Rock Hardness Ppediction Using Geophysical And Geochemical Data And Machine Learning - SME Annual Meeting 2022

A good understanding of the hardness of ore being handled and processed in a mining operation can significantly improve operational efficiencies by providing valuable data to support decision-making through the mining value chain (drilling, blasting, comminution). This study presents the results from the application of Machine Learning (ML) to predict rock hardness using various geophysical and geochemical features. Core samples from a mine site were logged using a multi-sensor core logging system. Measurements including ultrasonic p-wave and s-wave velocity, elemental concentration via portable XRF, and Leeb Hardness, were measured every 2 cm along the length of the core. The ML model was set up to predict the Leeb Hardness using the elemental concentrations and ultrasonic velocities as predictors. The Leeb Hardness values were grouped into three bins and used as a classification target for the ML models. Various ML models, including linear regression, support vector machines, decision tree, XGBoost, Random Forest, K-Nearest Neighbors, and Naïve Bayes were tested. The model performance demonstrated that the Random Forest and XGBoost models produced the highest accuracy of 74-91%. Random forest and XGBoost were then selected to perform regression models to estimate a Leeb hardness value rather than a hardness class. These models were able to estimate the Leeb hardness value by 4.57-4.87 % error.