Study On Theory And Method Of Coal-Gangue Interface Recognition On Fully Mechanized Caving Face

Coal-gangue interface recognition (CIR) in fully mechanized caving face was systematically studied in this thesis. Firstly, the basic principles of detect coal-gangue interface using vibration method and the structure of CIR experimental system were introduced. The theoretical model of CIR was also established. Then, the vibration features of coal-gangue were extracted by classic time-domain analysis and spectrum analysis. In addition, the intuitive time-frequency spectrum was given by STFT, Wigner-Ville distribution and wavelet transform techniques. The feature extraction methods of coal-gangue vibration signals based on Hilbert-Huang transform (HHT) were discussed emphatically. Considering the end effect of HHT, the selection algorithm of intrinsic mode functions (IMFs) based on correlation coefficient was given. Consequently, the four feature extraction methods based on IMFs energy, singular value of IMFs matrix, information entropy of Hilbert spectrum and marginal spectrum energy were proposed respectively. The simulation experiments of CIR were conducted under different feature extraction methods, which were proved that the classifiers based on Mahalanobis distance, BP neural network and support vector machine all possessed good classification performance. Lastly, the top-caving technology based on CIR was presented.