Open Collections

Xu, Yang

University of British Columbia

Sensor-based sorting (SBS) is a mineral concentration process where rocks or bulk materials are separated based on their differential sensor responses. SBS is essential in modern mineral processing flowsheets, aimed at reducing energy and water usage by rejecting waste early. Benefits include reducing gangue transport to the mill, improving mill feed grade, and enhancing economic efficiency. Sorting efficiency depends on sensor selection and the accuracy of algorithms in converting sensor responses to ore grades. X-ray fluorescence (XRF) is a cost-effective method for determining chemical compositions, widely used in various industries. However, in ore sorting, XRF faces challenges as it only measures surface composition, which may not represent the volumetric metal content. This research aims to utilize the range of detectable elements by XRF to predict target element concentrations. Machine learning approaches were developed to improve on conventional statistical models for ore classification. Logistic regression (LR) and support vector machine (SVM) algorithms were applied, with dimensionality reduction techniques like principal component analysis (PCA) and autoencoder neural networks (AEN) used for data preprocessing. Model performance was evaluated using receiver operating characteristic analysis and data visualization, with the best model determined based on mass grade recovery sorting results. The primary contribution is a methodology for ore classification using machine learning based on XRF data. This adaptable approach provides a systematic framework for developing ore sorting models applicable to various mining operations. Key findings include the high accuracy of LR models with PCA inputs and the robustness of AEN for feature extraction, demonstrating the effectiveness of integrating machine learning with XRF in mineral processing, potentially improving efficiency and reducing costs in the mining industry.

Text

2024-07-06

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/88584

Mining Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/