Development of recovery domains: Examples from the Prominent Hill IOCG deposit, Australia

Minerals Engineering 64 (2014) 7–14

Contents lists available at ScienceDirect

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

Development of recovery domains: Examples from the Prominent Hill IOCG deposit, Australia Julie Hunt a,⇑, Ron Berry a, Dee Bradshaw b, Brett Triffett c, Steve Walters b,d a

CODES ARC Centre of Excellence in Ore Deposits, University of Tasmania, Private Bag 79, Hobart, Tasmania 7001, Australia JKMRC (Julius Kruttschnitt Mineral Research Centre), University of Queensland, Indooroopilly, Queensland 4068, Australia c OZ Minerals, Melbourne, Victoria 3000, Australia d CRC ORE (Optimising Resource Extraction), University of Queensland, St. Lucia, Queensland 4067, Australia b

a r t i c l e

i n f o

Article history: Received 30 November 2012 Accepted 18 March 2014

Keywords: Recovery Geometallurgical modelling Sulphide ores

a b s t r a c t New methods and concepts for recovery domaining with outputs that are suitable for geometallurgical modelling are being investigated. In the work reported here, mineralogical and textural information obtained for drill core samples by a combination of techniques has been investigated to develop models for predicting recovery. The simplest and cheapest method involves assay data only and gives predicted results with ±4% RMS error. Since this model involves only assay data it could be used to predict Cu recovery values for the remainder of the deposit. The predicted Cu recovery values can be used to rank samples and divide the deposit into recovery domains that are suitable for integration into the planning process for mining, mineral processing and scheduling. Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction Prominent Hill is an iron oxide–copper–gold deposit in South Australia owned and mined by OZ minerals. Ore is obtained from open pit and underground mines and is treated via a grinding and flotation processing plant with a 10 Mt pa capacity (OZ Minerals, 2012). The first saleable concentrate was produced in February 2009 and a total (measured, indicated and inferred) resource of 186 Mt of 1.1 %Cu and 0.7 g/t Au were reported in June 2013 (OZ Minerals, 2013). The deposit is made up of copper- and gold-bearing hematiterich breccias (Belperio et al., 2007). Copper and gold principally occur in chalcocite + gold (±bornite ± covellite ± diginite) and chalcopyrite + gold (±bornite ± uraninite ± fluorite ± pyrite) style mineralisation within a hematite-dominant matrix (Belperio et al., 2007). Iron oxide-white mica-silica alteration is pervasive within and marginal to the main breccias and is surrounded by a wider zone of less intense alteration. An investigation of methods for predictive recovery of copper suitable for geometallurgical modelling was carried out on samples from Prominent Hill (Hunt et al., 2011a). The primary aims were to improve prediction of recovery and classify recovery domains by assessing the variability of different ore types through the use of

⇑ Corresponding author. Tel.: +61 362262782. E-mail address: [email protected] (J. Hunt). http://dx.doi.org/10.1016/j.mineng.2014.03.014 0892-6875/Ó 2014 Elsevier Ltd. All rights reserved.

parameters that can be routinely and cost effectively measured on ore material. Chemical and mineralogical information was obtained along with the results of batch flotation in order to develop a range of models for recovery of copper. Testing identified the simplest most effective recovery model which was then used to calculate predicted recovery values for samples in the site data base. This allowed the samples to be ranked and, based on the ranking, the deposit could be divided into potential recovery domains. The modelling and domaining should be viewed as an iterative process and can be modified as new areas of the mine are developed or additional ore types identified.

2. Sample selection and characterisation 2.1. Sample selection A sample set with highly constrained characteristics that covered a wide compositional range especially in Cu grade, sulphide speciation and gangue association was required to support the construction of a recovery model. A review of site ore type definition and multivariate (i.e. principle component) analyses of an assay data base provided by site, led to the development of a copper sulphide speciation diagram and a gangue discriminant diagram that could be used to aid sample selection (Fig. 1; Walters and Hunt, 2011). The gangue discriminant diagram is based on analysis of a set of site data from 313 drill holes and includes a full suite of

