Machine Learning-Based Optimization of High-Frequency Injection Amplitude in Sensorless PMSM Drives

Abstract

The methods known as high-frequency injection (HFI) are widely used in sensorless permanent magnet synchronous motor (PMSM) drives to provide adequate rotor position measurements in the low-speed and zero-speed ranges where back-EMF methods fail to work. Although effective, the choice of amplitude of injection is also a critical issue. A too low amplitude compromises the accuracy of position estimation and noise immunity and a too high amplitude increases the ripple on the torque, acoustic noise, and copper losses. Further, ideal amplitude changes dynamically with operating condition like load torque, rotational speed, variation of DC-link voltage, and machine parameter change and hence the approach of fixed or look-up table is not the best. This work gives a new system of optimization based on machine learning, which optimally identifies the amplitude of the HFI based on real-time operating state values. The suggested learning agent reduces structured multi-objective cost criterion which is the combined evaluation of rotor position estimation error, torque ripple magnitude, and incremental copper losses. The framework provides close-to-optimal operating space amplitude settings based on different conditions through data-driven exploration of the nonlinear operating space. The simulation findings identify stronger resistance to voltage drop across inverter fed and measurement noise, and lower steady-state torque oscillations relative to traditional techniques, confirming that simulation has been shown to be suitable in low-voltage electric mobility applications.