Research On Diagnosis Of Rotor Eccentricity Fault For Interior Permanent Magnet Synchronous Motor

In recent years,permanent magnet synchronous motors(PMSM)have started to be used in high-speed railways,high speed maglev trains and other speed control applications,and their reliable operation is becoming more and more significant to the human's society.Therefore it is of great necessity to study the condition monitoring and fault diagnosis of permanent magnet synchronous motors.As one of the common motor failures,rotor eccentricities lead to unbalanced magnetic pulls,which accelerate the eccentricity degree and cause stator-to-rotor rub if without timely detection or proper maintenance.Such rub can damage the stator insulation and the rotor magnets,which in turn would worsen the problem.Therefore this dissertation contributes to the diagnosis of eccentricity fault of the interior PMSM in stationary and nonstationary operating conditions,and a type of interior PMSM for electric vehicle is employed.The whole research is on three aspects,which are the modelling of faulty machine,the model-based approaches and the sophisticated signal and data processing techniques.For the modelling of faulty machine,the modified winding function approach(MWFA)is used to model an eccentric 2p-pole interior PMSM with distributed winding structure.The developed model is validated by 2-D finite element analyses(FEA).Through the simulations and analyses of the stator inductance,the permanent magnet(PM)flux and the no-load back-electromotive force in static,dynamic and mixed eccentricity cases,it is confirmed that only the mixed eccentricity can generate eccentricity-related characteristics in motor current signature analysis(MCSA),and the distortion of stator inductance is the primary source.Besides,the overall developed system model is used to verify the feasibility of subsequent fault diagnostic approaches and predict their performance.For the stationary operating condition of interior PMSM,two online model-based frequency-domain and time-domain eccentricity fault diagnostic approaches are proposed in this dissertation successively.First,an analytical inductance calculation model based on a non-approximation MWFA is proposed for the faulty interior PMSM,aiming to provide more precise quantitative relations between the stator inductances and the rotor eccentricity degree.The proposed inductance calculation model takes full consideration of the winding distribution structure and it does not requires any series approximation of the turns function or the eccentricity function in achieving the task.Then by using this model,the frequency-domain quantitative relations between the discovered eccentricity signatures in the stator inductances in dq synchronous rotating reference frame and the rotor eccentricity degree and the initial angle of eccentricity are studied in depth.Then an inductance matrix expression under eccentricity fault in dq reference frame is newly developed.Based on this and inspired by position sensorless control techniques,an on-line frequency-domain eccentricity fault diagnostic approach is then proposed based on high-frequency d-axis inductance fluctuation.It solves the problem that all the injection based eccentricity fault diagnostic approaches in the literature can be used only when the motor is offline.The approach also has two significant advantages.One is,lower amplitude of the injected high-frequency sinusoidal pulsating voltage gives better diagnostic performance if the accuracy of current sensors satisfies.Hence the injection has little effect on the system operation.The second is,its fault index gives clear distinction of different eccentricity degree and does not confuse the eccentricity and the(partial)demagnetization.Both the eccentricity and the PM demagnetization can be simultaneously diagnosed with the help of PM flux linkage amplitude estimation.Although this method does not require the motor to be offline,it requires a fast Fourier transform(FFT)of the real-time estimated high-frequency d-axis inductance.At present,the FFT implementated in embedded digital signal processor(DSP)for motor control usually cannot meet the requirements of fault diagnosis because of the limitation of storage space and calculation rate,which affects the real-time execution.Hence the last FFT step has to be accomplished via a host computer or an additional dedicated chip.In order to realize a completely real-time eccentricity monitoring approach,the aforementioned inductance matrix expression is further transformed to ?? stationary reference frame.Then by exploiting the transient current vector trajectory,an inverse transient complex inductance vector theory as well as a special high-frequency voltage sequence injection method is proposed for the eccentricity diagnosis of an interior PMSM.A fully real-time on-line visualized monitoring approach in pure time domain is reached finally.The approach is free of FFT and requires only several steps of algebraic operations,which can be totally implemented in DSP and does not increase the burden of DSP.The absolute value of its fault index is positively correlated with the total eccentricity and the fault index itself is able to distinguish between the eccentricity and the demagnetization significantly.Since the time-domain vector loci will have distortion with the increase of motor speed,the approach is more suitable for the low speed operating condition.Considering that the model-based approaches can only be used for the stationary operating condition while the diagnosis for nonstationary condition generally requires the use of signal processing techniques,a fault diagnostic approach is proposed for the nonstationary operating condition of interior PMSM based on sophisticated signal and data processing techniques.Firstly,a class of high-resolution parameter estimation approaches are improved and extended from the frequency domain to the time-frequency domain.Then a time-frequency analysis approach named self-commissioning short time zoom matrix pencil is proposed to realize the diagnosis in the nonstationary operating condition.It can maintain high computational efficiency under any operating conditions and does not require any manual commissioning.Secondly,in the quasi-stationary operating condition where the speed and load fluctuate,high-resolution parameter estimation approaches have the problem with inconsistent estimation results.To solve this problem,a fault feature extraction approach is also proposed by introducing the Monte Carlo estimation and distribution fitting.This approach is used to perform further data processing on the large number of estimated samples generated by the aforementioned proposed time-frequency analysis approach.In this way,nearly unbiased frequency and amplitude estimates of the fault signatures can be obtained.The entire approach can be used not only to detect rotor eccentricity faults,but also to detect other rotor faults.Meanwhile,the approach also inherits the deficiencies of signal processing techniques.That is,the diagnosis process requires large calculation and needs to be implemented in a host computer.It can not distinguish between the eccentricity and the demagnetization,and hence it has to be used in combination with a model-based approach.The proposed three approaches have their own advantages and disadvantages,and are applicable to different operating conditions and demands.The correctness and effectiveness of all the theories and methods proposed have been fully verified by simulations and experiments.