Abstract: The eigenvalue of impact signal extracted on the basis of wavelet packet is applied to the mass estimation of loose parts, for wavelet packet has good local properties because of its multi-scale feature. Eigenvalue vector is obtained through normalization after receiving wavelet packet coefficients of each frequency band by using wavelet packet decomposition and its energy spectrum calculation. The mass estimation of loose parts is executed, with the eigenvalue vector as SVM input parameter and mass of loose parts as SVM output parameter. The experiments carried out to verify the proposed method in the plate test bed indicate that this method has good quality and stable performance of mass estimation for loose parts with different mass.

Key words: Energy spectrum, Loose parts, Mass estimation, Support vector machine, Wavelet packet