A weighted multi-scale L1/L2 deconvolution method and its application for bearing fault diagnosis

Incipient weak fault diagnosis is is of paramount importance for the secure functioning of rotating equipment. Detecting masked weak fault impulses is challenging due to significant background noise and various interferences. Visual deblurring (BD) stands out as a proficient approach for extracting repetitive transients, but the performance of conventional BD methods is often compromised by strong interferences, such as pronounced harmonics and random impulses. This paper proposes a weighted multi-scale L1/L2 deconvolution (WMSD) method to address these challenges. The WMSD method utilizes the adjustable Q-factor wavelet transform to construct a multi-scale space. In this space, the weights of all scales are determined by power spectrum kurtosis (PSK) to retain the fault characteristic energy, forming the weighted multi-scale space. The scale containing abundant fault characteristic information is identified based on the combined amplitude-weighted PSK. The frequency band of this scale is then used to initialize the L1/L2 deconvolution filter, allowing the deconvolution model to focus on fault characteristic information. The weighted reconstructed signal is subsequently processed using the initialized deconvolution model, and the fault is diagnosed using the envelope spectrum. The resilience and efficiency of the WMSD technique are confirmed by its adept handling of fault signals that exhibit intense harmonic content and considerable erratic impulses.