Assessing the performance of XRF surface analysis for classifying copper ore into high-grade and low-grade categories

Abstract

Although promising, the use of x-ray fluorescence (XRF) for particle-by-particle copper ore sorting has been insufficiently explored. This study assesses the feasibility of XRF-based dry sorting for Moroccan copper ore hosted within sedimentary rocks, utilizing 464 samples from various lithologies with different copper contents. XRF is widely used for surface analysis across industries, but its limited penetration depth presents challenges in mining applications, where surface measurements may not reflect volumetric grades. To address this, a machine-learning model was developed to enhance XRF-based sorting. A detailed mineralogical study initially characterized the rock types and copper-bearing minerals. Portable XRF sensors then performed surface measurements in a laboratory, generating data to train machine-learning models for predicting volumetric copper content and classifying samples into high-grade and low-grade categories. Initial XRF-based classification achieved accuracies between 93% and 75% for copper cut-off grades of 0.2% to 0.8%. Using second-degree polynomial regression, the correlation between predicted and actual copper content improved to 77%. Logistic regression further enhanced classification accuracy to 92%-86% across the same thresholds. For a 0.2% copper threshold, the high-grade class contained 1.55% copper, compared to an average grade of 1.02% across all samples—a 52% enhancement. These findings highlight the promise of XRF-based sorting for classifying sedimentary copper deposits based on surface content. With further research, including real-time, high-speed detection systems, this approach could be validated at a pilot scale for broader industrial applications.