You are on page 1of 17

Core Program Theme 3 – Integrated Comminution Modelling

Integrated Comminution Modelling


Toni Kojovic and Pat Walters
JKMRC, The University of Queensland

3.0 Executive Summary tests. A simple protocol has been defined for the inte‐
gration of the ELF meter within the Ci test, which in‐
In 2010, the Theme 3 research team has worked on
cludes a no‐load calibration at the key crusher gaps
three separate tasks, as outlined below:
used in Ci testing for each machine. Power meter in‐
stallations have been adopted by JKtech and external
Theme Task Status
• Three labs have installed ELF labs using the Ci test, and other laboratories are ex‐
Improvements pected to follow. There are still limitations in dealing
power meter; protocol avail‐
to
CP3a able. with friable core samples, and further work is re‐
GeM Comminu‐
• Latest results from CdA con‐ quired to investigate the most appropriate way of
tion Index
firm value of power meter.
managing such cases.
• A*b Express method con‐
Use of non‐core firmed.
CP3b drilling • Bond proxy model validation RBT A*b Express Concept and RBT Bond Proxy
products complete; final report
The A*b Express test concept has been confirmed as a
drafted.
rapid means of quantifying the impact resistance of
• Experimental work delayed;
Comminution • Modelling approach shown core samples, using only one test at a single low RBT
CP3c
blend response to be viable for Bond (ball energy. Once calibrated to RBT Lite or full RBT test
mill) and AG/SAG milling. results, the A*b Express method should be able to
provide a robust mapping tool with a minimal sam‐
The team has largely completed its objectives and is ple requirement. This is ideally suited to hardness
now moving to a phase of applications support and characterisation at pre‐feasibility level. Alternatively,
method validation/refinement. New ideas, such as the integration with routine assaying of blast chips offers
A*b Express concept, will be evaluated once a suitable a number of potential applications suited to short‐
prototype device is made available in 2012. The im‐ term planning:
plications of the Theme 3 developments to date are
• Reconciliation of block model hardness estimates;
discussed below.
• Adjustment of cut‐off grade based on grade and
hardness of each hole;
Improvements to GeM Comminution Index • Optimisation of ore blending from different
The GeM Ci has been shown to be ideally suited for benches; or
high volume hardness mapping. With the addition of • Adjustment of explosive addition/strength to op‐
timize powder factors.
a simple power meter, the Ci test provides more use‐
ful information on the hardness of the core material, The A*b Express concept has been under develop‐
particularly the A*b index in the harder samples ment within the Amira P9O project and a prototype
design by RME is in progress. Field units are expected
where the Ci was previously unable to resolve the
by June 2012.
difference between two samples. The improvement
was not as significant when applied to Bond predic‐ Theme 3 has finalized the development of a Bond
tion, with the lowest error being around 10%. This is grindability proxy test, based on the high energy ap‐
comparable with commercial ore hardness ‘modified’ plication of the RBT. The results on 296 samples from

GeMIIi (Amira P843A) Technical Report 8 – December 2011 38


Core Program Theme 3 – Integrated Comminution Modelling
8 ore deposits suggest the Bond ball mill work index initiatives and consolidation of the three core research
can be reliably estimated using the product size dis‐ activities, highlighted in this chapter.
tribution from the 2.0 kWh/t RBT test on the
11.2+9.5mm fraction. Hence the test adds additional 3.2 Methodology
value to the routine GeM RBT Lite test (only requiring
3.2.1 Addition of Power Measurement in the Ci Test
2.0 kWh/t product to be fully sized – via mechanical
or optical means). The Bond proxy test is also appli‐ Preliminary trials using granite suggest the specific
cable to non‐core samples which previously were not work required to crush the samples at the Ci gap, and
suitable for GeM Ci testing (e.g. blast hole samples), resulting product size distribution, should provide a
and combines easily with RBT A*b Express method‐ more reliable link to the impact resistance as
ology (0.2 kWh/t). The rapid turn‐around time (~20 expressed by the standard Drop Weight Test index
minutes) and small sample mass requirement (~100g A*b. To confirm this potential, an experimental study
for RBT Lite ‐11.2+9.5mm) makes the test ideally using Ernest Henry and Cadia East core samples was
suited to the geometallurgical testing toolkit. initiated. A total of 138 samples were subjected to Ci
and RBT testing on the 11.2x9.5mm survivor
Comminution Blend Response fragments post crushing.

The theme has established two options to generate The power meter was placed between the wall socket
comminution ‘blend response curves’. For ball mill‐ and crusher plug, as shown in Figure 3.1, so that the
ing, small‐scale physical testing and modelling car‐ unit can measure the power draw directly without
ried out using the Bond ball mill appears to be the influencing the crushing equipment in any way. The
most practical method. Initial results suggest that the current methodology requires that the crusher is
blend response is variable and needs to be calibrated warmed up prior to starting the crushing tests as
to provide reliable predictions of ball mill through‐ there is a significant warm up period associated with
put. However, impact hardness A*b values generated the equipment. To ensure only the crushing energy is
using the JKRBT or JKDWT are more decoupled from measured the power reading is zeroed before the
the milling process and assessing blend additivity crushing tests are carried out. This step takes into
requires a different approach. account the vibrating feeder and splitter equipment
attached to the Boyd crusher assembly.
A novel to generate A*b blend response curves which
may estimate the effect of varying blend proportions
on AG/SAG mill throughput has been developed us‐
ing JKSimMet. In operational practice the number of
response curves required will be determined by the
number of distinct ore types each having a unique
impact resistance and density.

JKSimMet may also be a viable tool for estimating the


Bond BMWi blend response for a given set of ore
components. The proof‐of‐concept results suggest
that the Bond test can be reliably simulated in
JKSimMet, given the results from the full Bond test.
The results suggest simulation may provide an alter‐
native to the more costly experimental investigation
using a Bond ball mill.

