Guarded Two-Time-Scale UKF and Adaptive NMPC for Voltage-Disturbance Rejection in PMSM Drives

Abstract

Nonlinear model predictive control (NMPC) of permanent magnet synchronous motor (PMSM) drives depends on accurate state estimates and an accurate internal motor model. Slow changes in motor parameters and rapid inverter voltage disturbances can produce similar residuals in the measured currents. Therefore, an unrestricted augmented estimator misinterprets an actuator disturbance as a change in flux linkage or load torque. This leaves the controller with a biased prediction model. This paper coordinates a guarded two-time-scale unscented Kalman filter (UKF) with adaptive constrained NMPC. A two-level test based on the dq-axis current innovations identifies and confirms an electrical event candidate before enabling disturbance cancellation. The candidate logic protects the slow flux-linkage and load-torque estimates, while a confirmed event increases the update authority for the voltage-disturbance states. The NMPC system updates the motor model while maintaining flux and load torque values at a detected electrical event. Tightened constraints ensure that some inverter capacity remains available to correct the rated voltage error without exceeding the voltage limit. Sensitivity analysis indicates that voltage and flux errors can produce similar current responses. Deterministic simulations and 30 paired Monte Carlo trials confirm the overall performance. Compared with adaptive NMPC, the proposed controller reduces the mean speed RMS error by 33.8%, the error during the disturbance interval by 79.8%, and the RMS current by 31.1%. The results show that protecting the slow states improves error attribution and that disturbance cancellation after event confirmation is the main source of improved tracking.