Multisensory Approach to Diagnose Bearing Faults Using Cohen Class Bilinear Distributions
Avyner L. O. Vitor, Clayton Luiz Graciola, Alessandro Goedtel, Wesley Angelino de Souza, Marcelo Favoretto Castoldi, Daniel Moríñigo-Sotelo, Óscar Duque-Pérez, Tomas Alberto Garcia-Calva · 2024
Monitoring the bearings of induction motors is essential to avoid downtime and optimize maintenance schedules. Manufacturing processes are typically dynamic, requiring time-varying methods to diagnose faults. The Wigner-Ville distribution is renowned for generating high-resolution representations in both time and frequency. Unfortunately, cross-terms are the primary problem of this technique, impairing many monitoring methodologies. Therefore, this work compares several modified versions of the Wigner-Ville distribution, known as Cohen-class bilinear distributions, in a multisensory strategy to diagnose the severity of bearing wear. Then, the Shannon entropy of each Cohen-class bilinear distribution is tested as a fault indicator using the Bhattacharyya distance to evaluate the potential of separability between the classes. The results demonstrated that Born-Jordan and Zhao-Atlas-Marks stood out among the Cohen-class bilinear distributions, whereas the entropy of vibration promoted more effective diagnostics than audio.