Bearing fault diagnosis based on singular value distribution of impulse response segment.

Extracting periodic impact features from vibration signals has always been a key issue in the fault diagnosis of rolling element bearings. However, the repetitive impacts induced by localized defects are difficult to identify due to the presence of background noise and interferences. A novel approach for bearing fault diagnosis based on singular value distribution of impulse response segment is proposed. The characteristics of singular value decomposition (SVD) of the impulse response are analyzed, and the relationship between the matrix row number and the bandwidth of subspace is estimated quantitatively. According to the unique distribution of singular values, the double-order attenuation ratio (DAR) is designed to evaluate the transient component of the short-time segment. Then, by segmenting the vibration signal, the time-dependent DAR sequences are obtained, which can be used to locate the impacts in the signal. Eventually, the fault-related cyclo-stationarity in DAR sequences is enhanced by autocorrelation and measured by the Gini index. The effectiveness of the proposed method is verified by simulation and bearing fault datasets, in contrast to the state-of-art algorithms.