those described in the previous
section is that aspect ratio can be studied whilst keeping all other x2 = 80 per cent passing size for the product
variables constant, therefore obtaining unequivocal results x1 = 80 per cent passing size for the feed
concerning its influence. In particular, feed ore characteristics are
kept constant. With full-scale mills such control is not usually The predicted specific energies shown in Figure 4 were
possible and hence comparisons are made difficult by having to calculated using Equations 1 - 3, in which there is a function
use data from different sites with widely different operating related to aspect ratio, though there are no parameters which
conditions on top of differing aspect ratios. The solution to this represent different energy efficiencies for particular mill designs
problem revolves around finding a data reduction process that is or operating conditions, ie it is assumed that the energy
able to remove the influences of all relevant variables so that the efficiency is the same for all mills/circuits. The correlation is
impact of aspect ratio can be studied in isolation. Such a data very good, error analysis showing that there are no underlying
reduction technique can be found in Equations 1, 2 and 3. These biases. In contrast to this result Figure 5 shows the correlation
equations are used by SMCC Pty Ltd to estimate the specific using a similar equation to Equation 1 but with no allowance for
energy of AG/SAG mills under all normal operating conditions, aspect ratio. The results are very different; the scatter being quite
eg ball load, mill speed, feed size and ore hardness and for all pronounced in places, illustrating that aspect ratio has a
normal aspect ratios and mill sizes (Morrell, 2006). The significant influence with respect to the specific energies of the
equations are semi-empirical and are based on data from 43 mills in the database. The typical response of this relationship is
different circuits and 70 different ore types. The range of shown in Figure 5 and can be seen to work in a similar manner to
design/operating parameters covered by these data is shown in the way that the Adam and Hirte data indicate occurs in pilot
Table 1. mills.
45
TABLE 1
40
Range of circuits/ores.
35
Max Min
observed kWh/t
sg 4.63 2.5 30
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)
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
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
observed kWh/t
single stage. The aspect ratio relationship is only relevant to
open circuit mills where throughput and grind size vary in 25
association with one another. In single stage mills grind size is
20
independent of aspect ratio as it is controlled by the action of
the classifier. Specific energy is also controlled in one sense by 15
the classifier, in that the P80 it produces dictates the energy
requirement, in conjunction with the throughput. Looked at 10
without aspect relationship
another way, in a single stage high aspect ratio mill its initial 5 with aspect ratio relationship
tendency (ie when feed is first introduced into the mill circuit)
will be for it to give a higher throughput and coarser discharge. 0
0 5 10 15 20 25 30 35 40 45
However, due to the action of the classifier the coarser part of predicted kWh/t
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 FIG 7 - Observed versus predicted specific energy for single stage
come to equilibrium. If the mill were low aspect ratio the initial AG/SAG mills; with and without an aspect ratio term.
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 PILOT DATA
feed would be less than in the high aspect situation.
Equilibrium would eventually be reached at exactly the same Although the Adam and Hirte (1973) test work used a variable
throughput and power draw, providing the final grind was the length pilot mill, such installations are very rare and in most
same as that obtained in the high aspect ratio case. However, in cases a standard 6 ft × 2 ft (D × L) mill is used for pilot studies.
the low aspect ratio case it can be expected that the recycle load Due to our much improved understanding of how AG/SAG mills
would be lower. Overall energy efficiency would be the same, work and in particular how to characterise feed ore, pilot test
as the Adam and Hirte data indicate should be the case. The only programs are not as common as they used to be and hence data
situation where this similarity in overall performance may not are becoming increasingly scarce. However, over the last six
occur is where a very low aspect ratio mill was chosen in single years the author has collected data from nine pilot studies where
stage mode to grind to a relatively coarse grind. It is possible that subsequently a full-scale plant has been built. Data were then
by the time the feed were discharged from the mill it would have collected from these plants and compared with the pilot data. In
been overground to the point where the relatively coarse target the context of this paper, the value of doing so is that the pilot
P80 could not be achieved by the classifier (it would only be able tests were carried out using 6 ft × 2 ft mills which have an aspect
to obtain a P80 finer than the target). The overall energy ratio of three. Most full-scale mills tend to have aspect ratios of
efficiency would be no different; the mill would simply be two or lower and hence it is possible that in comparing the pilot
unable to obtain its desired target grind and have a higher and full-scale data aspect ratio effects may become apparent.
specific energy than the high aspect ratio alternative. As a first approach the raw pilot net kWh/t was plotted against
To illustrate the single stage situation further, the single stage the raw full-scale kWh/t at the pinion (Figure 8). To arrive at this
mill data were separated from the other data in the database. latter figure power at the pinion was assumed to be 93.5 per cent
A wide range of aspect ratios are included in this data set (from of the motor input power. The data indicate a definite correlation
0.5 - 2). Equations 1 - 3 were then used both with and without though with the full-scale values being on average 21 per cent
an aspect ratio relationship to predict the specific energies of larger than the pilot ones. Scatter in the data is also apparent and
each of the mills. These were then compared to the observed is attributed to the following causes:
values (Figure 7). As can be seen there is no significant • the full-scale plant data did not always reflect exactly the
difference between the two approaches, further supporting the same ore as treated during the pilot study, though as far as
assertion that the energy efficiency is not a function of aspect possible full-scale plant data were collected during periods
ratio. where the ore was as similar as possible;
16 16
15 15
y = 1.2099x y = 1.0932x
Full scale kWh/t @ pinion
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.
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.
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
OPTIMISATION OF FEED END LINERS was converted from standard chrome moly material to a
BullnoseTM rubber white iron composite (Figures 6 and 7), based
Wear on the feed end liners was occurring at the join between the on trials in the 40 foot Cadia mill, with an expectation to double
inner and outer liners as shown in Figure 4, in line with the the wear life. These were installed in September 2006.
operating mill charge level.
White Iron
Rubber
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
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
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).
Fine
Fine Rougher
Scavenger
Fine Cleaner
Fine Cleaner
Scavenger
Plant Feed
Concentrate Tailings
Coarse Co Cleaner
Cleaner Scavenger
Coarse Coarse
Rougher Scavenger
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.
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
0.20 0.40
0.10 0.20
0.00 0.00
+75 +53 +CS2 +CS5 +CS6 +75 +53 +CS2 +CS5 +CS6
Following this laboratory development, a pilot plant was bank could be diverted to either final concentrate or the coarse
constructed and operated to fine-tune the design and provide the cleaner scavenger concentrate stream. This was done to allow for
necessary process design criteria and mill sizing to enable a changes in the residence time distribution between cleaner and
prefeasibility engineering study. The piloting was also necessary cleaner scavenger duty in the pilot plant. Figures 5, 6 and 7 show
to generate an improved estimate of the potential recovery/grade photographs of various components of the pilot plant.
improvement that can be gained form installing a regrind circuit Reagents were added to the pilot plant feed and mill discharge
at LNO. These were both necessary steps in developing a streams in similar quantities to those used in the full scale plant.
business case for the project.
The pilot plant was constructed from four banks of flotation Operation, surveying and mass balancing
cells and a regrind mill together with associated reagent dosing
and slurry transport pumps. The pilot plant was constructed in The pilot plant was operated on a single shift (day) basis. For each
such a way that it could replicate both the existing LNO coarse on/off trial, the pilot plant was allowed more than five residence
cleaning circuit as well as a post-regrind flotation circuit (fine times to reach steady state before a survey was conducted.
cleaner). For all surveys, timed samples of each individual concentrate
Figure 4 shows a flow sheet of the pilot plant as operated with stream as well as combined streams and feed and tail streams
the regrind mill in operation. The dotted lines in the figure show were collected, wet and dry weighed and assayed. This allowed
alternate stream destinations. for greater confidence in the mass balancing of results. Each
Figure 4 also shows that the pilot plant was set up in such a sample consisted of a full stream cut to avoid any representative
way that the individual cell concentrates from the coarse cleaner 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
Plant Co Ro Con
Co Cleaner Scav Final Tail
Plant Co Sc Con
Co Cl Co Cl Scav
Fine Cleaner
Final Con
16
15
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
35
30
25
MgO Recovery
15
10
0
0 10 20 30 40 50 60 70 80 90 100
Nickel Recovery
Regrind Re-Clean (No Regrind) Plant
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
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.
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).
This effect is often quoted by inert grinding mill manufacturers the ground to purge the mill of material remaining from the
as a benefit over conventional grinding, which would not previous run. Mill product was then pumped to the discharge
typically increase the slope of this relationship, resulting in the hopper as in the previous run with all parameters recorded and
production of more slime material. density and sizing samples taken. The procedure was repeated
Further grinding of this tails material will not produce a until the target grind size was achieved. The results from each
significant amount of recoverable nickel, but rather ultra-fine or pass were used to construct the signature plot for the concentrate.
slimes material (due to there being few composite particles Throughout the Isa Mill operation Colorado sand was added
remaining in the tails stream). Thus, it was concluded that a through the media addition port to maintain constant power draw.
regrind to 20 µm is adequate to enhance liberation.
Grindability
Regrind mill sizing
Several signature plots were constructed during the piloting
Feed was pumped from the pilot plant concentrate launders or campaign. Of most interest is the one shown in Figure 13, which
the LNO concentrator concentrate hoppers to the Isa Mill feed was conducted on the pilot plant coarse cleaner scavenger
tank. The feed pump to the mill was turned on and once slurry concentrate stream at two different densities (32 and 40 per cent
was observed discharging from the mill the mill was started. The solids).
feed rate adjusted using the feed pump speed and the valve Of importance, Figure 13 shows that grinding efficiently in the
recycling feed to the feed tank. Discharge from the mill was Isa Mill was not significantly affected when the pulp density of
either discharged into the floor drains or was pumped to the the feed stream was reduced from 40 to 32 per cent solids.
‘fines’ cleaner flotation pilot cells. At the discharge tank a During the open circuit trial, the pulp density of the coarse
calibrated ten litre bucket was used to measure slurry flow rate cleaner scavenger concentrate was measured to determine
through the mill. Discharge density from the mill was measured whether or not a high enough density was achievable to maintain
using a Marcy scale. grinding efficiency in a potential regrind circuit. The density of
While indicative power consumption plots were determined coarse cleaner scavenger concentrate during the open circuit trial
when the mill was in circuit with the pilot plant, variations in was 42 per cent solids.
plant operating conditions made it difficult to produce Isa Mill Table 2 lists the signature plot equations and cumulative
signature plots that were used to determine cumulative energy energy consumption results.
consumption data.
To produce an Isa Mill signature, plot feed from the pilot plant
TABLE 2
(or concentrator) was fed into the mill feed tank. The mill was
determined to be at operating density when the mill feed tank Isa mill energy requirements.
was at the same density as mill discharge slurry. The feed tank
was continually topped up. When the mill was at operating Test condition P80 signature plot Power required to
density, the product from the mill was retained in the discharge equation achieve a P80 of 20 µm
tank. The mill discharge flow rate was measured and the density (kWh/t)
monitored. The power draw and slurry discharge temperature 40% solids Y = 15 850x-2.4 35.0
were monitored continuously. Samples for laser sizing were 32% solids Y = 25 100x-2.2 36.7
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 For design purposes, the higher energy consumption of
with water. The contents of the discharge tank were then pumped 36.7 kWh/t was used to determine full-scale requirements.
into the Isa Mill feed tank. Once all the product had been Scale-up from pilot to full scale was completed on the basis of
pumped into the feed tank, the discharge tank was hosed out and energy requirement alone. The flow sheet development exercise
the pumping line cleaned with water. determined a nominal milling requirement of 20 t/h, grinding to a
The mill feed pump was then restarted and the Isa Mill started. P80 of 20 µm. Table 3 summarises the milling requirements
The first 45 seconds of product from the mill was pumped onto around these nominal design values.
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.
for their general condition or for damage. Figure 4 illustrates the combination. A series of display tools demonstrated and explained
greyscale inspection capabilities of the software. The greyscale in the following figures optimise the use of this method as a
display can also assist in interpreting the thickness data display decision-making tool for milling engineers or management.
by providing a realistic visual liner-element cross-reference to Milling operators or liner manufacturers may want to assess
the thickness displays, which are explained further below. For overall liner wear distribution or shape as represented by
example, certain blue areas in the thickness display in Figure 5 MillMapper thickness displays (Figures 5 - 9). Depending on the
are in fact lifter bars; red areas represent grate holes, etc. actual liner age at the time of scanning this information may be
The primary purpose and therefore functionality of MillMapper used to support liner design improvements. It is therefore
is to display high density 3D mill liner thickness information at instructive to look at the isolated location of liner segments as
the particular time the scanner data were acquired. MillMapper this provides a more specific indication of patterns of wear.
achieves this by displaying liner thickness as a colour range Thickness displays also enable milling engineers or management
appropriate for the type of liner in question. Feed, belly to monitor critical liner portions if they have a history of failure,
and discharge surfaces can be displayed individually or in 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.
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 the percentage of points that lie below a critical threshold. For
in its traditional 3D display. MillMapper therefore features a example, 0.2 per cent of the two million points covering the belly
separate representation of belly liner thickness by unwrapping liner from one particular dataset, ie 495 points, are below a
the cylindrical liner surface onto a 2D plane (see Figure 9). 30 mm threshold. Since the example liner has only worn through
half a life cycle, the majority of these points represent gaps
Reports between liner segments, where the scanner detected the outer shell
of the mill. Milling engineers or managers may want to define a
Aside from all graphical mill liner displays, MillMapper also reline criterion for their particular mills, such that whenever a
provides a range of numerical wear/thickness statistics. A global certain percentage of acquired data is output to be thinner than a
‘one number’ indicator that can be used to quickly assess wear is particular liner thickness a reline should be scheduled.
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.
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 5. The measurements are empirically more accurate than
faster than any alternative method of inspection. Data capture those of the other techniques.
does not take one to 1.5 hours as erroneously reported by 6. The MillMapper 3D digital mill liner wear model
Chandromohan and Powell (2006). is quality controlled through scientifically-proven
3. MillMapper samples liner thickness at several million points calibration procedures and rigorous statistical analyses.
within the mill (the entire visible mill surface can be 7. Mill linear wear is visualised by a series of colour-
measured). Other techniques can sample liner thickness in, at coded thickness maps and respective numerical 3D
best, a few dozen manually-selected locations. thickness data for the shell, feed and discharge ends.
4. The relative location of all point samples is accurately known. Other methods cannot provide this depth of
The other methods do not deliver this information. visualisation.
8. The MillMapper 3D digital mill liner wear model enables Currently, operators determine ball charge volume based on crude
operators to correlate liner shapes with other pertinent mill tape measurement surveys, which in many circumstances can not
data such as energy consumption or milling efficiency at provide adequate accuracy to achieve volume determination to the
respective critical liner life times (ie energy consumption nearest one per cent of actual volume, the accuracy desired to
spikes or poor milling efficiency due to a particular liner achieve consistent breakage rates and product size.
shape), and therefore facilitate liner design changes so as to Other recently-implemented innovations include improved
improve these other factors. This in turn has the potential to visualisation tools that allow users to interactively extract
facilitate significant cost savings. cross-sections of areas of interest from the data. MillMapper will
also feature automatic discharge-end grate hole sizing, a
QUALITY CONTROL AND ASSURANCE functionality which is expected to be available mid 2007. This
will enable sites that have a need for monitoring the expansion of
It is important to remember that the measurements from any grate holes caused by wear so as to control product size to
instrument (ultrasonic probe, mechanical profiler or laser determine change-out schedules based on grate hole sizes rather
scanner) are subject to both random and systematic errors. than grate plate thickness.
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, SUMMARY
we have developed in-house methods, error models and software
for their calibration (Lichti and Franke, 2005; Lichti and Licht, Originally conceived as a tool for thickness gauging of liners
2006). This is an absolutely critical component to the thickness inside cylindrical comminution mills, MillMapper is changing
gauging methodology described herein; without it laser scanners into a production optimisation tool. It features a number of
would not likely be useful for mill liner thickness gauging. By advantages over existing thickness gauging methods, including
using our self-calibration procedure, the overall accuracy safety, speed, accuracy, complete data coverage and far superior
improvement can be as high 80 per cent, allowing assessment functionality. More recent developments will
millimetre-level positioning to ranges up to 10 m. Further details facilitate mill operation optimisation. Quality control is very high
can be found in Lichti and Franke (2005) and Lichti and Licht and provided by rigorous instrument calibration procedures and
(2006). statistical analyses. As mentioned earlier, MillMapper is already
in use at a number of Australian sites including those of major
CURRENT AND FUTURE DEVELOPMENTS mining operators ALCOA, BHP Billiton, Barrick, Newmont, and
KCGM.
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 REFERENCES
production optimisation system. Chandromohan, R and Powell, M S, 2006. A structured approach to
As reported by Russell (2006), mill throughput can be modelling SAG mill liner wear–monitoring wear, in Proceedings
significantly improved through optimisation of new liner design SAG 2006, Vancouver, Canada, 23 - 27 September, Volume 3,
in combination with shortened reline scheduling. Contrary to pp 133-148.
what one might expect, a higher frequency of reline shutdowns Franke, J, Lichti, D D and Stewart, M P, 2006. MillMapper: A new tool
may not lead to higher combined loss of production, when for grinding mill thickness gauging, in Proceedings SAG 2006,
compared to longer reline scheduling if latest efficient reline Vancouver, Canada, 23 - 27 September, Volume 3, pp 75-87.
technologies are deployed (Weidenbach, Hart and Griffin, 2006) Lichti, D D and Franke, J, 2005. Self-calibration of the iQsun 880 laser
and liner design is at optimum simplicity. Currently mine sites scanner, in Proceedings Optical 3D Measurement Techniques VII,
can not reliably determine the most efficient shape of liners to Vienna, Austria, 3 - 5 October, pp 112-121.
achieve optimum throughput without MillMapper because no Lichti, D D and Licht, M G, 2006. Experiences with terrestrial laser
other comprehensive three-dimensional tool is available to scanner modeling and accuracy assessment, The International
provide this information. MillMapper not only provides the wear Archives of the Photogrammetry, Remote Sensing and Spatial
tracking for reliable shortened reline scheduling but also very Information Sciences, 36 (Part 5):155-160.
detailed liner shape information so as to facilitate the design of Napier-Munn, T J, Morrell, S, Morrison, R D and Kojovic, T, 1999.
new liners corresponding to the shape of a worn liner at its peak Mineral Comminution Circuits: Their Operation and Optimisation,
throughput performance. Currently peak throughput performance pp 413 (Julius Kruttschnitt Mineral Research Centre, Indoroopilly).
typically occurs towards the later stages of the liner life cycle, Rajamani, R K, Mishra, B K, Latchrieddi, S, Patra, T N and Prathy, S K,
not when the liner is new and, ideally, the mill should be most 2005. On the dynamics of charge motion in grinding mills, Paper
productive. presented to SME Annual Conference, Salt Lake City, 28 February -
2 March, 9 p.
Current MillMapper developments to facilitate optimum mill Russell, J, 2006. Advanced grinding mill relining for process
operation include automatic ball charge volume measurement metallurgists and management, in Proceedings SAG 2006,
and liner volume and therefore weight measurement so as to Vancouver, Canada, 23 - 27 September, Volume 3, pp 11-22.
provide expert control-system operators with the calibration data Weidenbach, M, Hart, S and Griffin, P, 2006. Application of a 6T liner
for feed rate and ball charge replenishment settings during handler for Cadia Valley Operations’ 40’ SAG Mill, in Proceedings
milling. These MillMapper output features are derived from the SAG 2006, Vancouver, Canada, 23 - 27 September, Volume 3,
same laser scan data collected to determine liner thickness. pp 59-74.
ABSTRACT 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
Mill liners provide the wear-resistant surface within grinding mills, they
also impart motion to the mill charge, ie the grinding process, by any special benefits from new initial shell lifter profiles may be
providing the key between the charge and mill shell, and, at the discharge maintained only for a period, unless shell lifters are of substantial
end, they remove the ground contents of the mill. As semi-autogenous size, as the shell lifter profile changes through wear.
grinding (SAG) mills have grown in size the process aspects of liner Over-throwing is usually to be avoided. However, high levels
design and their impact on mill performance have become particularly of overthrow (if the liners and balls can withstand the impacts)
important. While feedback from practical experience is critical to good results in rebounded balls entering the toe region through most of
mill liner design, increasingly sophisticated computer-based tools are the shell lifter wear life and with high levels of impact energy.
being developed to assist in mill liner design.
Hitting the shell just above the toe is the worst situation as the
ball-on-shell impact energies are highest, leading to shell and
INTRODUCTION ball damage with no compensating high-energy rebound into the
charge.
SAG mill liner design, especially the impact of liner design on
the process aspects of mill operation, is critical to good SAG mill Increasing the spacing between the liners allows more balls to
operation. This paper draws on operating mill experience to fit in the ‘bucket’ (the space between the lifters), increasing
present a current understanding of SAG mill operation from the charge lifting rate and allowing more grinding media to be
perspective of both shell liner and grate/pulp lifter design. The thrown and delivered to the toe. This may result (for larger
paper provides practical observations and interpretations of SAG buckets) in a dispersion of impacts for the thrown charge with the
mill liner operation and performance: its goal is to identify and loss of focus on impact at the toe. However, a positive outcome
for fixed speed mills is that some balls may continue to be
describe the practical issues to consider in liner design for SAG
effective in hitting the toe region for longer through the lifter
mills.
wear cycle.
Mill speed can be increased along with a change in shell lifter
SHELL LIFTERS face angle to increase both impacts of balls at the toe and charge
turnover; these effects improve milling performance through
Shell liner design and mill performance increased ‘ball-charge participation’. However, as mill speeds
increase above 78 - 80 per cent critical speed, pulp lifter
The two main process objectives in SAG mill shell lifter design efficiencies could fall and affect overall mill performance.
are to maximise the impact and rate of delivery of grinding
media at the toe of the charge where most of the impact-induced Increasing mill speed progressively over the wear life of shell
rock-breaking action is assumed to occur, and to avoid liner and lifters can maintain a ‘sweet spot’ in mill performance by
ball damage. Changing the face angles of shell lifters alters continuing to direct ball impacts at the toe as both the lifter
grinding ball trajectories, and spacing between shell lifters height falls and face angle increases with wear. The ‘sweet spot’
(McPherson a:b ratio) affects charge lifting rate hence mill is usually tracked by the feedback of impact sound from
performance (McIvor, 1983; Powell and Nurick, 1996). Trajectory microphones mounted close to the mill. This approach coupled
and charge structure computer models incorporating these with a focused impact, using smaller not larger bucket size, is a
concepts, together with practical operating experience and promising current development in liner design and mill
engineering judgement are used in shell lifter design to meet the operation, providing the potential for lower energy consumption,
above objectives. Advanced computing tools such as discrete increased throughput and start-up ‘on-grind’ with new lifters.
element modelling (DEM), now incorporating slurry effects Current shell lifter designs commonly adopt large face angles
using smoothed particle hydrodynamics (SPH) (Cleary, Sinnott (typically 22° but up to say 30°, perhaps 35° with very large
and Morrison, 2006), contribute to mill liner design by providing lifters) with sufficient spacing between lifters to overcome
detailed illustrations and data on the interaction between liners packing. An aspect of this is that liners are being designed and
and mill charge motion. supplied with substantial sections that can change the bucket
Trajectory models generally track (for shell lifters of various capacity significantly. On the latter point, there is some merit in
maintaining a ‘2D row’ shell drilling in new mills such that 2D,
heights and face angles) the fate of the ‘lifter ball’, the ball that
(4/3)D or 1D liner spacing can be used: the fine tuning of
sits against the lifter and plate (Royston, 2001). At constant mill
spacing between lifters and the related bucket capacity can then
speed ball trajectories should degrade (ie balls fall more towards be accommodated in the liner design. (‘2D’ means twice the
the bulk charge than the toe) with time as lifters wear down. number of ‘feet’ in the mill diameter.)
Shell lifters with initially large face angles may start by directing
ball impacts at the toe and come ‘on-grind’ almost immediately For large mills using 125 - 140 mm feed ball size,
only to fall-off in performance. Any practical shell lifter design conventional new lifter-liner dimensions are around 300 -
350 mm overall height above the shell, around 100 mm plate
has to perform over the full life of the shell lifter/liner. For fixed
thickness and around 150 mm top width: detailed design depends
speed mills, a common practice is to design shell lifters with
on individual mill circumstances. Increasing lifter height usually
some (slight) overthrow when new, such that the liner ‘wears-in’
leads to increased shell liner wear life, and changing the
to come ‘on-grind’ early in the wear cycle with mill performance direction of mill rotation regularly may also increase lifter life by
up to 20 per cent.
1. MAusIMM, Principal, Royston Process Technology Pty Ltd, Unit 7, In traditional ‘HiLo’ liner systems, alternate rows of worn ‘Lo’
435 Gregory Terrace (off Bathgate Street), Brisbane Qld 4000. lifters are replaced with new ‘Hi’. This system appears to work
Email: royston@gil.com.au well in some smaller mills (say, around and below 24 ft),
especially where packing can be controlled to the level of the Mill charge levels
‘Lo’ lifters. Traditional ‘HiLo’ replacement systems fail in larger
mills, especially in high impact environments where large Within the mill charge, balls have to transfer energy into rock
grinding balls are used and it is difficult to avoid breakage of breakage via attrition and abrasion through charge motion in the
highly worn lifters. Usually such mills have been forced into mill (ie the lifter-induced swirl at the toe, the non-thrown
‘HiHi’ shell lifter systems to avoid liner breakage. cascading action and the relative motion of the rising charge).
A recent development in liner design has been to introduce a It is now a common practice, to improve milling rate, for SAG
form of the traditional ‘HiLo’ lifter system where the ‘Hi’ lifters mills to be run with low charge levels at maximum ball charge
are considerably larger than used in prior practice. The ‘Lo’ lifter levels. The outcome is to increase ball-charge participation by
is kept to a height similar to the ‘Hi’ in a prior ‘HiHi’ more turnover of the charge per rotation of the mill and by
arrangement (see Figure 1). The objectives here are to improve increasing the ball-to-rock ratio. In the past, such mills had limits
wear life, to increase lifting rate, to continue to direct ball on the total charge mass and/or motor capacity (and the resulting
impacts at the toe, and to change the shell liner wear (and low charge levels often resulted in increased liner damage, ball
packing) distribution, while (at least) preserving the ball-impact and bolt breakage hence challenges to liner design).
resistance of the original ‘HiHi’ lifter set. Such a ‘HiLo’ liner set With newer mills with higher load capacity, high ball charge
can be of substantial size with an asymmetric design that levels (say up to 18 per cent) have been used. The objective again
requires uni-directional rotation of the mill. 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).
Heavy packing, especially in larger mills, can reduce mill With the discharge end (DE), most large mills have adopted
charge lift and milling performance and increase lifter wear rates cantilever grates usually with large open areas, inter-grate gaps
significantly. An understanding of the packing characteristics of and a large proportion of pebble ports to promote pebble recycle
the ore is a critical aspect of shell liner design. If for example the for pebble crushing. Nevertheless, there are mills with restricted
ore supply is from a single source with constant but limited grate openings that aim to limit rock outflows in order to
packing characteristics it may be practicable simply to promote a fine grind size, and other mills that use a
accommodate some packing as part of the liner design; in lifter jet-return-trumpet system and have to limit in the opening size of
and plate 2D shell liner designs this approach can add grate slots to allow steel media to be returned into the mill by the
significantly to plate life. If packing is severe then spacing the water jet.
liners to say 4/3D (with associated changes in face angle covered The general structural principles outlined above for FE liners
above to avoid plate damage) may provide enough gap between apply also to the DE, noting for grates and plates, the fit should
the shell lifters to prevent packing; then the shell lifter-liner be a one-on-one match with the underlying pulp lifter (see later).
design has to be based on an impact environment where no Grate performance as a liner is affected by:
packing is present. Wider spacing to 1D may ensure no packing
can occur but at the risk of damage (through ball impact) to wide • peening (metal flow that closes the grate openings) – check:
exposed plates and of high liner wear rates due to charge ball-grate impact situation, material of construction, slot
slippage. opening size, wear rates across the surface of the grate, need
for the affected openings;
END LINER AND GRATE STRUCTURES • 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
Introduction type and worn shape, rock fracture at pegging size;
The overall charge level in the mill affects the wear patterns on • trailing-side outer-grate ball-impact damage – check: ball
the end-liners (and the charge transfer effectiveness of the grate throw, grate lifter throw, plate thickness, edge support,
through the degree of exposure of the grate openings to the casting integrity especially at the extremities of the grate,
charge). Ball trajectories together with charge level (especially at web thickness, web support, wear induced strain relief rates
low levels) affect end liner design through the nature of ball across plate;
impacts on end liners and grates.
• inner end grate damage – check: ball throw trajectory
especially if using large face-angle lifters near the head end,
End liner design/life issues local sources of plunging balls, protection provided by inner
An imaginary circular line that may be drawn on the rotating end plate and (worn?) lifters, metal flow between grate and next
inner plate, charge levels, casting integrity, web thickness,
of the mill by the stationary ‘eye’ of the mill charge is referred to
wear-relief rates across plate, glacis (protective ramp) on
here as the ‘eye-line’. The position of maximum wear on the feed inner plate; and
end (FE) lining is around this ‘eye-line’. To prevent (rapid)
abrasive wear of the end plates, lifter bars are used to deflect • general plate damage – check thickness, thicker plates of
charge from the plate. 100+ mm are used in larger mills.
The FE plate itself usually carries a central bar originally used
to stiffen separate plates and in integral FE lifter-liners, favoured PULP LIFTERS
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 Introduction
the plate. Recent trends include: increasing the size of this bar
to improve plate (and indirectly lifter) wear life, and for Material is discharged from a SAG mill using pulp lifters and
uni-directional mills repositioning the bar better to protect the pulp lifting effectiveness can affect the charge level in the mill,
high wear region immediately in front of the FE lifter. hence grinding efficiency. Installed at the discharge end of the
mill, pulp lifters are a radial array of channels separated by
With FE lifters the trend in recent years has been towards FE channel walls also known as vanes or septums. Each channel is
lifters with angled leading faces and outer taper; these designs open to the mill at the outer end to allow material inflow
aim to shed (ie avoid throwing) balls that could damage liners at (through a grate) and at the inner end to allow discharge from the
the head end, and to even out the wear along the FE lifter. mill through the mill trunnion (Napier-Munn et al, 1996).
The change to large face angles on shell lifters has resulted in Typically, the number of pulp lifters employed is the number of
a change towards radial-directed incoming trajectories of balls. feet in the mill diameter (a ‘1D array’).
