- Email: [email protected]
The PGM flotation predictor: Predicting PGM ore flotation performance using results from automated mineralogy systems
Minerals Engineering 36–38 (2012) 75–80
Contents lists available at SciVerse ScienceDirect
Minerals Engineering journal homepage: www.elsevier.com/locate/mineng
The PGM flotation predictor: Predicting PGM ore flotation performance using results from automated mineralogy systems Charles Bushell ⇑ Mineralogy Division, Mintek, Private Bag X3015, Randburg 2125, South Africa
a r t i c l e
i n f o
Article history: Available online 28 March 2012 Keywords: Precious metal ores Froth flotation Flotation modelling Process optimisation Flotation kinetics
a b s t r a c t Performance of froth flotation recovery plants for platinum group minerals (PGMs) is usually monitored by means of routine chemical assays of samples taken at various locations in the plant. Whilst these assays can alert the plant metallurgist to variations in recovery, the reasons for changes in recovery are not adequately revealed by the assay results. Assay-by-size analyses can help to diagnose whether PGM and/or base metal sulphide (BMS) liberation issues exist, but do not provide any information on mineralogical changes in the plant feed material. The flotation performance of an ore is determined by its mineralogy. Mintek’s Mineralogy Division is currently developing PGM flotation prediction software that uses data from automated mineralogy systems to provide valuable information to the plant metallurgist. Each PGM-bearing particle detected by the automated mineralogy system is individually evaluated. Particle floatability, based on the mode of occurrence of the PGM, the proportion of floatable component/s and the composition of constituent minerals in each PGM-bearing particle is calculated. These data provide a direct output that highlights the metallurgical properties and recoverability of the PGM-bearing particles in samples gathered from strategic locations in the recovery plant. Ó 2012 Elsevier Ltd. All rights reserved.
1. Introduction Automated mineralogy systems generally use a combination of scanning electron microscope (SEM), backscattered-electron (BSE) images, image analysis, and energy dispersive spectrometry (EDS), to provide very useful data that cannot be obtained from other analytical techniques, particularly in complex ores. Common results outputs include relative abundance of minerals (modal analysis), liberation characteristics of valuable minerals, mineral grain and particle size, and mineral association data. These results can help to pinpoint mineralogical changes in an ore that can often affect mineral recoveries. If these mineralogical changes can be quantified with a reasonable degree of confidence, then the effect of these changes can be compensated for in the recovery plant. The terms ‘‘mineral grain’’ and ‘‘mineral particle’’ are used frequently in this paper. It is important to distinguish between the two. A mineral grain is a homogenous unit of pure mineral. A mineral particle is made up of one or more mineral grains. In the case of a pure, liberated mineral, the terms ‘‘grain’’ and ‘‘particle’’ are equivalent. In ores that contain platinum group minerals (PGMs), the plant feed material generally contains less than 10 ppm of platinum ⇑ Tel.: +27 11 709 4528; fax: +27 11 709 4564. E-mail address: [email protected] 0892-6875/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.mineng.2012.02.016
group elements (PGEs). The low PGE grade makes it virtually impossible to provide statistically meaningful PGM data by using traditional manual techniques. Automated SEM/EDS systems help to alleviate this problem by searching several polished sections per sample in unattended runs. Potential PGM-bearing particles are located by means of the high BSE intensity produced due to the high average atomic number of PGE-bearing minerals, and are identified by means of automated EDS analyses, performed on the constituent mineral grains in the PGM-bearing particles. Analysis results are saved to a database during the automated runs. Results from automated mineralogy systems are typically presented as tables or charts that summarise particular sample characteristics, and represent the total population or a specific sub-set of analysed mineral particles in the sample. To reliably determine floatability from mineralogical data, however, each analysed mineral particle needs to be individually evaluated. The reason for this is that more than one particle characteristic often needs to be considered to determine floatability. For example, a liberated PGM grain would be expected to float, but might not if its grain size is either very small or very large. In the case of composite particles, floatability depends on the minerals present, the grain size, association and mode of occurrence of each of these minerals, and the total particle size (Chetty et al., 2009). These individual particle data are written to the results database during analysis, but are usually not resolved in the pre-defined tables
76
C. Bushell / Minerals Engineering 36–38 (2012) 75–80
and charts produced by the system’s data processing software. The ‘‘grains table’’ within the automated SEM results database provides a source of individual PGM-bearing particle data required for interrogation by the flotation predictor. Similar approaches to predicting floatability and recovery of ores using mineral particle properties determined by automated SEM systems have been documented (Evans et al., 2011; Evans, 2010; Ford et al., 2007; Hunt et al., 2008; Lotter et al., 2003; Wightman et al., 2010, 2008). These, however, apply largely to base metal sulphide recovery, and do not involve the development of custom designed software to interrogate the mineralogical data. The methodology discussed in this paper is a simple approach, which has been developed to cater specifically for South African Bushveld PGM ores with typical feed grades of less than 10 g/t. The concept has been implemented as a VBA module within Microsoft Excel, which can accept PGM mineralogical data from any of the current automated SEM platforms.
