Varian, J, McDougall, S, Ghadirianniari, S, Llewelyn, K, Campbell, R, Eberhardt, E & Moss, A 2022, 'Development of a wet muck spill susceptibility tool for short-term prediction through a logistic regression approach', in Y Potvin (ed.), Caving 2022: Proceedings of the Fifth International  Conference on Block and Sublevel Caving, Australian Centre for Geomechanics, Perth, pp. 1459-1470, https://doi.org/10.36487/ACG_repo/2205_102
(https://papers.acg.uwa.edu.au/p/2205_102_Varian/)
Abstract: The Grasberg mining complex in Papua, Indonesia, consists of three caving operations and an open stope mine. Two of the caves, the Grasberg Block Cave (GBC) and the Deep Mill Level Zone (DMLZ), are ramping up to full production, while the Deep Ore Zone (DOZ) cave will cease operations in 2022. The DOZ has had a history of production interruptions due to wet muck spill events. The newer caves expect to be affected by similar wet muck hazards due to the presence of fines and saturated material at the muck pile, overlying open pit in the case of the GBC and overlying caves in the case of the DMLZ, high annual rainfall, and complex topography at the subsidence that directs surface and groundwater into the cave.
To proactively manage this hazard, experience from the DOZ cave mine is being applied to improve the understanding of drawpoint wet muck spill susceptibility. The combination of fines generated through secondary fragmentation from the high draw columns and saturation from the intense surface and groundwater inflow results in wet muck material at the drawpoint, providing the cause while mucking activities provide the trigger. Other contributing factors included in the analysis are the uniformity of draw and neighbouring drawpoint conditions.
Although the consequences of wet muck spill events are high, they are still relatively rare, resulting in an imbalanced dataset. To overcome this challenge, cost-sensitive learning is incorporated into the logistic regression model for significant variables selection, thus developing a wet muck susceptibility tool. This tool aims to identify individual drawpoint susceptibility to wet muck spill events based on a simple material classification and measures of draw performance. The approach has been successful in describing historical drawpoint susceptibility at the DOZ. Furthermore, this study provides a concept applicable to other wet muck susceptible cave mines. 

Keywords: caving operation, wet muck spill, machine learning, wet muck susceptibility tool