Physics-Informed Neural Network Approach to Diagnosing Uniform Demagnetization Faults in Permanent Magnet Synchronous Machines

Permanent Magnet (PM) Demagnetization (PMD) faults in Permanent Magnet Synchronous Machines (PMSMs) pose significant challenges to their reliability and performance, necessitating advanced diagnostic techniques for early detection to prevent catastrophic failures and minimize downtime. This paper proposes a Physics-Informed Neural Network (PINN) approach for diagnosing PMD faults in PMSMs. The PINN integrates the dynamical model of the PMSM with measured data to estimate the PM flux linkage under healthy and uniform PMD conditions, thereby quantifying the severity of the PMD. By incorporating the physical equations of the motor, the PINN offers higher data efficiency and enhanced robustness compared to existing conventional data-driven methods. The estimation of PM flux linkage is validated through both simulation data obtained from a finite element analysis (FEA) model of the PMSM and experimental measurements from a laboratory test rig under steady-state and dynamic operating conditions. These evaluations demonstrate the method’s capability to accurately quantify the severity of demagnetization faults. Furthermore, the robustness of the proposed approach to measurement noise and model parameter variations and its data efficiency are assessed. Finally, a hyperparameter tuning strategy tailored for PINNs operating in parameter estimation mode is introduced.