Optimization of a coal mine roof characterization model using machine learning

Predicting areas of increased propensity for roof deformation is crucial for the proactive management of geotechnical risk in underground coal mines. Current practices rely largely on assessing rock mass strength or characterization indices in isolation. Validation of applied ground support systems typically, and in part, comprises a review of observed deformation and analysis of extensometer data from adjacent previously mined areas. This information is seldom routinely assessed relative to the rock mass strength or characterization data. Typically, roof deformation is a response to several combining factors such as mining-induced stress, roof lithology, rock mass strength, and roadway geometry. Therefore, it is optimal to develop an approach to integrate these factors to quantify their relative significance to causing roof deformation, with the goal of establishing a reliable quantitative model for predicting roof deformation. This paper provides a case study in machine learning using Artificial Neural Network (ANN) and geophysical log data information from an underground coal mine located in the Bowen Basin (Australia), whereby a suite of parameters is considered pertinent to excavation stability. This machine learning predictive model analyses key geotechnical parameters including field and mining-induced stress, rock mass strength, clay/shale content, and strata monitoring data to identify strata hazards. Finally, the classification model can indicate the risk of roof displacement and observed roof deformation as communicated through a contoured 'geotechnical hazard map'. This hazard map can assist site engineers understand roof conditions in underground coal mining and thus put in place operational mitigation processes to reduce geotechnical risk on mining. This could potentially improve the identification of high-risk areas and inform future bolting requirements to maintain a safe working environment and minimise production interruptions.