Advanced spatial processing of the cyclic spectral coherence map and its application to bearing fault detection in electric motors

Rolling element bearings are critical components in machinery whose early detection of faults is essential. Faulty bearings generate vibration signals characterised by cyclic impulses that correspond to the specific fault. Cyclic spectral coherence (CSC) is a valuable tool for identifying these faults; however, the automated interpretation of CSC results remains challenging, particularly when dealing with signals that have a low signal-to-noise ratio. This study aims to address this limitation by introducing statistical measures as weighting vectors to enhance fault detection. Specifically, the proposed method employs a frequency band selector (based on the carrier frequency f) and a fault frequency selector (based on the modulation frequency α ). The fault frequency selector can be considered an improved EES-like vector (IEES, Approach 1). Both selection vectors are used to reconstruct a map highlighting informative components. This denoised map is then aggregated to reconstruct the EES (REES, Approach 2). Finally, REES can be used as a 2D spatial filter applied to the original CSC map, resulting in a novel filtered CSC map. This filtered map is then aggregated to produce a filtered EES (FEES, Approach 3). The study presents three approaches, each utilizing two sparsity statistics: kurtosis and conditional variance. Validation using vibration data from two test rigs, as well as simulation and real vibration data, demonstrates that the proposed approach outperforms classical methods in terms of effectiveness.