2. Producing an input file for the flotation predictor The ‘‘grains table’’ produced by the automated SEM system is extracted from the sample results database, and is used to determine individual PGM grain and PGM-bearing particle characteristics. The first step is to determine the mode of occurrence of the PGM in each PGM-bearing particle. This is achieved by means of a Microsoft Excel spreadsheet, using a visual basic for applications (VBA) macro. The macro automatically classifies each PGM-bearing particle into one of six pre-defined mode of occurrence classes, described in Table 1 and illustrated in Fig. 1. Mineral ID and grain measurement data are also captured from the grains table. Grain areas are used to calculate a liberation index for each PGM-bearing particle. This measure of PGM grain liberation is calculated by dividing the area of potentially floatable component (PGM + BMS) by the total area of the particle (PGM + BMS + gangue). The resultant figure will range between 0 and 1, the latter indicating either a liberated PGM grain, or a binary particle containing PGM and BMS only. In contrast, the liberation index of a PGM grain totally enclosed within a BMS-barren silicate particle (low probability of flotation) will approach zero. Mineral ID, particle liberation index, grain and particle size, and PGM mode of occurrence for each PGM-bearing particle are used to produce a file that contains all of the necessary particle data to be interrogated by the flotation predictor software.
3. Flotation predictor operation and data flow The PGM flotation predictor predicts PGM recovery by producing an output based on the physical properties of all PGM-bearing particles detected in the sample. These properties include mode of occurrence of the PGM (as per descriptions in Table 1), the floatability of minerals associated with the PGM, liberation index (i.e. proportion of floatable component in the particle), and particle/ grain size. By classifying PGM-bearing particles into five classes that contain particles expected to float within particular time
Fig. 1. PGM-bearing particle mode of occurrence classes.
intervals, a recovery–time profile is produced. If desired, a flotation model can then be applied to this profile to determine flotation kinetics for the PGM bearing particles. The current version of the flotation predictor estimates fast and slow floating PGM fraction percentages and flotation rate constants by fitting the predicted recovery profile to the Kelsall flotation model (Kelsall, 1961) using a non-linear regression procedure. A simplified flow chart of the flotation predictor operation is provided in Fig. 2. The flotation predictor user interface, illustrated in Fig. 3, allows the user to set various flotation parameters prior to calculation. These include minimum and maximum particle sizes, choice of gangue minerals considered to be hydrophobic under the flotation conditions used, and liberation index cut-off values for AG and SAG class composite particles. SAG class liberation index cut-off values vary for the different BMS species, as the different BMS species have different floatability characteristics (Wiese et al., 2007; Penberthy et al., 2000). Initial ‘‘default’’ flotation parameter values are provided when the flotation predictor is run, and can be easily restored at any stage. These values are based on particle characteristics observed in samples generated by laboratory scale test-work that has been performed on various South African PGM ores. The parameters can be adjusted to suit a particular recovery plant, or circuit within that plant, according to mineralogical properties of PGM-bearing particles observed in representative samples of feed, concentrate and tailings gathered from the plant or circuit in question. The flotation predictor can thus produce results similar to those obtained from laboratory-scale rate tests from a full-scale operating plant. Once the plant feed has been properly characterised, the predictor output can be used as a diagnostic tool for routine plant monitoring and troubleshooting. Recovery via
Table 1 PGM-bearing particle mode of occurrence classes. PGM mode of occurrence class
Description
L SL AG SAG SG G
Liberated PGM PGM associated with BMS only (i.e. a binary PGM–BMS particle) PGM attached to silicate or oxide gangue (i.e. PGM exposed at particle perimeter) PGM associated with BMS attached to silicate or oxide gangue (i.e. BMS exposed at particle perimeter) PGM associated with BMS locked within silicate or oxide gangue (i.e. no exposure of BMS or PGM at particle perimeter) PGM locked within silicate or oxide gangue (i.e. no exposure of PGM at particle perimeter)
C. Bushell / Minerals Engineering 36–38 (2012) 75–80
77
Fig. 2. Flotation predictor flow diagram.
entrainment is not taken into account, as recovery via non-selective mechanisms cannot be predicted based on particle liberation properties. When evaluating concentrate samples, however, the presence of any SG or G class (i.e. non-floatable) particles, or PGM-bearing particles with very low liberation indices is considered to be due to entrainment, provided that the gangue component of the particle is not hydrophobic.
