An Extension of Correlated Kurtosis to Tunable Q-Factor Wavelet Transform in Fault Diagnosis of Rolling Element Bearings

In engineering practice, the type of rolling bearing failures are typical, and the fault signals are impulsive, which are influenced by the running status, the frequency component of the vibration signals become extremely complicated. Tunable Q-Factor wavelet transform (TQWT) method based on correlation kurtosis is proposed. This method decomposes the bearing fault signal into two different parts, high and low resonance components (HRC and LRC), basing on the different characteristics of signal oscillation. Then, calculating the correlation kurtosis of each decomposition layer of LRC to achieve signal reconstruction. Envelope demodulation analysis is performed on the reconstructed LRC. This method replaces the kurtosis index used in previous studies with correlated kurtosis index, which can effectively eliminate or weaken the influence of peak value interference. Finally, the simulation signals and wheelset bearing test signals are validated. The results show that this method can extract the features of bearing outer, inner and roller under strong background noise.