In these situations balls travel between (ie are not deflected by) Pulp lifters operate through a lifting and bailing action. Pulp
end lifters and hit the ends of exposed end-liners, resulting in lifters fill with pulp (fine rock slurry with pebbles) flowing in
significant, often premature, damage. through grates. The pulp lifters, as they rotate with the mill, lift
Some feed end issues: the pulp until it flows towards the centre of the mill along a pulp
• Ensure good fit of FE parts with the conical mill head (and lifter channel. The pulp exits the mill via an inner ‘discharge
mount on sound backing rubber); poor fit can lead to bolt cone’ that diverts the pulp out of the mill through the trunnion
failure and plate-cracking. opening. Pebbles that fail to discharge fall back down the pulp
lifters, lower pulp-lifter efficiency and cause pulp-lifter wear.
• Limit exposure of parts to radial incoming ball impacts, ie Curved pulp lifters can improve pebble discharge and the wear
avoid exposed ends, protrusions and large bolt hole openings lives of pulp lifters.
that provide ball impact points that can lead (through Pulp already in the pulp lifter can flow back into the mill
persistent impacts) to metal flow and/or fracture. through grate openings as the grates rotate out of the charge.
• For both the FE and DE, avoid mixing new with old parts in Control of this ‘pulp reverse flow’ is an important aspect of grate
ways that expose new parts to impact damage. Preferably and pulp lifter design.
capture most of the wear on a limited number of parts and
change out all these parts together. Pulp lifter charge motion
• Sequence change-out of FE lifters simultaneously with shell The following description of the operation of conventional (and
lifters; this avoids high wear on old shell lifters. curved pulp lifters) is based on video data of the discharge from
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
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
from front to back of around four per cent of mill diameter. For Particle track outputs (Royston, 2006) show different motions
mills with high recirculating loads, or high throughputs of say in large compared with small mills where both have the same
soft ores, the pulp lifter depth requirement should be increased to outer curve radius (see Figures 8 and 9). With proportionally
limit reverse flow. longer straight to curved section in the larger mill, inward motion
In existing mills, a partial retrofit solution for increasing pulp may not be fully sustained once the charge reaches the straight
lifter depth is to increase the depth of the pulp lifter channel portion of the pulp lifter. At this junction, the charge is subject to
towards the centre of the mill in a tapered pulp lifter (see the same forces as would be present on a straight-radial pulp
Figure 7). Another approach (for uni-direction mills) is to use lifter. The force balance may not strongly favour inward flow,
pulp lifters with pulp lifter channel walls (normally at a and the ‘rock’ could even be subject momentarily to slowing
right-angle to the head, Figure 5) angled downwards from the forces. It follows that with the relatively longer straight portion
grate towards the head (ie downwards to the right with respect to of the large mill, the charge would take longer (in the mills
Figure 5): this forms a trough to hold pulp on the head-side of rotation) to reach the centre. Note the comments above on
the pulp lifter and away from the grate during the lifting motion methods for increasing the curve radius, hence increasing the
of the pulp lifter.
proportion of curved to straight channel and improving pulp
While 1D pulp lifter arrays are typical, practical issues such as lifter performance. Nevertheless, a proportion of straight (with
the positioning of mill head-end joints can limit the number and the curved) appears important; the inner straight portion is
distribution of holes used to mount pulp lifters. With a lower
required to direct efficiently to the centre any pulp lifter charge
number of pulp lifters, the charge volume in each pulp lifter
that gets thrown to the other side of the pulp lifter during inward
chamber is increased (adding to the risk of reverse flow). A
lower number of pulp lifters also results in wider individual pulp charge motion after the pulp lifter rotates past the mill vertex. If
lifters and each takes longer to discharge and this increases the the curve extends to the centre, discharging rocks could have ‘to
risk of backflow. Increasing the number of pulp lifters adds to the climb the back’ of the curve towards the centre of the mill for
number of channel walls between pulp lifter chambers and, due discharge.
to the additional ‘dead’ volume of the walls, may lower the
overall pulp lifting capacity. 0.5
0.4
CURVED PULP LIFTERS
Curved pulp lifters can discharge rocks much earlier in the mill 0.3
DISCHARGE CONE WEAR PATTERNS AND out of favour. Various schemes have been tried over the years to
DESIGN prevent wear across bolt openings including inserts that were
held in the liner with the bolt head beneath and out of contact
In conventional straight-radial pulp lifter discharge cones, there with the insert (Denlay, Woods and Royston, 1997) and the
is significant wear of the channel walls that divide the pulp captive bolt systems described below. Current general practice is
lifters. This wear effect is higher on the longest channels in to bury the bolt-head well inside the part so that the head
discharge cones that merge two or more channels into one becomes exposed only at the point of maximum wear of the part.
towards the centre of the mill, especially where a lower fourth The common wedge-seat bolts require good bolt alignment and
channel receives all the charge from the upper three. Discharge seating; spherical bolt-seats are a variation on the wedge seat,
cone design with straight-radial pulp lifters is focused on wear of aimed to provide good seating even with the bolt slightly out of
the channel wall (vane). This may involve the use of alignment.
abrasion-resistant ‘white iron’ castings in discharge cones. In Common sealing washer practice is to use a rubber sealing
composite metal-rubber parts the channel walls may incorporate washer captured in compression under a strong stiff steel washer
wear-resisting alloys. Multiple merged-channel diversion cones cap. Many proprietary seal designs are available aimed
need also to be designed to allow ‘indexing’, ie the rotation of particularly at sealing worn bolt holes. ‘Nylok’ type self-locking
the cone on replacement. This allows the distribution of nuts are popular but plain black nuts are also used.
‘backflow’ to unworn pulp lifter bases, spreading the backflow Bolt removal from worn liners has been improved
wear amongst all pulp lifter chambers to improve the overall considerably with the advent of recoil-less impact hammers, their
wear life of straight-radial pulp lifters. use has improved safety and reduced downtime. Screw-out bolts
Curved pulp lifters produce a significantly different wear are an alternative offered by some suppliers; in this design, bolts
pattern in the discharge cone compared with straight-radial pulp with screwed-on heads are captured inside the liner. On liner
lifters. Heavy wear is predominantly at the diversion curve of the removal, the bolt shanks are screwed out through the shell
cone. Typically two flow-wear induced grooves develop on each allowing the liner to fall in. This type of design has the advantage
side of the channel wall: one is assumed to be associated with of avoiding wear through the lifter bolt-hole region (a source of
fluid pulp flow discharge before the pulp lifter reaches the mill weakness in conventionally bolted liners) as well as reducing
vertex, the other with pulp and solids reaching the cone after the liner removal downtime. One disadvantage is that if the bolt
pulp lifter has passed the mill vertex. In this situation, the design shanks will not come out, heavy liners with the bolting have to
effort is focused on the diversion and exit zones with the aim of be knocked-in together, which can be difficult.
limiting replaceable wear parts (in metal or rubber) to the
diversion zone.
LINER SIZE AND MATERIALS
BOLTING There is now good experience in the use of double-wide and
double-long large liner sizes in large mills. There is pull and
Typical bolt and nut combinations are in the metric system Class push in this development. The pull is the need to reduce
8.8 bolts with Class 8 nuts (SAE Grades 5 <bolt and nut> are downtimes and simplify change-outs by reducing the number of
similar). A higher Class (or Grade) provides the potential for liner movements in and out of the mill. The push is the
higher bolt tensions; however, with the higher tensile strength the availability of large capacity liner handling machines and
bolts can become vulnerable to stress raisers (eg from surface associated bolt-removal impulse hammers (as well as other
damage in handling and installation) and fracture once initiated equipment and procedures) aimed to mechanise and at the same
can lead to rapid failure. time to improve the safety of change-outs, especially of large
Good joint compression is critical to the long-term security of parts.
bolts. If compression is lost across the joint, even though the bolt ‘Double-wide’ parts are commonly used for end liners, grates
may still be in tension, there can be movement across the joint. and shell liners, and foundries have directed efforts towards
The cyclical bending moment can lead quickly to bolt failure supplying such parts. Important issues in moving to
(even in large diameter bolts) through fatigue cracking at a stress ‘double-wide’ parts are to ensure a full fit and for grates not to
raising point (say – one thread into the nut or where the threads overlap joints in the pulp lifters, ie one-on-one fit is required, or
end on the bolt shank). otherwise there is a risk of breakage at or near joints. A potential
A certain amount of preload is required on the lining bolts to negative is that by removing joints between shell liners, there is
ensure compression is maintained under varying liner loads. less capacity for impact strain relief through metal flow and this
Preload is a function of applied tightening torque to the nut and has to be considered in design. A practical issue with
bolt, and the torque is affected by the lubrication applied to the double-wide shell liners is the placement of the necessarily
bolt and nut. Bolt-supplier torque tables usually assume that the substantial lifting eyes; if these fill the space between the lifters
bolts and nuts have clean non-corroded metal surfaces with some they can provide an abrasive-wear path causing increased wear
light oil lubricant (sufficient to get consistent torque readings). on the lifters at that point.
Dry or rusty bolts will be under tensioned at the usual torque Traditional lifter and plate designs continue to have their place
table settings. If special low-friction lubricants are used the providing both opportunities to capture wear on smaller easily
applied torques must be adjusted downwards to match, otherwise removed parts, and at the extreme in very high impact
bolts will be over-tensioned and this could lead to breakage. environments, providing opportunities for a liner design solution
The bolt torque may be applied initially by an air-driven through opportunities to direct high metal flow and for inter-part
impact wrench ‘rattle-gun’, and completed by an hydraulic movement to relieve and to counteract strain build-up (leading to
torque wrench preferably by rotating the bolt without stopping stress fracture).
(ie tightening under constant dynamic friction) until the required ‘Chrome-Molybdenum’ (Cr-Mo) alloy steel combines high
torque is reached. As new liners settle with use, good practice to impact resistance and good wear life and is still the dominant
re-torque all bolts after a few hours of operation, the objective is material of construction for SAG mill liners. ‘White Iron’
to obtain stable bolt tension. continues to be a material of choice for non-impact highly-
The impact of grinding media directly on bolt-heads can lead abrasive wear-zones (where Cr-Mo alloy steel is less effective).
to bolt failure and for this reason bolts with deep heads (aimed to Specialised alloy formulations around the Chrome-Moly type
provide reduced wear across the bolt head opening) have fallen materials have been promoted: these are based on creating
microstructures and through-hardness sometimes by careful heat McIvor, R, 1983. Effects of speed and liner configuration on ball mill
treatment and are claimed to provide improved wear resistance in performance, Mining Engineering, 35(6):617-622.
abrasion-prone parts of SAG mills. Other liner suppliers claim Mokken, A H, 1978. Progress in run-of-mine: (autogenous) milling as
harder high-carbon versions of their ‘standard’ Cr-Mo products introduced and subsequently developed in the gold mines of the
Union Corporation Group, in Proceedings 11th Commonwealth
give similar outcomes; higher carbon appears to provide Mining and Met Congress, Hong Kong.
improved abrasion-resistance at similar hardness levels (assumed Napier-Munn, T J, Morrell, S, Morrison, R D and Kojovic, T, 1996.
to be related to carbide content). Mineral comminution circuits their operation and optimization, in
One interesting development is the bi-metallic liner using a JKMRC Monograph Series in Mining and Mineral Processing,
‘White-Iron’ insert that can give increased wear-life in Vol 2, Chapter 7 (Julius Kruttschnitt Mineral Research Centre:
low-impact abrasion-prone locations such as end liners. In Indooroopilly).
addition, metal-faced rubber ‘Polymet’ type products with Powell, M S and Nurick, G N, 1996. A study of charge motion in rotary
designs resistant to damage through ball impacts are growing in mills, Part 1: Extension of the theory, Minerals Processing,
application in SAG mills. 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
REFERENCES pulp lifters, Preprint 03-108, SME Annual Meeting, Cincinnati, 24 -
Cleary, P W, 2006. Advanced computer modeling for industrial granular 26 February.
flows and particle breakage in mills, crushers and screens, in Royston, D, 2000. Grate-pulp lifter interaction in SAG/AG mills, in
Proceedings IIR Crushing and Grinding Conference, Townsville, 29 Proceedings Seventh Mill Operators’ Conference, pp 63-67 (The
- 30 March. Australasian Institute of Mining and Metallurgy: Melbourne).
Cleary, P W, Sinnott, M and Morrison, R D, 2006. Prediction of slurry Royston, D, 2001. Interpretation of charge throw and impact using
transport in SAG mills using SPH fluid flow in a dynamic DEM multiple trajectory models, in Proceedings International Conference
based porous media, Paper presented to Comminution ’06, Perth, 15 - on Autogenous and Semiautogenous Grinding Technology
17 March (submitted for publication in Mineral Engineering). (SAG2001), Vancouver, 30 September - 3 October, Vol 4,
Denlay, D R, Woods, R J and Royston, D, 1997. Development of the ANI pp 115-123.
Bradken Vortex Pulp Lifter and spherical head bolt mill liner fastener Royston, D, 2004. Review of recent experience with large-angle, wide-
system, in Proceedings Sixth Mill Operators’ Conference, pp 67-72 spaced shell lifter-liners, Mining Engineering, 56(12):73-76.
(The Australasian Institute of Mining and Metallurgy: Melbourne). Royston, D, 2006. SAG mill pulp lifter design, discharge and backflow,
Dunn, R, Fenwick, K and Royston, D, 2006. Northparkes Mines SAG Mining Engineering, 58(9):57-62.
mill operations, in Proceedings International Conference on Taggart, A F, 1947. Handbook of Mineral Dressing Ores and Industrial
Autogenous and Semiautogenous Grinding Technology (SAG2001), Materials, Chapter 5, pp 78-79 (Wiley).
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.
ABSTRACT This paper details some of the findings of the study. This study
is part of a larger project aimed at evaluation of the metallurgical
In many mineral processing operations, the reported production figures
for an accounting period may not make the best use of available data.
accounting strategy of the entire Northparkes operation from
While most of the mathematical approaches, such as check-in/check-out, mine haulage to concentrate load-out.
mass balancing, used in effective metallurgical accounting strategies are
well understood, the sources and magnitude of measurement errors NORTHPARKES OPERATIONS
generally are not. As part of the ‘AMIRA P754-Metal Accounting and
Reconciliation’ project, a study was undertaken at the Northparkes Overview
copper/gold concentrator near Parkes, New South Wales to examine the
data used in their concentrator metallurgical accounting system. The Northparkes Mines (NPM) copper-gold operation is located
This paper summarises the results of a detailed examination of the approximately 27 kilometres northwest of Parkes in Central West
mass measurement systems around the concentrator as well as the New South Wales (NSW). Production presently comes from the
primary sampling system used for grade estimation. The goal of this work E26 underground mine, the first block cave mine in Australia,
was to develop practical techniques to quantify the error (ie both accuracy with make-up concentrator feed coming from surface stockpiles
and precision) of the measured metal accounting information. With this of ore from the now-defunct E22 and E27 open cut mines.
information, statistical confidence levels for this data can then be Mining commenced in 1995.
calculated. If the errors are small, the effect of error in each piece of
measured data can then be propagated to other calculated production The NPM grinding circuit consists of a primary SAG mill with
figures such as recovery and feed grade using a statistical technique oversize crusher followed by a secondary ball mill grinding
called ‘the propagation of error’. 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
INTRODUCTION of rougher-scavenger banks followed by Jameson cleaner,
As part of the AMIRA P754 – Metal Balancing and cleaner-scavenger and Jameson re-cleaner steps. The combined
Reconciliation study, evaluation of all sources of metallurgical flash flotation and flotation plant concentrate is thickened,
filtered and stockpiled before train load-out to the port.
accounting data from the Northparkes copper-gold concentrator
was performed. These data sources can be divided into two
general areas: Concentrator metal accounting
1. tonnage estimation through mass measurement, and Figure 1 shows an overview of the sources of metallurgical
accounting data in the Northparkes concentrator. The plant
2. grade estimation through sampling. 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
1. PhD Candidate, Julius Kruttschnitt Mineral Research Centre, Isles twice the throughput capacity of module 1. Metallurgical
Road, Indooroopilly Qld 4068. Email: w.jansen@uq.edu.au
accounting information for each module comes from two mass
2. MAusIMM, Chief Technologist, Julius Kruttschnitt Mineral measurement sources as well as samples from four slurry streams
Research Centre, Isles Road, Indooroopilly Qld 4068. in the plant.
Email: r.morrison@uq.edu.au The first mass measurement point is a weightometer (belt
3. MAusIMM, Concentrator Superintendent, Northparkes Mines, PO scale) measuring the ore fed to the concentrator from the primary
Box 995, Parkes NSW 2870. Email: rick.dunn@riotinto.com stockpiles, which are mainly fed by primary crushed ore from the
underground mine. Some make-up feed for these stockpiles also weightometer, or belt scale. It is a device attached to a conveyor
comes from surface stockpiles consisting of ore from surface that measures the weight of material on the belt over a certain
mines no longer in operation. The second mass measurement length and the speed of the belt. Using these measurements the
point used for metallurgical accounting is a weightometer that throughput, usually in tonnes per hour, and the total integrated
measures the mass of filtered concentrate from the combined flow, usually in tonnes, are determined.
flash flotation and flotation plant concentrate streams. All of the The estimate of concentrator feed tonnage reclaimed from the
grade measurement samples are taken by the AMDEL on-stream primary stockpiles comes from a weightometer on the SAG mill
analysis system housed in the flotation plant. The system takes feed conveyor belt to each module. Each weightometer supports
real-time grade measurements of concentrator streams for four weigh frame idlers and contains four load cells. Module 2
process control, but physical sample cuts are also taken, which has a typical mass flow of about 400 t/h. The weightometer is
are analysed in the on-site laboratory to obtain metallurgical located just outside of the stockpile reclaim tunnels on a
accounting data. horizontal section of the belt before the crushed SAG mill
recycle is added to the primary mill feed.
Objectives Concentrate production is measured by weightometers on two
belts that convey the filtered concentrate to the load-out
The objectives of this part of the Northparkes metallurgical stockpiles. These weightometers support two weigh frame idlers
accounting study are: each and contain two load cells each with a typical flow of
1. to better understand the error in metallurgical accounting 20 t/h. The smaller and more uniform flow of the filtered
data, and concentrate stream makes for favourable belt weighing
conditions. Therefore using a shorter weighing platform and
2. to better quantify the statistical confidence in the fewer load cells is normally sufficient for this application.
measurement and sampling processes in the concentrator. For all of the concentrator weightometers, a belt speed
In order to do this, it is necessary to examine each source of tachometer is located on the tail pulley. The information
metal accounting data and, if possible, develop techniques to provided by the integrators includes the total tonnes across the
measure or better estimate the error associated with each. weightometer, the throughput of material in tonnes per hour and
There are two basic types of error that can affect the quality of the belt speed in metres per second.
a measurement: systematic and random. These two errors can
affect measurement accuracy and precision. In the sense with Calibration
which it will be used in this document, precision is the measure
of how closely repeated measurements agree among themselves, The weightometers are calibrated during plant shutdowns.
and it is commonly expressed in terms of standard deviation (σ) Calibration involves two steps:
and variance (σ 2 ). Systematic errors produce a measurement • a zero calibration, and
bias. One definition of bias is the difference between the ‘true’
value and the average of a large number of replicate results. • a span calibration.
Thus, accuracy is, in this paper, the combination of lack of bias To perform the zero calibration, the feed of ore to the belt is
and high precision (Dodge, 2003). stopped while the belt is still running. Then an electrician
performs the zero calibration procedure built into the integrator
Structure of the paper software.
The span calibration procedure is nearly the same as the zero
This paper is the outcome of experiments and observations from span calibration. For the feed weightometers, a large metal
the aforementioned metallurgical accounting data sources, and cylinder of a known mass is loaded onto the weigh frame while
the results of this work are divided into three sections. The first the unloaded belt travels over the weightometer. The weight that
section details experiments performed on and observations made this static mass exerts on the weigh frame corresponds to a
of the two mass measurement systems in the concentrator: typical load on the weigh frame during normal operation. This
the mill feed weightometers and the filtered concentrate way the span calibration is performed as close to real operating
weightometers. The variation of wet mass measurements from condition as possible within the limitations of the system. There
these sources is discussed as well as the variation in moisture are a number of other calibration options for belt scales that more
contents used to calculate dry tonnages from the wet mass closely replicate real operating conditions, such as roller chains
measurements. running over the weightometer or material tests because these
The second section of the report takes a closer look at the more closely replicate belt tensions experienced during normal
AMDEL grade measurement system. The results of an operation. However, the required infrastructure for these tests
experiment undertaken to quantify the variation in sample grades does not exist for these weightometers, and thus static weight
from the system sampler is also presented. calibration must be used.
The third section of the report uses information from the first Span calibration of the filtered concentrate weightometers can
two sections of the report and proceeds step-by-step through be done a number ways that are preferable to those available for
various metallurgical accounting calculations that take into the feed weightometers. The weigh platform for these scales has
consideration the errors associated with measurements. Through a protruding peg upon which one can put increasingly heavier
these calculations, a better appreciation of the effect of errors in masses to obtain an incremental span calibration. However, the
various measurements can be gained as well as quantification of conveyor infrastructure allows for a material test, which is an
confidences associated with some of the reportable metallurgical improved calibration because it measures under actual belt
accounting figures. The technique of ‘propagation of error’ loading conditions instead of those simulated by a static weight.
will also be introduced with application to metal accounting The integrator software includes a module for performing a
calculations. material test, where the actual concentrate is conveyed over the
weightometer and into a container (usually the back of a large
MASS MEASUREMENT truck) for a certain period of time. The tared container is then
weighed on a certified weigh bridge. This value is compared with
Feed and concentrate weightometers the tonnage read by the weightometer. Thus, the accuracy of the
weightometers can be assessed. This also satisfies the definition
The predominant method of estimating mass flow of conveyed of accuracy as ‘comparison with a known measurement’ since a
material in a mining and processing operation is by means of certified weigh scale is used for comparison.
A series of material load tests were performed for one of the speed as a speed measurement device whose calibration has
filtered concentrate weightometers both as a means of calibration drifted has the potential to impart a substantial bias to tonnage
and as a check of its accuracy. Two runs were completed, and measurements. This experiment can help identify bias in this
Table 1 shows the results. These two runs equate to two measurement, but can not provide any information about the
calibrations, after the first of which, the weigh scale data was measurement variance. Routine simple checks such as this for all
entered into the integrator. site weightometers can help ensure minimal measurement bias
imparted from belt speed data.
TABLE 1
CV15 material load calibration results. Belt loading
The second measurement used to calculate the total tonnes and
Weightometer Weigh scale Difference throughput on a conveyor belt is the actual weight that the
reading (t) reading (t) (%)
material exerts on the weigh frame as it passes over the
Run 1 7.7 7.5 2.7 weightometer. These weight measurements are taken many times
Run 2 8.4 8.43 0.4 per second by the load cells supporting the weigh frame and
signals are converted to mass per length of belt. This value is
It can be reasonably assumed that the procedure adjusts the multiplied with the speed of the belt to obtain a throughput value,
weightometer readings by 2.7 per cent after the first run. and the throughput value versus time is integrated to give total
Therefore the unadjusted second reading would likely be 3.1 per tonnes. There are a number of issues to consider that can affect
cent, and hence the accuracy of the scale is about three per cent the belt loading value.
off and the precision is almost 0.5 per cent. This equates to a bias As mentioned, using a static mass to simulate weightometer
of about two to three per cent plus a standard deviation belt loading conditions does not accurately represent a belt
(precision) of about 0.5 per cent. A better estimate of the loaded with ore. Belt tension from a fully laden belt can have a
precision would require more data. significant effect on the total mass and throughput readings
While performing a material run calibration is preferable to a (Fristedt, 1986).
static load calibration, it is not always practical to do this due to Dirt and dust build-up on weigh frames and/or the calibration
time restraints or the possibility of a concentrate stockpile mass can gradually bias tonnage high by placing a constant extra
blocking truck access to the falling material. A parallel static weight on the weigh frame. Figure 2 shows a small build-up of
calibration was not performed on this occasion, so a comparison concentrate on the weigh frame of one of the filtered concentrate
between the material run calibration and static calibration weightometers. To illustrate the potential effect, a build-up of
methods was not possible. 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
Mass flow measurement
reported. Keeping a weigh frame dust free is a simple
A weightometer uses two pieces of data to calculate total housekeeping issue that can prevent one possible source of bias.
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.
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
accounting records. This type of information can help to provide and labour intensive to determine a relative standard deviation or
an estimate of the active bias in the mass measurement systems variance for a feed moisture and even more difficult to do this
and can be considered when trying to explain some production task with any regularity as the ore changes. This would require a
discrepancies between measurements. However, caution should large number of belt cuts from the same lot of ore with precise
be exercised as it can be difficult to tell over what period this determinations of moisture in each sample. Even with a
bias, or portion of it, has been active. streamlined moisture content determination procedure, this
Most of these means to combat the introduction of systematic process would be difficult to do regularly. First, one must
errors into the mass flow data via belt loading measurements can consider whether or not it is worth doing.
be routinely implemented by maintenance technicians or The vast majority of variance in the dry feed mass figure
operators for more reliable metallurgical accounting mass data. actually comes from the wet feed mass measurement from the
The majority of the error in a belt scale has been found to be weightometer. For typical Northparkes tonnage data using a feed
normally due to conveyor design and resulting belt effects and moisture of two per cent, the moisture variance accounts for less
user practices (Colijn, 1983). than 0.05 per cent of the variance of the dry tonnage figure.
Thus, regardless of the variance of a measured moisture factor,
Mass measurement variance this number is relatively insignificant. However, the variance of
the measurement is not to be confused with the accuracy of the
Ideally, one would like to be able to measure not only bias in the measurement as described earlier. A poor assumption will always
belt loading and the belt speed, but also the precision for each lead to a bias in the ‘mass processed’ estimate.
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.
Concentrate moisture content variance
Since belt speed and belt loading are the two raw measurements The filtered concentrate stream has different characteristics from
used in calculating the wet feed mass to the plant, one must the feed stream, which make it easier to measure in a number of
understand the variation of measurements in a statistical sense ways. Taking representative samples for moisture is much easier
before a confidence can be quantified for this figure. for this stream due to its much finer size distribution. Because of
In reality, it is difficult to experimentally quantify the variance this, sample handling is less complicated and loss of moisture is
of these measurements. Additional infrastructure would be easier to prevent. Northparkes take an hourly moisture cross-belt
required for a meaningful experiment. Furthermore, the raw grab sample from the filtered concentrate belt and add it to a
measurements taken by the weightometers are processed in a shiftly composite sample in a sealed container.
number of ways before being output. At the moment, the A simple experiment was conducted to quantify the variance
variance for equipment such as this is estimated by measurement of such moisture content measurements used in the calculation
experts, though the basis of the estimate is often unclear. of dry concentrate mass. Ten cross-belt grab samples of filtered
‘Accuracy’ is often claimed as a percentage of full scale without concentrate were taken from each of the two filtered
reference to what confidence limits might apply. concentrate belts in quick succession and placed individually in
sealed bags. Once the samples were taken, the wet and dry
Moisture content masses were weighed in the lab using the current moisture
The quantity of ore calculated from the measured speed and analysis procedure.
loading of the conveyor belts is the wet tonnage. As most sites, The goal of this experiment is to examine the variation in
including Northparkes, report their production values in dry moisture content between the ten samples for each belt in order
metric tonnes (dmt), a moisture content factor is used to convert to quantify the sampling variance that one can expect for a given
the wet tonnage measurements from the weightometers to dry moisture sample. The underlying assumption of this experiment
tonnages. Thus, accurate determination and reporting of the mass is that the concentrate sampled from each belt was from the same
of conveyed material relies on an accurate assessment of the batch and had the same moisture content. The results of the
moisture content of the material. experiment are shown in Table 3.
From the wet and dry masses of the samples, the average mass this error is significant. For high grade materials, measurement
per cut of the automated sampler can be calculated. Because each error increases with the assay value and will typically flatten out
sample in this experiment consisted of more than one cut of the to a maximum value eventually. These ‘end behaviours’ of an
sample stream, the average mass per cut had to be calculated by error model often necessitate the use of two-term error models,
dividing the total dry mass by the number of cuts in that sample. such as those seen commonly in analytical chemistry. These use
By doing this instead of taking one cut per sample, some of the an additive term for low-concentration measurements and a
variability between individual cuts is inevitably smoothed out. proportional term for high concentration measurements (Rocke
However the results are illustrative of the magnitude of and Lorenzato, 1995).
variability in cut mass that can be expected from each stream. A two-term model was considered, but in this case, it was
Table 5 shows the mean dry mass per sample cut and the cut deemed unnecessary because the standard deviation for the assay
mass relative standard deviation for each sampled stream. data at very low assays was always a very small number
approximating zero. The additive error term usually applied to
TABLE 5 account for the minimum error at low concentrations was taken
Standard deviations of AMDEL sample cuts. to be zero in this analysis. Thus, a one term proportional error
model is left.
Module 1 Module 2 The type of model selected satisfies the need for a single term
Avg dry Rel std dev Avg dry Rel std dev proportional model whose error levels off to a maximum value at
mass per cut (%) mass per cut (%) high grades. It is also the model that has the best fit regression
(g) (g) for the data. The data points in Figure 3 show the copper grade
Float feed 49 8.35 70 5.62 versus the corresponding standard deviations for the feed, final
concentrate and final tailings for each module. The model shown
Final con 125 8.15 56 4.25
on the figure is a least-squares regression for a power equation.
Final tails 45 3.98 108 3.42 (Note the use of logarithmic scales for the axes.)
Rougher T - - 123 5.93
Cl Sc feed - - 18 7.89
Fl Fl con 265 7.32 107 39.17
Sample variation models FIG 3 - Copper error model showing assay versus SD.
This study is primarily concerned with the use of overall
variance in a metallurgical accounting and reporting context. The The equation describing this regression line is an error model
variance for each stream of interest can be calculated from the which relates generically the grade of the sampled stream of
lab assay results from the AMDEL experiment. This information interest to the variation that one can expect at that grade using
can show the variance that may be expected for certain grades of the current sampling and assaying procedure. For the copper
material, but it is even more useful to be able to predict the error model shown in Figure 3, the mathematical model is of the
expected variance at any grade of material coming through the form:
system. If there was a relationship determined between the grade
SDabs = 0.0181 ∗ ACU
0 . 9332
and the variance that could be expressed in a single variation (1)
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 where:
relationship exists, and, if so, the best way to model this
relationship for general metallurgical accounting usage. SDabs is the absolute standard deviation of the assay value
Two common ways of expressing measurement variation are ACU is the copper grade in per cent copper (for example, a
using standard deviation or variance. In order to develop a 33 per cent copper assay is entered as 33 and not 0.33)
variation model from the experimental data, the standard For this model, an R-squared value of 0.948 was achieved;
deviations of the assay values for copper and gold were R-squared is a statistical measure of how closely the regression
calculated for each stream for both modules and then plotted in a line approximates the actual data points where 1.0 would be a
graph to determine if a relationship in the data exists. perfect fit of the data.