8

J. Hunt et al. / Minerals Engineering 64 (2014) 7–14

Fig. 1. Cu speciation (top) and gangue discriminant (bottom) diagrams developed for Prominent Hill samples. Left diagrams show all 169 GeM samples (labels as in right diagram); right diagrams show archetype samples. BN = bornite, CC = chalcocite, CPY = chalcopyrite, HG = high grade, PY = pyrite.

multi-element geochemistry. Chemical analyses that were selected as model parameters for the principal component analysis were Fe, Al, Si, K, Mg and Ca. The principal component analysis was carried out on 23,000 assay samples from 53 drill holes that were considered potential geometallurgical samples for the study reported on here. Each sample selected for this study consists of a contiguous 10 m zone of 1=4 drill core which was entirely from one compositional domain as defined by both the ore type classification and the domains in the principal component space shown in Fig. 1. Samples were chosen to cover the range of dominant ore types recognised (NB: as this was a geometallurgical study one of the main aims was to search for variable response from different ore types). The samples from this study are plotted in the field of the first two principal components (Fig. 1) showing the range in sample compositions tested. The gangue classification scheme was compared with the sulphide speciation classes to check for sulphide associations. This indicated, for example, that bornite–chalcocite assemblages can occur in a range of gangue lithologies including quartz-white mica dominant to hematite-siderite dominant. The different gangue mineralogies associated with the sulphides would be expected to have an influence on grinding and liberation behaviour and flotation recovery. Samples with a chalcopyrite dominant sulphide association also come from a wide range of gangue associations. Final selection of samples took place after discussions with site to ensure adequate coverage of open pit and underground areas of the mine and availability of drill core. One hundred and sixty-nine

(169) samples were selected for mineralogy and textural analysis. A subset of these (n = 24) was identified as ‘archetypes’ representing distinctive signatures in terms of the Cu and gangue classification schemes (Fig. 1) that would be expected to result in different recovery response. Batch flotation testing to determine Cu recovery was carried out on the ‘archetype’ samples.

2.2. Sample characterisation As one of the main aims of this study was to predict recovery from parameters that can routinely be measured on ore material, data was collected on all samples in several ways to test for cost effective options. High quality XRF assay data was collected for comparison with mine site assay data (mine site data was analysed via inductively coupled plasma optical emission spectrometry/ inductively coupled plasma mass spectrometry – with detection limits of 100 ppm for Si, Al, Fe & P, 50 ppm for Mg, Ca, K & Ti, 1 ppm for Ba – and modified aqua regia multi element inductively coupled plasma mass spectrometry – with detection limits of 10 ppm for Cu and 50 ppm for S) to check for any problems that may be present in the mine site assay data base in terms of analytical methods used. In addition, four primary sources of mineralogical information were obtained: SEM-based point counting (i.e. MLA-XMOD), automated optical microscopy (AOM; Berry, 2008, 2011), quantitative XRD (QXRD – e.g. Rietveld, 1969) and mineralogy calculated from assay (e.g. Berry et al., 2011).

9

J. Hunt et al. / Minerals Engineering 64 (2014) 7–14 Table 1 Correlation of XRF compositional estimates and with site data averaged over the same interval. Oxide

Comment

SiO2 Al2O3 Fe2O3 MgO CaO K2O Ba Cu S P2O5 TiO2

Mine Mine Mine Mine Mine Mine Mine Mine Mine Mine Mine

assay assay assay assay assay assay assay assay assay assay assay

R2 0.5% relative higher 0.4% relative higher 0.8% relative lower 2.4% relative lower 2.1% relative higher 1% relative higher 3.1% relative lower 1% relative lower 5.1% relative lower 1.7% realtive higher 2.2% relative higher

0.99 0.99 0.99 0.99 0.99 0.99 0.95 0.98 0.97 0.96 0.99

Examination of analytical results showed that the compositions of samples analysed by XRF were very close to those of mine site assays for the same samples (Table 1). The exception was S, where the mine site assay was on average 5% relative lower than the XRF total S analysis. NB: mine site assay data for each 1 m sample were averaged over the 10 m sample intervals to facilitate this comparison. Assessment of results from the four sources of mineralogical information showed that the calculation of mineralogy based on a combination of QXRD and assay data provided a robust, cost effective method suitable for large scale application (e.g. Berry et al., 2011). This method required the local calibration of chalcocite and bornite, i.e. the abundance of chalcocite and bornite measured by QXRD were replaced by the amount calculated from Cu/S values. Assuming the MLA-XMOD data (169 samples) are accurate this method gave a bornite estimate with a root mean square (RMS) error of 0.6% and chalcocite with a RMS error of 0.5% (this was possible for drill core samples because SEM-based mineralogical studies showed that, from the 169 samples, the 55 samples with more than 0.6% chalcopyrite had 60.1% chalcocite, and the 82 samples with >0.2% chalcocite had 60.5% chalcopyrite). The SEM-based point counting method was considered to give the best estimate of the proportion of minor and trace minerals and was

