Ore guiding in underground mining using MWD and machine learning

The combination of chemical analysis techniques as XRF and measurement-while-drilling (MWD) technology in underground mining is a novel approach to gather comprehensive on-line information about physical and mineralogical data. In this work their integration is investigated aiming to delineate the ore/waste boundary. The input data analyzed come from a narrow vein underground mine, where the ore correspond to fluorite, and gathered in two production levels with complex mineralogical characteristics. Two approaches are followed to build a generalizable classification methodology and a site-specific model: a k-means clustering heuristic algorithm to define rock classes based on their physicochemical features, and a machine learning (ML) ensemble model trained with drilling parameters. For an imbalance class problem, a good recognition is obtained for the ore when is related to a high content of silica and for the waste. To predict the ore with low silica content it is necessary to integrate an analysis-while-drilling method to recognize the fluorite content when this mineral is embedded in different rocks. MWD can be employing for the lithologies recognition when representing different structural and mechanical properties between them.