Unknown Bearing Fault Recognition in Strong Noise Background

The fault pattern of rolling bearing usually is unknown in heavy noise background, which means the bearing has two possibilities of single part and compound fault. A new diagnosis method of bearing unknown fault is proposed, which can adaptively diagnose the bearing unknown fault with single part or multi-site. The ensemble empirical mode decomposition with the decomposition cut-off frequency is used to decompose noisy signals. Combined with the sample entropy index, a target frequency band screening criterion is structured to select the target frequency band that contains the potentially fault feature. Meanwhile, the advantage of sample entropy in anti-noise interference is analyzed. Subsequently, the spectral magnification is designed as the quantitative indicator to get the optimal system output of target frequency band when the fault frequency is unknown. Finally, the resonance factor is designed to eliminate the false fault caused by coherence resonance. The feasibility of the proposed approach is confirmed with both simulated and experimental signals.

Automated summary

A new diagnosis method for bearing unknown faults is proposed in this study, which can adaptively diagnose faults in single or multiple parts of the bearing. The method utilizes ensemble empirical mode decomposition and sample entropy index to select the target frequency band containing potential fault features. The advantages of sample entropy in anti-noise interference are also analyzed. The proposed approach is validated using both simulated and experimental signals.