Deep Learning-Based Approach for Speed Estimation of a PMa-SynRM

Abstract

Synchronous motors require information about absolute rotor position to ensure full control. Different types of sensors connected directly to shaft are preferred for measuring rotor position. These sensors have some disadvantages such as more hardware complexity, high cost, increased volume, cable addition, decreased noise immunity, decreased reliability, and increased maintenance requirement. The best and only way to figure out these disadvantages is to use any sensorless method. There are various position-sensorless control techniques that can be grouped under two main categories as model-based methods and saliency tracking-based methods. This paper presents an approach to determine the rotor position of synchronous motor without any position sensor by using machine learning regression algorithms. Performance analysis was performed for different speed transitions by using different parameters of long short-term memory (LSTM). The most common metrics root mean squared error (RMSE), mean absolute error (MAE), and R-Squared (R-2) were examined to measure prediction performances.