EPEOgram: An Optimal Frequency Band Identification Method for Rolling Bearing Compound Fault Diagnosis

Spectral coherence (SCoh) is an effective tool for revealing the second-order cyclostationary of bearing faults. Identifying the optimal demodulation bands (ODBs) is key to accurately diagnosing bearing failures. However, most methods focus on the demodulation of a single fault information band, neglecting the discrete resonant frequency bands or multiple informative frequency bands, which may lead to the omission of critical fault information. In addition, the reliance on prior fault characteristic frequencies limits the practical application of complex fault diagnosis. To solve such issues, an ODBs identification method namely equal-proportional expansion optimization-gram (EPEOgram) is proposed for rolling bearing compound faults. First, the original SCoh is reconstructed to automatically identify the potential cyclic frequencies (PCFs) in each spectral frequency slice (SFS) based on the proposed autocorrelation cycle ratio (ACR) calibration algorithm. Next, the PCFs are used as fault-type labels to construct a weight metric to evaluate the abundance of the fault information contained in each SFS, and then to locate the initial center frequency band for optimization. Finally, an equal-proportional expansion frequency band optimization structure is developed to generate the EPEOgram for identifying the ODBs in each initial center band, and the ODBs with similar PCFs are clustered to construct ODBs cluster weighted envelope spectrum (OCWES) for bearing fault detection. This method does not require prior fault information and can accurately and quickly identify the ODBs for the discrete information frequency bands. The simulation and experimental results indicate that the proposed method can effectively identify multiple fault information frequency bands with rich fault information and less noise interference. In addition, comparison with state-of-the-art methods highlights the superiority of the proposed methodology.