PWM and Model Predictive MRAS-based Estimators for Sensorless Control of PMSM

A novel sensorless rotor position and speed estimator is presented in this thesis. The method is based on measuring the current response to conventional space-vector pulse width modulation (SV-PWM) for sensorless PMSM drive applications. Model-Reference Adaptive Systems (MRAS) estimators are frequently used for sensorless angle estimates. They are typically based on comparing two stator flux estimations based on current and voltage models. The voltage model uses an integration of stator voltages to calculate the stator flux. The integration process, however, results in phase delays. The proposed method uses oversampling and averaging over a switching SV-PWM cycle, eliminating the need for integrators. Extensive experiments are provided to assess the effectiveness of the proposed estimator. The results of the experiments demonstrate good performance at various speeds and under various load circumstances, in both motoring and regenerating mode. The proposed method also shows robustness to changes in motor parameters.
This thesis also introduces an innovative predictive model reference adaptive system position and speed estimators for Permanent Magnet Synchronous Machines. The speed estimator is based on the finite control set model predictive control (FCS-MPC) principle. A search method is utilized to find the best-estimated speed at each sampling interval that produces the smallest speed-tuning error signal. Consequently, the need for a PI controller in MRAS estimators is eliminated. The problem with a fixed PI controller is that it is unable to maintain optimal performance under all different operation conditions. Therefore, an adaptive PI controller is required for satisfactory performance, however, this is a difficult process that takes time and effort. In contrast to MRAS methods based on PI controllers, the suggested approach eliminates the need for gain tuning, simplifying the process and ensuring optimal performance. The experimental testing of the proposed estimator demonstrates enhanced performance compared to the PI-based MRAS method.