A Comprehensive Interturn Fault Severity Diagnosis Method for Permanent Magnet Synchronous Motors Based on Transformer Neural Networks

This article proposes a novel deep learning (DL)-based interturn short-circuit fault (ISCF) severity diagnosis method using the transformer neural network (TNN). The input features are the currents in alpha-beta reference frame while the outputs are the number of shorted turns and short-circuit (SC) current amplitude. By only monitoring the stator currents, this method provides a comprehensive overview of the fault severity. The proposed TNN creates multiple representations of the input using the multihead attention mechanism. This allows the network to focus on specific parts of the input signals and create an accurate estimate. The dataset was collected using a motor with rewound windings to simulate stator ISCF, which was operated under nine specific load and speed conditions and three numbers of shorted turns. Both the shorted turns and SC current amplitude estimations on the test dataset have higher than 96% accuracy. A comparison of several methods is presented based on various criteria. Based on the comparison and considering the achieved accuracy, the proposed method shows high potential in terms of comprehensiveness, accuracy, practicability, and cost.