The Early Crack Source Signal Extraction Of The Key Parts

With the advancement of technology and the rapid development of the productive forces, the mechanical parts carry the various faults and the higher risk of accidents in order to satisfy higher requirements. The direct losses can be avoided timely by early crack signal extraction. Therefore, the extraction of early crack source signal of the critical mechanical parts has a very important significance.The weak crack source signal processing and feature extraction are the key problems. The main research in the paper include:1) The lead-break signal and crack signal caused by metal tensile were extracted by wavelet analysis, which is effective for non-stationary signals. And it achieved good results. However, the choice of the threshold of Wavelet analysis is determined by experience. And the extraction is ineffective when the SNR is low.2) Aimed at the problem that the crack source signals of the mechanical parts are often mixed with various vibration and strong background noise, which caused great difficulties for extracting the characteristics of the signal, the independent component analysis (ICA) was used. It can extract signal on the conditions of band-aliasing with noise, strong non-steady-state and non-gaussian, or strong background noise. The fixed-point independent component analysis on the basis of negative entropy was used. The lead-break signal and steel crack signal with different strength Gaussian white noise were studied. The SNRs, correlation coefficients and spectrograms show that the method can obtain the better result for low SNR signal.3) On the condition of the number of observation signal is no more than the number of source signal, that is the under-determined condition, the sparse coding shrinkage denoising based on ICA was used to extract the weak signal. The essence of the method is to seek the most appropriate estimation for probability density function of source signal. The generalized Gaussian model was used to fit the probability density function. The maximum a posteriors was used to estimate the independent components and the nonlinear shrinkage functions were used to denoise. The denoised signal was obtained by pseudo inverse transformation. The signal with noise was studied by the method. The results show that the method is not better than the ICA, but it’s practical.