4. Flotation predictor output The flotation predictor produces a direct output from mineralogical data, highlighting the metallurgical properties of PGM-bearing particles in the sample in tabular and graphical format. The results file includes quantitative information on the potential recovery, the reasons for recovery or losses, particle size distribu-
78
C. Bushell / Minerals Engineering 36–38 (2012) 75–80
Fig. 3. Flotation predictor user interface. Note: numerical parameters are shown as examples, and do not necessarily represent realistic values.
tion, grouping of valuable species, and estimation of flotation kinetics. The output varies according to the sample type. For example, feed sample output focuses more on potential recovery and estimation of flotation kinetics, whilst tailing sample output focuses more on reasons for losses. Typical flotation predictor graphical outputs are illustrated in Figs. 4–8. These include: Potential PGM recovery, expressed as both volume and number percentage of the total PGM population assessed (Fig. 4). A predicted PGM flotation rate curve, based on the particle characteristics of the recoverable PGM particles detected (Fig. 5). Reasons for PGM losses (Fig. 6), categorised into the four major causes of losses to tailings: 1. PGM locked in gangue (i.e. no exposed floatable component). 2. PGM-bearing particles with low liberation indices (i.e. a particle with a very low proportion of floatable component).
Fig. 4. Potential PGM recovery (as % of PGM population assessed).
3. Particles too large to float. 4. Particles too small to float. PGM-bearing particle size distribution (Fig. 7), useful for comparison with assay-by-size results – favourable comparison validates the auto-SEM results. Grouping of similar PGM species (Fig. 8). As the dataset of PGMbearing particles obtained by the auto-SEM is small due to the low PGE grade of typical samples, grouping can help to achieve a mineral balance across the recovery circuit. The metallurgical significance of mineralogical results from automated mineralogy systems can often be difficult to interpret.
Fig. 5. Predicted PGM time–recovery profile for PGM feed sample.
C. Bushell / Minerals Engineering 36–38 (2012) 75–80
Fig. 6. Reasons for PGM losses.
Fig. 7. PGM distribution by particle size (for comparison with assay-by-size results).
predictor software is run. This was achieved by performing mineralogical investigations on concentrates collected at 1, 3, 7, 20 and 40 min during the flotation rate tests on each ore. The physical characteristics of PGM-bearing particles detected in each of the concentrate samples were noted, and these characteristics were used to establish the cut-off values for PGM size, mode of occurrence and liberation indices for particles that would be expected to report to individual concentrates within the five abovementioned time limits. Comparisons of flotation rate tests and flotation predictor rate estimation results on three different ores are presented in Fig. 9. The flotation response of 2 reasonably unaltered ores (A & B), and one highly altered ore, (C) is shown. Recovery profiles in these results were produced by using the same flotation prediction parameters, based on collective particle characteristics observed in all three ores. The results clearly show that the parameters used were far better suited to unaltered ores. By adjusting the liberation index threshold parameters to match particle characteristics observed in the concentrates produced by rate tests on ore C only, a much improved recovery–time profile for ore C, shown in Fig. 10, was produced. This shows the importance of matching the predictor parameters to the ore’s mineralogical characteristics. Three sets of initial default parameters (for unaltered, partially altered and highly altered ores) are planned for inclusion in the next version of the flotation predictor software, but these only provide a guideline towards the optimal values of the various parameters. Ideally, a specific set of flotation predictor parameters should be established for a particular ore by first performing a thorough PGM mineralogical analysis of feed, concentrate and tailing samples from the recovery plant. Flotation predictor parameters can then be derived from the PGM-bearing particle characteristics observed in the various recovery plant streams. The output produced by the optimised parameters defines a baseline mineralogical profile of the ore, against which future routine mineralogical analyses can be compared. This allows the flotation predictor output to be used as a diagnostic tool, providing a way to monitor the plant, and to establish whether any change in plant performance is due to changes in feed mineralogy. As long as plant reagent addition remains reasonably constant, the effect of reagents should not affect results negatively, as the baseline data will have been established with reagents present.
Fig. 8. Grouped PGM species.
The link between ore mineralogy and plant performance is not always clearly revealed by traditional liberation, particle size and mineral association results from these systems. The aim of the flotation predictor is to interpret the mineralogical results in a repeatable manner to produce meaningful metallurgical data in an easy to understand format. Although this has only been applied to flotation of South African PGM ores at this stage, the same principles should be equally applicable to other orebodies and commodities. 5. Comparison of flotation predictor results with laboratory scale flotation rate tests Laboratory scale rougher flotation rate tests on several PGM ores were used to establish the initial ‘‘default’’ flotation predictor parameters which appear in the user interface when the flotation
79
Fig. 9. Comparison of rate tests and flotation predictor output, ores A–C.