Application of a single term error model to data such as this To better illustrate the behaviour of this model Figure 4 was
must be carefully considered. For measurements of very low developed, which shows both the same absolute model as shown
concentrations (such as very low grades), a constant minimum in Figure 3 (solid line) as well as the calculated relative standard
error is often found. This means that the error, measured in this deviation (dashed line). The secondary y-axis on the right shows
case as standard deviation, does not grow continuously smaller as the scale for the relative standard deviation model. Note the
the measured grade of material decreases, but rather levels off to x-axis showing copper assay is no longer in log scale so as to
some minimum value. This can be very important to consider if better illustrate the end behaviour of the models.
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
FIG 5 - Gold error model showing assay versus SD. 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
For the gold data, the model describing this general regression error information from this study. Finally, there will be a
is: demonstration of how variance in individual measurements
propagates through metal accounting calculations to give overall
SDabs = 0.0354 ∗ AAU
0.8328
(2) errors and confidence limits for derived figures.
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:
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:
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.
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
People-planning is critical; failure to do so could mean large Allocate a set time for discussion with your subordinate and
amounts of unnecessary and urgent extra work for you. don’t change this time for other meetings. If you do change the
The most effective team must be prepared to believe in you time, it sends the message that you give low priority to this
and follow your lead. Make sure that you support them at every person. Whilst in discussion, focus your attention on that person
turn in public and, if you have to counsel them, do it in private. for the agreed period and don’t interrupt it to answer the phone
Try to make the job fun and have a sense of humour – it helps. or receive other visitors.
Be constructive and honest with feedback during performance
Master tip reviews. If you are required to give the person bad news, give
them good news first, then bad news and close on a positive
If you have a reasonably solid team, the answer to the following
note. It won’t be an effective performance review if you open
two questions should be in the affirmative:
proceedings negatively. Above all, remain positive – most
• Can you go on vacation and leave things in the hands of a situations are retrievable.
competent 2IC knowing that everything will be in a Try to get the employee to tell you what type of roles and
reasonable shape upon your return? tasks they want to do. Use your influence to make it happen, put
• If a person in a key role, such as a foreman, decides to their name forward for key positions for permanent roles, task
suddenly resign, do you have personnel ready and competent forces, projects or studies. Actively support them to achieve
to step up to that position? their goals.
Good managers maintain an ‘open door’ philosophy – it allows
Staff retention free communications at all levels. Those who do not, are
considered unapproachable by their staff.
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 Master tip
competent team members. Take time to ‘walk the talk’, find out what your subordinates are
Numerous employee turnover surveys (Brereton, Beach and talking about. You may have said or done something that has
Cliff, 2003) have been conducted in Australian mines over the been misinterpreted, causing concern amongst the workforce.
past few years and the results indicate a wide range of turnover
from five to 30 per cent, with some sites experiencing rates of Be aware of the big picture
over 60 per cent. In general, because of the various career
development opportunities that arise, it can be expected that Early on in a new position, learn as much as possible about the
most metallurgical professionals will stay in a position for company you work for. Be able to identify the key players
approximately two to four years. The surveys also indicated that (usually on the company web page) and know and understand
the turnover was less for mill and maintenance personnel. In your company’s strategic direction and corporate vision. Be sure
general, expect a retention time of between three and ten years. to attend all corporate visitor presentations. Feed this information
Support the staff by providing good tangible and intangible onto all of your people – most are usually interested. Openly
working conditions. Ensure that the basic tangible conditions are display interest in and support the big picture with those who are
covered, such as fair pay, appropriate crib room and changing in your department, and don’t criticise the company – you are
facilities, but don’t go overboard by providing everything that is seen as being its representative.
asked for by the operating crews – it will only lead to difficulties The upside of knowing about the big picture corporate
later on as they will have become spoilt. Ensure that your staff direction is that it should be a guide for you in your own
have the tools, support and information to do a good job direction and that of your department and sets the scene for the
(Dunn-Lampe, 2005). Increasingly important are the intangible type of people you need to employ, budget inputs and capital
benefits (Beams, 2005), including areas such as planned career requests.
development, supervisor support and flexible work practices, Metallurgists usually come into contact with many other
which are often more important than simply the tangible issues. professionals on site and in the corporate offices. It is not quite
In a survey conducted by The AusIMM on its members in late the case with most other professionals, eg accountants don’t
2005, the skills shortage was seen as the single most important usually deal with electrical or mechanical engineers and it is rare
issue currently facing the mining industry (Anon, 2006). The for mining engineers to work with instrument and control
second most important issue was managing staff turnover. All engineers. This means that you should have a reasonable
this makes a greater challenge to maintain a knowledgeable, understanding of what the others do, so you have a head start on
skilled and experienced workforce. viewing the big picture!
Practice at becoming an active listener to your subordinates,
Master tip peers and superiors.
There are a number of reasons why people leave and it is not Be aware of the overall cost breakdown for the site, and in
usually simply about money, but more due to a lack of particular have a sound understanding of the costs in your area.
challenging work and a lack of supervisor support. Remember It is almost guaranteed that some of your subordinates will be
that people join companies and leave bosses. brighter and more inventive than you, as may be some of your
peers, and perhaps even your boss! Your credibility becomes
Provide regular performance reviews/feedback tainted when your manager comes up with an idea you refuse to
use and in the end they were proved to be correct.
Different companies have different standards in this area, and it
is critical that you nevertheless take time out for ad hoc
Master tip
one-on-one chats with all your staff. Get to know something
interesting about each person, such as whether they are into Take care not to get bogged down by the technical issues. It may
horse racing, yachting, pigeon racing, etc and then brush up a seem like fun and it is after all what you know best and feel most
little on the subject so that you can continue the conversation at comfortable with. However, you are a manager and your job is to
another time. lead your team and influence others.
Acting department manager roles 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
There are several situations where a reliable and capable 2IC will decide to make wholesale changes and you probably won’t be
be acting in a role of the department manager. These include: asked to act as site manager again. Be briefed beforehand on the
• Whilst on R and R at a FIFO operation. It is common for a site manager’s expectations and make sure you fulfil them or
senior metallurgist to act as department manager, and an have watertight reasons if you do not.
individual may end up performing this role for significant You will be regularly required to provide an explanation of the
periods of time. The person in the acting role has all the plant’s performance. You need to get all your facts straight
responsibility but no permanent authority. For those in such a without stretching the truth or blaming other departments. Be
position, use this as an opportunity to practise the role. Leave sure to check out the facts and don’t be caught by the site
critical decisions for when the manager returns, ring the manager during one of their walk-arounds. Be careful of all the
manager or, better still, resolve with the site manager. It is excuses that you come up with for poor performance – they are
best to establish how to do this between the manager and 2IC usually remembered, and you will have difficulties if you cannot
early on in the proceedings. verify them.
• Whilst you are on leave or off-site for business reasons.
These are usually one-off occasions and it is advisable to Master tip
have a competent 2IC to manage the department in your
absence. Give written guidelines to them on what issues you Don’t ever upset the site manager’s secretary or personal
want to be contacted about and how you want to be assistant – they can be an extremely helpful person during key
contacted. times.
• Acting in the role for several months. Sometimes, the 2IC
could be in the role for extended periods whilst a Statutory and other responsibilities
replacement is being sought for the head of department. This
As the most senior metallurgist on site you will need to have a
is a tough situation as there are all the responsibilities, but
clear understanding of your responsibilities and what can get you
there is usually limited authority. Decisions cannot be
into serious trouble – this goes beyond the duty of care. Ensure
deferred and problems must be tackled. At the beginning of
that early in your tenure you find out and take appropriate action
this period, it is suggested that the 2IC has an open
regarding the following:
discussion with the site manager to establish the expectations
of the position, including future handover to the new • Know what inspections and checks have to be performed.
incumbent. It is advisable that suitable mentors or advisors, Always keep these up to date. Be seen doing the inspections
either inside or outside the organisation, are available to yourself and involve key people. Do as much as possible
provide advice on technical and personnel matters. yourself, but maintain a balance and don’t delegate it all.
• Keep up to date with changes in relevant legislation, in
Master tip particular the Mines Act and Regulations. Be aware of what
Ensure you have a trustworthy 2IC. There have been times when is in the legislation, because site managers are usually
the person acting in your position has a closer relationship with conversant with them.
the site manager than you do. In this situation damage may be • Know what can get you terminated, what can put you in
done to your credibility by a ‘white-anting’ 2IC who undermines court or, worse still, what can put you in gaol.
your authority without you being there to defend yourself.
• Always have a clear, simple and logical understanding of
why the plant is not performing to expectations and be
Managing upwards able to explain concisely to others who are not trained
You will experience a wide range of different styles of site metallurgists. This is especially important for presentations
managers, but most will be mining engineers. Broadly, there are and discussions with executive management.
three types of managers: All operations have been designed and developed to generate a
• Those that never visit the plant and leave you alone totally. profit. Ensure that your costs are under control and that you
clearly understand what the cost drivers are and how they are
• Those who constantly visit the mill, talk to operators and influenced. If you have a cost over-run in a particular area, be
fitters and think they are metallurgists. These can make your
sure to get it under control as quickly as possible.
life a misery.
One area where you have primary responsibility is for the
• Those that have confidence in their staff and mostly leave metallurgical accounting. It is critical to ensure that your
you alone, but want to be informed immediately when metallurgical accounting is maintained to an auditable standard.
something significant happens. Failure to do so can lead to serious difficulties, particularly if
Spend time with your manager as early as possible in your there is inventory missing and unaccounted for. This can lead to
appointment to get to know what type you have; what level of long explanations of where the company’s assets are and whether
problem they want to know about; and how often they want to you are in control or not.
see you. Don’t pop in to their office three times a day if they are There was an example where 900 ounces of gold could not be
happy with a once a week catch up. Ensure they find out accounted for at the end of a month’s operation despite numerous
problems from you (and not from some other source), and what checks and measurements. Embarrassingly, a month later the
type of communication method they prefer, be it phone, material was found in the circuit. That system was out of control
face-to-face, email or written memorandum). On a FIFO site you and it took almost a year to fine tune a robust and accurate
will probably have more than one manager, so you have to put system to prevent those kinds of issues occurring again. A key
time and effort into understanding how each one likes to manage measure of business unit performance is cash flow and ‘missing’
and be communicated with. inventory can have a major adverse effect on this measure. Spend
Sometimes you will have the privilege yourself to act as the time setting up a system that is fail-safe, using auditable
site manager, which can be tough if things are not going well in accounting methods to check for any discrepancies.
Often you will be called upon to conduct a tour of your plant explain the process in simple terms. Non-metallurgists think
with visitors ranging from analysts to mine inspectors. It is that metallurgy is a black art – don’t let them leave site still
worthwhile making sure that housekeeping is to an acceptable believing that.
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
• You need to act appropriately at official and social company
overboard. Keeping the plant clean will also assist in the functions outside of work hours. Inappropriate behaviour can
operational crews having pride in their workplace. lead to bad publicity which could undermine your authority
at work.
Setting an example – pride and passion • Do not abuse the privilege of free lunches or dinners,
company cars, phones, or credit cards. People watch your
There are a number of areas where you should set an example as every move and in most companies there will be a code of
people will be constantly watching your behaviour. Some people conduct issue.
will watch for weaknesses and seek to exploit them.
• Senior managers hate surprises. Be sure to communicate bad
• Ensure that you set a high visible personal standard in safety. news at an early stage and keep them informed on the
If you see something being done unsafely and don’t do progress.
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
Master tip
company safety initiatives. Never let alcohol affect your performance or behaviour. Never
• Maintain a reasonably appropriate dress standard and be arrive at work with a hangover and definitely not with a blood
neatly groomed. Always wear appropriate safety gear – never alcohol level above the company’s limit.
be caught out being below the accepted standard. It will be There will be times when it becomes all too much and you
impossible to address breaches if you don’t follow company need to take stock of your situation as to what is causing the
guidelines. stress. Don’t sweat the small stuff – it is only a job. However, it is
• Be aware of pedestal behaviour, where junior employees a reasonably high pressure job and if you are not having fun, get
consider you as a god and hang off every word you say. You out and try something different.
are just the same as everyone else, except that you have
different responsibilities. In these situations you may fall into Run effective meetings
the trap where a thoughtless ‘off the cuff’ comment can be There has been much written and said about running and
misquoted as being fact. managing meetings. They are an important part of the job and
• Do not become involved in spreading rumours or by the they take up a lot of busy people’s time. Some key guidelines to
inappropriate use of email and internet – it could become help make them more productive include:
awkward in counselling individuals. • Ensure that the meetings in your department have agendas –
• Do not speak ill of any company employees, especially your don’t waste people’s time, or they won’t come to the next
seniors and peers – you may think it, but don’t say it. Treat meeting.
others the way you would want them to treat you. • Ensure action items have names and due dates assigned to
• Show impartiality with your team. Don’t have favourites. deal with them. Follow up to ensure that the items have been
Remain somewhat aloof. It will be easier to delegate tasks actioned.
and counsel employees by doing so. • Production meetings can be tedious and boring, especially
• Be a mentor to others around you. Provide mentors from for non-metallurgists when going into specific details on
other areas for those in your department, and don’t forget to metallurgical issues. Make the meetings ‘to the point’.
have a mentor for yourself. Ideally, a mentor should not be • Do not attend or have meetings for meeting sake (just
your immediate superior. because we’ve always had one is not a good enough reason)
• Be punctual for appointments, particularly if you are chairing – challenge the status quo.
the meeting or where there are many company senior • A good meeting is a quick meeting – don’t let them drag on
managers in attendance. for hours.
• Get dirty if needed, particularly when there is a larger than
normal spillage and ‘all hands on deck’ are required. It will Master tip
be appreciated if you perform this task every now and again
to help out, but don’t make it a habit. Every two to three months organise a meeting with the shift
supervisors. The logistics can be complicated, but getting these
• Start and finish work at reasonable hours, as people will key production leaders in the same room for a few hours without
notice and make comments if you are consistently late for production pressures is invaluable. From the shift supervisor’s
work or leave consistently early. They don’t know or care point of view, they hardly ever have time to get to know their
how many other hours you are working. counterparts on the other shifts. Getting them together away from
• Be prepared to work occasionally after hours if necessary, normal production or operating pressures can bring a team
such as on weekends or in the evenings. Take a stroll through together to work cohesively. It also allows the supervisors to
the plant and talk to some of the operating and maintenance offload a number of issues which don’t often get discussed. At
crew. They will appreciate you taking the time to do so. 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
• You are responsible for the costs, equipment and people in
workshop environment.
your plant. Demonstrate that responsibility and when
appropriate, admit fault – do not abdicate the responsibility
onto someone else. Challenge the status quo
• Be prepared to show your plant to dignitaries, directors, A good plant manager will constantly challenge others in their
analysts and other visitors. Be proud of your plant, show area on the status quo. You will often hear: ‘we’ve always done it
passion in the work that you are involved in and be able to that way! Why should it change?’ Many people like to be in their
comfort zone and don’t want to make changes or ‘rock the boat’. plant operation only – you need to be aware of the big picture.
Unfortunately, life does not remain the same; there is always You can do nothing about the product specifications as they are
continual change and room for continuous improvement. controlled by sales contracts. However, it is your responsibility to
This can be illustrated by the following exercise. At a produce the product within these specifications to ensure that it is
workshop attended by one of the authors several years ago, each saleable. You won’t get much mileage from complaining about
group had to compete to build a tower out of paper and sellotape the ore that you are fed, particularly if you have nothing to back
to suspend a fifty cent piece as quickly as possible. Fortunately, it up with and given that most site managers are mining
one member of the group had partaken in a similar exercise and engineers, the mining department will almost always get the
quickly designed and built the required structure. As our team sympathetic ear. You need to be in a position where you can
came first, we shared the design and construction characteristics, influence your geology and mining colleagues so that the
before repeating the competition. Again we came first, but if we treatment plant receives ores that are of acceptable quality and of
had built it in the same time as in the first round we would have a consistent nature. Mine to mill is a good start, but the big
been in eighth place. As the most senior metallurgical picture is really mine to market. Be sure that the material you are
professional on site, it is your duty to seek out new ways of given is capable of making a product that is always within
improving performance and getting the most out of the ore, the specifications. Take time to understand the mineralogy in both
equipment and the people. the valuable product and the gangue.
Be prepared to make reasonable experimental changes to the Depending on the nature of the organisational structure, the
plant, and if it does not go according to plan, admit it and correct maintenance function can be part of the plant manager’s
it. Do not make wholesale changes from which you cannot responsibility. This can be quite daunting to the newer plant
recover. managers, as you sort out the balance between production and
maintenance. In a complicated processing operation the
If you want to implement a process change, be sure to actively maintenance section will probably report directly to the site
engage the operating personnel, particularly the shift supervisors, manager. This means that the plant manager has to deal
to see how they would go about making that change. Listen effectively with the maintenance manager in order to fulfil both
carefully to their answer as they will probably provide the most the production and maintenance requirements.
effective method. More importantly, as they gain ownership, they
will do their utmost to ensure that the process change does not
Master tip
fail. Ensure you have your shift supervisors’ support when
making major changes. If they do not support you, there will be Spend time gaining a better understanding of what is happening
problems with the operating crews in accepting them. in the mining and geology departments. Make regular trips to the
At a gold operation several years ago, there was a requirement pit or underground and have a clear understanding where the ore
to conduct a trial to retreat some tailings material over a several is and will be coming from. Bring these departments closer to the
week period. Presenting the crews with this challenge and asking processing plant by showing them what makes it work better and
for the most effective approach resulted in unexpected show them the impact of having to treat inferior quality ores.
enthusiastic support which provided a more than successful trial.
Do not make changes for changes sake. Sometimes you will be The future – Generations X and Y
pressured by your site manager to make some quick Future trends will be different from what has been experienced in
extraordinary changes. Think through carefully before you act to the past. Many professionals now have double degrees, usually
minimise the negative impact. Be alert for the controlling site with an affinity for the commercial side. Research has indicated
manager who thinks they are a plant manager as they sometimes (Stuart, 2005) that few of the younger people are prepared to
like to make wholesale changes in your absence. consider working in the minerals industry in either a professional
Identify people that support change and work with them. or trade role and many have no desire to work in regional or
Involve these people at the early stages. If they have buy-in they remote areas of Australia. It disturbingly indicates that they will
will help to make the changes and incremental improvements most likely not stay as site metallurgists for too long and will
will occur. want to relocate to the city to pursue other interests.
Traits of Generation Y (born between 1980 and 1994) are said
Master tip to be (Kotler, 2003; Stuart, 2005):
Do not make a whole series of process changes to conditions in • technological savvy,
the plant late on Friday afternoon or just before you fly out and • high expectations of the future,
then leave the crews with no one to contact for the weekend or R • an obsession with achievement,
and R break. This leaves the crews with little or no support and
could undermine your authority and respect. • emphasis on immediate gratification,
• greater acceptance of racial and ethnical differences,
Working with other departments • are at school now or have left in the last ten years,
You are in a unique position where your section is on the • network and communicate madly,
receiving end of the ore to be processed and the plant has to turn • gain knowledge on the need-to-know basis, and
out a product that usually has to meet tight specifications. This
can sometimes be seen as the ‘raw end of the pineapple’, • are concerned about the environment (but few are studying
particularly if there is a poor rapport between the mineral the earth sciences).
processing, geology and mining engineering departments. On the Younger generations have a tendency to focus on lifestyle
contrary, you are in a most powerful situation, where, if you issues rather than career development. This could make it
know the ore and how it behaves, you will end up with a great difficult to recruit potential employees in outback residential or
deal of influence and control. You need a supportive site manager FIFO roles as many of the younger generation are not keen to
for this to occur. move away from the cities and FIFO arrangements may have a
Unfortunately, it is rare for the mining and geological limited shelf life. This shortage of the younger generation
professionals to take more than a passing interest in what takes combined with older workers leaving the workforce to retire
place in a processing plant. Do not focus on the details of the further exacerbates the skills shortage.
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);
The Swan decline is located at the Kangaroo Flat portal site, Processing Plant (NMPP) on about 35 000 t of development ore.
the southern, and shallowest known part of the reef system, The test work included the following major activities:
to access the orebody for exploration, sampling and mine • comminution testing and physical characterisation of ore
development purposes. This intersects the Deborah and hardness and competency,
Sheepshead lines of reef, which will be mined and treated in the
• extensive gravity testing,
initial phase.
• flotation response,
Permitting and processing • recovery trends with grind size,
In accordance with the requirements of the 1998 Environmental • diagnostic size-by-size analyses on residues,
Effects Statement (EES) and the 2004 Supplementary • leach kinetics on flotation concentrates,
Environmental Effects Statement (SEES), surface storage of the • settling and thickening characterisation,
tailings from the processing facility at the Carshalton site are
permitted as follows: • cyanide detoxification, and
• Gravity tailings produced from gravity processing and which • tailings dewatering.
have no chemical treatment. The majority of the test work for the Feasibility Study was
• Flotation tailings produced from flotation and which have carried out between 2001 and 2003. A further series of test work
had contact with flotation chemicals. was carried out in 2004 to provide additional information for
flow sheet optimisation and process design purposes.
• Disposal of separate cycloned overflow and underflow Key outcomes and conclusions from the metallurgical test
products from flotation tailings.
work and pilot plant trials were:
• Gravity/flotation tailings stored in purpose-designed facility • Pilot plant trials treated ore with high head grades, 8 -
for possible future recovery and reuse as by-products.
14 g/t Au, consistent with typical feed grades expected from
• Tailings from float concentrates which have been treated the mine during the initial years of operation.
with flotation chemicals and the residue from these
• Bulk samples with grades from 3 - 20 g/t Au provided an
concentrates which have been treated with cyanide. Tailings
assessment of gravity recovery for the range of head grades
from the leaching of float concentrates will be detoxified
likely from the mine.
prior to storage. The weak acid dissociable (WAD) cyanide
level will be less than 50 mg/L, in accordance with the • The ore is considered moderately hard from work index
International Cyanide Management Code, at the point of testing on five samples. These produced a range of values of
discharge into the facility. 13 to 20 kWh/t for the rod mill work indices and 13 to
17 kWh/t for the ball mill work indices.
• Float concentrate leach residue stored in permanent (lined)
facility. Float concentrate, and the resultant leach residue, • The same samples showed a range of abrasion indices from
represents a small fraction of the total plant tailings, thereby 0.20 to 0.45.
reducing chemical use and storage. • Liberation of much of the free (nugget) gold occurs at coarse
• Leachate, if generated, will not be released to the sizes around 1 to 2 mm.
groundwater. • Gold and particle size distribution analyses show that metal
Consequently, a cyanide leach and carbon-in-pulp or value in fractions coarser than 1 to 2 mm can be variable, but
Merrill-Crowe type operation on whole ore treatment were not is typically 50 per cent of the total value in the head. This is a
applicable process routes. Preliminary metallurgical testing done higher proportion of metal value in these coarser sizes
on gravity and whole ore leaching showed comparable recoveries compared to most other mines in Australia (Johansen, 2005).
to gravity/float/intense leach of float concentrate. The latter • Gravity gold recovery from bulk samples is variable,
process route was considered to offer environmental advantages. particularly in lower grades, and generally in the range of 60
Concentrate processing at the Carshalton site, which includes to 90 per cent for the higher grades tested. A summary of the
cyanide leaching (of gravity and float concentrates), results of recovery by grade is shown in Figure 2. The
electrowinning, smelting and leach residue detoxification was proportion of feed to gravity concentrates, corresponding to
approved. these recoveries, is generally 0.3 to 0.6 per cent by weight.
• Gravity gold recovery from the pilot trials on parcels of ore
Project key stages from higher grade reef areas are shown in Table 1.
The Feasibility Study, managed by AMC Consultants Pty Ltd,
was completed in 1Q04. BML awarded an EPCM contract to TABLE 1
Ausenco Services (Ausenco) in 3Q04 for the design and Gravity recovery, New Moon Processing Plant.
construction of a 300 000 tpa process plant, with allowances for
future expansion to 600 000 tpa, and associated services to treat Stream Description S3-101 S3-Backs Garrards
the underground ore. Following an interim engineering and cost
review in 1Q05, BML increased the treatment plant capacity to Head Au g/t (calc) 9.50 8.71 14.0
Au g/t (assay) (9.74) (8.05) (14.2)
600 000 tpa. Mobilisation to site occurred in May 2005 and first
gold was poured in July 2006. Conc Recovery % Au 77.6 74.9 88.2
Conc Weight % 0.5 0.8 1.8
PROCESS CHARACTERISTICS OF
BENDIGO ORES • Recoveries are consistent with those achieved for the bulk
samples, but generally with a higher weight proportion from
Metallurgical testing feed to gravity concentrates.
Extensive metallurgical testing for ore characterisation and • Flotation gold recovery from the pilot trials treating screened
variability assessment was carried out at bench scale on more gravity tails (minus 300 µm) gave gold recoveries (from float
than 90 bulk samples, and by pilot testing at the New Moon feed) in the range 76 to 89 per cent.
100
90
80
70
Recovery, % Au 60
50
40
30
20
10
0
3.0 4.0 5.0 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
The Bendigo ores tested are notable for their relatively high Three and four stage crushing
grades, excellent gravity gold recovery response, extremely coarse
liberation size (by mining industry standards) and good response A flow sheet based on conventional three-stage crushing to
to flotation. The ores are relatively ‘clean’ metallurgically with nominally -10 to 12 mm would not achieve the desired product
minor amounts of arsenic (from arsenopyrite) as the only size. A further concern was that optimum plant metallurgical
significant deleterious element in the concentrate. performance in the ball mill and gravity recovery circuit would
not be achieved due to flattening, over-grinding or smearing of
Flow sheet development gold particles in the ball mill. This results in lock-up of gold in
the milling circuit and renders gravity and classification
As evident from the mineralisation characteristics and the testing processing less efficient as the flattened particles present a large
program, the main design consideration for the process plant is to surface area to flowing water films.
maximise gold recovery in the gravity circuit in a cost-effective A four-stage crushing circuit would produce a finer product
manner. Maximising gravity gold recovery is expected to
closer to the crush size required, but with an increasingly
maximise overall gold recovery as coarse, well liberated gold
complex circuit that required a large footprint within the limited
that is not recovered by the gravity recovery system goes to the
flotation circuit and is likely to be lost to flotation tailings. A area available for the processing plant in the site as permitted.
further consideration for overall gold recovery is to ensure Circuits of this type treating high silica ores typically have high
adequate liberation of gold is achieved as unliberated particles of operating costs and increased maintenance demands to maintain
gold or pyrite/gold may also be lost to flotation tailings. operating availability. The finer crush size may reduce, but not
To maximise gold recovery, the flow sheet was developed with fully alleviate the flattening and smearing issues in the ball
the following processing strategies: mill.
• fine crushing,
Two stage crush and HPGR
• maximum deportment of crushed ore to bulk gravity (coarse)
gold recovery prior to ball milling, A high pressure grinding rolls (HPGR) system was considered
for the tertiary comminution stage following a conventional two-
• upgrading of bulk gravity concentrate to a product suitable stage crushing circuit. Batch testing of about 1300 kg of ore was
for smelting,
carried out at a manufacturer’s laboratory. Unit performance
• intensive leaching of tail from gravity concentrate upgrade, parameters on power and wear rates were obtained. The HPGR
• grinding and liberation of non-gravity recoverable pyrite/ manufacturer testing showed an approximate ten to 15 per cent
gold for subsequent flotation, increase in the proportion of -2 mm product (‘natural’ nugget
size) with the HPGR in open circuit, and even higher with the
• flotation of a pyrite/gold concentrate, HPGR in closed circuit when compared to the product
• carbon-in-leach recovery of the gold from the flotation distribution predicted from tertiary crushing.
concentrate, and As plant testing with the HPGR during the plant trials at
• conventional gold recovery system from loaded carbon. NMPP was not able to be carried out, a series of comparative
option studies were carried out to assess the potential benefit of
DESIGN CONSIDERATIONS FOR THE increased recovery using a HPGR with various crushing
arrangements. With the HPGR in the circuit, a simple payback in
PROCESS PLANT less than three years required an increase in overall gold recovery
of one per cent (at assumed gold price US$400/oz).
Comminution circuit selection The decision to proceed with the HPGR in circuit was made
Analysis of the test work indicated that gravity gold recovery is on the basis of technical, operating, capital cost, operating cost,
strongly dependent on the proportion of -4 mm product that lead time, manufacturer support and performance warranty
could be generated by a crushing circuit. considerations (Johansen et al, 2005).
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
A sample cut from final float tailings, representing about three The final concentrate, loaded into trays, is placed in the calcine
per cent by weight, was provided to detect further potential oven and dried overnight before smelting.
gravity recoverable by treating this bleed stream in a centrifugal In the smelting furnace and, in some conditions in the calcine
bowl concentrator. oven, arsenopyrite can oxidise to form arsenic trioxide (arsenic
trioxide has a boiling point of 465°C but sublimes at 193°C).
Disposal of float tailings Smelt furnace fumes along with fumes from the calcine oven flue
and electrowinning cells are ducted to the gold room scrubber.
Disposal of tailings and management of the tailings area The wet scrubber system uses a weak sodium hydroxide solution
considered several objectives: circulated through the tower to remove contaminants in the gold
• able to recover a coarse sands product if, in the future, a room area as well as from the sodium metabisulfite and lime
market was identified and specifications were confirmed; mixing tanks.
• recover as much of the water (and residual flotation A test work program was carried out to provide information
chemicals) contained in the flotation tailings slurry directly on the amount of arsenic oxidation expected. A range of
back to the process water circuit; arsenopyrite contents in concentrate, degrees of oxidation and
• reduce process water accumulation in the pond in the tailings rates of oxidation were assessed to determine conservative
area; and arsenic trioxide concentration criteria for scrubber design.
• enhance deposition characteristics of the slimes, which Further, initial calcination temperatures are limited to 250°C to
would improve the settled density of these particles. allow monitoring and evaluation of smelting methods and calcine
conditions.
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
Noise attenuation
underflow, or potential sands product, is pumped to separate Noise targets are a high profile community issue for this plant
areas in the tailings storage facility. and targets have been set for maximum allowable noise levels at
various times of the day and weekend. These targets must be
Carbon-in-leach (CIL) circuit achieved to meet the conditions of the work plan and license.
A moderate preg-robbing effect was identified in the test work. A noise study for the project was undertaken for the SEES.
Further noise modelling work was carried out during the plant
Activated carbon is present in all tanks in the seven stage CIL
design stage to provide recommendations for building design and
circuit.
noise attenuation practices.
