Short Circuit Fault Detection in Permanent Magnet Synchronous Motor Based-on Group Model of Data Handling Deep Neural Network

Abstract

Short circuit fault (SCF) in stator coils is one of the most common types of electrical faults. The expansion of this fault leads to the permanent demagnetization of the magnet, and causes irreparable damage to the machine in a short period. With the development of artificial intelligence technologies and various machine learning and deep learning techniques, an increase in fault detection accuracy has been achieved. In this paper, permanent magnet synchronous motor (PMSM) is investigated under normal mode and fault conditions, namely SCF in winding loops, phase to phase SCF and open circuit fault of one of the phases. Group Model of Data Handling deep neural network (GMDH-DNN) is used to produce a SCF detection model. Results of simulating the proposed method and the data extracted from the PMSM reveals that the accuracy rate of SCF detection in the winding loops of the PMSM in the proposed method is equal to 99.2%, which constitutes an improvement of 1.7% compared to other existing methods such as conditional generative adversarial network (CGAN). Moreover, simulating other existing methods - namely support vector machine (SVM), k nearest neighbors (KNN), C4.5, multi-layer perceptron (MLP), recursive deep neural network (RDNN) and long short-term memory networks (LSTM) – and comparing them with the proposed method, unveil that the accuracy of the proposed method for SCF detection in winding loops overweigh those of aforesaid existing methods.