80
C. Bushell / Minerals Engineering 36–38 (2012) 75–80
a mineralogical profile for the plant. Future PGM mineralogy studies can be compared against this mineralogical profile to establish whether changes in feed mineralogy are responsible for variations in recovery. Where feed mineralogy remains constant, but concentrates and tailings differ from the baseline data, plant issues other than mineralogical changes are indicated. Maximum value can be obtained from mineralogical data by using the flotation predictor output in conjunction with other mineralogical information, such as relative mineral proportions (modal analysis) and mineral associations. Acknowledgements The author would like to thank Mr. Nigel Ramlall of Mintek’s Minerals Processing Division for performing the laboratory scale rate tests and for providing insight on various aspects of PGM flotation, as well as colleagues in Mintek’s Mineralogy Division for their support and encouragement during the developmental stage of the flotation predictor.
Fig. 10. Comparison of rate test and flotation predictor output for altered ore, after adjusting liberation index threshold parameters to suit the ore characteristics.
6. Future plans The current version of the flotation predictor (v 2.0, June 2011) operates as a VBA module within Microsoft Excel. Although this works adequately, VBA format is not considered a viable long-term development platform. The software will be upgraded to a standalone executable format in future. Capabilities will also be expanded to allow flotation prediction of commodities other than PGM. 7. Conclusions The flotation predictor produces results that show good correlation with results obtained from laboratory scale flotation rate tests. The predictor is not intended to replace rate tests, as these can be performed quicker and cheaper than automated SEM mineralogical analyses. The predictor can, however, produce flotation rate estimates directly from a full-scale operating flotation plant, and also links flotation performance to ore mineralogy, which is not achieved by traditional rate tests. This provides a means to monitor and troubleshoot plant performance based on ore mineralogy. Although the pre-set default flotation parameters produce useable results, prediction can be improved by establishing parameters through a detailed mineralogical study of the ore and plant products generated from it. Once these baseline data have been established, the flotation prediction parameters essentially define
References Chetty, D., Gryffenberg, L., Lekgetho, T.B., Molebale, I.J., 2009. Automated SEM study of PGM distribution across a UG2 flotation bank: implications for understanding PGM floatability. The Journal of the Southern African Institute of Mining and Metallurgy 109, 587–593. Evans, C.L., 2010. Development of a methodology to estimate flotation separability from ore microtexture. PhD thesis, University of Queensland. Evans, C.L., Wightman, E.M., Manlapig, E.V., Coulter, B.L., 2011. Application of process mineralogy as a tool in sustainable processing. Minerals Engineering 24, 1242–1248. Ford, F., Lee, A., Davis, C., Xu, M., Lawson, V., 2007. Predicting Clarabelle mill recoveries using Mineral Liberation Analyser (MLA) grade-recovery curves. In: Proceedings of the 48th Annual Conference of Metallurgists of CIM, Sudbury, Ontario, Canada, pp. 32–41. Hunt, J.A., Berry, R., Walters, S., Bonnici, N., Kamenetsky, M., Nguyen, K., Evans, C.L., 2008. A new look at mineral maps and the potential relationships of extracted data to mineral processing behaviours. In: Ninth International Congress for Applied Mineralogy. ICAM Australia 2008, Brisbane, QLD, 8–10 September, 2008, pp. 429–432. Kelsall, D.F., 1961. Application of probability in the assessment of flotation systems. Transactions of the Institution of Mining and Metallurgy 70, 191–204. Lotter, N.O., Kowal, D.L., Tuzun, M.A., Whittaker, P.J., Kormos, L., 2003. Sampling and flotation testing of Sudbury Basin drill core for process mineralogy modelling. Minerals Engineering 16 (9), 857–864. Penberthy, C.J., Oosthuyzen, E.J., Merkle, R.K.W., 2000. The recovery of platinum group elements from the UG-2 chromitite, Bushveld complex – a mineralogical perspective. Mineralogy and Petrology 68, 213–222. Wiese, J.G., Becker, M., Bradshaw, D.J., Harris, P.J., 2007. Interpreting the role of reagents in the flotation of platinum-bearing Merensky ores. The Journal of the Southern African Institute of Mining and Metallurgy 107, 29–36. Wightman, E., Evans, C.L., Vizcarra, T., and Sandoval, G. 2008. Process mineralogy as a tool in modelling mineral processing operations. In: Ninth International Congress for Applied Mineralogy. ICAM Australia 2008, Brisbane, QLD, 8–10 September, 2008, pp. 475–481. Wightman, E., Evans, C., Bradshaw, D., Manlapig, E.V., 2010. Predicting flotation response from liberation information. In: Proceedings XXV International Mineral Processing Congress, Brisbane, pp. 3403–3409.
