Feature extraction of vibration signal of roadheader based on singular value decompositio

In view of difficulty of dynamic load identification of roadheader, feature extraction method of vibration signal of roadheader based on singular value decomposition was proposed. Collected vibration signals is decomposed by wavelet packet, and node coefficients at different frequency bands of each bottom layer are reconstructed to construct the time-frequency matrix. Then singular value decomposition of the matrix is performed, and based on Fisher criterion, class separability criterion based on divergence matrix is used to select singular value which is sensitive to hardness of different cut rock walls, and the value serves as feature quantities of the vibration signal. The criterion value of divergence matrix is used to solve the problem that it is impossible to measure quantitatively sensitivity of singular values to cutting hardness. Analysis results show that for vibration signals of roadheader under three cutting conditions of horizontal cutting, vertical cutting and longitudinal drilling, compared with feature vectors extracted by wavelet packet frequency band energy method, the feature vectors extracted by the singular value decomposition method have better class separability.