Wavelet Transform-based Higher-order Statistics for Fault Diagnosis in Rolling Element Bearings

Signal processing plays a pivotal role in fault diagnostics of mechanical systems. An approach, viz. wavelet transform-based higher-order statistics, was developed in this paper for fault diagnosis in rolling element bearings. In the approach, wavelet transform (discrete wavelet and wavelet packet transform) was introduced into a fourth-order statistic, kurtosis. Thereinto, discrete wavelet transform-based kurtosis (DWTK) was applied to signals to get a higher resolution in low-frequency bands; on the other hand, wavelet packet transform-based kurtosis (WPTK) was applied to obtain a relatively high resolution in high-frequency bands in comparison with the DWTK. DWTK, WPTK and wavelet transform-based kurtosis (WTK) curves were introduced to calibrate the in-field signals in comparison with the benchmark signals, whereby the non-stationary transients and singularity in the vibration signals attributed to damage were detected. WTK curves of vibration signals collected from bearing with damage of different severities and locations were evaluated for damage detection and classification. The results demonstrated the excellent capability of the WTK in vibration signal processing and fault diagnosis.