Contents lists available at SciVerse ScienceDirect
Minerals Engineering journal homepage: www.elsevier.com/locate/mineng
The PGM flotation predictor: Predicting PGM ore flotation performance using results from automated mineralogy systems Charles Bushell ⇑ Mineralogy Division, Mintek, Private Bag X3015, Randburg 2125, South Africa
a r t i c l e
i n f o
Article history: Available online 28 March 2012 Keywords: Precious metal ores Froth flotation Flotation modelling Process optimisation Flotation kinetics
a b s t r a c t Performance of froth flotation recovery plants for platinum group minerals (PGMs) is usually monitored by means of routine chemical assays of samples taken at various locations in the plant. Whilst these assays can alert the plant metallurgist to variations in recovery, the reasons for changes in recovery are not adequately revealed by the assay results. Assay-by-size analyses can help to diagnose whether PGM and/or base metal sulphide (BMS) liberation issues exist, but do not provide any information on mineralogical changes in the plant feed material. The flotation performance of an ore is determined by its mineralogy. Mintek’s Mineralogy Division is currently developing PGM flotation prediction software that uses data from automated mineralogy systems to provide valuable information to the plant metallurgist. Each PGM-bearing particle detected by the automated mineralogy system is individually evaluated. Particle floatability, based on the mode of occurrence of the PGM, the proportion of floatable component/s and the composition of constituent minerals in each PGM-bearing particle is calculated. These data provide a direct output that highlights the metallurgical properties and recoverability of the PGM-bearing particles in samples gathered from strategic locations in the recovery plant. Ó 2012 Elsevier Ltd. All rights reserved.
1. Introduction Automated mineralogy systems generally use a combination of scanning electron microscope (SEM), backscattered-electron (BSE) images, image analysis, and energy dispersive spectrometry (EDS), to provide very useful data that cannot be obtained from other analytical techniques, particularly in complex ores. Common results outputs include relative abundance of minerals (modal analysis), liberation characteristics of valuable minerals, mineral grain and particle size, and mineral association data. These results can help to pinpoint mineralogical changes in an ore that can often affect mineral recoveries. If these mineralogical changes can be quantified with a reasonable degree of confidence, then the effect of these changes can be compensated for in the recovery plant. The terms ‘‘mineral grain’’ and ‘‘mineral particle’’ are used frequently in this paper. It is important to distinguish between the two. A mineral grain is a homogenous unit of pure mineral. A mineral particle is made up of one or more mineral grains. In the case of a pure, liberated mineral, the terms ‘‘grain’’ and ‘‘particle’’ are equivalent. In ores that contain platinum group minerals (PGMs), the plant feed material generally contains less than 10 ppm of platinum ⇑ Tel.: +27 11 709 4528; fax: +27 11 709 4564. E-mail address: [email protected] 0892-6875/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.mineng.2012.02.016
group elements (PGEs). The low PGE grade makes it virtually impossible to provide statistically meaningful PGM data by using traditional manual techniques. Automated SEM/EDS systems help to alleviate this problem by searching several polished sections per sample in unattended runs. Potential PGM-bearing particles are located by means of the high BSE intensity produced due to the high average atomic number of PGE-bearing minerals, and are identified by means of automated EDS analyses, performed on the constituent mineral grains in the PGM-bearing particles. Analysis results are saved to a database during the automated runs. Results from automated mineralogy systems are typically presented as tables or charts that summarise particular sample characteristics, and represent the total population or a specific sub-set of analysed mineral particles in the sample. To reliably determine floatability from mineralogical data, however, each analysed mineral particle needs to be individually evaluated. The reason for this is that more than one particle characteristic often needs to be considered to determine floatability. For example, a liberated PGM grain would be expected to float, but might not if its grain size is either very small or very large. In the case of composite particles, floatability depends on the minerals present, the grain size, association and mode of occurrence of each of these minerals, and the total particle size (Chetty et al., 2009). These individual particle data are written to the results database during analysis, but are usually not resolved in the pre-defined tables
76
C. Bushell / Minerals Engineering 36–38 (2012) 75–80
and charts produced by the system’s data processing software. The ‘‘grains table’’ within the automated SEM results database provides a source of individual PGM-bearing particle data required for interrogation by the flotation predictor. Similar approaches to predicting floatability and recovery of ores using mineral particle properties determined by automated SEM systems have been documented (Evans et al., 2011; Evans, 2010; Ford et al., 2007; Hunt et al., 2008; Lotter et al., 2003; Wightman et al., 2010, 2008). These, however, apply largely to base metal sulphide recovery, and do not involve the development of custom designed software to interrogate the mineralogical data. The methodology discussed in this paper is a simple approach, which has been developed to cater specifically for South African Bushveld PGM ores with typical feed grades of less than 10 g/t. The concept has been implemented as a VBA module within Microsoft Excel, which can accept PGM mineralogical data from any of the current automated SEM platforms.
