Improvement of Low-speed Sensorless Control with Multi-Layer Neural Network

Abstract

In recent years, there has been an increasing demand for position sensorless control in PMSM drives, and vari ous methods have been studied. Switching noise is a problem in the low-speed sensorless control method that uses the current slope during PWM. Furthermore, another problem is that the inductance does not appear in a sinusoidal distribution owing to magnetic saturation. In this paper, we improve the sensorless control method that estimates the position from the current slope during PWM, which is greatly affected by switching. Additionally, we build a multi-layer neural network (NN) that directly estimates the position signals by learning a large amount of current data, and verify the driving results in the low-speed range when the learned NN is incorporated into real-time control.