Mapping of a salient-pole PMSM and data-driven current vector selection - Development of an automated mapping system for PMSMs and evaluation of data-driven current selection

Abstract

Efficient operation of Permanent Magnet Synchronous Machines (PMSMs) requires accurate knowledge of the machine’s behaviour. Conventional control strategies often rely on linear machine models with constant parameters, which do not capture non-linear effects such as magnetic saturation, cross-saturation, speed-dependent losses, and temperature-dependent parameter variations. This thesis presents the development of an automated experimental mapping system for a salient-pole PMSM and investigates how measured machine data can be used for data-driven current vector selection. The automated mapping system was developed to gather data about machine behaviour through systematic measurements performed in a machine test bench. Measurements of currents, voltages and produced torque were collected at different speeds and temperatures. The data was used to investigate how the true machine behaviour deviates from the linear model. This was done using visualizations of torque surfaces, flux-linkage and efficiency maps and voltage ellipses compared for different speed and temperature conditions. The largest deviations from the linear model could be explained by magnetic saturation and cross saturation at large current magnitudes, noticeable both in the torque and voltage behaviour. Within the investigated operating region, the calculated inductances based on the measurements showed that the values of Ld and Lq could vary up to 50% and 60% respectively compared to the linear model, while the cross inductances Ldq and Lqd had absolute values up to 0.04mH and 0.03mH respectively. The mapped machine behaviour was used to develop a model for simulation in Simulink using Look-up Tables (LUT) for flux linkage and torque. Steady-state validation of the model yielded high accuracy for the case used in table generation 2000RPM and 50 ◦C, achieving mean current errors below 0.5A and a mean torque error of 0.087Nm, largely attributed to the introduced zero-torque compensation. The simulation model was used to evaluate a control method that utilizes LUTs to provide current references based on the given torque demand, as well as voltage and current limitations. While having some numerical instabilities, the developed current vector selector proved to be a valid approach, handling several limitation cases effectively while the voltage utilization is held accurately at 95% as limited by the control implementation. Furthermore, using mapped data to implement a Maximum Torque Per Watt trajectory yielded a mean machine efficiency of 65.26%, corresponding to a 0.21% increase compared to a mapped MTPA strategy.