Rotor position estimator based on machine learning

Abstract

As considered in this study, machine learning is a field of artificial intelligence which provides computation algorithms without explicit programming. The calculation rules are learned from a set of data that associates the inputs to the outputs. Machine learning based estimator is developed to predict the rotor position of an electrical machine. The estimator requires only the phase currents to be measured. An off-line training makes the model learn to associate rotor position to each instantaneous current measurement. The estimation precision depends on the quality of current measurements and on the effectiveness of the training. Compared to rotor position acquired using mechanical sensor, the estimation error doesn't exceed few percents which is enough to implement a sensorless control or to make safe redundancy of the mechanical sensor. The estimator is developed for a permanent magnets synchronous machine used on the power train of an electric vehicle. Potentially, machine learning based estimator can cover all the speed/torque range of the machine.