To solve the problem that it is difficult to accurately identify the rotor eccentric fault, demagnetization fault and hybrid fault of a permanent magnet synchronous motor (PMSM) with a slot pole ratio of 3/2 and several times of it, this paper proposes a method to identify the rotor fault based on the combination of branch current analysis and a machine learning algorithm. First, the analysis of the electrical signal of the PMSM after various types of rotor faults shows that there are differences in the time domain of the electrical signal of the PMSM after three types of rotor faults. Considering the symmetry of the structure of the PMSM with a slot-pole ratio of 3/2 and its integer multiples, the changes in the time domain of the phase currents cancel each other after the fault, and the time domain fluctuations of the stator branch currents that do not cancel each other are chosen as the characteristics of the fault classification in this paper. Secondly, after signal preprocessing, feature factors are extracted and several fault feature factors with large differences are selected to construct feature vectors. Finally, a genetic algorithm is used to optimize the parameters of a support vector machine (SVM), and the GA-SVM model is constructed as a classifier for multifault classification of permanent magnet synchronous motors to classify these three types of faults. The fault classification results show that the proposed method using branch current signals combined with GA-SVM can effectively diagnose faulty PMSM.

To address such challenges as an uncertain number of demagnetization poles of the permanent magnet synchronous motor (PMSM) and cases in which the fault cannot be located, this paper proposes a fault identification and location methodology based on the analysis of the motor stator current. First, the influence of the irreversible demagnetization of permanent magnets on the analytical model of the back electromotive force (Back-EMF) of the rotor in a single motor stator slot is analyzed. Moreover, considering the topology of the motor, the influence of the demagnetization fault on the stator phase current and branch current is analyzed. Since the stator phase currents cannot diagnose the partial demagnetization faults of PMSM with some topological structures, the stator branch current is selected as the signal for the identification and localization of the demagnetization fault. Secondly, the demagnetization fault diagnosis and mode recognition of the motor are carried out through the amplitude of the real-time branch current and the harmonic components of the PMSM. A sample database of demagnetization faults is established through calculation and normalization of the residual value of the stator branch current and the branch current of the healthy motor after demagnetization in one pole order. The fault threshold is obtained by analyzing the residual of the branch current of uniform demagnetization and the Pearson correlation coefficient of the fault sample database. Then, the correlation coefficient between the real-time branch current residual value of PMSM and the fault sample database is analyzed, and the number of demagnetization poles and the fault location are determined by the number and location of the calculated correlation coefficient exceeding the threshold. Finally, the feasibility and effectiveness of the proposed method are verified by the finite element analysis (FEA) results.