Figure 3.1. Photograph of power meter installation with


3.1 Introduction
Boyd crusher at Sumner Park laboratory
In 2010 the Theme 3 research team has been
A series of quick timed runs, ranging from 1 to 3
providing support across several case study
minutes, and their associated energy draws are used

39 GeMIIi (Amira P843A) Technical Report 8 – December 2011


Core Program Theme 3 – Integrated Comminution Modelling
to calibrate the operating parameters of the crusher. The first was Anglogold Ashanti who requested ALS
The resulting data essentially represent an energy Bogota to upgrade their machine currently being used
curve which can be expressed as follows: to process samples from the la Colosa deposit. The
second is Newcrest Telfer mine who have installed a
Whrbase = a × t + b (1)
new Boyd and power meter in their assay lab, ahead
where
of using the Ci test as a proxy comminution test on
a = Energy consumed per min of running time
t = Running time in minutes their VSC ore body. The research team selected an
b = Energy used in starting & stopping crusher inexpensive power meter.

The ‘a’ and ‘b’ parameters need to be calculated for As none of these installations has managed to provide
each crusher gap settings as the weight distribution of detailed information on the core samples treated,
the plates changes which affects the operating energy Teck granted permission to the Theme 3 research
of the crusher. Once derived, equation 1 can be used team to access the data from their Andacollo geomet
to determine the total input energy for a sample that comminution test program conducted by JKtech in
has been passed through the GeM Ci procedure. The Sumner Park. This program is outlined in the next
total specific input energy in kWh/t for a given section.
sample is calculated using the formula: 3.2.2 Results from Ci Tests on Andacollo Samples
Whr − Whrbase The Andacollo samples selected for the Ci power
Eis = (2)
m meter validation were subjected to the following tests:
where
Eis = Specific input energy (kWh/t) • Ci testing of 56 core samples across 6 UGs using
Whr = Energy recorded on the power meter standard GeM protocols for HQ half core.
m = Mass of crushed material in kilograms • Ci testing of SAG feed rock samples from 7 UGs
using standard GeM protocols for 26.5x22.4mm
The results from Ernest Henry and Cadia East fragments.
tests showed that the introduction of a simple power • Sub‐sampling of SAG feed and Ci products to
meter during Ci crushing has shown a consistent remove ‐11.2+9.5mm fractions for RBT Lite testing.
improvement in the A*b prediction when the specific • RBT testing of ‐11.2+9.5mm fractions using RBT
power terms are included in the calibration model. Lite protocol on all core and SAG feed sub‐
The improvement was particularly significant in the samples.
harder samples where the Ci was unable to resolve • DWT testing of SAG feed samples from 7 UGs
the difference between two samples. The success of using standard JKTech protocols. This data
the extended trial on 138 samples prompted two provided the A*b size scale‐up factor for each UG,
sponsors to install an ELF power meter on the applied to the core samples.
laboratory crusher used for Ci testing (see Figure 2). • Bond testing of SAG feed samples from 7 UGs
using standard and JKMRC Modified Bond
protocols. The difference between the two
protocols was around 3.3%.
• Modified Bond testing of 56 core samples.

The Ci power meter data includes several key inputs


which are used to calculate the net specific energy
(kWh/t) consumed per sample. An example of the
typical data values captured for each sample is shown
in Table 3.1. The distribution of specific energy
consumed per sample is wide, as evident in Figure
3.3. Overall the results conform to a normal
distribution. For competent samples, power data is
Figure 3.2. ELF power meter used in Ci testing directly related to ore hardness. If core is friable, the

GeMIIi (Amira P843A) Technical Report 8 – December 2011 40


Core Program Theme 3 – Integrated Comminution Modelling
results are unable to decouple hardness and degree of confirm the most appropriate means of managing
fines in the core interval. These samples may need friable samples submitted for Ci testing.
pre‐screening to quantify the degree of as‐received
fines. This issue requires further investigation to

Table 3.1. Example of Ci Power Meter Results Collected During Gem Ci Test

LIMS Crush time No Load Net kWh/


ID Mass [g] (min) Power [Wh] [Wh] [Wh] t
0091C 5000 2.16 29.54 26.94 2.6 0.52
0093C 7600 2.24 29.91 27.76 2.15 0.28

Andacollo Ci Power Measurements using two markers from the Ci product size
0.90
distribution:
0.80
• % ‐ 4.75mm
0.70
• % ‐ 150 microns
Net Specific Energy (kWh/t)

0.60
The ratio %‐4.75mm and specific energy is essentially
0.50
a crushing related A*b term, similar to the A*b
0.40
Express calculation using RBT test results at a low
0.30
energy:
0.20
A * b = T 10 / Ecs ≈ % − 4.75 / Eis (3)
0.10

0.00 Since the best form of the correlation between the


‐4 ‐3.5 ‐3 ‐2.5 ‐2 ‐1.5 ‐1 ‐0.5 0 0.5 1 1.5 2 2.5 3 3.5 4
DWT/RBT A*b and Ci behaviour was unknown,
N Score
LREG (Kojovic et al, 1989) was used in the multiple
Figure 3.3. Distribution of CdA Ci specific energy results
linear regression analysis to find the most accurate
Despite the potential disconnect between the crushing and robust relationship. The charts in Figure 3.4 show
behaviour of the core samples and impact resistance the quality of fit to the measured data for the seven
of the ‐11.2+9.5mm survivors, the specific energy SAG feed samples, with and without the inclusion of
consumed during crushing was expected to be highly the power measurement. The model accuracy
relevant. The key inputs in the regression analysis improvement is clearly evident, with a 50% reduction
were the specific energy and degree of size reduction in the relative error.
achieved during crushing. The latter was expressed

