Dynamic Failure Predictions Using Machine Learning and Geochemical Data - SME Annual Meeting 2026

Dynamic coal failure events remain a safety concern in underground coal mining due to their sudden nature and high fatality rate. Previous research utilized Principal Component Analysis on field samples revealing strong empirical correlations between positive failure history and low sulfur content (S) and high volatile matter (VM). Further studies indicated that bumping coals tend to be less mature and less well-cleated than non-failure coals. However, these in situ studies faced the challenge of distinguishing intrinsic coal properties from external geological and stress-related influences. This study augments previous findings through laboratory experimental testing of coal samples that included unconfined compressive strength, chemical composition analysis, and qualitative observations on cleating, combined with compositional data analysis and machine learning techniques, to isolate and understand coal’s inherent bursting capacity. These results further confirmed the relationship between coal chemistry and failure classification.