Research On Fault Diagnosis Of Permanent Magnet Synchronous Motor Based On Convolution Neural Network

Permanent magnet synchronous motor has the advantages of simple structure,small size high efficiency,high power factor and low vibration and noise.Therefore,permanent magnet synchronous motor is widely used in aerospace,transportation and other fields.In recent years,the state has vigorously developed the electric vehicle industry,and the permanent magnet synchronous motor is one of its preferred driving motors.Taking the electric vehicle as an example,due to the overload operation of permanent magnet synchronous motor,aging coil insulation,the motor short-circuit fault may be caused.The demagnetization fault of permanent magnet material will occur with the increase of temperature.These faults not only affect the operation efficiency and performance of the motor,but also easily affect the whole transmission system.Moreover,for electric vehicles,the motor operation reliability is particularly important.If the motor fault is not found in time,the motor stator current will greatly exceed the rated current,stop the motor and even burn the motor.This not only causes huge economic losses,but also may endanger the safety of human life.Therefore,the research on motor fault diagnosis method is of great significance.In view of the traditional fault analysis methods based on mathematical model and signal transformation,it is difficult to extract the fault characteristics of stator current signal of permanent magnet demagnetization,winding turn to turn short circuit and other faults of permanent magnet synchronous motor,and the complex calculation leads to the problems of low fault diagnosis accuracy and low efficiency,a fault diagnosis method of permanent magnet synchronous motor based on the combination of signal visualization and convolution neural network is proposed in this paper.The extracted stator current signal is transformed into an image by park transform.Using the advantages of convolutional neural network in image recognition,the classification of motor fault is obtained,and then the fault diagnosis of motor is realized.The main research contents of this thesis are as follows:Firstly,the thesis analyzes the causes of turn to turn short circuit fault and demagnetization fault of permanent magnet synchronous motor,and establishes the mathematical models of motor under normal operation,turn to turn short circuit fault and demagnetization fault;This thesis expounds the relevant principles of vector control,builds the inverter circuit and control circuit model of motor operation,obtains the stator current waveform under different operating states of motor through finite element simulation,and analyzes the current variation law of motor under different operating conditions.Then,the stator current data collected by finite element simulation is transformed into images by Park vector method,the relevant principle of current transformation into images is expounded,the images of the motor under different operating states are obtained,and the differences of image characteristics under different operating states of the motor are analyzed.Finally,the structure and basic principle of convolutional neural network are described,and the convolutional neural network model based on Google Net is established to recognize the images of motor in different operating states.The results show that this method has high recognition accuracy.The recognition results are compared with Alex Net network and other machine learning algorithms such as LS-SVM and PNN.It is proved that this method can be applied to the fault diagnosis of permanent magnet synchronous motor on a large scale.