Resonance-based bearing fault diagnosis using wavelet packet decomposition

This paper presents a new bearing fault diagnosis based on detecting the impulses in the vibration signal caused by the damage. These impulses excite the natural frequencies of the system to resonate. The proposed algorithm uses wavelet packet decomposition (WPD) to localize the subband containing the frequencies of the system resonance. Decomposition of the vibration signal is performed only for the best wavelet packet tree, which results in lower computational complexity compared with other conventional methods. Moreover, reliable detection is achieved by utilizing de-noising techniques. Finally, studies on simulated and real vibration signals from defective bearings reveal that the proposed method effectively identifies the bearing faults even in case that the desired signal is buried in background noise.