Features of the design include the bulk of the crushing circuit
Carbon stripping and regeneration located below ground level in an insulated building and the use
Gold is recovered from the loaded carbon in a conventional of insulated buildings to cover most remaining plant. Conveyors
pressure Zadra elution circuit. Stripped carbon is regenerated in have been covered and sound skirts installed to mitigate noise.
batches in a horizontal kiln. The regeneration system treats
100 per cent of the carbon inventory from stripping to remove PROCESS AND PLANT DESCRIPTION
organic fouling from flotation reagents. This ensures carbon
activity is returned to ‘near-new’ condition suitable for return to General
the last CIL tank. An after-burner treating the regeneration kiln
exhaust gases is included to remove odours from burning of The main activities for processing gold ore at BML’s operations
organics. are:
• crushing;
Disposal of CIL tailings • grinding and gravity concentration;
The SEES requirement was that CIL tailings are detoxified • gravity concentrate upgrade in gold room;
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 • flotation;
operating target of less than 20 mg/L CN-WAD from the • cyanidation of gravity concentrate upgrade tailing and
commencement of operations. flotation concentrate;
Bench scale, semi-continuous testing was carried out on • acid washing and elution of loaded carbon;
samples of float concentrate from NMPP to confirm the cyanide
• regeneration of eluted carbon;
destruction method selected and determine process parameters.
The copper catalysed, air/SO2 method was selected based on its • detoxification of cyanidation leach residue slurry,
proven performance, robust operating system and cost-effective • electrowinning and smelting to doré bullion;
plant. The detox test results were analysed and design criteria
selected to achieve sufficient residence time in the reactor, • de-sliming of the flotation tailings to produce a sand product;
and
adequate dosing capability for reagents, and supply of oxygen,
which would enable the BML target to be achieved. • reagent mixing, storage and distribution.
Due to the large air demands and the requirement to place the
detoxification facility in a building, the system was designed to Overview
utilise oxygen injection to replace the more typical use of low
pressure air for the reaction demands. 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
Scrubbing exhaust fumes from gold room
and centrifugal concentrators) to recover the majority (up to
The final gravity concentrate, from the various upgrading stages 70 per cent is expected) of the gold in the ore. The recoverable
in the gold room, contains remnant heavy minerals comprising remainder is collected by froth flotation. The gold production
mainly pyrite, but with varying amounts of minor arsenopyrite. routes are:
• Gravity concentrate is upgraded in the gold room using Spinner and centrifugal concentrator concentrates from the
shaking tables and a centrifugal bowl concentrator to produce gravity/grinding circuit are periodically flushed to the gravity
a final product suitable for calcination and smelting. concentrate surge tank and pumped to the gravity concentrate
tank in the gold room. Accumulated spinner and centrifugal
• Gold room tails (from shaking tables) are subject to intensive
concentrator concentrates together with gold room gravity
cyanidation. The soluble gold is electrowon from the leach
solution. Cathode clean-up is smelted in the gold room. spinner tails are fed to the gold room fine screen. Oversize
particles in the concentrates are fed to the concentrate roll
• Flotation concentrates are treated using the CIL process. crusher and then to the gold room gravity spinner. Spinner tails
Gold is eluted from loaded carbon and is electrowon from the are returned to the gold room fine screen.
pregnant eluate. Cathode clean-up is smelted in the gold
Fine concentrates (undersize) are fed to the concentrate jig. Jig
room.
concentrates are upgraded using two shaking tables in series,
whereby the final concentrates from the tables and the gold room
Process plant spinner are calcined prior to direct smelting with fluxes into doré
A brief description of the processing stages is provided below. bullion.
A plan view of the process plant general arrangement is shown in
Figure 6. Intensive leach and gold recovery
Jig tail and table tails from the gold room are leached in a
Crushing and HPGR proprietary system (intensive leach reactor, or ILR), where gold
Ore is delivered by truck from the underground mining operation dissolves under aggressive reaction conditions in an enclosed
to a run-of-mine (ROM) storage pad. A front-end loader transfers rotating drum. Cyanide and sodium hydroxide solutions are
ore from the ROM pad stockpile to the two stage crushing plant. added to the ILR. Clarified pregnant solution containing
A conveyor delivers crushed ore to the fine ore bin. Crushed ore solubilised gold (electrolyte) is circulated through an
reclaimed from the fine ore bin is fed to the high pressure electrowinning cell. Gold is deposited on the cathodes. The
grinding rolls (HPGR) operating in closed circuit with the ball barren eluate is recycled to the ILR circuit.
mill screen. At the completion of the electrowin cycle, the loaded cathodes
Extremely coarse gravity recoverable gold nuggets are are washed by high pressure water sprays and the washings
recovered in the ball mill screen gold trap in the screen feed box. drained from the cell and the sludge filtered. The filter cake is
dried prior to smelting with fluxes into doré bullion.
Gravity and grinding
Flotation concentrate leaching
The ball mill screen undersize is pumped to the primary jigs. Jig
tails gravitate to the centrifugal concentrator via a scalping The flotation concentrate slurry is thickened to remove excess
screen. Spinners upgrade primary jig concentrates. water and combined with tails slurry from the ILR. The
Tails from the centrifugal concentrator are pumped to the mill combined slurry is screened to remove trash, then mixed with
cyclone cluster. The cyclone underflow gravitates to the ball mill lime slurry for pH control and pumped to the CIL circuit.
feed chute. Ground ore slurry discharging from the ball mill Sodium cyanide is added to the slurry as it enters the first
combines with mill screen undersize in the mill pump hopper. agitated CIL tank. Oxygen, to aid dissolution, is added to the CIL
The slurry is pumped back to the gravity gold recovery circuit. tanks down-shaft in the agitator in each tank.
Dissolved gold is adsorbed onto activated carbon granules.
Thickening and flotation Carbon is advanced counter-current to the pulp flow by air lifts.
Air swept screens retain carbon in each tank.
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 Detoxification
control flotation feed slurry density, activator, collector and Leached slurry, discharged from the agitated CIL tanks, is
frother are added. The overflow of the conditioning tank feeds treated in the cyanide detoxification reactor using a modified
the flotation cells. Flotation is in a single bank of rougher cells in copper catalysed, air/SO2 process. This reduces the (weak acid
a 1/2/2 arrangement. Flotation concentrate slurry is pumped to dissociable) cyanide concentration to below permitted limits. The
the concentrate thickener. detoxified CIL tailings slurry is pumped to a separate lined
impoundment.
Tails storage facility
Float tailing slurry is pumped to the sand cyclone cluster. Fines Loaded carbon gold recovery
are removed in the cyclone overflow, which gravitates to the
Loaded carbon from the first CIL tank is recovered by air lift and
flotation tail thickener to remove excess water.
screened. Water sprays on the screen assist in removing slurry
The cyclone underflow (sands) slurry is diluted to a controlled from the carbon. The separated loaded carbon gravitates to the
pumpable density and pumped to a section of the tailings storage acid wash column. After acid treatment and water rinsing to
facility. Thickener underflow is pumped to the flotation tailings remove the residual acid, the carbon is transferred to the elution
storage facility. Supernatant water is decanted from the settled column.
solids on the TSF and pumped back to the process water tank.
Gold is stripped (eluted) from the loaded carbon into a gold
bearing solution (eluate) using a hot (130°C) caustic cyanide
Gold room gravity recovery solution. The partially cooled eluate is circulated through a
Concentrate from the mill screen gold trap discharges into a dedicated electrowinning cell. At the completion of the
kibble in the gold room. Coarse concentrates from the gold trap electrowin cycle, the loaded cathodes are washed by high
are hand sorted to recover coarse gold nuggets. After sorting, pressure water sprays and the washings drained from the cell and
coarse concentrates are crushed using the concentrate roll the sludge filtered. The filter cake is dried prior to smelting with
crusher and fed to the gold room gravity spinner. fluxes into doré bullion.
Regeneration ACKNOWLEDGEMENTS
Barren, or stripped, carbon is processed through a regeneration Bendigo Mining Limited’s assistance in preparation of this paper
kiln which operates at 750°C to restore (reactivate) the gold and permission to publish is acknowledged. The input of BML
adsorption capacity of the carbon. The reactivated carbon is and its consultants, for aspects of process design and project
water quenched, screened and returned to the last CIL tank. development in treating these ‘nuggety’ quartz reef ores and
meeting challenging site specific conditions, is also
Reagents acknowledged.
Cyanide Leaching
Tails Dam
Zinc Precipitation Cyanide Destruction
Based on the findings of surveys, several circuit and force as opposed to the standard 60 ‘G’ in order to increase
operational changes were implemented and include: the recovery of fine (<10 µm) gold.
• A 30” Knelson was installed in parallel to the in-line • Following the automation of the Gemini table, the Knelson
pressure jig, and was also fed from the mill discharge hopper. concentrators were set up on an 18 minute dump cycle to
The tailings from the 30” Knelson then reported to the flash maximise recovery and minimise any displacement of gold
float cell thereby limiting the misreporting of gravity by the coarse sulfides. Due to the competing trade-off
recoverable gold to the flotation concentrate. between short cycle times and increased ‘off-line’ periods,
• The mill discharge hopper was partitioned, such that all the 30” Knelson was fitted with a ‘dynamic braking module’,
ground product was treated through a gravity device prior to increasing the utilisation.
classification or flotation. The current gravity circuit flow sheet is shown in Figure 3.
• The gold room table was enlarged to a Gemini GT1000 (after Figure 4 trends the mill feed grade versus gravity gold
an unsuccessful attempt with a Wilfley table) and recovery as various process improvements were installed and
subsequently automated such that the middlings stream was commissioned. The trend highlights that gravity gold recovery
pumped back to the table feed. Tabling occurred 24 hours per increased with feed grade through 2003 - 2004. However, the
day without the need for an operator, resulting in improved gravity gold recovery has remained elevated at approximately
efficiency and security. An intensive cyanide leach reactor 50 per cent since early 2005 when feed grades treated were
was explored to replace the table; however, the potential risk similar to those treated in the first few years of the operation,
associated with a subsequent toxic effect to the ensuing prior to circuit changes, when gravity recovery averaged 27 per
bacterial oxidation circuit and long payback period ruled this cent. Gravity recoverable gold (GRG) tests on the BacOx feed
out. has indicated <0.2 per cent of the gold was recoverable to a
• The gold room table tailings stream was redirected to the gravity concentrator validating the success of the installed
flotation concentrate regrind circuit and not back to the ball circuits and operational strategies.
mill, as it contained sulfides and fine gold that was subject to
further treatment, reducing the potential of loss to the FLOTATION CIRCUIT
flotation tailings. The original flotation circuit was designed to recover the gold
• The in-line pressure jig was modified by cutting down the bearing sulfides with performance guarantees to produce a
internal segmented ring from 100 mm to 30 mm to reduce concentrate grade of 27 per cent sulfide, and a flotation tailings
the potential blinding of the ragging by shards of steel from <0.3 g/t gold. Actual performance achieved the target sulfur
the grinding media. The jig ragging and screen aperture was grade, but gold loss to tailings was initially 1 g/t gold.
also enlarged to increase the yield from 2 t/h to 6 t/h. The original flotation circuit consisted of a flash flotation
• The 20” Knelson was moved from the float tails to treat the cell in the primary grinding circuit in combination with a
increased jig concentrate mass flow. conventional flotation circuit treating cyclone overflow post
gravity gold recovery. The conventional float circuit consists
• The 12” Knelson, previously treating the jig concentrate, was of a bank of rougher/scavengers with the rougher/scavenger
moved to the regrind circuit to recover fine gold that reported concentrate being cleaned in cleaner cells to achieve the desired
to concentrate via flotation in the unit cell, conventional sulfide concentrate grade for optimal bacterial activity. Reagents
flotation circuit or from the gold room table tailings. The 12” used are copper sulfate for arsenopyrite activation, sodium
Knelson concentrator speed was increased to develop 120 ‘G’ isobutyl xanthate and interfroth IF106 frother.
Ball Mill
Mill Discharge Inline Jig 12” Knelson
Overflow
Classification Gemini Table
Underflow
FIG 2 - Original gravity gold circuit and gold balance (treating 200 per cent of new feed tonnage).
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
Surveys of the flotation circuit and subsequent mineralogical • installation of in stream analysis system for inferring a sulfur
analysis indicated that: grade,
• The majority (>50 per cent or 0.5 g/t) of the gold and sulfide • cycloning and regrinding the coarse or higher specific value
loss in the flotation tailings was <10 µm. Gold and sulfur minerals recovered to the scavenger concentrate prior to
recovery in the -10 µm fraction was low at 53 per cent (refer refloating,
Figure 5).
• installation of froth crowders, and
• The upgrade ratio of the gold and sulfides was poor in the • fully automating the circuit operation.
<10 µm fraction and exacerbated by entrainment and
subsequently poor selectivity.
Coarsening of the primary grind size
• Mineralogical analysis of the tailings highlighted that fine
gold and sulfides were encapsulated within ankerite The original comminution circuit was designed to achieve a
(carbonaceous gangue mineral), and there were lean cyclone overflow P80 of 75 µm. This figure was derived from
composites in the scavenger concentrate that required further laboratory test work. Comparison of the size distributions of the
liberation to permit recovery to final concentrate (refer plant cyclone overflow with that of metallurgical laboratory test
Figure 6). samples revealed that despite having the same P80 value, this was
• Froth structure was weak due to the large froth surface area to only one point on the distribution curve that was totally different.
cell volume ratio for the quantity of mineral to be recovered. 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.
• Control of the circuit was by manual sample collection, The laboratory rod mill product contained approximately
LECO analysis for sulfur followed by manual adjustment to
the air to achieve the target concentrate grades. This four 47.5 per cent <38 µm and 14 per cent <10 µm whereas the plant
hour turn around process was almost futile based on the product contained 58 per cent <38 µm, and 30 per cent <10 µm.
40 minute circuit residence time and cyclic feed grades. To reduce the fines generation prior to flotation, several trials
were undertaken and included, manipulation of the crushed ore
Based on the plant surveys and observations, the following feed size to the ball mill, ball charge, milling density and cyclone
changes were made to the circuit: inclination. All modifications had a minor benefit due to
• coarsening of the grind size from 75 µm to 170 µm, limitations by the cyclones. Ultimately the four operating 10”
70 100
60
80
50
% distribution
Recovery
40 60
30 40
20
20
10
0 0
-10 10 21 38 53 75 106 150
Size (µm)
Tailings mass % Tailings Au distribution Tailings S distribution
Flotation Gold recovery Flotation Sulfur Recovery
FIG 5 - Flotation size by size recovery and gold/sulfide distribution in the flotation tailings.
100
P80 correct @ 75 µm, but
large difference in % -10 µm
Cumulative % Passing
80
60
40
Feasability (lab)
20 Plant (Dec 00)
Plant (Oct 05)
0
10 100 1000
Size log (µm)
Warman Cavex cyclones were replaced by a single 15” cyclone. concentrate but failing to be recovered to the final concentrate
There was an immediate improvement with a reduction in the due to the low flotation kinetics. The lean composites ultimately
circulating load from 400 per cent to 250 per cent and the P80 reported to the flotation tailings. This situation was exacerbated
increased from 75 µm to 170 µm. by the progressive coarsening of the primary grind size. To
The progressive coarsening of the grind size resulted in the improve the probability of recovery of the value minerals within
following changes: the lean composites to the final concentrate, the scavenger
concentrate was cycloned, with the coarse and/or higher SG
• The P80 increased from 75 to 170 µm, and the mass per cent minerals reporting back to the primary ball mill feed. The regrind
<10 µm reduced from 30 per cent to 14 per cent. cyclone overflow reported back to the rougher feed.
• Flotation selectivity improved, resulting in lower gold and
sulfide loss to tailings. Froth crowders
• Mill throughput increased from 25 t/h to 38 t/h for the same Froth volume/height adjustment by crowders permits increasing
power draw (ie 35 per cent reduction in power per tonne or decreasing the retention time of solids in the froth as a
milled). The higher throughput rate has permitted separate measure of controlling selectivity. Flotation cells need
campaigning of the ore to provide opportunity for to be selected for the optimal surface area, carrying rate and
maintenance and improved labour utilisation.
concentration ratio for the specific task and particle size range.
• Grinding media consumption, pump wear rates reduced by The surface area per unit volume of the installed rougher and
50 per cent. scavenger cells at Beaconsfield for the required mineral recovery
was high at 0.66 m2/m3, resulting in shallow froth depths and cell
In stream analysis pulping due to lack of froth stability. Transverse froth crowders
In stream analysis (ISA) was reviewed and installed at were subsequently installed lowering the surface area per unit
Beaconsfield for the following reasons: volume in the roughers by 55 per cent and the scavengers by
80 per cent. The use of the crowders resulted in greater flotation
• To maximise gold and sulfide recovery at the target sulfur selectivity, circuit stability and a 40 per cent reduction in frother
grade with varying feed grades. usage.
• To provide consistent feed sulfur content to the bacterial
oxidation circuit to maintain ‘steady state’ operation. Flotation circuit automation
• To reduce operator judgement/intervention. Following the installation of the ISA, the float circuit was fully
• To instantly highlight process problems (reagent blockages, automated to maintain a constant concentrate sulfur grade
etc). irrespective of changes to the feed grade. Cascade control loops
• To reduce assaying costs in the laboratory (one or two adjusting air addition (primary control) and level (secondary
calibration check samples per day versus 18 to 24 process grab control) were set up to:
samples). The contract laboratory was reduced from 24 hour • Maintain the target unit cell concentrate sulfur grade.
coverage to 12 hour coverage saving over $100 000 pa.
• Maintain the target cleaner concentrate sulfur grade.
An AMDEL in stream analyser was installed to measure the
iron and density of the cleaner concentrate and flash float • Manipulate the rate of rougher concentrate recovery as the
concentrate. From the iron and density count rates, an inferred feed sulfur grade varied. This was achieved by using the
sulfide grade was determined and utilised as the process variable pinch valve positioning on the cleaner tailings as the process
for the flotation circuit automation. variable to control the rougher pull rates.
• Maintain a constant scavenger cell pull rate, using scavenger
Regrinding the scavenger concentrate regrind cyclone pressure as the feedback loop.
The mineralogical analysis of the flotation circuit highlighted Refer to Figure 8 for the current flotation circuit and control
that there were lean composites being recovered to the scavenger logic.
Cyclone
overflow Air
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
The result of the ISA installation and subsequent automation oxidises the sulfides. The mechanism is considered indirect
of the circuit has: since the bacteria do not need to attach to any solid surface.
• stabilised the feed grade to the BacOx circuit; The residence time of a continuous bacterial oxidation process
is dependent on two factors:
• provided greater control to lower the final sulfur concentrate
grade and to maximise gold recovery without detriment to • time required for the bacterial population to double, and
the bacterial activity; • time needed to oxidise the material of interest.
• reduced the annual contract laboratory cost; and Due to the time requirement for the bacterial population to
• reduced operator dependence, which was part of the double, the circuit was designed with three primary reactors all
justification to reduce manpower by 20 per cent. in parallel where 75 - 85 per cent of the oxidation occurs, with a
residence time of approximately three days. The slurry from the
The developments within the gravity and flotation circuits three primary reactors is transferred to a secondary, tertiary and
have resulted in a 40 per cent reduction in gold loss to tailings, finally a quaternary reactor in series prior to exiting the circuit,
equating to approximately $2.6 million dollars per annum. as shown in Figure 9. If the residence time in the primary
reactors was reduced there is a potential to ‘wash out’ the
BACTERIAL OXIDATION REVIEW bacteria, ie the bacteria are being displaced from the reactor by
new feed at a faster rate than the bacteria can double.
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). 1
The BacOx process is used to oxidise the sulfides (pyrite and
arsenopyrite) and render the encapsulated gold amenable for Oxidised
Product
dissolution by cyanide. Without oxidation of the sulfides, the gold
recovery via direct cyanidation ranged from five to 60 per cent 2 4 5 6
during the feasibility testing.
Feed
y
Tank
nd
ry
ar
rtia
rn
Se
ate
sulfide minerals. The primary oxidant is the oxygen in the air and 3
Te
Qu
The bacteria utilised are mesophilic iron and sulfur oxidising FIG 9 - BacOx reactor process flow.
cultures. Two major mechanisms have been postulated (Ralph
and Wensheng, 1993) to describe the oxidation of pyrite and
arsenopyrite by the bacteria: The bacterial oxidation plant was to be designed to reach a
level of 95 per cent oxidation processing 2.85 t/h of concentrate
• Direct attachment to the mineral surface, oxidising the containing 27 per cent sulfide. The plant, as designed and
sulfide for an energy source. installed, struggled to achieve 80 per cent oxidation at 1.6 t/h
• Indirect mechanism where the bacteria catalyse the (ie 47 per cent of target). The reasons for the failure to achieve
conversion of ferrous to ferric iron in solution, which then the oxidation performance at design throughput were:
• The design of the agitation system was defective, with the • Installation of spirals on the BacOx residue to recover
agitator blades being positioned more than halfway up the partially oxidised (short circuited) sulfides and liberated
tank, resulting in the bottom third of the tank filling with gold, which were subsequently returned to the regrind/
solids, thereby reducing the residence time for the bacterial gravity circuit prior to BacOx.
numbers to double. • Installation of a weir overflow system between reactors as
• Failure of the gearbox output shafts, internal gears, blades opposed to the requirement for manually operated airlifts for
and agitator shafts occurred. Subsequent engineering review level control, thereby stabilising residence time in each
via strain gauges installed on the reactor blades and shafts reactor.
indicated that the impeller loads acting on the reactor shafts
The current optimal conditions and controls for sulfide
were underestimated by up to eight times. The main cause
oxidation at Beaconsfield are shown in Table 1.
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 TABLE 1
agitator blade. Optimal conditions and controls for the Beaconsfield
oxidation circuit.
Cu(CN)32- → CN- + Cu(CN)22- The cyanate hydrolyses to form carbonate and ammonium
ions:
Variations in operator titrations for free cyanide were up to
1500 ppm on the same sample. Operator adjustment of the CNO- + 2H2O → CO32- + NH4+
cyanide levels resulted in wild swings from 1000 ppm to Caro’s acid also oxidises weakly bound metal cyanides (WAD
10 000 ppm in cyanide concentration where the target levels cyanide), resulting in the metals being precipitated as
were around 2500 ppm. The copper and cyanide level hydroxides.
fluctuations impacted on the subsequent detoxification circuit.
The initial detoxification process was problematic due to
The circuit improved via the following advancements:
erratic solution chemistry, high copper WAD cyanide levels, and
• increased levels of oxidation precyanide leaching, thereby high cyanide levels. The operation of the circuit was streamlined
reducing the competition for cyanide by the intermediate and costs reduced by:
sulfur species;
• increased levels of oxidation through BacOx circuit,
• alkaline preaeration of the slurry prior to cyanidation to
convert residual sulfur to sulfate, reducing SCN- generation; • installation of spirals on the BacOx residue to scalp any
partially oxidised sulfides for re-treatment,
• multiple point dosing of cyanide through the circuit to
control within a fine range; and • alkaline preaeration of the leach feed to reduce thiocyanate
generation,
• installation of two cyanide analysers to monitor the cyanide
concentration via potentiometric determination, with • automation of cyanide monitoring and addition through the
subsequent automatic dosing of cyanide to the individual circuit, and
tanks. • use of hydrogen peroxide as opposed to Caro’s acid due to
the presence of copper catalysing the cyanide oxidation.
ZINC CEMENTATION (MERRILL CROWE)
The zinc cementation process for the recovery of gold and silver CYANIDE MANAGEMENT
from cyanide solutions was introduced in 1890 and was
eventually replaced by CIP from 1949. Zinc cementation is Allstate Explorations became a signatory to the International
generally preferred over CIP for treating high silver:gold ratios, Cyanide Management Code in April 2006.
as well as treating high solution tenors as experienced at The site management have worked in collaboration with
Beaconsfield. ORICA to develop, trial and commission a ‘mini-sparge’ cyanide
Like cyanidation, cementation is a heterogeneous redox system, making the handling and use of one tonne boxes
process. Zinc is dissolved at the anodic areas of the surface and obsolete. The mini-sparge installation has resulted in the risk
the electrons released serve to reduce the aurous ions at the rating to site personnel and the environment reducing to less than
cathodic areas. The overall chemical reaction for the cementation two per cent of the original value compared to the use of one
of gold by zinc is: tonne boxes of cyanide.
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).
ABSTRACT zones – the Southern Zone and the Northern Zone. These contain
similar mineralisations and are separated by a fault. Ore was
Commissioning of the Cannington processing plant began in October
1997. The nameplate capacity (of 1.55 million tonnes of ore per annum) sourced only from the deeper, larger and higher grade Southern
was achieved relatively quickly. Since then (like many mines) the Zone in the early years of the mine. In the last three years ore
operation has strived to increase throughput and improve valuable metal production has been augmented by mining from the Northern
recoveries, to offset declining head grades and lift the value of the Zone. Prior to mining, the mean grades of the Southern Zone
operation. The processing plant now has the capability of treating over were 12.4 per cent Pb, 4.9 per cent Zn and 582 g/t Ag, and the
3.2 million tonnes of ore per annum and in financial year 2005/06 the mean grades of the Northern Zone were 8.9 per cent Pb, 3.0 per
combined concentrate production exceeded 600 000 wet tonnes, cent Zn and 371 g/t Ag.
approximately 80 per cent above concentrate output in the first full year
The ore has been divided into ten mineralisation types, on the
of operation (at the design feed rate). This improved performance was
achieved by making changes and additions to each of the major unit
basis of the Pb/Zn ratio and the gangue (either Fe-rich or
processes – primary grinding, flotation, leaching, filtering and paste fill siliceous). Four mineralisation types dominate mine production
preparation. The changes have included: adding a fines flotation circuit, and efforts are made to blend the ore, underground and on the
expanding the grinding circuit, adding a pebble crusher, switching from surface, to ensure that the Pb and Zn feed grades to the plant are
continuous to batch leaching, building paste conveyors and stable. Blending is also used to control the amount of talc and
decommissioning the column flotation cleaning circuits. This paper arsenic in the feed blend, as high amounts of these components
reviews these changes, outlines the current operating practices and can have an adverse effect on concentrate quality.
discusses the challenges or opportunities that remain. As an example of the Cannington mill feed ore blend, the
mean feed grades for calendar year 2005 (CY 05) are presented
INTRODUCTION in Tables 1 and 2 (for selected metals and minerals). The gangue
minerals are largely non-sulfides (or NSG) and these include
The Cannington Ag/Pb/Zn mine is located in NW Queensland fluorite, talc, quartz and magnetite. Fluorine is present in fluorite,
and is named after a nearby pastoral station. Apart from the talc and some other minor minerals. The ore generally contains a
mine, there is little infrastructure in the immediate area and the relatively small component of gangue sulfides, the most common
nearest provincial city, Mount Isa, is over 300 km by road to the is pyrrhotite. Arsenic in the ore is present as arsenopyrite and
north-west. As a result, the mine is a fly-in, fly-out operation and lollingite. The silver mineralogy is relatively complex. Silver is
most of the employees reside in Townsville, approximately
800 km to the north-east of the mine.
TABLE 1
The key features of the ore are the high silver and fluorine
CY 05 mill feed metal assays.
grades. Obviously the high silver grade is a big positive for the
operation and a significant portion of the mine revenue is derived Element Assay
from the silver. Fluorine is a negative and the recovery of Ag (ppm) 505
fluorine to the concentrates must be minimised; the processing
flow sheet includes extra or specialised stages to reject fluorine Pb (%) 10.2
from the products. Zn (%) 3.34
At Cannington, the ore is extracted from the underground mine Fe (%) 14.8
using an open stoping method. The ore is trucked and hoisted Sb (%) 0.046
to the surface stockpile. The processing flow sheet includes
grinding, flotation, fluorine leaching, dewatering and paste fill As (%) 0.106
preparation. The saleable products from the processing plant are SiO2 (%) 41.0
a Pb concentrate and a Zn concentrate. The majority of the Ag CaO (%) 8.02
reports to the Pb concentrate. The concentrates are trucked to a
MgO (%) 1.47
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 TABLE 2
nine years of operation since commissioning. CY 05 mill feed mineral distribution.
present in solid solution with all the sulfides and in distinct silver
minerals. The silver minerals include acanthite, dyscrasite, TABLE 3
freibergite, native silver and pyrargyrite. The most common (in Annual feed and concentrate tonnages.
CY 05) was freibergite.
Financial year Feed (t, wet) Pb conc (t, wet) Zn conc (t, wet)
FY 97/98† 1 023 130 163 988 61 944
HISTORY
FY 98/99 1 426 823 220 173 96 515
The deposit was discovered in 1990 as a result of drill testing of FY 99/00† 1 632 732 294 584 136 134
magnetic targets by BHP Minerals (Walters and Bailey, 1998).
