A Novel Method for Automatically and Accurately Diagnosing Demagnetization Fault in Direct-Drive PMSMs Using Three PNNs

Abstract

Direct-drive permanent magnet synchronous machines (DDPMSMs) have recently become an ideal candidate for applications such as military, robotics, electric vehicles, etc. These machines eliminate the need for a transmission mechanism and excitation coil circuits, which enhances the system’s overall efficiency and decreases the likelihood of failures. However, it may incur demagnetization faults. Due to the characteristic of having a large number of pole pairs, this type of machine exhibits numerous demagnetization fault modes, which poses a challenge in locating demagnetization faults. This paper proposed a probabilistic neural network (PNN)-based diagnostic system to detect and locate demagnetization faults in DDPMSMs, using information obtained through three toroidal-yoke-type search coils arranged at the bottom of the stator slot. A rotor partition method is proposed to solve the problem of demagnetization fault location difficulty caused by various fault modes. Demagnetization fault location is achieved by sequentially diagnosing the condition of each partition of permanent magnets. Three demagnetization fault identified signals (DFISs) are constructed by the voltage of the three toroidal-yoke coils, which are used as inputs of PNNs. Three PNNs have been designed to map the extracted features and their corresponding types of demagnetization faults. The database for training and testing the PNNs is generated from a DDPMSM with different demagnetization conditions and different operating conditions, which are established through an experimentally validated mathematical model, an FEM model, and experiments. The simulation and experimental test results showed that the accuracy in diagnosing the location of the demagnetization fault is 99.2% when the demagnetization severity is 10%, which demonstrates the effectiveness of the proposed three PNN-based diagnostic approach.