Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Amenability of low-grade ore stockpiles to sensor-based ore sorting technology Mazhary, Arvin 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2017_november_mazhary_arvin.pdf [ 6.14MB ]
Metadata
JSON: 24-1.0355396.json
JSON-LD: 24-1.0355396-ld.json
RDF/XML (Pretty): 24-1.0355396-rdf.xml
RDF/JSON: 24-1.0355396-rdf.json
Turtle: 24-1.0355396-turtle.txt
N-Triples: 24-1.0355396-rdf-ntriples.txt
Original Record: 24-1.0355396-source.json
Full Text
24-1.0355396-fulltext.txt
Citation
24-1.0355396.ris

Full Text

    AMENABILITY OF LOW-GRADE ORE STOCKPIES TO SENSOR-BASED ORE SORTING TECHNOLOGY by Arvin Mazhary M.A.Sc., McMaster University, 2010 M. di 2̊ Livello, La Sapienza University of Rome, 2008  A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in The Faculty of Graduate and Postdoctoral Studies (Mining Engineering)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) August 2017  © Arvin Mazhary, 2017  ii   ABSTRACT With sensor-based ore sorting attracting more attention among the industry leaders, and in an effort to show the potential for sensor-based ore sorting technology, this research takes a particle sorting approach and looks at sorting low-grade and waste rock stockpiles to concentrate the misplaced mineralized rocks and generate value. The results from the optical sensor showed that where there was a visual distinction between the mineralized and gangue material, this sensor managed to identify each group well.  Despite using a multivariate linear regression (MLR) analysis, the electromagnetic sensor did not predict the grades effectively. The X-Ray Transmission (XRT) sensor performed quite well for both base metal and gold samples. One recurring problem was the presence of iron minerals such as pyrite that, due to their relatively high atomic density, tarnished the sorting results. With elemental distinguishing capabilities, the X-Ray Fluorescence (XRF) sensor boasts great potential for ore sorting. Both single and multivariate linear regression analysis were used to analyse the results from the XRF sensor. Although, while overall satisfactory results were obtained from the XRF sensor, sensor capabilities in actual dynamic sorting cases need to be assessed. Recommendations for future work can be on different aspects of this work. One would be to try to improve the static, bench-top testing facilities so they represent dynamic sorting scenarios better, such as use of a conveyor-type platform where rocks can pass under a sensor. If a similar study is to be performed, it is highly suggested to focus the efforts on one mine, one size fraction (preferably -50 mm +37.5 mm) with a larger number of particles. In terms of continuation of this work, it would be best to take these tests to the next level and perform bulk sorting tests to determine how these bench-scale tests correlate with bulk dynamic sorting results. Also, a detailed economic analysis based on these results would yield valuable results.   iii  LAY SUMMARY Sorting rocks using various sensors is similar to scanning luggage with the airport x-ray scanners. Similar to those scanners, when a rock particle that contains metal (desired particle) goes under a sensor, X-ray as an example, it looks darker than a rock particle that contains no metal, such as quartz crystals. Once the sensor distinguishes this difference, it sends a message to a mechanical or pneumatic actuator to reject the rocks that are not desired. Pre-concentration of coarse particles has various economic and environmental benefits such as lower energy consumption, lower water consumption, lower chemical reagent consumption, lower carbon footprint, smaller waste disposal areas and higher recoveries. Overall, this technology shifts the mining industry toward a more sustainable practice.   iv  PREFACE This dissertation is original, partly published independent work of the author, Arvin Mazhary. During this work, some help with regards to sample preparation and data analysis, was sought from other co-op and/or scholarship students, however, all work was supervised and fact-checked by the author. The “Major Sorting and Sensor Providing Companies” section of Chapter 2 was originally submitted as part of completion requirements of a co-op term that the author performed at SRK Consulting. This section was updated for to be included in this thesis. Chapter 4 was originally published as a Conference Proceeding for the Canadian Mineral Processor’s Society conference in 2015 (CMP 2015) and was updated and modified slightly for this Thesis. All other data are unpublished and a result of independent work of the author, Arvin Mazhary, with some help from other students.   v  TABLE OF CONTENTS Abstract ..................................................................................................................................... ii Lay Summary ........................................................................................................................... iii Preface...................................................................................................................................... iv Table of Contents ...................................................................................................................... v List of Tables ........................................................................................................................... xi List of Figures ........................................................................................................................ xvi Acknowledgements .............................................................................................................. xxiv Dedication ............................................................................................................................. xxv 1. Introduction ....................................................................................................................... 1 2. Literature Review.............................................................................................................. 3  Optical Sensor (OPT) ................................................................................................. 5  Comex Optical Sorters ........................................................................................ 5  Steinert Optical Sorters ....................................................................................... 7  Tomra Optical Sorters ......................................................................................... 7  Electro-Magnetic Induction Sensor (EM) .................................................................. 9  MineSense Technologies EM Sorters ............................................................... 11  Steinert Induction Sorters ................................................................................. 11  Tomra EM Sorters............................................................................................. 11  X-Ray Transmission (XRT) ..................................................................................... 11  Comex XRT Sorters .......................................................................................... 12  Steinert XRT Sorters ......................................................................................... 15  Tomra XRT Sorters........................................................................................... 15 vi   X-Ray Fluorescence (XRF) ...................................................................................... 16  MineSense Technologies XRF Sorters ............................................................. 18  Rados XRF Sorters ........................................................................................... 18  Steinert XRF Sorters ......................................................................................... 19  Ore-Sorting Case Studies ......................................................................................... 19  Anglo American Platinum – UG2 Mine ........................................................... 19  Goldfields – Kloof Gold Mine .......................................................................... 24  Coeur Operations – Kensington Mine .............................................................. 29  Rocklands Mine (CuDECO Limited, 2015) ..................................................... 30  Mittersill Mine (Mosser & Robben, 2014) ....................................................... 31  Current Work............................................................................................................ 31 3. Experimental Methods .................................................................................................... 33  Sampling................................................................................................................... 33  Sample Collection ............................................................................................. 33  Sampling Error .................................................................................................. 34  Material Preparation ................................................................................................. 36  Sensor Tests.............................................................................................................. 36  Optical Tests ..................................................................................................... 36  Electromagnetic (EM) Tests ............................................................................. 37  XRT Tests ......................................................................................................... 37  XRF Tests ......................................................................................................... 38  Near Infrared (NIR) Tests ................................................................................. 39  Data Processing ........................................................................................................ 39 vii   Constitution Heterogeneity ............................................................................... 39  Metal Liberation Curve ..................................................................................... 39  Grade Estimation .............................................................................................. 39  Multivariate Linear Regression (MLR) Analysis ............................................. 40 4. Ore Characterization ....................................................................................................... 42  Introduction .............................................................................................................. 42  Optical Sensor Response Analysis ........................................................................... 42  Brenda – Size Fraction A (-25+19 mm) ........................................................... 43  Copper Mountain – Size Fraction B (-37.5+25 mm) ........................................ 46  Mount Polley – Size Fraction C (-50+37.5 mm) .............................................. 49  Myra Falls – Size Fraction D (-75+50 mm) ..................................................... 51  Electromagnetic Sensor Response Analysis ............................................................ 53  Copper Mountain – Size B ................................................................................ 54  Copper Mountain – Size D ............................................................................... 56  XRT Sensor Response Analysis ............................................................................... 59  Brenda ............................................................................................................... 59  Copper Mountain .............................................................................................. 62  Mount Polley ..................................................................................................... 66  Myra Falls ......................................................................................................... 69  XRF Sensor Response Analysis ............................................................................... 73  Comparison of Base Metal Grades with ICP .................................................... 73  Comparison of Light Elements ......................................................................... 82  Inductively Couple Plasma - Mass Spectroscopy (ICP-MS) ................................... 92 viii   Rietveld Refinement (XRD)................................................................................... 101  Constitution Heterogeneity .................................................................................... 106  Heterogeneity and Sorting Potential ............................................................... 110  Heterogeneity and Particle Size ...................................................................... 114  Heterogeneity of Gangue Elements ................................................................ 117  Conclusion ...................................................................................................... 118 5. Brenda Sortability Results ............................................................................................ 120  Optical Sortability Results ..................................................................................... 120  Electromagnetic Sortability Result......................................................................... 122  Size A (-25+19 mm) ....................................................................................... 122  Size B (-37.5+25 mm) .................................................................................... 123  Size C (-50+37.5 mm) .................................................................................... 124  Size D (-75+50 mm) ....................................................................................... 124  X-Ray Transmission Sortability Results ................................................................ 126  X-Ray Fluorescence Sorting Results...................................................................... 128  Size A (-25+19 mm) ....................................................................................... 129  Size B (-37.5+25 mm) .................................................................................... 131  Size C (-50+37.5 mm) .................................................................................... 134  Size D (-75+50 mm) ....................................................................................... 135  Conclusion .............................................................................................................. 137 6. Copper Mountain Sortability Results............................................................................ 139  Optical Sortability Results ..................................................................................... 139  Electromagnetic Sortability Results ....................................................................... 141 ix   X-Ray Transmission Sortability Results ................................................................ 142  X-Ray Fluorescence Sortability Results ................................................................ 145  Size A (-25+19 mm) ....................................................................................... 145  Size B (-37.5+25 mm) .................................................................................... 148  Size C (-50+37.5 mm) .................................................................................... 150  Size D (-75+50 mm) ....................................................................................... 152  Conclusion .............................................................................................................. 155 7. Mount Polley Sortability Results .................................................................................. 157  Optical Sortability Results ..................................................................................... 157  Electromagnetic Sortability Results ....................................................................... 160  Size A (-25+19 mm) ....................................................................................... 160  Size B (-37.5+25 mm) .................................................................................... 161  Size C (-50+37.5 mm) .................................................................................... 161  Size D (-75+50 mm) ....................................................................................... 163  X-Ray Transmission Sortability Results ................................................................ 165  X-Ray Fluorescence Sortability Results ................................................................ 170  Size A (-25+19 mm) ....................................................................................... 170  Size B (-37.5+25 mm) .................................................................................... 172  Size C (-50+37.5 mm) .................................................................................... 174  Size D (-75+50 mm) ....................................................................................... 176  Conclusion .............................................................................................................. 177 8. Myra Falls Sortability Results ...................................................................................... 179  Optical Sortability Results ..................................................................................... 179 x   Electromagnetic Sortability Results ....................................................................... 182  X-Ray Transmission Sortability Results ................................................................ 182  X-Ray Fluorescence Sortability Results ................................................................ 185  Size A (-25+19 mm) ....................................................................................... 186  Size B (-37.5+25 mm) .................................................................................... 188  Size C (-50+37.5 mm) .................................................................................... 190  Size D (-75+50 mm) ....................................................................................... 192  Conclusion .............................................................................................................. 194 9. Gold Sortability Results ................................................................................................ 196  Electromagnetic Sortability Results ....................................................................... 196  X-Ray Transmission Sortability ............................................................................. 197  X-Ray Fluorescence Sortability Results ................................................................ 198  Single-Variate Regression Analysis ............................................................... 199  Multivariate Linear Regression Analysis........................................................ 204  Conclusion .............................................................................................................. 208 10. Conclusion and Recommendations ............................................................................ 210 References ............................................................................................................................. 213    xi  LIST OF TABLES Table 2.1 Magnetic properties of dominant sulphides available in the tested material .......... 10 Table 2.2 Results from iron ore (type 1) sorting through X-ray Transmission (XRT) ........... 14 Table 2.3 Results from iron ore (type 1) sorting using both an X-ray Transmission (XRT) as well as an optical sensor utilizing pattern recognition techniques .......................................... 14 Table 2.4 Sorting iron ore (type 2) using XRT ....................................................................... 14 Table 2.5 Sorting iron ore (type 2) using XRT utilizing pattern recognition ......................... 14 Table 2.6 Coal sorting with conventional use of an XRT sorter ............................................ 14 Table 2.7 Coal sorting with an XRT sorter using pattern recognition .................................... 15 Table 2.8 Different types of XRF sorters provided by Rados ................................................ 19 Table 2.9. Gold grades of various rock types ......................................................................... 25 Table 2.10. Kloof Gold Mine sorting operating data – tonnage and availability ................... 27 Table 2.11. Kloof Gold Mine sorting operating data – grades and recovery.......................... 27 Table 2.12. Kloof Gold Mine sorting OPEX .......................................................................... 27 Table 2.13. Kloof Gold Mine sorting economic outcome ...................................................... 28 Table 2.14. 2009 Kloof bulk test work results (Von Ketelhodt, Falcon, & Falcon, 2011)..... 28 Table 4.1 Spectral Indices for select rocks from Brenda Size Fraction A .............................. 43 Table 4.2 Spectral Indices for select rocks from Copper Mountain Size Fraction B ............. 46 Table 4.3 Spectral Indices for select rocks from Mount Polley Size Fraction C .................... 49 Table 4.4 Spectral Indices for select rocks from Myra Falls Size Fraction D ........................ 51 Table 4.5 Copper and iron grades for select samples and corresponding Phase response ..... 56 Table 4.6 Copper and iron grades for select samples and corresponding Magnitude and Phase response................................................................................................................................... 58 xii  Table 4.7 XRT spectral indices and grades for select rocks in Brenda Size A ....................... 59 Table 4.8 Top 10 rocks based on XRT Spectral Index – Brenda Size A ................................ 62 Table 4.9 XRT spectral indices and grades for select rocks in Copper Mountain Size B ...... 63 Table 4.10 Top 10 rocks based on XRT Spectral Index – Copper Mountain size fraction B 66 Table 4.11 XRT spectral indices and grades for select rocks in Mount Polley size fraction C................................................................................................................................................. 67 Table 4.12 Top 10 rocks based on XRT Spectral Index – Mount Polley size fraction C ....... 69 Table 4.13 XRT spectral indices and grades for select rocks in Myra Falls size fraction D .. 70 Table 4.14 Top 10 rocks based on XRT Spectral Index – Mya Falls size fraction D ............ 73 Table 4.15 Correlation factors: XRF vs. ICP – Heavy Metals, Myra Falls ............................ 74 Table 4.16 Comparison of Grades: XRF vs. ICP, Myra Falls A, Zinc & Copper .................. 79 Table 4.17 Comparison of Grades: XRF vs. ICP, Myra Falls A, Lead & Iron ...................... 80 Table 4.18 Comparison of Grades: XRF vs. ICP, Myra Falls D, Zinc & Copper .................. 81 Table 4.19 Comparison of Grades: XRF vs. ICP, Myra Falls D, Lead & Iron ...................... 82 Table 4.20 Correlation factors: XRF vs. ICP – Light Elements, Myra Falls .......................... 83 Table 4.21 Comparison of Grades: XRF vs. ICP, Myra Falls A, Sulfur & Magnesium ........ 88 Table 4.22 Comparison of Grades: XRF vs. ICP, Myra Falls A, Calcium & Potassium ....... 89 Table 4.23 Comparison of Grades: XRF vs. ICP, Myra Falls D, Sulfur & Magnesium ........ 90 Table 4.24 Comparison of Grades: XRF vs. ICP, Myra Falls D, Calcium & Potassium ....... 91 Table 4.25 Brenda ................................................................................................................... 94 Table 4.26 Copper Mountain .................................................................................................. 95 Table 4.27 Mount Polley......................................................................................................... 96 Table 4.28 Myra Falls ............................................................................................................. 97 xiii  Table 4.29 XRD analysis for Brenda .................................................................................... 102 Table 4.30 XRD analysis for Copper Mountain ................................................................... 103 Table 4.31 XRD analysis for Mount Polley.......................................................................... 104 Table 4.32 XRD analysis for ................................................................................................ 105 Table 4.33 XRD analysis for two visually mineralized gold samples .................................. 106 Table 4.34 XRD analysis for two visually gangue gold samples ......................................... 106 Table 4.35 Summary of the calculated grades of collected material (all size fractions) ...... 110 Table 4.36 Recovery at different mass pulls for with regard to their CH value for all samples............................................................................................................................................... 113 Table 4.37 Measurements for each size fraction .................................................................. 115 Table 5.1 Optical sorting summary – Brenda mine .............................................................. 122 Table 5.2 MLR statistical data for EM – Brenda-A (-25+19 mm) ....................................... 123 Table 5.3 MLR statistical data for EM – Brenda-B (-37.5+25 mm) .................................... 124 Table 5.4 MLR statistical data for EM – Brenda-D (-75+50 mm) ....................................... 125 Table 5.5 Summary of the XRT sorting results – Brenda..................................................... 126 Table 5.6 Predicted Cu grades vs. true grades – Brenda A 15 highest grade rocks ............. 130 Table 5.7 MLR analysis coefficients and statistical data – Brenda A .................................. 131 Table 5.8 Predicted Cu grades vs. true grades – Brenda B 18 highest grade rocks .............. 133 Table 5.9 MLR analysis coefficients and statistical data – Brenda B .................................. 133 Table 5.10 Predicted Cu grades vs. true grades – Brenda C 10 highest grade rocks ............ 134 Table 5.11 MLR analysis coefficients and statistical data – Brenda C ................................ 135 Table 5.12 Predicted Cu grades vs. true grades – Brenda D 15 highest grade rocks ........... 136 Table 5.13 MLR analysis coefficients and statistical data – Brenda D ................................ 137 xiv  Table 6.1 Optical sorting summary – Copper Mountain ...................................................... 141 Table 6.2 MLR statistical data for EM – Copper Mountain-D ............................................. 142 Table 6.3 Summary of the XRT sorting results – Copper Mountain .................................... 143 Table 6.4 Predicted Cu grades vs. true grades – Copper Mountain A 15 highest grade rocks............................................................................................................................................... 147 Table 6.5 MLR analysis coefficients and statistical data – Copper Mountain A ................. 148 Table 6.6 Predicted Cu grades vs. true grades – Copper Mountain B 15 highest grade rocks............................................................................................................................................... 149 Table 6.7 MLR analysis coefficients and statistical data – Copper Mountain B .................. 150 Table 6.8 Predicted Cu grades vs. true grades – Copper Mountain C 15 highest grade rocks............................................................................................................................................... 151 Table 6.9 MLR analysis coefficients and statistical data – Copper Mountain C .................. 152 Table 6.10 Predicted Cu grades vs. true grades – Copper Mountain D 15 highest grade rocks............................................................................................................................................... 154 Table 6.11 MLR analysis coefficients and statistical data – Copper Mountain D ............... 155 Table 7.1 Optical sorting summary – Mount Polley ............................................................. 160 Table 7.2 MLR statistical data for EM – Mount Polley-A ................................................... 161 Table 7.3 MLR statistical data for EM – Mount Polley-C ................................................... 163 Table 7.4 MLR statistical data for EM – Mount Polley-D ................................................... 165 Table 7.5 Summary of the XRT spectral sorting results – Mount Polley ............................. 168 Table 7.6 Predicted Cu grades vs. true grades – Mount Polley A 15 highest grade rocks ... 171 Table 7.7 MLR analysis coefficients and statistical data – Mount Polley A ........................ 172 Table 7.8 Predicted Cu grades vs. true grades – Mount Polley B 15 highest grade rocks ... 173 Table 7.9 MLR analysis coefficients and statistical data – Mount Polley B ........................ 174 xv  Table 7.10 Predicted Cu grades vs. true grades – Mount Polley C 15 highest grade rocks . 175 Table 7.11 MLR analysis coefficients and statistical data – Mount Polley C ...................... 176 Table 7.12 Predicted Cu grades vs. true grades – Mount Polley D 15 highest grade rocks . 177 Table 7.13 MLR analysis coefficients and statistical data – Mount Polley D ...................... 177 Table 8.1 Optical sorting summary – Myra Falls ................................................................. 182 Table 8.2 Summary of the XRT spectral sorting results – Myra Falls ................................. 183 Table 8.3 Top 15 rocks based on XRT Spectral Index – Mya Falls size fraction A ............ 184 Table 8.4 Predicted Zn grades vs. true grades – Myra Falls A 15 highest grade rocks ........ 187 Table 8.5 MLR analysis coefficients and statistical data – Myra Falls A ............................ 188 Table 8.6 Predicted Zn grades vs. true grades – Myra Falls B 15 highest grade rocks ........ 189 Table 8.7 MLR analysis coefficients and statistical data – Myra Falls B............................. 190 Table 8.8 Predicted Zn grades vs. true grades – Myra Falls C 15 highest grade rocks ........ 191 Table 8.9 MLR analysis coefficients and statistical data – Myra Falls C............................. 192 Table 8.10 Predicted Zn grades vs. true grades – Myra Falls D 15 highest grade rocks ...... 193 Table 8.11 MLR analysis coefficients and statistical data – Myra Falls D .......................... 194 Table 9.1 EM sensor’s MLR coefficients and statistical data for the gold sample .............. 196 Table 9.2 Regression model coefficients for XRF datasets with and without interaction effects............................................................................................................................................... 205 Table 9.3 Regression model coefficients for ICP datasets with and without interaction effects............................................................................................................................................... 207    xvi  LIST OF FIGURES Figure 2.1 List of sensors and their corresponding operating wavelenngths ............................ 4 Figure 2.2 Components of a Comex OSX Sorter ..................................................................... 6 Figure 2.3 Components of a Comex VSX Sorter ..................................................................... 7 Figure 2.4 Components of a Comex OCXR Sorter ................................................................ 13 Figure 2.5 Schematic of XRF detection principles (Portable Analytical Solutions, 2017) .... 17 Figure 2.6. RadosTM XRF ore sorter setup (Rule, Fouchee, & Swart, 2015) ......................... 21 Figure 2.7. Simplified PFD of the Proof of Concept plant (Rule, Fouchee, & Swart, 2015) . 22 Figure 2.8. Proof of Concept plant mass and metal balance used in business case (Rule, Fouchee, & Swart, 2015) ........................................................................................................ 22 Figure 2.9. Correlation between PGM and copper for PGM sample considered (Rule, Fouchee, & Swart, 2015) ........................................................................................................................ 23 Figure 2.10. First PGM deportment results from 4 channels of Sorter 1 (Rule, Fouchee, & Swart, 2015) ............................................................................................................................ 24 Figure 2.11. Kloof Gold Mine optical sorter flowsheet .......................................................... 26 Figure 2.12. Left, sorted product, right, rejected waste .......................................................... 26 Figure 2.13. Au grade and recovery vs. mass-pull (Von Ketelhodt, Falcon, & Falcon, 2011)................................................................................................................................................. 29 Figure 4.1 Processed optical image for Brenda A Rock #29 .................................................. 44 Figure 4.2 Processed optical image for Brenda A Rock #38 .................................................. 44 Figure 4.3 Processed optical image for Brenda A Rock #47 .................................................. 45 Figure 4.4 Processed optical image for Brenda A Rock #2 .................................................... 45 Figure 4.5 Processed optical image for Brenda A Rock #39 .................................................. 46 Figure 4.6 Processed optical image for Copper Mountain B, Rock #20 ................................ 47 xvii  Figure 4.7 Processed optical image for Copper Mountain B, Rock #58 ................................ 47 Figure 4.8 Processed optical image for Copper Mountain B, Rock #36 ................................ 48 Figure 4.9 Processed optical image for Copper Mountain B, Rock #86 ................................ 48 Figure 4.10 Processed optical image for Copper Mountain B, Rock #47 .............................. 49 Figure 4.11 Processed optical image for Mount Polley C, Rock #40 ..................................... 50 Figure 4.12 Processed optical image for Mount Polley C, Rock #11 ..................................... 50 Figure 4.13 Processed optical image for Mount Polley C, Rock #69 ..................................... 50 Figure 4.14 Processed optical image for Mount Polley C, Rock #52 ..................................... 51 Figure 4.15 Processed optical image for Mount Polley C, Rock #29 ..................................... 51 Figure 4.16 Processed optical image for Myra Falls D, Rock #63 ......................................... 52 Figure 4.17 Processed optical image for Myra Falls D, Rock #56 ......................................... 52 Figure 4.18 Processed optical image for Myra Falls D, Rock #4 ........................................... 52 Figure 4.19 Processed optical image for Myra Falls D, Rock #52 ......................................... 53 Figure 4.20 Processed optical image for Myra Falls D, Rock #22 ......................................... 53 Figure 4.21 Magnitude vs. Frequency for Copper Mountain Size B – select rocks ............... 55 Figure 4.22 Phase vs. Frequency for Copper Mountain Size B – select rocks ....................... 56 Figure 4.23 Magnitude vs. Frequency for Copper Mountain Size D – select rocks ............... 57 Figure 4.24 Phase vs. Frequency for Copper Mountain Size D – select rocks ....................... 57 Figure 4.25 Processed Spectral XRT image for Brenda A, Rock #29 .................................... 60 Figure 4.26 Processed Spectral XRT image for Brenda A, Rock #47 .................................... 61 Figure 4.27 Processed Spectral XRT image for Brenda A, Rock #27 .................................... 61 Figure 4.28 Processed Spectral XRT image for Brenda A, Rock #34 .................................... 62 Figure 4.29 Processed Spectral XRT image for Copper Mountain, Rock #21 ....................... 63 xviii  Figure 4.30 Processed Spectral XRT image for Copper Mountain, Rock #4 ......................... 64 Figure 4.31 Processed Spectral XRT image for Copper Mountain, Rock #54 ....................... 65 Figure 4.32 Processed Spectral XRT image for Copper Mountain, Rock #60 ....................... 