Mine development proceeded swiftly and the first concentrate FY 00/01 1 837 464 316 779 165 351
was produced in 1997. FY 01/02 2 238 715 376 438 155 416
Commissioning the processing plant was completed quickly FY 02/03 2 382 983 377 992 169 053
by standards of complex base metal operations, however not FY 03/04 2 689 262 406 593 146 641
without encountering problems (Leung et al, 2000). To expedite
the production of concentrate and efficiently utilise resources, FY 04/05 3 183 628 457 317 141 089
the plant was commissioned in stages. The primary grinding FY 05/06 3 114 153 430 458 182 674
circuit was commissioned in late September 1997. The first
saleable Pb concentrate was produced on 25 October 1997. The † Annualised.
plant was fully commissioned by late December 1997. The
primary grinding circuit operated for two weeks at the design The steady increase in mill throughput was achieved by
throughput in November 1997, but ore availability prevented completing numerous capital projects. Many of these were
continuous operation at this rate until mid 1998. Ag recoveries relatively minor de-bottlenecking projects, but two significant
exceeded the design target virtually from day one, which was an plant expansions have been completed, in 2001 and 2003. In
important financial milestone for the operation. Pb and Zn addition to the plant expansions in 2001 and 2003, a significant
recoveries matched or exceeded design targets within six months. capital project was completed in 2001 to increase recovery of
The target recoveries were 80 per cent, 85 per cent and 75 per fine Pb and Ag particles.
cent for Ag, Pb and Zn, respectively. In the original design, the autogenous grinding mill (AG mill)
One significant hurdle encountered during the commissioning and the concentrate filters were deliberately oversized to provide
phase was difficulty rejecting fluorine from the concentrates. The operational flexibility and mitigate the start-up risks (Leung
main cause of the problem was that the feed blend contained et al, 2000). The marginal cost of increasing the size of these
higher than expected amounts of naturally floating talc, which items was relatively minor.
contains fluorine. The fluorine leaching circuit is effective for In mid 2001 mine output was nominally increased from 1.5 to
removing only the fluorite from the concentrate. To remove talc, 1.8 million tonnes per annum, by expanding the mine into the
part of the flotation circuit was reconfigured to act as a upper levels of the Southern Zone footwall (Smith, 2002).
preflotation stage. This step significantly reduced the fluorine Modifications to the paste plant and an expansion of the grinding
content of the Pb concentrate; however, fluorine rejection was circuit were required to treat the extra ore. The paste plant was
still unsatisfactory. modified by installing a novel 350 m conveyor system to allow
In the first months of production, a satisfactory product could delivery of paste to the new part of the mine. The conveyor and
only be made by targeting a high Pb grade (resulting in recovery associated equipment was commissioned in December 2001. The
loss). Residual fluorine in the Pb concentrate (at the target grade) grinding circuit was expanded by adding a pebble crusher. A
was due to fluorine bearing silicates locked with galena that temporary pebble crusher was installed in September 2001 and
could not be removed by leaching. After discussions with the used until November 2002. A permanent pebble crusher was
operation’s customers, the fluorine specification was raised, commissioned in early 2003.
which allowed improved Pb recovery. The processing plant was expanded or upgraded in the second
Since commissioning (like many mines) the operation has half of CY 03 to allow sustainable treatment of 2.4 million
tonnes of ore per annum (some details of the expansion have
strived to raise throughput and improve valuable metal
been reported earlier, see Okely and Fargher, 2004). The 2003
recoveries, to increase the value of the mine and offset declines
project was associated with an expansion of the mine to allow
in feed grades. Over the years, steady progress has been made.
ore production from the Northern Zone. Just prior to the
The annual feed and concentrate wet tonnages are presented in expansion the processing plant was operating near the target rate
Table 3 (note, typical moisture contents are three per cent, seven (of 2.4 million tonnes per annum) and most of the capital works
per cent and nine per cent for the feed, Pb concentrate and Zn were justified on the basis of improved metal recoveries and/or
concentrate, respectively). In financial year 2005/06 (FY 05/06) improved concentrate quality. In early 2004 additional capital
the feed rate exceeded the design of 1 550 000 wet tonnes by was approved to expand the processing plant to nominally
approximately 100 per cent. Between September 2005 and 2.8 million tonnes per annum. These additional works were
February 2006 the average feed rate was 450 t/h (wet), not relatively minor de-bottlenecking exercises.
including periods when the plant was shut down. This gives an Work completed during the 2003 expansion included upgrades
indication of the capability of the processing plant in the future. to the grinding and Pb regrind circuits, installation of a new
In FY 05/06 the combined concentrate production exceeded larger zinc cleaner circuit, expansion of the Pb fines flotation
600 000 wet tonnes, approximately 80 per cent above circuit, de-commissioning of the Pb flotation columns, a new
concentrate output in the first full year of operation (at the design paste conveyor, an expansion of the Pb concentrate filter and an
feed rate). The relative gain in concentrate production is less than upgrade to the reagent system. The major components were
the feed rate gain due to a decline in the feed grades. commissioned in October 2003.
In FY 06/07 the ore tonnage to be processed is expected to be The fine Pb or split float circuit was commissioned in July
below FY 05/06, due to a halt in ore production from the 2001, to increase fine Ag and Pb recovery (Torrisi and Smith,
Southern Zone of mine from May to November 2006. The halt 2003). The split float circuit project comprised installation of
was required to complete an underground rehabilitation program classification stages to separate the fine Pb from the feed, and
to improve ground conditions and ensure employee safety (see installation of rougher and cleaner flotation cells to separately
news release, Anon, 2006). Beyond FY 06/07 annual ore upgrade the fine material. Overall Ag and Pb recovery (to the Pb
production is expected to return to normal levels. concentrate) was improved by installing this circuit.
PROCESSING DETAILS tank). The tank has a residence time of approximately one hour.
It is used to smooth the feed to the flotation circuit and mitigate
Grinding the effect of short grinding circuit stoppages.
Since commissioning the new grinding configuration, the feed
The current grinding circuit consists of two stages, an AG mill
rate has been gradually stepped up from the pre-expansion rate of
followed by a Vertimill or tower mill; see Figure 1. This
configuration has been in operation since October 2003, when about 300 t/h (wet), to over 450 t/h (wet). The increase has been
the Vertimill and associated equipment was installed. The design achieved by increasing the fraction of pebble ports, improving
and commissioning of the Vertimill has been described by the utilisation of the pebble crusher and installing an expert
Jankovic, Valery and Clarke (2006), and selected equipment control system (the Metso OCS). The downside of the
details are presented in Appendix 1. throughput increase has been an increase in the flotation feed
size, from the target 80 per cent passing size (P80) of 90 microns
The grinding circuit feed is stockpiled on the run-of-mine pad
to up to 170 microns (see Figure 2, which shows the average
(or ROM pad). Typically 100 000 to 300 000 tonnes of ore is
stored on the ROM pad, sorted into three ore types – hoisted, monthly data since October 2003).
hauled and high-grade zinc ore. The hoisted ore is crushed The increase in flotation feed size has resulted in a drop in Pb
underground and trucked from the shaft or hoist to the ROM pad and Ag recovery; the correlation between Ag recovery and P80 is
(a distance of approximately 1.6 km). The hoisted ore is shown in Figure 3. A 40 micron increase in feed size results in
predominantly Southern Zone material. The hauled ore is trucked approximately a two per cent drop in Ag recovery and a one per
to and crushed on the surface, and largely originates from the cent drop in Pb recovery. No correlation between Zn recovery
Northern zone. The required blend of ore is stacked onto the and feed size has been observed. The recovery drop is thought to
feeders using a front end loader. The three reciprocating plate be largely due to a decrease in the liberation of the valuable
feeders are located in a tunnel below the ROM pad. minerals in the feed, rather than due to a drop in flotation
During the 2003 expansion the AG mill circuit cyclone pack residence time. Galena liberation in the feed is correlated to P80,
was relocated to the Vertimill circuit. A new pack containing the and finer grinding of plant feed samples in the laboratory leads to
larger diameter Cavex 400 cyclones was installed in closed higher Ag and Pb recoveries. No correlation between Ag mineral
circuit with the AG mill. The expansion also included the liberation and P80 has been observed. The increase in mine
installation of a large surge or conditioning tank between the revenue due to the throughput increase has more than
grinding and flotation circuits (colloquially known as the ‘JA’ compensated for the recovery loss.
Primary Cyclones
Feed
Vertimill
AG mill
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.
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.
Paste fill preparation a number of projects to lift utilisation. Paste plant production has
been limited by availability of tailing in the past.
The paste plant treats the thickened flotation tailing (65 per cent
Liberation analyses of monthly plant products show that
solids) using a batch process, to produce fill for the underground
significant amounts of liberated sphalerite report to the Pb
mine (75 to 80 per cent solids). The tailing is first filtered using concentrate. Typically 70 per cent of the sphalerite in the Pb
two parallel vacuum disc filters. The filter cake is then combined concentrate is liberated. Similarly much of the galena that reports
with cement and water in a spiral mixer, to the required yield to the Zn concentrate is liberated. There is an opportunity to
stress. The cement addition is typically three to five per cent of improve the mine’s revenue by improving Pb/Zn selectivity.
the solids content. The paste is distributed to the underground The improvement in mine output has been significant in the
stopes using a combination of holding hoppers, screw and belt last nine years. While a similar increase in output is unlikely in
conveyors, positive displacement pumps, fully cased boreholes the future, there remain many ways to improve performance.
and distribution pipes.
In FY 05/06 the paste plant produced 1 900 000 tonnes of ACKNOWLEDGEMENT
paste using approximately 58 per cent of tailing. In this year
paste production (and tailings recovery) was slightly less than the The authors wish to thank BHP Billiton for permission to publish
mine requirement. In broad terms the goal is to recover two this paper.
thirds of the tailing to paste. Although progress has been made in
the recent years, the paste plant remains a bottleneck in the REFERENCES
overall mine and mill operation. Improvements have been made Anonymous, 2006. BHP Billiton news release to the Australian Stock
by reconfiguring the plant to allow paste to be directed to Exchange, 26 April.
more than one bore line at a time, by improving mine to mill Greet, C, Kinal, J and Steiner, P, 2005. Grinding media – Its effect on
communications and by increasing paste plant utilisation. pulp chemistry and flotation behaviour – Fact or fiction?, in
Proceedings Centenary of Flotation Symposium, pp 967- 972 (The
FUTURE DIRECTIONS AND CONCLUDING Australasian Institute of Mining and Metallurgy: Melbourne).
Jankovic, A, Valery, W and Clarke, G, 2006. Design and implementation
REMARKS of an AVC grinding circuit at BHP Billiton Cannington, SAG 2006
The operation seeks to improve in three ways: Conference.
Leung, K, Andreatidis, J, Duffy, M and Torrisi, C, 2000. Commissioning
1. reduce the flotation feed size (and increase the liberation of the Cannington project – Creating success, in Proceedings Seventh
the Ag and Pb minerals); Mill Operators’ Conference, pp 123-126 (The Australasian Institute
of Mining and Metallurgy: Melbourne).
2. reduce the cross contamination of the concentrates (by Pb
Okely, A and Fargher, D, 2004. Cannington growth project – A partnering
and Zn), that is, the loss of Zn to the Pb concentrate and approach to design and implementation, in Proceedings
visa versa; and Metallurgical Plant Design and Operating Strategies, pp 623-636
3. increase the capability of the paste plant. (The Australasian Institute of Mining and Metallurgy: Melbourne).
Smith, R, 2002. Conveying of paste backfill – Focusing on zero harm of
At present the Cannington metallurgy group is working on a the environment at Cannington, in Proceedings Green Processing
number of projects aimed at reducing the flotation feed size or Conference, pp 271-277 (The Australasian Institute of Mining and
more precisely, increasing the liberation of the valuable minerals Metallurgy: Melbourne).
in the flotation feed. These include: fine tuning of the grinding Torrisi, C and Smith, R, 2003. Bringing split float to reality – The
circuit expert system, introduction of flash flotation to the Cannington Experience, in Proceedings Eighth Mill Operators’
grinding circuit and bleeding low grade pebbles from the circuit. Conference, pp 73-80 (The Australasian Institute of Mining and
Metallurgy: Melbourne).
One way to increase paste plant utilisation (and make minor Walters, S and Bailey, A, 1998. Geology and mineralization of the
improvements in flotation feed size) is by increasing the utilisation Cannington Ag-Pb-Zn deposit: An example of Broken Hill-type
of the processing plant. There is scope to improve the utilisation of mineralization in the eastern succession, Mount Isa inlier, Australia,
the processing plant and the maintenance group have implemented 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
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.
Ausenco managed the feasibility study with input from third The flow sheet is comprised of:
party consultants and engineers in the design of mining, • ore delivery to a primary gyratory crushing;
geotechnology, metallurgical test work, tailings storage facility,
high voltage power supply, environmental assessment and • primary crushed ore stockpiling;
concentrate marketing. • apron feeder reclaim;
In 2006, PBM commissioned Ausenco to commence detailed
• a SABC grinding circuit;
design, procurement and construction of the project. The two
13 MW grinding mills were purchased in June 2006 with the • rougher and three stage cleaning flotation circuit;
plant scheduled for commissioning in the first quarter of 2008. • rougher concentrate regrind;
• concentrate thickening, filtration and load-out;
PROCESS DESIGN
• direct flow of unthickened tailings to the tailing storage
The process design program consisted of metallurgical and facility (TSF); and
mineralogical studies to define the ore processing characteristics
followed by a concept definition phase where concepts that • recovery of process water from the tailings dam.
would optimise capital cost were developed to suit the location The plant layout was constrained by the topography. Plant site
of the project. selection involved an iterative process with the plant location
The metallurgical performance was a function of the moving several times due to ground condition, proximity to local
mineralogy. The Phu Kham copper-gold mineralisation communities, proximity to open pit mine and environmental
comprises two principal styles: stockwork-disseminated and control considerations.
banded-massive skarn. Primary copper in both mineralisation The final location allows gravity flow of tailings and all plant
styles occurs as predominantly chalcopyrite in the primary zones run-off to the tailings dam. Thus, the environmental impact of the
and chalcocite, digenite and other secondary sulfides in the project is constrained to the mine and adjacent tailings dams
transition ore zone. Skarn ores included the copper and gold areas.
mineralisation contained in pyrite, silicate, magnetite and
haematite skarns, and in quartz-pyrite altered sediments. Both
the stockwork-disseminated and banded-massive skarn ores OREBODY DESCRIPTION
contain relatively high proportions of pyrite relative to the copper
and gold mineralisation. Geology and mineralogy pertaining to metallurgy
Upper zone transition ores were relatively soft with altered In a simplistic sense the host rock is an altered porphyry
copper mineralisation and complex associations with pyrite. This overlaying redbeds. The orebody consists of two zones, thrust
resulted in higher predicted throughputs, poorer flotation and block zones (Figure 2). Within these zones the ore
recoveries and relatively low concentrate copper grades is comprised of stockwork-disseminated sulfide and banded-
considering the dominant chalcocite mineralogy. massive sulfide mineralisation. Overlaying the rock types and
The primary ore zones were comprised of more competent ore, mineralisation events is secondary alteration that has resulted in:
resulting in lower grinding circuit throughputs and better copper
and gold recoveries in flotation. • a copper deficient oxide layer containing gold that is
currently being heap leached (Phu Bia project),
The flow sheet selection was typical of a large copper
concentrator except for the inclusion of a third stage of cleaning • a mixed ore zone of oxide and transition sulfide
and the use of sand mill based regrind. mineralisation and host rock (not included as ore),
• a transition ore zone containing dominant chalcocite Assuming a rougher flotation feed sizing of about P80 106 µm,
(denoted ‘c’ ore), the sizing of the rougher concentrate was expected to be
approximately P80 75 µm. These rougher concentrates require
• a lower transition zone with mixed chalcocite and regrinding. A regrind target of P80 of less than 38 µm is required
chalcopyrite (denoted ‘t’ ore), and
ahead of cleaner flotation.
• a primary ore zone containing dominant chalcopyrite Head samples based on mine production and mineralogical
(denoted ‘p’ ore). classes were ground to a P80 of 106 µm and the MLA extended
The supergene zone is relatively small and for the purposes of liberation analysis technique used to assess the liberation of five
the study was incorporated in the transitional zone. size fractions:
Pyrite is the dominant sulfide mineral in both the transition • >106 µm, 20 per cent of mass;
and primary ore zones resulting in typical sulfur:copper grade
ratio within the economic ore zone of between ten and 12. • <106 >38 µm, 36 per cent of mass;
The term ‘skarn’ is used interchangeably to describe particular • <38 >C5, 24 per cent of mass; and
types of mineralogy within the banded/massive mineralisation • <C5, 19 per cent of mass.
category. The skarn ore exists as a mixture of mostly pyrite and
silica based skarns with some magnetite skarn and quartz-pyrite The size fraction assays indicated a bias of copper deportment
altered sediments. to the finer fractions and non-copper sulfides (pyrite) to the <106
Disseminated stockwork mineralisation was described as >38 µm fraction.
stockwork-disseminated (abbreviated to ‘SWD’) and suffixed as Chalcopyrite was the dominant copper mineral with minor
either upper transition, lower transition or primary alteration chalcocite, bornite and tennantite in primary ore. Transition ore
(abbreviated to ‘c’, ‘t’ or ‘p’, respectively). was dominated by chalcocite, with minor bornite and tennantite.
Banded/massive mineralisation was described by a Approximately 15 per cent of the mineral suite was pyrite.
differentiated ‘skarn’ type, eg sediment (abbreviated as SDM), Chalcopyrite liberation in the <38 µm fractions was
pyrite (abbreviated as ‘PSK’), magnetite (abbreviated as ‘MSK’) 83 per cent reducing to 20 per cent in the >106 µm fraction.
and silicate (abbreviated as ‘SSK’). Likewise, each differentiated Bornite, chalcocite and tennantite showed poorer liberation,
skarn type is suffixed according to alteration type. indicating more difficult metallurgy in treating the transition
A number of rock type based and production period based ores.
samples were assessed using either optical or automated Principal non-Cu binary associations of chalcopyrite were with
mineralogical analysis methods. The mineralogical characteristics non-sulfides (estimated at 70 to 80 per cent) and pyrite (20 to
of these samples suggested that problems would be encountered 30 per cent). Chalcocite showed a higher tendency to be
in the design of a treatment process. The copper sulfide to pyrite associated with pyrite in composites, again presumably due to
mass ratios for these samples, which ranged from 1:5 to 1:14, copper mineral alteration in the transition zone.
were appreciably greater than those displayed for most other In summary, the MLA assessments indicated that:
porphyry skarn deposits worldwide.
Secondary enriched copper sulfide minerals, chalcocite and • transition zone ore with higher chalcocite may have higher
covellite, indicated altered mineralisation within the deposit. Cu mineral and pyrite association and, thus, result in more
Since chalcocite and covellite occurred in close proximity to difficult flotation separation; and
pyrite, often as rims, some activation and subsequent flotation of • supported the optical mineralogy conclusion that a regrind to
pyrite is likely. The principal objective of process design was to approximately 38 µm will provide sufficient liberation to
ensure that adequate flotation differential is maintained between achieve a 23 per cent Cu concentrate grade and reasonable
copper sulfides and pyrite in the rougher flotation stage. recovery.
The low copper sulfide to pyrite mass ratio in the samples
suggested that establishing and maintaining a sufficient Mine schedule
differential flotation rate between pyrite and the copper sulfides
to achieve high concentrate grades would be difficult at high The mineralogy (based on ore type code) of the feed to the plant
copper sulfide recovery. This indicated the need for strong as a function of ore schedule is summarised in Table 1.
suppression of pyrite in the cleaner stages and some selectivity in
the rougher flotation stage. PROCESSING REQUIREMENTS
Based on mineralogical investigations, a flotation feed sizing
P80 in the range of 100 to 120 µm resulted in the copper sulfide Key test work outcomes
minerals being adequately liberated from the non-sulfide gangue
host for rougher flotation. Of the unliberated copper sulfides, the The principal challenge with the Phu Kham ore was the high
majority are locked in structurally simple binary composites with pyrite content and resulting high sulfur:copper grade ratio and
non-sulfide gangue. A typical particle of this class contained consequential difficulties in separating the sulfide minerals.
about 20 per cent by weight copper sulfides and the majority Pyrite must be rejected from concentrate to yield a target
should be readily captured in the rougher flotation circuit. 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
is typically associated with the rejection of some copper and gold Phase 3 of the test work program was undertaken using
and can result in modest metallurgy with some pyrite skarn ores. composite samples representing mining periods defined by the
The feed sulfur:copper ratio for Phu Kham is in the range eight interim mine schedule. The samples were composited based on
to 14 on a blended annual basis, but may be >20 in some the volume defined by the pit outlines and the available core by
locations. The greater the pyrite content, the lower the expected PMB geologists. The Phase 3 program was entirely centred on
copper and gold recoveries. flotation metallurgy and the preparation of samples for
The metallurgical test work program for the feasibility study engineering design test work.
was completed in five phases. Samples for the test work program Phase 4 included all test work conducted for engineering
were selected from within the Central Zone of the orebody by design on flotation circuit products (ie regrind mill specific
PBM geologists based on general criteria agreed by Ausenco and energy) and the feed material (eg materials handling test work).
PBM. During the feasibility study it became apparent that core
Phase 1 included an initial assessment of comminution and oxidation was impacting on metallurgy. To evaluate the
flotation characteristics using samples selected in the basis of ore metallurgy of fresh core, a fifth phase of metallurgy was added to
types and preliminary oxidation profiles. the test work program. The Phase 5 program focused on the
Phase 2 of the test work program was principally comminution metallurgy of the transition zone material that was most
test work, but flotation test work was added during the study to impacted by ex situ core oxidation.
allow a preliminary assessment of the impact of core oxidation.
Comminution test work outcomes
• Phase 2A was comprised of comminution test work on fresh
core from selected areas and ore types, thought to be the The ore characteristics for comminution circuit design were
most competent. These samples were to set the lower defined as a function of ore type and mining period. In addition,
throughput limits of comminution circuit capacity and were samples selected to represent the highest competency and major
sourced from new holes that twinned existing holes to enable ore types (at the middle and base of the proposed pit) were
the use of whole core. assessed. The results are summarised in Table 2, Table 3 and
Table 4.
• Phase 2B was comprised of comminution test work on
samples selected to represent period composites split into The Phase 1 samples (Table 2) were heavily weighted towards
primary ore and did not reflect the less competent material in the
skarn and stockwork ore types. The samples were
mine schedule. Table 3 summarises the comminution data from
composited from the general core inventory by PBM
the SSKp and SWDp ore types in the primary ore and indicates
geologists to represent the mining periods based on an significant variation in the competency of SSKp ore, presumably
interim mine schedule. associated with rock fracture and/or the level of silicification.
• Phase 2C test work was conducted to compare the flotation Table 4 summarises the data selected on the basis of the
response of fresh core with that of the PBM core inventory interim mine schedule. This reinforced the expected lower
(used for the remainder of the test work program). The core competency of the ores to be mined early in project life and
drilled for the Phase 2A samples was used to prepare fresh resulted in decision to design the concentrator for a maximum
(unoxidised) core samples, and the core from the existing throughput of 14 Mtpa to allow for higher throughput and cash
twin hole used to represent ‘oxidised’ core. flow during the early payback period.
TABLE 2
Summary of Phase 1 comminution results.
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
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 (%)
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.
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 The overland conveyor delivers ore from the primary crusher
to the crushed ore stockpile. The conveyor is a single run
Metallurgical test work was also completed on the following: approximately 1.2 km long and 1200 mm wide. A 1350 kW
• ore samples for materials handling and mass flow design, and drive system runs the conveyor at 3.5 m/s for a maximum rate of
• rougher concentrate to determine regrind specific energy. 2500 t/h. The drive system consists of three drives and
accompanying gearboxes.
No test work was conducted on concentrate dewatering or
tailings thickening. The concentrate dewatering test work was The conveyor is covered to limit water runback to the crushing
not conducted due to the absence of pilot plant derived station during the wet season.
concentrates and the size distribution dependence of dewatering The coarse ore stockpile was designed as a conical structure
circuit performance. Tailings thickening test work was not done with two reclaim apron feeders. The cheaper option of the
due to the absence of a tailings thickener in the flow sheet. installation of a single belt feeder and two vibrating feeders was
Tailings deposition and environmental test work (acid mine considered and rejected due to issues associated with belt
drainage) was completed by the tailings dam and environmental feeder design and vibrating feeder operation on potentially
engineers engaged by PBM. ‘sticky’ ore.
Future
Pebble Crushers
Primary
Crusher SAG Mill
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
Trash
Screen Concentrate
Cleaner 2 Cleaner 3 Thickener
Cleaner
Screen
Conditioner
Concentrate
Storage
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
TABLE 6
Summary of SABC circuit grinding mill specific energy values by year.
An IPM pressure filter was selected and discharges filter cake Ore conveying
to a stockpile below the filter. The filter was sized by the vendor
based on experience and test work on a concentrate from a The overland conveyor is designed to process a peak flow of
comparable operation. 2750 t/h and conveys crushed ore from the primary crusher to the
crushed ore stockpile. The overland conveyor is fitted with a belt
The covered storage facility for filter cake has been minimised
condition monitoring system and weightometer for mass flow
to reduce the capital cost associated with the concentrate storage
measurement of ore transferred to the stockpile.
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 Stockpile and reclaim
concentrate to be sheeted temporarily to protect the stockpile The crushed ore stockpile is located just over a ridge. This allows
from rain and reduce dust emissions. the stacking conveyor to project horizontally to the stacking
position.
Reagents The reclaim system consists of two apron feeders each capable
The reagent’s make up and storage facilities included lime slaking of delivering the full feed rate. The apron feeders are located in
based on powdered quick lime supply, a sodium cyanide mixing close proximity to maximise the live capacity of the stockpile.
and storage system and DTC and frother, methyl isobutyl Whilst the reclaim chamber is constructed of concrete, ‘Armco’
carbinol (MIBC), storage tanks. style conveyor and emergency escape tunnels are provided. Dust
control is via a large mobile exhaust fan adjacent the emergency
exit tunnel.
Tailings dewatering and disposal
Both, rougher scavenger and cleaner scavenger, tailings report to Grinding
the tailings collection hopper and directly to the tailings dam
without prior thickening. The layout of two twin pinion drive mills with outboard
discharge end motors is relatively complex if footprint and layout
An improvement in water quality is expected by the use of are to be optimised. Access to the grinding floor by mobile
recovered process water from the tailings dam. This is equipment is provided by a ramp and cutting the mill feed end
considered important in maintaining high selectivity against into the sloping terrain. This reduces the storage area required for
pyrite in the flotation circuit. liner/lifters.
The cyclone feed pumps’ location evaluation study took some
Sampling time to complete. Numerous options were considered with all
The sampling system is comprised of multi-fin, multi stage static options complicated by the location of the motor plinths or the
samplers for the primary metal accounting streams. Secondary motors and the need to split the cyclone underflow between the
streams, the data from which is required to define the flotation SAG and ball mills. The final solution was to put the line
circuit mass balance, are sampled using static dual fin launder overhead between the adjacent SAG and ball mill drives. Access
samplers. Tertiary streams, the data from which is required for to the cyclone feed pumps, motors and lines and mill motors and
process monitoring, will be sampled using pressure pipe gear boxes is by mobile crane.
samplers.
All the samples are pumped to a Courier on-stream-analyser Pebble crushing
for continuous analysis and monitoring of the plant performance. The pebble crushing circuit is yet to be detailed. The conveyors
The Courier return samples are gravitated back to the process may be installed for mill start-up to allow the trommel oversize
point of origin. to be recycled and the current pebble jet return system to be
The design and layout currently makes space allowance for the removed.
future installation of a particle size analyser on the classification
cyclone overflow.
Flotation plant
DESIGN AND LAYOUT 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
General elevation of the flotation circuit thus saving capital.
The plant site is located on a ridge (Figure 6) and slopes from the The individual flotation cells are elevated on low steel
coarse ore stockpile to the tailings discharge. The sloping site structures that enable gravity flow of tailings away from the plant
presents some advantages due to the use of gravity flow and area without the use of tailings discharge pumps.
reduced pumping requirements. However, the earthworks are The compact nature of the flotation circuit resulted in
relatively complex to set the various relative levels and optimum restricted mobile maintenance access to the flotation circuit. As
slopes required by the layout. a result a tower crane has been allowed for to provide
maintenance crane access. The selected tower crane has a 55 m
Primary crushing boom length with a maximum lifting capacity of 3.6 tonne when
the boom is fully extended.
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 Regrind
crusher is an open face design (eg Cadia) rather than the The selection of the regrind circuit was an iterative process with
in-ground variety (eg Telfer, Ernest Henry). a number of various regrind technologies being considered. The
The crusher discharges directly to a heavy duty sacrificial feasibility test work indicated that a specific energy requirement
conveyor that delivers ore to the overland conveyor. An apron of between 8 kWh/t and 10 kWh/t was sufficient to produce a
feeder was not included to reduce capital cost. product size with a P80 of 38 µm.
Originally it was envisaged that five Metso 350 kW SMDs Maintenance access to the concentrate handling area is
would be utilised in the Phu Kham regrind circuit. The limited provided by the site tower crane.
available footprint and the potential for better mineral liberation,
grade and recovery associated with finer grinding culminated in Tailings disposal and water management
the selection of a single 2600 kW M10000 ISAMill for the
regrind circuit. The lack of real estate, the minimum capital approach to the
project, and the opportunity to improve the quality of the process
The proposed regrind circuit will operate in open circuit with water resulted in the exclusion of a tailings thickener. In addition,
rougher concentrate being submitted to the regrind stage, prior to the positive water balance and monsoon seasons dictated that
cleaner flotation. Cleaner scavenger concentrate can also be harvesting and managing process water could be conducted at
pumped to the regrind circuit to assist with middlings liberation. the tailings storage facility.
The regrind circuit is located adjacent to the flotation circuit The plant has two tailings streams, rougher tails and cleaner
and maintenance access is provided by the tower crane. scavenger tails. Both the rougher tails and cleaner scavenger tails
are individually sampled, prior to being combined in the final
Concentrate handling tails collection hopper.
The fall from the plant to the tailings storage facility (TSF) is
The concentrate handling area is located adjacent to the flotation approximately 150 m. The process plant tail streams gravitate to
circuit and is approximately four metres lower than the flotation the TSF via a series of break tanks in order to control the slurry
circuit. This area is comprised of a high rate thickener and velocity and vent line pressure to atmosphere.
vertical plate pressure filter. Concentrate is stockpiled in a The tailings reclaim water is harvested using pontoon pumps,
storage shed and loaded to concentrate trucks by front-end which transfer the reclaim water to a decant water transfer tank.
loader.
As with the tailings disposal system, for the decant water
The concentrate thickener is fitted with a scum recovery return system there is approximately 150 m of head from the
system. Froth scum is returned to the thickener feed. Thickener surface of the TSF to the plant. In order to minimise the pumping
overflow is collected in a separate thickener overflow tank and head required, the decant water is returned to the process plant
pumped to the SAG mill discharge for use as mill dilution water. via a series of water transfer stations.
APPENDIX 1
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.
ABSTRACT Of course, what higher quality mill feed means will vary from
operation to operation. In some cases it is finer fragmentation, in
Metso Minerals Process Technology Asia-Pacific (MMPT-AP) has been
working with many mining companies worldwide performing ‘Mine to others it is well blended for grade and lacking in contaminants
Mill’ or Process Integration and Optimisation (PIO) studies. MMPT-AP and it can even indicate that certain ore types are, in fact, not
has developed a proven methodology to improve the efficiency of the profitable and should be considered mineralised waste.
mine-mill interface and gain maximum benefit. The PIO methodology Yet for such a simple concept, the application of Mine to Mill
involves rock characterisation, mathematical modelling and simulation to has proven to be very difficult. The staff members of Metso
generate a list of operational and control changes for both the mine and Minerals Process Technology Asia-Pacific (MMPT-AP) have
concentrator. been involved in a significant portion of these efforts either as
Following the methodology, the list of recommendations is developed MMPT-AP projects or in our former roles as operations and
relatively easily; the more challenging part depends on the company. research engineers. What have we learned over the past decade?
Success of a PIO project relies on the ability of an organisation to That the issues remain human ones and the engineering
implement change – both operationally and culturally: the human challenges are readily overcome.
element.
MMPT-AP has developed a proven methodology for applying
When a PIO project is started, there is initial interest and commitment
from the company. But how can this be maintained? How does an
what we call Process Integration and Optimisation or PIO. PIO
operation know if the benefit continues? MMPT-AP is now developing reflects the fact that optimising concentrator feed goes beyond
long-term relationships with clients to ensure that the increased run-of-mine (ROM) fragmentation and considers all aspects of
production and improved efficiency continues. Direct and indirect improving mill performance from throughput, recovery and final
measurements are used to determine the key factors involved and to concentrate grade to lower operating costs.
understand the cause/effect relationships. Simulation tools also can be However, almost all of our clients are focusing their attention
used to compare current with recommended practices – so when material on increased throughput, so the examples included in this paper
properties vary over time, the benefits can still be quantified. reflect that interest.
