Research on coal gangue recognition based on multi-layer time domain feature processing and recognition features cross-optimal fusion

Abstract

Accurate and rapid recognition of coal gangue is an important link to realize automatic mining, which is helpful to improve the efficiency of coal mining and the safety of mining workers. Relevant studies have confirmed the effectiveness of coal gangue recognition based on tail beam vibration signal, but the recognition accuracy of a single point source vibration signal of tail beam is slightly low. In order to improve the practicability of coal gangue recognition method based on tail beam vibration signal, this paper takes the single source tail beam vibration signal as the object, improves the coverage of tail beam vibration characteristics by conducting experiments and extracting multiple vibration signals at different positions, and conducts the research on coal gangue recognition technology by multi-layer time domain feature processing and cross-optimal fusion of multiple signals. Firstly, through analysis of classification algorithm ability and coal gangue recognition principle, the coal gangue recognition process is formulated and the classification model is constructed. Then, based on the two data samples of a single signal eigenvector and multi-signal eigenmatrix obtained by Direct Data Statistics (DDS), the recognition research is carried out, and the recognition effectiveness of the eigenmatrix is verified. Based on this conclusion, the three-layers eigenmatrix data samples of DDS, EMD Data Statistics (EDS) and HHT Data Statistics (HDS) are obtained by multi-layer time domain feature processing, and the coal gangue recognition based on multi-layer eigenmatrix is carried out respectively. DDS is the most accurate signal processing method. On this basis, the construction method of DDS + EDS and DDS + HDS recognition eigenmatrix based on array series is studied, and the recognition ability of DDS + HDS recognition eigenmatrix construction method and Logic Regression algorithm is proved. Recognition accuracy based on DDS + HDS recognition eigenmatrix and Logic Regression algorithm is up to 95.25 %. Finally, in order to further improve the recognition accuracy and response speed, this paper proposes a method of feature fusion based on cross-optimal selection (FFCOS). Features of DDS and HDS are sorted by recognition sensitivity, and then cross-selected and fused according to this sort. The results show that the recognition accuracy of the serial construction method of DDS + HDS recognition eigenmatriX array and the Logistic Regression algorithm can reach 97 %, which proves the effectiveness of the FFCOS method and realizes the accurate recognition of coal gangue based on the vibration signal of single source tail beam.