Intelligent Fault Diagnosis of Rolling Element Bearing Based on SVMs and Statistical Characteristics

In this paper, the statistical characteristics of time, frequency and time-frequency domain are applied to discriminate various fault types and evaluate various fault conditions of rolling element bearing, and the classification performance of them is evaluated by using SVMs. Experimental results showed that the statistical characteristics Mean, Variance, Root, RMS and Peak of the 2⁵ sub frequency bands in frequency domain obtain higher classification accuracy rate on all the fault datasets than the statistical characteristics in the whole time and frequency domain. Wavelet packet decomposition is an efficient time-frequency analysis tool, and it can decompose the original signal into independent frequency bands. Experiment on the statistical characteristics of the 5th level wavelet packet decomposition showed that the statistical characteristics Variance, Root, RMS and Peak can discriminate various fault types and evaluate various fault conditions well on all the datasets. Compared with the statistical characteristics of sub frequency bands in frequency domain, the classification performance of the statistical characteristics of the wavelet packet transform is a little lower than that of the statistical characteristics of sub frequency bands in frequency domain.