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 PIO METHODOLOGY
variable or poor concentrator performance, operations can predict when The methodology involves a number of steps: benchmarking,
production will be affected and put controls in place to prevent it from rock characterisation, measurements, modelling/simulation and
happening. This requires support and commitment from the company or where required, material tracking. A PIO project is normally
‘buy in’. Ensuring this buy in continues is the challenge facing every
company involved in PIO work – but the benefits are well worth it.
comprised of a number of site visits spaced over a few months.
The first site visit is to establish current operating practice,
This paper will review MMPT-AP’s experiences in implementing PIO initiate rock characterisation and collect measurements of blast
projects in the long term: from both the consulting and operational
viewpoints. MMPT-AP’s involvement with a number of companies will
fragmentation and mill performance. This is followed by
be discussed as examples of what works and what does not. 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,
INTRODUCTION monitor the results and ensure the improvements are maintained
over time.
Believe it or not, the term ‘Mine to Mill’ has been used in the
mining industry for over a decade with many operations dabbling
Benchmarking and process audits
in the concept and resulting in some successes and many failures.
The idea is a simple but effective one: how can the various stages The first step of a PIO project is to benchmark the current
of the mining process be aware of possible upstream or practices by auditing the operation and control of the blasting,
downstream issues and work together for greater efficiency? This crushing, grinding and flotation processes.
typically relates to the mine-mill interface or how the mine can The quality of blast pattern implementation is assessed and the
produce a more suitable, higher value or higher quality resulting ROM fragmentation measured using image analysis.
concentrator feed. 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
1. Manager – Process Integration and Optimisation, Metso Minerals
Process Technology Asia-Pacific, Unit 1, 8 - 10 Chapman Place,
of operational changes in the mine or concentrator on the entire
Eagle Farm Qld 4009. Email: adrian.dance@metso.com process.
2. FAusIMM, General Manager, Metso Minerals Process Technology
Asia-Pacific.
Rock characterisation
3. MAusIMM, Manager – Development and Process Engineering, Once the current operating performance has been measured
Metso Minerals Process Technology Asia-Pacific. under one set of conditions, the effect of changing rock
properties can be quantified. This involves rock characterisation
4. Manager – Process Control and Information Engineering, Metso or defining domains of similar properties. In this paper, this will
Minerals Process Technology Asia-Pacific. be discussed in terms of blast fragmentation. The process could
5. Crushing Process Technology Engineer, Metso Minerals Process equally be applied to other quality parameters such as flotation
Technology Asia-Pacific. performance, leach recovery, lump to fines ratio, etc.
The MMPT-AP methodology for rock characterisation utilises importantly, as the rock properties have now been well
simple and inexpensive measurements that can be performed by characterised and the processes modelled, the variable nature of
trained site personnel. Quite often the measurements are already the material can now be compensated for.
being collected by the operation. The advantage of simple
measurements is the amount of data that can be collected in a Modelling and simulation
very short time frame, as the samples do not require shipping to
an outside laboratory. When attempting to characterise an entire The measurements collected while at site are combined with the
orebody, the density of data is very important. rock characterisation domains to model the complete process
For rock characterisation, MMPT-AP use measurements of chain. MMPT-AP use these data to develop site-specific models
rock strength (point load index, PLI and/or UCS) and rock of blast fragmentation, crushing, grinding and flotation. This
structure (rock quality designation, RQD and/or fracture allows customised blast patterns to be developed that optimise
frequency). Both PLI and RQD measurements can be taken on both crushing and grinding performance. For each domain, blast
drill core and point load tests can also be performed on irregular designs are recommended to generate the optimal fragmentation
shaped samples of material. size for downstream processes. This may involve an increase or
decrease in energy level (or powder factor) depending on the
The PLI value can be correlated to unconfined compressive rock characteristics of each domain.
strength (UCS) as well as the JKMRC drop weight test
parameters A and b. The drop weight parameters are necessary in The objective is to minimise the overall cost for the entire
order to model the crushing and grinding circuits. Therefore, the process by distributing the energy required sensibly and
effectively where it is best applied. Near-field vibration
use of the point load index allows sites to characterise their rock
measurements and models are used to confirm that pit wall
properties quickly and easily while still making use of the
stability issues are considered in the blast designs.
sophisticated grinding models that are available.
In addition, the crushing and grinding models allow the impact
The rock structure is represented by the RQD value that
of operational and control strategies to be investigated. For
indicates the fracture frequency present in the drill core. This
example, what is the best closed-side-setting to operate my
measurement is routinely taken at operations for geotechnical
primary crusher at in terms of production and product size? What
purposes but has been shown to be very useful in blast
target load should I use in my SAG mill when processing this
fragmentation modelling in the absence of detailed rock mass
domain? What is the tendency for this material to be SAG mill,
structure mapping.
ball mill or recycle crusher limited?
Once the PLI and RQD data are available, the range of rock
All of these questions can be evaluated using the model of all
properties are mapped out and domains are defined (see
the stages of comminution (blasting, crushing and grinding).
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. BENEFITS OF PIO TO AN OPERATION
The domain structure shown in Figure 1 follows the existing It is quite extraordinary how the mining industry has allowed
ore type characterisation used by the site but expands further into uncertainty and unpredictability to be an accepted part of normal
areas of structure (coarse, medium and fine) and strength (soft, operation. It is still common for the concentrator to be unaware
medium and hard). The ranges of strength and structure used are of changes in material properties and how they influence mill
based on the variability of the orebody. The more variable the performance. In addition, a lack of accurate material tracking so
PLI and RQD values measured in the orebody, the greater that prediction of when these changes will happen is very
definition required for domains. In the example given in unlikely.
Figure 1, the RQD values were divided into ranges of zero to This perhaps could be understood before the application of
30 per cent, 30 - 60 per cent and >60 per cent while the PLI technologies such as online image analysis, GPS equipment
values were divided into 0 - 3 MPa, 3 - 6 MPa and >6 MPa. tracking and sophisticated geological modelling software.
Once the domains have been defined, different blasting However, these tools are now readily available and can answer so
practices, crushing and milling operational strategies are many outstanding questions: ‘Why is the mill tonnage down?’,
established. Through modelling and simulation studies, the ‘How long will it last?’ and more importantly, ‘What can be done
impact of blending different domains can be reviewed. Most to prevent it in the future?’
D-FS D-FM
RQD 0 - 30
Diorite
RQD 30 - 60
D-CS D-CM D-CH
RQD >60
RQD 0 - 30
Intermediate IT-MM
RQD 30 - 60 IT-MH
Tonalite IT-CH
RQD >60
RQD 0 - 30
Young YT-CH
RQD 30 - 60
Tonalite
RQD >60
Time and time again, the explanation of poor performance is effects of material hardness and size. It is very important to
related to ‘bad ore’ without any measurements to back it up. We isolate the two as they can often be confused. In addition,
as ‘metal manufacturers’ operate in a manner that other measurement of hardness and size must be done objectively and
manufacturing sectors could never do – and yet, we still manage if possible, automatically. At one operation, the changes in mill
to make a profit (particularly in times of high metal prices). tonnage observed were related to rock hardness – high tonnage
Imagine the increase in efficiency possible through the was ‘soft’ ore and low tonnage was ‘hard’ ore; and of course,
understanding of what material parameters are important to there was nothing that could be done about that. After installing
concentrator performance and the accurate measurement, an online image analysis system and measuring rock hardness
tracking and control of those parameters. This is achievable by independently, it was determined that hardness had almost no
every operation worldwide. effect – the changes in mill tonnage were due almost entirely to
feed size. An example of how feed size and the amount of ‘fines’
Performance improvements affected mill tonnage is shown in Figure 2 over a 24-hour period.
Needless to say, the performance improvements experienced as a It can be seen in Figure 2 that feed size controlled mill
result of PIO projects have been significant; typically, five to behaviour – more fines, higher tonnage. There was almost no
20 per cent increases in throughput through controlled blasting effect of hardness for this operation. As far as the mill was
and crushing practices. In fact, these improvements can occur concerned, if it was the same size, it processed at the same rate.
without any increase in operating cost but instead, through the Actually measuring material properties – online and
understanding of where to add energy and where to reduce automatically – led to an understanding of what was actually
energy. In one case, the concentrator experienced a throughput important. In addition, it provided greater predictability. It was
increase of 33 per cent at no added cost. found that X per cent fines would always lead to Y throughput.
When finer ROM fragmentation is required through higher- In other words, prepare the feed material and the mill
energy and controlled blasting, quite often the benefits solely in performance was entirely predictable.
the mine outweigh any added blasting costs. That is, higher
This is very empowering. Instead of a situation where
equipment availability, better digging rates, lower haulage costs,
improved crusher production … the list goes on. With the current concentrator performance was changing due to an abstract
mining and crushing methods, finer material tends to be easier hardness factor, it was changing due to material size – that is
and cheaper to handle. entirely under the control of the operation. In other words,
Mother Nature makes the orebody hard or soft, but blasting and
Of course, the benefits downstream in the concentrator are
magnified through consistent, high quality feed that leads crushing practices make the material big or small. (Stockpile
to a stable and more predictable performance. In the PIO design and operation can moderate these effects somewhat.) By
methodology discussion above, the definition of blasting knowing how feed size influenced mill performance, it was then
domains allows different blasting and crushing strategies to be possible to regulate the feed size to achieve the highest possible
developed. The reason is that SAG and AG mills are no different mill throughput – every day of the year.
from the rod and ball mill circuits operated in the past – all Using MMPT-AP’s PIO methodology detailed above, it is also
grinding circuits benefit from consistent and prepared feed possible to predict mill performance until the end of mine life
material. After compensating for the harder/softer and blockier/ based on drill core measurements of PLI and RQD with models
finer nature of the material, the end product through blasting and of blast fragmentation, crushing and milling circuits linked
crushing is more consistent – leading to steady and predictable together. Such predictions of mill tonnage can be entered into
concentrator performance. every ore block of the mine planning software to create a
What is the value to an operation to say with confidence: ‘We geometallurgical model.
know what the mill tonnage will be tomorrow, next week, next
year and until the end of mine life as we can control it.’? Methods to monitor and control
Introduction of measurements As described above, measurements are the key to understanding
what material properties are important to concentrator
Before concentrator performance can be controlled, the performance and in what priority. It is typical that rock
important material properties affecting mill performance must hardness and feed size are the two key parameters affecting
first be defined and more importantly, measured. In this paper, concentrator throughput; however, other factors such as clay
the discussion will focus on mill throughput and be limited to the content and specific gravity may also have an effect.
60 1800
1600
50
1400
Weight % in Fraction
40 1200
% Fines
1000 % Medium
30 % Coarse
Mill t/h
800
20 600
400
10
200
0 0
Material hardness can be measured in a variety of ways. For transponders for use in blasted material monitoring. These RF
blast domain definition, MMPT-AP relies on point load index tags are small, robust and inexpensive and can be dropped into
and the ability to correlate the PLI values with drop weight test the blasthole stemming column or placed on the muck pile
parameters. This is very useful for simulation as the models surface post-blast (see Figure 3).
require estimates of rock strength and breakage characteristics. The tags are not powered but are detected by antennas placed
For relative comparison, other methods can be used such as over conveyor belts. Each tag has a unique identifying number
drilling penetration rates and blastability indices. What is that the antenna transmits to a remote computer for recording
important is that the values used are reproducible and not along with the date/time. By noting the initial position of each
subjective (that is, potentially prejudiced or biased). tag (ie blasthole ID), an estimate of the origin of the material
ROM fragmentation size measurements are now possible using being processed can be made. By tracking the actual material
commercial image analysis systems that have been around for itself, concerns about estimating stockpile volumes, mixing and
the past ten years. The strengths and weaknesses of these retention times can be avoided.
systems have been well documented and, provided the user
understands them, the systems provide reliable and accurate Throughput forecasting and geometallurgical
data. (Figure 2 is a good example of applied image analysis
measurements.)
modelling
As most operations blend different ore sources in their mill The measurement and understanding of how rock hardness and
feed, an understanding of where the material originates is size affect mill performance is needed to optimise ROM material
absolutely essential. This allows the actual concentrator properties on a daily basis. Operations personnel can adjust the
performance on this material to be reported back into the blast pattern, explosive distribution and primary crusher operation
geological block model for comparison with any estimates or in order to achieve a consistent feed size to the mill. If necessary,
prediction of similar material nearby in the future. Material harder and softer ore types can be blended or campaigned
tracking is a necessity for geometallurgical modelling. separately under different grinding circuit operating conditions.
There are currently two methods to track material movements The essential feature of this scenario is the ability to predict
from the mine to the concentrator: model-based and sacrificial how the material will perform. That includes the situation when
instruments. the material was not treated correctly – that is, when tailored
The first method of tracking material involves the development blasting and crushing practices were not followed (as can happen
of a software program to record the movements of material from at times).
the open pit or underground to the intermediate or long-term The ability to predict concentrator performance under different
stockpiles, through the crusher and coarse ore piles and into the operating conditions and well into the future allows different
concentrator. Each stockpile can be represented by simple perfect mine plans to be investigated. For example, what will be the
mixing models or if necessary, more sophisticated three- demand on trucks and shovels if an operation decides to reduce
dimensional models. the blast energy? What will happen over time as the material
The models allow the effect of material mixing and delays to properties change? This is the aim of throughput forecasting and
be incorporated and provide a reasonably accurate estimate of geometallurgical modelling.
mill feed. Such a system provides much greater definition or After defining representative blasting domains, MMPT-AP
detail on changes in concentrator feed and can be updated as simulate the impact of different rock properties and operating
frequently as every 15 minutes. A daily summary will not conditions using the customised models of blasting, crushing and
provide such a degree of detail. grinding. Mapping the orebody using rock strength and structure
Another method for tracking material movements being measurements allows the mill throughput to be simulated as far
employed by MMPT-AP are ore block markers called SmartTags. into the orebody as the drill core has penetrated. An example of
MMPT-AP has developed passive radio frequency (RF) this is shown in Figure 4.
POST BLAST
[ID,x,y,z]
POST CRUSHER
DETECTOR
[ID, time]
RECEPTOR
DETECTOR
[ID, time)
FIG 3 - Use of radio frequency tags to track material movements (Dance et al, 2006).
In Figure 4, point load index and RQD values are plotted different mine plans and blasting scenarios to be simulated
against mine elevation. It became very clear that in this case, quickly and easily.
rock strength and structure would vary quickly between 800 and Figure 5 shows an example of throughput forecasting where
650 m in elevation and then stabilise. In the early stages of the daily concentrator tonnages over a three and a half year period
operation, blasting energy could be minimised but in order to are plotted as blue diamonds. For this operation, throughput
sustain expected mill throughputs, would need to increase as the varied from below 4000 tph to over 7000 tph due to a wide range
mine deepened. In other words, the initial blast patterns would of ore strengths and structures mapped by PLI and RQD values.
not adequately fragment the material to maintain mill tonnage MMPT-AP developed a throughput model based on rock
and adjustments would need to be made. In addition, simulations strength and structure as well as blasting conditions. (In this case,
indicated that additional mill power would need to be installed in ore grade was also used to account for hardness variations not
order to keep the final grind size from coarsening. picked up by the PLI results.) Material tracking was necessary in
The data presented in Figure 4 – along with a detailed mine order to estimate the daily distribution of mill feed from the
plan and blast conditions – can be used to generate a trend of different ore sources as well as assign PLI and RQD values to
expected mill throughput and final grind size. This allows each source.
120 12
RQD
RQD
Is50
100 10
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).
The model predictions are shown in Figure 5 as an upper champion comes into play: ‘What can we do straight away based
and lower ‘operational envelope’. The operational envelope on the comments and conclusions made by the consultants?’ and
represents small changes in the input parameters that would ‘Why are we waiting for the report? Why don’t we act
result in higher or lower expected mill tonnage. For example, for immediately?’
the upper boundary, the rock properties were softened and the In order to sustain a level of interest at site, regular
blasting energy increased within an expected variation. The maintenance visits need to be scheduled by the outside group.
reverse was done for the lower boundary. Within these expected This is not only to ensure the benefits continue, but also to
variations for normal operation, the predicted concentrator motivate people despite the belief that ‘The ore has changed
tonnage varied by 800 tph. Over the three and a half year period,
since you were here last’.
the average relative error for the model was 0.84 per cent or
±46 tph.
Key performance indicators
HOW TO ENSURE SUCCESS To reinforce the changes required, key performance indicators
(KPIs) – or how senior staff are measured – need to be aligned
Over the past decade, MMPT-AP personnel have been involved with the project’s objectives. That is, KPIs for the mine
in many Mine to Mill or PIO projects – some successful and personnel need to include the quality parameters identified at the
some not. In every case, issues were identified and solutions start of the project. This is where measurement of these
presented to the operation that represented significant properties is so important. For example, if finer fragmentation of
improvements in throughput or efficiency. Following a certain ore types is found to be vital for concentrator
prescribed methodology, the engineering solutions fall out and performance, online measurement of ROM fragmentation size
become clear very quickly. Whether or not the operation decides and the monitoring of these measurements can be used to set up a
to implement the recommendations is typically the first hurdle KPI and a success rate. Of course, it is equally important when
as, at times, the solutions are a step-change from current setting up this KPI to quantify what is expected from the mine
operating practices. The second challenge is sustained and not deviate over time (or ‘move the goal posts’). It cannot be
implementation over time. When MMPT-AP has been involved described as ‘good’ fragmentation but rather ‘a P80 of no greater
in successful projects, a number of common traits were in place: than…’ as measured by an automated system.
management support and the presence of a site champion,
assistance from an outside agency and ownership of the situation An interesting concept is establishing a contract between the
through changes in key performance indicators. 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
Site champion and management support requirements are and, in the event of not meeting these
The selection of a site champion is a bit of a misnomer – a true specifications, the penalties involved. It would certainly change
champion cannot be selected, they must take it upon themselves the economics of a mine if it had to sell its product to the
to support the project wholeheartedly despite the complaints and concentrator rather than simply send whatever is available –
resistance of others. They cannot be told to be the champion. worth considering no matter how radical it may sound.
They must also possess a combination of technical skills (to be The mine needs to truly believe that the important parameters
able to explain the benefits) and persuasive skills (in order to for downstream performance are under their control and that they
negotiate with the parties involved). It is all about dealing with can satisfy the agreed-to KPI and perhaps even exceed it. If they
cultural change and how the key personnel handle it. The do not consider it important, then the project is likely to fail.
champion must also have sufficient seniority (or ready access to
seniority) so that they cannot be ignored. An easier option is for Financial incentives
managers to voice their total support of the work and agree
completely with the decisions of the champion. For the work However crass this may sound, if an operation is to benefit
environment is not a democracy and ultimately whatever the boss through change then it is sensible to let the employees share in
says, goes. the wealth created. In general, there is no greater motivator than
Initial expectations are that if the message is made clear and money. This can be as simple as establishing bonuses based on
the benefits demonstrated, people will decide themselves to KPIs related to mill feed quality (once again, well defined). This
adopt the new practices and support the project. Unfortunately, will raise the importance of such KPIs as there is a financial
this is rarely the case and the voice of senior management incentive.
ultimately needs to ring resoundedly in everyone’s ears that MMPT-AP has observed the approach of a general workforce
voluntary support is not an option. The economic benefits of PIO gainsharing program and unfortunately, it was not successful.
projects can gain the acceptance of senior personnel as they see When someone does not truly believe they can make a
more of the ‘big picture’ than others do. However, commitment difference, it is hard to motivate them to implement change.
or ‘buy in’ from workers on the grinding floor is unlikely unless Work ethic is one thing, but implementing change in the face of
financial incentives are put in place (see below). peer pressure is another; particularly in a unionised environment.
However, monetary bonuses targeting the key players in the mine
Outside assistance and concentrator are a persuasive tool.
Despite the greatest intentions, PIO efforts rarely succeed
without some level of outside assistance or agency. This is
Regular audits/benchmarking
because generally in-house projects do not attract enough In all the PIO projects that MMPT-AP has been involved with,
attention from upper management. The advantage of dealing most implement the necessary changes and immediate successes
with a consultancy group experienced in this field is the are gained, a few continue over time but it is rare that the benefits
confidence their personnel bring with them to site. They have all are maintained in the long term. To address this, MMPT-AP is
done this before and can quickly identify where the benefits can now strongly recommending that regular site visits be agreed to
be made and put systems in place to measure the gains. at the outset of a project.
You can almost measure the growing level of interest while the One of the reasons for the long-term difficulties is continuity.
group are at site – and feel the collective sigh of relief when the A champion involved with a successful PIO project can be
consultants depart to generate the report. This is where the rewarded with a promotion or a position at another operation.
Alternatively, they may simply decide to move on. What As an example, the data presented in Figures 6 and 7 cover the
continuity is in place to ensure the benefits are sustained? The period from 2004 to mid-2006 for an operation that was involved
outside consultancy – with their standard operating protocol – in a PIO project in late 2003. In both figures, the daily
can provide that continuity. MMPT-AP are working more and concentrator tonnage, average mill feed rock strength (RQD
more with operations establishing support contracts with a value) and blasting powder factor (in kg/t) are shown. In late
regular number of site visits per year. Perhaps this will increase 2003, it was identified that rock strength played a significant role
the success rate in the long-term of PIO projects. in blast fragmentation and to achieve a more consistent ROM
During these site visits, process audits and benchmarking will size, blasting patterns should be adjusted to match the RQD
reveal if the systems put in place (ie blast implementation, value. In Figure 6 for 2004, it is clear that powder factor
crusher operation) are still working. Has the material changed followed RQD very closely and, as a consequence, mill
enough in its characteristics to warrant further changes? The throughput was reasonably steady.
modelling/simulation tools can be used again to verify if the
benefits have degraded or if opportunities lie elsewhere. Finally, Therefore, a PIO project was successfully implemented with
it revives interest in the project and maintains communication blast patterns modified to suit rock conditions and a consistent
between all parties. mill feed size resulting in a steady and higher mill tonnage.
8000 1.4
Mill Tonnage
Powder Factor
7000 RQD
1.2
Concentrator Tonnage (t h)
/
6000
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
8000 1.4
Tonnage
Powder Factor
7000 RQD 1.2
Concentrator Tonnage (t / h)
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.
The data in Figure 7 shows that in late October-early November characterisation step defines blasting domains and allows
2005, the rock structure became consistently blockier and steadily different blasting and crushing strategies to be developed.
worsened to the end of 2005 and into the middle of 2006. As the PIO is about empowerment: by measuring and understanding
material became more difficult to fragment effectively using the what parameters affect concentrator performance, site personnel
same blast patterns, concentrator throughput was affected. now have the knowledge to make a difference. Following a
This is the point where regular audits are useful. With the prescribed methodology, the engineering solutions are quickly
material becoming more difficult to process, what is the identified while the human issues are more difficult to overcome.
decision? Should further changes to blasting and crushing When MMPT-AP has been involved in successful projects, a
practices be trialled or not? With measurements of how much the number of common traits were in place: management support
material properties are changing and how long the conditions and the presence of a site champion, assistance from an outside
will continue (based on drill core data), a decision can be made agency and ownership of the situation through changes in key
that makes economic sense for the entire operation. For the performance indicators.
personnel involved, where are the KPIs? Where are the The ultimate objective of any Mine to Mill exercise is for the
individual financial incentives to investigate this problem and to mine and concentrator to communicate with one another and
implement further changes? better appreciate each other’s problems and requirements. Also,
The ultimate objective of any Mine to Mill exercise is for the to use the economics of the entire operation as the basis for any
mine and concentrator to communicate with one another and decisions involving operating practices.
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 REFERENCES
reducing powder factor for this blockier material demonstrated Dance, A, 2001. The importance of primary crushing in mill feed size
on the concentrator throughput, there might be scope for optimisation, in Proceedings SAG 2001, Vancouver, Canada.
alternative blasting designs and/or crushing practices. 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,
SUMMARY Vancouver, Canada.
The concept of Mine to Mill has been applied in the mining Renner, D, La Rosa, D, DeKlerk, W, Valery, W, Sampson, P, Noi, S B and
industry now for over a decade with some successes and many Jankovic, A, 2006. AngloGold Ashanti Iduapriem mining and
failures. MMPT-AP has been involved with many of these milling process integration and optimisation, in Proceedings SAG
2006, Vancouver, Canada.
projects and developed a proven methodology called process
Tondo, L A, Valery, W, Peroni, R, La Rosa, D, Silva, A, Jankovic, A and
integration and optimisation or PIO. The methodology involves
Colacioppo, J, 2006. Kinross’ Rio Paracatu Mineração (RPM)
benchmarking, rock characterisation, measurements, modelling/ mining and milling optimisation of the existing and new SAG mill
simulation and where required, material tracking. The rock circuit, in Proceedings SAG 2006, Vancouver, Canada.
investigation of aggregating paramagnetic particles occurred. The mineralogy of the gold that resulted in it having a high
Svoboda (1987) gives a very good outline of the factors effecting magnetic susceptibility is not clearly demonstrated. Low levels
magnetic aggregation of paramagnetic particles. From a practical of gold make mineralogical analysis problematic. However,
perspective most of the literature outlines laboratory experiments explanations by the above authors range from: gold in sulfide
or theoretical modelling of the process, and do not generally minerals, gold in silicates, gold in oxide minerals, gold
apply the process to the valuable paramagnetic sulfide particles containing iron or gold with iron or iron oxide coatings.
that are recovered by flotation. Nor does the extensive literature Martinez (1988a, 1988b) applies a coating of colloidal iron or
apply the process to flotation. However, the theoretical and iron hydroxide to ores and tailings to magnetise the gold
laboratory basis is equally applicable to sulfide as well as particles so that they can be recovered by magnetic separation.
non-sulfide minerals.
Gold may also be closely associated with sulfide minerals such
Engelhardt, Ellis and Lumsden (2005) have clearly as chalcopyrite, galena, bornite, pyrite and arsenopyrite (Henley,
demonstrated that magnetic aggregation of fine paramagnetic 1975). Some of these minerals are paramagnetic, so that gold
particles can be used to increase the flotation recovery of the fine
may be concentrated magnetically if the associated sulfide can be
paramagnetic minerals chalcopyrite and sphalerite. Their paper
summarises the literature that outlines the theoretical basis of concentrated magnetically.
aggregation of fine paramagnetic particles. Bheemalingeswara (1995) has shown that the coating of gold
particles with iron oxide/iron hydroxide gave the particles
Gold mineralogy sufficient magnetic susceptibility to be concentrated in the
laboratory by magnetic separation.
Gold, particularly when found in base metals ores is often In summary, gold’s disposition in the ore as well as in the
recovered by froth flotation. Gold mineralogy, whether in base mineral flotation slurry will determine whether the gold is
metal ores or other ores, can be complex. Though mostly amenable to magnetic methods. Other researchers have shown
occurring as gold metal, rather than a gold mineral, it can occur that gold can be associated with paramagnetic minerals, or with
as free native gold or be associated with different minerals in the
magnetic minerals, or has iron impurities within the gold matrix,
ore, depending on the nature of the ore formation. The difficulty
or is coated with a magnetisable iron coating. Gold in any of
in being precise about gold mineralogy is that in most flotation
these forms has been shown to respond to magnetic methods. In
ore the gold is in such low concentrations that seeing the gold is
the flotation process where paramagnetic minerals such as
difficult, therefore understanding much about its occurrence and
associated mineralogy is also difficult (Henley, 1975). chalcopyrite are being recovered then increasing the recovery of
fine chalcopyrite would be expected to increase the recovery of
Table 1 shows that gold is not paramagnetic but it is
the fine gold associated with the chalcopyrite. As well, in the
diamagnetic. This means that with magnetic fields of industrial
flotation slurry there is fine metallic iron and iron hydroxides,
strength gold is not magnetised. However, gold in its distribution
within the ore may be associated with paramagnetic minerals generated from the grinding process and the normal alkaline
such as chalcopyrite or marmatite. The literature contains a flotation conditions, so that it is possible that the free gold or
number of studies where magnetic separation methods have been even gold on the surface of a silicate particle may absorb iron or
used to concentrate and recover gold (Corrans and Levin, 1979; iron hydroxides and so be able to be magnetised and aggregated.
Corrans, 1980; Corrans and Dunne, 1985; Ciesla, 2003). Gold The gold in the ore at Northparkes mine is mostly found
that can be recovered by magnetic separation is gold that occurs disseminated as fine particles, most particles are less then 5 µm.
in minerals with a strong magnetic susceptibility, because the The particles are generally associated with the copper bearing
concentration of paramagnetic minerals is only moderately sulfide minerals; however, there is a significant percentage found
successful by magnetic separation. along the grain boundaries of the silicate host rock.
fine mineral could be determined separately from the coarse inherently inferior to an automatic sample, it was decided to use
mineral. All samples were sized at 20 µm and 38 µm. As the the automatic final concentrate sample. Automatic samplers took
results show this significantly reduced the variance and allowed normal shift composite samples and these samples were assayed
the difference to be ‘seen’ much more readily. Particularly in the as well as subsamples being split for size analysis. Samples were
case of determining gold recovery this was effective. Coarse gold sized at 20 µm and 38 µm with the +38 µm sizes being combined.
in a sample is notoriously difficult to accurately measure and so Plant operation was monitored so that on days where a mill
removing this coarse gold noise from the testing was vital. This was shut down or some other major change occurred samples
sizing step was a significant change in methodology to normal were discarded. Note that the block design makes it highly
trials and may have more widespread implications. unlikely for the results to be biased by the flash float concentrate
being included in the concentrate sample. Changes in flash float
Northparkes mine and operation performance will only increase the noise in the data.
The Northparkes copper-gold mine is located in the central west
of New South Wales and has been operating since 1993. The RESULTS AND DISCUSSION
operation is a joint venture between Rio Tinto and the Sumitomo The results from the randomised block testing are summarised in
Group. Table 2 for copper and Table 3 for gold.
A diagram of the Northparkes Module 2 flow sheet is given in There is no statistical difference between the feed grades,
Figure 1. The processing plant comprises two parallel modules, concentrate grade or copper and gold distribution in the different
Module 1 with a design capacity of 245 t/h and Module 2 with a size fractions.
design capacity of 423 t/h. Ore is crushed and then ground in a The analysis of variance of the recovery for the different
SAG mill and two stages of ball milling to a flotation feed size fractions for copper and gold are given in Table 4 for copper and
(P80) of 95 - 110 µm. Flash flotation is used in the grinding Table 5 for gold.
