Fault Diagnosis of Low-severity Demagnetization in Permanent Magnet Synchronous Motors Using Numerical Data

This paper presents a fault diagnosis method of low-severity demagnetization in permanent magnet synchronous motors (PMSMs) using numerical data. Simulations have been carried out on a motor that has dimension and physical parameters in our experimental setup by 2-D finite element analysis (FEA) to obtain the healthy and faulty current and torque signals under different speeds. Supervised machined learning methods are used to classify single and mixed demagnetization faults based on this numerical data to overcome the difficulty of implementing such faults in the lab and to tackle the problem of the availability of the historical data under faulty conditions. Four classification algorithms, support vector machine (SVM), K-nearest neighbours (KNN), ensemble and neural network (NN), with 22 classifiers are used in this study to evaluate the performance and the adaptability of different classifiers for diagnosing these faults. It is found that the supervised machine learning (SML) algorithms have a higher accuracy when the training is based on features extracted from the current signal than the torque signal, and NN classifiers provide near 100 % classification accuracy for the single and mixed demagnetization faults. Therefore, it is recommended to be used for the classification of these faults based on the current signal.