No Power Meter With Power Meter


80 80
Rel Error = ±24.0% Rel Error = ±12%
70 70

60 60
A*b (pred from Ci)

A*b (pred from Ci)

50 50

40 40

30 30

20 20

10 10

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

A*b (DWT) A*b (DWT)

Figure 3.4. Summary of A*b model relations for Andacollo SAG feed samples, with and without Ci Power data

41 GeMIIi (Amira P843A) Technical Report 8 – December 2011


Core Program Theme 3 – Integrated Comminution Modelling
Initial attempts at developing a single universal source of error was confirmed when the same
relation for the 56 core samples were unsuccessful, so comparison was made between the estimated and
specific relations were derived for each UG ore type. actual A*b values for the SAG feed samples, where
The charts in Figure 3.5 show the quality of fit to the DWT, RBT Lite and SMC estimates were all available.
measured data, with and without the inclusion of the This simply means that the A*b values used in the
power measurement. The core results also show a regression analysis are probably not as accurate as
consistent improvement in the A*b prediction when first expected. Hence the best possible accuracy of the
the power terms are included, though not as Ci based model is unlikely to be better than 10%.
significant. This may be related to another potential This is comparable with commercial ore hardness
source of uncertainty, the scale‐up from the RBT ‘modified’ tests.
11.2+9.5mm results to the DWT equivalent size. This

No Power Meter With Power Meter


80 80
Rel Error = ±12.0% Rel Error = ±10%
70 70

60 60
A*b (pred from Ci)

A*b (pred from Ci)


50 50

40 40

30 30

20 20

10 10

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

A*b (DWT equiv) A*b (DWT equiv)

Figure 3.5. Summary of A*b model relations for Andacollo core samples, with and without Ci Power data

Another aspect from the Andacollo testwork that variables; constant, normalized Ci‐GRD index and a
warrants mention is that soft and friable core intervals material hardness index determined using the
do not generate sufficient coarse survivors for RBT measured specific energy consumed, and Ci feed and
testing post Ci crushing. This was particularly evident product sizes.
with UG7 core samples which only registered RBT
With Power Meter
samples for the harder examples in this domain, 25

having A*b values below 50. However the UG7 SAG


20
feed sample was found to be extremely soft, with a
BMWi ‐ kWh/t (pred)

DWT A*b value of 300. The RBT A*b value for the 15
SAG feed 11.2+9.5mm fraction was also very soft,
with an A*b value of 250. No material as soft as this 10

was found in any core intervals from UG7.


5
The relationship between the Andacollo Ci and
Bond test results were also investigated using LREG 0
0 5 10 15 20 25
analysis. Various model forms, including linear and
BMWi ‐ kWh/t (meas)
non‐linear expressions, were evaluated. The analysis
suggests the improvement in model accuracy with Figure 3.6. Summary of Bond BMWi model quality for
Andacollo SAG feed and core samples with Ci Power data
power measurement is not as significant as it was for
A*b, with the best possible error being around 9‐10%, 3.2.3 RBT A*b Express Concept
in line with historical P843 results. Figure 3.6 shows
The validity of A*b Express concept as a rapid
the accuracy of the Ci based model, using only three
means to estimate the A*b index was investigated

GeMIIi (Amira P843A) Technical Report 8 – December 2011 42


Core Program Theme 3 – Integrated Comminution Modelling
using Ernest Henry, Cadia East and Andacollo a single test at the specific energy of 0.2 kWh/t. Figure
samples. In each instance, RBT Lite testing was 3.11 highlights the simple workflow of the A*b Ex‐
carried out to determine the A*b for the 11.2x9.5mm press test.
size fraction. The T10 value for the 0.2 kWh/t test was
Ernest Henry
then used to estimate the A*b, knowing that the A*b 120
can be approximated reliably using the slope of the
100
T10‐Ecs curve at 0.2 kWh/t (see Figure 3.7).
80
T 10
A* b ≈

A*b Lite
(3) 60
0.2
where
40
T10 = percent passing 1mm at 0.2 kWh/t
20 y = 1.0027x + 4.0318
T10 (%) R² = 0.9922
0
38.1
0 20 40 60 80 100 120
19.8 A*b Express

10.4
Figure 3.9. Comparison of A*b values derived from RBT
Lite and A*b Express tests on Ernest Henry samples
4.4
A*b =23 (hard ore)
Andacollo ‐ SAG Feed
0.2 0.5 1.0 2.0 80
Energy (kWh/t) 70

Figure 3.7. A*b estimated using slope at Ecs=0.2 kWh/t 60


(A*b Express = 4.4/0.2 = 22) 50
A*b Lite

Figures 3.8 and 3.9 show the Cadia East and 40

Ernest Henry results respectively, confirming the 30

strong correlation between the two measures across 20

both sets of samples. 10

0
Cadia East 0 10 20 30 40 50 60 70 80
50
A*b Express

40
Andacollo ‐ Core
80
30
A*b Lite

70

60
20
50
A*b Lite

10 40
y = 0.8772x + 3.2861
R² = 0.9544 30
0
20
0 10 20 30 40 50
10
A*b Express
0
Figure 3.8. Comparison of A*b values derived from RBT 0 10 20 30 40 50 60 70 80
Lite and A*b Express tests on Cadia East samples A*b Express

The Andacollo results also provided the opportunity Figure 3.10. Comparison of A*b values derived from RBT
to check the validity of the A*b Express test. Figure Lite and A*b Express tests on Andacollo samples
the results from the A*b Express and RBT Lite com‐
parison on both SAG feed and core samples, which
support the previously reported finding that the A*b
Express test can provide a reliable A*b estimate using

43 GeMIIi (Amira P843A) Technical Report 8 – December 2011


Core Program Theme 3 – Integrated Comminution Modelling
size distribution and the corresponding Bond ball mill
work Index (BMWI) of the core sample.

