Abstract

The brittleness index is a critical parameter for evaluating rock bursts and catastrophic failures in deep underground mining. Accurate prediction of this index is vital for monitoring rock bursts, which pose risks to miners and structural integrity in engineering projects. This study introduces a new mathematical brittleness index, incorporating crack initiation, crack damage, and peak stress, tailored for sandstones with varying water contents. The proposed index is benchmarked against established brittleness indices (B1, B2, B3, B4) using infrared radiation (IR) characteristics, specifically the variance of infrared radiation temperature (VIRT), and advanced artificial intelligence (AI) models, including k-nearest neighbor (KNN), extreme gradient boost (XGBoost), and random forest (RF). Key findings include: (1) crack initiation, elastic modulus, crack damage, and peak stress decrease as water content increases; (2) brittleness indices B1, B3, and B4 exhibit a strong positive exponential correlation (R² = 0.88) with water content, while B2 shows a negative exponential correlation (R² = 0.82); (3) the new brittleness index demonstrates robust linear correlations with B1, B3, and B4 (R² > 0.85) but a weaker negative correlation with B2 (R² = 0.61); (4) the RF model outperforms others in predicting the brittleness index, achieving R² = 0.999, RMSE = 0.383, MSE = 0.007, and MAE = 0.002. Thus, the RF model is recommended for precise brittleness prediction. These findings provide valuable insights for developing effective brittleness indices to mitigate rock burst risks in water-influenced rock engineering projects.