2. Producing an input file for the flotation predictor The ‘‘grains table’’ produced by the automated SEM system is extracted from the sample results database, and is used to determine individual PGM grain and PGM-bearing particle characteristics. The first step is to determine the mode of occurrence of the PGM in each PGM-bearing particle. This is achieved by means of a Microsoft Excel spreadsheet, using a visual basic for applications (VBA) macro. The macro automatically classifies each PGM-bearing particle into one of six pre-defined mode of occurrence classes, described in Table 1 and illustrated in Fig. 1. Mineral ID and grain measurement data are also captured from the grains table. Grain areas are used to calculate a liberation index for each PGM-bearing particle. This measure of PGM grain liberation is calculated by dividing the area of potentially floatable component (PGM + BMS) by the total area of the particle (PGM + BMS + gangue). The resultant figure will range between 0 and 1, the latter indicating either a liberated PGM grain, or a binary particle containing PGM and BMS only. In contrast, the liberation index of a PGM grain totally enclosed within a BMS-barren silicate particle (low probability of flotation) will approach zero. Mineral ID, particle liberation index, grain and particle size, and PGM mode of occurrence for each PGM-bearing particle are used to produce a file that contains all of the necessary particle data to be interrogated by the flotation predictor software.
3. Flotation predictor operation and data flow The PGM flotation predictor predicts PGM recovery by producing an output based on the physical properties of all PGM-bearing particles detected in the sample. These properties include mode of occurrence of the PGM (as per descriptions in Table 1), the floatability of minerals associated with the PGM, liberation index (i.e. proportion of floatable component in the particle), and particle/ grain size. By classifying PGM-bearing particles into five classes that contain particles expected to float within particular time
Fig. 1. PGM-bearing particle mode of occurrence classes.
intervals, a recovery–time profile is produced. If desired, a flotation model can then be applied to this profile to determine flotation kinetics for the PGM bearing particles. The current version of the flotation predictor estimates fast and slow floating PGM fraction percentages and flotation rate constants by fitting the predicted recovery profile to the Kelsall flotation model (Kelsall, 1961) using a non-linear regression procedure. A simplified flow chart of the flotation predictor operation is provided in Fig. 2. The flotation predictor user interface, illustrated in Fig. 3, allows the user to set various flotation parameters prior to calculation. These include minimum and maximum particle sizes, choice of gangue minerals considered to be hydrophobic under the flotation conditions used, and liberation index cut-off values for AG and SAG class composite particles. SAG class liberation index cut-off values vary for the different BMS species, as the different BMS species have different floatability characteristics (Wiese et al., 2007; Penberthy et al., 2000). Initial ‘‘default’’ flotation parameter values are provided when the flotation predictor is run, and can be easily restored at any stage. These values are based on particle characteristics observed in samples generated by laboratory scale test-work that has been performed on various South African PGM ores. The parameters can be adjusted to suit a particular recovery plant, or circuit within that plant, according to mineralogical properties of PGM-bearing particles observed in representative samples of feed, concentrate and tailings gathered from the plant or circuit in question. The flotation predictor can thus produce results similar to those obtained from laboratory-scale rate tests from a full-scale operating plant. Once the plant feed has been properly characterised, the predictor output can be used as a diagnostic tool for routine plant monitoring and troubleshooting. Recovery via
Table 1 PGM-bearing particle mode of occurrence classes. PGM mode of occurrence class
Description
L SL AG SAG SG G
Liberated PGM PGM associated with BMS only (i.e. a binary PGM–BMS particle) PGM attached to silicate or oxide gangue (i.e. PGM exposed at particle perimeter) PGM associated with BMS attached to silicate or oxide gangue (i.e. BMS exposed at particle perimeter) PGM associated with BMS locked within silicate or oxide gangue (i.e. no exposure of BMS or PGM at particle perimeter) PGM locked within silicate or oxide gangue (i.e. no exposure of PGM at particle perimeter)
C. Bushell / Minerals Engineering 36–38 (2012) 75–80
77
Fig. 2. Flotation predictor flow diagram.
entrainment is not taken into account, as recovery via non-selective mechanisms cannot be predicted based on particle liberation properties. When evaluating concentrate samples, however, the presence of any SG or G class (i.e. non-floatable) particles, or PGM-bearing particles with very low liberation indices is considered to be due to entrainment, provided that the gangue component of the particle is not hydrophobic.
