Diagnosis And Identification Of Muitiple Rotor Faults In Pmsms Based On Branch Current Analysis And Machine Learning

Permanent magnet synchronous motor(PMSM)is widely used in sustainable energy wind power generation,new energy vehicles and railway transportation due to its high power,high torque,high efficiency,reliability and good dynamic performance.The rotor is responsible for the input or output of the mechanical energy in PMSM,Therefore,it is an important part of the PMSM.The health condition of the rotor directly affects the operating performance of the motor and plays a crucial role in the motor’s performance evaluation.However,PMSMs are prone to rotor eccentricity,permanent magnet demagnetization and mixed faults in operation.Unfortunately,the fault characteristics of each type of rotor faults have similarity,and there are problems such as difficult diagnosis and identification.Therefore,it is a high demanding to develop a sensitive and accurate technologies of online PMSM rotor fault diagnoses and investigate the working condition of PMSM.The specific research methods are as follows:Firstly,this paper introduces the structure of PMSM and the mechanism of each type of rotor fault,and then compares the signal changes of back electromotive force(Back-EMF),phase current,and branch current after each type of fault.In this paper,it is found that due to the symmetry of the PMSM structure with a slot-pole ratio of 3/2 and its integer multiples,the time-domain changes in the phase currents after a fault cancel each other out.However,the fault characteristics are retained in the branch currents as well as the advantage that the branch currents enable online fault diagnosis.Therefore,the time-domain variation characteristics of branch currents are chosen as the signal for fault diagnosis and identification in this paper.Secondly,the models for the PMSM’s health and various rotor fault kinds are then established and solved using finite element software,and the motor branch current signals for each model are gathered.Based on the existence of particular harmonic components in the frequency domain of the branch current signal,a motor’s rotor failure can be identified.On the obtained branch current signals,pre-processing activities like residualizing and normalizing are carried out.Following that,the signal’s time domain feature factors are extracted,and feature vectors are built.Next,a GA-SVM model is constructed by combining Genetic Algorithm(GA)to obtain the optimal values of parameters in Support Vector Machine(SVM).The GA-SVM model is adopted to classify and identify rotor faults by combining the feature vector samples with the GA-SVM model.Combining the feature vector samples with the GA-SVM model for multifault classification and identification of rotor faults.The classification results are compared with those of commonly used machine learning classification models,such as K-nearest neighbor algorithm(KNN)and linear discriminant analysis(LDA),to prove the correctness and superiority of the method.Implementing PMSM rotor multi-fault diagnosis and identification.Finally,the demagnetization fault for the motor is identified,and the demagnetization fault location issue is studied.A demagnetization fault sample database is built and the fault thresholds are determined.The number of demagnetized poles of the PMSM demagnetization fault and the fault localization are efficiently determined by analyzing the Pearson correlation coefficients of the branch currents and the fault sample database during real-time operation of the motor and comparing them with the fault threshold.The issue of the degree of eccentricity fault is studied for PMSM after determining the occurrence of eccentricity faults.A method for estimating the degree of eccentricity is studied using the relationship between the amplitude variation of specific harmonics in the branch current and the degree of eccentricity fault.