Failure prediction of gas-bearing coal–rock combination with different high ratios based on acoustic emission response characteristics

Abstract

To address the problems of unclear main control factors, incomplete acquisition of precursory feature information and poor model prediction effects are needed in the prediction of damage and failure of coal–rock combinations. Based on the multi-index characteristic data of acoustic emission (AE) during the uniaxial compression process of gas-bearing coal–rock combinations, the time–frequency characteristics of AE signals in the whole process of damage and failure are analyzed, and a prediction model of the deep residual attention denoising network (DRADN) is proposed. The dominant factors influencing the damage of coal–rock assemblages are identified using the ReliefF algorithm. A Conv-Sparse attention module (CSAM) is designed to extract the temporal feature information of the main controlling factors in the process of damage and failure, and a soft threshold is introduced to optimize the time–frequency features for eliminating noise information. A time–frequency feature transfer module (TFFTM) is established to learn the spatial structure of features. The nonlinear relationship between the mapping feature information and the risk type of the multilayer perceptron network is used to predict the risk level of damage and failure of the gas-bearing coal–rock combination. Experimental results demonstrate that the constructed damage and failure indices effectively guide the stage division of uniaxial compression. Comparative evaluation reveals that the DRADN achieves accuracy, precision, and recall rates of 98.63%, 98.73%, and 98.63%, respectively. The generalizability and stability of the DRADN under the experimental data of different high-ratio combinations, as well as its better noise resistance in noisy environments, are verified via cross-domain experiments and simulated noise experiments. This study expands the application of AE technology in the prediction of coal–rock risk levels and provides a technical reference for the prevention and control of gas-bearing coal–rock dynamic disasters.