Research Of Rotating Machinery Fault Feature Extraction Based On Vibration Signal

Rotating machinery faults must be diagnosed as early as possible to avoid industrial accidents, and then it is the key of fault diagnosis how to extract useful fault feature. Having strong real-time, high reliability, convenience and other advantages, vibration signal analysis has been widely applied in the field of fault feature extraction. However, only a little useful fault feature information is usually submerged in the complicated observation signal, resulting in the difficulty of fault feature extraction by traditional or single signal technology. In the thesis, morphological wavelet and FastICA are studied in depth as the primary signal analysis means. Based on the advantages of adaptive morphological gradient lifting wavelet(AMGLW),robust fixed-point algorithm(rFastICA) and kurtosis, AMGLW-rFastICA-Kurtosis as a novel comprehensive analysis method is proposed to extract mechanical fault feature. The main study contents are as follows:(1) Fault mechanism of core parts in gearbox,such as gear and rolling bearing are presented, fault signal features are also analyzed to provide basis for latter studies.(2) So as to overcome the trouble of strong background noise contained in observed signal, AMGLW is employed as a de-noise method. It's steps and advantages are analyzed in detail. Finally, AMGLW is proved to have a better de-noising performance, comparing with the traditional morphological filtering and morphological wavelet in the simulation results.(3) Aiming at the problem of observed signal generated by nonlinear mixture of multiple vibration sources, rFastICA is used to separate structural vibration and fault frequency components from observed signal. The simulation proves rFastICA has a better separation.(4) In order to identify the separation signal including fault feature information, kurtosis index is introduced, which is highly sensitive to fault impact signal. The simulation results verify that kurtosis owns the excellent ability to identify fault signal.(5) From the perspective of complementary advantages, a novel comprehensive analysis method is thus proposed to extract the weak fault feature information, viz.AMGLW-rFastICA-Kurtosis. The experimental results show that the proposed scheme is effective and prominent for fault feature extraction.