65 Figure 4.33 Processed Spectral XRT image for Mount Polley, Rock #40 ............................. 67 Figure 4.34 Processed Spectral XRT image for Mount Polley, Rock #11 ............................. 68 Figure 4.35 Processed Spectral XRT image for Mount Polley, Rock #30 ............................. 68 Figure 4.36 Processed Spectral XRT image for Mount Polley, Rock #47 ............................. 69 Figure 4.37 Processed Spectral XRT image for Myra Falls, Rock #23 .................................. 71 Figure 4.38 Processed Spectral XRT image for Myra Falls, Rock #39 .................................. 71 Figure 4.39 Processed Spectral XRT image for Myra Falls, Rock #68 .................................. 72 Figure 4.40 Processed Spectral XRT image for Myra Falls, Rock #26 .................................. 72 Figure 4.41 Comparison: XRF vs. ICP – Myra Falls, Zinc, size fraction A .......................... 75 Figure 4.42 Comparison: XRF vs. ICP – Myra Falls, Zinc, size fraction D .......................... 75 Figure 4.43 Comparison: XRF vs. ICP – Myra Falls, Copper, size fraction A ...................... 76 Figure 4.44 Comparison: XRF vs. ICP – Myra Falls, Copper, size fraction D ...................... 76 Figure 4.45 Comparison: XRF vs. ICP – Myra Falls, Lead, size fraction A .......................... 77 Figure 4.46 Comparison: XRF vs. ICP – Myra Falls, Lead, size fraction D .......................... 77 Figure 4.47 Comparison: XRF vs. ICP – Myra Falls, Iron, size fraction A ........................... 78 Figure 4.48 Comparison: XRF vs. ICP – Myra Falls, Iron, size fraction D ........................... 78 Figure 4.49 Comparison: XRF vs. ICP – Myra Falls, Sulfur, size fraction A ........................ 84 Figure 4.50 Comparison: XRF vs. ICP – Myra Falls, Sulfur, size fraction D ........................ 84 Figure 4.51 Comparison: XRF vs. ICP – Myra Falls, Magnesium, size fraction A ............... 85 Figure 4.52 Comparison: XRF vs. ICP – Myra Falls, Magnesium, size fraction D ............... 85 xix  Figure 4.53 Comparison: XRF vs. ICP – Myra Falls, Calcium, size fraction A .................... 86 Figure 4.54 Comparison: XRF vs. ICP – Myra Falls, Calcium, size fraction D .................... 86 Figure 4.55 Comparison: XRF vs. ICP – Myra Falls, Potassium, size fraction A ................. 87 Figure 4.56 Comparison: XRF vs. ICP – Myra Falls, Potassium, size fraction D ................. 87 Figure 4.57 ICP vs. Pulverized XRF copper correlation for Brenda Size A .......................... 98 Figure 4.58 ICP vs. Pulverized XRF iron correlation for Brenda Size A ............................... 98 Figure 4.59 ICP vs. Pulverized XRF sulfur correlation for Brenda Size A ............................ 99 Figure 4.60 ICP vs. Pulverized XRF Zinc correlation for Myra Falls Size D ........................ 99 Figure 4.61 ICP vs. Pulverized XRF iron correlation for Myra Falls Size D ....................... 100 Figure 4.62 ICP vs. Pulverized XRF sulfur correlation for Myra Falls Size D .................... 100 Figure 4.63 Mount Polley copper grade distribution (size fraction B) ................................. 108 Figure 4.64 Mount Polley copper grade distribution Histogram (size fraction B) ............... 108 Figure 4.65 Myra Falls zinc grade distribution (size fraction C) .......................................... 109 Figure 4.66 Myra Falls zinc grade distribution histogram (size fraction C) ......................... 109 Figure 4.67 Ideal copper grade-recovery curves for Brenda (all size fractions) .................. 110 Figure 4.68 Ideal copper grade-recovery for Copper Mountain (all size fractions) ............. 111 Figure 4.69 Ideal copper grade-recovery curves for Mount Polley (all size fraction) .......... 112 Figure 4.70 Ideal grade-recovery curves for Myra Falls (all size fractions) ........................ 112 Figure 4.71 Gold grade distribution histogram ..................................................................... 114 Figure 4.72 Ideal gold grade-recovery curve ........................................................................ 114 Figure 4.73 Grade-recovery curves for all size fractions for Brenda and their CH values... 115 Figure 4.74 Grade-recovery curves for all size fractions for Copper Mountain and their CH values .................................................................................................................................... 116 xx  Figure 4.75 Grade-recovery curves for all size fractions for Mount Polley and their CH values............................................................................................................................................... 116 Figure 4.76 Grade-recovery curves for 4 size fractions for Myra Falls and their CH values117 Figure 5.1 Optical recovery of copper – Brenda Size A (-25+19 mm) ................................ 120 Figure 5.2 Optical recovery of copper – Brenda Size B (-37.5+25 mm).............................. 121 Figure 5.3 Optical recovery of copper – Brenda Size C (-50+37.5 mm).............................. 121 Figure 5.4 Optical recovery of copper – Brenda Size D (-75+50 mm) ................................ 122 Figure 5.5 EM recovery of copper – sensor for Benda-A (-25+19 mm) .............................. 123 Figure 5.6 EM recovery of copper – sensor for Benda-B (-37.5+25 mm) ........................... 124 Figure 5.7 EM recovery of copper – sensor for Benda-D (-75+50 mm) .............................. 125 Figure 5.8 XRT recovery of copper – Brenda Size A (-25+19 mm) .................................... 127 Figure 5.9 XRT recovery of copper – Brenda Size B (-37.5+25 mm) ................................. 127 Figure 5.10 XRT recovery of copper – Brenda Size C (-50+37.5 mm) ............................... 128 Figure 5.11 XRT recovery of copper – Brenda Size D (-75+50 mm) .................................. 128 Figure 5.12 XRF Copper Recovery, Brenda A – Single vs. Multivariate Analysis ............. 130 Figure 5.13 XRF Copper Recovery, Brenda B – Single vs. Multivariate Analysis ............. 132 Figure 5.14 XRF Copper Recovery, Brenda C – Single vs. Multivariate Analysis ............. 134 Figure 5.15 XRF Copper Recovery, Brenda D – Single vs. Multivariate Analysis ............. 135 Figure 6.1 Optical recovery of copper – Copper Mountain Size A (-25+19 mm)................ 139 Figure 6.2 Optical recovery of copper – Copper Mountain Size B (-37.5+25 mm) ............. 140 Figure 6.3 Optical recovery of copper – Copper Mountain Size C (-50+37.5 mm) ............. 140 Figure 6.4 Optical recovery of copper – Copper Mountain Size D (-75+50 mm)................ 141 Figure 6.5 EM recovery of copper – Copper Mountain-D (-75+50 mm) ............................. 142 xxi  Figure 6.6 XRT recovery of copper – Copper Mountain Size A (-25+19 mm) ................... 143 Figure 6.7 XRT recovery of copper – Copper Mountain Size B (-37.5+25 mm) ................ 144 Figure 6.8 XRT recovery of copper – Copper Mountain Size C (-50+37.5 mm) ................ 144 Figure 6.9 XRT recovery of copper – Copper Mountain Size D (-75+50 mm) ................... 145 Figure 6.10 XRF Copper Recovery, Copper Mountain A – Single vs. Multivariate Analysis............................................................................................................................................... 147 Figure 6.11 XRF Copper Recovery, Copper Mountain B – Single vs. Multivariate Analysis............................................................................................................................................... 149 Figure 6.12 XRF Copper Recovery, Copper Mountain C – Single vs. Multivariate Analysis............................................................................................................................................... 151 Figure 6.13 XRF Copper Recovery, Copper Mountain D – Single vs. Multivariate Analysis............................................................................................................................................... 153 Figure 7.1 Optical recovery of copper – Mount Polley Size A (-25+19 mm) ...................... 157 Figure 7.2 Optical recovery of copper – Mount Polley Size B (-37.5+25 mm) ................... 158 Figure 7.3 Optical recovery of copper – Mount Polley Size C (-50+37.5 mm) ................... 158 Figure 7.4 Optical recovery of copper – Mount Polley Size D (-75+50 mm) ...................... 159 Figure 7.5 EM recovery of copper – Mount Polley-A (-25+19 mm) ................................... 160 Figure 7.6 EM recovery of copper – Mount Polley-C (-50+37.5 mm) ................................ 162 Figure 7.7 EM recovery of copper – Mount Polley-D (-75+50 mm) ................................... 164 Figure 7.8 XRT Spectral recovery of copper – Mount Polley Size A (-25+19 mm) ............ 168 Figure 7.9 XRT Spectral recovery of copper – Mount Polley Size B (-37.5+25 mm) ......... 169 Figure 7.10 XRT Spectral recovery of copper – Mount Polley Size C (-50+37.5 mm) ....... 169 Figure 7.11 XRT Spectral recovery of copper – Mount Polley Size D (-75+50 mm) .......... 170 xxii  Figure 7.12 Mount Polley Size A – XRF Copper Recovery – Single vs. Multivariate Analysis............................................................................................................................................... 171 Figure 7.13 Mount Polley Size B – XRF Copper Recovery – Single vs. Multivariate Analysis............................................................................................................................................... 172 Figure 7.14 Mount Polley Size C – XRF Copper Recovery – Single vs. Multivariate Analysis............................................................................................................................................... 175 Figure 7.15 Mount Polley Size D – XRF Copper Recovery – Single vs. Multivariate Analysis............................................................................................................................................... 176 Figure 8.1 High-grade rock #45 with 34% zinc .................................................................... 179 Figure 8.2 Low-grade rock #95 with 36% iron and 0.3% zinc ............................................. 180 Figure 8.3 Optical recovery of Zinc – Myra Falls Size A (-25+19 mm) .............................. 180 Figure 8.4 Optical recovery of Zinc – Myra Falls Size B (-37.5+25 mm) ........................... 181 Figure 8.5 Optical recovery of Zinc – Myra Falls Size C (-50+37.5 mm) ........................... 181 Figure 8.6 Optical recovery of Zinc – Myra Falls Size D (-75+50 mm) .............................. 182 Figure 8.7 XRT recovery of copper – Myra Falls Size A (-25+19 mm) .............................. 183 Figure 8.8 XRT recovery of copper – Myra Falls Size B (-37.5+25 mm) ........................... 184 Figure 8.9 XRT recovery of copper – Myra Falls Size C (-50+37.5 mm) ........................... 185 Figure 8.10 XRT recovery of copper – Myra Falls Size D (-75+50 mm) ............................ 185 Figure 8.11 Myra Falls Size A – XRF Copper Recovery – Single vs. Multivariate Analysis............................................................................................................................................... 186 Figure 8.12 Myra Falls Size B – XRF Copper Recovery – Single vs. Multivariate Analysis............................................................................................................................................... 189 Figure 8.13 Myra Falls Size C – XRF Copper Recovery – Single vs. Multivariate Analysis............................................................................................................................................... 191 xxiii  Figure 8.14 Myra Falls Size D – XRF Copper Recovery – Single vs. Multivariate Analysis............................................................................................................................................... 192 Figure 9.1 Grade-recovery curves based on the EM sensor for the gold sample ................. 196 Figure 9.2 Grade-recovery outcome of dynamic XRT sorting of gold-bearing material ..... 197 Figure 9.3 Comparative arsenic grade-recovery curves ....................................................... 200 Figure 9.4 Gold grade-recovery curves based on arsenic XRF readings .............................. 201 Figure 9.5 Comparative iron grade-recovery curves ............................................................ 202 Figure 9.6 Gold grade-recovery curves based on iron XRF readings ................................... 202 Figure 9.7 Comparative sulfur grade-recovery curves ......................................................... 203 Figure 9.8 Gold grade-recovery curves based on sulfur XRF readings ................................ 203 Figure 9.9 Gold grade-recovery curves based on silica XRF readings ................................. 204 Figure 9.10 Multivariate regression Gold grade-recovery curves with and without interaction effects .................................................................................................................................... 208   xxiv  ACKNOWLEDGEMENTS I would like to extend my appreciations to my thesis supervisor, Prof. Bern Klein of Mining Engineering Department of University of British Columbia, for his continuous support of me and the project by providing insightful advice and appropriate direction wherever necessary. I would also like to thank Matthew Kowalczyk, of then Tomra, for introducing me to the world of sensor-based ore sorting and allowing me to use Tomra’s facility in Surrey, British Columbia, to conduct bench-scale amenability tests using their available sensors. Similarly, I would like to thank Dr. Andrew Bamber of MineSense Technologies Inc. for allowing me to use their facility in Vancouver, British Columbia, to conduct other parts of my research. I would like to express my sincere gratitude to Dr. Adrian Dance whose intelligent and invaluable comments and advice not only contributed to better conducting this research but also strengthened my critical thinking approach to the field of mineral processing. My deepest appreciation to Natural Sciences and Engineering Research Council of Canada (NSERC) and Mitacs for financially assisting with this research. A financial assistance without which examining the results to this detail would have not be feasible. I would also like to thank my fellow students who helped me during this research by assisting with sample preparation or data analysis, in order of appearance, Tugba Cebeci, Xu Bo, Hiten Sulakh, Emilio Adriano and Huaizhe Li. Last but not least, my deepest gratitude toward my parents and family whose life-long support was an undeniable contributor to this success. I would also like to extend my appreciation to Dr. J. M. Cheung for standing with me throughout the years and her continuous encouragement for me to complete this degree. Ultimately, I would like to thank all my friends who had various roles in helping me accomplish this, Brandon Nichols, Densie Nunes, Hassan Ghaffari and others.   xxv  DEDICATION   To my beloved mother, whose untimely demise left us with an irreparable void.  1  1. INTRODUCTION Over the past few years, sensor-based ore sorting technology has been gaining more attention. These days, major mining companies are interested in implementing this technology to overcome the challenge of ever-decreasing ore grades and higher mining and processing costs. Mining companies are not only dealing with both those challenges, but also environmental regulations are also becoming tighter and mining companies are responsible for more sustainable and cleaner mining practices. Therefore, to tackle all these challenges, mining companies have finally started to embrace this not-so-new technology and investigate the implementation of this technology at their various operations. While there are many sorting companies that are continuously improving the algorithms, increasing the throughput and overcoming other major setbacks of this technology, this work focuses on the practical side of sensor-based ore sorting and examines the amenability of low-grade, waste rock stockpiles to sensor-based ore sorting. This work looks at low-grade samples from four base metal stockpiles, as well one from a gold stockpile. These samples are examined using a range of sensors, including optical (OPT), electromagnetic (EM), X-Ray Transmission (XRT) and X-Ray Fluorescence (XRF). The literature review for this thesis briefly introduces the sensing technologies used in this work. Following introduction of each sensor, the major companies that provide those sorters are introduced. The information provided for each company was based on available information online and through informal meetings. Therefore the depths of the information may not be the same for all companies. In addition, since there is little data available of actual sorting operations, the literature review chapter summarizes the few operational data published to-date. These operations only included those that had performed testwork in larger than lab-scale tests (i.e. pilot plant and larger). The objective of this research is to investigate the applicability of sensor-based ore sorting to low-grade stockpiles for base metal and gold operations. By studying these five operations against the four different sensors, the effort is to one, show if sensor-based ore sorting technique can be applied to such stockpiles, and two, which types of deposits respond better to which type of sensor. 2  Another topic that this research tackled was the effects of heterogeneity of the sample on sorting outcomes. To do that, the Constitution Heterogeneity (CH) concept of Pierre Gy was explained and discussed. Thereafter, each sample was studied for its CH values and these values were then correlated with their potential for ideal sorting. This research also looks at the effects of particle size on sorting efficacy and amenability. For this purpose, four size fractions from each mine is analysed. For the gold samples however, due to time and financial limitations, only one size fraction was studied. Having defined the objectives, it is important to note that the focus of this study was neither on developing nor improving sorting algorithms, but rather to have an impartial look at the sensor’s capabilities and characteristics of rocks that would make them amenable, or not amenable, to each specific sensor. During this process, where there were algorithms available (for the case of XRT and OPT) they were used to analyse the data, and if there was no algorithm available, either a Single- or a Multivariate Linear Regression (MLR) analysis was performed (XRF and EM). It is also noteworthy that although economics of such studies are of utmost importance, it was not part of the scale of this research. In the end, the purpose of this work was to not only familiarize the industry with sorting concept, but also delve into significant factors that could benefit or potentially tarnish sorting outcomes. With examining a comprehensive suite of commodities, it is believed that the general guidelines presented in this work will help the mining companies to quickly screen possible sensors that could potentially benefit them and their operations.   3  2. LITERATURE REVIEW Sensor-based ore sorting is not a new concept. Although there were attempts in utilizing sensors to pre-concentrate ore as early as the Second World War (Salter & Wyatt, 1991), it was not until recently that the technology attracted significant attention by the major mining companies around the world. Sensor-based ore sorting can be applied to particles, called particle sorting, or used to sort bulk of material, which is called bulk sorting. This research only examines particle sorting and excludes all the information and sensors that can be used for bulk sorting. Sensor-based ore sorting for particles involves identification of individual particles and rejection of the particles that are identified as gangue by the sensor (Napier-Munn & Wills, 2016). Salter and Wyatt (Salter & Wyatt, 1991) identified four sub-processes for sensor-based ore sorting:  Particle presentation  Particle identification  Data analysis  Particle separation Feed (particle) preparation is a critical step in ore sorting as the particles should be in a specific size range and have surface characteristics suitable for sensing. For an efficient sorting to occur, particles need to maintain a 3:1 or 2:1 ratio in size. Also, for some sensors the surface of the particles needs to be reflective (e.g. for Optical) and clean of dust (e.g. for XRF). In addition, for a proper analysis by the sensor, each particle needs to be presented to the sensor properly. Therefore feed rate and material handling methods are of utmost importance (Wotruba, 2006). The particle identification stage is performed by using a combination of a sensor and a processing unit. Once the decision is made whether a particle is mineralized or barren, the processing unit sends a signal to the ejection unit where the particles are separated. The physical separation can be performed either by high-pressured compressed air jets, water jets or mechanical arms/flaps (Weatherwax, 2007). 4  An array of sensors can be used in a sorter machine to examine the particles. Of all the sensors available, only four of them are used and discussed in this thesis. These four sensors include Optical (OPT), Electromagnetic (EM), X-Ray Transmission (XRT), and X-Ray Fluorescence (XRF). Figure 2.1 shows a range of sensors used in this field along with their corresponding operating wavelengths. Figure 2.1 List of sensors and their corresponding wavelengths (Von Ketelhodt L. , 2009)  The technology, its evolution, principles, benefits and shortcomings have been extensively discussed in various literature (Salter & Wyatt, 1991) (Robben, Wotruba, Robben, von Ketelhodt, & Kowalczyk, 2013), (Knapp, Neubert, Schropp, & Wotruba, 2014), therefore this chapter confines its content to briefly explaining the technologies that were used in these studies, a brief description of major companies that provide each of those technologies and a summary of case studies that were conducted in a larger scale than static lab tests and have been published so far. Below is a short list of the companies presented in this thesis and the sensors they provide.   5  Table 2.1 Summary of companies and the sensors they provide Company Optical EM XRT XRF Comex ■  ■  MineSense  ■  ■ Rados    ■ Steinert ■ ■ ■ ■ Tomra ■ ■ ■   Optical Sensor (OPT) In optical sorting systems, ore particles are generally tested by a digital line-scan camera. The separation decisions are typically made based on differences in color, reflectance, or transparency between the particles (Bamber, 2008). Optical sorting is one of the most cost effective sensor-base sorting systems, however, this method is confronted with several challenges as most minerals are optically complex and therefore require much more sophisticated illumination sources than just the visible light spectrum. In such instances, other light sources such as Ultraviolet (UV), fluorescent or Near Infrared (NIR) excitation can be utilized. Another challenge with optical sensors is that this technique can only scan the surface of the rock and has no penetrative capabilities. Thus, like any other surficial technology, optical sorters are prone to blind spots (Mazhary & Klein, 2015). This shortcoming is often overcome through installation of multiple cameras and sensors to cover two or more sides of the rocks. Although this partially overcomes that shortcoming, the technology is still incapable of detecting minerals if they are trapped inside a rock. The Near Infrared (NIR) sensor is another senor that falls under the category of Optical sensors. This sensor works based on reflection or absorption of near infrared light. The NIR sensor was used at the beginning of this study, however due to its inability to distinguish between gangue and valuable rocks for the samples in this research, the experiments were not continued.  Comex Optical Sorters Comex is a company based in Poland and headquartered in Norway. The company provides particle separation and classification solutions. Comex has running installations in Australia, Austria, China, Korea, Norway and Turkey. These installations are mostly optical sorting and in one case it is a combination of Optical and X-ray transmission sorting. 6  The OSX sensor series are the standard optical (visible light) sorters. The sorters can incorporate infrared cameras (ISX) as well as ultraviolet (USX). They operate on a range of belt widths from 600 to 1500 mm and can handle belt velocities of 3-5 m/s leading to throughputs of 5 to 250 tph based on size and density of the materials being sorted. They can identify materials by their color, size, shape, structure and texture. The minimum size that can be sorted is 10-15 mm depending on the model. Ejection mechanisms includes air jets for particles below 100 mm and pneumatic flaps for particles above 100 mm (up to 300 mm). Figure 2.2 Components of a Comex OSX Sorter  The VSX sensor series are another type of optical sorters by Comex and they are different from the OSX series in that they take advantage of two additional mirrors that provide images from three sides of the rock. The name comes from the specific conveyor design that consists of two belt conveyors arranged in the shape of a V (Figure 2.3). Material is delivered to the analyzing unit in a single line and one particle at a time. Although sensors have a full view of the rock, the throughput is quite low and is in the range of 5-40 tph depending on size and density. The unit can handle particles between 20 and 300 mm. With the VSX model, it is possible to have a full and detailed 3D scan of each particle and it can be equipped with all the sensors 7  mentioned earlier. The VSX series have no limitations when it comes to number of possible sorting fractions. Optical sensors by Comex have been used in the following applications.  Quartz  Limestone  Dolomite  Feldspar  Iron ore (in one case optical sorting + XRT)  Fluorite  Gemstone sorting Figure 2.3 Components of a Comex VSX Sorter    Steinert Optical Sorters Steinert’s optical sorters can be installed in the KSS models where multiple sensors can be utilized at the same time. Steinert optical sorters come in three different types, the color sensors where particles are separated based on their visual differences, the laser where separation is based on the laser diffraction, and the NIR Hyper Spectral Imaging (HSI) sensors. Steinert suggest that the color/laser combination best suits precious metals and industrial mineral applications.  Tomra Optical Sorters Tomra Sorting is headquartered in Germany with test centers in Germany and Australia. Tomra sorting machines come in different series: Feed Material V-shaped conveyor Analyzer Unit Separation Unit 8   PRO Series: The processing series consists of sorting machines based on free fall (chute principle).  COM Series: The Common belt series consists of general-purpose belt sorters providing solutions for heterogeneous feed or moisture content.  GEM Series: The GEM series are used for gemstone and diamond applications. These categories differ from each other in terms of positioning of the sensor, whether on the belt (COM), chute (PRO) or the application (gem stone vs. minerals) (GEM). Each of these sorters can vary based on the application and capacity and can be equipped with a variety of sensors. The COLOR (COL) sensors consist of a line-scan camera with high resolution and precise color detection. Various surficial characteristics of rocks including size, shape, brightness and color can be processed simultaneously. Tomra also provides the COLOR DUAL technology in which two cameras scan both sides of the rock, covering over 80% of the surface of the rock to decrease the chances of blind spots. The PRO series can also incorporate the NIR sensors to sort certain gold deposits depending on gold association with other rock such as quartz. Depending on the machine these sensors are used in, the particle size and throughput will vary. Particle size can be in the range of 20 to 120 mm resulting in a throughput up to 100 tph. Table 2.2 summarizes the types and standard applications of Tomra’s optical sorters.   9  Table 2.2 Applications of Tomra’s optical sorters Minerals Standard Applications Models Standard Premium COLOR DUAL NIR COLOR DUAL NIR Magnesite Waste rock reduction / Product Pre-concentration █      Calcite Production of high quality products / Contaminant removal █ █ █    Talc Production of high quality products / Contaminant removal █  █    Gold Waste rock reduction / Pre-concentration     █ █ Quartz Production of high quality products / Contaminant removal █ █  █   Phosphate Waste rock reduction / Pre-concentration █ █     Borate Production of high quality products / Contaminant removal / Pre-concentration █  █   █ Feldspar Waste rock reduction / Pre-concentration █ █      Electro-Magnetic Induction Sensor (EM) The electromagnetic sensor is a sensor with penetrative capabilities. This capability allows the sensor to “scan” the bulk of the particle eliminating chances of blind spots as is the case with surficial sensors such as optical and XRF. The magnetic induction principles is similar to those of electric conductivity sensors. The difference is that in an induction sensor, the disturbance in the magnetic field is measured. An oscillating magnetic field is generated through a coil powered by a high frequency AC signal. When a conductive particle is placed inside the magnetic field, an electric current is generated inside the rock because of the coil’s oscillating magnetic field. The generated electrical current in a conductive rock in turn will generate a second magnetic field that has a certain magnitude and is different from the original magnetic field by a certain phase angle. The change in magnitude and phase angle compared to a reference coil is then measured and is taken as the basis for material identification. Although it is believed that this technique would be applicable for conductive sulphide minerals of VMS origin (Klein & Bamber, TBD), the current study does not necessarily confirm this claim. Of various rock types from porphyry to VMS and from low-grade copper minerals to very high-grade sphalerite, the inductive sensor failed to distinguish between 10  gangue and valuable rocks. The inability of such sensors in distinguishing the rocks is typically associated with the minerals’ matrix within the rock. To understand the electromagnetic behaviour of rocks, it is necessary to have an understanding of magnetism and magnetic susceptibility properties of different minerals. Magnetic materials are those whose magnetic field increases when placed in an external magnetic field, and are all called either ferromagnetic, ferrimagnetic or antiferromagnetic materials. Those materials whose magnetic field is decreased in similar circumstances are identified as non-magnetic and are called either para or diamagnetic. With that introduction, Table 2.3 lists the dominant base metal minerals available in all samples examined, along with their magnetic susceptibility values. Table 2.3 Magnetic properties of dominant sulphides available in the tested material (Pearce, Pattrick, & Vaughan, 2006) Mineral Magnetic Category Magnetic susceptibility in k×10−6 cgs (except pyrite) Bornite Paramagnetic n/a Chalcocite Diamagnetic n/a Chalcopyrite Antiferromagnetic K~30 Galena Diamagnetic K=-3 to +84 Molybdenite Diamagnetic K=-63 to -77 Pyrite Diamagnetic K=107 Sphalerite Diamagnetic K=-25 to -60 Therefore, of the major minerals in the samples from all these mines, only chalcopyrite is magnetic. It is important to reiterate that magnetic field in para and diamagnetic material weakens as they are placed in an external magnetic field. This will interact with the increased magnetic field of chalcopyrite and the extent of this interaction depends on the constituents of each rock and the matrix that these constituents form inside the rock. Therefore, pinpointing the exact reason why the EM sensor does not work for certain deposits is impossible. For a detailed list of minerals and their magnetism the reader is advised to study the paper by Caroline Pearce (Pearce, Pattrick, & Vaughan, 2006) 11   MineSense Technologies EM Sorters MineSense is a start-up Canadian company that is mainly focusing on bulk sorting applications of sensors. The company’s main focus is on two sensors, EM and XRF, and are also developing applications for Laser Induced Breakdown Spectroscopy (LIBS). The High Frequency Electromagnetic Spectroscopy, or HFEMS, sensor, is based on the EM sensor and uses electromagnetic waves in the frequency range of 100 kHz to 1400 kHz. This range places the sensor in the short-to-medium frequency range therefore HFEMS would be a misnomer. Despite the bulk sensing capabilities of this sensor, only material with a magnetic or conductive susceptibility would respond to such excitation methods. As a result, an assumption regarding a specific element’s behavior can never be made. It was observed that some high-grade copper particles did not show any response to this type of sensor while some others did. SortOre, one of the MineSense products, is the particle sorting test unit. SortOre utilizes two sensors, EM and XRF, simultaneously to provide real-time information on the particles on conveyor belt. The data can also be used to assess semi-bulk and bulk sorting (with the use of diverters) solutions.  Steinert Induction Sorters Steirnet Induction Sorting System, or ISS, can be used to sort a range of material as long as they have electrical or magnetic properties. These systems can handle particles in the size range of 1 to 200 mm and at and at belt widths of 600 to 3000 mm.  Tomra EM Sorters Tomra Sorting is another company that provides electromagnetic sorting machines. The EM sensor is mounted on the COM (common belt) series which can handle a particle size range of 5 to 30 mm and a throughput capacity of up to 100 tph.  X-Ray Transmission (XRT) XRT sensors are similar to the x-ray scanner available at the airports. The sensor utilizes a dual energy high intensity x-ray beam that penetrates through material and can differentiate material based on their atomic density. While the penetrative feature is a great advantage, this sensor as well comes with its shortcomings. Failure to distinguish between elements within 2 atomic 12  numbers from each other and non-elemental specificity are the major issues with this sensor. These shortcomings can lead to low-value, high-atomic-density particles (e.g. pyrite) report to the concentrate and negatively affect the sort. However, if, as an example, gold is associated with dense (or for the reverse case with light) material, an XRT sensor can prove effective.  In terms of difference in density of ore and waste material, a minimum threshold of 0.3-0.4 kg/L should exist in order for a good separation. For example a pure product density of 4.5 kg/L and pure waste density of 2.7 kg/L will lead to very good separation results while a density of 3.1 kg/L and 2.7 kg/L will lead to very poor sorting efficiencies, if at all possible.  This type of sorters is well suited for base metal applications. For precious metals, due to their very low grades, an association with another element must be established.  Comex XRT Sorters Comex XRT sorters are introduced as the CXR series. The sorter can also incorporate an optical (OCXR) or an infrared (ICXR) sensor, as well. The sorters also have the flexibility of adding other sensors such as visible light or ultraviolet (UV) tailored to the clients’ needs.  The CXR sorters can accommodate belt speeds of up to 5 m/s. The throughput ultimately depends on the material top size and density but as a reference point the following throughputs were achieved for an iron ore sample with an average density of 3.8 kg/L:  +8 - 20mm - 20t/h +20 - 50 mm - 40 t/h +50 - 100 mm - 80-90 t/h +100 - 250 mm - 150- 200 t/h These values may vary depending on application requirements. In general, material must be supplied to the belt conveyor on a uniform layer with small gaps between particles. It is recommended that feed size ratio of 2.5 to 3 to be followed to achieve better sorting efficiencies. The ejection mechanism can be either compressed air or mechanical flaps based on the particle size. 13  The minimum size of particles for CXR series was reported as 8mm. However, no information was available regarding the top size the sorters can handle. If all conditions are met, sorters can achieve an ejecting efficiency of up to 99%. CXR sorters have been used in the following applications:  Iron ore (in one case + optical sorting)  Fluorite  Barite  Chrome ore  Manganese ore  Tungsten ore  Coal sorting  Gemstone sorting Figure 2.4 Components of a Comex OCXR Sorter  Comex uses a common technique for detection and recognition of objects called Pattern Recognition. Comex uses pattern matching to improve the efficiency of their sorters. Below are a few operations that used Comex sorters with and without the pattern matching algorithm (Kolacz, 2014). 