Cyclic band Box-Cox sparse measures based blind filtering and its application to bearing fault diagnosis

Blind filtering is one of the most important techniques for bearing fault diagnosis. Among these techniques, the Box-Cox sparse measures (BCSM) filter shows its effectiveness for bearing fault extraction. However, in practical applications, numerous cyclostationary sources can exist in a rotating machine, and the collected signals may not be Gaussian distributed. The abovementioned problems can decrease the effectiveness of the BCSM filter. This study overcomes these challenges by proposing a novel norm, cyclic band Box-Cox sparse measures, for blind filtering to extract the bearing fault more robustly. The proposed method aims to maximize sparsity in the envelope spectrum at particular frequencies of interest so that the influence of unrelated cyclostationary interferences can be minimized. In addition, by assuming the signal to be α stable distributed, the proposed method can extract the fault feature under non-Gaussian conditions. The robustness of the proposed method is validated through simulations and case studies.