4. Flotation predictor output The flotation predictor produces a direct output from mineralogical data, highlighting the metallurgical properties of PGM-bearing particles in the sample in tabular and graphical format. The results file includes quantitative information on the potential recovery, the reasons for recovery or losses, particle size distribu-
78
C. Bushell / Minerals Engineering 36–38 (2012) 75–80
Fig. 3. Flotation predictor user interface. Note: numerical parameters are shown as examples, and do not necessarily represent realistic values.
tion, grouping of valuable species, and estimation of flotation kinetics. The output varies according to the sample type. For example, feed sample output focuses more on potential recovery and estimation of flotation kinetics, whilst tailing sample output focuses more on reasons for losses. Typical flotation predictor graphical outputs are illustrated in Figs. 4–8. These include: Potential PGM recovery, expressed as both volume and number percentage of the total PGM population assessed (Fig. 4). A predicted PGM flotation rate curve, based on the particle characteristics of the recoverable PGM particles detected (Fig. 5). Reasons for PGM losses (Fig. 6), categorised into the four major causes of losses to tailings: 1. PGM locked in gangue (i.e. no exposed floatable component). 2. PGM-bearing particles with low liberation indices (i.e. a particle with a very low proportion of floatable component).
Fig. 4. Potential PGM recovery (as % of PGM population assessed).
3. Particles too large to float. 4. Particles too small to float. PGM-bearing particle size distribution (Fig. 7), useful for comparison with assay-by-size results – favourable comparison validates the auto-SEM results. Grouping of similar PGM species (Fig. 8). As the dataset of PGMbearing particles obtained by the auto-SEM is small due to the low PGE grade of typical samples, grouping can help to achieve a mineral balance across the recovery circuit. The metallurgical significance of mineralogical results from automated mineralogy systems can often be difficult to interpret.
Fig. 5. Predicted PGM time–recovery profile for PGM feed sample.
C. Bushell / Minerals Engineering 36–38 (2012) 75–80
Fig. 6. Reasons for PGM losses.
Fig. 7. PGM distribution by particle size (for comparison with assay-by-size results).
predictor software is run. This was achieved by performing mineralogical investigations on concentrates collected at 1, 3, 7, 20 and 40 min during the flotation rate tests on each ore. The physical characteristics of PGM-bearing particles detected in each of the concentrate samples were noted, and these characteristics were used to establish the cut-off values for PGM size, mode of occurrence and liberation indices for particles that would be expected to report to individual concentrates within the five abovementioned time limits. Comparisons of flotation rate tests and flotation predictor rate estimation results on three different ores are presented in Fig. 9. The flotation response of 2 reasonably unaltered ores (A & B), and one highly altered ore, (C) is shown. Recovery profiles in these results were produced by using the same flotation prediction parameters, based on collective particle characteristics observed in all three ores. The results clearly show that the parameters used were far better suited to unaltered ores. By adjusting the liberation index threshold parameters to match particle characteristics observed in the concentrates produced by rate tests on ore C only, a much improved recovery–time profile for ore C, shown in Fig. 10, was produced. This shows the importance of matching the predictor parameters to the ore’s mineralogical characteristics. Three sets of initial default parameters (for unaltered, partially altered and highly altered ores) are planned for inclusion in the next version of the flotation predictor software, but these only provide a guideline towards the optimal values of the various parameters. Ideally, a specific set of flotation predictor parameters should be established for a particular ore by first performing a thorough PGM mineralogical analysis of feed, concentrate and tailing samples from the recovery plant. Flotation predictor parameters can then be derived from the PGM-bearing particle characteristics observed in the various recovery plant streams. The output produced by the optimised parameters defines a baseline mineralogical profile of the ore, against which future routine mineralogical analyses can be compared. This allows the flotation predictor output to be used as a diagnostic tool, providing a way to monitor the plant, and to establish whether any change in plant performance is due to changes in feed mineralogy. As long as plant reagent addition remains reasonably constant, the effect of reagents should not affect results negatively, as the baseline data will have been established with reagents present.
Fig. 8. Grouped PGM species.
The link between ore mineralogy and plant performance is not always clearly revealed by traditional liberation, particle size and mineral association results from these systems. The aim of the flotation predictor is to interpret the mineralogical results in a repeatable manner to produce meaningful metallurgical data in an easy to understand format. Although this has only been applied to flotation of South African PGM ores at this stage, the same principles should be equally applicable to other orebodies and commodities. 5. Comparison of flotation predictor results with laboratory scale flotation rate tests Laboratory scale rougher flotation rate tests on several PGM ores were used to establish the initial ‘‘default’’ flotation predictor parameters which appear in the user interface when the flotation
79
Fig. 9. Comparison of rate tests and flotation predictor output, ores A–C.