14  Table 2.4 Results from iron ore (type 1) sorting through X-ray Transmission (XRT) Stream Mass Recovery (%) Fe Content (%) S Content (%) Fe Recovery (%) S Recovery (%) Waste 18.17 21.45 0.165 7.0 20.1 Product 81.83 63.6 0.146 93.0 79.9 Feed 100 55.96 0.149 100 100 Table 2.5 Results from iron ore (type 1) sorting using both an X-ray Transmission (XRT) as well as an optical sensor utilizing pattern recognition techniques Stream Mass Recovery (%) Fe Content (%) S Content (%) Fe Recovery (%) S Recovery (%) Waste 41.87 47.01 0.31 35.2 85.8 Product 58.13 62.4 0.037 64.8 14.2 Feed 100 55.96 0.149 100 100 In the following example, the objective was to upgrade iron to above 63%. A task that was not possible using the XRT sensor with conventional algorithms. Table 2.6 Sorting iron ore (type 2) using XRT Stream Mass Recovery (%) Fe Content (%) Waste 9.5 30.1 Product 90.5 61.7 Feed 100 58.8 Table 2.7 Sorting iron ore (type 2) using XRT utilizing pattern recognition Stream Mass Recovery (%) Fe Content (%) Waste 21. 7 41.2 Product 78.3 63.6 Feed 100 58.8 Table 2.8 Coal sorting with conventional use of an XRT sorter Stream Mass Recovery (%) Ash Content (%) Calorific Value (LCV) (kJ/kg) Waste 26.2 37.11 14627 Product 73.8 17.57 20517 Feed 100 25.8 18034   15  Table 2.9 Coal sorting with an XRT sorter using pattern recognition Stream Mass Recovery (%) Ash Content (%) Calorific Value (LCV) (kJ/kg) Waste 32.2 47.9 11088 Product 67.8 16.71 21371 Feed 100 25.8 18057  Steinert XRT Sorters Steinert’s XRT sorters, the XSS T model, uses a Dual Energy x-ray system to identify the material based on their density. These sorters can process particles in a size range of 10 to 200 mm on belt widths of 1 to 2 meters. An example of image processing for this sensor is used in Figure 2.5. Figure 2.5 Image analysis procedure for Steinert’s XRT sorter   Tomra XRT Sorters Tomra’s XRT sensors are incorporated in the COM (common belt) series and can handle the particle size range of 8 to 150 mm. Depending on the width of the belt, these sorters can manage feed throughputs of up to 150 tph (XRT 1200) or 300 tph (XRT 2400). Belt speed is generally 16  determined after some medium-scale test runs but it is generally around 2.5 m/s. XRT sorters are replacing the old X-Ray Luminescence (XRL) originally used in diamond mining.  X-Ray Fluorescence (XRF) XRF sensors are surficial sensors capable of detecting elements with an atomic number greater than 20 (Ca). The limit of detection for XRF is fairly low, however, the lower the atomic number, the longer exposure time is necessary to detect low-grade, light elements. This exposure time proves to be the most challenging aspect of using an XRF sensor for sorting purposes. XRF works by measuring the secondary x-ray emitted from the sample after it is irradiated by a primary x-ray source. The primary x-ray should be stronger than the atom’s K or L shell binding energy. When a solid (or liquid) sample is irradiated by a sufficiently strong primary x-ray, an electron from inner orbital shells is removed. The atom stabilizes itself by replacing the vacancy with another electron from higher energy orbital shells. By regaining stability, the atom releases energy in the form of fluorescent x-ray. The released energy is characteristic of each element and therefore it is the basis for of XRF analysis (Portable Analytical Solutions, 2017).   17  Figure 2.6 Schematic of XRF detection principles (Portable Analytical Solutions, 2017)  With elemental-analysis capabilities, XRF sensors have the potential for elemental discrimination in complex ores but are historically known to be slow in detection. However, with the recent advances in computer power and more efficient and sophisticated algorithms, the computational procedures take no longer than a few milliseconds. Having said that, the XRF sorters are still considered to have low throughputs but not because of sensing time required, rather the sample presentation to the sensor. In this sorter, to ensure that particles pass through the sensor properly, the particles have to travel through specific channels. The ore feed channels vibrate at a certain frequency to ensure each particle lies along its length so that the maximum area is visible to the sensor. The vibration can also control the free fall trajectory of each particle via controlling the linear velocity. This process limits the amount of rocks that can go through a sorter at any given time and therefore reduces sorter throughput compared to other sorting technologies. A decision to keep or reject the particle is based on the thresholds determined for specific elements. Figure 2.7 shows the XRF peaks for some common elements.   18  Figure 2.7 XRF peaks for a number elements (Demographer, n.d.)   MineSense Technologies XRF Sorters As mentioned in the Electromagnetic Sorters section, MineSense uses XRF sensors in two of their products. The SortOre, which is a sorting machine, and BeltSense, which is a sensing-only device. XRF measures the grade of base metals and although it is claimed to measure precious metals as well, no limit of detection (LOD) is provided. Considering the precious metals’ grades are often too low to be detected by XRF or any other sensors, the precision of this product for measuring precious metals should be investigated. In MineSense XRF sorters laser height telemetry is used to measure the distance between the surface of the rock and the XRF sensor through which correction factors can be applied to the XRF readings.  Rados XRF Sorters Rados is a sorting company based in Russia specializing in XRF sorting. Their technology is claimed to be in operation in over 20 countries and 50 operational sites. They have overcome 19  the exposure time barrier through sophisticated algorithms. The available installations are over a range of commodities from base and precious metals to industrial minerals, uranium and Rare Earth Elements (REE). 60% of the installations are on greenstone gold deposits and the systems has detection limits of 0.1 to 0.2 g/t and recoveries of up to 90%. Generally the XRF detection limit for elements is about 0.01% to 0.1%. For copper, the detection limit is generally about 0.1% - 0.5%. Table 2.10 summarizes the sorter types Rados offers and some of the specifications. Particles are exposed to x-ray for only a matter of milliseconds (30 ms for a 30 mm particle) on their free fall. The ejection system is based on mechanical flaps due to lower power consumption costs as well as claimed higher ejection efficiencies. Table 2.10 Different types of XRF sorters provided by Rados Technical Characteristics Separator Type SXF-4-50 SXF-4-150 SXF-4-300 Size Class Ranges (mm) 10 – 60 30 - 150 60 - 300 Throughput (tph) 3 – 8 10 – 25 20 – 50 Sorting Chanel Numbers 4 4 2  Steinert XRF Sorters The Steinert XRF sorters, XSS F systems, similar to the XSS T systems come widths of 1-2 meters and can handle particles grain sizes of 10 to 200 mm.  Ore-Sorting Case Studies  Although there are claimed to be many operating sorting systems, due to various reasons including confidentiality, most of them either remain anonymous or very little of their operational data and results are available. It was not until recently that some operational data were published. Some of the most recent results are presented below.  Anglo American Platinum – UG2 Mine The notion of using a pre-concentration method at the Anglo American Platinum UG2 mine came from the inevitable grade dilution during mining. Characteristically, PGM ore bodies typically exist as 1-meter thick visible reefs of highly mineralized material. Although narrow-vain mining seems like a good option, for ease of mining and safety of workers, wider areas should be mined to make room for future mining activities. This means that a large enough 20  section of the reef that is placed between barren hanging wall and footwall should be mined. (Rule, Fouchee, & Swart, 2015). However, grade dilution associated with this method became a major issue for this company as it directly affected the productivity and hence economics of the mine (Neingo & Cawood, 2011). Moreover, additional barren extraction occurred after mechanized mining was applied, in which method machines require much bigger pathways to move in and out. Therefore, the company sought a pre-concentration method to counteract the effects of dilution. Initially, other physical separation methods such as screening and Dense Media Separation (DMS) were investigated. The separation achieved through screening was not as sharp and as effective as expected, due to the fact that a certain portion of high-grade material reported to the screen oversize. In the case of DMS, however, apart from high operating costs of the media, the main challenge was the presence of composite particles on the interface between the chromite layers and silicate, which resulted in PGM losses. (Rule, Fouchee, & Swart, 2015) Therefore, in an effort to lower the operating cost and, most importantly, increase the feed grade of PGM, a MikroSortTM optical sorting system was installed at Waterval UG2 concentrator. The optical sorter was designed to sort the rocks based on the visual differences between mineralized chromite and waste silicates, and reject the waste rock from a moving ore stream using pulsed air jet streams. However, application of this system was discontinued shortly after for economic concerns as the rejection rate was much lower when compared with the downstream production rate, and the maintenance cost was relatively high (Rule, Fouchee, & Swart, 2015). After the initial setback, a RadosTM XRF sorter was utilized. The XRF sorter analyzed and differentiated between individual rocks based on the algorithm which identified chrome, copper and nickel content, which had proven to be successful proxies for PGMs. After identification, the sorter rejected particles with grades below a set threshold value using actuators. The setup for the sorter is shown in Figure 2.8. Along with the three elements mentioned, iron and calcium were also detected.   21  Figure 2.8. RadosTM XRF ore sorter setup (Rule, Fouchee, & Swart, 2015)   Proof of Concept Plant Figure 2.9 illustrates the simplified process flowsheet of the Proof of Concept plant which included a crushing plant and four RadosTM sorters operating continuously 24 hours a day. The crusher reduced the size of the rocks to 30-80 mm according to the preliminary tests conducted. The feed preparation stage included a washing screen where fines were removed, and the surface of the larger particles were rinsed ahead of the analysis (Rule, Fouchee, & Swart, 2015). Fines washed from the surface as well as fines generated as a result of breakage during the sorting process reported to concentrate stream based on the hypothesis that fines are usually higher grade since highly mineralized ore are usually more breakable. The coarse particles were then sent to the sorter for analyses and rejection. Each machine could sort eight rocks per second per channel and each rock was analysed for five elements. Therefore, a strong information and data processing system was installed on the sorting plant to handle the large amount of data collected.   22  Figure 2.9. Simplified PFD of the Proof of Concept plant (Rule, Fouchee, & Swart, 2015)   Early Results and Findings  Mass balance of the Proof of Concept which was used in the business case is illustrated in Figure 2.10. An overall mass yield of 85.2% was achieved while 97% of the PGMs were recovered between the screening and XRF sorting processes. In addition, an increase of 15.3% was also observed in the mill PGM grade. (Rule, Fouchee, & Swart, 2015). Figure 2.10. Proof of Concept plant mass and metal balance used in business case (Rule, Fouchee, & Swart, 2015)  As mentioned earlier, Mintek used the copper content as an indication of PGM presence. Figure 2.11 displays the correlation between the measured copper content and the PGM content that was used in the sorting algorithm.   23  Figure 2.11. Correlation between PGM and copper for PGM sample considered (Rule, Fouchee, & Swart, 2015)  In the first run, 8725 individual particles were analyzed using one of the sorters. The average grade of these particles were predicted to be 1.28g/t 2PGM+Au. With only 1 false negative incident, a sorting execution accuracy of 99.93% was achieved. (Rule, Fouchee, & Swart, 2015).  As shown in Figure 2.12, the sorter managed to successfully upgrade the material received to 1.62g/t. The average rejected mass percentage was 35% (vs. 40% predicted), with a grade of 0.65g/t 2PGM+Au (vs. a predicted value of 0.59g/t 2PGM+Au) (Rule, Fouchee, & Swart, 2015).   24  Figure 2.12. First PGM deportment results from 4 channels of Sorter 1 (Rule, Fouchee, & Swart, 2015)  In conclusion, although the preliminary results are achieved with non-optimal sorter performance, the sorting outcome was impressive. These tests showed that 15% of run of mine mass could be rejected and only 3% of PGMs would be lost. Pre-concentration using the XRF sorter proved to have improved the overall PGM recovery by 17.5%, while the increase in OPEX was marginal, therefore significantly increasing overall economics of the mine. (Rule, Fouchee, & Swart, 2015)   Goldfields – Kloof Gold Mine Multiple companies around the Witwatersrand complex in South Africa have tried various ways to recover misplaced gold in the waste rock dumps, including hand sorting (Kloof Gold Mine & Buffelsfontein Gold Mine), screening for fines or even sending the whole waste rock dump back to the mill such as the case with Goldfields’ Driefontein Mine and Anglo Gold Ashanti’s Vaal Reef operations. These operations generally failed due to inefficiencies and inconsistencies associated with such operations specially hand picking the ore (Von Ketelhodt L. , 2009). In more advanced approaches, where an association between gold and other proxies was available, sensor-based ore sorting methods such as radiometric or optical was utilized. In case of Kloof Gold Mine, extensive testwork including an amenability phase with about 500 kg as well as a 6-ton bulk test was performed to examine the reproducibility of the initial testwork. 25  Various rock types were identified at the site and average grade of each was calculated (Von Ketelhodt L. , 2009). The rock types and their associated grades are summarized in Table 2.11. Table 2.11. Gold grades of various rock types Rock Types Au (g/t) VCR 14.50 Cobble 3.70 Marginal 0.55 Dolomite <0.08 Lava <0.08 Green Quartz <0.08 Grey Quartz <0.08 During the initial testwork, Ventersdorp Contact Reef (VCR) with highest grade and being the minority element was chosen to be ejected. Cobble and marginal were also included in the algorithm to be ejected along with VCR. Through a single pass sorting method with the objective of maximum VCR recovery, it was possible to recover 73% of the gold in 19% of the feed mass. While high recoveries of VCR and Marginal rocks were achieved (97% and 100%, respectively), only 10% of the cobble reef was recovered. The misplaced cobble reef accounted for 21% of total gold losses. This testwork increased the gold grade from a calculated 1.77 g/t in the feed to 6.95 g/t in the product stream. During the 6-ton bulk testwork, recovering VCR and cobble reef resulted in 60% gold recovery in 13.5% mass, improving the gold grade from 0.24 g/t in the feed to 1.06 g/t in the product stream (Von Ketelhodt L. , 2009). These promising results propelled the installation of a pilot plant. The pilot plant was installed in September 2003, and was in operation until June 2004, aiming at -75 mm +16 mm size fractions. The plant mainly operated from September 2003 to February 2004 before a series of detailed process evaluation, performance, gold recoveries and profitability of operation was performed. After this period, the plant operated for another month during June 2004 (Von Ketelhodt L. , 2009). The outline of the sorting plant and process is shown in Figure 2.13. The plant included a feed hopper, conveyors, wet screen and water circulation sump, power generator, and material handling equipment for feed product and waste.   26  Figure 2.13. Kloof Gold Mine optical sorter flowsheet  Figure 2.14. Left, sorted product, right, rejected waste  During the operation of the pilot plant, nearly 110,000 tons of materials were processed. Operating cost for the plant averaged at $2.47 USD/t and gold in 2004 was experiencing the low price of $380 USD. Unfortunately due to low gold prices this project was deemed uneconomical. This was despite the fact that grade upgrade was almost 20 fold (from 0.27 g/t in the feed to 5.7 g/t in the product). Re-analysis of the same data with 2009 gold prices and a 60% increase in operating cost showed that with higher gold prices, the same project could have been profitable and although simplistic, it showed the great potential for sorting waste 27  rock dumps at Kloof Gold Mine (Von Ketelhodt L. , 2009). The following tables summarize operating results and economic analysis of the project. Table 2.12. Kloof Gold Mine sorting operating data – tonnage and availability Period Days of Operation Feed Rate (tph) Total Throughput (ton) Conc. (ton) Slimes product (ton) Total product (ton) Total waste (ton) Available Hours per day Availability (hours) Total Downtime Overall Availability (%) Sep/Oct ‘03 24 67 12.526 743 251 994 11.532 366 186 180 50.85% Nov/Dec ‘03 32 68 27.657 1.428 553 1.981 25.676 622 406 216 65.29% Jan ‘04 22 83 29.185 1.442 584 2.026 27.159 446 353 93 79.08% Feb ‘04 17 74 20.995 530 420 950 20.045 348 283 65 81.44% Jun ‘04 13 82 19.321 522 386 908 18.412 284 236 48 83.10% Total 108 75 109.683 4.665 2.194 6.859 102.824 2066 1464 602 70.88% Table 2.13. Kloof Gold Mine sorting operating data – grades and recovery Period Head Grade (g/t) Conc. Grade (g/t) Slimes Grade (g/t) Total Product Grade (g/t) Tailings Grade (g/t) Recovery (%) Yield (%) Sep/Oct 2003 0.32 3.06 1.41 2.65 0.12 69.55% 7.93% Nov/Dec 2003 0.30 3.42 1.40 2.86 0.11 71.28% 7.16% Jan 2004 0.29 3.40 1.40 2.82 0.10 71.72% 6.94% Feb 2004 0.25 5.00 1.40 3.41 0.10 70.33% 4.53% Jun 2004 0.27 5.76 1.50 3.89 0.08 68.45% 2.78% Total 0.29 4.13 1.42 3.12 0.10 70.27% 5.87% Table 2.14. Kloof Gold Mine sorting OPEX Cost structure  USD Depreciation (60 months) ($/day) 846 Finance cost ($/day) 402 Operating Cost (SPH) ($/day) 1.633 Accommodation and S&T ($/day) 70 Kloof processing/transport ($/day) 415 Salaries ($/day) 204 Assays ($/day) 114 Maintenance ($/day) 379 Hire of ablution ($/day) 3 Hire of office ($/day) 30 Total ($/day) 4.095    Total cost per ton feed ($/t) 2.47 Total cost per day ($/day) 4.095 Total cost per month ($/month) 81.902   28  Table 2.15. Kloof Gold Mine sorting economic outcome Gold price June 2004 $/ounce 380 R/$ exchange rate R/$ 6.6 Gold Price R/g 81 Feed rate (t/hr) 82 Mass percent to sorter concentrate (%) 2.78 Head grade of feed (g/t) 0.27 Sorter concentrate grade (g/t) 5.7 Fines grad (g/t) 1.5 Overall concentrate grade (g/t) 3.9    Value of product per month ($/month) 72.474 Total costs per month ($/month) 81.902 Loss per month ($/month) (9.428) In 2009, after the rise in the gold prices, Goldfields decided to conduct a new series of sorting tests not only to assess the economics of the project but also to confirm the data from the pilot plant back in 2003/2004 (Von Ketelhodt, Falcon, & Falcon, 2011). For the initial test, a few samples of fine dark and light Reef, Lava waste and quartz waste were scanned under the optical sorter. These initial tests were performed to develop a sorting algorithm. After these tests, 45 tons of Surface Rock Dump (SRD) was separated into two size fractions of -78+50 mm and -50+20 mm and were fed to the sorter at different settings. These settings were based on the color thresholds defined for the Reef material and aimed at a high, medium and low mass pulls of 30%, 15% and 8%, respectively (Von Ketelhodt, Falcon, & Falcon, 2011). As expected, higher grades and lower recoveries were achieved at lowest mass pull (8.3% for the coarse and 11.6% for the fines fraction). The highest recoveries achieved were 82.3% for the fines and 68.7% for the coarse fraction (Von Ketelhodt, Falcon, & Falcon, 2011). The details of the tests are summarized in Table 2.16 below. Table 2.16. 2009 Kloof bulk test work results (Von Ketelhodt, Falcon, & Falcon, 2011) Feed Size Sorter Setting Feed Mass [kg] Conc. Mass [kg] Tails Mass [kg] Conc. Mass Pull Calc. Head Grade [g/t] Conc. Grade [g/t] Tails Grade [g/t] Au Recovery -78+50mm Low Mass pull 11,213 935 10,278 8.3% 0.19 1.45 <0.08 62.2% Medium Mass pull 9,749 1,139 8,430 13.5% 0.18 0.66 0.10 50.8% High Mass pull 12,165 3,397 8,768 27.9% 0.33 0.82 0.15 68.7% -50+20mm Low Mass pull 4,186 487 3,699 11.6% 0.43 2.53 0.16 68.2% Medium Mass pull 3,052 551 2,501 18.1% 0.26 1.07 <0.08 74.6% High Mass pull 3,115 923 2,192 29.6% 0.54 1.49 0.14 82.3% 29  Figure 2.15 below shows the grade-recovery curves of the three runs for each of the two size fractions. Figure 2.15. Au grade and recovery vs. mass-pull (Von Ketelhodt, Falcon, & Falcon, 2011)  These results were not only in-line with the pilot plant production data, but also showed that the coarse size fraction contained less gold. After a comprehensive grade and size distribution analysis, Goldfields decided that the +50 mm size range does not contain enough gold and therefore the focus was placed on -50 +20 mm range (Von Ketelhodt, Falcon, & Falcon, 2011). Based on von Ketelhodt et al. a full size plant was installed at the Kloof Gold Mine and it was during the commissioning phase, however, there is no new information available as what the performance and sorting recovery looked like (Von Ketelhodt, Falcon, & Falcon, 2011).  Coeur Operations – Kensington Mine Coeur Operations saw a slow start to their Kensington Mine in 2011. Due to such a difficult start, the plant did not manage to reach target production values, which resulted in unexpectedly high costs per ounce of gold. After overcoming these hurdles in 2012, Coeur Operations faced another setback. The gold grades were much lower than what they had hoped 30  for, and with the plant already running at above its nameplate capacity, the only option for Coeur Operations was to find higher grade gold resources. Although these efforts eventually paid off, high grade gold resources are depletable and other strategies needed to be investigated (Lasley, 2014). With 15% of the material being rejected between mill and the flotation circuit, some high grade material were inevitably sent to the waste pile. Therefore Coeur Operations focused their attention on ore sorting techniques. They investigated the use of a Tomra XRT sorter to identify and separate these high grade pebbles. At the end of 2013, a pilot plant XRT sorting test was conducted at Kensington Mine. The pilot plant managed to recover 65% of the gold in high grade pebbles in only 10% of the mass and generating a stockpile of 0.3 oz./t material from an initially waste stream of 0.04 oz./t. The mill already had a high recovery rate of 96% but taking advantage of the sorter, this recovery can be increased further to 98%. With a head grade of 0.2 oz./t and 2000 tons per day mill capacity, the sorter can potentially recover 7.8 ounces of gold per day (Lasley, 2014). After the successful pilot project, a full-scale ore sorter was planned to be added in 2015 which would be fully integrated into the main processing flowsheet by 2016 (Lasley, 2014). The recent results of these full-scale sorters was not available at the time of writing this thesis.  Rocklands Mine (CuDECO Limited, 2015) Considering the high content of the native copper in the Rocklands Mine, CuDECO planned to install an ore sorter to treat the +40 mm fraction before the main processing plant. The sorter was anticipated to upgrade the copper grade in the native copper concentrate to 90%, with the rate up to 200 tph. In early 2014, German manufacturer Steinert offered an extensive bulk test program and ore-sorter test-work, significant success was achieved on the sorter operated in the Rocklands mine. Although the sorter was operating at the rate of 15-30 tph, it upgraded the copper grade to 75-90% copper in the concentrate. As a result, a full-scale ore sorter, with the capacity up to 200 tph, was purchased. The main processing plant was expected to treat the -40 mm native copper product with the throughput of 9,000 tpd, while the ore sorter would assist in processing the +40 mm size fraction in the crusher circuit. 31   Mittersill Mine (Mosser & Robben, 2014) Scheelite (CaWO4) has been extracted and produced in Mittersill, Austria since 1976. However, tailing disposal became an issue because of a limited capacity of the tailing sedimentation pond, restricted by the rugged landscape and environment effects in this area. Thus, waste rejection was necessary to upgrade the run of mine prior to the mill, to decrease the tailing discharge and also reduce the operating cost of the mill.  Sensor-based sorting using XRT technology was found to be the most applicable on the account of the fact that the atomic density of tungsten is much higher than that of surrounding rocks. After preliminary batch and bulk testing in Tomra’s test center facility, a pilot plant operation using a belt sensor sorter was installed at the Mittersill mine to handle the size 16-28 mm. After the optimization phase, the results surpassed the expectations. Therefore in 2012, to establish a larger production plant, a second sorter was installed to handle coarser material between 28 to 60 mm. After some modifications in the crushers before and after the sorting unit, the throughput was incrementally increased to 65 tph. Over the 13 weeks of pilot plant testing, the sorters managed to concentrate the run of mine feed from an average of 0.21% WO3 to anywhere between 0.34% and 0.43% WO3 in 45% to 54% of the mass. The sorting stage, including the undersize bypass resulted in recoveries averaging at 97.3% (Mosser & Robben, 2014).  Current Work Considering all the potential that sensor-based ore sorting technologies have to offer, and the suspicion and lack of attention from the industry, this research aimed to change the industry’s image and perception of sensor-based ore sorting. By focusing on low-grade and waste rock stockpiles, this research tries to show the added value that sensor-based ore sorting can bring to new, operating as well as non-operating mines. The main objective for this work is to show and prove that sensor-based ore sorting can be applied specifically to low-grade material, and more comprehensively, to all other material regardless of grade. This work tries to prove that overall, and with the right approach, this 32  technique can be applied to most deposits and stockpiles as long as there is sufficient heterogeneity within the material. This research also investigates Constitution Heterogeneity and its correlation with sortability. In addition, the correlation between size fraction and heterogeneity of the material is also studied to confirm either of the previously claimed statements with respect to that. Another important objective of this work is to demonstrate that utilizing Multivariate Linear Regression (MLR) analysis, as a more sophisticated data analysis technique, will result in better and more efficient sorting algorithms. While acknowledging that economics is of utmost importance, this research does not investigate the economic benefits of ore sorting for either of these materials. Economic analysis can be performed based on the data from this research in future works.   33  3. EXPERIMENTAL METHODS  Sampling  Sample Collection Sampling and sample representativeness of the lot is the biggest assumption in this research. Although efforts were made to minimize sampling bias, the sampling procedure comprised of hand picking quantities far too small in comparison to millions of tons of waste or low-grade stockpiles. Since following a correct sampling procedure and acquiring the right sample quantities was not feasible in this project, true sampling error of our technique cannot be identified. Some authors regard such samples as “specimens” (Petersen, Minkkinen, & Esbensen, 2005), however, in this thesis they are referred to as “sample” for simplicity. At the beginning of this research program, four mine properties agreed to provide samples for the research. The mines included Copper Mountain and Mount Polley (low-grade stockpiles) and Myra Falls and Brenda (waste rock dumps). Around 500 – 700 kg of rocks in a size range around 15 to 25 cm was collected from different locations of the stockpile/dump in an effort to maintain objectivity in sampling. Although the initial intention was to use all this material for the research, only a portion (100 – 150 kg) was eventually used due to the scale of the experiments and time and financial restriction around the project. The samples were brought back to the lab facility at the University of British Columbia (UBC) for experiments. Another set of samples were included in this project. These samples were received from an operating gold mine, which for confidentiality purposes, its name cannot be disclosed. No information on how these rocks were collected is available. The samples were low grade and around cut-off grade and were from a sulphide zone in the deposit. The samples used in the study represent a range of deposit types all of which have low metal grades. Below is a short description of each of these operations. Brenda: copper-molybdenum porphyry – waste rock Copper Mountain: copper porphyry with gold – low-grade Mount Polley: copper porphyry with gold – low-grade Myra Falls: polymetallic (Zn, Cu, Pb, with Au and Ag) massive sulphide – waste rock Gold Mine: gold ores –  selected rock samples 34   Sampling Error Due to the small size of the sample collected compared to the size of the stockpile, it was evident that the collected sample could not have been representative of the material in the stockpile. However, due to simplicity for this work, it is assumed that collected sample represents the stockpile fairly well. Nevertheless, simple calculations below show the amount of sample that should have been collected to secure a 90% precision in terms of sample representativity. When the mass of the collected sample is significantly smaller than the mass of the stockpile, Pierre Gy (Gy, 1982) states that the variance of fundamental error 𝜎𝐹𝑆𝐸2  can be expressed as 𝜎𝐹𝐸2 = (1𝑀𝑆−1𝑀𝐿) 𝑓𝑔𝑐𝑙𝑑𝑁3       Equation 3-1 Where: MS mass of the of sample (grams) ML mass of the stockpile (grams) f shape factor g particle size distribution factor c mineralogical composition factor l liberation factor dN Nominal size of fragments in the sample Although the lots where the samples for this study were collected from were of different sizes, for the sake of calculation of sampling error it is assumed that the lot had a total mass of 1 million tons. Therefore ML = 1 x1012 grams. The total sample used for each of these mines in this research was about 150 kg. Therefore MS = 150 x 103 grams. Since ML >> MS, the above equation simplifies to: 𝜎𝐹𝐸2 =𝑓𝑔𝑐𝑙𝑑𝑁3𝑀𝑆        Equation 3-2 The particle shape factor, f, ranges from 0.1 for needles to 1 for cubes, however, it is generally in the range of 0.3 to 0.5. For most ores a value of 0.5 is considered and therefore the same is used for these calculations. 35  Particle size distribution factor (or granulometric factor) is defined as the average fragment volume divided by the nominal fragment volume and it takes values between 0 and 1. While low values of g denotes a large distribution range, higher g values indicate a smaller distribution range. Since during sample collection only rocks between 15-25 cm in size were collected, the ratio of d (nominal top size) to d’ (lower size limit) is 1.67 (𝑑 𝑑′⁄ =2515⁄ =1.67). Therefore, according to Barry A. Wills (Wills, 2016) for the size ratio between 1 and 2, g can be taken equal to 0.75. Mineralogical composition factor (c) in its simplest form can be expressed as: 𝑐 =(1−𝑎)𝑎[(1 − 𝑎)𝜌𝑚 + 𝑎𝜌𝑔]      Equation 3-3 where: a fractional average mineral content of the material ρm Density of the mineral ρg Density of gangue material Assuming that the main mineral is chalcopyrite and a copper grade of 0.2% in the stockpile, density of 4.19 g/cm3 for chalcopyrite and 2.65 for gangue material, then: 𝑎 =  𝐶𝑢 + 𝐹𝑒 + 2(𝑆)𝐶𝑢∗0.2100=63.55 + 55.85 + 2(32.06)63.55∗0.2100= 0.0058 𝑐 =(1 − 𝑎)𝑎[(1 − 𝑎)𝜌𝑚 + 𝑎𝜌𝑔]=(1 − 0.0058)0.0058[(1 − 0.0058)4.19 + 0.0058 ∗ 2.65]= 716.75 𝑔𝑐𝑚3⁄  The liberation factor (l) is a dimensionless number between 0 and 1 and depends on the nominal size of the fragment as well as the size of fully liberated mineral grain. According to Pierre Gy (Gy, 1982), l can be calculated as 𝑙 = (𝑑𝑙𝑑𝑁⁄ )0.5. Despite the literature debate whether a 0.5 exponent is accurate or not, Gy’s model have proven as a useful tool and therefore the equation is used here as he defined. Since the exact liberation size (dl) was not available, it was assumed 36  that the complete liberation is achieved at 150 μm. With the assumption of nominal particle size of dN =20 cm (average 0f 15 and 25 cm), l would be equal to: 𝑙 = (𝑑𝑙𝑑𝑁⁄ )0.5= (0.015 20⁄ )0.5= 0.027 With all the variables defined and calculated, the amount of sample that needs to be collected from a 1 million ton stockpile to achieve a 95% confidence level of ±0.1% (2σ = 0.1/0.2 = 0.5, i.e. 25% error) is: 𝑀𝑆 =fgcldN3σFE2 =0.5x0.75x716.75x0.027x2030.252⁄ =  928908 gr =  928.9 kg Therefore, about 1 ton of material in the size range of 15-25 cm needed to be collected to achieve the said precision. This was not feasible due to various limitations around this research.  Material Preparation For the base metal samples, a jaw crusher was used to break down the rocks into smaller fragments with a top size of almost 75 mm. The crushed samples were then screened into 5 size fractions, -75 mm +50 mm (referred to as size D), -50 mm +37.5 mm (referred to as size C), -37.5 mm +25 mm (referred to as size B), -25 mm +19 mm (referred to as size A), and -19 mm, of which the 4 largest size fractions were used for analysis. From each size fraction, 100 pieces of rocks were then randomly selected using a riffle. Rocks then were washed, dried, weighed, and numbered for sensor analysis. The same jaw crusher was used to break the gold-bearing rocks into smaller fractions. For these samples however, 100 rocks from only one size fraction (-50 mm to +37.5 mm) were collected. Rocks were washed, dried, weighed and labeled for further analysis.  Sensor Tests  Optical Tests These tests were performed at Tomra’s facility in Surrey, British Columbia. An RGB camera was used to capture the photos of four rocks at a time. An image analysis software developed by Tomra was used and kindly provided to analyze the photos. By Tomra’s suggestion, the 37  rocks were rinsed before capturing their images. The reason behind rinsing the rocks is that a clean, wet surface intensifies the color gradients and therefore facilitates the identification by the camera. The images of the rocks as well as their final assays were then used in the image analysis algorithm to generate optical indices. These indices, as will be explained in the next chapter, associate different colors to high or low grades rocks. The rocks were then sorted based on these generated indices and the corresponding rock grades were used to create grade-recovery curves.  Electromagnetic (EM) Tests These tests were performed at MineSense Technologies’ facility. The rocks were placed on a sensing coil and the change in the magnitude and the phase compared to a reference coil was measured. Since there was no algorithm to analyse the sensor’s response at the time, a Multivariate Linear Regression (MLR) analysis approach was taken to investigate the applicability of the sensor. Since the EM sensor deemed not applicable to most of these samples, the tests for some material was skipped.  XRT Tests XRT tests were performed at Tomra’s facility in Surrey, British Columbia, on a modified dual energy x-ray airport luggage scanner. The image analysis software application was developed by Tomra and generously was made available to be used for this research. The model number of the x-ray scanner is not available. Rocks were placed in a wooden frame with a thin sheet of plastic at the bottom (where the rocks would sit). The tray was then ran under the x-ray scanner, low and high energy images of the rocks were captured and saved. These images, along with the final assays of the rocks were then used in the image analysis software to generate XRT indices and to develop a correlation between each index and the grade of the rocks. The rocks were then sorted based on these indices in a descending manner and the corresponding rock grades were used to develop grade-recovery curves. 