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- Copper results
carbamate as chemical additives. The rougher concentrate is then The copper recovery increase in the <20 µm fraction and the 20 -
cleaned by Jameson cells to produce the final copper gold 38 µm fraction has been shown by ANOVA analysis to be, in
concentrate. Napier-Munn’s (1995) terms, real. The best estimate of the
Concentrate is thickened and filtered and sold by contract and increase in copper recovery is the difference in the means, which
on the spot market for smelting. Tailings after thickening report is 2.08 per cent for the <20 µm fraction and 0.98 per cent for the
to the tailings dam. 20 - 38 µm fraction. The results, increased recovery in the finer
Copper mineralogy is a mixture of chalcopyrite and bornite. fractions and no increase in recovery in the coarser fractions, are
There is little associated pyrite. consistent with a mechanism of magnetic aggregation of fine
paramagnetic minerals improving their flotation. The results are
Testing methodology consistent with those of Engelhardt, Ellis and Lumsden (2005).
Although the grind size is not particularly fine at Northparkes
Testing was carried out on Module 2. The flotation module has it is interesting to note that about 70 per cent of the copper is less
two large existing conditioning tanks that were used for NaSH than 38 µm. It is also worth noting that sizing the sample allowed
conditioning when high oxide ore was treated. The ProFlote® for a significant reduction in the noise with the standard
magnetic conditioning devices were installed in the second of deviation of the paired differences being much lower in the
these conditioners where the slurry residence time is 2.5 minutes. <38 µm fractions compared to the >38 µm fractions. This meant
The ProFlote® conditioning devices consist of high strength high that the statistical significance in the <38 µm fractions was much
gradient rare earth permanent magnets arrayed within a stainless higher than for the total sample, shortening the trial and giving
steel assembly. The strong magnetic field of about 5000 gauss greater confidence in the results.
contacts the slurry. At frequent and regular intervals the magnets
are automatically removed from the slurry to allow for the
Gold results
removal of ferromagnetic material that may blind the magnets.
A randomised block design program was used. The first day The gold recovery increase in <20 µm fraction has been shown
after turning the magnetic conditioners ‘on’ or ‘off’ was an by ANOVA analysis to be ‘real’ with the best estimate of the
equilibration day and the second day was a test day. The nature increase being the difference in the means, which is 3.51 per cent.
of randomised block design testing is that paired samples are Again, the increase in the gold recovery was for the finest
taken. Randomising and pairing is established at the beginning of fraction with no real increase in the coarser fractions. This is
the testing and locked in for the trial period. If during one of the consistent with an aggregation mechanism. Of particular note is
pairs there is a major plant shutdown or disturbance for a the substantial increase in the standard deviation for the coarser
prolonged period of time, or sampling or laboratory issues occur size fractions. Despite as large or larger differences in the means
then both data points in the pair are discarded. Once the testing the results cannot be said to be real because the standard
was standardised and underway and a few teething problems deviations for the courser size fractions are so large. Given that
solved about 70 per cent of the data was considered useful. Most about 50 per cent of the gold in the feed is in the >20 µm
of the discarded data related to an automatic sampler problem, or fraction, the very large standard deviations in gold recovery for
one of the ball mills being down for maintenance. The more these fractions means that testing to demonstrate a real difference
intensive sample preparation in the laboratory did not present a in recovery over the whole sample (without sizing) would have
problem after the teething period. required a very long trial.
The plant automatic samplers sample the final concentrate, There is not sufficient data, nor sufficient gold in this very
final tail, plant feed and flash float tail (flotation feed). This low-grade gold ore to determine on which gold the ProFlote is
presented a problem in the testing because the final concentrate impacting. Whether magnetic aggregation is impacting on the
includes the flash concentrate. For the test work either the final fine gold associated with copper sulfide minerals; bornite and
concentrate could be used or the concentrate excluding the flash chalcopyrite; or with fine free gold that has a magnetite or
concentrate sampled manually. Because the flash concentrate has iron-hydroxide coating or whether it is some other paramagnetic
a very low quantity of fines, and because the manual sample is or ferromagnetic mineral it is impossible to say.
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
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
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
Concentrate Concentrate
Stockpile 2 Stockpile 1
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.
TABLE 5
ANOVA analysis of gold results.
Engelhardt, Ellis and Lumsden (2005) showed that magnetic Because magnetic aggregation targets the fine fractions these
aggregation could increase the recovery of silver, a metal that fractions were separately assayed and the improvement was
like gold often occurs in elemental form in ores and is also shown to occur in these fractions alone. Targeting the fine
diamagnetic. These gold results are then consistent with the fractions in carrying out the trial reduced the plant noise and
results presented by Engelhardt, Ellis and Lumsden (2005). allowed the improvement to be ‘seen’ more quickly.
The randomised block design method has also been shown to
Economic impact be effective in ‘seeing’ these improvements.
Perhaps one of the major innovations of this work is the use of
The economic impact of these improvements in recovery to sized samples to reduce the plant noise. While this has been
Northparkes mine are substantial. This technology is now part specifically applied to a process that targets fine mineral, the
of the Northparkes flow sheet, and testing on Module 1 will methodology may also be applicable to other processes that don’t
commence in 2007. necessarily target fine mineral. For instance, in testing a process
change that the laboratory has shown to improve the process
CONCLUSION 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.
Magnetic aggregation has been shown to improve the efficiency Particularly in the separation of gold, measuring the effect on
of separation of flotation by improving the recovery in the fine the fine fraction where the variance is much smaller might be an
fractions. Both copper and gold recovery in the fine fractions has indicator of what is happening across all sizes, especially if the
increased. These increases are real and beneficial. The results are laboratory demonstrates this is an appropriate conclusion. Or it
consistent with a magnetic aggregation mechanism. may be sufficient to the plant that the process improves the fine
gold recovery alone. Whichever, looking at changes in size Corrans, I and Levin, J, 1979. Wet high-intensity magnetic separation for
fraction recovery could be a preferred or additional method of the concentration of Witwatersrand gold-uranium ores and residues,
testing that allows the operator to see small ‘real’ differences J S Afr Inst Min Metall, 79(2):210-228.
more readily, than with a non-sizing testing method. 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
ACKNOWLEDGEMENTS Australasian Institute of Mining and Metallurgy: Melbourne).
Fuerstenau, D, Chander, S and Abouzeid, A, 1979. The recovery of fine
Barry Lumsden would like to thank Professor Jameson, Director particles by physical separation methods, in Beneficiation of Mineral
of the Centre for Multiphase Processes at the University of Fines Problems and Research Needs (eds: P Somasundaran and N
Newcastle for his assistance with Barry’s research and Rick Arbiter), pp 3-61 (The American Institute of Mining, Metallurgical
Dunn, formerly of Northparkes mine, for his assistance in the test and Petroleum Engineers, Inc: New York).
work. Barry Lumsden would like to thank AusIndustry for its Henley, B, 1975. Gold-ore mineralogy and its relation to metallurgical
financial support of his research and Northparkes mine for treatment, Minerals Sci Engng, 7(4):289-312.
allowing this work to be presented. Martinez, E, 1988a. Recovery and magnetic concentration of gold, South
African Patent No ZA 8708828.
REFERENCES Martinez, E, 1988b. Process for concentration of gold and uranium
magnetically, US Patent No US 4726895.
Bheemalingeswara, K, 1995. Possible effects of iron oxide coating in the Napier-Munn, T, 1995. Detecting performance improvements in trials
recovery of particulate gold from stream sediments, Journal of with time-varying mineral processes, Mineral Processing, 8(8):
Geochemical Exploration, 52(3):373-380. 843-858.
Ciesla, A, 2003. Use of a strong magnetic field to recover weakly Sivamohan, R, 1990. The problem of recovering very fine particles in
magnetic grains dispersed in some mineral wastes, Prace Naukowe mineral processing – A review, International Journal of Mineral
Instytutu Podstaw Elektrotechniki I Elektrotechnologi Politechniki Processing, 28:247-288.
Wroclawskiej, 38:262-292. Svoboda, J, 1987. Magnetic Methods for the Treatment of Minerals
Conway, T and Dunne, R, 1986. The influence on gold recovery of (Elsevier: Amsterdam).
de-sliming before wet high-intensity magnetic separation, in Svoboda, J, Corrans, I and Spitze, M, 1986. The effect of pH on the
Proceedings International Conference on Gold, Volume 2: Extractive recovery of uranium and gold by high-gradient magnetic separation,
Metallurgy of Gold, pp 419-428 (South African Institute of Mining International Journal of Mineral Processing, 17(1-2):83-98.
and Metallurgy: Johannesburg).
Svoboda, J, Lazer, M and Te Riele, W, 1987. Selective magnetic
Corrans, I, 1980. A development in the application of wet high-intensity separation of uranium and gold, IEEE Transactions on Magnetics,
separation, in Proceedings International Symposium Fine Part MAG-23(1):283-293.
Process (ed: P Somasundaran), Vol 2, pp 1294-1309 (The American
Institute of Mining, Metallurgical and Petroleum Engineers, Inc: Trahar, W, 1981. A rational interpretation of the role of particle size in
New York). flotation, International Journal of Mineral Processing, 8:289-327.
Corrans, I and Dunne, R, 1985. Optimisation of the recovery of gold Trahar, W and Warren, L, 1976. The floatability of very fine particles – A
and uranium from Witwatersrand residues, in Proceedings XV review, International Journal of Mineral Processing, 3:103-131.
International Mineral Processing Congress, Vol III, pp 497-509 Watson, J, Rassi, D and Bahaj, A, 1983. The recovery of gold
(Bureau de Recherches Geologiques et Minieres). and uranium from gold ore leached residues by HGMS, IEEE
Corrans, I, Gilbert, W, Liddell, K and Dunne, R, 1984. The performance Transactions on Magnetics, MAG-19(5):2136-2138.
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.
The amount of pyrite dissolving is limited by calibrating the RESULTS AND DISCUSSION
model against plant data.
Traditionally, new circuit water is added to the grind. One way of
Cyanide addition is controlled to a titratable cyanide set-point
reducing the WAD cyanide at the discharge spigot is to add new
with a control loop. Titratable cyanide is calculated from the CN-
water to the tails and increase the level of recycle. The WAD
+ one cyanide from Cu(CN)43-, but excludes HCN(aq). WAD
cyanide concentration at the discharge spigot is thus reduced
cyanide is calculated by combining the CN-, HCN(aq), and
through dilution. Obviously, this is only likely to be feasible with
dissociable copper cyanides.
a tailings thickener.
Lime is added as a slurry containing 22 per cent CaO to a pH
The thickener overflow water is recycled to the front of the
set-point, with the pH calculation taking into account the
circuit. This water obviously contains cyanide, so not only is the
buffering effect of magnesium in hyper-saline water.
concentration of cyanide discharged to the tails spigot reduced,
The control of reagents to a set-point is a key feature of the but the total amount of cyanide discharged is reduced. Recycling
model, allowing reagent consumption to be calculated under the cyanide reduces the amount of new cyanide that needs to
different model scenarios. be added and further compounds the reduction in cyanide
discharged to the spigot.
Adsorption An added advantage is that the water that would otherwise be
The reactions occurring in the adsorption circuit are largely a discharged to the tails spigot is low in magnesium. Through
continuation of those occurring in the leach circuit. Although the dilution, some of this water is now recycled resulting in a lower
model has the capability of following gold, it is not significant in lime demand.
the context of this study. The exception to this is the gold in Obviously the flow of streams, their chemistry and the impact
solution tails. The recycle of solution gold can significantly of recycles becomes very complex very quickly which is why we
reduce solution gold losses and play an important role in the have used models to develop a quantifiable understanding of the
process economics. For the purpose of this exercise the gold in processes occurring.
solution tails has been set at 0.025 mg/L. The model also serves to remind us that there may be
One reaction that is important in the adsorption circuit is the operational issues associated with the proposed changes in
adsorption of copper on carbon. At this stage the model includes circuit configuration. The introduction of cyanide to the grind
copper adsorption rates and equilibrium loading profiles and so and perhaps more importantly to the classification circuit, where
copper in the adsorption circuit is well modelled. What is not HCN(g) volatilisation may become a safety issue, are important
modelled is the behaviour of copper in the elution circuit and its considerations to take into account.
recycle. Quantifying the consequences of the dilution strategy
demonstrates that the benefits can be significant. Under the
TSF conditions modelled, the combined effect of dilution plus recycle
The physical and chemical processes occurring in a tailings dam reduces the WAD cyanide at the discharge spigot from 124 to
are difficult to understand with certainty due to the myriad of 61 mg/L and the amount of cyanide discharged from 45 to
environmental and process interactions. Diurnal and seasonal 18 kg/h (see Figure 2).
fluctuations have a large impact, changing beach angles, Not only are there environmental benefits; it is also calculated
changing water hold-up and understanding the settling processes that adding new water to the tails thickener may reduce operating
are all important in a TSF model. For the purpose of this work, costs from $1.63 to $1.23/t of ore treated (see Figure 3). Solution
plant data indicating a water recycle of 31 per cent was used. gold lost to tails has been counted as an operating ‘cost’ such that
This leaves a settled mud density of 68 per cent, a value within the benefits of recycling solution gold to the head of the leach
the range indicated by Newson and Fahey’s (2001) study of core circuit and providing a second opportunity to capture this gold
samples taken from gold tailings dams. become clearly apparent.
An excess of carbon dioxide at atmospheric concentration was The model also enables the sensitivity of the circuit to
added to the TSF, lowering the pH of the tails decant water to operational changes to be assessed. For example, increasing the
7.2. This matched closely the pH of plant decant water. density of the tails thickener underflow from 65 to 70 per cent
HCN(g) loss was proportional to the HCN(aq) concentration may not be beneficial. Although the intuitive conclusion is that
after the pH had fallen due to CO2 adsorption and calibrated the increased density should lead to a further reduction in WAD
against the decant cyanide concentration. cyanide at the discharge spigot, the model suggests that water
The Mg(OH)2 precipitated in the circuit was assumed to settle balance requirements limit the amount of new bore water that
to the floor of the tailings dam and not interact with the can be added in this scenario and thus the expected benefits (see
atmospheric CO2 saturated decant water. This resulted in the Figure 4).
decant water having a similar magnesium concentration to the The model can also be modified to consider more complex
adsorption circuit tailings, a conclusion that matched measured circuits. A major complexity that has been considered is the
plant data. introduction of a flotation circuit. Flotation complicates the
picture substantially in that recycle of cyanide to a float circuit is
Detox generally not acceptable. To overcome this, a Detox circuit may
be considered and/or separate water circuits may be an option.
In some of the models a Detox circuit was included. Detox was In the ‘two separate water circuits’ scenario the tails thickener
with peroxide on a clear solution achieving 30 per cent peroxide overflow is used to dilute the feed thickener underflow (see
utilisation. Figure 5). This scenario is limited by the amount of new water
Due to the complexity of the model, not all the details are that can be added to the tails thickener. In fact in this scenario,
included here but in all cases the model was calibrated against insufficient water can be added to the tails thickener to
plant data to ensure that the base case conditions matched those compensate for water losses in the TSF and consequently
prevailing in an operating plant. The model does, however, additional water is required to top up the process water tank. The
remain weak when extrapolating to a significantly different set of environmental benefits and cost savings are substantial, but
operating conditions and it is only with further experience that reduced when compared to those available in a circuit without
the robustness of the model will improve in this respect. flotation (see Figure 6).
140 50
WAD cyanide concentration
WAD cyanide discharged 45
120
40
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.
1.40
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.
1.40
100 35
1.20
Cost ($/t of ore treated)
30
80 1.00
25
60 0.80
20
0.60
40 15
0.40
10
20
5 0.20
0 0 0.00
Base case Water to tails thickener 70 % underflow density Base case Water to tails thickener 70 % underflow density
FIG 4 - Environmental and economic consequences of increasing the tails thickener underflow density from 65 to 70 per cent solids.
In the ‘overlapping water circuits’ scenario the tails thickener loss of some of the solution gold to the leach feed thickener
overflow is directed to the leach feed thickener feed (see overflow, from which it is assumed to be lost (see Figure 8).
Figure 7). The overflow from the leach feed thickener is then Another potential disadvantage to this circuit is that gold
bled back to the front of the circuit. This circuit demonstrates leaching may commence in the leach feed thickener and this
benefits in reduced WAD to the discharge spigot, versus the solution gold will be ‘discarded’ with the leach feed thickener
base case, reduced cyanide consumption, but suffers from the overflow.
Detox costs
140 50 1.80
WAD cyanide concentration Solution gold lost
WAD cyanide discharged 45 1.60
Cyanide cost
120 Lime cost
40
WAD cyanide discharged (kg/hr)
WAD cyanide concentration (mg/L)
1.40
100 35
1.20
Cost ($/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.
Detox costs
140 50 1.80
WAD cyanide concentration Solution gold lost
WAD cyanide discharged Cyanide cost
45 1.60
120 Lime cost
40
1.40
100 35
1.20
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.
ABSTRACT carrier minerals such as sulfides, KCs are less effective, as the
amount of material that can be trapped inside the grooves of the
Knelson concentrators are commonly used to recover fine liberated gold
in gold processing plants. The decision to employ a gravity gold circuit bowl is small. In such cases where higher mass recovery is
consisting of centrifugal devices such as Knelsons is primarily taken on required, other gravity units such as in-line pressure jigs (IPJ) are
the results of GRG tests, among other considerations. typically preferred.
While the GRG test is considered a material characterisation test, Laplante, Woodcock and Huang (2000) devised a laboratory
which is useful in the design stages, its use in evaluating the performance procedure to determine the amount of gold in an ore that can be
of an existing unit is questionable. The problem is exacerbated by the fact recovered by centrifugal gravity methods and termed it gravity
that it is time consuming and costly. recoverable gold (GRG). It is known as the GRG test and has
This paper examines the validity of the GRG test for performance found wide acceptance in the mineral industry as a measure of
evaluation and discusses alternative test work that can be performed to the amenability of an ore to gravity gold recovery. This paper
evaluate the performance characteristics of an operating machine.
Principles that are useful in selecting optimal operating parameters, such
discusses the merits and drawbacks of the test and discusses an
as optimal fluidising water flow rate for a given ore/machine combination, alternative procedure to measure and improve the performance of
are also discussed and illustrated with laboratory and plant data. KCs used in gravity gold circuits.
Generally, the liberation size is indicated by the stage with the content of the softer ores are therefore more susceptible to
highest gold recovery. A higher gold recovery in stage one changes in the fluidising water flow rates used by individual
indicates a coarser liberation size and that at stage three indicates operators.
a finer liberation size. Figure 1 shows typical cumulative Another ambiguity arises in estimating the non-GRG fraction
recovery curves resulting from GRG tests on different ore types. of the coarse particles recovered in a LKC. It is subjective and
The position and curvature of the plots indicate not only the requires a highly skilled operator for the assessment.
liberation size of the gold but also the ease with which it can be While the GRG test results may indicate the amenability of an
recovered from the associated gangue particles. The shape of the ore to centrifugal gravity concentration, it would not be useful in
curve also offers insights into the possible recovery methods. A tuning an operating KC for maximum gold recovery due to the
GRG content less than 30 per cent indicates that gravity differences in the operating conditions of the LKC and the plant
concentration may not be an effective option; however, for KC. In the light of variability in ore hardness characteristics
Cu-Au ores flash flotation may be a suitable option. If the GRG and/or grind size changes in the comminution circuit, the
content is above about 60 per cent and the shape of the curve is characteristics of the feed particles that enter the KC will vary,
more convex, which indicates a coarser liberation size, then a resulting in varying gold recoveries. The extent to which these
high yield device such as an in-line pressure jig may be an changes occur in a KC cannot, however, be extracted by standard
option. For intermediate GRG contents gravity concentrating GRG tests.
units such as Knelson or Falcon concentrators may be more
effective.
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,
FIG 1 - GRG versus particle size for different ore types
the release analysis procedure for flotation. The separation
(Laplante, Woodcock and Huang, 2000).
achieved in these tests is generally presented as separability
curves and they indirectly represent the degree of liberation of
The GRG data has been used to evaluate gravity gold circuits the feed material. In this context, the GRG test is such a
and determine the benefits of employing gravity concentration procedure and hence may be considered a material
devices such as Knelson or Falcon concentrators, as well as characterisation test.
evaluating alternative circuit configurations. This involves the On the other hand, the machine characteristics are obtained by
evaluation of recovery effort, which essentially is the product of subjecting fully liberated material or artificially reconstituted
the process efficiency (viz GRG recovery) and the fraction of the feed mixtures to the machine. The separation data is often
circulating load treated by gravity (Laplante and Xiao, 2001). presented as performance curves, which are commonly known as
partition curves or Tromp curves. In assessing the performance
Drawbacks of the GRG test of an existing machine it is the machine characteristics that need
The main drawback of the concept of GRG stems from its to be established first.
definition. Gravity concentration devices achieve mineral Coulter and Subasinghe (2004) proposed a method of
separation by applying forces on particles. Thus, particles quantifying separation data of a Knelson concentrator by
recovered in one machine may not necessarily be recovered in analysing the forces acting on a particle within a KC. They
another due to the differences in the extent of the forces applied, proposed a performance criterion (X), defined as the ratio of the
the way they are applied and particularly if the feed size is drag force (Fd) on a particle due to fluidising water to the
changed. According to the GRG concept, the percentage of gold combined effect of centrifugal (Fc) and Bagnold (FB) forces,
recovered in a three inch LKC from a stage-ground feed material which are functions of particle size, bowl rotational speed,
is taken as the GRG content. However, the feed material fed to an particle density and solids concentration. The complete
industrial scale KC is of a different size and the separating forces derivation may be found elsewhere (Coulter and Subasinghe,
exerted by the machine on the particles are different to those of a 2004).
LKC. Hence the GRG content must be different in different Figure 2 shows a schematic view of how heavy mineral
machines. The proponents of the GRG test believe that about two
concentration takes place within a KC. The feed material, most
thirds of the GRG recovered in a LKC would be recovered in an
industrial scale unit. However, there are industrial KC units of commonly cyclone underflow or ball mill discharge, enters
various sizes that inherently recover different amounts of gold. through a central feed tube as a slurry. The slurry descends onto
Thus, the concept of GRG must at least in part be machine the base plate at the bottom of the rotating conical bowl and is
dependent. thrown outwards as initially there is no centripetal force acting
The second drawback of the GRG test is that there is some on the particles to maintain a circular path. As a result, some of
amount of subjectivity associated with the test itself. The the particles get trapped within the grooves of the bowl forming a
fluidising water flow rates in the three stages are not strictly concentrate bed while some get carried upward into the tailings
stipulated and can be varied according to the size of the feed stream by the rising current of water. Injection of water through
material (Guerney, Laplante and O’Leary, 2003). This would fluidisation ports located in the ribs of the bowl prevents
mean that the GRG content would vary, particularly if the ore compaction of the concentrate bed. This creates a fluidised bed
hardness and the breakage characteristics of the ore are such that (Harris, 1984), which acts as a concentrating chamber for heavier
it produces more fines during the staged grinding step. The GRG gold particles under enhanced gravitational forces.
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
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.
40
30 Probability of retention
Run No Magnetite Silica 1 Silica 2 Silica 3
20 1 0.05 0.013 0.142
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
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).
(63 mm PQ and above) compared to exploration holes (often diameter. For shallower and softer ores (eg nickel laterites) pilot
48 mm HQ, or below). The aim of drilling metallurgical samples are often collected using a 1 m diameter Caldwell drill.
holes is to generate significant sample mass to generate ore The largest tonnage pilot samples are usually extracted by
type composites and allow an extensive test program to be opening up the orebody by trench, trial shaft, winze or adit
carried out on a single sample. mining. The choices of locations for accessing the orebody are
• If pilot plant trials are to be conducted, source a large sample limited by geology, topography, timing and cost. Typically, bulk
of ore (20 to 100 t). pilot samples are representative of the conveniently available ore,
rather than representative of the orebody or an ore type. Of the
• For a ‘comprehensive’ pilot test program, source tonnage size available methods for extracting pilot plant feed samples, the
samples of each major ore type. most representative is large diameter drilling at a relatively large
• Conduct characterisation testing to define the properties of number of locations.
the major geological ore types. If samples of individual ore types are extracted for pilot
• Possibly conduct tests on ore blends representative of mining testing, the same representation problems described above apply.
time periods. However, the problems are magnified because each ore type
makes up only a proportion of the orebody and some ore types
• Conduct a variability test program making sure that all ore may be more accessible than others.
types are represented a number of times.
Characterisation of ore types by test work is typically
The data from this test program is then analysed to estimate restricted to what are seen as the essentials. If large diameter
critical design factors including SAG and ball mill specific cores are extracted specifically for drop weight testing (DWT),
energies, crusher design parameters, flotation recoveries and for example, a decision is often made, especially on smaller
leach recoveries. These values are used to set process design projects, that there is no need to conduct variability testing in
criteria and then to size process equipment. respect to SAG mill competence. In the case of SAG milling, this
There are a number of significant issues inherent in this means that the mill design will have limited (or no) variability
approach, all of which can detrimentally impact on the project. information. As the SAG mill is one of the highest capital cost
There is a high probability that the ore types determined on the items in the circuit, and a likely throughput bottleneck, this is a
basis of geological classifications have little or no metallurgical short-sighted approach.
relevance. This leads to a lack of metallurgical definition The preparation of time-period blends is often done with the
between ore types because the ranges of the metallurgical available samples or is done mathematically using the
properties for each lithology overlap and they are highly variable ‘representative’ ore-type data. Again this approach is
within an ore type. Acceptance of the domains and associated questionable because it relies upon the assumption that the
average test results as definitive regularly creates confusion and available or tested samples of the ore types are representative.
myths about ore type properties that mislead designers and, Unfortunately, even if the available ore type samples are
ultimately, operators.
‘typical’, the chances that the ore type is consistent throughout
The sourcing of specific ‘metallurgical’ samples, by drilling a the orebody are low. Often there are significant trends in
few large diameter holes, limits the spatial distribution of what is breakage properties, even within well defined ore types, that are
considered the most definitive set of test results available for related to factors independent of the geological divisions.
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’ DATA COLLECTION DRIVERS
examples of those ore types. All of the above situations, resulting from typical design
In focusing on ‘metallurgical’ samples, the much larger practice, are regularly encountered by the author. Unfortunately
resource of well-distributed samples in the core library is often such situations are often discovered after commissioning and
ignored. The ability to utilise these samples in a meaningful during an investigation into the reasons for a circuit failing to
manner is strongly influenced by the choice of characterisation meet performance expectations. Because fixing the problems
test methods and the available processing options for the type of during operations is common enough to be considered normal,
ore (ie flotation for base metal sulfides or leaching for a the fundamental question that needs to be addressed is: ‘What
free-milling gold ore). are the reasons that owners and process engineers are satisfied
Core samples extracted for metallurgical purposes often with such risky practices being applied during the design phase
bypass the normal geological characterisation steps because the of a project?’
overriding aim is to use 100 per cent of the core for test work.
The extraction of metallurgical test samples and the generation
There may also be other factors, such as immediate sample
of results both suffer from a number of inconvenient timing
refrigeration to minimise possible oxidation, that prevent
meaningful characterisation. It is then largely a matter of issues. The problems are mostly related to the reasons the data is
assumption that the samples do actually represent typical collected in the first place.
examples of the major ore types. By the time this assumption is One of the biggest barriers to conducting an appropriate
checked the samples have usually been transported large metallurgical test program is the fact that, in the eyes of the
distances and been through a number of processing steps. If the owners, the value in the orebody has already been established to
assumption proves incorrect, then the ability to extract additional the satisfaction of the market and the regulators by complying
sample is likely to be limited by drill rig availability and drill with the appropriate reporting code. Compliance with these
target priority. Even more problematic, if ore type domain codes (such as JORC) requires that an extensive and expensive
assumptions prove incorrect, then there is a similar likelihood drilling program be conducted. In the value establishment phase
that the second set of samples will also be unrepresentative. the owner is rightly focused on the reporting of resources and ore
Tonnage-scale samples extracted for pilot testing are, for many reserves. A positive market response to the geological data places
reasons, likely to be unrepresentative. Because large masses are a present value on the resource, a value that incorporates
required for pilot testing it is necessary to either open up the estimates of capital and operating cost. However, as the purpose
orebody or conduct a program of drilling large diameter holes. of getting the market to respond is to raise funds for a feasibility
For hard rock deposits, a drilling program for pilot work would study (or establish value for a sale), the cost estimates must
typically consist of core in the range 83 mm to 150 mm in always be dominated by assumption.
By the time a process study team has a chance to look at the ‘similar’ metallurgical properties are simply grouped together
orebody there are already a large number of process assumptions giving fewer equally uninformative domains.
in place. There is also a timeline for development (often The definition of ‘similar’ in the context of domain assignment
unrealistic) and marketing may also be underway for is often misleading. The metallurgical similarities between
concentrates. There is no denying that all these factors are lithological domains often exist because each of the lithologies
essential for a successful project. However, if there is no exhibits a similar, but broad, range of metallurgical properties.
allowance for adequate process development in the timeline, then The root cause of the problem is usually that the lithological
the project risk increases significantly. In some instances, domains are not appropriate and an unrelated set of
metallurgical test work is seen as a necessary evil that simply metallurgically relevant domains can be found within each of the
delays the time to plant commissioning. major lithologies.
As there are no regulatory codes governing process To simply group the lithologies together is to ignore the fact
development reporting, the owner returns to a mode where that the lithological domains may be meaningless in a
standard market rules of disclosure apply. This results in the metallurgical sense and an inappropriate starting point. The
process development stage being treated quite differently to the ultimate result of this grouping process is to discard the
exploration phase by both the owner and the market. geological ‘baby’ with the geological ‘bathwater’, potentially
Another key driver for data collection is the recognised ignoring the majority of the vast database of geological
complexity of the orebody or the process. If the orebody has information. Having declared the lithological domains redundant
obvious variability issues or contains potential problem elements, the next unfortunate step is to create a metallurgical database,
often it is recognised in the exploration phase and the exploration independent of the geology. Typically this involves large
budget is expanded accordingly. The additional exploration numbers of specialist tests designed to map the metallurgy of the
requirements often alert the owners to seek expert process advice deposit.
early. This is not geometallurgy. It is essentially rendering the
A further driver is the complexity of the known process. For geological database to being a source of assays to complement
example, in the case of low grade nickel laterites the complexity the new and expensive metallurgical database.
of the PAL (pressure acid leach) process is well recognised and a Rather than improving the relationship between process
significant budget and time frame are usually allocated for engineers and geologists, this process has the strong potential to
process development. build a new barrier. A new set of metallurgical terms is
Conversely, if the process is recognised as simple (gold CIL, introduced that must then be imposed on the geologists and
copper flotation, etc) then a relatively small process development miners that, unfortunately, have no geological relevance and little
budget is set and the development time frame is compressed. It is mining relevance.
these projects that often face the greatest risk if the process
development allowance (both budget and time) is inadequate. Geometallurgical domains
Additional test work that causes a schedule delay is always
To assign geological domains requires a geologist. Assigning
bad news. The later in a project, the worse it is received. The
metallurgical domains requires a metallurgist. To assign geo-
process engineer must either live with the imposed schedule or
metallurgical domains, by extension, requires a geometallurgist.
attempt to extend the schedule by a number of months so that
adequate test work, or even pilot testing, can occur. The latter A geometallurgist, in the majority of cases, should start out
approach is rarely successful and the outcome depends mainly being a metallurgist or process engineer. The aim is to assign
upon the magnitude of the identified risk in the context of the domains that are metallurgically relevant while utilising the
project. The outcome is also strongly dependent on how geology. This is best done by someone who first understands the
successfully the process engineer has communicated the issues to process intimately and, secondly, has the capacity and experience
the owners. to understand the geology, especially the process implications of
the geological data.
