Fault diagnosis of rolling element bearing using cyclic autocorrelation and wavelet transform

Fault diagnosis of bearings under localized defects is essential in the design of high performance rotor bearing system. Traditionally, fault diagnosis of rolling element bearings is carried out performed by the use of signal processing methods, which assume statistically stationary signal features. This paper presents a feature-recognition system for rolling element bearings fault diagnosis, which utilizes cyclic autocorrelation of raw vibration signals. Cyclostationary analysis of non-stationary signals clearly indicates the appearance of several distinct modulating frequencies. The coefficients of wavelet transform are calculated using six different base wavelets, after calculating cyclic autocorrelation of vibration signals. The base wavelet that maximizes the Energy to Shannon Entropy ratio is selected to extract statistical features from wavelet coefficients. Finally, a comparative study is carried out with the calculated statistical features as input to soft computing techniques. Three soft computing techniques are used for faults classifications, out of which two are supervised techniques i.e. Support vector machine, Artificial Neural Network and other one is an unsupervised technique i.e. Self-Organizing Maps. The Complex Gaussian wavelet is selected based on maximum Energy to Shannon Entropy ratio. The results show that the support vector machine identifies the fault categories of rolling element bearing more accurately and has a better diagnosis performance.