Fractal Modelling And Algorithm Optimization For Bulk Ore Sorting Process

Bulk ore sorting is considered of significant advantages in improving the profitability of the mining industry.However,its application has been impeded due to the difficulty to assess the sortability of an ore with current approaches and the inefficiency of algorithms in the sorting practices.In this thesis,a bulk ore sorting model was developed where a number of factors were included and related to the sortability assessment,including ore properties,sorting cut-off grade,mass balance and metal price.Metal distribution within the ore deposits were recognized as a fractal structure.Grade Factor and Fractal Dimension parameters were introduced as indicators for grade magnitude and heterogeneity,respectively.A sorter mass balance model was derived,and sorter mass yield,sorting concentrate grade,metal recovery were predicted based on characterized ore properties.The optimum sorting cut-off grade was determined as function of sorting costs,metal price,refinery cost and flotation recovery,equation as SCOG=(cc-cr)/y·(s-r).The developed model was validated with case studies on Cadia Ridgeway belt cut ore,New Afton draw points and New Afton drill cores,respectively.The validation results showed that the developed model was reliable in predicting the bulk ore sorting performance.The developed bulk ore sorting model has provided a powerful tool for ore sortability assessment.The feasibility of BOS implementation for two scenarios,namely ore preconcentration and ore waste recycle,were investigated by use of the developed bulk ore sorting model.The influence of ore properties,characterized by the Grade Magnitude and Fractal Dimension,on the BOS feasibility were studied.Results showed that the greatest positive change in economics after BOS implementation in ore preconcentration was for the ores where metal was distributed in higher heterogeneity and lower magnitude(indicated by lower G and higher D).Such ores were therefore considered to be more sortable.For BOS implemented for ore waste recycle,ores with a higher G and higher D were considered as more sortable.With the use of this model,the bulk ore sortability would be easier to assess in terms of ore intrinsic properties and systematic economics.More academic studies and industrial applications were expected to be encouraged by use of the developed model.In practice,the actual performance of a sorting system would also depend on the accuracy of the sensor responses and the sorting algorithm predictions.In this sense,a lab-scale experiment was conducted where ores are tested using XRF.Errors associated with the XRF measurements were categorized and analyzed.The inefficiency of the conventional linear regression algorithms were illustrated.Heterogeneity error was identified as the most significant influence on the sorting efficiency due to its large magnitude and irregularity.Receiver Operating Characteristics was applied as the tool in classification of bulk materials.The performance of this approach was compared to conventional sorting algorithm,namely,simple linear regression(SLR)and multiple linear regression(MLR),where XRF responses were linearly correlated to the assay grades.It was observed that economics was improved using ROC compared to the linear regression models.At Q=5,10 and 20,NSR was increased by 3.1,0.9 and 1.1 $/t,respectively,compared to that by SLR;compared to MLR,NSR was increased by 3.0,0.7,0.9 $/t.At Q=20,NSR up to 7.6$ was achieved by the ROC approach comparing to 8.3$/t in the ideal case.The ROC approach was found to result in more accurate classification and enhanced economics performance of the bulk ore sorting system.