There are many factors that have a high probability of forcing
the job of the process engineers to fit into an inadequate time The first step in assigning geometallurgical domains is having
frame and budget. Opposing this, the process engineer must an understanding of what is important to the process. Each process
consider contractual and reputation risk exposure for the has its own characteristics. A few of these are listed in Table 1.
engineering company should the circuit fail to meet expectations.
TABLE 1
DOMAINS Important process issues to understand on an orebody-basis
for a number of circuit types.
The fundamental building block of geometallurgical analysis is
the domain. For the purposes of this paper, a domain is an Circuit type Critical properties
exclusive classification of ore by a common property or set of SAG mill circuit • Competence
properties. • Grindability
Stage crush and mill circuit • Crusher work index
Lithological domains • Grindability
The simplest domain set is lithologically based, typically Iron ore (haematite) • Lump/fine ratio
assigned by the geologists early in the exploration phase. The • Impurity levels (S, P, etc)
common link is the rock type assignments. Typically, some Copper sulfide ore • Copper mineralogy
lithologies will also be described as ore and some as waste, • Alteration
providing a rudimentary metallurgical basis to the domain • Impurity levels (F, As, etc)
allocations. Gold ore • Gravity recoverable gold
In many projects this is where the classifications start and end. • Cyanide recoverable gold
The description of the domains by lithology lasts for the life of • Refractory gold
the project regardless of the appropriateness for the process • Copper grade
engineer. • Preg-robbing properties
In some instances lithological domaining is eventually Nickel laterite • Beneficiation characteristics
recognised as being inadequate, but the solution offered is worse • Acid consumption
than the initial problem. The domains that are found to have • Slurry viscosity
Few of these properties are measured directly in the geological From this point on, the analysis can take a number of
analysis of the core intervals. However, a large number of these pathways. If the lithological domains do have metallurgical
properties have the potential to be inferred by using one or more relevance then lithology may be sufficient for geometallurgical
of the pieces of information within the geological database. domain definition. Unless the relationships between metallurgical
parameters and lithology are highly correlated it is prudent to
The second step is to select a number of samples from
continue to search for additional correlations, such as with grade,
throughout the orebody that represent a reasonable spatial
alteration, etc. If there are a number of other geological data sets
distribution and cover the lithological domains. Priority must
that appear to influence the metallurgy then the relevance of
also be given to testing ore that is to be treated early in the life of these can be examined using techniques such as multiple linear
the operation. To select the test samples it is essential to gain an regression and non-linear data fitting.
understanding of the most likely mine plan together with
alternative plans that may gain ascendancy during the study If the lithology has no relationship to the properties, and the
metallurgical properties are variable beyond the limits of
phase of the project.
experimental error, then the search for relationships must be
The number of samples to be selected will be determined by wide ranging. Rather than adopt a trial and error approach there
the complexity and size of the orebody. The number should be are some logical starting points:
high enough to allow a wide spread of likely ores (and some
probable dilution waste) to be tested. It is much better to test • Ignore geological data sets that show little or no variability as
localised samples than to generate composites and test these. The they are unlikely to relate to a variable metallurgical property.
localised samples can be directly related to the geology and will • Unless there is a logical reason or a probable correlation it is
have a higher natural variability than composites. A minimum of usually safe to ignore trace elements.
20 samples should be metallurgically tested as this should
• Analyse the metallurgical data set for clusters and determine
provide a basis for statistical analysis and pattern analysis.
if any geological data may be a predictor for those clusters. If
The selected samples must then be either extracted from the distinct clusters are found for the metallurgical properties
orebody or sourced from within the available core library. Given then it is likely that a basis for geometallurgical domaining
that the project schedule is usually very high priority, the focus has been found.
should be on sourcing samples already extracted from the
orebody as core. In addition, it is essential to ensure that the Important geological factors to incorporate in the investigation
samples have been subjected to the full suite of geological include:
analyses. Note that RC drill samples are generally unsuitable for • major valuable elements;
this test program unless the critical properties are independent of • related elements such as sulfur;
comminution characteristics and are unaffected by a potential
loss of fines during extraction. • major non-valuable elements (typically Si, Ca, Mg, Al and
The samples are then metallurgically tested to determine the Fe);
critical parameters. Critical properties are typically those that • ratios between elements (eg Cu:S, Mg:Si, Zn/Fe, Fe:MgO);
determine throughput, final product rates and saleability of • alteration reporting – note that alteration reporting may be
product. Increasingly, some environmental properties can also be difficult to interpret, especially if there are related alteration
considered critical in some situations. Critical properties relate types that should be combined for metallurgical purposes;
directly to key process design criteria. The most appropriate
metallurgical tests are standard procedures that provide • core fracture data such as RQD (the percentage of core in a
fundamental characteristics rather than flow sheet specific given length that is in pieces greater than 100 mm in length,
procedures that may be subject to unintended variability. For a potential pointer to competence), fracture frequency and
example, it is better to conduct rougher only flotation tests that point load test results; and
determine floatable and non-floatable ore components than to do • other potentially useful data such as colour, Mohs hardness,
a complex rougher/cleaner procedure that measures the response depth, northing, easting, etc.
of the ore to the test conditions regardless of characteristics such
as feed grade or mineralogy. From an understanding of the physics of the process,
determine all the factors likely to contribute to the metallurgical
After the metallurgical testing is completed and the results are property in question and begin with these, both in isolation and
analysed the process engineer will have available for analysis a using multiple regression techniques.
set of geological data and an associated set metallurgical data. Look for unusual non-linear data dependencies that may mask
The most important investigative work from a geometallurgical an important relationship. For example, a relationship may be
viewpoint follows this milestone. valid below a certain point but becomes scattered or constant
above that point. Recovery by flotation is one property that is
Geometallurgical investigation typically low at low feed grades, increases to a maximum at a
higher feed grade and then stays at the maximum at still higher
The geometallurgical investigation phase is an examination of feed grades. Assuming it exists, a graphical analysis of flotation
data relationships. The first step is to look for correlations test recovery versus core grade would reveal this relationship and
between the geological data sets and the related critical the correct response may be to apply a constant recovery to cores
metallurgical test results. This can be done either visually, using above a threshold grade. Below that threshold a mathematical
graphs, or using statistical tools such as analysis of variance relationship between core assay and recovery can be applied.
(ANOVA) tables. Usually it is better to look at the data Look for relationships within geological groups, other than the
graphically because this allows the early detection of non-linear lithologies, such as alteration types. These relationships may be
relationships, it allows outliers to be identified and investigated distinctly different between geological groups and may need to
and it allows for the detection of incorrect or assumed data. be described by a number of independent equations which, for
To determine if the existing geological domains are appropriate example, are invoked depending on the alteration type.
the metallurgical results are grouped into their lithologies and Hopefully at the end of the investigation one or more
examined for clusters and patterns. If the critical properties predictive relationships with correlation coefficients (R2) greater
overlap in a number of the geological domains then it is highly than about 0.75 have been found. Alternatively, clustered
probable that the geological domains are metallurgically metallurgical properties have been related to geological
meaningless. predictors, which may or may not be numerical.
The correlation coefficient is a useful value for screening Excel, to extract, group, count and average data within potential
possible predictive relationships. It is most appropriate when geometallurgical domain definitions.
comparing alternatives for the prediction of a common If a chosen system of geometallurgical domains is to be
parameter. It is less useful when comparing the prediction of one metallurgically useful then it is necessary that each of the
parameter with another. domains exhibit distinct clusters of metallurgical properties. If
Judgement of the prediction outcome is best done by the clusters are diffuse and overlap then the chosen domains are
evaluating the standard error, or the standard deviation (SD) and not likely to be any better metallurgically than the original
then calculating the relative error or relative SD. Relative error or lithological domains.
SD both indicate the inaccuracy that can be expected when using If distinct and valid data clusters are not found then the
the predictive method. predictive technique is not verified and the analysis must either
At this point the data relationships are mostly theoretical and be repeated or abandoned.
they must be verified. It is essential that the relationships are On the assumption that a reasonable prediction method has
checked to ensure that they are valid beyond the data set from been established, and that a reasonable basis exists for
where they were developed. It is also essential to check the effect geometallurgical domaining, it is then possible to utilise this in
of extrapolating the relationship beyond the limits of the base the design and mine planning process. The advantage of
data ranges. following this methodology is that the existing geological
database becomes a metallurgical resource and there is also
Verifying geometallurgical relationships potential to incorporate some of the metallurgical findings back
into both the geological modelling and the mine planning
Potential geometallurgical relationships are relatively easy to processes. In addition, the process engineer has gained an
verify, mainly because of the large size of the geological excellent picture of the orebody geology and the geologist has
database. probably learned a large amount about what is important for the
However, the first verification step is to check the standard future process plant.
error for your prediction within the metallurgical database. If the
standard error is large compared to the accuracy you require (for APPLYING GEOMETALLURGY
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 The engineering application of the geometallurgical findings can
within 1 kWh/t) then a prediction basis has not been established. be the difference between success and failure or the difference
If the standard prediction error within the data set is acceptable between a good result and an excellent result.
then the predictive method should be applied across the whole, or
part of, the geological database. Study phases
As the geological database is not a single database but a The study phases of a project are the most crucial as they
collection of related databases, it is first necessary to arrange the determine the form of the final design. Investigations during the
data into an analysable form. The problem encountered for study phases are, in the context of the project, inexpensive. It is
spreadsheet users is that values such as assays are typically of interest to note how it is relatively easy to justify funds to fix
measured every metre, SG may be once every 20 m, RQD is an operational problem when it could have been solved for a
likely to be by the core tray and lithology is likely to be by fraction of the price during the study phases.
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, Options study
usually intervals of 1 or 2 m. Geological analysis software is An options study compares, at a high level, the various options
generally able to generate normalised databases in text or Excel available for extraction and processing of the orebody and the
format directly from the geological data sets. options for marketing the products. At the options study level it
Once all the geological data is in one place it is important to is not necessarily useful to have too much information because
clean up the database. Unfortunately, as it is not possible to this can slow the evaluation process and mask the high level
predict all the problems and errors that may exist in such a issues. However, in the options phase preparation must be made
database, the clean-up process is something that is difficult to for the next study phases, the prefeasibility and the definitive
automate. The database must be examined for blank records, feasibility studies. However, money for an extensive metallurgical
missing critical data, duplicated data, inconsistent data (ie a mix test program may not be available at this point. Funding is
of ppm and per cent values in the same column) and many other generally dependent upon the outcome of each study phase.
varieties of error. During the options study phase it is important to be alert to
Use a feature such as conditional formatting, data sorting and early indications of disconnects between the geology and the
data filtering to quickly identify data points that lie outside metallurgy. For example, if the lithologies seem to have similar
the possible measurement range (eg assays >100 per cent, metallurgical properties or if ores defined as the same ore type
SG values >5). Once a problem is identified, action can be taken have vastly different properties then geometallurgical analysis is
probably essential.
to either delete or repair the record. Repair within the
geometallurgical database is only possible in obvious cases, such A key outcome from an options study phase should be a set of
as where ppm and per cent are mixed as mentioned above. Any preliminary investigations designed to test geometallurgically
instances where repairs have been made should be documented relevant possibilities. This is likely to include extracting more
and the owner of the geological database informed. Ideally, the samples from each of the lithologies and extracting some key
owner of the geological database will correct the errors at source samples from significant zones that appear to be different to
and re-issue the database. those already tested.
The outcome from the options study phase should be a short
Once a clean database has been developed it is then possible to
list (preferably only two alternatives) of economically attractive
conduct predictive verification and domain allocation. options.
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
Prefeasibility study (PFS)
predicted properties are in place it is best to use relatively The prefeasibility study is intended to arrive at a preferred option
sophisticated techniques, such as pivot tabling in Microsoft out of the shortlist and to begin flow sheet definition. The level
of detail required is related directly to the accuracy of the A definitive study must establish the real value of the resource.
estimate required. Typically the prefeasibility estimate accuracy This requires a well developed mine plan with an associated ore
is in the region of ±25 - 30 per cent. delivery schedule. To establish a cash flow prediction for the life
A geometallurgical view of the orebody provides the engineer of the operation based on the mine plan it is essential that the ore
with additional process selection drivers not necessarily delivery schedule informs the process engineers of variability in
considered in a more superficial design procedure. For example, the critical process parameters. Realistic cash flow projections
when choosing between single and two-stage milling options a are only possible with an understanding of the period by period
detailed understanding of the ore types, their variability and the variations in head grade, throughput rate, operating cost per
stability needs of the downstream process are essential to ensure tonne and product quality.
the decision is technically correct. In situations where the grade, It is also necessary to estimate the ‘noise’ that should be
mineralogy and hardness variability of the ore is significant but expected for each of the critical process parameters on an
the downstream process is robust over a wide range of operating operational time scale. In the DFS stage, the shortest predictive
conditions then a single stage mill is the most likely time frame within an ore delivery schedule is usually one month.
recommendation. However, when the process requires stability Late in the production schedule the time scale is usually one
and accurate control over particle size then a two-stage mill is year. Worse for judging variability, a mine plan is usually
the likely recommendation, together with an ore blending constrained, in part, by the need to smooth head grade and
strategy. product output. It is not appropriate to assume that the flexibility
Because the required geometallurgical testing burden is small required of the plant design can be determined from the month
compared to some other methodologies, the majority of the by month variability of the mine plan. Operational flexibility
required test program can be performed during prefeasibility. If must be based on the expected shift by shift variability.
correctly planned during the options study phase, and if focused The issue is then to find a data source that does have the
on available samples rather than new core, the total test program possibility of representing the variability of the critical process
can be completed in two to three months. In many cases, the parameters. The most obvious source is the library of core data
resulting data set will be adequate not only for the prefeasibility and a second, related, source is the block model.
study, but also for the definitive feasibility study. If the data set is The core library represents metre by metre variability within
subsequently found not to be definitive, the results will provide the orebody. This is too detailed to use directly as a
an excellent basis for planning the definitive test work program. representation of shift by shift variability. The core library data is
It is recommended that by the end of the prefeasibility study the best examined as a set of location-based groups, an example of
chosen domains for the orebody are geometallurgically relevant. It which is already utilised in the mining world as the block model.
is also recommended that the plan for full confirmation and If available, the block model data can be used but a few checks
characterisation of the geometallurgical domain selections are are required first. The block size should represent between ten
included as recommendations in the prefeasibility report. 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
Definitive feasibility study (DFS) grouped data that can be used to calculate the critical parameters
The definitive study phase is often also called the Bankable for the block. The blocks should also be relevant in a
Feasibility Study. The accuracy of such a study is, at a geometallurgical sense. If these conditions are met then an
maximum, ±15 per cent. Considering the financial uncertainty examination of the variability in the block data can provide a
can be as high as ±10 per cent (due to interest rates, exchange meaningful basis for process design.
rate fluctuations and equipment/materials/labour prices) there is The most difficult part of this step is to ensure that the relevant
little, if any room for technical uncertainty. Ideally, at the data is calculated on a block basis. To do this an understanding is
commencement of a DFS the basic flow sheet has been finalised required of the linkage between process response and geological
and only one option is ‘left on the drawing board’. However, it is data and the geological model must be able to provide the critical
typical at the commencement of a DFS for trade-off studies to be process parameters. Without a geometallurgical analysis none of
this is possible.
conducted to determine the most appropriate equipment types for
the selected unit process. Examples of trade-offs are Jaw versus In one recent example, which will not be identified, a plant
Gyratory primary crushing and conventional versus high rate was designed for an average grade in the region of 1.5 per cent
versus paste thickening. and an upper expected feed grade of about 2.3 per cent. During
start-up the plant was required to deal with grades fluctuating
At the definitive study phase the client is acting on board between a low of about 0.5 per cent and a high of eight to nine
approval to proceed, usually with the intention of transitioning per cent. The basis of design was estimated from the mine plan
into funding approval for the project as a whole. The date for variability. However, the certainty of receiving a much wider
first production is probably set and a tight time frame is imposed range of feed grades was painfully obvious for anyone that
for the study, usually aiming for a funding decision at a particular wished to examine the core library.
board meeting. From the engineer’s perspective the client At the end of the DFS the process engineer must understand
expectation is that the test work carried out in the previous study the critical process parameters in terms of their operational-
stages, together with any additional test work already approved, timescale variability and in terms of their relationship to the
is going to be sufficient for design. Again, the engineer must mine delivery schedule. The geometallurgical domains should be
determine if this is the case. If more test work is essential and the well established in all aspects of the project thinking.
potential exists for extending the study time frame then the
engineer has some significant justification work ahead.
Design phase
The key issue in a DFS is having definitive data for design.
Declaring data to be definitive is firstly the job of the design The design phase of a plant is when uncertainty is extremely
engineer. The engineer must be satisfied that design decisions are unhelpful. If design questions and assumptions persist at this
based on fact and able to withstand scrutiny. Scrutineers include stage, then the previous stages were probably carried out
the client representatives, the vendor’s technical teams, the inappropriately. Unfortunately, many projects suffer because
Bank’s resource analysts and the Board’s consultants. If there are myths and assumptions have not been properly tested before it is
performance warranties then scrutiny within the engineering too late. These myths sometimes underpin key design decisions
company should be the toughest of all to pass. and if they are disproved late in the project development cycle
they can be damaging to both schedule and reputation. It is this is unlikely to be the same as the mine plan available for the
last point that most often results in myths persisting, even when definitive study and will change regularly on a short-term tactical
proven beyond doubt to be false. basis. Without a means of predicting the performance of the ore
The process engineer that inherits a project at the design phase that has actually been scheduled, often ready for treatment within
is in a difficult position. Not having been responsible for the the week, the plant is destined to operate in reactive mode. With
study phases, the engineer has not had the opportunity to select a meaningful method for the prediction of critical process
samples for test work or test assumptions and myths. The properties the plant can be a proactive participant in the short-,
engineer must, during bid preparation, examine the data made medium- and long-term production planning processes, rather
than a victim.
available by the client and dig deep enough to be satisfied that
the detailed design path is clear. However, if serious problems Many operations have found, usually a few years after
are detected during the evaluation phase that demand more test commissioning, that the ore they are supposed to receive never
work then the engineer is in a dilemma. Depending on the client, comes. The revelation then dawns that the ore types they have
the message may or may not want to be heard. As the engineer is relied on since the earliest days of design do not possess
primarily in the business of building plants the focus may be predictable sets of process properties and are more often
directed at winning the job first and sorting out the problems misleading than useful. It is at this point that many operations
employ someone to conduct a geometallurgical analysis.
later. The decision will be based on risk exposure for the
Addressing geometallurgical issues at an operating site with no
engineer and the contractual basis upon which the job is carried
historically effective geometallurgical practices is a much more
out. Ultimately, it is likely the client will pay the price, either if
difficult problem than building geometallurgy into the design
some additional test work and an associated schedule delay or process and is beyond the scope of this paper.
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 CONCLUSIONS
they were carried out competently, then there should be far less The implementation of geometallurgical analysis should result in
technical and schedule risk in the transition to project. It is close cooperation and a common language being developed
recommended that as a minimum, the definitive study phases and between the geologists, mining engineers, process engineers and
then move into the design and construct phase. It is strongly project managers.
suggested that the engineering firm be involved from the earliest The earlier that geometallurgical analysis is implemented in a
stages of the project so that geometallurgical thinking (together project’s life the more value it will add to the project.
with its extensive design experience) is incorporated from the During the study phases, geometallurgical analysis should
start. Unravelling the geometallurgical issues at the definitive progress in line with the accuracy of estimation required and
stage and beyond is difficult and expensive. Building with the associated level of funding that is available. At options
geometallurgical thinking into the project from its inception will study level the metallurgical appropriateness of the geological
provide a strong foundation for team-based thinking across the view of the orebody must be established. At the prefeasibility
owner’s and engineer’s teams. stage the geometallurgical domains should be defined and the
definitive test program should be outlined. At the definitive stage
Operational phase the domains should be entrenched in the project, the domain
properties defined and the variability of the properties
The operational phase is where the appropriateness of the established.
engineer’s designs, especially those parts based on assumption At plant commissioning the ore domains the plant receives
and unproven myth, are exposed for all to see. During should not be a surprise and the critical metallurgical properties
commissioning it is expected that mechanical issues will arise, should be adequately defined.
minor design problems will need rectifying and operation will
not be smooth. However, it is not appropriate that problems be During operations the owner’s team should be able to plan
production on a reliable basis so that what the plant actually
caused by the plant being fed ore that it didn’t expect to see.
receives as feed in the following weeks, months and years is
Examples include:
what they planned.
• Incompetent oxide ore being fed to a SAG mill that has been
set up at commissioning for competent fresh ore, resulting in ACKNOWLEDGEMENTS
damage to mill liners.
• Overloading of the flotation circuit, especially the cleaning The author acknowledges the permission of GRD Minproc to
stages, due to unrecognised variability in the mill feed. The publish this paper and the assistance of the GRD Minproc
result is usually a significant loss of recovery or a significant Mining and Geology section, with a special mention to Annick
reduction in plant throughput (if throughput turn down is Manfrino, for helping to make geometallurgical analysis a reality
possible with the selected equipment). within our organisation. Most importantly acknowledgement is
due to many colleagues over the years, metallurgists, miners and
• The generation of unsaleable products due to mineralogical geologists, with a desire to talk a common language so that we
problems or the presence of deleterious elements. can work together to maximise the value of our orebodies.
All these outcomes are predictable and examples of instances
like these are not uncommon. REFERENCES
As the engineer is usually responsible for commissioning it is Burger, B, Hatta, H, McGaffin, I and Gaffney, P, 2006. Batu Hijau –
in its best interest to understand, as fully as possible, what is to Seven years of operations and continuous improvements, in
be expected during the commissioning phase. As the client wants International Autogenous and Semiautogenous Grinding Technology,
to be earning cash as rapidly as possible it is also in their best 2006 (eds: J Allan et al), Vol 1, pp 120-132 (Department of Mining
interests to allow the engineer to do the work that is necessary to Engineering, University of British Columbia).
eliminate any of these surprises. Even a week of commissioning Dance, A, Valery, W, Jankovic, A and La Rosa, D, 2006. Higher
delay is likely to result in losses that dwarf the expenditure productivity through cooperative effort: A method of revealing and
correcting hidden operating inefficiencies, in International
necessary to detect these losses ahead of time. Autogenous and Semiautogenous Grinding Technology, 2006 (eds: J
Post commissioning the owner will finally have a mine plan Allan, et al), Vol 1, pp 120-132 (Department of Mining Engineering,
that is correct, at least for a couple of weeks into the future. This University of British Columbia).
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
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
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)
[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+.
pH’s below 9.2 the presence of carbonate may enhance the gold -
2Cu(CN)32-
diff mV
-200 Ag(CN)2 +
mV
-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.
Decreasing the addition rate reduced the overestimation; determination), and PbOH+ and Pb2+ become soluble, which
however, the end point becomes very difficult to interpret below interferes with the rhodanine indicator (gives the salmon pink
2 ml/min due to the very slow colour change. A ten-fold dilution end point colour). This can be avoided by maintaining the sample
was found to exacerbate the overestimation with the apparent pH >11 or by the addition of a concentrated lead nitrate solution
cyanide concentration increasing by more than 15 per cent. prepared in 1 M NaOH. Addition of solid PbO or PbCO3 could
A possible explanation is that the silver complexes faster with also be used.
cyanide from the Cu(CN)32- complex than with rhodanine and The results of measurements made using a concentrated lead
the equilibration (Equation 9) with the indicator complex nitrate solution prepared in 1 M NaOH are shown in Figure 8.
(Rh-Ag+) is slow and also influenced by the concentration. The slightly lower cyanide titration values after lead addition
could be attributed to the low solubility of lead cyanide or the
Rh − Ag + + 2Cu(CN)3 ⇔ Rh + Ag(CN)2 + 2Cu(CN)2
2− − −
(9) oxidation of some sulfide to form thiocyanate. This is supported
by the potentiometric titration value determined without lead
Comparisons between the titration values and the addition. This determination is possible as silver sulfide was
thermodynamic titratable cyanide for two different cyanide found to form preferentially to silver cyanide and thus it is
titrated before the cyanide (Figure 9). Hence, the sulfide does not
concentrations with copper present are shown in Figure 7.
need to be removed for the potentiometric titration. This has the
Clearly, the rhodanine titration significantly overestimates the
added benefit that the sulfide concentration can also be
available cyanide for gold leaching when copper is present, the determined concurrently using this method. However, online
magnitude of which appears to be related to the copper potentiometric titration set-ups and the amperometric method are
concentration. The potentiometric and amperometric techniques affected by sulfide ions; online potentiometric titrations
on the other hand closely measure the available cyanide for gold generally only detect one end point and hence would include
leaching. sulfide ions in the cyanide determination, and the build-up of
silver sulfide precipitate effects subsequent measurements for
6000
both techniques unless removed.
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
sulfide ions and the presence of lead can assist in overcoming the using the potentiometric method is not possible due to the
effect of sulfide ions. Recent laboratory test work confirmed that concurrent titration of the third cyanide associated with copper
gold dissolution does not occur to any appreciable extent when and thiosulfate. Thiosulfate does not interfere with the
sulfide ions are present in the leach (Figure 10). Once sufficient amperometric cyanide measurement when -50 mV is utilised.
oxygen was added (at the 24 hour mark) to oxidise the sulfide However, interference and contribution to silver oxidation by
ions, improved gold recovery was observed. Sulfide ions are not thiosulfate does occur at the higher potentials used to estimate
typically found in gold leach solutions unless the ore or the copper concentration.
concentrate contains appreciable reactive sulfides. For the latter,
being able to quantify sulfide ions is important for controlling CONCLUSIONS
reagent addition to optimise gold recovery.
Potentiometric titrations are recommended in preference to
Thiocyanate and thiosulfate ions rhodanine titrations when measuring the cyanide concentration in
circuits treating sulfide or copper rich ores and concentrates.
The interference of thiocyanate ions on the rhodanine and The potentiometric titration may also be tailored to yield
potentiometric titration cyanide measurements was investigated sulfide and copper ion concentration as part of the cyanide
with a solution containing 4000 mg/L NaCN and 100 mM KSCN measurement procedure.
(5.8 g/L SCN). No interference of thiocyanate was identified in
Using default set-up procedures, the online potentiometric
the determination of cyanide by either titration method.
titration measurement of cyanide will be affected by sulfide ions
The potential for interference of thiocyanate on the concurrent and a silver sulfide precipitate will eventually foul the electrode.
cyanide and copper determinations using the potentiometric The amperometric technique potentially offers comparable
titration was investigated with solutions containing 22 500 mg/L capabilities to the potentiometric titration, though careful set-up
NaCN, 2000 or 8000 mg/L Cu+ and 100 mM KSCN. In of the amperometric sensor is required for these complex leach
comparison to when no thiocyanate was present, a small increase solutions.
was found in the cyanide determination which was reflected by a
similar decrease in the copper determination. This is possibly due Gold leaching is hindered by the presence of sulfide ions and
to a complexation equilibrium of thiocyanate with copper (cyanide hence sufficient air or oxygen addition is required in the leach to
complexes much stronger than thiocyanate), which displaces a eliminate or minimise the presence of sulfide ions. The gold
small amount of CN-. Thus, the cyanide determination most likely leach rate does not always follow the ionic cyanide or any of the
reflects the ionic cyanide available for gold leaching. However, cyanide measurements from the various techniques with pH.
the copper estimation is slightly low and at high thiocyanate Above pH 9.2 the gold leach rate is constant, and only decreases
concentrations this can be as much as 15 per cent. with decreasing pH below 9.2 when cyanide is limited.
Thiosulfate ions have also been found in cyanide leach solutions
where reactive sulfides are present in the ore. Thiosulfate was ACKNOWLEDGEMENT
found to interfere with the cyanide determination using the
rhodanine titration method, resulting in an over-determination of The authors would like to thank Newcrest Mining Ltd for their
the cyanide concentration. AgCN(s) was found to precipitate financial support for a portion of the work presented in this
before the change in the rhodanine indicator colour was paper.
observed, which correlated with silver having formed AgCN(s)
and Ag(S2O3)-. REFERENCES
The potentiometric method was effective in distinguishing Breuer, P L, Jeffrey, M I and Dai, X, 2005. Leaching and recovery of
between the end points of the silver reactions with cyanide and copper during the cyanidation of copper containing gold ores, in
thiosulfate, with accurate determination of the ionic cyanide Treatment of Gold Ores (eds: G Deschenes, D Hodouin and
concentration possible. However, in copper cyanide solutions L Lorenzen), pp 279-291 (The Canadian Institute of Mining,
containing thiosulfate an estimate of the copper concentration Metallurgy and Petroleum: Montreal).
8000 Cu
Fe 50
7000
Sulfide ions
Cu, Fe, S2- (mg/L)
Au recovery (%)
6000 Au recovery 40
5000
30
4000
3000 20
2000
10
1000
0 0
0 5 10 15 20 25 30 35 40 45 50
Time (hr)
Dai, X, Jeffrey, M I and Breuer, P L, 2005. The development of a Jeffrey, M I, 1998. A kinetic and electrochemical study of the dissolution
flow injection analysis method for the quantification of free of gold in aerated cyanide solutions: The role of solid and solution
cyanide and copper cyanide complexes in gold leaching solutions, phase purity, PhD thesis, Curtin University of Technology, Bentley,
Hydrometallurgy, 76(1-2):87-96. Werstern Australia.
Heath, A R, Rumball, J A and Bailey, S, 1999. An amperometric method Jeffrey, M I and Breuer, P L, 2000. The cyanide leaching of gold
for measuring cyanide in CIP/CIL circuits, Minerals Engineering, in solutions containing sulfide, Minerals Engineering, 13(10-11):
12(11):1313-1326. 1097-1106.
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