Comminution hardness forecasting from hyperspectral drill core image derived textural clusters

In the mining industry, the analysis of rock characterisation data aims to estimate the profitability of a project across all mine-life stages, from assessing the fertility of a prospect to the quantification of metal tonnage in an operating mine. Nevertheless, as the grade of ore bodies decrease with time, the importance of metallurgical processes in the cost/value equation increases. This is because as ore grades decrease it becomes necessary to comminute larger volumes of rock for the extraction of a smaller amount of valuable material, while also seeking to achieve smaller liberation sizes suitable for recovery and processing. As a result, current comminution stages represent 70% of the total energy costs of a mine and metallurgical plant with this expected to substantially increase in the coming decade. In this study, we demonstrate an increase in the accuracy and precision of rock hardness predictive modelling, represented by the SPI (Semi-Autogenous Grinding Power Index), through the incorporation of automated textural clustering based on the Mineral Co-occurrence Probability Fields (MCOPF) method. The SPI parameter relates the energy needed to reduce the size distribution of a specific rock sample in a semi-autogenous mill. The utility of MCOPF was established on a large dataset of hyperspectral drill core images from a porphyry Cu deposit in Chile where it was used to quantitatively model rock textures. When incorporating MCOPF derived textures into the SPI model a robust increase in both accuracy and precision of SPI predictions was observed. This implies that the routine acquisition of hyperspectral drill core imagery and subsequent processing using MCOPF can significantly improve the forecasting of hardness and potentially other metallurgical parameters reliant on rock textures