Weak fault feature extraction of rolling element bearings based on ensemble tunable Q-factor wavelet transform and non-dominated negentropy

Tunable Q-factor wavelet transform (TQWT) has been proven to be usable in the fault diagnosis of rolling element bearings; however, its performance is heavily dependent on the selection of the Q-factor for decomposition and the optimal subband for reconstruction. In this paper, a novel method based on ensemble TQWT and non-dominated negentropy is proposed for weak repetitive transient extraction. Firstly, the vibration signal is decomposed with couples of Q-factors and redundancies to match the fault-induced oscillatory behaviors. Then, negentropy is utilized to evaluate the square envelopes and square envelope spectra of all subband signals from impulsiveness and cyclostationarity, respectively. After that, Pareto filtering is performed to search for the non-dominated set, and the knee point in the Pareto front is drawn on a distance metric for decision-making of the optimal subband. Finally, single branch reconstruction of the optimal subband is conducted to identify the fault characteristics for diagnosis. The effectiveness of the proposed non-dominated negentropy in weak fault feature extraction of rolling element bearings is verified by both simulation and experimental case studies. Furthermore, comparative studies also demonstrate its superiority over three peer ensemble TQWT methods.