REPESgram: A Novel Method for the Optimal Demodulation Band Selection in Rolling Bearing Fault Diagnosis

Due to the modulation of hardware transmission paths and interference from environmental noise, identifying early localized faults in rolling bearings is challenging. Fixed-bandwidth filter banks are used in traditional narrowband demodulation methods, which mean that the ideal demodulation frequency band does not fully involve fault information. An optimal demodulation-band selection method based on adaptive filter banks is proposed in this article. First, in response to the limitations of the Locmaxmin method, it is proposed to utilize the Welch spectral density estimation (WPSD) method to address the issue of overly concentrated spectral peaks and to calculate the central local minimum of adjacent local maxima to determine the segmentation boundaries. This results in a bandwidth-adaptive filter bank that preserves the richness of fault information in the target frequency band. Afterward, an index called the ratio of energy in the PES (REPES) is proposed for identifying sensitive fault frequency bands, which can extract the periodic cyclic impact information from signals across various aspects based on the product envelope spectrum’s (PESs) superior cyclic stationarity detection capability. Lastly, the method was compared with other advanced methods and validated via data from the bearing simulation signal, artificial fault bearing experiments, and accelerated degradation test. The results demonstrate that this method is superior to other fault diagnosis methods and can reliably identify fault-related spectral bands, effectively achieving rolling bearing fault diagnosis.