Figure 3.12 compares three samples with a very


wide range in Bond grindability and corresponding
RBT product size distributions.

M −1mm
T10 = 100 × %
M

T10
A*b = Figure 3.12. Comparison of RBT 2.0 kWh/t product size
0.2
distributions for 3 samples with wide range in BMWi
Figure 3.11. Flowchart of A*b Express test
The investigation suggests that the BMWi can be re‐
3.2.4 Bond Proxy Model Validation liably estimated using the associated RBT product
The JKRBT is conventionally used in the predic‐ size distribution from the highest energy (2.0kWh/t)
tion of an A*b hardness index for a given ore. How‐ GeM RBT Lite test.
ever, Theme CP3b investigations have found that the
size distributions produced through this mode of RBT Wi (kWh/t) = k1 × RBTGRD k2 (4)
breakage contained more information than first ex‐ where
pected, suggesting the potential to reliably derive an k1, k2 = site specific constants
equally important comminution parameter, the Bond RBTGRD = Index calculated using RBT 2.0 kWh/t
Ball Mill Work Index (BMWi). Figure 3.11 shows the product size distribution
two test devices, JKRBT and Bond Ball Mill.
The investigation found that the highest RBT Lite en‐
ergy had the strongest correlation with the BMWi.
One of the key reasons believed to be behind this cor‐
relation may be due to the similarities between the
energy inputs of the highest energy JKRBT test (2.0
kWh/t) and the standard 100 revolutions for a Bond
Test (~1.96 kWh/t). The fact that hard survivors from
any core sample tend to dominate the Bond test
grindability, would also support the strong correla‐
tion. Finally, it is well known that most ore types
show a strong correlation between the Bond Wi and
impact hardness A*b. The outcomes of this develop‐
Figure 3.11. JKRBT used for A*b testing and Bond Ball Mill ment provide the following benefits:
used for grindability tests
1. Potential new opportunities for testing non‐core
Since the start of P843A, the research team has fo‐
drill products which, conventionally, could only
cused on the potential link between the RBT product
be characterized using a full Bond ball test.

GeMIIi (Amira P843A) Technical Report 8 – December 2011 44


Core Program Theme 3 – Integrated Comminution Modelling
2. A rapid testing option for the mapping and classi‐ The results suggest that with a site specific calibra‐
fication of the ore grindability ideally suited to tion, the relative error can be as low as 4.6% (see Table
the high volume needs of geomet variability test‐ 3.3). On average, across all eight sites, the relative er‐
ing. ror was 6.8%. The RBT Wi proxy method was not as
3. New testing applications made possible through successful for the Los Grande and La Colosa Cu por‐
the minimal mass requirements of the Bond phyry deposits, reflected in the highest relative errors.
proxy test. Additionally, the method had to be corrected to take
into account the coarse 300 micron closing size for Los
The prediction of grind response from the RBT prod‐
Grande, since the RBTGRD index is calculated using
uct size distribution, referred to as the RBT Wi, has
data down to 106 microns. Fine closing sizes
been validated using a comprehensive dataset con‐
(<106μm) still need to be tested.
sisting of 296 samples sourced from the AMIRA
GeMlll P843/P843A archives. The range of the data‐
Table 3.2. Comparison of RBT Wi Test Accuracy (Site Spe‐
base is illustrated in Figure 3.13.
cific Models)

Site Relative Error


Aqqaluk 4.6%
Bingham BHC 5.3%
Bingham Core 5.3%
Boddington 7.8%
Cowal 5.6%
La Colosa 8.2%
Los Grande 8.1%
Resolution 5.1%
Sudbury 4.7%
All Samples 6.8%

Figure 3.13. Dataset used in validation of RBT Wi Concept


The RBT Wi proxy method is more accurate than the
The results indicate a high level of accuracy, as illus‐ GeM Ci test, and has the added benefit of already be‐
trated in Figure 3.14, believed to be sufficient for ing included in the GeM RBT Lite test. The only re‐
mapping the ore deposit variability. The Bond proxy quirement is that the product from the highest energy
test requires only a small sample volume and offers a RBT test must be sized fully sized, either using me‐
rapid turnaround time; thereby enabling processing chanical sieves or optical systems like the Camsizer.
of a large number of samples and a significant reduc‐
The RBT Wi method, combined with the recent
tion in cost over full Bond ball mill work index test‐
JKMRC development of the A*b Express test, means
ing.
that with only two single energy RBT tests, one high
(2.0kWh/t) and one low (0.2kWh/t), the A*b and
BMWi of the ore sample can be estimated in a fraction
of the time required using standard tests. This has
historically never been possible. The two tests provide
a host of new potential characterization opportunities,
ideally suited to high volume geometallurgical test‐
ing. Figure 3.15 shows the simple workflow of the
RBT Wi proxy test.

Figure 3.14. Comparison of predicted versus measured


Bond BMWi values for 296 samples, across 8 sites

45 GeMIIi (Amira P843A) Technical Report 8 – December 2011


Core Program Theme 3 – Integrated Comminution Modelling
two gold ores from Western Australia, having BMWi
FEED values of 6.6 and 14.0 kWh/t, concluded that there
was a strong interaction between the components of
the blend, the harder material having the greater
JKRBT influence on the overall breakage properties and
@ 2.0 kWh/t
BMWi. For example, at 25% hard component, the
actual BMWi was 30% higher than expected from a
linear response. At Sudbury the effect of the hard
Mechanical or
Optical Size component is much lower, less than 7%.
Analysis

Product Size
Distribution

GRD Index

BMWi = k 1 × GRD k 2

Figure 3.15. Flowchart of RBT Wi Proxy test


Figure 3.16. Bond grindability index response curves for
3.2.5 Modelling the Comminution Blend Response the Sudbury ore and dilution rock blends