used to check the values obtained by other methods, however it is not considered suitable for widespread geometallurgical application due to the high cost of analysis. The AOM method is cost effective and worked well for distinguishing chalcopyrite, pyrite and bornite but was unable to routinely distinguish finely intergrown chalcocite from hematite and for this reason it is not considered appropriate for Prominent Hill ore. 2.3. Copper recovery A subset of 24 ‘archetype’ samples was selected from the Prominent Hill sample set (n = 169) to give an indication of copper recovery. Full kinetic batch flotation tests were carried out on these samples; a summary of results are in Table 2. Flowsheet development including the mineralogical characterisation and batch flotation tests is described in Barnes et al. (2009) and Colbert et al. (2009). A grind time to reach a P80 of 106 microns (site target size for rougher flotation) was established for each sample with 1 kg milling at 60% solids. For each test 1 kg of ore was milled with 15 g/t sodium ethyl xanthate (SEX; 1% solution) added to the mill for the specified time. The pulp was transferred to a flotation cell (bottom driven 3L Runge cell) and diluted to 30% solids with 35 g/t SEX (1% solution) added and then conditioned for one minute. The impeller speed was turned up to 700 rpm and an air rate of 7L/minute was used. Flotation was done at natural pH which was between 7.8 and 8.2. The froth height was 1 cm. Concentrates were collected after 1, 2, 5 and 10 min, weighed, dried and assayed. This work built on earlier testing as reported in Bradshaw (2011), Bradshaw et al. (2012) and Hunt et al. (2011b,c). The collector addition was not varied to accommodate the wide range in Cu grades (0.59– 10.6% Cu) as the aim of the tests was to evaluate the effects of ore variability rather than optimising processing conditions. However as the highest recovery was obtained with the ore with the highest grade, this was not considered performance limiting. Table 2 shows that there was a high variability in batch flotation recoveries for the 24 tests (59.97–97.81%) and no clear correlation between these recoveries and copper feed grades, Cu/S ratios or

Table 2 Results of batch flotation tests on ‘‘archetype’’ samples. Feed characterisation is from analysis of batch float feed. Concentrate characterisation results are shown as cumulative after 10 min. Feed characterisation

Concentrate characterisation

Sample

Fe (%)

Cu (%)

S (%)

Cu/S ratio

Mass recovery (g)

Water recovery (g)

Cu recovery (%)

Cu Grade (%)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

31.30 28.60 24.70 18.85 34.00 16.60 15.80 25.45 14.50 6.95 7.20 15.70 14.25 3.75 11.35 13.20 15.75 29.00 21.95 6.55 16.05 4.10 2.40 9.85

1.18 0.59 3.34 2.10 1.62 0.59 1.26 5.03 2.54 3.69 0.95 0.64 0.55 1.84 1.63 5.50 0.72 2.96 3.04 2.45 3.90 1.16 4.92 10.56

1.21 0.35 1.02 0.83 1.37 0.46 0.60 1.47 1.06 1.97 0.91 0.95 0.30 0.57 0.50 2.15 0.25 0.78 0.75 0.62 0.95 0.31 1.19 2.51

0.98 1.69 3.27 2.53 1.18 1.28 2.10 3.42 2.40 1.87 1.04 0.67 1.83 3.23 3.26 2.56 2.88 3.79 4.05 3.95 4.11 3.74 4.13 4.21

8.92 8.08 14.85 12.84 10.09 7.55 5.79 12.94 15.38 16.13 8.22 8.51 6.66 10.62 9.76 16.70 6.31 11.34 11.47 7.99 9.21 6.45 10.13 20.23

