Rock hardness and copper recovery prediction using textural clustering

The benefit obtained from the exploitation of an ore body is a result of the balance between its content of valuable metals and their recovery, and the cost of all the activities associated to the obtention of a commercial product. Rock characterisation data is crucial to understand, evaluate and optimise this balance. The description of rock composition through geochemistry and mineralogy, thanks to the past century's technological development, is well understood. Nevertheless, rock texture plays an important role in many stages of the business, but it remains elusive in its modelling and further quantitative application. In this study, the Mineral Co-Occurrence Probability Fields feature extraction method has been applied to a large dataset of hyperspectral imaging of rocks belonging to a South American porphyry copper deposit, with the aim of incorporating textural features into predictive modelling of rock comminution hardness (Semi-Autogenous Grinding Power Index – SPI) and copper flotation recovery. The results show a robust increase in precision and a potential increase in accuracy of the predictions when incorporating textural data, implying that the routine acquisition of hyperspectral imagery from drill cores can significantly improve the forecasting of metallurgical parameters reliant on rocks' textures.