Application of near-infrared (NIR) spectroscopy to sensor based sorting of an epithermal Au-Ag ore OCM conference presentation 19-03-2015 TU Delft, Resource Engineering M.Dalm MSc. 1
Contents of the presentation Introduction Test work approach NIR spectral analysis Classification on spectral data Results Conclusions Final remarks 2
Introduction Sensor based sorting Sensor based sorting represents all applications in which singular particles are mechanically separated on certain physical or chemical properties after determining these properties with a sensor 3
Introduction Sensor based sorting Sensor based sorting can be implemented as a pre-concentration step into ore processing operations to remove sub-economic ore material prior to conventional treatment Potential benefits: Reduce the costs of ore processing Decrease cut-off grade Increase mineable resources Decrease water and energy consumption Source: www.tomrasorting.com/mining/ 4
Introduction Sensor based sorting Sensor based sorting requires real-time sensors that detect physical and/or chemical parameters that indicate the economic value of individual rock particles No real-time sensors are currently known that can directly detect the Au or Ag contents of epithermal ores Optical X-ray transmission Electromagnetic 5
Introduction Sensor based sorting Why test the near-infrared (NIR) sensor? The motivation is based on the geology of epithermal ore deposits Figure from Hedenquist et al. (1998) 6
Introduction Sensor based sorting Why test the near-infrared (NIR) sensor? The deposition of Au and Ag in epithermal deposits is related to the formation of certain alteration minerals Many alteration minerals can be detected with NIR sensors It is possible that detection of alteration mineralogy can be used to predict the economic value of individual ore particles 7
Introduction Near-InfraRed(NIR) spectroscopy The NIR sensor records reflected electromagnetic radiation (light) at wavelengths ranging from 350 to 2500 nm Certain minerals produce absorptions of NIR radiation at specific wavelength locations 8
Introduction Aim & Objectives The aim of the test work was to investigate if the NIR sensor can detect alteration minerals that can be used as indicators of ore value Test work objectives: Discriminate ore particles with low Au and Ag grades Discriminate ore particles with high carbon contents Discriminate ore particles with high sulphide contents 9
Test work approach Dataset 94 ore particles were collected from a South American mine that exploits a high sulphidation epithermal Au-Ag deposit 10
Test work approach Dataset Grade distributions of elements of interest & sorting requirements Au cut-off = 0.20 ppm C cut-off = 0.5 % S cut-off = 0.25 % 11
Test work approach Dataset On all samples: NIR reflection spectra were measured with an ASD Fieldspec3 On a subset of 36 samples: Fire assay for determination of Au and Ag grades LECO analyser for determination of C and S contents (combustion infrared detection technique) X-Ray Diffraction (XRD) for determination of mineralogy 12
NIR spectral analysis Subdividing the spectral range The measured NIR spectra were subdivided into two regions: Visible region (VIS) Wavelength range: 350 1300 nm Dominated by charge transfer absorptions of the Fe ion (Fe 2+ & Fe 3+ ) Determination of Fe-bearing mineralogy Short Wave InfraRed region (SWIR) Wavelength range: 1300 2500 nm Dominated by absorptions of molecule bond vibrations Determination of alteration mineralogy 13
NIR spectral analysis Determination of mineralogy Minerals were determined from the measured NIR spectra by comparing the spectra to spectral libraries (G-MEX & USGS) Not shown: spectra on carbonaceous samples with average reflections < 11 % XRD validated the occurrence of all minerals except dickite (clay mineral) 14
Classification on spectral data Partial Least Squares Discriminant Analysis (PLS-DA) Partial Least Squares Discriminant Analysis (PLS-DA) was applied to investigate the potential of using the NIR data to distinguish between pre-defined groups of samples PLS-DA is based on Partial Least Squares (PLS) regression PLS regression is a method that finds a linear relationship between predictor variables X and a response variable Y: ˆ = β0 + β1 1 + + βp p y x x 15
Classification on spectral data Partial Least Squares Discriminant Analysis (PLS-DA) Difference between PLS regression and PLS-DA: PLS regression predicts one or more response variables PLS-DA predicts a class analogy to two or more classes PLS-DA is performed by calibrating a PLS regression model in which the response variables Y are replaced by a dummy matrix with assigned class memberships Sample Class nr 1 1 2 1 3 2 4 1 5 2 6 2 Dummy Matrix 1 0 1 0 0 1 1 0 0 1 0 1 16
Classification on spectral data Partial Least Squares Discriminant Analysis (PLS-DA) Advantages of PLS-DA over most other classification models: PLS-DA models the common structure between the predictor and class data Scores and loadings are calculated that allow visualisation of data patterns that are relevant for the classification Loadings contain information on the wavelength regions that are important for the classification Scores describe the relationship between the loadings and the measured spectra 17
Classification on spectral data Partial Least Squares Discriminant Analysis (PLS-DA) PLS-DA was performed on the following definition of classes: Au grade C content S content Class 1 < 0.20 ppm > 0.50 % > 0.25 % Class 2 > 0.20 ppm < 0.50 % < 0.25 % 25% of the samples were excluded during model calibration to allow validation of the resulting classification 18
Classification on spectral data Classification on Au grade: PLS-DA on SWIR spectra Calculated responses of the PLS-DA model on SWIR spectra for classification on the Au grade Threshold for classification 19
Classification on spectral data Classification on Au grade: PLS-DA on SWIR spectra Scores and loadings on the 1 st and 2 nd latent variable 20
Classification on spectral data Classification on Au grade: PLS-DA on SWIR spectra Scores and loadings on the 1 st and 2 nd latent variable 21
Results Classification results Results of PLS-DA classification on Au (cut-off grade: 0.20 ppm) Averages before classification Averages Class 1 (45%) Averages Class 2 (55%) Au (ppm) 0.261 Au (ppm) 0.086 Au (ppm) 0.405 Ag (ppm) 1.594 Ag (ppm) 0.000 Ag (ppm) 2.906 Results of PLS-DA classification on C (cut-off grade: 0.50 %) Averages before classification Averages Class 1 (14 %) Averages Class 2 (86 %) C (%) 0.722 C (%) 5.186 C (%) 0.002 Results of PLS-DA classification on S (cut-off grade: 0.25 %) Averages before classification Averages Class 1 (29 %) Averages Class 2 (71 %) S (%) 0.758 S (%) 2.147 S (%) 0.190 22
Conclusions NIR spectroscopy can be used to distinguish ore samples with: Low Au and Ag grades by using the occurrence of specific alteration minerals It works!!! High C contents by using the overall reflection of the spectra High S contents by using the occurrence of specific Fe-bearing minerals PLS-DA is an effective technique to investigate classification possibilities based on NIR spectroscopic data 23
Final remarks Representativity of the sample set is unknown Samples used in the research are probably not representative for the entire ore deposit Additional research is currently ongoing on a larger and more representative sample set Testwork with other sensor types is currently ongoing to investigate if discrimination on sub-economic ore material can be improved Special thanks to Barrick Gold for providing funding and access to samples 24
Thank you for your attention! 25