38   XRF Tests The XRF benchtop tests were performed at MineSense Technologies Ltd. using an Olympus Delta series handheld XRF Analyzer. Depending on the size of the rocks, 2 to 4 sides of the rocks were scanned using the analyzer. In an unpublished study performed by MineSense Technologies Ltd., average of measurements of more than 4 sides (up to 16) did not deviate significantly from that of the 4 sides. Therefore, for the smallest size fraction, a two-side scan was performed and up to 4 sides for the largest fraction. The measured grades from all sides of a rock were then averaged and used as the XRF sensor’s response for that rock. One test property that was modified for these tests was the exposure time. The XRF providing companies suggest a standard x-ray exposure time of 90 seconds including a 60-second primary beam and a 30-second secondary beam. While the primary beam is a higher energy beam aiming to detect heavier elements, the secondary beam is of lower energy and detects lighter elements. However, for the base metal samples, it was determined that with dropping the secondary beam, which is only used to increase counts on light elements, and decreasing the exposure time of the primary beam to only 5 seconds, the device would still be able to detect copper levels of as low as 0.05% reliably. For the gold-bearing samples however, a different approach was taken in terms of exposure times. For these measurements the exposure times for each beam were set at 5 seconds totalling an exposure time of 10 seconds per measurement. Otherwise, similarly to the base metal samples, each rock was measured on 4 sides and an average value was taken as the sensor’s response. It is important to mention that the static tests are different in essence compared to the actual sorting scenario in that they take advantage of longer exposure time but, in this instance, a less sophisticated detection/identification algorithm is used. While the higher exposure time yields better detection on static test, this is compensated by more efficient algorithms in actual sorting processes. Longer exposure times (10 seconds) were tested to account for this compensation. 39   Near Infrared (NIR) Tests NIR tests were performed at Tomra’s facility but since the early tests proved that material was not amenable to this type of sensor, further tests were not performed and hence the results are not presented in this thesis.  Data Processing  Constitution Heterogeneity The concept of Constitution Heterogeneity (CH) was derived from Pierre Gy’s sampling theory explained in the book Pierre Gy's Sampling Theory and Sampling Practice by Francis F. Pitard (Pitard, 1993). Constitution Heterogeneity has been previously shown to indicate the potential for ore sorting (Mazhary & Klein, 2015). In simple terms, while high heterogeneity values suggest good potential for sorting, samples with low heterogeneity values require more investigation to determine whether sorting could be an upgrading solution. The following formula was used to calculate CH values for different sample sets. 𝐶𝐻𝐿 = 𝑠2(ℎ𝑖) =1𝑁𝐹∑ ℎ𝑖2𝑖 =1𝑁𝐹∑(𝑎𝑖−𝑎𝐿)2𝑎𝐿2𝑖 .𝑀𝑖2𝑀𝐿2    Equation 3-4 When talking about heterogeneity, it is important to note that there is the heterogeneity of the desired element, base metals or gold in this instance, and also heterogeneity of the sensor’s response (e.g. XRT, XRT etc.). While the desired element should have a high heterogeneity for efficient sorting, the sensor’s response also needs to be distinct enough so that the detection and separation could occur.   Metal Liberation Curve Metal liberation curve (or ideal recovery curve) demonstrates the ideal scenario for sorting. It shows the grade-recovery curve assuming that all particles are sorted out based on their desired metal grade, highest to lowest. The struggle in sorting is to try and match the liberation curve using one or multiple sensors or methods.  Grade Estimation For the base metal samples, due to financial restriction, it was not possible to assay every single rock. Therefore, after the sensor tests were completed, all the rocks were pulverized to approximately 85% passing 75 µm, placed in separate Ziplock bags and then were analyzed 40  using the same XRF device. Since this measurement was meant to be used as a measurement of true grade, longer exposure times of 30 seconds for the first beam and 30 seconds for the second beam were used. A quick study (results not presented) showed that a 30-second primary beam and a 30-second secondary beam would perform as well as 60-30. Ziplock bags were scanned on both sides and the average value was calculated. To examine the validity of the pulverized XRF readings, around 20% of the rocks from each size fraction from each of the base metal mines were sent for ICP and Fire Assays. The ICP results for the main elements (e. g. Cu, Zn, Pb, Fe, S) were used to create correlation curve with those of the pulverized XRF readings. The correlation equation was then used to extrapolate these grades for the rocks that were not assayed (~80% of remaining rocks) based on their pulverized XRF readings. These grades were used as the actual grades of base metals in the samples. To predict the gold and silver grades, this time, pulverized XRF readings of those 20% of the rocks were used in an MLR analysis against their gold and silver fire assay results to create a correlation equation. This equation was subsequently used to extrapolate gold and silver grades for the rest of the rocks. In the case of the gold samples, all the 100 rocks were sent for Fire Assay and ICP, therefore no extrapolation was necessary.  In terms of assumptions for the grades, wherever the grades reported for an element were below the limit of detection, for the sake of calculations, they were assumed to be zero. For the over-limit grades (in case of the gold samples only), extrapolation techniques were used to estimate the grades. Over-limit grades for the base metal samples were measured in the assay lab.  Multivariate Linear Regression (MLR) Analysis Multivariate Linear Regression analysis was performed in an attempt to improve the sortability of the rocks by taking multiple factors (e.g. XRF grades for multiple elements) into account. MLR analysis can also be used to investigate a combination of sensors. However, this approach should not be mistaken with using multiple sensors in series. When used in series, each sensor’s response is evaluated individually before a decision is made, while using an MLR analysis on multiple sensors suggests that all the responses from the sensors are combined, analysed 41  against each other and then a decision is made based on the most relevant set of factors. This method has shown to exclude sensors such as XRT in favor of XRF as clearly an XRF reading correlates more with actual grades of copper, as an example. MLR analysis is performed using MATLAB. The ICP (or Fire Assay in case of gold) grades of rocks from each size fractions were used as the Y variable in the algorithm. The sensors responses, such as grades of various elements for the XRF or Magnitude/Phase responses for the EM, were used as the X variables. The “stepwise(x,y)” algorithm in MATLAB was run under default conditions with –value of 0.05 as the threshold for inclusion or exclusion of each variable. The generated coefficients for the significant factors by the algorithms were then used to calculate the predicted grades (?̂?) of the desired element (i.e. copper, zinc or gold). The resultant grades (?̂?) were used to generate the grade-recovery curves. 42  4. ORE CHARACTERIZATION  Introduction All the rocks in this study were subject to tests using various sensors and analytical tests. The base metal samples were subject to tests using Optical (OPT), Electromagnetic (EM), X-Ray Transmission (XRT) and X-Ray Fluorescence (XRF). The gold sample were also tested using the same sensors, except for the Optical sensor. In this section, first the responses from each of these sensors are analysed and the reason behind false negatives and false positives are discussed. False negatives are defined as the valuable rocks not being identified as valuable by the sensor. Similarly, false positives are defined as gangue material misidentified as valuable.  The algorithm for the Optical and XRT sensors performs an image analysis technique which generates indices that are correlated with the grades of each corresponding rock. The details of how these indices are generated were never disclosed by Tomra. But it is believed that the software sees the colors and develops an association between certain color spectra and the grades and thereby it generates Spectral Index values. The average of these values are then correlated with the grades and used to generate grade-recovery curves. The generated indices are not unique and can mean differently for different samples. One example for this can be seen below where for Brenda Size A, light regions are associated with high grade and given higher Spectral Indices while for Copper Mountain Size B, dark regions are considered high grade and therefore given higher Spectral Indices. After discussing the sensors’ responses, the analytical results, including ICP and XRD, are presented.  Optical Sensor Response Analysis To investigate the performance of the sensor and determine the reason behind false negative and false positive samples, a few rocks from only one size fraction of each mine is investigated. The rocks are selected based on their spectral indices generated by the algorithm and are tested against their corresponding grades. The spectral index is the criteria through which the sorting threshold can be determined, and therefore it was used here in determining whether a rock was 43  categorized as high-grade or low-grade and hence determining which one was a false negative or positive. Among the rocks selected are 1 true positive 1 true negative 1 false positive 1 false negative with 1 true negative matching its spectral index for comparison  Brenda – Size Fraction A (-25+19 mm) Table 4.1 summarizes the five rocks as explained above. As it can be seen in the table, rock #29 with the highest grade of copper has the highest spectral index assigned to it. Similarly, rock #38 with 0% copper has the lowest spectral index of 2. Below, these two rocks are examined against each other before discussing the false positive and negatives. Table 4.1 Spectral Indices for select rocks from Brenda Size Fraction A Spectral Index Sample ID # Grade Cu (%) Description 23 29 8.98% True Positive 2 38 0.00% True Negative 18 47 0.00% False Positive 7 2 1.21% False Negative 7 39 0.00% True Negative Figure 4.1 shows the processed image for rock #29. From the image on the left, crystals of chalcopyrite can be seen, which in the image they look bright. Therefore, the algorithm associated bright spots as high grade and marked them red in the image on the right.   44  Figure 4.1 Processed optical image for Brenda A Rock #29  Based on the same argument above, rock #38 was low-grade and looks quite dark (Figure 4.2) and therefore it was assigned a low spectral index, categorizing it as waste. Figure 4.2 Processed optical image for Brenda A Rock #38  Examining rock #47, indicates that the algorithm assigned a spectral index of 18 to it, identifying it as high-grade while this rock had no copper content. The reason for this misidentification can be attributed to the lighting for this specific rock (Figure 4.3). Although lighting for all rocks were similar, some reflected the light directly back at the camera and therefore appeared brighter in the image. As mentioned earlier, the algorithm associated brightness with high-grade material and therefore identified this rock as high-grade. The white spots that can be observed in Figure 4.3 and Figure 4.4 are parts of the image where the 45  algorithm failed to capture properly and therefore the resultant Spectral Index does not reflect a realistic reading for the rock. Figure 4.3 Processed optical image for Brenda A Rock #47  Based on the same reasoning, the reason behind rock #2 having been identified as waste can be attributed to the fact that this rock appeared dark under the camera. Both rock #2 and #39 have the same spectral index of 7 and it can been seen from Figure 4.4 and Figure 4.5 that they look similar. Figure 4.4 Processed optical image for Brenda A Rock #2    46  Figure 4.5 Processed optical image for Brenda A Rock #39  At the end, it should be mentioned that the host rock at Brenda was granodiorite, with mineralized vein structure. Therefore, there was always the risk of blind spots. This could have been the case with rock #2. However, most optical sorters these days take advantage of multiple cameras and therefore eliminate such possibilities significantly.  Copper Mountain – Size Fraction B (-37.5+25 mm) Similarly to Brenda, five rocks were selected from the Size Fraction B of Copper Mountain material. The highest grade rock was rock #20 with 2.11% copper and rocks #58 and #47 with 0% (true negatives) and #19 with only 0.009% copper was a false positive. There were not many high-grade rocks and therefore rock #86 with 0.289% copper was taken as a false negative. Table 4.2 Spectral Indices for select rocks from Copper Mountain Size Fraction B Spectral Index Sample ID # Grade Cu (%) Description 26 20 2.11% True Positive 2 58 0.00% True Negative 19 36 0.009% False Positive 6 86 0.289% False Negative 6 47 0.00% True Negative Based on Figure 4.6, it seems that the darker spots are associated with high grade, mineralized rocks. This was despite the presence of a bright mineralized spot on the surface of rock #20.   47  Figure 4.6 Processed optical image for Copper Mountain B, Rock #20  Now looking at rock #58, it was obvious that the algorithm considered brighter areas as low-grade (Figure 4.7) and therefore no correlation with the high-grade material or areas on the rock. Figure 4.7 Processed optical image for Copper Mountain B, Rock #58  Now that the optical criteria for high- and low-grade rocks are established, it can easily be understood why rock #36 (Figure 4.8) was identified as false positive. Type of the rock, with dark features, and also shadows played a role in its misidentification.   48  Figure 4.8 Processed optical image for Copper Mountain B, Rock #36  Rocks #86 and #47 were given the same spectral index of 7 by the algorithm, despite the difference in their grades. Looking at Figure 4.9 for Rock #86, it is quickly realized that the rock was brighter than most high-grade rocks in the sample which resulted in its lower spectral index value. Also, the image for rock #47 (Figure 4.10) shows some shaded areas which overall made the rock look darker and therefore gave it a higher spectral index. Rock #47 should have ideally had the same spectral index as rock #58 because they are the same type of rock (Figure 4.7 vs. Figure 4.10). Therefore, the higher spectral index of the low-grade rock and lower spectral index of high-grade rock place misplaced them in the sorting algorithm resulting in rock #86 potentially reporting to waste. Figure 4.9 Processed optical image for Copper Mountain B, Rock #86    49  Figure 4.10 Processed optical image for Copper Mountain B, Rock #47   Mount Polley – Size Fraction C (-50+37.5 mm) Generally Mount Polley rocks were quite homogenous in terms of both appearance and grades, therefore a good potential for sorting was not expected. Here the same trend was observed, with less than ideal recoveries for copper. Same procedure as for previous rocks was taken to compare five rocks from the Size Fraction C of the Mount Polley material. Table 4.3 Spectral Indices for select rocks from Mount Polley Size Fraction C Spectral Index Sample ID # Grade Cu (%) Description 14 40 0.95% True Positive 4 11 0.02% True Negative 14 69 0.01% False Positive 6 52 0.27% False Negative 6 29 0.02% True Negative By examining rocks #40 (Figure 4.11) and #11 (Figure 4.12), the true positive and negative, respectively, it can be concluded that the algorithm correlated the dark green tone colors to high grade and orange/pink colors to low-grade and thereby assigning corresponding spectral indices to these rocks.   50  Figure 4.11 Processed optical image for Mount Polley C, Rock #40  Figure 4.12 Processed optical image for Mount Polley C, Rock #11  Consequently, rock #69, false positive, which had a dark green color (Figure 4.13) was misidentified as high-grade while it only contained 0.01% copper. Figure 4.13 Processed optical image for Mount Polley C, Rock #69  A quick comparison between the two rocks with the same spectral index of 6 but different grades shows how similar they are in terms of their colors. Rock #52 with 0.27% copper could, therefore, be potentially misplaced in the reject stream. Analysis of processed images for these 51  two rocks, Figure 4.14 and Figure 4.15, explains why these two rocks are seen rather similar under the camera. Both rocks contain pink/beige colors, and although the rocks are different in nature, this difference was not picked up on by the sensor. Figure 4.14 Processed optical image for Mount Polley C, Rock #52  Figure 4.15 Processed optical image for Mount Polley C, Rock #29   Myra Falls – Size Fraction D (-75+50 mm) Color associations for Myra Falls were similar to those of Mount Polley. While dark green/grey was correlated with high-grade rocks (Figure 4.16), orange/beige colors were indicative of low-grade material as can be seen in Figure 4.17 for rock #56. Table 4.4 Spectral Indices for select rocks from Myra Falls Size Fraction D Spectral Index Sample ID # Grade Zn (%) Description 36 63 28.88% True Positive 2 56 0.02% True Negative 25 4 0.05% False Positive 6 52 8.81% False Negative 6 22 0.04% True Negative 52  Figure 4.16 Processed optical image for Myra Falls D, Rock #63   Figure 4.17 Processed optical image for Myra Falls D, Rock #56  This color association would easily explain the misidentification of #4 (dark green look) as high-grade (Figure 4.18) as well as the misplacement of rock #52 (orange/brown) in the waste fraction (Figure 4.19). Figure 4.18 Processed optical image for Myra Falls D, Rock #4    53  Figure 4.19 Processed optical image for Myra Falls D, Rock #52  It is important to shift the attention to the shadow effects in rock #22 (Figure 4.20) and how these effects led to this rock getting a higher spectral index than it should have. Figure 4.20 Processed optical image for Myra Falls D, Rock #22  There was only a few cases of false negatives in the Myra Falls dataset with regards to the optical sensor. Therefore, as it will be seen later in this thesis, optical sorter worked fairly well with this material, although for some size fractions it was better than the others.  Electromagnetic Sensor Response Analysis At the time of performing the test, there was not any signal analysis algorithm available. Since developing an algorithm was not considered in the scope of this research, rudimentary measures were taken to determine whether the EM sensor was applicable to the material or not. This method compared the average magnitude/phase values for rocks of around different grade ranges at different frequencies. This method showed no significant difference for any of the 54  samples. Therefore it was determined that the EM sensor was not applicable to these rock types and as a result, the tests for some other samples were not performed. At a later time during this research, multivariate linear regression analysis was applied on the magnitude/phase responses to check the applicability of the electromagnetic sensor. MLR analysis confirmed the preliminary analysis done earlier for most of the cases with a few exceptions that will be discussed later in this thesis. Two cases are analysed here to show what the sensor responses were like in cases were EM proved applicable and when it could not offer any solutions. The two cases presented here are Copper Mountain size B fraction (-37.5+25 mm) and the same mine but size fraction D (-75+50 mm).  Copper Mountain – Size B Multivariate Linear Regression did not identify any significant factors among the 28 responses (14 Magnitude and 14 Phase) and their interaction effects, a total of 406 factors. To check the validity of this, each of the two responses from the EM sensor, Magnitude and Phase, were plotted against the frequency across which they were generated. Since plotting the response curves for all 100 rocks in graph would have lost its clarity and would not necessarily show any useful information, only the responses for a few select rocks (10) in three grade ranges are shown in each graph. To build a case, first the sensor responses for the highest and lowest grade rocks were plotted to see if they differ in patterns. Afterward, more rocks from the highest, lowest and around the cut-off grade (0.1% to 0.2% Cu) were added to the plot and were determined whether they followed the high grade pattern/location on the graph or not. Figure 4.21 shows the magnitude for response over the frequency range of 100 kHz to 1400 kHz for ten rocks. The legend indicates the rock numbers along with their copper grade. A quick look at the graph reveals that there is no difference among any of the responses even though the grades of particles are different.   55  Figure 4.21 Magnitude vs. Frequency for Copper Mountain Size B – select rocks  Figure 4.22 demonstrates the Phase part of the response from the electromagnetic sensor for the same rocks as in Figure 4.21. Although at first sight the responses look to be different, a close look proves otherwise. Again, for the purpose of keeping the graph simple to read, the responses for only ten rocks are presented. The rocks were selected from high-grade (x3 rocks), Low-grade (x3 rocks) and around the cut-off grade (x4 rocks). Each of these groups of rocks, although similar in grade, showed up on different regions of the graphs, indicating that a clear cut separation based on Phase cannot be determined either. Although only ten rocks are shown on the graph, but they do represent the general trend of the rocks. Table 4.5 summarizes the grades of copper and iron for the elements in the graph and whether or not they were classified as “high grade” by the sensor.   051015202530Magnitude (mV)Frequency (kHz)#20 - Cu: 2.11%#60 - Cu: 1.50%#11 - Cu: 1.35%#92 - Cu: 0.21%#37 - Cu: 0.17%#80 - Cu: 0.17%#70 - Cu: 0.17%#16 - Cu: 0.01%#35 - Cu: 0.00%#81 - Cu: 0.00%56  Figure 4.22 Phase vs. Frequency for Copper Mountain Size B – select rocks  Table 4.5 Copper and iron grades for select samples and corresponding Phase response Rock Cu Fe Classified as High-Grade? #20 2.11% 5.98% Yes #60 1.50% 4.72% Middling #11 1.35% 4.13% No #92 0.21% 1.29% Middling #37 0.17% 1.23% Yes #80 0.17% 1.04% Middling #70 0.17% 1.06% No #16 0.01% 0.56% No #35 0.00% 2.67% Yes #81 0.00% 3.48% Middling  Copper Mountain – Size D For size fraction D of Copper Mountain material, MLR analysis identified only one factor out of the 406 available as significant. Meanwhile, looking at Figure 4.23 and Figure 4.24, representing Magnitude and Phase responses for a few select elements, demonstrated some pattern, specifically in the case of the Phase response.   -3-2.5-2-1.5-1-0.50Phase (degrees)Frequency (kHz)#20 - Cu: 2.11%#60 - Cu: 1.50%#11 - Cu: 1.35%#92 - Cu: 0.21%#37 - Cu: 0.17%#80 - Cu: 0.17%#70 - Cu: 0.17%#16 - Cu: 0.01%#35 - Cu: 0.00%#81 - Cu: 0.00%57  Figure 4.23 Magnitude vs. Frequency for Copper Mountain Size D – select rocks  Figure 4.24 Phase vs. Frequency for Copper Mountain Size D – select rocks  Table 4.6 summarizes the copper and iron grades for the selected elements and their corresponding response for both Magnitude and Phase. Unlike the case with Copper Mountain Size B fraction presented earlier, both Magnitude and Phase followed a similar trend in terms of identifying rocks as high or low grade. Therefore they both identified the same rocks as high or low-grade. -30-20-1001020304050Magnitude (mV)Frequency (kHz)#28 - Cu: 1.30%#43 - Cu: 1.13%#51 - Cu: 0.47%#46 - Cu: 0.33%#82 - Cu: 0.33%#79 - Cu: 0.21%#19 - Cu: 0.21%#73 - Cu: 0.12%#41 - Cu: 0.10%#29 - Cu: 0.00%#85 - Cu: 0.00%-4-3.5-3-2.5-2-1.5-1-0.500.51Phase (degrees)Frequency (kHz)#28 - Cu: 1.30%#43 - Cu: 1.13%#51 - Cu: 0.47%#46 - Cu: 0.33%#82 - Cu: 0.33%#79 - Cu: 0.21%#19 - Cu: 0.21%#73 - Cu: 0.12%#41 - Cu: 0.10%#29 - Cu: 0.00%#85 - Cu: 0.00%58  Similar to the case with other sensors, and despite a more clear-cut difference in the trends of low and high-grade rocks, several cases of false positive and false negative were observed. Such as rock #51 that with 0.4% copper and 1.71% iron being identified as low-grade and rock #29 with 0% copper and 2.41% iron as high-grade. At first it might seem that iron content is the culprit, but looking at rock #85 with 0% copper and 3.34% iron also reports to low-grade, rejects this hypothesis. The iron minerals available at Copper Mountain include magnetite and hematite as magnetic ones and augite, clinochlore, ankerite and siderite as non-magnetic. Without the mineralogical results for individual rocks, it is difficult to pinpoint the reason behind the discrepancies observed. The sensitivity to magnetic excitation can be related not only to the elements/minerals present in a rock but also the matrix of such elements/minerals inside the rock. Investigating the matter in more detail was out of scope of this study. Table 4.6 Copper and iron grades for select samples and corresponding Magnitude and Phase response Rock Copper Iron  Identified as high-grade? #28 1.30% 3.15% Yes #43 1.13% 2.81% Yes #51 0.47% 1.71% No #46 0.33% 1.83% Yes #82 0.33% 1.04% No #79 0.21% 3.02% No #19 0.21% 1.49% Yes #73 0.12% 1.39% No #41 0.10% 1.58% Yes #29 0.00% 2.41% Yes #85 0.00% 3.34% No Based on the two sets of results presented here, one size B and the other size D, one might tend to believe that size of the rocks matter, however, other sets of results, presented later in this thesis will reject this hypothesis. It is the author’s conclusion that a sophisticated algorithm to analyse the sensor’s response is necessary to take advantage of possible potentials of the EM sensor. 59   XRT Sensor Response Analysis To investigate the XRT sensor performance, results of tests on the low grade was were studied in further detail. It is important to mention that while the XRT tests for the base metal mines were done on a rock by rock bass, the gold samples were scanned dynamically in bulk and therefore the XRT indices for individual rocks for the gold samples are not available. Another point that is worth mentioning is that the XRT sensor generates “Spectral” and “Texture” responses, as well as a “Combined” signal which takes into account the former two. On Tomra’s suggestion, only the “Spectral Index” was considered for the material under investigation (base metal) as it had been shown to correlate better with the grades of base metals. To compare some of the rocks, a true positive and a true negative as well as a false positive and false negative were chosen from the “Spectral” response. Since only the indices for each of these rocks were available (as opposed to having an already categorized stream as in the case with the gold samples), defining a false negative could be subjective. For the analysis of these rocks, the main metal (copper or zinc), iron and any other heavy element with a significant grade were considered.  Brenda Table 4.7 summarizes the grades as well as the spectral indices for four rocks from the Brenda waste dump, size fraction A (-25 mm +19 mm). Table 4.7 XRT spectral indices and grades for select rocks in Brenda Size A Sample ID Spectral Index Cu Mo Au [g/t] Ag [g/t] Fe Denomination 29 53 8.98% 0.22% 12.760 52.15 7.47% True Positive 47 2 0.00% 0.00% 0.000 0.000 1.52% True Negative 27 8 0.11% 0.006% 0.006 0.528 5.73% False Positive 34 4 0.52% 0.004% 0.059 2.516 2.61% False Negative To form a frame of reference for the measurement, processed images for rock #29 (true positive, Figure 4.25) and rock #47 (true negative, Figure 4.26) are compared. Although the low-energy photos of the rocks do not show a significant difference, the difference is quite discernible in the high-energy photos. The gray-scale image is then processed to develop the heat map image on the photo on top left corner of each figure. In the heat map image, the high-grade rock shows 60  a number of dark red pixels that are positively correlated to high-grade material. The reverse holds true for rock #47. Based on the same reasoning, looking at rocks #27 and #34 (Figure 4.27 and Figure 4.28), it can be understood why they were reported as false positive and negative, respectively. Examining the grades would shed more light on the reason behind this misidentification. As mentioned earlier, presence of iron as a heavy element can tarnish the accuracy of the XRT sensor. This can be seen in the case of rock #27 (Figure 4.27) where the copper grade is below the cut-off grade but the presence of iron makes the rock look darker under the x-ray and therefore a higher spectral index is assigned to it. Similarly, in the case of rock #34 (Figure 4.28), although the copper grade is 0.5%, lower iron content makes the rock look brighter under the x-ray and therefore a lower spectral index assigned to it. Figure 4.25 Processed Spectral XRT image for Brenda A, Rock #29    61  Figure 4.26 Processed Spectral XRT image for Brenda A, Rock #47  Figure 4.27 Processed Spectral XRT image for Brenda A, Rock #27    62  Figure 4.28 Processed Spectral XRT image for Brenda A, Rock #34  Table 4.8 summarizes the XRT spectral indices for the top 10 rocks in this size fraction. Table 4.8 Top 10 rocks based on XRT Spectral Index – Brenda Size A Sample ID (#) Spectral Index Cu (%) Fe (%) 29 53 8.98 7.47% 96 16 0.60 5.94% 2 14 1.21 3.48% 44 12 0.95 4.57% 53 9 1.13 4.18% 27 8 0.11 5.73% 21 7 0.22 3.96% 68 7 0.20 3.55% 97 7 0.19 3.20% 19 6 0.00 3.54%  Copper Mountain The second sample that is discussed here is Copper Mountain size fraction B. Since the XRT sensor’s principal is based on the atomic density of the rocks, therefore the same reasoning as with the case of Brenda applies. With that in mind, four rocks from Copper Mountain size fraction B were analysed. Table 4.9 summarizes the spectral indices and grades for the selected rocks. 63  Table 4.9 XRT spectral indices and grades for select rocks in Copper Mountain Size B Sample ID Spectral Index Cu Mo Au [g/t] Ag [g/t] Fe Denomination 21 20 2.11% 0.004% 1.25 3.55 3.87% True Positive 4 2 0.01% 0.001% 0.00 0.00 0.47% True Negative 54 7 0.01% 0.001% 0.06 0.11 1.38% False Positive 60 6 1.50% 0.007% 0.76 2.44 4.72% False Negative Figure 4.29 shows the low and high energy images of rock #21 along with its processsed image. Comparing that with Figure 4.30 that shows the same images for rock #4, a clear difference in the darkness and hence, intensity of the sensor’s response can be observed. With 2.11% copper and 3.87% iron, rock #21 appeared dark under the x-ray while rock #4 with insignificant amounts of either element, looked bright in comparison. This distinct difference in appearance resulted in a correct identification of these rocks as true positive and true negative, respectively. Figure 4.29 Processed Spectral XRT image for Copper Mountain, Rock #21    64  Figure 4.30 Processed Spectral XRT image for Copper Mountain, Rock #4  The case for the false positive and false negative rocks was different, however. Looking at their copper and iron grade, it was expected that the rocks should have been identified correctly. Rock #54, with only 0.01% copper and 1.38% iron should have been given a low spectral index, while similarly, rock #60 with 1.50% copper and 4.72% iron should have been given a high spectral index. Further investigation showed that there was rather large difference between the sizes of these two particles. Although the purpose of a Dual Energy XRT is to account for this difference, at times where the difference in thickness is significant, misidentifications such as this could happen. The processed images of these two rocks can be seen in Figure 4.31 and Figure 4.32, below.    65  Figure 4.31 Processed Spectral XRT image for Copper Mountain, Rock #54  Figure 4.32 Processed Spectral XRT image for Copper Mountain, Rock #60  66  Table 4.10 summarizes the grades as well as the spectral indices for the top 10 rocks for this size fraction. Table 4.10 Top 10 rocks based on XRT Spectral Index – Copper Mountain size fraction B Sample ID (#) Spectral Index Cu (%) Fe (%) 21 20 1.38% 3.9% 31 10 1.85% 5.3% 20 8 2.11% 6.0% 11 7 1.35% 4.1% 16 7 0.01% 0.6% 54 7 0.01% 1.4% 77 7 0.02% 0.5% 79 7 0.37% 2.6% 97 7 0.21% 1.3% 5 6 0.05% 0.9%  Mount Polley Examining the grade-recovery graphs for Mount Polley (Figure 4.33 to Figure 4.36) indicates that unlike the other three cases of base metals studied here, there are no distinct indications of a false positive or a false negative. This is due to the fact that most of the Mount Polley rocks in these samples shared similar grades. As it was mentioned earlier, Mount Polley material was quite homogeneous. The effect of the homogeneity in the appearance was observed in the case of the optical sensor, and that of constituents can be seen here in the case of the XRT sensor (and XRF for that matter).  Table 4.11 summarize the grades along with the corresponding spectral indices for the selected rocks in Mount Polley size fraction C. Considering the principle behind the XRT sensor is the same as before, even a quick look at the table below explains the denomination of each rock. Rocks #40 and #11 with the high and low grades of copper and iron, respectively, have a known fate in spectral identification by the XRT sensor. Rock #30 however, with one of the highest iron grades, it is expected to appear dark under the XRT sensor, and examining Figure 4.35, confirms this as well. Rock #47 although with 0.31% copper, because it had a low iron content, it showed brighter under the x-ray and therefore got a lower spectral index that in turn led to its misidentification by the sensor. The processed images for all these rocks can be seen in 67  Figure 4.33 to Figure 4.36. Table 4.12 summarizes the grades and spectral indices for the top 10 dense rocks in this size fraction. Table 4.11 XRT spectral indices and grades for select rocks in Mount Polley size fraction C Sample ID Spectral Index Cu Mo Au [g/t] Ag [g/t] Fe Denomination 40 14 0.951% 0.001% 1.041 2.572 5.65% True Positive 11 2 0.023% 0.001% 0.000 0.333 3.34% True Negative 30 14 0.100% 0.001% 0.000 0.467 8.19% False Positive 47 4 0.310% 0.001% 0.398 0.518 2.41% False Negative Figure 4.33 Processed Spectral XRT image for Mount Polley, Rock #40    68  Figure 4.34 Processed Spectral XRT image for Mount Polley, Rock #11  Figure 4.35 Processed Spectral XRT image for Mount Polley, Rock #30    69  Figure 4.36 Processed Spectral XRT image for Mount Polley, Rock #47  Table 4.12 Top 10 rocks based on XRT Spectral Index – Mount Polley size fraction C Sample ID (#) Spectral Index Cu (%) Fe (%) 72 17 0.36% 4.98% 59 16 0.23% 3.38% 50 15 0.33% 8.92% 30 14 0.10% 8.19% 40 14 0.95% 5.65% 43 14 0.30% 4.60% 65 14 0.33% 4.18% 3 13 0.18% 4.62% 20 13 0.15% 7.01% 21 13 0.17% 6.52%  Myra Falls For Myra Falls samples, Size fraction D was chosen for a detailed analysis. Since the reasoning behind the denomination of rocks determined by XRT is the same as before, these rocks are not individually explained, however some interesting facts about are mentioned. Looking at Table 4.13, the first thing that possibly attracts attention is the high grades of zinc and iron in some of the samples. These high grades (e.g. 49% zinc and 35% iron) make grades 70  such as 5% and 6% seem relatively low. And that was what caused misidentification of, for example, rock #26 as low grade. Table 4.13 XRT spectral indices and grades for select rocks in Myra Falls size fraction D Sample ID Spectral Index Zn Cu Pb Au [g/t] Ag [g/t] Fe Denomination 23 71 48.74% 0.09% 4.32% 3.51 98.34 3.22% True Positive 39 2 0.06% 0.05% 0.00% 0.00 0.55 5.57% True Negative 68 60 0.26% 0.12% 0.03% 0.63 2.99 34.89% False Positive 26 8 2.09% 0.10% 0.37% 0.08 5.64 6.58% False Negative The other interesting fact that was noticed about Myra Falls samples was that due to the high grades of zinc, false negatives were barely observed. As an example, rock #26 with 2.09% zinc (just around the cut-off grade) was the only rock with a zinc grade above 2% in the bottom four fifths of the samples. Figure 4.37 to Figure 4.40 demonstrate the processed and unprocessed images of these four rocks. It only requires a quick glance to realize that how darker high-grade rocks look under the x-ray compared to the samples from previous materials. Table 4.14 summarizes the top 10 dense rocks in this size fraction based on their spectral indices. It is important to mention that all the rocks that were placed before rock #11 had much higher iron grades that this rock. Therefore they received a higher spectral index and were placed before rock #11.   