Blend Response for Bond Ball Mill Grindability


More tests were slated, but the Queensland
Theme CP3c aims to develop methods to floods in January 2011 impacted the P843A sample
generate comminution ‘blend response curves’. The storage facilities in Sumner Park, and delayed plans
initial work focused on small‐scale physical testing to continue this work. In the absence of additional
and modelling carried out using the Bond ball mill. experimental Bond blend response data, the research
Sudbury ores were selected for the case study since team moved their effort to an alternative approach, a
the Clarabelle operation sources ore from several methodology to quantify the blend response using
mines and has reported large throughput variability. simulation of the Bond process. JKSimMet was used
As the mill has little control over rock dilution in the to essentially calibrate the Bond mill process using the
feed, a series of blends were tested to see if there was detailed results for the two main components that in
a response to the ore to wall rock dilution. Four practice will be blended. The ‘proof‐of‐concept’ was
dilution rock types were used, mixed with massive based on Bond test data from Vale Inco, Sudbury,
sulphide ore, from 0 to 100% in 20% increments. where hard waste rocks appear with the massive
The results, summarized in Figure 3.16, suggest sulphide ore as part of the feed to the Clarabelle
that the Bond blend response is variable and needs to milling plant. The results from a series of Bond tests
be calibrated to provide reliable predictions of ball run on selected blends of wall rock and sulphide were
mill throughput. However, the testing requires reported in Feb 2010. Details of the Bond process
significant care in sample preparation and handling, modelling are presented below.
otherwise the outcome is highly questionable. Proof of Concept
The relatively low non‐linearity of the blend Before explaining the modelling process, it is
response curves for Sudbury is not consistent with important to highlight the key elements of the
published research on multi‐component Bond tests standard Bond ball mill work index test.
(Yan and Eaton, 1993). Yan and Eaton, working with The standard Bond mill is 0.305m by 0.305m. It

GeMIIi (Amira P843A) Technical Report 8 – December 2011 46


Core Program Theme 3 – Integrated Comminution Modelling
has a revolution counter and runs at 70rpm. The given ore sample are:
grinding charge consists of 285 balls, the total
1. Frequency of breakage and discharge rate for each
weighing 20.125 kg. The standard feed is prepared by
particle size ‐ (primarily a function of ore hardness
stage crushing to pass a 3.35 mm sieve. To derive a
and feed size distribution/density)
Bond Ball Mill Work Index (BBMWi) a series of
consecutive batch grinds are conducted using the 2. Classification performance of the closing screen
Bond mill. (function of screen integrity and operator method).

The initial feed material, equivalent to 700cm3 of The above mechanisms are easily handled using
packed material, is ground for a specified number of the JKSimMet ball mill model:
cycles and the material passing the nominated closing 1. The breakage rates are inherently combined with
size is removed. New feed is then added to the the discharge rates in a normalised single
oversize to maintain the initial mass, and the process parameter represented by (R/D*) as a function of
is repeated. The number of revolutions needed in particle size. Four size knots at 0.075, 0.15, 0.60
each cycle is determined from the results of the and 1.18mm were selected to cover the size range
previous period to produce sieve undersize equal to of the Bond ball mill feed, as illustrated in Figure
1/3.5 of the total charge in the mill. This is continued 3.17 for the Sudbury sulphide ore.
until a stabilization point has been reached, where the
2. The closing screen can be modelled using a
net grams of sieve undersize produced per mill
simple efficiency curve described by three
revolution reaches equilibrium, and the mass of the
parameters:
oversize is consistently 2.5 times that of the undersize
• α ‐ sharpness of the split
(representing the nominal 250% recirculating load in
• C ‐ fines split to undersize
a closed ball mill circuit).
• d50c ‐ cut‐size
At the end of the test, the undersize product is
The efficiency curve for the final cycle Bond test
screen analysed, and the average of the last three net
screening analysis is shown in Figure 3.17, reflecting a
grams of final product size generated per revolution
very efficient process.
(Gbp) is defined as the ball mill grindability. The ball
mill work index (kWh/t) is calculated from the The calibration of the Bond test was carried out
following equation (Bond, 1961): for the final cycle, where the screen analysis is
49.1 conducted on the product. The feed size distribution
WI = (4)
⎛ 10 ⎞ is also known, as is the final cycle recirculating load
10
P10.23 x Gbp0.82 x ⎜ − ⎟
and number of revolutions. The only missing was the
⎜ P F80 ⎟
⎝ 80 ⎠
actual screen oversize size distribution. The approach
where used to estimate this stream was simple. Firstly, the
WI = Bond ball mill work index ‐ a material‐ final cycle data was initially fitted with the oversize
specific constant (kWh/t) recycling back to the feed, only knowing the quantity
P80 = 80% product passing size (μm) of recycle but not the actual size distribution.
F80 = 80% feed passing size (μm) Secondly, once the size distribution of the final cycle
Gbp = grams per revolution when test stabilizes oversize was established, the recycle stream was
P1 = closing screen size (μm) decoupled and the fitting was repeated in open
Based on historical data, the average P80 for a circuit, as shown in Figure 3.17. This approach was
150 μm closing sieve size is 114 microns; at 106 μm it found to give the most reliable set of parameters as
is 76 microns, at 75 μm it is 50 microns, and at 45 μm the initial calibration was too sensitive to the starting
it is 26.7 microns. These relations are used in parameter values.
equation 4 when the P80 is not measured.

