Objective The dynamic hazards of coals are intimately associated with the complexity and uncertainty of coal failure, while accurately identifying the propagation states of coal cracks serves as a critical approach to investigating the instability failure of coals. However, the current identification methods based on the physical quantities of acoustic emission suffer from slow speeds and low efficiency, failing to meet the demand for safe, efficient, and intelligent mining of coal mines.

Methods To achieve the intelligent identification of the crack propagation states of coals, this study constructed an optimization model based on the Newton-Raphson-based optimizer (NRBO) and the eXtreme Gradient Boosting (XGBoost)—a distributed gradient boosting library (also referred to as the NRBO-XGBoost model). Then, this study determined the corresponding relationships of multivariate parameters (i.e., values b representing the relationship between acoustic emission magnitude and frequency, S representing the activity, and RA and AF representing the fracture types) with the crack propagation states of coals. Using experiments on acoustic emission monitoring of coal failure, this study acquired the multivariate signal data of acoustic emission, which were subsequently used as training samples. Employing the NRBO-XGBoost model, this study performed intelligent identification of the stable and unstable crack propagation stages (stages Ⅲ and Ⅳ, respectively) of coals under varying loading rates, achieving adaptive optimization of XGBoost parameters. Furthermore, this study evaluated the performance of various models using four metrics: accuracy, precision, recall, and F1 score.

Results and Conclusions  The results indicate that in stage Ⅲ, coals exhibited slightly increased values b and S and a decreased proportion of shear cracks. After entering stage Ⅳ, coals manifested opposite variation trends in value b and the proportion of shear cracks, along with significantly increased value S. With an increase in the loading rate, value b and its fluctuation amplitude decreased, value S showed a reduced increment, and the proportion of shear cracks exhibited a rising increment. The NRBO-XGBoost model enabled the intelligent identification of the crack propagation states of coals based on multivariate acoustic emission parameters. Compared to the XGBoost and PSO-XGBoost models, the NRBO-XGBoost model yielded higher accuracy, precision, recall, and F1 scores, which were 90.69%, 88.79%, 99.00%, and 93.62%, respectively under low loading rates and 82.76%, 86.76%, 88.50%, and 87.62%, respectively under high loading rates. The results of this study provide a novel philosophy for the identification and intelligent monitoring of the crack propagation states of coals using the acoustic emission technique. Establishing prediction models based on the measured acoustic emission data allows for the monitoring and early warning of dynamic hazards of coals.