71  Figure 4.37 Processed Spectral XRT image for Myra Falls, Rock #23  Figure 4.38 Processed Spectral XRT image for Myra Falls, Rock #39    72  Figure 4.39 Processed Spectral XRT image for Myra Falls, Rock #68  Figure 4.40 Processed Spectral XRT image for Myra Falls, Rock #26    73  Table 4.14 Top 10 rocks based on XRT Spectral Index – Mya Falls size fraction D Sample ID (#) Spectral Index Cu (%) Fe (%) 76 59 28.75% 9.03% 71 23 48.74% 3.22% 60 68 0.26% 34.90% 53 63 28.88% 8.72% 45 85 0.45% 37.25% 44 73 8.87% 5.15% 40 91 0.94% 27.11% 39 52 8.81% 5.23% 36 81 0.21% 33.81% 34 98 0.64% 36.66%  XRF Sensor Response Analysis  Comparison of Base Metal Grades with ICP To investigate the accuracy of the XRF readings, rock XRF (Rock) and pulverized XRF (Pul.) readings were correlated with the ICP assays. Also to investigate the effects of size, and to see whether or not there is a change in correlation with size, the smallest and largest size fractions of Myra Falls are examined here. Since XRF readings are independent of the type of rocks, Myra Falls is the only material that is investigated here. The reason that Myra falls was selected for these investigations was that it contained all the base metals (Zn, Cu, Pb) and therefore XRF accuracy could be examined against a suite of major desired elements. Table 4.15 summarizes the correlation coefficients between the two XRF readings and the corresponding ICP assays. ICP assays are explained in Section 4.6 of this Chapter. As each metal is examined, the reader is referred to the corresponding figure where all these values are plotted against their corresponding rock ID. First the desired base metals along with iron as heavier elements are analysed and then afterward, lighter elements such as sulfur along with magnesium, calcium and potassium as gangue indicators are studied. Focusing on the heavy elements, first, the effects of size on the correlation factors is analysed. Initially it was expected that due to “sampling error” (four measurements on a small rock vs. on a big rock), the correlation between the rock XRF and ICP would be weaker for the larger size fractions compared to the small size fraction. However, a quick comparison between the rock XRF correlation coefficients for the two sizes and the four metals suggests that size did not play a role in determining these factors. While these correlation coefficients were all quite 74  high for zinc and copper, it was a different case for lead. For the size fraction D, a weak correlation was observed between rock XRF and the ICP assays.  This could have been due to a blind spot when rock XRF tests were being performed. Iron on the other hand showed a generally weaker correlation compared to the other base metals. Iron with atomic weight of 55.85 g/mol is lighter than copper (63.54 g/mol) and zinc (65.38 g/mol). As it will be presented later in this section, XRF works best for heavier elements. Although iron is not significantly lighter than copper or zinc, XRF might not have been accurately reading iron grades which in turn would lead to a weaker correlation between the XRF readings and the actual iron grade. Table 4.15 Correlation factors: XRF vs. ICP – Heavy Metals, Myra Falls  Size Correlations with ICP Zinc Copper Lead Iron  Rock Pul. Rock Pul. Rock Pul. Rock Pul. Size A 0.972 0.999 0.905 0.989 0.915 0.999 0.867 0.960 Size D 0.986 0.996 0.958 0.973 0.765 0.997 0.923 0.964 It is important to mention that a high correlation factor does not necessarily mean that the measured values are close to each other, rather that the trend is similar. As it can be seen in Figure 4.41, these three measurements do not overlap (except for very low grades), neither have they followed a consistent trend in over- or underestimating the grades. This was the case with both size fractions for all the four heavy elements studied here (Figure 4.41 through Figure 4.48). Therefore, it cannot be said for sure that either of these measurements consistently under or overestimates the actual grades.   75  Figure 4.41 Comparison: XRF vs. ICP – Myra Falls, Zinc, size fraction A  Figure 4.42 Comparison: XRF vs. ICP – Myra Falls, Zinc, size fraction D    0%10%20%30%40%50%60%0 20 40 60 80 100Zinc GradeRock IDRock XRFPul. XRFICP0%10%20%30%40%50%60%0 20 40 60 80 100Zinc GradeRock IDRock XRFPul. XRFICP76  Figure 4.43 Comparison: XRF vs. ICP – Myra Falls, Copper, size fraction A  Figure 4.44 Comparison: XRF vs. ICP – Myra Falls, Copper, size fraction D    0%2%4%6%8%10%12%14%0 20 40 60 80 100Copper GradeRock IDRock XRFPul. XRFICP0%1%2%3%4%5%6%7%8%0 20 40 60 80 100 120Copper GradeRock IDRock XRFPul. XRFICP77  Figure 4.45 Comparison: XRF vs. ICP – Myra Falls, Lead, size fraction A  Figure 4.46 Comparison: XRF vs. ICP – Myra Falls, Lead, size fraction D    0.0%0.5%1.0%1.5%2.0%2.5%3.0%3.5%4.0%0 20 40 60 80 100Lead GradeRock IDRock XRFPul. XRFICP0.0%0.5%1.0%1.5%2.0%2.5%3.0%3.5%4.0%4.5%5.0%0 20 40 60 80 100 120Lead GradeRock IDRock XRFPul. XRFICP78  Figure 4.47 Comparison: XRF vs. ICP – Myra Falls, Iron, size fraction A  Figure 4.48 Comparison: XRF vs. ICP – Myra Falls, Iron, size fraction D  The grade comparison results for these studies are summarized in Table 4.16 and Table 4.17 for size fraction A, and Table 4.18 and Table 4.19 for size fraction D. A look at Table 4.19 indicates the reason behind the weak correlation between Rock XRF and the ICP results for lead. For rock #73, XRF only reads 0.16% lead while the rock contains just over 3% lead. This was the only rock that caused the poor correlation.   0%5%10%15%20%25%30%35%0 20 40 60 80 100Iron GradeRock IDRock XRFPul. XRFICP0%5%10%15%20%25%30%35%40%0 20 40 60 80 100Iron GradeRock IDRock XRFPul. XRFICP79  Table 4.16 Comparison of Grades: XRF vs. ICP, Myra Falls A, Zinc & Copper Sample ID Zinc  Copper Rock Pul. ICP  Rock Pul. ICP 2 0.09% 0.03% 0.01%  0.04% 0.02% 0.02% 3 0.11% 0.07% 0.04%  0.00% 0.01% 0.01% 4 0.04% 0.02% 0.01%  0.01% 0.01% 0.01% 6 0.05% 0.05% 0.03%  0.00% 0.01% 0.00% 12 0.10% 0.09% 0.06%  0.00% 0.01% 0.01% 17 0.10% 0.04% 0.03%  0.03% 0.03% 0.02% 25 43.54% 47.43% 52.60%  0.18% 0.05% 0.11% 31 0.04% 0.08% 0.06%  0.00% 0.00% 0.00% 34 0.00% 0.02% 0.01%  0.00% 0.00% 0.00% 39 0.03% 0.03% 0.02%  0.00% 0.03% 0.02% 41 35.13% 27.72% 28.74%  2.86% 3.89% 4.75% 45 24.27% 32.61% 35.86%  1.41% 2.00% 2.59% 47 1.75% 1.70% 1.49%  0.12% 0.18% 0.17% 58 0.25% 0.54% 0.35%  0.01% 0.05% 0.04% 59 0.00% 0.01% 0.01%  0.00% 0.00% 0.00% 69 3.31% 9.95% 9.02%  0.06% 0.30% 0.36% 76 0.04% 0.80% 0.59%  0.00% 0.06% 0.05% 77 0.20% 0.23% 0.18%  0.00% 0.01% 0.01% 84 2.65% 1.51% 0.97%  4.69% 7.57% 6.48% 86 0.43% 1.11% 0.74%  11.60% 8.29% 7.23% 87 0.03% 0.02% 0.01%  0.01% 0.01% 0.01% 92 0.04% 0.10% 0.05%  0.00% 0.01% 0.01%    80  Table 4.17 Comparison of Grades: XRF vs. ICP, Myra Falls A, Lead & Iron Sample ID Lead  Iron Rock Pul. ICP  Rock Pul. ICP 2 0.02% 0.00% 0.00%  20.49% 26.32% 28.03% 3 0.00% 0.00% 0.00%  5.08% 5.42% 6.10% 4 0.00% 0.00% 0.00%  1.32% 1.48% 1.38% 6 0.00% 0.00% 0.00%  5.70% 5.22% 6.31% 12 0.00% 0.00% 0.00%  9.26% 6.14% 6.84% 17 0.00% 0.00% 0.00%  9.00% 7.11% 7.51% 25 0.49% 0.16% 0.19%  2.16% 1.66% 4.72% 31 0.04% 0.01% 0.01%  5.03% 3.10% 3.10% 34 0.00% 0.00% 0.00%  0.74% 1.07% 1.26% 39 0.01% 0.01% 0.01%  3.20% 2.58% 2.56% 41 1.64% 3.06% 3.72%  8.24% 7.41% 14.23% 45 3.41% 2.91% 3.57%  14.77% 4.36% 8.66% 47 0.31% 0.46% 0.49%  5.35% 2.82% 4.42% 58 0.06% 0.07% 0.07%  5.51% 4.65% 7.01% 59 0.00% 0.00% 0.00%  5.24% 5.41% 7.12% 69 1.37% 2.40% 2.87%  3.66% 6.67% 11.93% 76 0.01% 0.02% 0.02%  3.31% 3.12% 4.18% 77 0.01% 0.02% 0.02%  2.04% 2.08% 2.46% 84 0.04% 0.05% 0.04%  12.52% 17.29% 27.13% 86 0.30% 1.02% 1.00%  21.04% 20.37% 33.12% 87 0.01% 0.02% 0.01%  13.92% 25.11% 29.68% 92 0.01% 0.01% 0.01%  4.82% 5.09% 7.75%    81  Table 4.18 Comparison of Grades: XRF vs. ICP, Myra Falls D, Zinc & Copper Sample ID Zinc  Copper Rock Pul. ICP  Rock Pul. ICP 2 0.08% 0.08% 0.04%  0.01% 0.01% 0.01% 7 0.01% 0.01% 0.01%  0.00% 0.01% 0.01% 11 11.55% 10.11% 8.93%  0.07% 0.11% 0.13% 13 0.00% 0.00% 0.00%  0.00% 0.00% 0.00% 19 0.65% 1.16% 0.74%  0.56% 0.75% 0.61% 22 0.07% 0.04% 0.02%  0.36% 0.60% 0.38% 23 54.10% 43.62% 51.42%  0.13% 0.07% 0.21% 25 0.06% 0.06% 0.04%  0.00% 0.00% 0.00% 45 1.20% 2.20% 1.73%  0.23% 0.11% 0.10% 59 22.25% 25.73% 25.79%  3.71% 4.02% 4.66% 64 0.17% 0.04% 0.02%  0.02% 0.01% 0.00% 73 0.98% 7.92% 7.31%  0.05% 0.17% 0.17% 74 0.10% 0.26% 0.17%  0.01% 0.02% 0.02% 75 0.05% 0.03% 0.02%  0.00% 0.02% 0.01% 78 0.03% 0.02% 0.01%  0.01% 0.01% 0.01% 79 4.93% 1.79% 0.97%  4.79% 5.96% 4.27% 80 0.10% 0.10% 0.06%  0.08% 0.09% 0.08% 81 0.10% 0.19% 0.10%  0.02% 0.03% 0.02% 88 0.23% 0.07% 0.04%  0.07% 0.01% 0.01% 91 0.35% 0.84% 0.41%  2.91% 7.18% 4.97% 98 1.38% 0.58% 0.33%  0.21% 0.10% 0.08% 99 1.06% 0.50% 0.25%  0.09% 0.06% 0.03%    82  Table 4.19 Comparison of Grades: XRF vs. ICP, Myra Falls D, Lead & Iron Sample ID Lead  Iron Rock Pul. ICP  Rock Pul. ICP 2 0.00% 0.01% 0.01%  4.64% 5.01% 6.45% 7 0.01% 0.00% 0.00%  7.98% 7.39% 5.13% 11 0.02% 0.02% 0.02%  3.30% 2.26% 4.04% 13 0.00% 0.00% 0.00%  1.52% 0.84% 0.60% 19 0.03% 0.06% 0.05%  2.31% 3.08% 4.12% 22 0.01% 0.00% 0.00%  14.13% 18.89% 26.87% 23 1.23% 3.70% 4.68%  1.79% 1.85% 4.74% 25 0.01% 0.00% 0.00%  4.25% 3.79% 4.45% 45 0.40% 0.67% 0.69%  2.68% 2.42% 4.43% 59 2.10% 2.88% 3.05%  11.07% 6.61% 14.77% 64 0.00% 0.00% 0.00%  5.36% 4.87% 5.21% 73 0.16% 3.03% 3.40%  2.72% 3.43% 6.35% 74 0.02% 0.01% 0.01%  3.47% 2.40% 3.10% 75 0.01% 0.01% 0.01%  10.59% 20.18% 26.96% 78 0.00% 0.00% 0.00%  2.04% 2.17% 2.41% 79 0.07% 0.08% 0.07%  14.79% 20.17% 29.64% 80 0.00% 0.00% 0.00%  5.11% 5.83% 7.97% 81 0.04% 0.01% 0.01%  19.89% 26.93% 27.61% 88 0.02% 0.00% 0.00%  5.91% 4.66% 4.71% 91 0.02% 0.02% 0.02%  18.86% 21.43% 34.19% 98 0.03% 0.02% 0.02%  28.06% 29.27% 31.32% 99 0.03% 0.02% 0.02%  23.55% 25.61% 33.33%  Comparison of Light Elements XRF is said to be most accurate for elements above the atomic number of 20 (calcium). This can be observed in Table 4.20 where the correlation factors are significantly lower than those of heavier elements already studied, with sulfur being an exception. Interestingly, the correlation factors for sulfur was fairly high and that could be attributed to overall high grades of sulfur. A closer look at the distribution of sulfur grades (Figure 4.49 and Figure 4.50) however, proves that despite the close correlation, the XRF readings, whether rock or pulverized, consistently underestimated the actual sulfur grades. This was because of low molecular weight of sulfur that makes its detection harder and less accurate for the XRF sensor. Detection for magnesium (atomic weight of 24) was the least accurate, even in the case of pulverized XRF vs. the ICP assays. While Rock XRF failed to detect any magnesium for most 83  cases (reason: much shorter exposure time than the pulverized XRF), pulverized XRF almost consistently overestimated the magnesium grades with no proper correlation between the two grades (Figure 4.51 and Figure 4.52). After sulfur, XRF appeared to predict calcium grades best, with relatively higher correlation factors and the predicted grades were closer to the actual ones (Figure 4.53 and Figure 4.54).  In the case of potassium, both XRF readings managed to detect the element with mostly similar grades, however, they both overestimated the potassium content (Figure 4.55 and Figure 4.56). This study briefly shows to what extent the XRF readings can be relied on. It is important to mention that these tests were performed statically and that the both XRF readings were averages of 2 to 4 reading per sample. Therefore, while this can be a starting point in evaluating the XRF response, detection of these elements, specifically the light ones, on a dynamic sorter must be investigated. Table 4.20 Correlation factors: XRF vs. ICP – Light Elements, Myra Falls Sample Correlations with ICP Sulfur Magnesium Calcium Potassium Rock Pul. Rock Pul. Rock Pul. Rock Pul. Size A 0.921 0.985 0.242 0.332 0.640 0.741 0.847 0.702 Size D 0.947 0.974 -0.206 0.333 0.644 0.941 0.644 0.941   84  Figure 4.49 Comparison: XRF vs. ICP – Myra Falls, Sulfur, size fraction A  Figure 4.50 Comparison: XRF vs. ICP – Myra Falls, Sulfur, size fraction D    0%5%10%15%20%25%30%35%40%45%50%0 20 40 60 80 100Sulfur GradeRock IDRock XRFPul. XRFICP0%10%20%30%40%50%60%0 20 40 60 80 100Sulfur GradeRock IDRock XRFPul. XRFICP85  Figure 4.51 Comparison: XRF vs. ICP – Myra Falls, Magnesium, size fraction A  Figure 4.52 Comparison: XRF vs. ICP – Myra Falls, Magnesium, size fraction D    0%1%2%3%4%5%6%7%8%9%0 20 40 60 80 100Magnesium GradeRock IDRock XRFPul. XRFICP0%1%2%3%4%5%6%7%8%9%0 20 40 60 80 100Magnesium GradeRock IDRock XRFPul. XRFICP86  Figure 4.53 Comparison: XRF vs. ICP – Myra Falls, Calcium, size fraction A  Figure 4.54 Comparison: XRF vs. ICP – Myra Falls, Calcium, size fraction D    0%1%2%3%4%5%6%0 20 40 60 80 100Calcium GradeRock IDRock XRFPul. XRFICP0%1%2%3%4%5%6%7%0 20 40 60 80 100Calcium GradeRock IDRock XRFPul. XRFICP87  Figure 4.55 Comparison: XRF vs. ICP – Myra Falls, Potassium, size fraction A  Figure 4.56 Comparison: XRF vs. ICP – Myra Falls, Potassium, size fraction D    0%1%2%3%4%5%6%0 20 40 60 80 100Potassium GradeRock IDRock XRFPul. XRFICP0.0%0.5%1.0%1.5%2.0%2.5%3.0%3.5%4.0%4.5%5.0%0 20 40 60 80 100Potassium GradeRock IDRock XRFPul. XRFICP88  Table 4.21 Comparison of Grades: XRF vs. ICP, Myra Falls A, Sulfur & Magnesium Sample ID Sulfur  Magnesium Rock Pul. ICP  Rock Pul. ICP 2 6.97% 9.40% 42.70%  0.00% 8.47% 1.72% 3 0.42% 0.28% 2.42%  0.00% 6.94% 6.58% 4 0.49% 0.14% 1.20%  0.00% 0.00% 0.56% 6 0.27% 0.21% 2.27%  0.00% 4.67% 5.27% 12 0.13% 0.13% 0.39%  5.50% 5.47% 2.17% 17 0.00% 0.00% 0.00%  4.03% 8.47% 4.05% 25 10.27% 6.49% 33.10%  0.00% 0.00% 0.13% 31 1.53% 0.48% 1.71%  0.00% 7.59% 2.14% 34 0.00% 0.27% 1.31%  0.00% 3.22% 0.10% 39 0.83% 0.57% 1.68%  0.00% 3.02% 1.19% 41 14.46% 8.20% 34.30%  0.00% 6.25% 0.00% 45 9.00% 8.10% 31.60%  0.00% 8.35% 0.00% 47 3.59% 1.53% 5.29%  0.00% 3.21% 0.28% 58 3.39% 1.51% 7.86%  0.00% 0.00% 0.13% 59 2.67% 1.15% 7.65%  0.00% 4.30% 0.23% 69 3.81% 3.80% 20.50%  0.00% 0.00% 0.68% 76 1.99% 0.99% 4.77%  0.00% 0.00% 0.17% 77 0.62% 0.41% 2.37%  0.00% 5.40% 0.61% 84 10.90% 7.56% 35.90%  0.00% 0.00% 0.00% 86 13.27% 8.84% 39.40%  0.00% 0.00% 1.28% 87 9.73% 7.58% 43.70%  0.00% 0.00% 0.00% 92 2.13% 0.76% 7.12%  0.00% 0.00% 5.05%   89  Table 4.22 Comparison of Grades: XRF vs. ICP, Myra Falls A, Calcium & Potassium Sample ID Calcium  Potassium Rock Pul. ICP  Rock Pul. ICP 2 0.28% 0.47% 0.34%  0.00% 0.41% 0.00% 3 0.00% 0.23% 0.20%  0.00% 1.93% 0.09% 4 0.84% 4.85% 4.79%  3.58% 4.07% 0.16% 6 0.00% 0.21% 0.19%  1.11% 2.60% 0.09% 12 2.77% 0.00% 2.93%  0.00% 0.90% 0.00% 17 0.28% 1.04% 0.49%  0.00% 0.32% 0.00% 25 3.13% 1.97% 1.81%  0.58% 0.34% 0.00% 31 0.00% 0.19% 0.16%  1.81% 3.29% 0.10% 34 0.00% 0.05% 0.00%  4.66% 3.91% 0.10% 39 0.00% 0.34% 0.34%  3.90% 0.00% 0.18% 41 0.00% 0.14% 0.00%  0.00% 0.00% 0.00% 45 0.00% 0.15% 0.00%  0.00% 0.00% 0.00% 47 0.00% 0.19% 0.12%  2.31% 2.90% 0.07% 58 0.26% 0.53% 0.45%  3.59% 4.58% 0.11% 59 0.49% 0.31% 0.30%  5.03% 4.96% 0.22% 69 0.34% 0.59% 0.54%  1.08% 2.15% 0.08% 76 0.00% 0.15% 0.10%  3.32% 3.27% 0.16% 77 0.00% 0.21% 0.18%  3.06% 4.00% 0.08% 84 1.92% 1.08% 0.71%  0.00% 0.00% 0.00% 86 2.35% 0.00% 2.35%  0.00% 0.36% 0.00% 87 0.00% 0.00% 0.06%  3.20% 3.64% 0.07% 92 0.14% 0.30% 0.33%  4.13% 4.75% 0.11%   90  Table 4.23 Comparison of Grades: XRF vs. ICP, Myra Falls D, Sulfur & Magnesium Sample ID Sulfur  Magnesium Rock Pul. ICP  Rock Pul. ICP 2 0.57% 0.40% 4.35%  0.00% 5.46% 4.46% 7 0.00% 0.00% 0.00%  0.00% 4.04% 2.40% 11 3.47% 1.93% 6.67%  0.00% 1.23% 1.27% 13 0.18% 0.09% 0.56%  7.65% 3.98% 0.06% 19 0.61% 0.63% 3.46%  0.00% 1.15% 1.44% 22 5.59% 5.50% 31.80%  0.00% 1.28% 0.63% 23 10.48% 4.94% 28.00%  0.00% 2.33% 0.10% 25 0.79% 0.40% 2.39%  0.00% 2.15% 3.23% 45 1.50% 1.16% 5.42%  0.00% 0.78% 0.88% 59 8.85% 7.95% 34.00%  0.00% 0.00% 0.00% 64 0.35% 0.33% 0.82%  0.00% 2.70% 3.51% 73 0.96% 2.83% 9.44%  0.00% 0.99% 2.80% 74 1.12% 0.55% 3.41%  0.00% 3.00% 0.10% 75 3.16% 4.86% 30.80%  0.00% 8.08% 3.25% 78 0.13% 0.15% 1.59%  0.00% 4.38% 1.38% 79 9.14% 10.00% 38.90%  0.00% 4.37% 0.00% 80 0.61% 0.39% 4.38%  0.00% 4.34% 4.28% 81 11.30% 8.16% 46.10%  0.00% 5.05% 0.17% 88 0.19% 0.09% 0.27%  0.00% 4.63% 2.24% 91 10.10% 10.02% 41.90%  0.00% 1.58% 0.00% 98 14.97% 11.13% 52.00%  0.00% 1.49% 0.22% 99 13.68% 7.74% 44.30%  0.00% 4.80% 0.00%   91  Table 4.24 Comparison of Grades: XRF vs. ICP, Myra Falls D, Calcium & Potassium Sample ID Calcium  Potassium Rock Pul. ICP  Rock Pul. ICP 2 0.11% 0.45% 0.39%  0.50% 1.97% 0.10% 7 4.30% 3.48% 1.11%  0.00% 0.42% 0.05% 11 2.76% 2.68% 2.30%  1.58% 1.24% 0.06% 13 0.13% 0.41% 0.27%  2.24% 3.84% 0.24% 19 2.01% 2.48% 2.45%  0.00% 2.16% 0.11% 22 0.14% 0.30% 0.13%  0.00% 1.17% 0.07% 23 1.26% 1.10% 0.88%  0.00% 0.61% 0.00% 25 0.33% 0.51% 0.44%  1.36% 2.87% 0.19% 45 0.16% 1.84% 1.55%  0.81% 1.82% 0.07% 59 0.00% 0.24% 0.00%  0.00% 0.07% 0.00% 64 4.83% 3.96% 3.46%  0.41% 1.04% 0.05% 73 1.98% 5.82% 4.81%  0.00% 1.66% 0.06% 74 0.00% 0.18% 0.11%  2.05% 4.63% 0.24% 75 0.70% 0.56% 0.39%  0.85% 0.54% 0.08% 78 1.00% 1.72% 1.81%  2.66% 3.65% 0.24% 79 0.00% 0.53% 0.24%  0.00% 0.00% 0.00% 80 0.00% 0.34% 0.28%  0.36% 2.51% 0.14% 81 0.00% 0.15% 0.05%  0.57% 2.21% 0.08% 88 0.84% 2.41% 2.63%  0.00% 2.87% 0.15% 91 0.89% 0.95% 0.42%  0.00% 0.00% 0.00% 98 1.08% 1.08% 0.79%  0.00% 0.00% 0.00% 99 0.10% 0.14% 0.05%  0.00% 2.42% 0.13%   92   Inductively Couple Plasma - Mass Spectroscopy (ICP-MS) To estimate the true grades of each rock and compare them to their corresponding XRF readings, a 50-element ICP assay was performed on about 20% of the rocks from each size fraction for all the base metal samples. These rocks, roughly 20 from each size fraction, were selected so that they would cover a range of grades of the main metal based on their pulverized XRF readings. The pulverized XRF readings were used on the basis of their close correlation with the actual grades. Table 4.25 to    93  Table 4.28 summarize the average overall grades for each mine. These values are averaged for all the ICP results for all four size fractions. It is important to mention that while the listed values do not represent the actual average grade of the lot, they are listed here solely to present the reader with an idea of the constituent elements in each sample lot. The Limit of Detection (LOD) for each element is also presented in the tables and all the values below the LOD were assumed zero.   94  Table 4.25 Brenda Element Au Ag Al As B Ba Be Bi Ca Cd Ce Unit ppm ppm % ppm ppm ppm ppm ppm % ppm ppm LOD 0.005 0.01 0.01 0.1 10 10 0.05 0.01 0.01 0.01 0.02 Grade 0.22 3.19 1.48 4.20 0.17 179.71 0.09 3.62 1.05 0.88 14.95             Element Co Cr Cs Cu Fe Ga Ge Hf Hg In  Unit ppm ppm ppm ppm % ppm ppm ppm ppm ppm  LOD 0.1 1 0.05 0.2 0.01 0.05 0.05 0.02 0.01 0.005  Grade 12.02 131.78 3.06 4459.57 3.48 6.40 0.13 0.06 0.06 0.09              Element K La Li Mg Mn Mo Na Nb Ni P  Unit % ppm ppm % ppm ppm % ppm ppm ppm  LOD 0.01 0.2 0.1 0.01 5 0.05 0.01 0.05 0.2 10  Grade 0.81 7.84 12.58 1.08 502.53 440.37 0.07 0.14 24.44 626.46              Element Pb Rb Re S Sb Sc Se Sn Sr Ta  Unit ppm ppm ppm % ppm ppm ppm ppm ppm ppm  LOD 0.2 0.1 0.001 0.01 0.05 0.1 0.2 0.2 0.2 0.01  Grade 12.27 48.68 0.02 0.67 0.24 4.68 8.53 0.75 32.83 0.00              Element Te Th Ti Tl U V W Y Zn Zr  Unit ppm ppm % ppm ppm ppm ppm ppm ppm ppm  LOD 0.01 0.2 0.005 0.02 0.05 1 0.05 0.05 2 0.5  Grade 1.48 4.96 0.17 0.32 2.33 85.69 18.03 7.18 58.80 1.27     95  Table 4.26 Copper Mountain Elements Au Ag Al As B Ba Be Bi Ca Cd Ce Unit ppm ppm % ppm ppm ppm ppm ppm % ppm ppm LOD 0.005 0.01 0.01 0.1 10 10 0.05 0.01 0.01 0.01 0.02 Grade 0.22 0.92 0.57 14.37 0.30 30.70 0.56 0.07 2.83 0.51 24.25             Elements Co Cr Cs Cu Fe Ga Ge Hf Hg In  Unit ppm ppm ppm ppm % ppm ppm ppm ppm ppm  LOD 0.1 1 0.05 0.2 0.01 0.05 0.05 0.02 0.01 0.005  Grade 13.57 44.68 1.31 3864.70 1.86 3.37 0.08 0.17 0.21 0.11              Elements K La Li Mg Mn Mo Na Nb Ni P  Unit % ppm ppm % ppm ppm % ppm ppm ppm  LOD 0.01 0.2 0.1 0.01 5 0.05 0.01 0.05 0.2 10  Grade 0.10 10.74 10.96 0.91 563.18 30.79 0.06 0.07 18.81 1317.07              Elements Pb Rb Re S Sb Sc Se Sn Sr ppm  Unit ppm ppm ppm % ppm ppm ppm ppm ppm ppm  LOD 0.2 0.1 0.001 0.01 0.05 0.1 0.2 0.2 0.2 0.01  Grade 9.49 3.92 0.08 0.91 6.32 10.99 4.64 0.54 122.54 0.00              Elements Te Th Ti Tl U V W Y Zn Zr  Unit ppm ppm % ppm ppm ppm ppm ppm ppm ppm  LOD 0.01 0.2 0.005 0.02 0.05 1 0.05 0.05 2 0.5  Grade 0.30 2.73 0.05 0.02 0.90 74.10 0.27 9.87 72.11 5.87     96  Table 4.27 Mount Polley Elements Au Ag Al As B Ba Be Bi Ca Cd Ce Unit ppm ppm % ppm ppm ppm ppm ppm % ppm ppm LOD 0.005 0.01 0.01 0.1 10 10 0.05 0.01 0.01 0.01 0.02 Grade 0.28 0.71 1.55 8.44 4.59 111.85 0.73 0.06 1.99 0.16 16.08             Elements Co Cr Cs Cu Fe Ga Ge Hf Hg In  Unit ppm ppm ppm ppm % ppm ppm ppm ppm ppm  LOD 0.1 1 0.05 0.2 0.01 0.05 0.05 0.02 0.01 0.005  Grade 16.42 31.52 1.78 2974.51 5.91 9.43 0.47 0.28 0.06 0.16              Elements K La Li Mg Mn Mo Na Nb Ni P  Unit % ppm ppm % ppm ppm % ppm ppm ppm  LOD 0.01 0.2 0.1 0.01 5 0.05 0.01 0.05 0.2 10  Grade 0.16 8.65 11.92 0.87 417.05 46.60 0.06 0.06 16.59 1046.72              Elements Pb Rb Re S Sb Sc Se Sn Sr Ta  Unit ppm ppm ppm % ppm ppm ppm ppm ppm ppm  LOD 0.2 0.1 0.001 0.01 0.05 0.1 0.2 0.2 0.2 0.01  Grade 7.29 7.13 0.25 0.25 0.23 3.60 4.73 1.99 229.14 0.00              Elements Te Th Ti Tl U V W Y Zn Zr  Unit ppm ppm % ppm ppm ppm ppm ppm ppm ppm  LOD 0.01 0.2 0.005 0.02 0.05 1 0.05 0.05 2 0.5  Grade 0.16 1.03 0.12 0.01 0.96 160.41 0.48 9.91 37.89 10.38     97  Table 4.28 Myra Falls Elements Au Ag Al As B Ba Be Bi Ca Cd Ce Unit ppm ppm % ppm ppm ppm ppm ppm % ppm ppm LOD 0.005 0.01 0.01 0.1 10 10 0.05 0.01 0.01 0.01 0.02 Grade 1.44 33.03 1.33 210.55 0.00 18.23 0.02 7.17 0.86 199.61 6.06             Elements Co Cr Cs Cu Fe Ga Ge Hf Hg In  Unit ppm ppm ppm ppm % ppm ppm ppm ppm ppm  LOD 0.1 1 0.05 0.2 0.01 0.05 0.05 0.02 0.01 0.005  Grade 12.25 81.97 0.13 8882 12.74 6.74 0.40 0.01 1.98 1.56              Elements K La Li Mg Mn Mo Na Nb Ni P  Unit % ppm ppm % ppm ppm % ppm ppm ppm  LOD 0.01 0.2 0.1 0.01 5 0.05 0.01 0.05 0.2 10  Grade 0.07 2.28 7.01 1.54 868.81 29.97 0.00 0.00 57.16 461.11              Elements Pb Rb Re S Sb Sc Se Sn Sr Ta  Unit ppm ppm ppm % ppm ppm ppm ppm ppm ppm  LOD 0.2 0.1 0.001 0.01 0.05 0.1 0.2 0.2 0.2 0.01  Grade 4637.01 1.30 0.01 17.82 36.07 3.21 10.41 0.23 25.09 0.00              Elements Te Th Ti Tl U V W Y Zn Zr  Unit ppm ppm % ppm ppm ppm ppm ppm ppm ppm  LOD 0.01 0.2 0.005 0.02 0.05 1 0.05 0.05 2 0.5  Grade 2.06 0.08 0.02 0.93 0.29 37.10 0.57 2.99 47641 0.53  For each size fraction, a fitted equation was developed relating the ICP results and the pulverized XRF readings for the main elements (e.g. copper, iron, sulfur, etc.). This equation was then used to estimate the grades for these elements for the rest of the rocks in that size fraction. To show these correlations, only the results from Brenda (smallest size fraction, -25 mm +19 mm) for copper, iron and sulfur, and Myra Falls (largest size fractions, -75 mm +50 mm) for zinc, iron and sulfur, are presented here.   98  Figure 4.57 ICP vs. Pulverized XRF copper correlation for Brenda Size A   Figure 4.58 ICP vs. Pulverized XRF iron correlation for Brenda Size A    y = 1.0265xR² = 0.99940123456789100 2 4 6 8 10ICP Grade (%)XRF Pulverized Grade (%)CopperLinear (Copper)y = 1.0726x - 0.6795R² = 0.76210123456789100 2 4 6 8ICP GradeXRF Pulverized Grade (%)IronLinear (Iron)99  Figure 4.59 ICP vs. Pulverized XRF sulfur correlation for Brenda Size A  Figure 4.60 ICP vs. Pulverized XRF Zinc correlation for Myra Falls Size D    y = 1.3717x + 0.2008R² = 0.87320.00.51.01.52.02.53.03.54.04.50.0 1.0 2.0 3.0 4.0ICP Grade (%)XRF Pulverized Grade (%)SulfurLinear (Sulfur)y = 1.1173xR² = 0.991301020304050600 10 20 30 40 50IMS Grade (%)XRF Pulverized Grade (%)ZincLinear (Zinc)100  Figure 4.61 ICP vs. Pulverized XRF iron correlation for Myra Falls Size D  Figure 4.62 ICP vs. Pulverized XRF sulfur correlation for Myra Falls Size D    y = 1.2197x + 0.9633R² = 0.9305101520253035400 10 20 30 40IMS GradeXRF Pulverized Grade (%)IronLinear (Iron)y = 4.6329x + 1.0617R² = 0.949201020304050600 2 4 6 8 10 12ICP Grade (%)XRF Pulverized Grade (%)SulfurLinear (Sulfur)101  To estimate the gold and silver grades, the fire assay results for the 20 % assayed rocks, for each size fractions, were used in a Multivariate Linear Regression analysis against their pulverized XRF readings. The resultant equations from the MLR analysis were then used to estimate the gold and silver grades for the rest of the rocks. The results for the gold and silver grades are not discussed in this thesis, however can be used to perform hypothetical economic studies.  Rietveld Refinement (XRD) For the Rietveld Refinement tests, two rocks, one low-grade and one high-grade in their respective main base metal (i.e. copper or zinc), were selected from the base metal mines material. Since the assayed gold samples were not available, four rocks that visually looked low/high-grade were selected and analysed, as well. The samples were ground under ethanol in a vibratory McCrone Micronizing Mill for 10 minutes to reduce to the optimum grain size of <10 m for quantitative X-ray analysis. Step-scan X-ray powder-diffraction data were collected over a range 3-80°2 with CoKα radiation on a Bruker D8 Advance Bragg-Brentano diffractometer equipped with an Fe monochromator foil, 0.6 mm (0.3°) divergence slit, incident- and diffracted-beam Soller slits and a LynxEye-XE detector. The long fine-focus Co X-ray tube was operated at 35 kV and 40 mA, using a take-off angle of 6°. The X-ray diffractograms were analyzed using the International Centre for Diffraction Database PDF-4 and Search-Match software by Bruker. X-ray powder-diffraction data of the samples were refined with Rietveld program Topas 4.2 (Bruker AXS). The results of quantitative phase analysis by Rietveld refinements are given in the tables below. The grade for the main metal in each rock is also included in the tables, except for the gold samples. The values presented in the tables below indicate the relative amounts of crystalline phases normalized to 100%.   102  Table 4.29 XRD analysis for Brenda BR-D01 - Cu: 5.2% Mineral Formula Grade Quartz  SiO2 31 Chalcopyrite CuFeS2 15.5 Plagioclase (albite, oligoclase) NaAlSi3O8 – CaAl2Si2O8 20 K-feldspar KAlSi3O8  9 Biotite 1M K(Mg,Fe2+)3AlSi3O10(OH)2 9.3 Molybdenite  MoS2 0.2 Gypsum CaSO4·2H2O 1.1 Calcite CaCO3 6.3 Clinochlore  (Mg,Fe2+)5Al(Si3Al)O10(OH)8 1.6 Illite/Muscovite 2M1 K0.65Al2.0Al0.65Si3.35O10(OH)2 – KAl2AlSi3O10(OH)2 1.7 Kaolinite 1A Al2Si2O5(OH)4 1 Pyrite FeS2 0.4 Clinozoisite Ca2Al3(SiO4)3(OH)  3    BR-C58 - Cu: 10 ppm Mineral Formula Grade Quartz  SiO2 24.8 Plagioclase (albite, oligoclase) NaAlSi3O8 – CaAl2Si2O8 54.2 Actinolite Ca2(Mg,Fe2+)5Si8O22(OH)2 4.9 K-feldspar KAlSi3O8 11.2 Biotite 1M K(Mg,Fe2+)3AlSi3O10(OH)2 2.2 Clinochlore  (Mg,Fe2+)5Al(Si3Al)O10(OH)8 0.6 Illite/Muscovite 2M1 K0.65Al2.0Al0.65Si3.35O10(OH)2 – KAl2AlSi3O10(OH)2 2.1    103  Table 4.30 XRD analysis for Copper Mountain CM-B20 – Cu: 2.1% Mineral Formula Grade Pyrite FeS2 2 Quartz  SiO2 6.4 Chalcopyrite CuFeS2 8.2 Plagioclase (albite, oligoclase) NaAlSi3O8 – CaAl2Si2O8 29.2 K-feldspar KAlSi3O8 29.1 Augite (Ca,Na)(Mg,Fe,Al,Ti)(Si,Al)2O6 4.9 Magnetite Fe3O4 0.6 Clinochlore  (Mg,Fe2+)5Al(Si3Al)O10(OH)8 13.3 Illite/Muscovite 2M1 K0.65Al2.0Al0.65Si3.35O10(OH)2 – KAl2AlSi3O10(OH)2 1.4 Clinozoisite Ca2Al3(SiO4)3(OH) 4.8    CM-D85 – Cu: 373 ppm Mineral Formula Grade Quartz  SiO2 33 Plagioclase (albite, oligoclase) NaAlSi3O8 – CaAl2Si2O8 19.8 Calcite CaCO3 8.5 Kaolinite 1A Al2Si2O5(OH)4 3.2 Ankerite Ca(Fe2+,Mg,Mn)(CO3)2 3.9 Hematite Fe2O3 1.1 Siderite Fe2+CO3  0.6 Illite/Muscovite 2M1 K0.65Al2.0Al0.65Si3.35O10(OH)2 – KAl2AlSi3O10(OH)2 11.3 Anatase TiO2 0.9 Illite/Muscovite 1M K0.65Al2.0Al0.65Si3.35O10(OH)2 – KAl2AlSi3O10(OH)2 17.7    104  Table 4.31 XRD analysis for Mount Polley MP-D55 – Cu: 1.3% Mineral Formula Grade Clinochlore  (Mg,Fe2+)5Al(Si3Al)O10(OH)8 6 Plagioclase (albite, oligoclase) NaAlSi3O8 – CaAl2Si2O8 18.5 Magnetite Fe3O4 16.8 K-feldspar KAlSi3O8 34.5 Diopside CaMgSi2O6 7.1 Hematite Fe2O3 1.9 Chabazite Ca2Al4Si8O24 13H2O 2.1 Chalcopyrite CuFeS2 3.8 Molybdenite  MoS2 0.1 Prehnite Ca2Al2Si3O10(OH)2 5.2 Clinozoisite Ca2Al3(SiO4)3(OH) 4    MP-D48 – Cu: 114 ppm Mineral Formula Grade Plagioclase (albite, oligoclase) NaAlSi3O8 – CaAl2Si2O8 14.5 K-feldspar KAlSi3O8 9.5 Diopside CaMgSi2O6 43.9 Hematite Fe2O3 5.3 Clinochlore  (Mg,Fe2+)5Al(Si3Al)O10(OH)8 17 Calcite CaCO3 1.3 Vermiculite 2M (Mg,Fe2+,Al)3(Si,Al)4O10(OH)2∙4H2O  3.1 Quartz  SiO2 0.4 K-feldspar KAlSi3O8 4.8 Siderite ? Fe2+CO3 0.3    105  Table 4.32 XRD analysis for  MF-D59 - Zn: 25.8% Mineral Formula Grade Quartz  SiO2 10 Pyrite FeS2 24.9 Sphalerite (Zn,Fe)S 36 Barite BaSO4 8 Galena PbS 2.7 Chalcopyrite CuFeS2 13.5 Anglesite PbSO4 1.3 Illite/Muscovite 2M1 K0.65Al2.0Al0.65Si3.35O10(OH)2 – KAl2AlSi3O10(OH)2 3.7    MF-D13 - Zn: 28 ppm Mineral Formula Grade Quartz  SiO2 31.4 Illite/Muscovite 2M1 K0.65Al2.0Al0.65Si3.35O10(OH)2 – KAl2AlSi3O10(OH)2 29.4 Plagioclase  (albite, oligoclase)  NaAlSi3O8 – CaAl2Si2O8  36.4 Pyrite FeS2 1.3 Dolomite CaMg(CO3)2 0.7 Rutile TiO2  0.5 Calcite CaCO3  0.3    106  Table 4.33 XRD analysis for two visually mineralized gold samples Mineral Ideal Formula Rock #1 Rock #2 Andalusite Al2SiO5  2.1 Biotite K(Mg,Fe2+)3AlSi3O10(OH)2 3.6 4.5 Calcite CaCO3 2.2  Clinochlore (Mg,Fe2+)5Al(Si3Al)O10(OH)8 2.2 4.4 Cordierite Mg2Al4Si5O18  2.4 Dolomite CaMg(CO3)2 25.3  Ilmenite Fe2+TiO3 0.1 0.7 Magnetite Fe3O4 0.1 0.5 Plagioclase (Anorthite) NaAlSi3O8 – CaAl2Si2O8 6.4 18.9 Pyrite FeS2 0.7 11.3 Pyrrhotite Fe1-xS 0.4 2.0 Quartz SiO2 59.0 53.2 Total  100.0 100.0 Table 4.34 XRD analysis for two visually gangue gold samples Mineral Ideal Formula No.3 No.4 Actinolite Ca2(Mg,Fe2+)5Si8O22(OH)2  6.3 Andalusite Al2SiO5 0.6  Anthophyllite Mg7Si8O22(OH)2  1.1 Arsenopyrite FeAsS 9.0  Biotite K(Mg,Fe2+)3AlSi3O10(OH)2 1.5 8.1 Clinochlore (Mg,Fe2+)5Al(Si3Al)O10(OH)8 1.4 3.1 Cummingtonite Mg7Si8O22(OH)2  3.0 Dolomite CaMg(CO3)2  18.6 Illite/Muscovite 2M K0.65Al2.0Al0.65Si3.35O10(OH)2 /KAl2AlSi3O10(OH)2 23.3  Ilmenite Fe2+TiO3  0.6 Magnetite Fe3O4  0.6 Plagioclase (Anorthite) NaAlSi3O8 – CaAl2Si2O8 1.9 24.4 Pyrrhotite Fe1-xS 6.0  Quartz SiO2 56.0 34.2 Rutile TiO2 0.3  Total  100.0 100.0  Constitution Heterogeneity Of all the factors that are mentioned as reasons for sorting inefficiency, heterogeneity has never attracted much attention. Heterogeneity is generally considered in Theory of Sampling to measure the sampling error, but it is important to consider its effects on sorting efficiency. 107  When speaking of heterogeneities of a lot (stockpile, in our instance), two types of heterogeneities, Constitution Heterogeneity (CH) and Distribution Heterogeneity (DH), can be considered. For a lot (stockpile) of particulate materials, constitution heterogeneity is the difference between the composition of various units of the lot (aka particles) and not within the units, for example the grade of each rock compared to the other rocks and average grade of the lot. The distribution heterogeneity relates to the spatial distribution of grades in a lot due to their intrinsic properties such as shape, density, weight, etc. (Pitard, 1993). Distribution heterogeneity is not the focus in this thesis. Constitution heterogeneity was first introduced in the Theory of Sampling by Pierre Gy (Gy, 1982) to measure the uncertainty that is caused by sampling operations and is defined as the relative and dimensionless variance of the heterogeneities associated with each individual fragment (Fi) making up the lot (Nf fragments) (Gy, 1992) 𝐶𝐻𝐿 = 𝑠2(ℎ𝑖) =1𝑁𝐹∑ ℎ𝑖2𝑖 =1𝑁𝐹∑(𝑎𝑖−𝑎𝐿)2𝑎𝐿2𝑖 .𝑀𝑖2𝑀𝐿2   Equation 4-1 Where αi and αL are the grades of the fragment i and the lot, respectively, while Mi and ML are the masses of fragment i and the lot. One should also remember that in the above equation, NF should be substituted by NF-1 if the fragments are part of a much larger lot. To put the concept of heterogeneity into perspective, consider the following grade distributions for two different samples and their associated constitution heterogeneity values. It can be seen from Figure 4.63 that all the grades except for one are well spread between grades of zero and 0.76% with only one rock at 2.47. This distribution, considering the mass of each rock, leads to a constitution heterogeneity value of 0.82. Now looking at Figure 4.65, except for almost a dozen particles, the rest contain around 0% zinc. This significant difference in the grades leads to a constitution heterogeneity value of 21.42. Please note that although the scales of grades for these two deposits are different, they do not affect the value of heterogeneities, as constitution heterogeneity is a relative value.   108  Figure 4.63 Mount Polley copper grade distribution (size fraction B)  Figure 4.64 Mount Polley copper grade distribution Histogram (size fraction B)    00.511.522.530 20 40 60 80 100 120Copper Grade (%)Rock #Cu Distribution05101520253035400.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1.2 1.6 2 2.2 2.5Frequency (%)Copper Grade (%)Cu Distribution109  Figure 4.65 Myra Falls zinc grade distribution (size fraction C)  Figure 4.66 Myra Falls zinc grade distribution histogram (size fraction C)  Explaining the concept of heterogeneities in greater detail is out of the scope of this study and it is suggested that interested individuals should refer to the books by Pierre Gy (Gy, 1982) (Gy, 1992) and Francis Pitard (Pitard, 1993) for further information. In this study, only the main metal, aka copper and zinc, for the respective deposits are considered and the secondary metals were not included. Table 4.35 summarizes the calculated 05101520253035400 20 40 60 80 100 120Zinc Grade (%)Rock #Zn Distribution0102030405060708090100Frequency (%)Zn Grade (%)Zn Distribution110  average grades of each material based on the calculated grades through the pulverized XRF data. Table 4.35 Summary of the calculated grades of collected material (all size fractions) Operation Commodity  Copper Zinc Brenda 0.14% - Copper Mountain 0.13% - Mount Polley 0.19% - Myra Falls - 1.81%  Heterogeneity and Sorting Potential In this section, the grade-recovery curves for all 4 size-fractions combined for each mine, along with their constitution heterogeneity values, are demonstrated. This is a crude way of looking at the recovery curves in an ideal separation situation regardless of which sensor is used. As mentioned earlier, constitution heterogeneity is a relative number and independent of the actual grade ranges. Brenda rocks were quite heterogeneous (CH=37) and as it can be seen in Figure 4.67, high recoveries at fairly low mass pulls can be achieved. Figure 4.67 Ideal copper grade-recovery curves for Brenda (all size fractions)  For the case of Copper Mountain, although the heterogeneity value was not as high as that of Brenda, still with CH value of 4.53, good separation potential was observed (Figure 4.68). 