Research Of Signal Processing In Fault Diagnosis

With the development of science and technology and modern industry, equipment structure increasingly more complicated, systems of nonlinear stronger, outward features caused by system fault more complex, the diagnostic technique based on linear analysis generally is difficult to solve the diagnosis problems of large and complex equipments. Therefore, nonlinear and non-stationary signal analysis and processing become a key issue in fault diagnosis. Usually through the in-depth analysis of signal characteristics, people get signal source or system characteristics, operation and fault information. Important means to get signal source or signaling system characteristic information is signal analysis. Therefore fault diagnosis is based on signal analysis.This dissertation combines the traditional signal processing method and based on the latest non-stationary signal processing method to analyze the fault signal, so as to solve the actual problem. This dissertation studies the methods of signal analysis on identification of damage location. The main research contents can be summarized as follows:Traditional method. In this part, the method of the Fourier analysis and pattern recognition is mainly discussed. And some parameters of common identification are offered.Wigner distribution. In this part, the emphasis is placed on studying the characters and the fundamental thought methodology of Wigner distribution, and the correspond result of time-frequency distribution is discussed. Alao on the basis of nonlinear and interesecting term interference of Wigner distribution, and combining simulation and experimental analysis, we discuss the application of Wigner distribution in the rubbing fault rotor system, proved the validity of non-stationary signal and non-sensitivity to noise.Wavelets analysis. In this part, the emphasis is placed on studying the character and feature of wavelet theory. According to the excellent time-frequency localization analysis properties of the wavelet packet, this thesis describes the distribution of the transient signal quantitatively in time-frequency space by introducing the norm.Then proposes a method of noise cancelling and spectral analysis of transient signal. Compared with the Donoho soft-threshold method, the SNR is further increased. Theoretical analysis and the results of experiments both indicate that this method ia simple and reliable.