Research on fault diagnosis in the operation monitoring of permanent magnet synchronous motors through deep learning

Background Permanent magnet synchronous motor (PMSM) may develop faults during long-term operation, affecting the stability and safety of the drive system. Objective This paper aims to identify the types of PMSM operation faults using a deep learning algorithm. Methods The convolutional neural network (CNN)-gated recurrent unit (GRU) algorithm was compared with the support vector machine (SVM), random forest (RF), and back-propagation neural network (BPNN) algorithms. Ablation experiments were conducted. Finally, the Shapley additive explanations algorithm was used to calculate the importance of feature indicators. Results The CNN-GRU algorithm had better fault-diagnosis performance compared with the other three algorithms and was easier to make an accurate diagnosis of inter-turn short-circuit faults in stator windings. The precision, recall rate, and F-score of the CNN-GRU algorithm were 0.950, 0.948, and 0.949, respectively; the corresponding values of the BPNN algorithm were 0.823, 0.819, and 0.821, respectively; the corresponding values of the RF algorithm were 0.719, 0.713, and 0.716, respectively; the corresponding values of the SVM algorithm were 0.707, 0.700, and 0.703, respectively. Ablation experiments verified the effectiveness of the CNN and GRU algorithms for the entire algorithm. Stator current and voltage were of the highest importance in the fault diagnosis model, followed by motor torque, and motor temperature was least important. Contribution The contribution of this paper lies in improving the recognition performance of fault types by combining two intelligent algorithms, CNN and GRU, and taking into account both local features and time-series features. It provides an effective reference for ensuring the stable operation of motor drive systems.