In terms of the mechanisms in the Bond test mill,


the two key aspects that need to be modelled for a

47 GeMIIi (Amira P843A) Technical Report 8 – December 2011


Core Program Theme 3 – Integrated Comminution Modelling

breakage rate curve


100.0
Sulphide
Greenstone
10.0

R/D*
1.0

0.1

0.0
0.01 0.10 1.00
Particle Size (mm)

efficiency curve
100

80

60

% to O/Size
40

20

Sulphide 0
0.150

0.01 0.10 1.00

Particle Size (mm)

Figure 3.17. JKSimMet model representation of final cycle in the Bond test for Sulphide ore, showing ball mill model
parameters and efficiency curve for typical Bond test 150 micron closing screen

Note that the feed was normalized to 100% to parameters for the greenstone and sulphide ore
simplify the modelling process and subsequent use of samples, the next step was to construct the complete
the model in simulating the complete Bond test. And 7‐cycle Bond test flowsheet in JKSimMet, as
similarly, the JKSimMet Bond mill was assumed to be illustrated in Figure 3.19. The flowsheet was then
3m diameter x 3m long. validated for both greenstone and sulphide, using the
model as the virtual process. The calculation of the
The key question was whether the conceptual
required number of revs per cycle was carried out
model could adequately describe the actual Bond mill
using the same protocols as per the standard Bond
performance, specifically the mill product size
test (see Figure 3.20). However, the most appropriate
distribution and recycle load percentage. Figure 3.18
way to translate the number of revs into JKSimMet
shows the final calibration results suggesting the
equivalent terms was initially unknown. It was found
JKSimMet model is capable of representing the Bond
that the most convenient scaling method would be via
mill process sufficient well. The calibration included
the effective length of the mill. Hence setting 3m to
four R/D* parameters plus three screen parameters.
represent the number of revs in the final cycle was the
100

90
key. The number of revs for any cycle could then be
80 directly calculated from the number of cycles
Cumulative % Passing

70
required to reach the equilibrium.
60

50
This approach was tested on the greenstone and
40

30
sulphide data and found to be in close agreement,
20 given the inherent uncertainty in the physical Bond
10
test (see Table 3.2).
0
0.01 0.1 1
Size (mm) Table 3.2. Comparison of Bond BMWi Results derived from
FEED Exp Product Exp Product Cal SCR O/S Exp SCR O/S Cal
Actual and Virtual Bond Tests (Single Feed)
Figure 3.18. Quality of JKSimMet model fit to Bond Sample ID Actual Virtual Expected
greenstone final cycle test data kWh/t kWh/t Error
Greenstone 15.8 15.7 0.55
Having calibrated the JKSimMet model Sulphide 9.5 9.3 0.33

GeMIIi (Amira P843A) Technical Report 8 – December 2011 48


Core Program Theme 3 – Integrated Comminution Modelling

Figure 3.19. JKSimMet model representation of 7‐cycle Bond test flowsheet

Figure 3.20. Excel template for Bond virtual test using the same protocols

Generating the Bond Blend Response Curve already shown that the contents in the Bond mill are
biased towards the hard component when treating
Once the Bond mill performance was calibrated
mixtures of ore. Therefore, the key was to find the
using the JKSimMet model, and the simulation of the
mixing rule for any two components when tested
complete test was shown to be viable, the approach
under similar conditions. This rule was derived from
was extended to a blend of greenstone and sulphide.
the Sudbury Bond data using four dilution rock types
The goal was to simulate a blended feed using only
and sulphide ore. The approach provides the
one JKSimMet flowsheet, as keeping track of parallel
weighting factor, given the feed composition of hard
simulations would be very complex and difficult to
and soft components:
run. However, knowing that the mill breakage rates
were different for greenstone and sulphide presented Pb = Ps + ( Pg − Ps ) * %G * F (5)
a problem. Could they be simply combined in a where
weighted average given the feed composition? Pb = model parameter for feed blend
This was not expected to work since P843 has Pg = model parameter for greenstone

49 GeMIIi (Amira P843A) Technical Report 8 – December 2011


Core Program Theme 3 – Integrated Comminution Modelling
Ps = model parameter for sulphide estimating the Bond BMWi blend response for a given
%G = proportion of greenstone in feed set of ore components. To carry out such simulations
F = weighting factor the full Bond test results for each ore component are
required. Ideally the final cycle screen oversize
The weighting factor is a function of greenstone should be sized to simplify the application. To
and sulphide BMWi values, and the feed composition. validate the mixing rule, one Bond test on a feed
This approach needs further work to validate the blend should also be included.
underlying mixing rule across other ore type
Though the approach has limitations, it provides
mixtures. The virtual test is also run until a
a simple basis for investigating the multi‐component
stabilization point is reached:‐
behaviour of two ores of differing Bond hardness
1. Grams per Rev reaches equilibrium, and where only the ore blend is varied. The Bond Blend
2. Mass of the Oversize is 2.5 times that of the response model will be further tested using a
Undersize (the target 250% recycle load in a selection of samples to be tested as part of the P843A
closed ball mill circuit). case studies.

Figure 3.21 shows a plot of the simulated blend


Blend Response for A*b Hardness
response for the Bond test treating different blends of
greenstone and sulphide. This theme has shown that experimental
determination of the blend response in the Bond ball
mill, treating two components of significantly
17

16
different impact resistance, is viable. However,
15 experimental determination of the blend response for
BBMWi (kWh/t)

14 an AG/SAG mill treating two components of


13
significantly different impact resistance is challenging
12
for two main reasons,
11

