During the process of top coal caving, sensors are used at the coal discharge outlet to identify coal and gangue, which can enhance the mining efficiency and ensure the safety of personnel. Due to the extremely poor visual and auditory conditions in the underground environment, vibration signal monitoring technology has been introduced into the coal gangue identification task. This study focuses on three-axis vibration signals as the research object, aiming to fully explore the potential information of three-axis vibration signals and improve the reasoning ability of the model by integrating the multi-layer time-frequency information of three-axis vibration signals. A preprocessing module for vibration signals has been designed. This module achieves fine denoising by performing different denoising processes on each component after modal decomposition. Given the shortcomings of traditional wavelet thresholding, this study designs an optimized wavelet threshold function. A multi-layer time-frequency information fusion framework for coal gangue identification is proposed. This framework integrates information in different ways at the data layer, feature layer and decision layer, ultimately achieving an accuracy of 95.50%. The proposed identification framework can provide a design basis for the subsequent end-to-end model design based on neural networks and enhance the interpretability of the design process. Moreover, this study opens up a new path for coal gangue identification based on three-axis vibration information and lays a solid foundation for the exploration of multi-modal fusion technologies in the future.
Jw et al. (2026) studied this question.