01234567891001020304050607080901000 20 40 60 80 100Concentrate Grade (%)Copper Recovery (%)Mass Pull to Concentrate (%)Cu Recovery (%)Concentrate GradeCH=37 111  Figure 4.68 Ideal copper grade-recovery for Copper Mountain (all size fractions)  Mount Polley on the other hand, had the lowest degree of heterogeneity (CH value of 2) among all materials and as observed in Figure 4.69, a less efficient recovery compared with the previous two operations. This simply can lead to two facts, one, that the grades of the rocks are too close to each other to be differentiated efficiently by the sensor, and two, even if the sensors are capable of recognizing the difference, the even distribution of the grades would make it an inefficient process to sort this specific ore because there is a significant amount of copper in the rocks that are individually just below the cut-off grade. The concentrate grade of the ore showed that at 100% recovery, the ore grade is only slightly above the cut-off grade (0.1%) and the grades are so well distributed among the rocks that separation at 40% mass pull, as an example, results in losing a significant amount of copper (30%). Now comparing this result to the first two operations, one would conveniently conclude that it is possible to separate 85% to 90% of the copper in only 30% mass, while this number is only around 60% for Mount Polley’s case.   0.00.51.01.52.02.53.03.54.04.55.001020304050607080901000 20 40 60 80 100Concentrate Grade (%)Copper Recovery (%)Mass Pull to Concentrate (%)Cu RecoveryConcentrate GradeCH=4 112  Figure 4.69 Ideal copper grade-recovery curves for Mount Polley (all size fraction)  The last operation examined here is the waste rock samples from Myra Falls which was a massive sulfide zinc deposit. The grade distribution of rocks indicated a significant difference between the mineral-bearing rocks and the gangue material. This difference generated a constitution heterogeneity value of 39 that in turn hinted at a very effective recovery curve. This can be observed by the steep recovery curve in Figure 4.70. Figure 4.70 Ideal grade-recovery curves for Myra Falls (all size fractions)  0.00.51.01.52.02.53.03.54.04.55.001020304050607080901000 20 40 60 80 100Concentrate Grade (%)Copper Recovery (%)Mass Pull to Concentrate (%)Cu RecoveryConcentrate Grade010203040506070809010001020304050607080901000 20 40 60 80 100Concentrate Grade (%)Zinc Recovery (%)Mass Pull to Concentrate (%)Zn RecoveryConcentrate GradeCH=39CH=2 113  Table 4.36 summarizes CH values as well as the recoveries for each of these operations at different mass pull ratios. Table 4.36 Recovery at different mass pulls for with regard to their CH value for all samples Operations CH Mass Pull 5% 10% 20% 30% 40% 50% Mount Polley 2 23% 31% 47% 59% 69% 77% Copper Mountain 4 39% 55% 73% 85% 92% 96% Brenda 37 68.5% 76% 85% 90% 94% 96% Myra Falls 39 83% 93% 97% 98% 98% 99% One important fact that can be concluded here is that, since massive sulfide deposits there are significant grade differences between the desired ore and the gangue, separation of particles based on sensor responses seems more promising. Meanwhile, for porphyry deposits this difference is not as accentuated and that could make the recovery of mineral-bearing metals less effective. Also for a porphyry deposit, there may be a significant number of low-grade rocks that although they contain a large amount of metal in total, individually their metal contents are below the detection limits of the sensors. Overall, although heterogeneity can be used as a fair estimation of sorting recovery, it is by no means a definitive factor. For one reason, although constitution heterogeneity is independent of grade ranges, different sets of data can produce similar heterogeneity values corresponding to different recovery curves. Therefore, CH values should be considered more as a range to indicate recoveries qualitatively rather than an absolute value to predict the actual recovery.  For the only gold sample examined, the CH value of 8.826 was obtained, which could be translated into good potential for efficient sorting. The grade histogram (Figure 4.71) shows a skewed, narrow unimodal distribution which is favorable in sorting situations. Also, with 80% of the rocks below 1 g/t and only a small percentage between 2 to 4 g/t (around the cut-off grade of 3 g/t), a good sorting efficiency can be expected. The good sorting potential can be seen in Figure 4.72, gold liberation curve, with a sharp increase in recoveries at low mass pulls. Figure 4.72 shows grade-recovery curve for the most ideal case of gold concentration. 98% gold recovery can be achieved at only 40% mass pull. It is important to remember that this is the ideal, theoretical case and actual sorting procedure always falls short from this.   114  Figure 4.71 Gold grade distribution histogram  Figure 4.72 Ideal gold grade-recovery curve   Heterogeneity and Particle Size When it comes to heterogeneity and comminution, there are two different views as whether reduction in average grain-size would increase or decrease constitution heterogeneity. While Petersen et al (Petersen, Minkkinen, & Esbensen, 2005) mentioned that the reduction in average particle size is the most important factor in reducing CH, Pitard (Pitard, 1993) stated that comminution can only increase the value of constitution heterogeneity. In this section, 0%10%20%30%40%50%60%70%80%90%1 2 3 4 5 7.5 10 15 20 25.3FrequencyAu Grade [gpt]Au Distribution051015202530354045500%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Au Grade [gpt]RecoveryMass PullAu-FA RecoveryAu-FA Grade115  datasets from all operations are presented with heterogeneities of each size fraction for comparison. The grade-recovery curves for the four size fractions from each mine are presented in the following four figures. Table 4.37 summarizes the actual sizes for each of the four size fractions. Table 4.37 Measurements for each size fraction Size Fraction Measurements A -25 mm +19 mm B -37.5 mm +25 mm C -50 mm +37.5 mm D -75 mm +50 mm For Brenda, as observed in Figure 4.73 size fraction B (-37.5+2.5 cm) had the lowest CH while the other size fractions have fairly similar values (10 – 15). Figure 4.73 Grade-recovery curves for all size fractions for Brenda and their CH values  For Copper Mountain, CH values seemed to almost increase with decreasing size fractions with the two finest fraction showing CH values of 6.62 and 6.41 Figure 4.74.   0%1%2%3%4%5%6%7%8%9%10%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Concentrate gradeCopper Recovery (%)Mass Pull to Concentrate (%)Ideal Rec. (D)Ideal Rec. (C)Ideal Rec. (B)Ideal Rec. (A)Con. Grade (D)Con. Grade (C)Con. Grade (B)Con. Grade (A)CH=15.37 CH=10.37 CH=3.75 CH=14.26 116  Figure 4.74 Grade-recovery curves for all size fractions for Copper Mountain and their CH values  The four size fractions for Mount Polley did not seem to differ in their CH values significantly. All CH values were between 0.82 and 1.31. This closeness in CH values, which also means weak sorting potential, could also be seen by the similar recovery curves in Figure 4.75. Figure 4.75 Grade-recovery curves for all size fractions for Mount Polley and their CH values  0.0%0.5%1.0%1.5%2.0%2.5%3.0%3.5%4.0%4.5%5.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Concentrate GradeCopper RecoveryMass Pull to ConcentrateIdeal Rec. (D)Ideal Rec. (C)Ideal Rec. (B)Ideal Rec. (A)Con. Grade (D)Con. Grade (C)Con. Grade (B)Con. Grade (A)0.0%0.5%1.0%1.5%2.0%2.5%3.0%3.5%4.0%4.5%5.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Concentrate GradeCopper RecoveryMass Pull to ConcentrateIdeal Rec. (D)Ideal Rec. (C)Ideal Rec. (B)Ideal rec. (A)Con. Grade (D)Con. Grade (C)Con. Grade (B)Con. Grade (A)CH=1.22CH=0.94CH=0.82CH=1.31CH=2.61 CH=3.28 CH=6.62 CH=6.41 117  Myra Falls by far was the most heterogeneous sample. With CH values ranging from 17 to 21, although high sorting potential was expected, a clear trend among the individual size fractions was not observed (Figure 4.76). Figure 4.76 Grade-recovery curves for 4 size fractions for Myra Falls and their CH values  Based on these results, a definitive conclusion cannot be drawn to relate CH values to size fractions. A clear trend was not observed within either of these sample sets and that could be attributed to a few factors. The most probable cause for not seeing a pattern among the size fraction could be the less than ideal sampling procedure that failed to represent the stockpile. Other reasons could be that the size fractions studies were either two narrow or too wide for CH values to differ inherently.   Heterogeneity of Gangue Elements When it comes to sorting of a specific ore deposit, we can either look for the metal of interest, such as copper, zinc or focus on the rejection of waste, such as silica. So far this research focus has been on the metal of interest, however, it could be benefitial at times to reject what is considered gangue. The gangue elements generaly in base metal deposits are Si, Mn, Mg, and Ca. Iron is not considered as the major copper mineral in the deposites studied here was chalcopyrite. 0%10%20%30%40%50%60%70%80%90%100%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Concentrate Grade (%)Zinc Recovery (%)Mass Pull to Concentrate (%)Ideal Rec. (D)Idea Rec. (C)Ideal rec. (B)Ideal Rec. (A)Con. Grade (D)Con. Grade (C)Con. Grade (B)Con. Grade (A)CH=20.8 CH=21.4 CH=17 CH=19 118  To investigate if there is any correlation between copper grades and gangue grades, gangue grades were plotted versus the base metal of interest (i.e. copper and zinc). For all of the samples studies, a clear negative trend between the gangue elements and base metals were observed. The gangue-to-metal correlation coefficients were too low to be significant (<0.3). For the purpose of keeping this thesis brief, these results are not presented here. These results clearly indicated that there were no close correlations between the gangue minerals and the metals of interest. The constitution heterogeneity values of the gangue elements also re-confirmed that there was not enough heterogeneity in grades of those elements for them to be differentiated from each other, suggesting a weak degree of sortability if the basis for separation is their grade.  Despite these observations, isorting based on the gangue material is an important perspective to keep in mind. These studies were based on grades of the rocks only, while a piece of quartz, as an example, will look completely different under an x-ray transmission (XRT) sensor and that would give more feasible options regarding rejecting the waste as opposed to accepting the ore. Therefore, there will always be the possiblity of using a different sensor for this purpose.  Conclusion In this work, the relations between the consititution heterogeneity values and corresponding recoveries of the metal of interest were explored. It was observed that more effiecient sorting recoveries were achieved where the CH values were higher. It is important to remember that CH values are relative and a function of grade, particle mass and number of particles. For this reason, similar CH values can be generated from different distributions. Therefore, CH values cannot be used as a definitive measure to predict the recoveries, but rather as an indicator to qualitatively expect potential for sorting.  The other concept that was investigated in this chapter was a supposed correlation between particle size and constitution heterogeneity. Although there are two different ideas on the possible effects of comminution, no specific trend was observed for these sets of samples. This was attributed to a non-ideal sampling method that was practiced, due to various limitations. It is believed that had proper sampling techniques been used, an increase in the values of heterogeneity would have been observed with reduction in particle size. Other possible causes 119  were identified as too narrow, or two wide, of a size range for the CH values to show any difference. With regards to sorting (rejectcing) gangue material, it was observed that although focusing the sorting solely based on the grades of gangue minerals did not look promising, it would be worthwile to investigate the use of other sensors such as XRT to distinguis between the gangue and mineralized material.   120  5. BRENDA SORTABILITY RESULTS  Optical Sortability Results The host rock at Brenda mine is granodiorite with mineralized vein structure throughout the rocks. Therefore the general visual characteristics of the rocks are quite similar. In the optical tests performed for this research, only one side of the rock was photographed and efforts were made to do so objectively. Therefore, chances of the “wrong” side of the rock being exposed to the camera did exist. This could have been one of the reasons behind the overall unsatisfactory performance of the optical sensor for the material from Brenda mine, despite the rather heterogeneous nature of the rocks in terms of the grade. It was briefly mentioned in the Heterogeneity chapter that, in addition to a heterogeneous nature of the rocks, the sensor response needs to also be heterogeneous as well for an effective sorting to occur. The figures below show the ideal copper recovery and the corresponding sensor’s response for different size fractions for Brenda mine. Constitution Heterogeneity values for both copper grades and sensor’s spectral indices as well as the correlation of the sensor’s response to the grade are also included in each figure. Figure 5.1 Optical recovery of copper – Brenda Size A (-25+19 mm)    0%1%2%3%4%5%6%7%8%9%10%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu GradeCH=14.26 CH (OPT)=0.51 Correlation: 0.347 121  Figure 5.2 Optical recovery of copper – Brenda Size B (-37.5+25 mm)  Figure 5.3 Optical recovery of copper – Brenda Size C (-50+37.5 mm)    0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu GradeCH=3.75CH (OPT)=0.31Correlation: 0.3820%1%1%2%2%3%3%4%4%5%5%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu GradeCH=10.37CH (OPT)=0.09Correlation: 0.205122  Figure 5.4 Optical recovery of copper – Brenda Size D (-75+50 mm)  Of all the size fractions, the sensor responses for only size fractions A and D showed some potential. The spectral indices, although not having the highest heterogeneity value, managed to replicate the ideal recovery curve better than the other two cases. Almost 90% copper recovery could be achieved at 60% of the mass. Table 5.1 summarizes the optical sorting results for Brenda. Table 5.1 Optical sorting summary – Brenda mine Size Fractions CH OPT CH Recovery at Mass Pulls 20% 40% 60% 80% A (-25+19 mm) 14.3 0.51 55% 77% 89% 94% B (-37.5+25 mm) 3.8 0.31 46% 60% 70% 86% C (-50+37.5 mm) 10.4 0.09 48% 66% 74% 88% D (-75+50 mm) 15.4 0.35 74% 80% 88% 94%  Electromagnetic Sortability Result  Size A (-25+19 mm) MLR analysis identified two significant variates, Phase at 200 kHz frequency and interaction effects of Phases at 400 kHz and 600 kHz frequency. The resultant adjusted R2 value also was too low to show any significant correlation between grades and the MLR analysis. 0%2%4%6%8%10%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu GradeCH=15.37CH (OPT)=0.35Correlation: 0.485123  Based on the ideal recovery curve, a good potential for sorting was expected, however, the EM response deviated from the ideal curve resulting a weak recovery potential until around 60% mass pull to the concentrate. These curves can be seen in Figure 5.5 and Table 5.2 summarizes the MLR coefficients and other statistical data. Figure 5.5 EM recovery of copper – sensor for Benda-A (-25+19 mm)  Table 5.2 MLR statistical data for EM – Brenda-A (-25+19 mm) Brenda-A Variate Coefficient P200 -2.326 P400-P600 2.121 Intercept 0.743 RMSE 0.875 Adj. R2 0.083  Size B (-37.5+25 mm) For Size fraction B of Brenda, Magnitude at 100 kHz as well as the interaction effects of Magnitude at 1200 kHz with Phase at 900 kHz were identified as significant factors. The resultant predicted copper grades deviated from the actual grades with an adjusted R2 value of 0.131. As expected, the generated copper recovery curve based on these predicted grades did 0%1%2%3%4%5%6%7%8%9%10%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (EM)Cu GradeCu Grade (EM)124  not yield a good potential for separation. These graphs can be seen in Figure 5.6 with the details of the MLR analysis summarized in Table 5.3. Figure 5.6 EM recovery of copper – sensor for Benda-B (-37.5+25 mm)  Table 5.3 MLR statistical data for EM – Brenda-B (-37.5+25 mm) Brenda-B Variate Coefficient M100 0.514 M1200-P900 -0.426 Intercept 0.329 RMSE 0.238 Adj. R2 0.131  Size C (-50+37.5 mm) No significant variate was identified for Brenda Size C fraction  Size D (-75+50 mm) The results for Brenda Size fraction D looked more promising. With 18 significant variates, the predicted copper recovery curve followed the ideal recovery curve better than the other size fractions, achieving 90% recovery in about 70% mass pull. The list of these variates along with other statistical values are summarized in Table 5.4. It is to be determined whether larger size does actually play a role in the difference in sensor’s response among various size fractions. Figure 5.7 demonstrates the recovery curves based on the actual and predicted copper grades. 0.0%0.2%0.4%0.6%0.8%1.0%1.2%1.4%1.6%1.8%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (EM)Cu GradeCu Grade (EM)125  Figure 5.7 EM recovery of copper – sensor for Benda-D (-75+50 mm)  Table 5.4 MLR statistical data for EM – Brenda-D (-75+50 mm) Brenda-D Variate Coefficient M100 -3.032 M200 28.955 M300 -22.015 M400 -15.846 M700 21.981 M1000 7.836 M1200 -12.044 P700 11.146 P900 -45.024 P1000 15.701 P1200 24.158 M200-M500 22.679 M200-M1000 35.218 M300-M700 -51.918 M900-P600 -17.770 M900-P900 -84.202 M900-P1000 91.511 M1200-P1400 16.586 Intercept -0.202 RMSE 0.205 Adj. R2 0.909 0%1%2%3%4%5%6%7%8%9%10%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (EM)Cu GradeCu Grade (EM)126   X-Ray Transmission Sortability Results Considering that Brenda had a mineralized vein structure, the XRT sensor had the best potential in identifying the mineralized rocks without the risk of blind spots. As can be seen from Table 5.5, XRT yields good recoveries for all four size fractions. Although the two size fractions in the middle (B and C) seemed to have lower recoveries until 40% mass pull, but by 60% mass pull at least 80% of the copper could be recovered. This inefficiency in detecting copper can be seen through lower correlation coefficients of size B and C, as well. The reason behind lower correlation coefficients can be attributed to their fairly consistent iron content (3%-4%). This consistency caused the spectral index to have a rather narrow range. For Brenda size fraction B, the spectral range for the 100 rocks is spread narrowly between 2 to 17, and in the case of size fraction C, 96 out of a 100 has a narrow spectral index range of 2 to 10. This in turn results in a more homogenous distribution of the spectral indices which would result in a less efficient sorting. The distribution of the spectral indices for size fractions A and D are also narrow however there is a more number of true positives that helps achieve better sorting results. Table 5.5 Summary of the XRT sorting results – Brenda Fractions Responses CH Correlation Cu Recovery at Set Mass Pulls 20% 40% 60% 80% Size A Ideal 14 - 87% 96% 99% 100% Spectral 0.42 0.961 73% 79% 89% 94% Size B Ideal 3.75 - 79% 93% 98% 100% Spectral 0.09 0.633 53% 65% 88% 93% Size C Ideal 10.37 - 72% 86% 94% 98% Spectral 0.38 0.653 47% 69% 80% 90% Size D Ideal 15.37 - 87% 95% 98% 99% Spectral 0.33 0.741 75% 84% 90% 96% Figure 5.8 to Figure 5.11 demonstrate the grade-recovery curves for Brenda’s four size fractions.   127  Figure 5.8 XRT recovery of copper – Brenda Size A (-25+19 mm)  Figure 5.9 XRT recovery of copper – Brenda Size B (-37.5+25 mm)    0%1%2%3%4%5%6%7%8%9%10%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade0.0%0.2%0.4%0.6%0.8%1.0%1.2%1.4%1.6%1.8%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade128  Figure 5.10 XRT recovery of copper – Brenda Size C (-50+37.5 mm)  Figure 5.11 XRT recovery of copper – Brenda Size D (-75+50 mm)   X-Ray Fluorescence Sorting Results Two different XRF results are analysed here, one, is the single-variate analysis in which the copper Rock XRF readings were used as the sensor’s response, rocks were ordered based on this response in a descending manner, and the grade-recovery curves were generated based on the corresponding true grade of each rock. This set is annotated by “(XRF)” on the graphs and tables. For the second set of the results, the full Rock XRF readings of all elements and their 0.0%0.5%1.0%1.5%2.0%2.5%3.0%3.5%4.0%4.5%5.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade0%1%2%3%4%5%6%7%8%9%10%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade129  interaction effects were used in a Multivariate Linear Regression (MLR) Analysis algorithm against the true copper grade of the rocks. The algorithm, predicted the copper grade using multiple factors. This predicted copper grade then was used as the sensor’s response, and the grade-recovery curves were developed similar to the first case. This set of results is annotated by “(MLR)” on the graphs and tables. These two datasets are then compared against the ideal copper recovery. As it was mentioned in the Experimental Methods Chapter, the advantage of using MLR analysis is that since it has more degrees of freedom, it can possibly predict the true grades based on multiple factors better.  Size A (-25+19 mm) Figure 5.12 compares the XRF (blue) and MLR (green) datasets with the ideal copper recovery curve (yellow). As it was expected, the MLR manages to predict the copper grades better than just the XRF reading of copper. However, the advantage of MLR for this size fraction is only up to about 35% mass pull. From that point on both curves follow each other fairly closely. Neither XRF nor MLR, despite its rather high adjusted R2, managed to follow the ideal curve closely. Analysing Figure 5.12, it is observed that the MLR curve overlaps the ideal copper recovery curve at low mass pulls, but then as the mass pull increases, the MLR curve deviates from the ideal recovery curve. Examining the data points out to a few high-grade particles that are very well predicted by the algorithm and hence the overlapping of two curves at low mass pulls. However, for the lower grade particles, the algorithm did not predict their grades as well and therefore started to deviate from the ideal recovery curve. The reason behind a weak prediction is believed to be presence of the few high grade particles. It is understood that the algorithm prioritizes matching the higher grade particles rather than the lower grades. Through this, the predicted values for some of the lower grades deviate more from the actual grades. This can be seen in Table 5.6 where rocks #77 and #16 are predicted to have higher grades than they actually do.   130  Figure 5.12 XRF Copper Recovery, Brenda A – Single vs. Multivariate Analysis   Table 5.6 Predicted Cu grades vs. true grades – Brenda A 15 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade 29 8.978% 8.979% 2 1.188% 1.209% 53 1.143% 1.128% 44 0.945% 0.945% 96 0.621% 0.604% 26 0.608% 0.608% 34 0.494% 0.522% 97 0.309% 0.193% 59 0.245% 0.247% 12 0.198% 0.167% 64 0.182% 0.689% 98 0.181% 0.198% 77 0.176% 0.034% 16 0.171% 0.082% 63 0.165% 0.175% The MLR algorithm identified eleven significant factors as seen in Table 5.7. While more mineralogical data is required to completely interpret the results, closely examining this list reveals interesting points. For example, the presence of interaction of copper and iron (Cu-Fe) indicates the presence of copper-iron minerals, potentially chalcopyrite and bornite as they were visually observed in the rocks. Another example would be the interaction of iron and 0%1%2%3%4%5%6%7%8%9%10%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)131  sulfur (Fe-S) which indicated presence of iron sulphide minerals (pyrite or pyrrhotite) which has a negative effect on predicting copper grades. The other example would be interaction effect of molybdenum and sulfur (molybdenite) which has a positive effect on copper, meaning presence of molybdenite indicates presence of copper. Of the 4 major coefficients, the interaction of copper and molybdenum was identified to have a negative effect on copper. The other interactions are not as significant and require more mineralogical data to be interpreted. Table 5.7 MLR analysis coefficients and statistical data – Brenda A Element Coefficient Bi 0.285 Cd-Mo 0.312 Cd-Pb 0.413 Cu-Fe 7.744 Cu-Mo -2.212 Fe-S -1.937 Fe-Sb -0.092 Hf-Si -0.947 K-Mo -0.450 K-Zn 0.219 Mo-S 1.853 Intercept 5.262 RMSE 0.075 Adj. R2 0.993  Size B (-37.5+25 mm) For size fraction B, there was not a significant improvement between the XRF and MLR datasets. This can be observed in Figure 5.13. Similar problem as in size fraction A can be seen here as well. The algorithm can predict the higher grades with great accuracy but for some of the lower grade rocks it loses its accuracy. The reason for this is sensitivity of lower grades to differences in predicted grades. For example if rocks #7 and #89 from Table 5.8 are compared, it is concluded that the fact that rock #7 has a lower molybdenum content causes it to predict higher copper grades. Both rocks have similar silica contents (14.9% for #7 and 16.44% for #89), however rock #7 only had 0.0016% molybdenum while rock #89 has 0.0698%. With interaction effects of Mo-Si having a negative effect on predicted copper grade, the product of these two numbers for each rock becomes important. With higher Mo and Si content, the product of these two numbers results in a larger negative number which in turn reduces the 132  predicted grade to closer to the actual value. However, for rock #7, this negative number is small and fails to bring the predicted value closer to the actual grade of 0.046%. Overall, size fraction B had fewer significant factors and also lower adjusted R2 value. These values are summarized in Table 5.9. The significant factors for Brenda size fraction B cannot be interpreted as clearly as the case with size fraction A, detailed mineralogical data is required before making any interpretations. Figure 5.13 XRF Copper Recovery, Brenda B – Single vs. Multivariate Analysis     0.0%0.2%0.4%0.6%0.8%1.0%1.2%1.4%1.6%1.8%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)133  Table 5.8 Predicted Cu grades vs. true grades – Brenda B 18 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade 31 1.814% 1.865% 19 1.024% 1.067% 91 1.015% 1.020% 43 0.518% 0.499% 70 0.450% 0.293% 40 0.298% 0.310% 89 0.285% 0.269% 61 0.268% 0.107% 59 0.245% 0.075% 7 0.226% 0.046% 66 0.223% 0.076% 51 0.215% 0.417% 3 0.183% 0.183% 35 0.177% 0.206% 75 0.172% 0.205% 81 0.156% 0.002% 53 0.155% 0.048% 45 0.153% 0.024% Table 5.9 MLR analysis coefficients and statistical data – Brenda B Elements Coefficients Cd-Mo 0.217 Cu-Si 1.785 Cu-Zr -1.168 Hf-Mg -0.311 Mn-Si 0.088 Mo-Si -0.222 Pb-Ti -0.114 Intercept 0.252 RMSE 0.101 Adj. R2 0.843   134   Size C (-50+37.5 mm) Size fraction C also behaved similarly to size fraction B. Therefore not much explanation is given here as it would be rehashing what was discussed earlier. Figure 5.14 XRF Copper Recovery, Brenda C – Single vs. Multivariate Analysis  Table 5.10 Predicted Cu grades vs. true grades – Brenda C 10 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade 49 2.398% 2.405% 62 0.575% 0.586% 52 0.237% 0.265% 32 0.208% 0.177% 55 0.141% 0.022% 45 0.126% 0.226% 31 0.121% 0.212% 36 0.120% 0.031% 48 0.111% 0.055% 40 0.097% 0.075%   0.0%0.5%1.0%1.5%2.0%2.5%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)135  Table 5.11 MLR analysis coefficients and statistical data – Brenda C Elements Coefficients Ti 0.027 Ca-Cu -0.310 Cu-S 0.227 Cu-Si 0.364 Cu-Zn 0.835 Fe-Hf -0.071 K-Mg -0.038 Mg-S 0.337 Intercept 1.372 RMSE 0.041 Adj. R2 0.972  Size D (-75+50 mm) Both XRF and MLR recovery curves predict the copper grades well. The advantage and accuracy of the MLR curve can been seen below 15% mass pull, beyond which both curves almost overlap. As it can be seen in Table 5.12, although there are some false positives among the top 15 sorted rocks, the grades for most of the rocks are predicted fairly accurately. Figure 5.15 XRF Copper Recovery, Brenda D – Single vs. Multivariate Analysis    0%1%2%3%4%5%6%7%8%9%10%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)136  Table 5.12 Predicted Cu grades vs. true grades – Brenda D 15 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade 1 5.312% 5.318% 7 3.069% 3.124% 28 2.484% 2.484% 3 0.746% 0.801% 66 0.634% 0.632% 42 0.464% 0.461% 12 0.231% 0.112% 46 0.223% 0.133% 56 0.221% 0.084% 75 0.219% 0.091% 72 0.215% 0.122% 48 0.195% 0.211% 59 0.182% 0.103% 26 0.171% 0.169% 76 0.169% 0.227% Table 5.13 summarizes the 19 significant factor in prediction of copper grades. This number of factors along with the high adjusted R2 also suggested a good prediction. As it can be seen from the table, the most significant factor is copper itself followed by interaction effects of Cu-Zr and hafnium both with negative effects.   137  Table 5.13 MLR analysis coefficients and statistical data – Brenda D Elements Coefficients Cu 12.775 Hf -6.687 Al-Cu -0.432 Al-Si 0.049 Ca-Cu -0.807 Ca-K 0.354 Cd-Pb -0.177 Cu-K -0.740 Cu-Sb 3.823 Cu-Si -3.513 Cu-Sn -0.697 Cu-Ti 1.431 Cu-Zn -0.987 Cu-Zr -9.210 K-Mn -0.246 K-V -0.125 Mn-Mo 0.823 Pb-V 0.294 Pb-Zn 0.072 Intercept -4.024 RMSE 0.059 Adj. R2 0.993  Conclusion The graphs showed that optical sorting seemed inefficient for the material from Brenda. This was attributed to the similar looking host rock (granodiorite) and the mineralized vein structure that could have fallen under the blind spot of the camera. Based on the results presented in this chapter, three out of four size fraction showed some correlation between the EM response and the copper grades. This correlation was best, based on the adjusted R2 value, for the largest size fraction (-75 mm +50 mm) and the smallest size fraction had the weakest correlation. However in terms of recovery efficiency, the smallest size fraction showed the best recovery (91% Cu recovery in 60% mass) and size fraction B had the lowest recovery of 80% copper in 60% mass. Overall, XRT predicted the grades more efficiently than the Optical or Electromagnetic sensors. The major issue that was observed, as expected, was presence of iron tarnishing sorting results 138  for base metals. Despite this shortcoming, XRT sensor proved to be a potential sensor in detecting base metals, especially in cases where there is a possibility of blind spots such as Brenda. The XRF analysis, whether single or multivariate proved to replicate the ideal recovery curves better than any other sensors, regardless of their degree of heterogeneity. For Brenda, it was observed that the single-variate vs. multivariate algorithms performed fairly similarly. For the case of the XRF sensor, it was observed that both responses (XRF and MLR) performed similarly. Although there were small differences between the curves at times, the difference was not significant enough to overrule the use of a single element proxy (i.e. copper for copper). The reason behind this is that copper, itself, as a single-variate, is a good indicator of the true copper grade in the samples. Another important point to mention here is the presence of heavy elements such as hafnium, antimony, tin and tungsten among the significant elements identified by the MLR algorithm. While there may be mineralogical reasons behind the presence of heavy minerals in the equation, their occurrence be simply because they are detected better by the XRF sensor due to their high atomic numbers and they are almost always present for every sample and therefore show up as important factor. MLR analysis was performed on the XRF readings for all elements. Some of these elements are present at very low concentrations (in the order of ppm) and may not be detected in an actual sorting scenario. This work assumed that all these elements will be detected and therefore proceeded with including them in the MLR analysis. It is important to note that the CH values for the spectral response cannot be analysed using the same scale or logic as the grade heterogeneity. In grade heterogeneity, the dispersion of values is greater than that of the spectral index. Spectral indices are natural numbers while grades are real numbers. Hence, the spread for the grades is wider therefore it leads to potentially higher CH values.   139  6. COPPER MOUNTAIN SORTABILITY RESULTS  Optical Sortability Results Based on results presented in Figure 6.1 through Figure 6.4, the recoveries predicted from optical sorting followed the ideal recovery fairly well therefore suggesting the potential for optical sorting. Of all the size fractions, the largest size fraction showed the weakest potential. This could be attributed to the fact that for larger size fractions not all of the rocks were photographed (camera view limitations) and therefore the photo might have not been a good representation of the whole visual features of the rocks. Table 6.1 summarizes the optical sorting results for Copper Mountain. Figure 6.1 Optical recovery of copper – Copper Mountain Size A (-25+19 mm)    0%1%1%2%2%3%3%4%4%5%5%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu GradeCH=6.41CH (OPT)=0.25Correlation: 0.440140  Figure 6.2 Optical recovery of copper – Copper Mountain Size B (-37.5+25 mm)   Figure 6.3 Optical recovery of copper – Copper Mountain Size C (-50+37.5 mm)    0.0%0.5%1.0%1.5%2.0%2.5%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu Grade0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu GradeCH=6.62 CH (OPT)=0.31 Correlation: 0.470 CH=3.28 CH (OPT)=0.16 Correlation: 0.466 141  Figure 6.4 Optical recovery of copper – Copper Mountain Size D (-75+50 mm)  Table 6.1 Optical sorting summary – Copper Mountain Size Fractions CH OPT CH Recovery at Mass Pulls 20% 40% 60% 80% A (-25+19 mm) 6.41 0.25 60% 76% 86% 97% B (-37.5+25 mm) 6.62 0.31 59% 77% 84% 93% C (-50+37.5 mm) 3.28 0.16 48% 72% 90% 96% D (-75+50 mm) 2.61 0.18 25% 55% 77% 99%  Electromagnetic Sortability Results Of all the four size fractions, MLR found that only the largest fraction showed correlation between the EM response and the copper grades. Even in that case, only one variate was identified as significant with a very low adjusted R2 value of 0.062. Therefore, it can be concluded that the EM sensor will not work for the Copper Mountain samples of this project. Figure 6.5 shows the two curves associated with the ideal and predicted copper recoveries, while Table 6.2 summarizes the statistical model’s coefficients.   0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu GradeCH=2.61 CH (OPT)=0.18 Correlation: 0.314 142  Figure 6.5 EM recovery of copper – Copper Mountain-D (-75+50 mm)  Table 6.2 MLR statistical data for EM – Copper Mountain-D Copper Mountain-D Variate Coefficient M100-M400 0.159 Intercept 0.086 RMSE 0.204 Adj. R2 0.062 The Reitveld tests performed on the low and high copper grade particles indicated that both hematite and magnetite were present in both rocks. Although the Reitveld tests were limited and performed on only two particles, the presence of hematite and magnetite in both rocks could be the reason why the EM sensor failed to perform satisfactorily.   X-Ray Transmission Sortability Results The ideal recovery curves show that for all size fractions, by around 60% mass pull, most of the copper (over 97%) could be potentially recovered. However, as expected, the case for the sensor recovery is never as clean. For these samples, the efficiency of sorting seemed to diminish with increasing size (Table 6.3). The reason behind lower recoveries for the case of size fraction D (Figure 6.9) was the effect of size on x-ray attenuation. This was also seen earlier in a case with Copper Mountain size Fraction B. Larger rocks absorb the x-ray more and therefore appear darker under the x-ray. Now if these large rocks contain heavy elements 0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (EM)Cu GradeCu Grade (EM)143  such as iron, this darkness will intensify. As a result of this, a large number of rocks were misidentified as high-grade while they were either large or large and richer in iron. In general, Copper Mountain rocks were relatively low-grade and homogeneous in iron content. Therefore, it is expected that size difference between the rocks was the major contributor to the inefficiency was the size of the rocks. Figure 6.6 to Figure 6.9 show the corresponding grade-recovery curves for these size fractions. Table 6.3 Summary of the XRT sorting results – Copper Mountain Fractions Responses CH Correlation Cu Recovery at Set Mass Pulls 20% 40% 60% 80% Size A Ideal 6.41 - 72% 95% 99% 100% Spectral 0.41 0.778 66% 81% 88% 94% Size B Ideal 6.62 - 81% 94% 98% 99% Spectral 0.47 0.557 56% 74% 84% 94% Size C Ideal 3.28 - 72% 94% 98% 99% Spectral 0.27 0.726 56% 68% 86% 94% Size D Ideal 2.61 - 70% 90% 97% 99% Spectral 0.11 0.482 46% 70% 82% 94% Figure 6.6 XRT recovery of copper – Copper Mountain Size A (-25+19 mm)    0%1%1%2%2%3%3%4%4%5%5%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade144  Figure 6.7 XRT recovery of copper – Copper Mountain Size B (-37.5+25 mm)  Figure 6.8 XRT recovery of copper – Copper Mountain Size C (-50+37.5 mm)    0.0%0.5%1.0%1.5%2.0%2.5%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade145  Figure 6.9 XRT recovery of copper – Copper Mountain Size D (-75+50 mm)   X-Ray Fluorescence Sortability Results Two different XRF results are analysed here, one, is the single-variate analysis in which the copper Rock XRF readings were used as the sensor’s response. Then rocks were ordered based on this response in a descending manner, and the grade-recovery curves were generated based on the corresponding true grade and weight of each rock. This set of results is annotated by “(XRF)” on the graphs and tables. For the second set of the results, the full Rock XRF readings of all elements and their interaction effects were used in a Multivariate Linear Regression (MLR) Analysis algorithm against the true copper grade of the rocks. The algorithm, predicted the copper grade using multiple factors. This predicted copper grade then was used as the sensor’s response, and the grade-recovery curves were developed similar to the first case. This set of results is annotated by “(MLR)” on the graphs and tables. These two datasets are then compared against the ideal copper recovery. As it was mentioned in the Experimental Methods Chapter, the advantage of using MLR analysis is that since it has more degrees of freedom, it can possibly predict the true grades based on multiple factors better.  