Data driven approaches to designing large open pit slopes - lessons from engineering geology

摘要

As mine open pits become increasingly aggressive and deep (beyond 400 m) due to exhaustion of near-surface resources or implausibility of underground excavation, significant challenges emerge using standard slope stability analysis techniques. Typically during overall failure analysis, factor of safety calculations are conducted. Although quite useful and despite recent advances in characterizing insitu stresses, the factor of safety approach has its inadequacies. For example, single factor of safety values cannot characterize an entire pit sector under varying geotechnical and environmental conditions. In this paper we draw on lessons learned from large dataset techniques in engineering geology to assess landslides. The proposed approach utilizes the emerging field of deep learning using artificial neural networks. Deep learning uses data-driven tools to continually update algorithms used to conduct computations resulting in high levels of accuracy and precision. Using our results and relevant examples from the literature, we discuss the benefits and shortcoming of the proposed approach, the appropriate conditions and types of environments for application and suggested modifications and improvements.