Data analytics for open-pit mining: examining vehicle interactions, material movement and compositional uncertainty with bucket inference and Monte Carlo simulation

This article examines the use of data analytics in mining automation. It considers the benefits of integrating telemetry data in computational systems that model the state of a mine. Specifically, it focuses on interactions between load-haul vehicles and the information that may be harvested to facilitate material tracking in surface mining. Bucket dig positions are inferred from GPS data by analysing wheel-loader and excavator interactions with haul trucks. Evaluation shows this approach achieves higher precision and load-haul cycle recall, as extra buckets missing from the existing OEM source are discovered. Analysis reveals a consistent pattern of behaviour from equipment operators which justifies the use of a simple model that neglects the articulated motion of the excavator arm. A major contribution is the integration of two technical objectives: the ability to consistently locate excavation points within mining pits based on vehicle interactions, and describe compositional variation in the excavated material using accurate local grade models informed by geochemical assays. This allows material movement associated with individual load-dump events to be tracked and linked with the underlying geology via kriging estimates. The case study highlights the importance of having high-resolution data, as it enables transient ore dilution events (e.g. inadvertent transferral of waste to stockpiles) to be identified and factored into risk assessment. The information generated may be used for ore-properties tracking in a dynamic graphical model to capture correlations, uncertainties and compositional changes induced by material movement. Recent development and application opportunities including the role of machine learning are briefly discussed.