431 864 1102 951 245 569 657 793 1083 680 533 367 479 622

87.58 94.52 87.34 83.47 83.45 87.96 87.93 82.73 89.34 90.88 94.67 87.05 83.46 74.10 77.94 84.75 80.35 95.06 75.55 77.36 73.91 82.19 59.97 97.81

10.81 6.91 17.78 14.08 13.73 8.07 19.21 32.11 17.00 21.46 10.93 7.13 6.72 11.36 12.12 25.75 8.44 21.95 16.34 23.90 31.96 13.40 29.07 49.28

758 423 740 805 441 580 556 593 805

S by LECO, DL 0.05–30%; Cu by ICP, DL 1–10,000 ppm; Fe by ICP, DL 0.01–50%; relative analytical error ±5%.

10

J. Hunt et al. / Minerals Engineering 64 (2014) 7–14

Table 3 Data for ‘‘archetype’’ samples. Abbreviations: HMBX = hematite breccia, VOLC = volcanic, HQBX = hematite-quartz breccia, SNST = sandstone, HG = high grade, CPY = chalcopyrite, BN = bornite, CC = chalcocite. Mineralogy from QXRD (relative errors ±5%). Lithology as logged by site

Sulphide class (from Fig. 1)

Gangue class (from Fig. 1)

Group (From Fig. 2)

Avg. Cu (ppm)

Avg. Fe (ppm)

Avg. Al (ppm)

Fe Oxide (wt%)

1

HMBX

CPY HG

Lo Al–Hi Fe

11,557

417,890

17,540

56.3

2

HMBX

BN-CPY

Lo Al–Hi Fe

5974

377,200

21,360

3

HMBX

BN-CC HG

Lo Al–Hi Fe

27,740

437,880

4

HMBX

BN-CPY HG

Hi Al–Lo Fe

20,930

5

HMBX

CPY HG

Lo Al–Hi Fe

6 7

VOLC HMBX

CPY BN-CC HG

8

HMBX

CC HG

9

HMBX

BN-CPY HG

10

HMBX

BN-CPY HG

11

HMBX+HQBX

CPY HG

12

HMBX+VOLC

CPY HG

13 14 15 16

VOLC SNST HMBX HMBX

BN-CPY BN-CPY HG BN-CC HG CC HG

17 18

VOLC HMBX

BN-CC CC HG

19

VOLC

BN-CC HG

20 21

HMBX HMBX

CC HG CC HG

22 23 24

HMBX VOLC HMBX

BN-CC HG CC CC HG

Hematite breccia Hematite breccia Hematite breccia Hematite alteration Hematite breccia Igneous Hematite breccia Hematite breccia Hematite breccia Hematite breccia Hematite quartz Hematite alteration Igneous clastic igneous Hematite breccia Igneous Hematite quartz Hematite alteration Igneous Hematite alteration Igneous clastic Hematite breccia

Sample

K Feldspar (wt%)

Kaolinite (wt%)

