Weak Fault Diagnosis Of Rolling Element Bearing Based On Improved Spectrum Kurtosis

Diagnosis technology is based on seeing, hearing, smelling in early years. It needs rich experience, and these methods have low efficiency. Mechanical equipment become more and more large-scale, integrated, intelligent and precise. And also the requirement of the reliability and stability of the equipment is constantly improving, the diagnosis research of mechanical equipment become more and more important. It can't meet the requirements of modern industry to maintenance and repair instruments after a fault has occurred, and it is very necessary for industry to monitor and recognize fault in early stage, which can reduce the incidence of failure, reduce maintenance cost and increase economic benefits. This paper has researched a rolling element bearing fault diagnosis method based on spectrum kurtosis. It proposed three methods:a spectrum kurtosis based on FFT, enhanced single bandwidth demodulation spectrum kurtosis, particle filter spectrum kurtosis. These method have been validated by several vibration data.Spectrum kurtosis based on FFT has been proposed, in order to reduce influence of noise, the EEMD has been used before FFT. After FFT, calculate FFT spectrum kurtosis value of each sub-band in different layers according the principle of Wavelet Package Transform(WPT), then gather spectrum kurtosis value from different layers to obtain kurtogram. Sub-band which has the maximum kurtosis value will be selected for filtering and demodulation, at last calculate power spectrum of envelop from filtering. This method has been validated by simulated signal and experimental data.Put forward a method named single bandwidth demodulation spectrum kurtosis. Do WPT for original data, the decomposition layer N can obtained by narrow band principle from Protrugram. Calculate kurtosis value of each sub-band wavelet packet coefficient in layer N. Select the band that has the maximum kurtosis value to reconstruct, then calculate envelop spectrum and power spectrum. At last calculate autocorrelation of power spectrum, that is the enhanced spectrum. Simulated signal and engineering data have been used to validate the performance of this single bandwidth demodulation spectrum kurtosis. The result shows that this method is effective.Use particle filter together with fast kurtogram to detect weak fault information. Build state equitation of original data based on AR model, the sum of state equitation and background noise is set as the observation equitation. Put these two equitation together to build state space equitation. Put parameters into state space equitation to estimate original vibration signal, the estimated signal is the noise-reduced signal. Calculate fast kurtogram of noise-reduced signal, select the band which has the maximum for envelop analysis, then calculate power spectrum of envelop signal. Several vibration data have been used to validate this method. The result of fast kurtogram and EMD-kurtogram are shown in this chapter, so it can has a comparison analysis.