Enhanced Speed Control of SPMSMs via Physics-Informed Neural Network-Based Active Disturbance Rejection Control

This paper presents a novel Active Disturbance Rejection Control (ADRC) strategy enhanced by a Physics-Informed Neural Network (PINN), designed to compensate for the rotor speed derivative-i.e., speed acceleration, which is typically neglected in conventional ADRC implementations. Owing to the difficulty in accurately measuring speed acceleration and its inherent nonlinear characteristics, the motor's dynamic equations are embedded within the neural network, and automatic differentiation is utilized to estimate acceleration. A physics-guided loss function is constructed to ensure the convergence of the network output. The PINN architecture is specifically designed to address complex nonlinear mappings while operating within the computational constraints of microcontrollers. In comparison with two representative ADRC configurations, the proposed method demonstrates improved dynamic performance. Its effectiveness is experimentally verified on a 1-kW permanent magnet synchronous motor (PMSM) drive platform.