10
1) small scale laboratory tests have yet to be
9 confirmed as valid representations of the full scale
0% 20% 40% 60% 80% 100% milling environment, and
% Wall Rock
2) pilot plant testing requires 10‐20 tonnes of ore
Actual Bond Response Bond Simulation Linear Mixing
per scenario so preclude the use of drill core.
Figure 3.21. Blend response curve for the Bond mill In order to mitigate this impasse, modelling was
for different greenstone and sulphide feed blends
considered as potential way forward. However, as
There is clearly an interaction between the JKSimMet currently has no capacity to simulate the
components, the harder material having the greater interaction of two ore types having different impact
influence on the overall Bond work index. For resistance, density and/or feed size distributions, an
comparison, the linear response is superimposed alternative modelling approach was considered. This
showing 5‐6% under‐estimation in the expected Bond approach aims to essentially calibrate the AG/SAG
work index at the 60/40 blend. process under a set of given conditions for which
This bias is relatively low compared to the multi‐ survey data and ore properties of the main
component Bond tests conducted by Yan and Eaton components are known. The ‘proof‐of‐concept’ was
(1993), claiming the BMWi was 30% higher than based on data from an operation treating a blend of
expected at a 25% hard blend. The higher degree of two ore types. Details of this case study are presented
non‐linearity seen in Yan and Eaton’s work may be a below.
function of the ore types and range of BMWi used in
Proof of Concept
their study (6.6 and 14.0 kWh/t). Further work is
The milling operation in question was surveyed
therefore required within P843A to explore this topic.
by JKMRC to generate detailed sizing and
The success of the ‘proof‐of‐concept’ study
performance data on a 21 ft. x 19 ft. autogenous (AG)
suggests JKSimMet may be a viable tool for

GeMIIi (Amira P843A) Technical Report 8 – December 2011 50


Core Program Theme 3 – Integrated Comminution Modelling
mill treating a blend of 85% soft and 15% hard ore. and 2) the grate classification behaviour, as illustrated
The product is discharged to a trommel that screens by Figure 3.23. As an initial concept model, these two
out the pebbles for use downstream. The fine product mechanisms were represented in JKSimMet using the
is classified in a large screw classifier, with the simplest breakage model (T10) applied to the size
oversize reporting back to the mill. The fine product ranges defined by the standard 0.25, 4.0, 16.0, 44.8
reports to a pebble mill. The AG mill pebble discharge and 128mm particle sizes. The product from each
is screened to provide the pebble feedstock. Figure breakage event/size range reports to the grate. The
3.22 shows the JKSimMet schematic of the surveyed particles finer than the grate contribute to the final
AG mill circuit. AG mill product, whereas the coarser particles are
recycled and form the mill load. This conceptual
Feed model is illustrated in Figure 3.23.
The main question was whether the conceptual
model could adequately describe the actual AG mill
AG Mill performance, specifically the mill product and load
size distributions. Figure 3.24 shows the final
Trommel
calibration results suggesting the conceptual model is
capable of representing the complex mechanisms of
the AG mill process. Surprisingly, the quality of fit is
Pebbles excellent for the product and mill load size
distributions. The calibration included five screen
Spiral
sizes, the t10 levels for each size and grate
Fine Prod-
Oversize uct classification parameters.
Recycle
100

90
Figure 3.22. Schematic of AG mill circuit used in A*b 80
blend response proof‐of‐concept study
Cumulative % Passing

70

60

50
In terms of the mechanisms in the AG mill, the 40

key aspects that need to be considered in the model 30

20
include:
10
1. frequency of breakage events for each particle 0
size ‐ breakage rate (function of speed) 0.01 0.1 1 10 100 1000
Size (mm)
2. energy associated with each breakage event – AG Feed Exp AG Product Exp AG Product Cal AG Load Exp AG Load Cal

function of mill load and diameter


Figure 3.24. Quality of conceptual model fit to AG
3. size distribution of the products from each
mill survey data
breakage event ‐ appearance function
The quantity of recycle did not match the load
4. classification performance of the grate mass in the mill, but this is not surprising since the
5. relationship between slurry hold‐up and breakage in the mill was modelled using a series of
transport (flowrate) out of the mill ‐ discharge simple impact models which assume each particle in
function
the feed is broken once, at the same T10 (or Ecs,
In attempting to model an AG mill process, energy per unit mass).
either using JKSimMet or other means, the model In the real process, the breakage combines
essentially has to quantify the above mechanisms. impact, attrition and abrasion, and the applied energy
This calibration step represents an opportunity to is distributed across all three mechanisms. As such,
develop a simplified model that describes the two key the simulated recycle better reflects the theoretical
aspects for a given set of mill and feed ore conditions, recycle of material in the mill should the process ever
namely 1) the breakage energy‐by‐size distribution, be operated solely in impact model at six nominal Ecs

51 GeMIIi (Amira P843A) Technical Report 8 – December 2011


Core Program Theme 3 – Integrated Comminution Modelling
levels.

Feed Grate
T10128

Product
128mm T1044.8

Recycle
T1016.0
44.8mm
Spiral
Oversize
16mm T104.4
Recycle

4.4mm T100.25

0.25mm T10-0.25

Breakage-
by-Size

Figure 3.23. Schematic diagram of AG/SAG mill process (Napier‐Munn et al, 1996), and conceptual model of the key
breakage and classification mechanisms

Generating the A*b Blend Response Curve 1.0


Once the AG mill performance was calibrated using 0.9
0.8
the conceptual model, the effective impact energy at
TPH (normalized)