Size A (-25+19 mm) For this size fraction, the MLR performed better at lower mass pulls but as the mass pull increased both curves performed similarly with MLR slightly yielding better recoveries (Figure 6.10). 0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade146   Table 6.4 shows that the MLR predicted grades for the top 15 rocks in the series quite well. There were some false positive that caused the curve to eventually deviate from the ideal curve. The reason behind these false positives was the inefficiency of the MLR algorithm in predicting lower-grade copper particles as precisely as the higher grade ones. Examining the data points out to a few high-grade particles that are very well predicted by the algorithm and hence the overlapping of the two curves at low mass pulls. However, for the lower grade particles, the algorithm did not predict the grades as well and therefore started to deviate from the ideal recovery curve. The reason behind a weak prediction is believed to be presence of the few high grade particles. It is understood that the algorithm prioritizes matching the higher grade particles rather than the lower grades. Through this, the predicted values for some of the lower grades deviate more from the actual grades. Among the 20 significant factors identified, Al-Zn, Pb-S (potentially presence of galena), and Fe-S (pyrite) were the major positive contributors while Mo-Zn and Si-Zn were the major negative ones. Although copper, itself, was not identified as a contributor, its interaction effects with manganese (positive) and tin (negative) were. Without the detailed mineralogical data available, it is difficult to interpret all of the interactions, and anything at this point would be mere speculations.   147  Figure 6.10 XRF Copper Recovery, Copper Mountain A – Single vs. Multivariate Analysis  Table 6.4 Predicted Cu grades vs. true grades – Copper Mountain A 15 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade 67 2.666% 2.674% 36 2.425% 2.432% 17 1.920% 1.923% 57 1.774% 1.764% 5 0.888% 0.947% 24 0.759% 0.739% 75 0.390% 0.376% 86 0.374% 0.397% 21 0.363% 0.190% 52 0.358% 0.348% 73 0.311% 0.293% 96 0.279% 0.361% 71 0.255% 0.251% 64 0.212% 0.237% 4 0.206% 0.129%    0.0%0.5%1.0%1.5%2.0%2.5%3.0%3.5%4.0%4.5%5.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)148  Table 6.5 MLR analysis coefficients and statistical data – Copper Mountain A Elements Coefficients Zr -0.049 Al-Zn 1.997 As-Zn -0.741 Ca-V -0.046 Cd-Mo 0.261 Cu-Mn 0.720 Cu-Sn -0.543 Fe-Mn -0.129 Fe-S 1.829 Hf-Sn -0.242 K-Mn -0.052 K-Zn -0.208 Mn-Zn 0.363 Mo-P 0.659 Mo-Zn -2.799 P-S -0.173 P-Zn -0.242 Pb-S 1.982 S-Zr -0.962 Si-Zn -1.485 Intercept 0.299 RMSE 0.078 Adj. R2 0.970  Size B (-37.5+25 mm) Size fraction performed similarly as size fraction A and therefore, only the results are presented here in Figure 6.11, Table 6.6 and Table 6.7. Of the major MLR factors, Cu-Si and S-Sb (possibly stibnite) are positive contributors while Al-S, Cu-Sb, S-Zr and Fe-Hf are negative contributors.   149  Figure 6.11 XRF Copper Recovery, Copper Mountain B – Single vs. Multivariate Analysis  Table 6.6 Predicted Cu grades vs. true grades – Copper Mountain B 15 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade 20 2.119% 2.111% 31 1.849% 1.850% 60 1.479% 1.500% 21 1.379% 1.383% 11 1.358% 1.353% 30 0.622% 0.748% 79 0.412% 0.368% 28 0.361% 0.411% 100 0.295% 0.297% 41 0.284% 0.069% 92 0.272% 0.214% 48 0.257% 0.155% 86 0.250% 0.289% 53 0.220% 0.105% 61 0.212% 0.221%    0%1%1%2%2%3%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)150  Table 6.7 MLR analysis coefficients and statistical data – Copper Mountain B Elements Coefficients Zr -0.050 Al-S -3.587 Cd-Cu -0.805 Cu-Pb -0.950 Cu-Sb -2.133 Cu-Si 3.761 Cu-V 0.989 Fe-Hf -1.458 Mo-Zn -0.298 Pb-S 0.820 S-Sb 5.167 S-Zr -2.328 Intercept -0.792 RMSE 0.068 Adj. R2 0.965  Size C (-50+37.5 mm) For size fraction C, the MLR response performed consistently better than the XRF but at  a 35% mass pull, the two responses almost overlapped (Figure 6.12). For this size fraction, since the XRF and MLR responses for below 35% mass pull were quite different, both sets of predicted grades by XRF and MLR are presented in Table 6.8 for comparison purposes. Examining the rock IDs and their corresponding grades explains why the two recovery curves are different at low mass pulls. The XRF response fails to correlate to higher copper grade particles as well as the MLR does and therefore, it lags behind in recovery.   151  Figure 6.12 XRF Copper Recovery, Copper Mountain C – Single vs. Multivariate Analysis  Noteworthy significant elements identified through the MLR analysis included Cu-Mo with a positive effect and Fe-Mo, Hf-Mo and Pb-S with negative effects. Table 6.8 Predicted Cu grades vs. true grades – Copper Mountain C 15 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade  Sample ID XRF Predicted Cu Grade True Copper Grade 15 1.427% 1.431%  40 4.714% 0.305% 48 0.898% 0.893%  70 4.216% 0.141% 58 0.687% 0.687%  44 2.950% 0.542% 37 0.673% 0.679%  61 1.817% 0.189% 44 0.542% 0.542%  2 1.617% 0.152% 23 0.523% 0.565%  58 1.585% 0.687% 45 0.483% 0.466%  9 1.543% 0.078% 94 0.463% 0.489%  15 1.415% 1.431% 31 0.444% 0.444%  77 0.944% 0.241% 41 0.304% 0.358%  37 0.736% 0.679% 40 0.287% 0.305%  4 0.701% 0.225% 42 0.252% 0.329%  41 0.695% 0.358% 73 0.249% 0.157%  48 0.505% 0.893% 77 0.242% 0.241%  17 0.484% 0.048% 68 0.219% 0.164%  56 0.469% 0.136%    0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)152  Table 6.9 MLR analysis coefficients and statistical data – Copper Mountain C Elements Coefficients Al-Mn -0.083 Co-Mo 0.113 Co-V 0.127 Cu-Mo 5.699 Cu-Pb 0.707 Fe-Mo -3.302 Fe-Si 0.309 Fe-Ti -0.245 Fe-Zn 0.118 Hf-Mg 0.272 Hf-Mo -4.067 Hf-Zn -0.057 K-Mg -0.067 K-Mo -0.209 K-S 0.125 Mn-Mo 1.171 Mo-Sb 0.417 Mo-Zn -0.567 P-S 0.062 P-Sn -0.049 Pb-S -1.015 Sb-Zr -0.159 Ti-Zr 0.110 Intercept -0.431 RMSE 0.054 Adj. R2 0.938  Size D (-75+50 mm) The performance of the two curves for size fraction D were similar to that of size fraction C. while the MLR performed better at low mass pulls, by around 40% mass pull both curves almost overlapped (Figure 6.13).   153  Figure 6.13 XRF Copper Recovery, Copper Mountain D – Single vs. Multivariate Analysis  For comparison reasons, the predicted copper grades by MLR and XRF are presented in Table 6.10. Similar to the case with size fraction C, a number of false positives and failure to predict higher copper grade particles led to inefficiency of the XRF dataset to yield high recoveries in lower mass pulls.   0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)154  Table 6.10 Predicted Cu grades vs. true grades – Copper Mountain D 15 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade  Sample ID XRF Predicted Cu Grade True Copper Grade 28 1.297% 1.297%  10 3.736% 0.764% 43 1.133% 1.126%  43 2.074% 1.126% 10 0.759% 0.764%  5 1.365% 0.288% 30 0.663% 0.634%  12 1.364% 0.082% 50 0.484% 0.525%  2 0.762% 0.237% 51 0.455% 0.474%  28 0.719% 1.297% 42 0.437% 0.437%  46 0.625% 0.334% 46 0.334% 0.334%  67 0.597% 0.188% 5 0.298% 0.288%  51 0.547% 0.474% 39 0.297% 0.283%  30 0.443% 0.634% 52 0.261% 0.261%  75 0.427% 0.257% 75 0.252% 0.257%  37 0.405% 0.043% 67 0.221% 0.188%  50 0.379% 0.525% 79 0.218% 0.208%  39 0.315% 0.283% 25 0.212% 0.227%  73 0.261% 0.117% Interactions of copper with other elements were some of the major contributors to predict the true copper grades. Among them Cu-S, Cu-Mo and Cu-Cd had negative and Cu-Ca, Cu-Si and Cu-Sn had positive effects. Interaction of molybdenum and zinc was also another significant positive contributor.   155  Table 6.11 MLR analysis coefficients and statistical data – Copper Mountain D Elements Coefficients Ni -0.482 As-Mo -0.869 Ca-Cu 0.768 Cd-Cu -0.968 Co-Mo -0.736 Co-Zn 0.127 Co-Zr -0.083 Cu-S -2.136 Cu-Si 0.948 Cu-Sn 2.837 Cu-Ti -0.249 Fe-S 0.157 Fe-Zr -0.026 Hf-Sn -0.923 Hf-V 0.598 Mg-S -0.079 Mo-S -0.486 Mo-Ti 0.535 Mo-Zn 1.631 P-Pb -0.049 Intercept 0.550 RMSE 0.045 Adj. R2 0.953  Conclusion Copper Mountain, showed fairly good potential for optical sorting as there was a distinction in appearance between the high-grade and low-grade rocks. Results showed that about 85% to 90% of the copper could be recovered in about 60% of the mass. Based on the results presented in this chapter, the EM sensor failed to predict the desired grades based on multiple signals for Magnitude and Phase. Of all the four size fractions, only one showed some correlation with the copper grades and even in that case the correlation was very weak. This inefficiency could be attributed to the fact that none of the minerals in the matrix were strongly magnetic or the interactions of various minerals with each other cancel out overall electromagnetic properties of the rocks. 156  XRT performed reasonably well and recoveries from 82% to 88% was achieved in 60% mass pull. When targeting base metals, the major issue that was observed, was the presence of iron tarnishing sorting results for base metals. Another interfering factor that was observed for some samples was when a great difference in size caused a rock to appear relatively darker or lighter under the x-ray and therefore causing it to be misidentified. Despite these two major shortcomings, the XRT sensor proved to be a potential sensor in detecting base metals. The XRF analysis, whether single or multivariate proved to replicate the ideal recovery curves better than any other sensors, regardless of their degree of heterogeneity. For Copper Mountain, it was observed that the single-variate versus multivariate algorithms performed similarly. Although there were small differences between the curves at times, the difference was not significant enough to overrule the use of a single element proxy (i.e. copper for copper). The reason behind this is that copper, itself, as a single-variate, is a good indication of the true copper grade in the samples. Overall the XRF with multivariate regression analysis yielded the best outcome. Another important point to mention here is presence of heavy elements such as hafnium, antimony, tin and tungsten among the significant elements identified by the MLR algorithm. While there may be mineralogical reasons behind their presence, it might as well be simply because of they are detected better by the XRF sensor due to their high atomic numbers and they are almost always present for every sample and therefore show up as important factor. The last issue to point out here is that the MLR analysis was performed on the XRF readings for any element that was not zero. Some of these elements have very low concentration (in the order of ppm) and may not be at all detected in an actual sorting scenario. This work assumed that all these elements will be detected and therefore proceeded with including all of them in the MLR analysis. It is important to note that the CH values for the spectral response cannot be analysed using the same scale/logic as the grade heterogeneity. In grade heterogeneity, the dispersion of values is greater than that of the spectral index. Spectral indices are natural numbers while grades are real numbers. Hence, the spread for the grades is wider therefore it leads to potentially higher CH values.   157  7. MOUNT POLLEY SORTABILITY RESULTS  Optical Sortability Results Mount Polley rocks have similar visual features in terms of color and texture and therefore optical sorter was found to be difficulty. The similar orange/pink color in most of the rocks made optical sorting for this material ineffective. The recovery curves for all four size fractions showed very weak potential for sorting through an optical sensor. Figure 7.1 Optical recovery of copper – Mount Polley Size A (-25+19 mm)    0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu GradeCH=1.31 CH (OPT)=0.30 Correlation: 0.310 158  Figure 7.2 Optical recovery of copper – Mount Polley Size B (-37.5+25 mm)   Figure 7.3 Optical recovery of copper – Mount Polley Size C (-50+37.5 mm)    0.0%0.5%1.0%1.5%2.0%2.5%3.0%3.5%4.0%4.5%5.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu Grade0.0%0.2%0.4%0.6%0.8%1.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu GradeCH=0.82 CH (OPT)=0.19 Correlation: 0.283 CH=0.94 CH (OPT)=0.12 Correlation: 0.411 159  Figure 7.4 Optical recovery of copper – Mount Polley Size D (-75+50 mm)  Table 7.1 summarize the optical sorting results for Mount Polley. As expected, and already seen form the corresponding Figure 7.1 through Figure 7.4, only weak recoveries are achieved through this sensor for all size fractions. These weak recoveries are mainly as a result of the homogeneous nature of the rock, both in terms of grade and appearance, rather than inefficiency of the sensor.   0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (OPT)Cu Rec.Cu Grade (OPT)Cu GradeCH=1.22 CH (OPT)=0.15 Correlation: 0.387 160  Table 7.1 Optical sorting summary – Mount Polley Size Fractions CH OPT CH Recovery at Mass Pulls 20% 40% 60% 80% A (-25+19 mm) 1.31 0.30 34% 50% 70% 86% B (-37.5+25 mm) 0.82 0.19 25% 49% 68% 83% C (-50+37.5 mm) 0.94 0.12 28% 52% 70% 86% D (-75+50 mm) 1.22 0.15 32% 48% 72% 87%  Electromagnetic Sortability Results  Size A (-25+19 mm) With the lowest heterogeneity values among all 5 samples, expectations for a possible sorting were not high. As explained in the Constitution Heterogeneity chapter, this can be seen in the ideal copper recovery curves in the figures below. Only two significant variates were identified for Size A of Mount Polley resulting in a low adjusted R2 value of 0.152. The resultant predicted recovery curve also deviated significantly from the ideal curve and therefore eliminating EM as a potential sensor. Figure 7.5 EM recovery of copper – Mount Polley-A (-25+19 mm)    0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (EM)Cu GradeCu Grade (EM)161  Table 7.2 MLR statistical data for EM – Mount Polley-A Mount Polley-A Variate Coefficient P100-P300 -19.749 P200-P300 20.296 Intercept 0.715 RMSE 0.152 Adj. R2 0.152  Size B (-37.5+25 mm) No significant factors were observed for Size B of Mount Polley.  Size C (-50+37.5 mm) Although only one significant variate was observed for Size C of Mount Polley, the predicted copper recovery followed the ideal recovery slightly better than the case with size fraction A. However, the recovery was still far from ideal. The graph and the details of the MLR analysis are shown in Figure 7.6 and   162  Table 7.3, respectively. Figure 7.6 EM recovery of copper – Mount Polley-C (-50+37.5 mm)    0.0%0.1%0.2%0.3%0.4%0.5%0.6%0.7%0.8%0.9%1.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (EM)Cu GradeCu Grade (EM)163  Table 7.3 MLR statistical data for EM – Mount Polley-C Mount Polley-C Variate Coefficient M100-M1300 0.104 Intercept 0.128 RMSE 0.139 Adj. R2 0.043  Size D (-75+50 mm) Six significant variates were picked up by the MLR analysis. The predicted copper recovery was closer to the ideal recovery than the other cases but still far from ideal. Similar to other cases, a high adjusted R2 (0.63) did not necessarily correspond to a close fit with the ideal recovery curve. These results are shown in Figure 7.7 and   164  Table 7.4. Figure 7.7 EM recovery of copper – Mount Polley-D (-75+50 mm)    0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (EM)Cu GradeCu Grade (EM)165  Table 7.4 MLR statistical data for EM – Mount Polley-D Mount Polley-D Variate Coefficient P100 7.531 P400 -6.086 M100-P100 10.002 M300-P200 -7.637 P900-P1300 4.236 P900-P1400 -3.268 Intercept 0.192 RMSE 0.136 Adj. R2 0.630  X-Ray Transmission Sortability Results Due to its low degree of heterogeneity, Mount Polley yielded lower recoveries even in the ideal case. This homogeneity was observed both in grade and appearance of the rocks This homogeneity is augmented through the sensor and therefore expecting a good recovery would be futile. This is also confirmed by studying   166  Table 7.5 that summarizes the spectral indices and sorting outcome for Mount Polley’s four size fractions.  Examining the table below shows that the most efficient sorting potential belongs to the smallest size fraction, most probably because the effects of particle size is less significant. In either case, only around 70% of copper can be retrieved in 60% mass pull to concentrate. Whether this would be justified economically or not is to be investigated. Also as mentioned earlier in the Sensor’s Response Analysis section in chapter 4, distinct false positive/negative cases did not exist in this material, which again was a result uniformity in heavy metal constituents of the rock, aka its homogeneity. As it can be seen in   167  Table 7.5, the correlation coefficients are also marginally low which could indicate a weaker correlation with the actual copper grades. These results are all shown in Figure 7.8 to Figure 7.11.   168  Table 7.5 Summary of the XRT spectral sorting results – Mount Polley Fractions Responses CH Correlation Cu Recovery at Set Mass Pulls 20% 40% 60% 80% Size A Ideal 1.31 - 50% 70% 84% 95% Spectral 0.43 0.647 37% 55% 72% 91% Size B Ideal 0.82 - 42% 64% 81% 94% Spectral 0.30 0.420 29% 52% 69% 85% Size C Ideal 0.94 - 46% 67% 83% 95% Spectral 0.11 0.403 30% 47% 72% 87% Size D Ideal 1.22 - 49% 69% 85% 96% Spectral 0.38 0.575 35% 56% 70% 84% Figure 7.8 XRT Spectral recovery of copper – Mount Polley Size A (-25+19 mm)    0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade169  Figure 7.9 XRT Spectral recovery of copper – Mount Polley Size B (-37.5+25 mm)  Figure 7.10 XRT Spectral recovery of copper – Mount Polley Size C (-50+37.5 mm)    0%1%1%2%2%3%3%4%4%5%5%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade0.0%0.1%0.2%0.3%0.4%0.5%0.6%0.7%0.8%0.9%1.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade170  Figure 7.11 XRT Spectral recovery of copper – Mount Polley Size D (-75+50 mm)   X-Ray Fluorescence Sortability Results Two different XRF results are analysed here, one, is the single-variate analysis in which the copper Rock XRF readings were used as the sensor’s response. Then rocks were ordered based on this response in a descending manner, and the grade-recovery curves were generated based on the corresponding true grade and weight of each rock. This set of results is annotated by “(XRF)” on the graphs and tables. For the second set of the results, the full Rock XRF readings of all elements and their interaction effects were used in a Multivariate Linear Regression (MLR) Analysis algorithm against the true copper grade of the rocks. The algorithm, predicted the copper grade using multiple factors. This predicted copper grade then was used as the sensor’s response, and the grade-recovery curves were developed similar to the first case. This set of results is annotated by “(MLR)” on the graphs and tables. These two datasets are then compared against the ideal copper recovery. As it was mentioned in the Experimental Methods Chapter, the advantage of using MLR analysis is that since it has more degrees of freedom, it can possibly predict the true grades based on multiple factors better.  Size A (-25+19 mm) The two responses for size fraction A of Mount Polley performed very similarly and with a deviation from the ideal recovery curve (Figure 7.12).   0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec. (Spect.)Cu Rec.Cu Grade (Spect.)Cu Grade171  Figure 7.12 Mount Polley Size A – XRF Copper Recovery – Single vs. Multivariate Analysis  Table 7.6 Predicted Cu grades vs. true grades – Mount Polley A 15 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade 72 1.408% 1.410% 53 0.798% 0.859% 16 0.731% 0.723% 77 0.453% 0.425% 95 0.408% 0.579% 21 0.373% 0.269% 56 0.360% 0.449% 32 0.321% 0.275% 15 0.307% 0.325% 18 0.298% 0.480% 46 0.296% 0.226% 65 0.296% 0.115% 74 0.269% 0.314% 70 0.259% 0.200% 79 0.256% 0.609% Only five significant factor were identified by the MLR algorithm, four of which were copper interactions with Cu-S and Cu-Zr being both positive and most dominant. The small number of significant factors along with a relatively lower adjusted R2 values suggested a weaker correlation and therefore, a weaker yield (Table 7.7). 0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)172  Table 7.7 MLR analysis coefficients and statistical data – Mount Polley A Elements Coefficients Ca-Cu 0.085 Cu-S 0.336 Cu-Zr 0.369 Mn-Ti -0.041 S-Zr -0.075 Intercept 0.741 RMSE 0.080 Adj. R2 0.823  Size B (-37.5+25 mm) The first thing that can be seen in Figure 7.13 is that the fit for both XRF and MLR curves is better for size fraction B compared to size fraction A. This can also be concluded from Table 7.8. A close analysis of the 15 highest grades rock in this table indicated that they were very well predicted by the algorithm. The efficiency and accuracy of the MLR algorithm can be inferred from the number of significant factors identified, as well as the high adjusted R2 value (Table 7.9). There were some similarities between these factors and the one identified before. As an example, interaction effects of copper and sulfur with a negative effect, and copper-iron and copper-silica with positive effects. Figure 7.13 Mount Polley Size B – XRF Copper Recovery – Single vs. Multivariate Analysis  0.0%0.5%1.0%1.5%2.0%2.5%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)173  Table 7.8 Predicted Cu grades vs. true grades – Mount Polley B 15 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade 31 2.476% 2.476% 46 0.758% 0.758% 13 0.609% 0.661% 18 0.535% 0.600% 68 0.510% 0.483% 78 0.449% 0.502% 75 0.436% 0.467% 16 0.431% 0.465% 17 0.379% 0.367% 34 0.353% 0.315% 23 0.334% 0.284% 72 0.326% 0.351% 100 0.321% 0.349% 54 0.319% 0.303% 19 0.306% 0.298%    174  Table 7.9 MLR analysis coefficients and statistical data – Mount Polley B Elements Coefficients Sb -0.155 Zr -0.592 Al-S 0.877 Ca-S -0.307 Co-Mg 1.276 Cu-Fe 0.276 Cu-K 0.130 Cu-S -1.581 Cu-Si 0.229 Fe-Mg -0.075 Hf-S 0.406 Mn-S 0.220 Mn-Sn -0.061 Mo-Pb -0.108 Pb-V 0.144 S-Sn -0.655 Sb-Zr 0.772 Intercept 1.089 RMSE 0.050 Adj. R2 0.963  Size C (-50+37.5 mm) For size fraction C, MLR consistently performed better throughout all mass pulls although with a deviation from the ideal curve (Figure 7.14). This deviation was also expected through examining both Table 7.10, comparing predicted grades with true ones, as well as the low adjusted R2 value for this dataset (Table 7.11).    175  Figure 7.14 Mount Polley Size C – XRF Copper Recovery – Single vs. Multivariate Analysis  Table 7.10 Predicted Cu grades vs. true grades – Mount Polley C 15 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade 40 0.953% 0.951% 48 0.555% 0.569% 26 0.467% 0.513% 85 0.438% 0.543% 77 0.385% 0.354% 74 0.350% 0.239% 72 0.336% 0.362% 53 0.326% 0.298% 50 0.306% 0.329% 65 0.283% 0.327% 64 0.277% 0.241% 18 0.268% 0.107% 56 0.258% 0.211% 80 0.251% 0.155% 61 0.251% 0.409%    0.0%0.1%0.2%0.3%0.4%0.5%0.6%0.7%0.8%0.9%1.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)176  Table 7.11 MLR analysis coefficients and statistical data – Mount Polley C Elements Coefficients Al-Fe -0.103 Cu-Mn -0.070 Cu-S 0.376 Cu-Si 0.281 Cu-Zr -0.104 Fe-Zr 0.181 Mo-S -0.122 Mo-Sn 0.200 Intercept 0.658 RMSE 0.067 Adj. R2 0.777  Size D (-75+50 mm) Similar to the case with size fraction C, MLR curves performed consistently better than the XRF one until about 65% mass pulls (Figure 7.15). Figure 7.15 Mount Polley Size D – XRF Copper Recovery – Single vs. Multivariate Analysis  Details of the predicted grades as well as the significant factors identified by the MLR algorithm are summarized in Table 7.12 and Table 7.13, respectively.   0.0%0.4%0.8%1.2%1.6%2.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullCu Rec.Cu Rec. (MLR)Cu Rec. (XRF)Cu GradeCu Grade (MLR)Cu Grade (XRF)177  Table 7.12 Predicted Cu grades vs. true grades – Mount Polley D 15 highest grade rocks Sample ID MLR Predicted Cu Grade True Copper Grade 55 1.409% 1.377% 51 1.177% 1.198% 15 0.292% 0.171% 63 0.280% 0.348% 34 0.238% 0.135% 35 0.236% 0.392% 29 0.232% 0.593% 66 0.229% 0.146% 39 0.222% 0.119% 6 0.218% 0.252% 47 0.215% 0.360% 64 0.215% 0.210% 44 0.214% 0.294% 37 0.213% 0.084% 67 0.213% 0.201% Table 7.13 MLR analysis coefficients and statistical data – Mount Polley D Elements Coefficients Hf -0.157 Mn -0.069 Cu-Fe 0.445 Cu-Mo 0.411 Fe-Sb -0.112 Intercept 0.752 RMSE 0.096 Adj. R2 0.813  Conclusion Overall, the results for Mount Polley material showed weak potential for sorting. The weak results were not necessarily due to shortcomings of the sensors but rather homogenous nature of the samples. The sensors with the best possible outcome included XRF with both single and multivariate regression analysis, and XRT. For the case of the XRF sensor, it was observed that both responses (XRF and MLR) performed rather similarly. Although there were small differences between the curves at times, the difference was not significant enough to overrule the use of a 178  single element proxy (i.e. copper for copper). The reason behind this is that copper, itself, as a single-variate, is a good indication of the true copper grade in the samples. Another important point to mention here is the presence of heavy elements such as hafnium, antimony, tin and tungsten among the significant elements identified by the MLR algorithm. While there may be mineralogical reasons behind their presence, it might as well be simply because of they are detected better by the XRF sensor due to their high atomic numbers and they are almost always present for every sample and therefore show up as important factor. The last issue to point out here is that the MLR analysis was performed on the XRF readings for any element that was not zero. Some of these elements have very low concentration (in the order of ppm) and may not be at all detected in an actual sorting scenario. This work assumed that all these elements will be detected and therefore proceeded with including all of them in the MLR analysis. Of the four sensors, the electromagnetic showed the weakest potential which can also be attributed to weak magnetic properties of these rocks. Also, due to similar visual features of the rocks, the optical sensor failed to yield promising results. It is important to note that the CH values for the spectral response cannot be analysed using the same scale/logic as the grade heterogeneity. In grade heterogeneity, the dispersion of values is greater than that of the spectral index. Spectral indices are natural numbers while grades are real numbers. Hence, the spread for the grades is wider therefore it leads to potentially higher CH values.   179  8. MYRA FALLS SORTABILITY RESULTS  Optical Sortability Results With high heterogeneities and great potential for sorting, Myra Falls exhibited good sorting potential. The major problem with Myra Falls rocks was presence of pyrite that under the camera looked like sphalerite. These two can be compared in Figure 8.1 (sphalerite with 34% zinc) and Figure 8.2 (pyrite with 36% iron). This similarity can result in rocks reporting to concentrate while not contributing to the zinc recovery. A perfect example for this misidentification can be seen in Figure 8.5, grade-recovery curves for Size C of Myra Falls, as horizontal flat lines. It is also important to mention that the vertical jumps in the recovery curves in the graphs below indicate rocks with relatively high zinc grades that are reported at later time due to misidentification. Figure 8.1 High-grade rock #45 with 34% zinc    180  Figure 8.2 Low-grade rock #95 with 36% iron and 0.3% zinc  Figure 8.3 through Figure 8.6 show the recovery curves generated based on the spectral indices generated by the optical sensor. Figure 8.3 Optical recovery of Zinc – Myra Falls Size A (-25+19 mm)    0%10%20%30%40%50%60%70%80%90%100%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec. (OPT)Zn Rec.Zn Grade (OPT)Zn GradeCH=19 CH (OPT)=0.47 Correlation: 0.640 181  Figure 8.4 Optical recovery of Zinc – Myra Falls Size B (-37.5+25 mm)  Figure 8.5 Optical recovery of Zinc – Myra Falls Size C (-50+37.5 mm)    0%5%10%15%20%25%30%35%40%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec. (OPT)Zn Rec.Zn Grade (OPT)Zn Grade0%5%10%15%20%25%30%35%40%45%50%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec. (OPT)Zn Rec.Zn Grade (OPT)Zn GradeCH=17 CH (OPT)=0.89 Correlation: 0.745 CH=21.4 CH (OPT)=0.46 Correlation: 0.657 182  Figure 8.6 Optical recovery of Zinc – Myra Falls Size D (-75+50 mm)  Table 8.1 summarizes the optical sorting results for Myra Falls. As expected, and seen in the graphs above, zinc recoveries of over 80% can be achieved in only 20% mass pull to concentrate. Table 8.1 Optical sorting summary – Myra Falls Size Fractions CH OPT CH Recovery at Mass Pulls 20% 40% 60% 80% A (-25+19 mm) 19 0.47 86% 89% 95% 99% B (-37.5+25 mm) 17 0.89 82% 85% 93% 97% C (-50+37.5 mm) 21.4 0.46 83% 87% 88% 90% D (-75+50 mm) 20.8 0.56 84% 90% 93% 99%  Electromagnetic Sortability Results The early electromagnetic tests conducted on similar samples from Myra Falls, did not produce any significant results. As a result, it was decided not to pursue further EM tests on Myra Falls samples. Therefore, no EM data is available for Myra Falls.  X-Ray Transmission Sortability Results High heterogeneity values, high-grade rocks, as well as high ratio of gangue to mineralized particles, all were promising facts to expect impressive sorting outcome. Examining Table 8.2 further confirms this hypothesis. High recoveries (over 90%) are achieved in just 20% of the mass of the material and by 40% almost complete recoveries are obtained. 0%10%20%30%40%50%60%70%80%90%100%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec. (OPT)Zn Rec.Zn Grade (OPT)Zn GradeCH=20.8 CH (OPT)=0.56 Correlation: 0.455 183  Table 8.2 Summary of the XRT spectral sorting results – Myra Falls Fractions Responses CH Correlation Zn Recovery at Set Mass Pulls 20% 40% 60% 80% Size A Ideal 19 - 97% 98% 99% 99% Spectral 1.63 0.772 93% 96% 98% 99% Size B Ideal 17 - 98% 99% 99% 100% Spectral 2.70 0.870 92% 99% 99% 100% Size C Ideal 21.4 - 96% 98% 99% 100% Spectral 1.79 0.726 92% 95% 97% 99% Size D Ideal 20.8 - 97% 98% 99% 100% Spectral 3.11 0.708 90% 96% 98% 100% With all four size fractions performing quite similarly, only their grade-recovery curves are briefly examined. There were no significant observation in any of the grade-recovery curves except very distinct false positive/false negatives. As an example, in the case of size fraction A (Figure 8.7), the small jump in recoveries is due to a 14-gram rock with 10% zinc (rock #69) that was identified as high-grade but was placed later in the queue. The reason for this “delay” in identification was presence of high-iron-content rocks, as summarized in Table 8.3. The table summarized the top 15 rocks in size fraction A of Myra Falls that were identified as high-grade, along with the grades of major elements in them. It is interesting to see how these rocks alternate between having high-grade zinc (true positive) and high-grade iron (false positive). Figure 8.7 XRT recovery of copper – Myra Falls Size A (-25+19 mm)    0%10%20%30%40%50%60%70%80%90%100%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec. (Spect.)Zn Rec.Zn Grade (Spect.)Zn Grade184  Table 8.3 Top 15 rocks based on XRT Spectral Index – Mya Falls size fraction A Sample Combined Zn Fe Cu Pb Weight (#) Index (%) (%) (%) (%) (g) 25 57 51.94% 2.94% 0.10% 0.19% 21.84 41 55 29.99% 9.72% 3.56% 3.69% 29.07 45 55 35.07% 6.05% 1.87% 3.49% 33.8 53 54 30.16% 8.57% 1.11% 3.01% 15.41 87 38 0.02% 31.65% 0.07% 0.02% 45.57 96 36 0.37% 37.21% 0.13% 0.02% 36 86 35 1.21% 25.60% 7.59% 1.21% 50.86 2 33 0.03% 32.43% 0.08% 0.00% 41.2 80 30 0.05% 29.85% 0.08% 0.01% 29.7 84 25 1.62% 21.80% 6.88% 0.05% 36.5 20 23 0.02% 27.34% 0.38% 0.00% 18.45 69 21 10.98% 8.99% 0.35% 2.90% 14.11 22 20 0.07% 22.80% 1.51% 0.01% 22.6 7 19 0.01% 18.69% 0.24% 0.00% 30.1 8 18 0.10% 21.37% 0.11% 0.01% 14.3 10 16 0.06% 15.79% 0.07% 0.06% 19.9 61 14 2.04% 11.96% 0.71% 0.01% 18.86 73 13 0.05% 12.70% 0.87% 0.01% 12.8 9 11 0.03% 9.25% 0.49% 0.00% 28.7 47 11 1.88% 4.43% 0.23% 0.55% 22.34 Figure 8.8 XRT recovery of copper – Myra Falls Size B (-37.5+25 mm)    0%5%10%15%20%25%30%35%40%45%50%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec. (Spect.)Zn Rec.Zn Grade (Spect.)Zn Grade185  Figure 8.9 XRT recovery of copper – Myra Falls Size C (-50+37.5 mm)  Figure 8.10 XRT recovery of copper – Myra Falls Size D (-75+50 mm)   X-Ray Fluorescence Sortability Results Two different XRF results are analysed here, one, is the single-variate analysis in which the copper Rock XRF readings were used as the sensor’s response, rocks were ordered based on this response in a descending manner, and the grade-recovery curves were generated based on the corresponding true grade of each rock. This set is annotated by “(XRF)” on the graphs and tables. For the second set of the results, the full Rock XRF readings of all elements and their 0%5%10%15%20%25%30%35%40%45%50%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec. (Spect.)Zn Rec.Zn Grade (Spect.)Zn Grade0%10%20%30%40%50%60%70%80%90%100%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec. (Spect.)Zn Rec.Zn Grade (Spect.)Zn Grade186  interaction effects were used in a Multivariate Linear Regression (MLR) Analysis algorithm against the true copper grade of the rocks. The algorithm, predicted the copper grade using multiple factors. This predicted copper grade then was used as the sensor’s response, and the grade-recovery curves were developed similar to the first case. This set of results is annotated by “(MLR)” on the graphs and tables. These two datasets are then compared against the ideal copper recovery. As it was mentioned in the Experimental Methods Chapter, the advantage of using MLR analysis is that since it has more degrees of freedom, it can possibly predict the true grades based on multiple factors better. Both XRF and MLR curves perfectly matched the ideal recovery for almost all cases. Therefore, not all results are discussed in further detail here as it would a repetition of the same reasoning multiple times.  Size A (-25+19 mm) Both the XRF and MLR curve match the ideal zinc recovery curve perfectly and both curves almost overlap the ideal curve (Figure 8.11). The zinc grades in these particles are quite high, therefore even short time exposures of XRF are capable of identifying them. Also as observed in the Ore Characterization Chapter, the MLR algorithm predicts these high grades with great accuracy. The predicted zinc grades along with their corresponding true grades are summarized in Table 8.4. Figure 8.11 Myra Falls Size A – XRF Copper Recovery – Single vs. Multivariate Analysis  0%10%20%30%40%50%60%70%80%90%100%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec.Zn Rec. (MLR)Zn Rec. (XRF)Zn GradeZn Grade (MLR)Zn Grade (XRF)187  Table 8.4 Predicted Zn grades vs. true grades – Myra Falls A 15 highest grade rocks Sample ID MLR Predicted Zn Grade True Zinc Grade 25 51.938% 51.938% 45 35.071% 35.071% 53 30.161% 30.161% 41 29.993% 29.992% 69 10.983% 10.982% 61 2.029% 2.040% 47 1.908% 1.880% 72 1.907% 1.907% 84 1.611% 1.625% 86 1.208% 1.213% 58 0.566% 0.575% 94 0.428% 0.518% 100 0.369% 0.346% 76 0.335% 0.881% 77 0.288% 0.259% Examining the significant factors identified by the MLR algorithm pointed out to interesting facts. Both zinc and lead were identified as single elements with significant positive and negative effects, respectively. Presence of tungsten and its interaction effects was noteworthy, however the Reitveld tests did not indicate the presence of any tungsten mineral. Other elements that had a significant effect on the algorithm are aluminum, hafnium, tin, antimony manganese and cadmium. Similarly to the tungsten, the Reitveld tests did not show any indication of any of these elements except for aluminum. A closer look at the ICP grades of these elements and comparison to the high-grade zinc particles showed that only cadmium had a good correlation with zinc. The three rocks where cadmium grades were above 0.1%, they all had the highest zinc grades, as well. The other elements however, had very low grades and did not correlate with the high-grade zinc particles. The MLR analysis for these samples were all-inclusive, meaning all the elements as long as they were not zero were included in the algorithm. This observation leads to the conclusion that for future tests, a more selective approach with regards to inclusion of elements in the MLR analysis should be taken. To be selective however, the Limit Of Detection (LOD) of elements in actual XRF sorting cases should be determined. Since these LODs were not available, this work did not focus on various 188  processes of elimination for the elements and confines the scope of these studies to the all-inclusive approach. Table 8.5 MLR analysis coefficients and statistical data – Myra Falls A Elements Coefficients Pb -6.578 Zn 51.637 Al-W 38.535 As-K 0.686 Ca-K -0.056 Cd-Hf 220.069 Mn-W -29.876 Mn-Zn -2.574 Pb-Sn 2.126 Pb-V 26.009 Sb-W -210.866 Si-V 0.223 Si-W -41.234 Sn-W 11.756 Sn-Zn -1.209 Intercept 58.612 RMSE 0.082 Adj. R2 1.000  Size B (-37.5+25 mm) Similar case as before was observed with Myra Falls size fraction B with almost perfect recoveries.   