Mine fires and other hazards caused by spontaneous coal combustion are a pervasive and longstanding issue in Jharia coalfields, India. This study proposes a novel approach to classify coal seams based on their propensity to spontaneous combustion using the intrinsic properties of 30 coal samples from different seams. This method eliminates the need for expensive and time-consuming experimental determinations of susceptibility indices (SI) such as crossing point temperature (CPT), critical air blast (CAB), and differential thermal analysis (DTA). All clustering models, viz. hierarchical, k-means, and multidimensional scaling, aptly classify coal seams into three categories: highly risky, medium risky, and low risk in terms of the tendency for spontaneous coal combustion. The results from unsupervised clustering for predicting the fieriness of coal seams match with on-field reports based on the history and nature of seams. The clustering results are also in concurrence with the SI which are generated through lab investigations. Furthermore, three machine learning (ML) algorithms, namely support vector machines (SVM), random forests (RF), and elastic net regression (EN), are used to comprehend the relationship between the coal's intrinsic properties of coal and SI. The actual nature of coal in these seams on the ground confirmed the findings of this study. The proposed methodology has practical implications for mine managers, as it can quickly provide safety risk assessment information to maintain safety and minimize economic losses due to unforeseen incidents.

Leveraging intrinsic properties for classification of coal seams towards spontaneous combustion proclivity and predicting susceptibility using machine learning: smart and sustainable mining approach