80
C. Bushell / Minerals Engineering 36–38 (2012) 75–80
a mineralogical profile for the plant. Future PGM mineralogy studies can be compared against this mineralogical profile to establish whether changes in feed mineralogy are responsible for variations in recovery. Where feed mineralogy remains constant, but concentrates and tailings differ from the baseline data, plant issues other than mineralogical changes are indicated. Maximum value can be obtained from mineralogical data by using the flotation predictor output in conjunction with other mineralogical information, such as relative mineral proportions (modal analysis) and mineral associations. Acknowledgements The author would like to thank Mr. Nigel Ramlall of Mintek’s Minerals Processing Division for performing the laboratory scale rate tests and for providing insight on various aspects of PGM flotation, as well as colleagues in Mintek’s Mineralogy Division for their support and encouragement during the developmental stage of the flotation predictor.
Fig. 10. Comparison of rate test and flotation predictor output for altered ore, after adjusting liberation index threshold parameters to suit the ore characteristics.
6. Future plans The current version of the flotation predictor (v 2.0, June 2011) operates as a VBA module within Microsoft Excel. Although this works adequately, VBA format is not considered a viable long-term development platform. The software will be upgraded to a standalone executable format in future. Capabilities will also be expanded to allow flotation prediction of commodities other than PGM. 7. Conclusions The flotation predictor produces results that show good correlation with results obtained from laboratory scale flotation rate tests. The predictor is not intended to replace rate tests, as these can be performed quicker and cheaper than automated SEM mineralogical analyses. The predictor can, however, produce flotation rate estimates directly from a full-scale operating flotation plant, and also links flotation performance to ore mineralogy, which is not achieved by traditional rate tests. This provides a means to monitor and troubleshoot plant performance based on ore mineralogy. Although the pre-set default flotation parameters produce useable results, prediction can be improved by establishing parameters through a detailed mineralogical study of the ore and plant products generated from it. Once these baseline data have been established, the flotation prediction parameters essentially define
References Chetty, D., Gryffenberg, L., Lekgetho, T.B., Molebale, I.J., 2009. Automated SEM study of PGM distribution across a UG2 flotation bank: implications for understanding PGM floatability. The Journal of the Southern African Institute of Mining and Metallurgy 109, 587–593. Evans, C.L., 2010. Development of a methodology to estimate flotation separability from ore microtexture. PhD thesis, University of Queensland. Evans, C.L., Wightman, E.M., Manlapig, E.V., Coulter, B.L., 2011. Application of process mineralogy as a tool in sustainable processing. Minerals Engineering 24, 1242–1248. Ford, F., Lee, A., Davis, C., Xu, M., Lawson, V., 2007. Predicting Clarabelle mill recoveries using Mineral Liberation Analyser (MLA) grade-recovery curves. In: Proceedings of the 48th Annual Conference of Metallurgists of CIM, Sudbury, Ontario, Canada, pp. 32–41. Hunt, J.A., Berry, R., Walters, S., Bonnici, N., Kamenetsky, M., Nguyen, K., Evans, C.L., 2008. A new look at mineral maps and the potential relationships of extracted data to mineral processing behaviours. In: Ninth International Congress for Applied Mineralogy. ICAM Australia 2008, Brisbane, QLD, 8–10 September, 2008, pp. 429–432. Kelsall, D.F., 1961. Application of probability in the assessment of flotation systems. Transactions of the Institution of Mining and Metallurgy 70, 191–204. Lotter, N.O., Kowal, D.L., Tuzun, M.A., Whittaker, P.J., Kormos, L., 2003. Sampling and flotation testing of Sudbury Basin drill core for process mineralogy modelling. Minerals Engineering 16 (9), 857–864. Penberthy, C.J., Oosthuyzen, E.J., Merkle, R.K.W., 2000. The recovery of platinum group elements from the UG-2 chromitite, Bushveld complex – a mineralogical perspective. Mineralogy and Petrology 68, 213–222. Wiese, J.G., Becker, M., Bradshaw, D.J., Harris, P.J., 2007. Interpreting the role of reagents in the flotation of platinum-bearing Merensky ores. The Journal of the Southern African Institute of Mining and Metallurgy 107, 29–36. Wightman, E., Evans, C.L., Vizcarra, T., and Sandoval, G. 2008. Process mineralogy as a tool in modelling mineral processing operations. In: Ninth International Congress for Applied Mineralogy. ICAM Australia 2008, Brisbane, QLD, 8–10 September, 2008, pp. 475–481. Wightman, E., Evans, C., Bradshaw, D., Manlapig, E.V., 2010. Predicting flotation response from liberation information. In: Proceedings XXV International Mineral Processing Congress, Brisbane, pp. 3403–3409.