Motor Fault Monitoring For Electric Vehicles Based On Convolutional Neural Network

Interior Permanent Magnet Synchronous Motors(IPMSMs)have been used in electric vehicle(EV)due to their higher output torque,better dynamic performance compared to the motors with electromagnetic excitation,simple construction,and easy maintenance.Under the working conditions of EVs,IPMSM is prone to demagnetization failure due to the narrow installation space of the motor which leads the poor heat dissipation conditions.Otherwise,there exist frequent start and emergency stop during the running EV,which induces the armature reaction easily and leads to demagnetization further.Moreover,the vibration occurring in running EV and violent collisions leading by the road condition or other environment element are easy to cause motor mechanical failures like air-gap eccentricity and bearing damage.Therefore,IPMSM used in electric vehicle is prone to occur demagnetization,eccentricity and other faults.However,current faults diagnosis algorithms in the literature are not suitable in the diagnosis for IPMSM used in EV.Based on this,this paper proposes a faults diagnosis method based on deep learning and image recognition,which can detect multiple IPMSMs’ faults with the same method.Moreover,the proposed method could be used under stationary conditions and nonstationary conditions and is immune to speed and load.In this paper,firstly,the existing diagnosis methods of demagnetization and eccentricity faults of PMSM are analyzed by comparing characteristics of various diagnosis methods;Secondly,various common motor modeling methods are compared.Then,the most accurate method,finite element method,is selected to establish the health model of IPMSM and fault models of different fault degrees which are analyzed by the co-simulation based on the finite element soft Flux and MATLAB-Simulink.;Thirdly,the structure,characteristics,derived networks and constructing methods of convolutional neural networks(CNNs)are introduced;Fourthly,the proposed method based on deep CNN and image recognition is elaborated,and the stator current of various of IPMSMs(including two degrees of eccentricity and two degrees of demagnetization faults)under various of operating conditions were divided into train dataset and test dataset;Finally,the train dataset was fed into designed CNN model to train the model.The test accuracy of 98.62% has been obtained.