7.4

1.2

0.9

52.4

10.6

1.2

1.8

23,570

59.8

10.0

0.7

3.1

261,170

57,480

33.9

29.3

0.4

4.6

15,340

454,150

17,310

37.7

7.6

0.2

1.5

Hi Al–Lo Fe Lo Al–Hi Fe

4823 11,776

153,450 326,170

67,930 30,330

12.8 44.1

38.2 17.4

0.1 0.3

3.0 0.8

Lo Al–Hi Fe

49,340

422,270

26,520

58.9

13.3

0.0

2.6

Lo Al–Hi Fe

26,225

386,290

25,000

46.6

13.0

0.5

1.5

Lo Al–Hi Fe

37,740

437,940

20,890

60.2

9.9

0.5

1.7

Lo Al–Hi Fe

9737

429,700

6280

59.5

2.5

0.2

0.4

Hi Al–Lo Fe

6183

204,600

56,790

16.0

32.7

0.1

2.3

Fe Fe Fe Fe

4823 16,078 13,138 52,280

147,930 52,578 142,050 383,870

67,930 88,344 74,563 24,550

5.1 4.9 18.8 53.4

41.8 43.7 31.6 11.7

0.0 1.8 3.2 1.3

2.1 4.9 6.1 2.0

Hi Al–Lo Fe Lo Al–Hi Fe

6404 26,070

163,590 446,970

66,960 6318

1.9 63.0

41.0 1.9

0.0 0.0

1.3 0.8

Hi Al–Lo Fe

24,147

236,310

56,100

20.9

32.8

0.0

1.8

Hi Al–Lo Fe Hi Al–Lo Fe

23,020 37,895

149,870 217,330

74,380 56,360

19.1 28.2

37.4 24.4

2.2 0.1

4.7 7.8

Hi Al–Lo Fe Hi Al–Lo Fe Lo Al–Hi Fe

44,590 10,743 92,105

62,240 66,920 305,560

88,080 72,560 22,202

6.0 7.1 41.6

49.4 37.8 10.3

0.8 0.2 0.4

4.0 6.0 3.2

Hi Hi Hi Lo

Al–Lo Al–Lo Al–Lo Al–Hi

predominant copper mineral present. The variability was higher with ores containing chalcocite than those with chalcopyrite and the mass versus water recovery relationships indicate that the variability is driven by variability in froth stability and that higher froth stability results in a higher copper recovery. The mean Cu recovery obtained (83%) was equivalent to that obtained in the site feasibility testwork (Triffett, 2011). 3. Copper recovery modelling and prediction The batch flotation results from the archetype samples were used to develop models for recovery of copper for Prominent Hill samples. The simplest of these models involves assay data only (Cu, Fe, Al) and because of this predicted recovery values can be

White mica (wt%)

calculated for all samples in the mine site assay data base. This facilitates the ranking of samples and division of the deposit into potential recovery domains. 3.1. Selection of model parameters Sample characterisation data were compared to Cu recovery data to identify parameters that could be used in model development. Copper recovery as determined by the 10 min batch flotation test was chosen as it is considered to better represent copper recovery that will be achieved in the mine flotation plant. The approach in our research (i.e. GeMIII project – AMIRA P843 and P843A) is to look for rapid, inexpensive ways to obtain necessary data, so although copper recovery correlated well with oxides

Table 4 List of models for Cu recovery determined for Prominent Hill archetype samples (see text for details). Fe and Al in ppm. RMS errors calculated by 12-fold cross validation. Group

Cu recovery

Avg. relative error (%)

RMS error (%)

Equation #

All 24 samples Low Al–High Fe High Al–Low Fe High Al–Low Fe

=( =( =( =(

±4.7 ±2.6 ±3.6 ±2.2

±6.6 ±3.3 ±4.5 ±3.4

1 2 3 4

9.72E 6.54E 4.88E 4.84E

5  Fe) + ( 7.17E 4  Al) + 143.58 4  Al) + ( 9.53E 5  Fe) + 140.68 4  Cu) + ( 2.47E 4  Al) + 105.08 4  Cu) + (2.04E 4  Al) + ( 89.84  GRD.INDX) + 143.75

J. Hunt et al. / Minerals Engineering 64 (2014) 7–14

11

and mineral values as well as assay elements (e.g. Fe: R = 0.6, Al: R = 0.7, K: R = 0.6, Fe oxide: R = 0.6, white mica: R = 0.6, K-feldspar: R = 0.7, kaolinite: R = 0.5) we decided to first test the use of assay elements only in our subsequent modelling. If this was unsuccessful we would then try including parameters from more expensive testing (e.g. mineralogy from QXRD). This was possible at Prominent Hill because site routinely analyses samples for a wide range of elements, thus, these data were readily available for a large part of the Prominent Hill mine. 3.2. Deriving copper recovery models

Fig. 2. Archetype samples plotted on an Al versus Fe diagram.

Models for copper recovery for Prominent Hill were developed using data for the 24 archetype samples. These data included site assay results for drill core (i.e. average values over 10 m derived from individual 1 m assay samples) and 10 min batch flotation results for Cu recovery. In drill core logs completed by site the samples were logged as dominantly hematite breccia with lesser amounts of hematite-quartz breccia, volcanic rock and sandstone

Fig. 3. Comparison of measured and predicted values for copper recovery for archetype samples. Top left = all samples modelled as one group (Eq. (1) in Table 4). Top right = low Al–high Fe samples modelled as a separate group (Eq. (2) in Table 4). Middle left = high Al–low Fe group samples modelled as a separate group (Eq. (3) in Table 4). Middle right = high Al–low Fe samples modelled as a separate group with a grinding term included in the model (Eq. (4) in Table 4). Bottom left = both groups modelled separately using assay data only. Bottom right = both groups modelled separately using assay data plus a grinding term included in the model for the high Al–low Fe group.

