3D Multibody Simulation of Realistic Rolling Bearing Defects for Fault Classifier Development

Rolling bearing faults stand out as the most prevalent type of fault in electrical machines. In this study, we leveraged geometry-based 3D multibody simulation to facilitate data-driven fault diagnosis. A comprehensive dataset was generated, encompassing data from both healthy and faulty bearings with realistic outer ring and inner ring faults of different types and sizes, operating at varying rotational speeds. Spectral analyses of the simulated bearing shaft displacement data proved that the bearing faults consistently appear at expected characteristic fault frequencies, with peak amplitudes correlating to the given fault size and rotation speed. Using the simulated data, we evaluated numerous feature engineering methods for machine learning-based fault classification. The classification results demonstrated a successful differentiation of simulated faults, whether on the outer ring or inner ring, from the healthy counterparts.