Rolling element bearing fault diagnosis using simulated annealing optimized spectral kurtosis

Abstract

To diagnose the bearing fault using vibration signal, methods like envelope analysis have been used. These methods need to locate the optimum frequency band to perform the analysis. Researchers have developed spectral kurtosis through kurtogram to detect the optimum frequency band. However, kurtogram uses a rigid structure of frequency filter bank and when the optimum frequency band does not match any of the frequency bands in the structure the fault may not be detected. In this paper a method based on simulated annealing is developed to locate the optimum frequency band. The method models spectral kurtosis as a function of the variables of a band-pass filter. Firstly the analysis result from the kurtogram is obtained as a start point, and then the central frequency and the bandwidth are optimized by maximizing spectral kurtosis through simulated annealing. Finally, the test signal is band-pass filtered by the optimized filter, and the envelope analysis is applied to complete the diagnosis. Experimental study shows that the method can diagnose the fault for different fault types. Being able to detect the real optimum frequency band, this method can strengthen the detection of the fault feature frequency component.