Graph and frequency leverage: a novel coal gangue identification method

Abstract Coal holds a significant position in the energy structure of China. Top-coal caving mining is one of the classical methods in coal mining technology. However, considering the hazardous environment of actual working faces, the identification accuracy can be greatly reduced. In this paper, with the aim of improving the identification accuracy, a graph conversion model based on the impact and vibration signals of coal and gangue in the tail beam of hydraulic supports was proposed. The signals utilized in this coal and gangue identification test comes from the caving face of Shanxi Tashan Mine No.8222. The acceleration sensor was attached to the tail beam of the hydraulic support by magnetic force to collect the impact and vibration signals of coal and gangue from the tail beam. Fast Fourier transform was firstly utilized to extract frequency domain features. After that, the features were extracted in the frequency domain and converted into a graph data structure. The extracted features were then utilized for constructing graph representations based on their similarities utilizing the proposed graph generation model. The identification of coal and gangue was conducted on the constructed graph representations by graph convolution networks. Finally, several comprehensive experiments were conducted to prove the effectiveness of the proposed method, with the results showing that the proposed method achieves higher accuracy compared to baseline methods, and requires relatively fewer parameters. By the proposed of baseline methods, this paper provided a potential direction for the accurate coal and gangue identification.