12

J. Hunt et al. / Minerals Engineering 64 (2014) 7–14

Fig. 4. Prominent Hill samples (n = 169) plotted on an Al versus Fe diagram using categories as in Fig. 2. Large symbols = archetype samples.

(Table 3). The archetype samples separate into two groups when plotted on an Al versus Fe plot (Fig. 2). This division is likely a reflection of the mineralogy: samples in the low Al–high Fe group contain more iron oxide (avg. 55 wt% versus 16 wt%) and less white mica (avg. 10 wt% versus 37 wt%) than samples in the high Al–low Fe group. The low Al–high Fe samples are logged as hematite breccia or hematite-quartz breccia in mine site logs and in the gangue discriminant diagram shown in Fig. 1. Logged lithologies for the high Al–low Fe group are hematite breccia, volcanic rock and sandstone (these correspond to hematite alteration, igneous and clastic respectively in Fig. 1 for these samples). Modelling was carried out by treating all of the archetype samples as one group and then as individual groups to determine if dividing the samples would significantly improve the model results. The best result obtained when all samples were treated as one group and only assay data were used as input gave predicted Cu recovery values with a RMS error of ±6.6% (Table 4). When the samples were divided into groups as shown in Fig. 2 the best results for the low Al–high Fe group and the high Al–low Fe group were RMS errors of ±3.3% and ±4.5% Cu recovery respectively. Fig. 3 shows a comparison of measured and predicted copper recovery. If the archetype samples are modelled as one group then the correlation between measured and predicted values is moderate (R2 = 0.66). Separating the samples into two groups and modelling them individually improves the correlation, especially for the high Al–low Fe group (i.e. low Al–high Fe group R2 = 0.69; high Al–low Fe group R2 = 0.79). This improvement gives an overall R2 = 0.85. Further improvements to the model for the high Al–low Fe group were possible by including a ‘‘grinding term’’ in the model (Eq. (4) in Table 4). This term is the ‘‘grinding index’’ (GRD.INDX) derived from the GeM Comminution Index test (GeMCi). The GeMCi test has been designed to be inserted into routine assay sample preparation and is based on constrained jaw crushing protocols linked to analysis of resultant size distributions (Kojovic et al., 2010). Adding this term to the model improved the correlation with measured values to R2 = 0.94 for the high Al–low Fe group, and the overall R2 for all samples to 0.92 (Fig. 3).

3.3. Modelling copper recovery Data for all samples in the GeM sample set (n = 169) were plotted on an Al versus Fe plot (Fig. 4) with fields as shown in Fig. 2. In general, the GeM samples plot within the categories determined using the archetype samples suggesting that for most samples

Fig. 5. Sample density plot of Prominent Hill samples above Cu cut off (>2000 ppm Cu) from 53 drill holes on an Al versus Fe diagram using categories as in Fig. 2. Hotter colours = higher sample density. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 6. Example showing a portion of a drill hole with copper recovery domains identified.

potential copper recovery could be estimated using the models determined for the low Al–high Fe and high Al–low Fe categories. If the samples from the 53 drill hole sample set (see sample selection) are plotted on a similar Al versus Fe plot most of those samples also plot in the categories determined using the archetype samples (Fig. 5). However there are clearly some samples that plot outside the low Al–high Fe and high Al–low Fe categories in Figs. 4 and 5. Some of these samples are logged as hematite breccia in drill core logs and plot as HT BX in the current gangue classification diagram (Fig. 1) but on average contain significantly more Ca (9 wt%) compared to other hematite breccia samples (1 wt% Ca). Other samples that plot outside the low Al–high Fe and high Al–low Fe categories are those with gangue class ‘‘other’’ or ‘‘quartz-hematite’’. High Ca samples and those of gangue classes ‘‘other’’ or ‘‘quartz-hematite’’ were not identified as ore types of interest in our initial evaluation and no archetype samples of this material were tested. In our research ore type classification and predictive modelling are viewed as iterative processes that should be modified to incorporate additional ore types, e.g. ore types that have not been previously recognised or those from new areas of mining. In the case here, to develop predictive copper models for samples that fall outside the defined categories, the gangue classification diagram

