Sparsity-driven multimodal narrow-band deconvolution for rolling bearing compound fault diagnosis

Abstract The diagnosis of compound faults in rolling bearings presents significant challenges, particularly in extracting discriminative features under low signal-to-noise ratio conditions and decoupling features embedded within overlapping resonance bands. To address these challenges, this paper proposes a new deconvolution inverse filter. First, a multimodal narrow-band filter is designed based on the shock pulse function for the initial deconvolution filter. Second, an objective function named envelope spectrum interference-to-harmonic ratio (ESIHR), to promote spectral decoupling, is designed to update the filter parameters via gradient descent. Finally, the filter parameters are optimized by minimizing the ESIHR. This process enhances the sparsity of the envelope spectrum by increasing fault-related information and reducing interference components, resulting in the effective separation of combined faults and noise suppression within overlapping resonance bands. The effectiveness of the proposed method is demonstrated through both simulation and experimental data. Comparison with common deconvolution methods confirms the superiority of the approach described in this article.