Position Estimation and Sensorless Control of PMSM Using Neural Network in Low-Speed Region

This paper introduces a neural network (NN)-based method to enhance the position estimation performance of a permanent magnet synchronous motor in the low-speed region. The goal is to expand the operable region into the extremely low-speed region when compared to conventional physical model-based methods. In contrast to prior NN-based methods, this paper presents three key contributions. First, the input features, consisting of input data composition and preprocessing methods, are examined based on the grid-search method to enhance the estimation performance. Next, sensitivity analysis for the developed NN model is performed according to the input variables. This enhances the explainability of the NN-based model, which has previously been considered a black-box model. Finally, the developed model is deployed on the digital signal processor (DSP) to demonstrate real-time behavior and shows that the proposed method reduces the uncontrollable region by 50.20%.