A novel data-driven method for ball bearing impedance modelling

Nowadays, electrical machines are commonly driven by voltage source inverters in which the intrinsic switching phenomenon can result in common mode voltage which can further lead to high frequency bearing currents. To accurately model the bearing impedance and in turn bearing currents, this paper proposes a probabilistic framework for bearing impedance modeling in inverter-driven applications. The impedance distribution is decomposed using the chain rule into a phase model and an amplitude model. A multi-layer perceptron (MLP)-based network is employed to predict the impedance phase distribution under given conditions and the corresponding amplitudes for different phases. This approach effectively captures both the transition from capacitive to resistive states, through phase behavior, and the associated amplitude responses, making it applicable across a wide range of shaft speeds, voltage amplitudes, and excitation frequencies. This modular approach aligns well with the physical processes, underlying the bearing breakdown. Additionally, it could be readily extended to incorporate additional parameters such as temperature and lubrication condition.