Feature Extraction for Rolling Element Bearing Faults Using Resonance Sparse Signal Decomposition

Abstract

Rolling element bearings are widely used in a variety of rotating machineries. If the rolling bearing elements are damaged, a cyclical impact transient signal and the vibration signal modulation phenomenon appears when the fault surface contacts other components of the rolling element bearing. To demodulate the cyclical impact signal and extract the bearing fault information, this paper proposes a new method based on resonance-based sparse signal decomposition (RSSD). First, the bearing vibration signal is decomposed into three components via RSSD. The high-resonance component contains a sustained oscillation cycle signal, the low-resonance component contains the impact transient signal, and the final component is the residual. The sub-bands near the natural bands are extracted for demodulation into two components. Two main sub-bands are obtained by summing these sub-bands. Next, these two main sub-bands are summed to obtain the original signal’s main sub-band. Finally, the auto power spectrum is extracted using envelope signal autocorrelation processing, and it reflects the degree and location of the fault in the rolling bearing. To verify its effectiveness in extracting fault information, the proposed method is applied to two practical application examples with an inner race fault and an outer race fault in a rolling bearing, respectively. Compared with envelope analysis and wavelet analysis, the results indicate that the spectra obtained with this method exhibit less burrs and a higher signal-noise ratio, and outperforms the other spectra in terms of revealing the amplitude modulation frequency of the fault impact response.