0.7
0.6
each size range was determined for the mill and 0.5
calibration ore blend. This energy‐by‐size array can 0.4
0.3
then be applied to any other feed blend by simulating 0.2
0.1
the change in energy caused by the change in ore
0.0
composition in the mill load (or recycle). This requires 0 25 50 75 100
% Soft (Ore Type B)
the hard and soft ore components to be simulated
Simulated Blend Response Calibration Point Linear Mixing
separately, iterating until the combined volumetric
filling (load) reaches the target at the set feed blend. Figure 3.25. Blend response curve for the AG mill
In each iteration, the applied energy is scaled treating two ore types mixed at different blends
according to the average density of the mill load.
The above results are consistent with
The resulting energy‐by‐size trend was found to
observations made in the Multi‐Component research
be consistent with the typical form of the AG/SAG
study currently being carried out as part of the Amira
breakage rate curve (Napier‐Munn et al, 1996), which
P9O project.
is encouraging.
The success of the ‘proof‐of‐concept’ study
Figure 3.25 shows a plot of the simulated blend suggests the proposed conceptual model may be a
response for the AG mill treating different blends of viable tool for estimating the A*b blend response for a
the two ore components in the feed. There is clearly a given mill configuration. The information and data
strong interaction between the components of the required to carry out such simulations is the
blend, the harder material having the greater following:
influence on the overall breakage properties and mill
throughput. For comparison, the linear response is 1. JKSimMet model of the AG/SAG mill circuit
superimposed which shows a 30% over‐estimation in treating a known blend of two distinct ore
components.
the expected throughout at the 50/50 blend.
a. Ideally the model should be calibrated to
detailed survey data which includes

GeMIIi (Amira P843A) Technical Report 8 – December 2011 52


Core Program Theme 3 – Integrated Comminution Modelling
measured feed ore parameters A*b and SG. been completed. The study was based on a compre‐
b. The model provides the key inputs of mill hensive database consisting of 296 samples across
load and product size distributions at the eight deposits. The results suggest that with a site
survey conditions. specific calibration, the relative error can be as low as
2. JKRBT or JKDWT ore parameters for the survey 4.6%. On average, across all eight sites, the relative
AG/SAG mill feed sample, including SG. error was 6.8%. The strong correlation appears to be
3. JKRBT or JKDWT ore parameters for both ore related to the similarity of the energy applied in the
components, including SG. Bond mill and the fact that the Bond work index is
In greenfield cases, where the circuit has yet to dominated by the harder components in any given
be designed, a JKSimMet model scaled from an sample. The combination of the A*b Express and RBT
appropriate baseline to design conditions can be Wi proxy tests offers a rapid system for mapping the
used. The feed ore hardness parameters for the design A*b and Wi, ideally suited to the geometallurgical
case would need to be assumed, based on the variability testing.
expected typical feed blend and nominal hard and CP3c/1 – Experimental determination of the
soft ore types in the deposit. blend response in the Bond ball mill, treating two
Though the conceptual model has limitations, it components of significantly different impact resis‐
provides a useful basis for investigating the multi‐ tance, is viable. However, in the absence of additional
component behaviour of two ores of differing experimental Bond blend response data, the research
hardness/SG in a given AG/SAG milling environment, team has developed a JKSimMet approach to quantify
where only the ore blend is varied; a scenario typical the blend response of the Bond process. The proof‐of‐
to geometallurgical throughput modelling concept results suggest that the Bond test can be re‐
applications. liably simulated in JKSimMet, given the results from
the full Bond test. The results from the case study
3.4 Conclusions using Sudbury ores are very encouraging, but further
validation is warranted.
CP3a ‐ The introduction of a simple power meter
during Ci crushing has shown a consistent improve‐ CP3c/2 ‐ The difficult task of quantifying the A*b
ment in the A*b prediction when the specific power blend response in AG/SAG milling through physical
terms are included in the calibration model. The im‐ experiments may be resolved through a series of vir‐
provement was particularly significant in the harder tual experiments based on a simulation model of the
samples where the Ci was unable to resolve the dif‐ process on a given ore blend. The proof‐of‐concept
ference between two samples. The improvement was results suggest that a simplified model framework in
not as significant when applied to Bond prediction. JKSimMet can reproduce the key mechanisms in the
Overall, measurement of energy consumed per unit mill expressed in terms that can be easily changed
weight should become routine practice when consid‐ when considering other feed blends. The resulting
ering Ci testing. blend response curve using the new conceptual
model appear to be realistic and provide a more ro‐
CP3b/1 ‐ A*b Express has been confirmed as a
bust approach for investigating the multi‐component
rapid means of quantifying the impact resistance of
behaviour of two ores of differing hardness/SG in a
core samples, using only one RBT test at the 0.2 kWh/t
given AG/SAG milling environment.
specific energy. Once calibrated to RBT Lite or full
RBT test results, the A*b Express method should be
3.5 Acknowledgements
able to provide a robust mapping tool with a minimal
sample requirement. This is ideally suited to hardness The information used in the development of the
characterisation at pre‐feasibility level or short‐term conceptual model was made possible through Julius
planning. Kruttschnitt Mineral Research Centre and P9O
AMIRA project.
CP3b/2 ‐ The development of a Bond proxy test
based on the high energy application of the RBT has

53 GeMIIi (Amira P843A) Technical Report 8 – December 2011


Core Program Theme 3 – Integrated Comminution Modelling
3.6 References
Bond, F.C., (1961). Crushing and Grinding Calcula‐
tions, Allis‐Chalmers publication, No.
O7R9235B.

Kojovic, T., and Whiten, W.J., 1989. Automated model


building in mineral processing, Proc Comp.
Tech. and Appl. Conference, Brisbane, July, 479‐
486.

Napier‐Munn, T.J., Morrell, S., Morrison, R.D. and


Kojovic, T., 1996. Mineral Comminution Cir‐
cuits: Their Design and Optimisation. JKMRC,
University of Queensland, Brisbane, 413pp.

Stark, S., Perkins, T. and Napier‐Munn, T.J., 2008. JK


Drop Weight Parameters: A Statistical Analysis
of Their Accuracy and Precision and the Effect
on SAG Mill Comminution Circuit Simulation,
published in AusIMM MetPlant Proceedings,
18‐19 August, pp 147‐156.

Yan, D & Eaton, R 1993, ʹBreakage properties of ore


blendsʹ, Minerals Engineering, vol. 7, no. 2, pp.
185‐99.

GeMIIi (Amira P843A) Technical Report 8 – December 2011 54

You might also like