Abstract
Hardness, as one of the most critical properties of rocks, has a significant role in different stages of mining projects in which its footprints can be tracked from the drilling to the comminution processes. An accurate and on-time estimation of the rock hardness of the mined materials would enable engineers and decision-makers to improve the operational efficiencies of different mining stages. Hence, this paper aimed to predict rock hardness using geochemical data. To do this, 162 rock samples were collected from the Canadian Malartic and Barnat open pits in Quebec, Canada, including different metamorphic and sedimentary rocks. The Leeb Rebound Hardness (LRH) was used to measure the rebound hardness of the samples due to its flexibility, practicality, and non-destructivity. An integrated statistical technique called small sample theory and associated confidence interval was used to account for the heterogeneity of rock samples and ensure the representativity of the measured mean Leeb hardness value. On the other hand, a multi-element geochemical analysis using a portable X-ray fluorescence spectrometry (pXRF) was performed to measure the chemical elements' bulk concentration of rock samples. The main reason to use the pXRF data as a proxy for characterizing rock hardness is that a pXRF device is a common tool among geologists and mining practitioners to assess rock geochemistry and can be used in different mining stages. Various machine-learning models, including Random Forest, Decision Tree, and Support Vector Machine, were used to predict the mean Leeb hardness value and rock hardness class. In all models, input parameters were a subset of either the original or the transformed geochemical observation, and the outputs were mean rock hardness value and rock hardness class. Results demonstrated that rock hardness could be reasonably predicted using the developed models based on pXRF data, even in the early stage of a mining project.