J. Hunt et al. / Minerals Engineering 64 (2014) 7–14

should be modified, or an additional diagram developed, so that these samples are identifiable, suitable archetype samples should be identified, batch flotation testing completed and modelling carried out to allow prediction of Cu recovery for these ore types. Also of interest, for the low Al–high Fe group, the model for estimating copper recovery involves Fe and Al only. Adding Cu does not improve the model, indicating that copper recovery for this group of samples is strongly controlled by gangue mineralogy. This suggests that a definition of ore type that relies strongly on copper grade and/or copper mineralogy may not be the most effective method to use for ore of this type. 3.4. Copper recovery domain definition Estimated copper recovery values were determined for samples in the site assay data base that plotted in the low Al–high Fe and high Al–low Fe groups using the ‘‘assay data’’ model. The predicted copper recovery values were calculated on a drill hole by drill hole basis to assess spatial continuity using the ‘‘Domain Definition Tool’’ (Keeney, 2008; Keeney and Nguyen, 2010, Keeney and Walters, 2011; Walters, 2011) which applies the standard time series analytical method of cumulative sums (e.g. Subba Rao, 1981). Fig. 6 shows an example of a drill hole where three domains have been identified with mean predicted copper recovery values of 89%, 86% and 96%. Once recovery domains have been determined for each drill hole the results can be visualised in 3D software packages. This will enable recovery domains to be visualised for the entire deposit and make them available for use in mine planning and optimisation. 4. Summary and conclusions A sample set that covered a wide compositional range was selected from Prominent Hill drill core using robust and systematic rules for representivity of dominant ‘‘ore types’’. Chemical and mineralogical information (assay, XRF, QXRF, MLA-XMOD, calculated mineralogy) was obtained for all the samples. Copper recovery was measured by batch flotation on a subset (24) of these samples selected to represent this range of compositions. Several models to predict potential copper recovery were developed. Samples were divided into two groups: high Fe–low Al and low Fe–high Al. In general, the two groups reflect the mineralogy of the samples (i.e. Fe-oxide rich or white-mica rich). The most successful models were developed by treating these two groups separately. The simplest of these models involves assay data only (Fe, Al, Cu) and gives predicted results with ±4% RMS error when compared to the results of batch flotation tests. Several factors enabled the recovery model to be propagated into the mine site data base: (1) only assay data are needed as input to the model, (2) a wide suite of assay elements was readily available at the Prominent Hill mine site as these are part of the routine assay procedure on site, and (3) a large proportion of samples in the site data base fall within the two categories identified in the calibration data (Fig. 5). The predicted Cu recovery was calculated for samples in the site data base and used to define recovery domains for each drill hole. This, in turn, allows 3D modelling and visualisation of the potential recovery domains. It was expected that the copper mineral speciation would influence the copper recovery as this was the basis of the original process and flowsheet design but this was shown not to be the case and rather the gangue mineralogy, was a stronger predictor of copper recovery. In summary, a rapid, cost effective method to estimate potential copper recovery and outline recovery domains was derived to represent much of the Prominent Hill site data set. The process is