189  Figure 8.12 Myra Falls Size B – XRF Copper Recovery – Single vs. Multivariate Analysis  Table 8.6 Predicted Zn grades vs. true grades – Myra Falls B 15 highest grade rocks Sample ID MLR Predicted Zn Grade True Zinc Grade 41 35.435% 35.399% 44 34.090% 34.076% 34 30.533% 30.581% 39 30.165% 30.177% 81 12.288% 12.288% 85 8.733% 8.744% 19 3.935% 3.977% 95 3.846% 3.851% 73 2.367% 2.386% 82 2.365% 2.302% 48 2.289% 2.327% 45 2.237% 2.140% 67 1.511% 1.477% 63 0.974% 1.240% 74 0.509% 0.520% Similar elements are seen here as significant factors. And overall the reasoning that was given for size fraction A will hold true for size fraction B as well. Interesting points to mention however, are that while lead (Pb) shows very good correlation with zinc, copper does not necessarily follow the same trend as zinc. Also, the presence of phosphorous and its 0%5%10%15%20%25%30%35%40%45%50%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec.Zn Rec. (MLR)Zn Rec. (XRF)Zn GradeZn Grade (MLR)Zn Grade (XRF)190  interactions with major elements is worth mentioning. However visually examining the ICP assays, phosphorous did not seem to have any correlation with zinc. As for the interactions with other elements, more detailed and perhaps a new algorithm could determine whether these interactions are founded or not. Table 8.7 MLR analysis coefficients and statistical data – Myra Falls B Elements Coefficients Hf 0.160 Zn 16.432 Al-K -0.123 As-Cd -1.610 As-P -0.460 Cd-P 1.007 Co-P 0.567 Co-Zr -0.091 Cu-P -13.990 Fe-Mo -0.199 K-Mn -0.209 K-W -10.034 K-Zn 8.310 P-Pb 30.883 P-Sn -1.491 P-V -0.261 P-W -18.863 P-Zn 7.470 Si-W 2.184 Sn-W 1.533 Sn-Zn -2.259 Intercept 19.028 RMSE 0.083 Adj. R2 1.000  Size C (-50+37.5 mm) Size fraction C performed similarly to the previous Myra Falls cases.   191  Figure 8.13 Myra Falls Size C – XRF Copper Recovery – Single vs. Multivariate Analysis  Table 8.8 Predicted Zn grades vs. true grades – Myra Falls C 15 highest grade rocks Sample ID MLR Predicted Zn Grade True Zinc Grade 45 35.062% 35.054% 55 27.500% 27.501% 49 25.821% 25.827% 91 6.194% 6.203% 99 2.622% 2.627% 79 1.786% 1.781% 84 1.556% 1.545% 6 1.417% 1.408% 89 1.403% 1.879% 21 0.853% 0.829% 85 0.623% 0.609% 97 0.458% 0.528% 75 0.451% 0.423% 52 0.409% 0.023% 82 0.407% 0.510% Among the significant factors identified by the MLR algorithm, presence of interaction of Fe-Zn (pyrite-sphalerite association) was important, so was the negative effects of interaction of Mn-Pb, while manganese grades are in the order of a few hundred ppm. 0%5%10%15%20%25%30%35%40%45%50%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec.Zn Rec. (MLR)Zn Rec. (XRF)Zn GradeZn Grade (MLR)Zn Grade (XRF)192  Table 8.9 MLR analysis coefficients and statistical data – Myra Falls C Elements Coefficients As-V -0.329 As-Zr 0.808 Cd-Fe -0.481 Cd-Pb -5.304 Co-Zn -0.610 Cr-K -0.165 Cu-K 0.657 Fe-Zn 7.961 Mn-Pb -4.253 Mo-Zn -0.310 Pb-Sn 0.544 Pb-Zr 19.059 Ti-Zr -0.580 Intercept 17.085 RMSE 0.111 Adj. R2 1.000  Size D (-75+50 mm) Similar trends and recoveries were observed for size fraction D, as well. Figure 8.14 Myra Falls Size D – XRF Copper Recovery – Single vs. Multivariate Analysis    0%12%24%36%48%60%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%GradeRecoveryMass PullZn Rec.Zn Rec. (MLR)Zn Rec. (XRF)Zn GradeZn Grade (MLR)Zn Grade (XRF)193  Table 8.10 Predicted Zn grades vs. true grades – Myra Falls D 15 highest grade rocks Sample ID MLR Predicted Zn Grade True Zinc Grade 23 48.739% 48.737% 63 28.875% 28.879% 59 28.756% 28.749% 11 11.290% 11.295% 73 8.829% 8.847% 52 8.805% 8.809% 43 2.864% 2.838% 45 2.411% 2.459% 26 2.029% 2.086% 79 1.982% 1.997% 51 1.978% 1.970% 20 1.578% 1.860% 19 1.339% 1.300% 40 1.053% 1.053% 14 0.983% 1.297%   194  Table 8.11 MLR analysis coefficients and statistical data – Myra Falls D Elements Coefficients Si 0.083 Zn -6.422 Al-Cu 2.748 Al-Mg 1.954 Al-Mo -0.471 Al-Pb 2.480 Al-Zn 3.825 As-K -0.396 Ca-Zn 29.226 Cd-Mo 1.560 Cd-Sb -0.191 Co-V -0.263 Cr-K 5.468 Cu-K -2.680 Cu-Mg 0.425 Fe-K 0.160 K-Mg -6.010 Mg-Ni -1.174 Mg-Ti -0.973 Mg-Zn -1.592 Mg-Zr 3.598 Mn-Pb 5.207 Mn-Zn -4.711 Pb-Si 1.162 S-Sb 0.459 Sn-W -6.718 Sn-Zn -1.853 V-W 18.235 Intercept 43.016 RMSE 0.125 Adj. R2 1.000  Conclusion Myra Falls material showed great potential for sorting. The best results were obtained using the XRF sensor where both single and MLR curves almost identically matched the ideal recovery curve. It was observed that for zinc both responses (XRF and MLR) performed rather similarly. Although there was small difference between the curves at times, this difference was not significant enough to overrule the use of a single element proxy (i.e. zinc for zinc). The 195  reason behind this is that zinc, itself, as a single-variate, is a good indication of the true zinc grade in the samples. Another important point to mention here is presence of heavy elements such as hafnium, antimony, tin and tungsten among the significant elements identified by the MLR algorithm. While there may be mineralogical reasons behind their presence, it might as well be simply because of they are detected better by the XRF sensor due to their high atomic numbers and they are almost always present for every sample and therefore show up as important factor. The last issue to point out here is that the MLR analysis was performed on the XRF readings for any element that was not zero. Some of these elements have very low concentration (in the order of ppm) and may not be at all detected in an actual sorting scenario. This work assumed that all these elements will be detected and therefore proceeded with including all of them in the MLR analysis. In case of the XRT sensor, although a great potential for sorting was observed, this potential was mostly due to the highly heterogeneous nature of the rocks rather than a good correlation of XRT spectral indices with rock grades. Presence of high-grade pyrite pieces interfered with the optical sensor, misplacing them in the accept category and at times pushing high-grade zinc particles further down the queue. While the results from the optical sensor was quite satisfactory as well, these results were less efficient than the XRF and XRT sensors. However, the electromagnetic tests were not performed on these samples when the initial tests proved futile. It is important to note that the CH values for the spectral response cannot be analysed using the same scale/logic as the grade heterogeneity. In grade heterogeneity, the dispersion of values is greater than that of the spectral index. Spectral indices are natural numbers while grades are real numbers. Hence, the spread for the grades is wider therefore it leads to potentially higher CH values.   196  9. GOLD SORTABILITY RESULTS The gold samples were prepared in only one size fraction, -50+37.5 mm. Also, the gold samples were only anlaysed using the Electromagnetic, XRT and XRF sensors and optical tests were not performed on these samples.  Electromagnetic Sortability Results The EM results for gold were unsatisfactory. Only one variate was identified as significant and the algorithm failed to predict the gold grades closely, which led to an inefficient recovery as can be seen in Figure 9.1. These results eliminate this sensor as an option to sort this specific gold-bearing material. Figure 9.1 Grade-recovery curves based on the EM sensor for the gold sample  Table 9.1 EM sensor’s MLR coefficients and statistical data for the gold sample Gold-Bearing Material Variate Coefficient P400 0.104 Intercept 9.797 RMSE 3.940 Adj. R2 0.157 051015202530354045500%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Grade [gpt]RecoveryMass PullAu Rec.Au Rec. (EM)Au GradeAu Grade (EM)197   X-Ray Transmission Sortability As explained in the Experimental Methods chapter, the XRT tests performed on the gold samples differed in procedure and therefore the individual XRT indices are not available. Hence, a detailed analysis of why each rock reported, or did not report, to the concentrate streams cannot be performed. Studying Figure 9.2, it can be seen that the best sorting outcome would be on iron with a steep increase in recoveries and a constant drop in grades. In the case of gold, although it looks promising, the economics of it should be studied. There will definitely be an upgrade but it is important to note that at 64% mass pull, 86% of gold was recovered and increasing the mass pull to 77% did not improve recovery significantly. This can be attributed to the rocks wrongly reporting the various sorted streams. Examining the rocks reported to the two concentrate streams (Cleaner Product and Cleaner Waste) indicated that 50 rocks that reported to the two concentrate streams were false positives (33 to Cleaner Product and 17 to Cleaner Waste), while only one rock was found to be false negative (in Scavenger Waste). This greatly tarnished the sorting outcome. There is a chance that having better separation criteria could potentially resolve this problem and yield better sorting outcomes. Figure 9.2 Grade-recovery outcome of dynamic XRT sorting of gold-bearing material  0123456789100%10%20%30%40%50%60%70%80%90%100%40% 60% 80% 100%Grade [gpt/%]RecoveryMass PullAu RecoveryS RecoveryFe RecoveryAs RecoveryAu GradeS GradeFe GradeAs Grade198  Focusing on the only instance of false negative, this can be attributed to low iron grade (3.71%) and rather high silica content (17.49%). Average silica content among the false positives was 13.63% in Cleaner Product (CP) and 14.57% in Cleaner Waste (CW). This was expected because lower silica content could mean denser particles, and vice versa.  In the Cleaner Product (CP) stream 33 out of 41 rocks were misidentified as false positive. This meant that only 8 particles had a higher grade than 3 gpt. Examining each individual rock for their iron grade quickly pointed to the culprit. Most of the particles in this stream had high iron grades except for seven rocks with iron grades lower than 5% and the average grade of iron for the whole stream was 7.58%. This explains why most of the rocks, whether containing gold or not, reported to this stream. For some of the rocks however, a grade analysis was inconclusive. For example, rock #37 with only 0.02 gpt gold and 1.17% iron should not have been identified as high-grade. Investigating the grades of other heavy elements was inconclusive as they were in the order of ppm and were not believed to have had any impacts. The other possible reason could have been an ejection error, which caused this rock to end up in the concentrate stream. In Cleaner Waste (CW), there were only two rocks out of nineteen that contained gold (rock #56 with 7.47 and rock #88 with 25.3 gpt gold). The average iron grade was 5.43%, lower than that of Cleaner Product stream, which does make sense. Both these two rocks with high-grade gold, also contained rather high grade iron (7.6% both), as well. Therefore, it was expected for them to be identified as high-grade. Similar to the case of the Cleaner Product stream, while some of these misplacements could be attributed to low or high iron grades, some might have been due to sorter ejection inefficiencies and without individual rock’s sensor response data, it is impossible to interpret this behavior in all rocks.  X-Ray Fluorescence Sortability Results The XRF results for gold were processed in slightly different manner. Since gold cannot be detected by the XRF sensor directly, other elements such as arsenic, iron, sulfur and silica were used for the single-variate regression analysis. These tests were performed at 10 second exposure times and therefore are annotated by “10”. On the other hand, the MLR analysis used the non-zero XRF readings of all elements as the “X” variables and the Fire Assay gold grades as “Y”. The analysis, as explained in the Experimental Methods Chapter was done by 199  MATLAB and using the default settings. For the gold samples, in addition to the single and multivariate analysis, the impacts of interaction effects in the MLR analysis was also examined. For the single variate XRF analysis, first the correlation between the rock XRF readings are compared to the calculated ICP grades of each proxy element. Then the grade-recovery curves based on these proxy elements are presented and discussed. This set of data is annotated by the corresponding element’s IUPAC name. For these sets of data, the Rock XRF readings of each proxy element was used as the sensor’s response, rocks were ordered based on this response in a descending manner, and the grade-recovery curves were generated based on the corresponding true grade of each rock. For the second set of the results, the non-zero Rock XRF readings of all elements without their interaction effects were used in the MLR analysis as explained earlier. This set of data is annotated by “MLR”. The last dataset, where the non-zero XRF reading and their interaction effects were used in the MLR analysis is annotated by “MLR Int.”. As it was mentioned in the Experimental Methods Chapter, the advantage of using MLR analysis is that since it has more degrees of freedom, it can possibly predict the true grades based on multiple factors better.  Single-Variate Regression Analysis  Arsenic Analyses To check the accuracy of Rock XRF readings, arsenic grades were plotted against the ICP assays and, as can be seen in Figure 9.3, it resulted in a close correlation. This close correlation can be identified through the overlapping of the grade-recovery curves in Figure 9.3. In this set of samples, only 4 out of 97 rocks had a grade over 1% arsenic and had to be extrapolated from the available data.   200  Figure 9.3 Comparative arsenic grade-recovery curves  Plotting gold grade-recovery curves based on the arsenic XRF readings did not generate a curve as it was expected. The reason behind this, and the two jumps at higher mass pulls (Figure 9.4), was that there were two rather high gold grade rocks (4.68 and 7.47 gpt) that were not positively correlated with arsenic. Therefore they report to the concentrate stream at higher mass pulls. Because of their high grade, they contribute to recovery significantly and that was why the recovery for Au-As lagged behind the ideal case until above 60-80% mass pulls.   0.0%0.5%1.0%1.5%2.0%2.5%3.0%3.5%4.0%4.5%5.0%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%As GradeRecoveryMass PullAs-ICP RecoveryAs-10 RecoveryAs-ICP GradeAs-10 Grade201  Figure 9.4 Gold grade-recovery curves based on arsenic XRF readings   Iron Analyses The XRF reading predicted the iron grades consistently though with a deviation from the actual ICP assays. A quick look at the grade-recovery curves suggested that iron had low heterogeneity values and could not be used as a potential proxy for gold. In this set of rocks, two rocks had iron grades greater than 15%. To estimate the grades of these two rock, the average of two XRF readings was used. The two grades were estimated at 29.5% and 18.95%.   051015202530354045500%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Au Grade [gpt]RecoveryMass PullAu-FA RecoveryAu-As RecoveryAu-FA GradeAu-As Grade202  Figure 9.5 Comparative iron grade-recovery curves  Figure 9.6 reveals how poorly iron was associated with gold, therefore eliminating it as a potential proxy element.  Figure 9.6 Gold grade-recovery curves based on iron XRF readings   Sulfur Analyses The XRF readings showed good correlations with the ICP assays for sulfur despite the low atomic number of sulfur. The close correlation can be seen in Figure 9.7, below.   0%5%10%15%20%25%30%35%40%45%50%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Fe GradeRecoveryMass PullFe-ICP RecoveryFe-10 RecoveryFe-ICP GradeFe-10 Grade051015202530354045500%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Au Grade[gpt]RecoveryMass PullAu-FA RecoveryAu-Fe RecoveryAu-FA GradeAu-Fe Grade203  Figure 9.7 Comparative sulfur grade-recovery curves  Although with some deviation at the low mass pulls, sulfur seemed to have a good correlation with gold. This is demonstrated by almost overlapping grade-recovery curves at above 35% mass pull. Based on the results shown in Figure 9.8, sulfur could potentially be used as a good proxy element for gold. Figure 9.8 Gold grade-recovery curves based on sulfur XRF readings  0%2%4%6%8%10%12%14%16%18%20%0%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%S GradeRecoveryMass PullS-ICP RecoveryS-10 RecoveryS-ICP GradeS-10 Grade051015202530354045500%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Au Grade [gpt]RecoveryMass PullAu-FA RecoveryAu-S RecoveryAu-FA GradeAu-S Grade204   Silica Analyses There were no ICP results for silica available therefore no comparative study could be performed. On the other hand, silica had a poor correlation with gold as it can be seen in Figure 9.9. Therefore, silica is eliminated from the list of elements to be used as a proxy for gold. Figure 9.9 Gold grade-recovery curves based on silica XRF readings   Multivariate Linear Regression Analysis Figure 9.10 compares the following 4 regression models with gold liberation function (idealistic separation scenario). 1. XRF readings at 10 seconds exposure time – annotated as “Au-XRF” 2. Same set of elements as above but with their actual ICP assays – annotated as “Au-ICP” 3. Same as #1 but with two-element interaction effects – annotated as “Au-XRF Int.” 4. Same as #2 but with two-element interaction effects – annotated as “Au-ICP Int.”  XRF and XRF Int. The summary of important factors as well as the intercept and adjusted R2 for “XRF” and “XRF Int.” can be seen in Table 9.2. Each model predicted different elements as significant contributors to the fit. But what was obvious was how including the interaction effects improved the adjusted R2 from 0.59 to 0.99. 051015202530354045500%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Au Grade [gpt]RecoveryMass PullAu-FA RecoveryAu-Si RecoveryAu-FA GradeAu-Si Grade205  Table 9.2 Regression model coefficients for XRF datasets with and without interaction effects XRF  XRF Elements w/ interaction w/o interaction  Elements w/ interaction w/o interaction Al - -3.657  Mg-Co 0.400 - Al-Mg 1.643 -  Mg-Ni -0.396 - As - 2.741  Mg-Sb -0.989 - As-Co 3.615 -  Mn-Cd 0.532 - Ca-Co -0.434 -  Mn-Ni -10.866 - Ca-Cr 1.164 -  Mn-Ti 1.648 - Ca-Sn 0.501 -  Ni-As 25.377 - Ca-Ti -1.907 -  Ni-Co 0.418 - Ca-V -1.052 -  S - 3.396 Cr-As -6.790 -  S-Al -3.301 - Cr-Mg -1.009 -  S-Ca -1.599 - Cr-Mn -0.799 -  S-Cd -8.518 - Cu - 2.741  S-Cr 12.420 - Cu-As 12.354 -  S-Mg -2.850 - Cu-Mg 0.568 -  S-Sb 4.126 - Cu-Mn -1.641 -  S-V 4.394 - Fe-As -21.109 -  Sb - -3.996 Fe-Ni 2.151 -  Si-Fe -1.689 - K-As 3.034 -  V-As 3.095 - K-V -4.324 -  V-Zn 1.389 - Mg - -1.079  V-Zr 0.534 - Mg-As 15.792 -  Zn-Co -6.249 - Intercept 20.188 12.8172  Adj. R2 0.9961 0.5934 Turning the focus on “XRF”, six elements, aluminum, arsenic, copper, magnesium, sulfur and antimony were determined as significant. Out of these, the importance of elements such as sulfur, arsenic and iron is easily understood and can be attributed to the presence of arsenopyrite which has a positive effect on gold. Aluminum and magnesium also can be attributed to a number of minerals such as biotite, clinochlore, cordierite and others that are present in this material. However, there was no indication of presence of any mineral that would contain copper or antimony. Based on the ICP assays, both copper and antimony are present in these samples however in small quantities (Cu<383 ppm and Sb<93 ppm), it indicates that these minerals were below the detection limit for the Reitveld tests and therefore were not pick up by the test. 206  For the XRF with interaction effect (“XRF Int.”), there were 39 interaction effects that were significant and no individual element was among them. While some combinations of elements, such as Al-Mg and Fe-As, are worthy of attention (arsenopyrite and other aluminum-magnesium minerals as mentioned above), some others might not make sense at the first look such as copper interactions with other elements considering there was no copper-containing mineral detected in the Reitveld tests. Further tests are necessary to investigate the presence of copper minerals in these sample before a definitive conclusion can be made. Same applies to other interaction effects such as V-Zr. Therefore a complete mineralogical data is necessary to interpret these element interactions. As stated above, the model with interaction effects improved the fit significantly. The fitted model overlaps gold liberation curve in Figure 9.10.  ICP and ICP Interactions. As mentioned earlier, two data sets that are also examined here are ICP results of the same elements that were detected by XRF, and their interaction effects, in the MLR analysis. The reason behind this study was to compare the resultant MLR model of the true grades (“ICP” dataset) versus the XRF grades. The study showed that while the number of significant factors for the “ICP” dataset was smaller, there were a number of common factors between the two models. The factors for “ICP” datasets with and without interaction effects are summarized in Table 9.3 .  207  Table 9.3 Regression model coefficients for ICP datasets with and without interaction effects ICP  ICP Element w/ interaction w/o interaction  Element w/ interaction w/o interaction Al -2.921 -  Mn-As 25.709 - As -31.534 3.333  Mn-Sn -1.418 - As-Co -14.345 -  Mn-Zr -17.665 - Ca-Ti 2.773 -  Ni-Co 3.978 - Cr -2.330 -  S - 4.140 Cr-Mn 2.978 -  S-Cd -5.029 - Cr-V 2.047 -  S-Co -3.063 - Cu-As 39.675 -  S-Fe 4.769 - Cu-Sn 1.080 -  S-K -5.263 - Cu-Ti -3.656 -  S-Zn 7.649 - Mg - -4.985  S-Zr 4.754 - Mg-As -14.379 -  Zn-As -15.908 - Mg-Mn 1.580 -  Zr 15.277 - Mn - 6.016  Zr-As 21.916 - Intercept 19.872 9.329  Adj. R2 0.975 0.486 There were 4 significant elements identified for the “ICP” dataset compared to 6 for the “XRF” dataset. Of the 4 elements, arsenic, magnesium and sulfur are in common and manganese is different. With arsenic, sulfur and magnesium having already been explained, more mineralogical data is required to interpret the presence of manganese and its correlation with gold. For the same set of results but with interaction effects (i.e. “ICP Int.” dataset), there were 25 elements identified as significant, among which some single elements were also identified. The single elements identified were aluminum, arsenic, chromium and zirconium. Similar to the case with “XRF Int.”, presence of aluminum and arsenic are justified, however, the presence of chromium and zirconium needs more mineralogical data to interpret. Comparing the four grade-recovery curves in Figure 9.10, it is evident that the curves that considered interaction effects yielded better grade-recovery curves and more similar to the gold liberation curve. For this set of data, The XRF with interaction effects (i.e. “XRF Int.” dataset) yielded the best results.   208  Figure 9.10 Multivariate regression Gold grade-recovery curves with and without interaction effects   Conclusion From the tests performed it was observed that the electromagnetic sensor did not offer any solution to concentrate the ore. The weak magnetic properties of the rocks was the reason behind this inefficiency. Although magnetite was present in the gangue, its low concentration as well as the presence of other non-magnetic minerals might have masked its detection by the sensor. Although the XRT sensor seemed to have performed well, the weak association of iron with gold was the inhibiting factor in better gold recoveries. There were a large number of false positive in the mix that diluted the concentrated product. A combination of XRT and XRF could possibly result in better sorting outcomes. In the case of the XRF sensor, generally finding a single element as a proxy is rather difficult. Although at times, such as the case studied here, a single element such sulfur or arsenic can be found, chances of false negatives still exist (e.g. in case of arsenic). Or sulfur detection at extreme short exposure times of dynamic sorting might be an issue. Although it seemed possible to achieve satisfactory gold recoveries using these single-variate regression models, the use of MLR analysis proved to improve gold prediction significantly and therefore highly suggesting the use of this technique in gold detection.  051015202530354045500%10%20%30%40%50%60%70%80%90%100%0% 20% 40% 60% 80% 100%Grade [gpt]RecoveryMass PullAu-FA Rec.Au-XRF Rec.Au-ICP Rec.Au-XRF Int. Rec.Au-ICP Int. Rec.Au-FA GradeAu-XRF GradeAu-ICP GradeAu-XRF Int. GradeAu-ICP Int. Grade209  Another important point to mention here is presence of heavy elements such as hafnium, antimony, tin and tungsten among the significant elements identified by the MLR algorithm. While there may be mineralogical reasons behind their presence, it might as well be simply because of they are detected better by the XRF sensor due to their high atomic numbers and they are almost always present for every sample and therefore show up as important factor. The last issue to point out here is that the MLR analysis was performed on the XRF readings for any element that was not zero. Some of these elements have very low concentration (in the order of ppm) and may not be at all detected in an actual sorting scenario. This work assumed that all these elements will be detected and therefore proceeded with including all of them in the MLR analysis.   210  10. CONCLUSION AND RECOMMENDATIONS This research investigated the amenability of low-grade and waste rock stockpiles to sensor-based ore sorting. Materials from a total of five stockpiles (or deposits) were investigated using four sensors, Optical (OPT), Electromagnetic (EM), X-Ray Transmission (XRT) and X-Ray Fluorescence (XRF).  The materials were analysed for their Constitution Heterogeneity values and these values were then correlated to their corresponding ideal recovery scenarios. These investigations indicated that while a high heterogeneity number demonstrates good potential for sorting, low heterogeneity values should not disappoint. Heterogeneity is a function of grade, weight and number of particles in each sample set and in certain cases the combination of these three variates could lead to low heterogeneity numbers while there is still a good potential for sorting present. Therefore, a low heterogeneity number solely indicates that more in depth tests are needed before one can draw any conclusion on sorting.  The results from the optical sorter were as expected. The sorter would work best for materials that are visually distinctive. There were two types of issues that were faced using this sensor. In case of the samples from Brenda mine, due to mineralized vein structure of the deposit, chances of blind spots existed. Although in dynamic sorting, the use of multiple sensors drastically reduces this possibility, in these bench-scale tests only one side of the rock was photographed and therefore, as observed earlier, there were instances of false negative throughout this sample. Another issue that arose with the optical sensor was inconsistency with the lighting. Some rocks, depending on how they faced the camera, were either in shadowed areas or reflected the light directly back into the camera. These cases caused misidentification of such rocks and therefore they could report to the wrong stream. While the optical sensor worked fairly well for Copper Mountain and Myra Falls, they were still not free of misrepresented rocks. In case of Mount Polley, due to visual similarity of the rocks, the optical sensor proved ineffective. The electromagnetic sensor’s performance was overall disappointing. Although magnetite and hematite were present in some samples, their grade was very low. Therefore it is safe to say that of all the minerals available in these rocks, only chalcopyrite has magnetic properties but it is believed that its magnetic properties is influenced by presence of para- or diamagnetic 211  minerals. The complex matrix of these minerals within the rocks made it impossible to further investigate the cause behind the subpar performance of the EM sensor. The XRT sensor was the one with great promise. This sensor is known to be the best sensor when it comes to sorting base metals. While this sensor generated impressive results specifically in case of Myra Falls, it was not free from error. The presence of iron minerals as rather high density materials decreased the efficiency of the XRT. The other issue that was identified with the XRT sensor was the uncommon incidents of a large rock being identified as high-grade due to its higher x-ray attenuation. Despite these two challenges, the XRT sensor performed really well disregarding the original heterogeneity of material. This sensor still performed acceptably for the gold samples, although the weak correlation of iron and gold caused a large number of false positives to report to the concentrate streams and therefore diluting the sorting outcome. The last sensor that was investigated was X-Ray Fluorescence. With elemental recognition and ability to measure the grades of element, this sensor on paper seems to be the perfect candidate. Based on the static, bench-scale tests that were performed, the XRF sensor performed outstandingly well for all five types of material. The sensor’s response was analysed through two different algorithms, single and multivariate linear regression (MLR). While the single versus multivariate linear regression analysis looked to perform rather similarly for base metals, there was a significant improvement in grade prediction by the MLR for gold. It is recommended that while for base metals using an MLR analysis might not be necessary, it can make the difference in sorting for gold. It is believed that this comprehensive thesis would benefit the industry by providing a guidance on various commodities and ore sorting. One important factor to consider though is that all these tests (except XRT for gold) were performed statically on a bench-scale sensor setup. This is the primary step in performing the amenability tests and by no means should it be used for economic predictions. For one, the representativeness of the collected sample is always under question for small scale tests such as these, and second, performance of sensors in dynamic conditions can vary significantly based on the desired targeted element and type of the ore being investigated. This difference can be significant when it comes to XRF detection of lighter elements such as sulfur, calcium, etc. 212  A number of recommendations can be suggested in terms of the experimental methods. Although the bench-top scale amenability tests are losing their attraction among the sensor-based ore sorting companies, these companies should not forget the benefits of such testwork despite their cumbersome nature. To better simulate the actual dynamic sorting, semi-dynamic conditions is suggested. The rocks can be mounted on a railed platform and moved under a stationary sensor. This can be applied to all sensors, and most importantly can be beneficial for sensors such as XRF and EM where the speed and movement of the rocks can affect sensor’s detection. In the case of the optical sorter, a better lighting system and possibly use of two cameras are suggested to eliminate the chances of blind spots. Also when capturing images of the larger rocks, it should be tried to capture as much of the rock as possible. With the EM sensor, the most important point to be investigated is the effect of size on the electromagnetic response. Based on the laws of electromagnetism, there should not be an effects, however, it may have to do with the relative size of the EM generating coil and the rock. In which case, smaller coils need to be used for smaller size fractions. With these recommendations in mind, it is hoped that this report has been useful to the reader and opens up new perspective in terms of current and future work in this field.   213  REFERENCES Bamber, A. S. (2008). Integrated Mining, pre-concentration and waste disposal systems for increased sustainability of hard rock metal mining. (4). Comex Sorting Technology. (2017). Retrieved from Comex Innovative Industrial Technologies: http://www.comex-group.com/en/products-and-solutions/sorting-technology CuDECO Limited. (2015). Quarterly Report ending 31th December 2014. CuDECO Limited. Demographer, L. (n.d.). XRF Spectroscopy. Retrieved April 29, 2017, from Superior X-Ray Tube: https://en.wikipedia.org/w/index.php?title=File:XRFScan.jpg Gy, P. M. (1982). Sampling of particulate materials: theory and practice. Amsterdam, Netherlands: Elsevier Scientific Publishers. Gy, P. M. (1992). Sampling of heterogeneous and dynamic material systems: Theories of heterogeneity, sampling and homogenizing. In P. M. Gy, Logical analysis of the concept of homogeneity and heterogeneity (pp. 48-55). Amsterdam, Netherlands: Elsevier Scientific Publishers. Klein, B., & Bamber, A. (n.d.). Comparison of Pre-concentration Technologies for Mine-Mill Integration. Knapp, H., Neubert, K., Schropp, C., & Wotruba, H. (2014). Viable Applications of Sensor-Based Sorting for the Processing of Mineral Resources. ChemBioEng Reviews, 86-95. Kolacz, J. (2014). Sensor-Based Sorting with Signal Pattern Recognition: The New Powerful Tool in Mineral Processing. IMPC, (pp. 106-115). Santiago, Chile. Lasley, S. (2014, October 26). Mining News: Coeur rewards mine success at Kensington. Retrieved from Petroleumnews: http://www.petroleumnews.com/pntruncate/884547412.shtml Mazhary, A., & Klein, B. (2015). Heterogeneity of low-grade ores and amenability to sensorbased sorting. Annual Meeting of the Canadian Mineral Processors. Ottawa, Canada. 214  Minnitt, R. C., Rice, P. M., & Spangenberg, C. (2007). Part 1: Understanding the components of the fundamental sampling error: a key to good sampling practice. The Journal of The Southern African Institute of Mining and Metallurgy, 505-511. Mosser, A., & Robben, C. (2014). X-ray transmission based sorting at the Mittersill tungsten mine. Retrieved from CEEC the future: http://www.ceecthefuture.org/comminution-2/x-ray-transmission-based-sorting-mittersill-tungsten-mine/ Napier-Munn, T., & Wills, B. A. (2016). Ore Sorting. In B. A. Wills, & J. A. Finch, Mineral Processing Technology–An Introduction to the Practical Aspects of Ore Treatment and Mineral Recovery (pp. 373-377). Oxford: Butterworth-Heinemann. Neingo, P., & Cawood, F. (2011). Correlation of productivity trends with market factors at three selected platinum mines. Statistics, 4. Pearce, C. I., Pattrick, R. A., & Vaughan, D. J. (2006). Electrical and Magnetic Properties of Sulfides. Reviews in Mineralogy & Geochemistry, 127-180. Petersen, L., Minkkinen, P., & Esbensen, K. H. (2005). Representative sampling for reliable data analysis: Theory of Sampling. Journal of Chemometrics and Intelligent Laboratory Systems, 261-277. Pitard, F. (1993). A Logical Introduction to the Notion of Heterogeneity. In F. F. Pitard, Pierre Gy's Sampling Theory and Sampling Practice: Heterogeneity, Sampling Correctness, and Statistical Process Control (pp. 57-62). CRC Press. Portable Analytical Solutions. (2017, February 15). Retrieved from http://portableas.com/index.php/technique/x-ray-fluorescence/ Robben, C., Wotruba, H., Robben, M., von Ketelhodt, L., & Kowalczyk, M. (2013). Potential of sensor-based sorting for the gold mining industry. CIM Journal, 4(3). Rule, C., Fouchee, R., & Swart, W. (2015). RUN OF MINE ORE UPGRADING – PROOF OF CONCEPT PLANT FOR XRF ORE SORTING. SAG Conference.  Salter, J. D., & Wyatt, N. P. (1991). Sorting in the Mineral Industry: Past, Present and Future. Minerals Engineering, 4(7-11), 779-796. 215  Von Ketelhodt, L. (2009). Viability of optical sorting of gold waste rock dumps. World Gold Conference.  Von Ketelhodt, L., Falcon, L., & Falcon, R. (2011). Optical sorting of Witwatersrand gold ores: An Update – Waste rock dump sorting at Goldfields; Run-of-mine sorting at Central Rand Gold. ALTA 2011 Gold Conference (pp. 18-33). Perth: ALTA Metallurgical Services. Weatherwax, T. W. (2007). Integrated Mining and Preconcentration Systems for Nickel Sulphide Ores. Vancouver, Canada: University of British Columbia. Wills, B. A. (2016). Sampling, Control, and Mass Balancing. In B. A. Wills, & J. A. Finsh, Mineral Processing Technology (pp. 41-90). Oxford: Butterworth-Heinemann. Wotruba, H. (2006). Sensor sorting technology - is the minerals industry missing a chance? XXIII International Mineral Processing Congress, (pp. 21-29). Istanbul, Turkey.   

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0355396/manifest

Comment

Related Items