Machine Learning for Inter-Turn Short-Circuit Fault Diagnosis in Permanent Magnet Synchronous Motors

Abstract

This work presents an approach to detect the inter-turn short-circuit (ITSC) faults in permanent magnet synchronous motors (PMSMs) using machine learning. PMSMs possess many attractive advantages such as high efficiency and high power density. However, with their widespread applications, preventive fault diagnosis to secure safe operations and reduce maintenance costs becomes increasingly critical. To achieve this objective, in this article, machine learning is employed for fault diagnosis, wherewith minor ITSC faults can be detected in a preventive manner. Therefore, the faulty motors can be replaced early to ensure system healthy operations. Moreover, machine learning is also utilized here to classify fault levels (e.g., minor, moderate, serious faults) for reference to appropriate maintenance actions. The proposed method adopts support vector machines (SVMs) and convolutional neural networks (CNNs) for training the diagnosis model with experimental data collected from tests in the laboratory. In the SVM, appropriate features required for training are selected through analysis with the PMSM mathematical model considering ITSC faults. This allows for more efficient training with much fewer data. In contrast, CNNs are purely a data-driven approach that requires much more data for training. The accuracy of both methods is validated with another set of measured data. It is found that both methods can achieve an accuracy of 99%; however, the SVM requires much less amount of data to achieve the same accuracy. This demonstrates the advantages of the model-aided machine learning method, which would be more suitable for applications where faulty motor data are not easily acquired.