13

intended to be an iterative one and it is recommended that samples that did not fall into the defined categories be assessed by the identification of additional archetype samples, further batch flotation testing and modelling. Acknowledgements This research is part of a major collaborative geometallurgical project being undertaken at CODES and SES (University of Tasmania), JKMRC, BRC and CMLR (Sustainable Minerals Institute, University of Queensland) and Parker Centre CRC (CSIRO). The authors acknowledges financial support and permission to publish from industry sponsors of the AMIRA International P843 and P843A GEMIII projects – Anglo Gold Ashanti, Anglo American, ALS, Barrick, BHP Billiton, Boliden, CAE Mining (Datamine), Codelco, Geotek, Gold Fields, Golder Associates, ioGlobal, Metso Minerals, Minera San Cristobal, Newcrest, Newmont, OZ Minerals, Peñoles, Quantitative Geoscience, Rio Tinto, Teck, Vale and Xstrata Copper (MIM). Financial support is also being provided by the Australian government through the CODES ARC Centre of Excellence in Ore Deposits and CRC ORE. The analytical facilities at the Central Science Laboratory, University of Tasmania, were used for MLA measurements in this project. QXRD results were from McKnight Mineralogy. Special thanks are offered to staff of Prominent Hill mine for their help with this project. References Barnes, K.E., Colbert, P.J., Munro, P.D., Designing the optimal flotation circuit – the Prominent Hill case. In: Tenth Mill Operators Conference, Adelaide, 12–14 October 2009. Belperio, A., Flint, R., Freeman, H., 2007. Prominent Hill: a hematite-dominated, iron oxide copper–gold system. Econ. Geol. 102, 1499–1510. Berry, R., Automated mineral identification by optical microscopy. AMIRA P843 GEMIII Internal Report (Technical, Report 1), 2008, 7.1-7.12. Berry, R., Prominent Hill case study – ore characterisation. AMIRA P843A GEMIII Internal Report, 2011, 39p. Berry, R., Hunt, J., McKnight, S., Estimating mineralogy in bulk samples. In: Proceedings The First AusIMM International Geometallurgy Conference. AusIMM, Australia, 2011, pp. 153–156. Bradshaw, D., Prominent Hill case study – batch flotation and small scale mineral separability testing. AMIRA P843A GEMIII Internal Report, 2011, 24p. Bradshaw, D., Wightman, E., Evans, C., Triffett, B., Characterising the reasons for the difference in performance of two ore blends in early processing at OZ Minerals Prominent Hill operation. In: Proceedings Mill Operators Conference, AusIMM, Australia, 2012, pp. 20–25. Colbert, P.J., Munro, P.D., Yeowart, G., Prominent Hill Concentrator – Designed for Operators and Maintainers Tenth Mill Operators Conference, Adelaide, 12–14 October, 2009, pp. 23–31. Hunt, J., Berry, R., Bradshaw, D., Prominent Hill case study – liberation and recovery modelling and prediction. AMIRA P843A GEMIII Internal Report 2011, 21p. Hunt, J., Berry, R., Bradshaw, D., 2011b. Characterising chalcopyrite liberation and flotation potential: examples from an IOCG deposit. Miner. Eng. 24, 1271–1276. Hunt, J., Berry, R., Bradshaw, D., Characterising liberation and flotation potential using image analysis, simulated fragmentation and small-scale flotation. In: Proceedings The First AusIMM International Geometallurgy Conference. AusIMM, Australia, 2011, pp. 331–334. Keeney, L., Stochastic trend analysis: theory and procedures. AMIRA P843A GEMIII Internal Report (Technical, Report 2) 2008, 15(1–15), pp. 14. Keeney, L., Nguyen, K., Domain analysis and interpretation (DomAIn). AMIRA P843A GEMIIIInternal Report (Technical, Report 7), 2010, pp. 40–42. Keeney, L., Walters, S., A methodology for geometallurgical mapping and orebody modelling. In: Proceedings The first AusIMM International Geometallurgy Conference. AusIMM, Australia, 2011, pp. 217–225. Kojovic, T., Michaux, S., Walters, S., Development of new comminution testing methodologies for geometallurgical mapping of ore hardness and throughput. In: Proceedings International Mineral Processing Congress – IMPC 2010. AusIMM, Australia, 2010, pp. 891–899. OZ Minerals web page, Prominent Hill. , 2012. Oz Minerals. ASX release: Prominent Hill reserves and resources and production outlook. , 2013.

14

J. Hunt et al. / Minerals Engineering 64 (2014) 7–14

Rietveld, H.M., 1969. A profile refinement method for nuclear and magnetic structures. J. Appl. Crystallogr. 2, 65–71. Subba Rao, T., 1981. A cumulative sum test for detecting change in time series. Int. J. Control 34, 285–293. Triffett, B., Personal communication, 2011.

Walters, S., Integrated industry relevant research initiatives to support geometallurgical mapping and modelling. In: Proceedings The First AusIMM International Geometallurgy Conference. AusIMM, Australia, 2011, pp. 273–278. Walters, S., Hunt, J., Prominent Hill case study – sample selection for liberation and recovery modelling. AMIRA P843A GEMIII Internal Report, 2011, 14p.