Abstract
Accurate resource estimates are critical for a good mine plan to be operationally efficient and mitigate financial risk. Traditional estimation methods such as kriging or inverse distance weighted can provide a sufficient overall average representation for resource modeling and on a micro-scale for ore control. However, the challenge in developing shorter-range models is balancing their timely completion with the availability of adequate localized data for accurate estimation. A few contributing factors include fixed automated modeling parameters, short durations between the drills and the shovels, sampling resolution, and turnaround time on sample analysis. This paper introduces an innovative in-house developed platform, airth Act Mine ForesightTM, which utilizes machine learning (ML) techniques that enhance short-range model estimation accuracy and adaptability by leveraging a combination of resource model data and historical blasthole sampling data to make grade predictions for target planning areas below the active mining bench.
To support the above hypothesis, airth.io collaborated with Capstone Copper, where Mine Foresight was implemented, and the results of the ML-based grade prediction models were reconciled against a standard interpolation approach. The case study provides evidence of the accuracy and reliability improvements made in the short-range estimation model.
Looking forward, we discuss the beneficial implications of these findings which should impact mine planners in the designing of shovel advances, material routing, and maximization of production plans. The ML-based approach can improve short-range grade estimations, but also sets the stage for further innovations in iterative resource modeling and reconciliation.