Study On Fault Diagnosis Method Of Permanent Magnet Motor Based On Improved Convolutional Neural Network

As the main power equipment of production and living equipment,the health of motor operation has a crucial impact on the normal operation of production equipment.It can reduce the damage of the motor and the economic loss of the enterprise to find the fault point and determine the cause of the fault in time to ensure the normal operation of the motor.Therefore,in this paper,the problem of motor fault diagnosis is studied deeply.In view of the problems of less data of turn to turn short circuit of permanent magnet motor,unstable training of generative counter neural network and poor data expansion effect,etc.In this paper,a data expansion method based on deep convolution generation is proposed to deal with the problems of poor timeliness and low accuracy of single feature in the motor fault diagnosis of big data,multiple fault types,multiple faults or concurrent faults,as well as the difficulties in deep neural network training,gradient disappearing or exploding,and high error rate Fault diagnosis method of permanent magnet motor based on network.In view of the problems of less inter turn short circuit data,unstable training of generative counter neural network,and poor data expansion effect,vibration signal and stator current signal are selected as fault diagnosis features of permanent magnet motor.Because deep learning needs a lot of data support,in order to make up for the shortage of turn to turn short circuit fault data of permanent magnet motor,as well as the problems of poor network stability and low quality of "forged data" in the sample expansion of generative neural network.This paper is based on the neural network sample data expansion method of the depth convolution generator.In this method,convolution layer and deconvolution layer are used to replace the pool layer of the discriminator and generator,so as to improve the learning efficiency of "fake data" of the generator.The simulation of tensorflow framework in Python environment shows that compared with the generative neural network and the generative neural network,the method in this paper has better learning efficiency and shorter training time.For shallow neural network training big data,there are problems such as slow learning rate,over fitting and so on.For traditional deep neural network training big data,there are problems such as gradient disappearance and explosion,network training difficulty and so on.In this paper,we introduce cross connect module,and study the principle of each cross connect module.The simulation results show that the deep convolution neural network with cross link module has high accuracy.In this paper,the effects of different initial functions,learning rate and batch size on the accuracy of the model are studied,and the optimal super parameters of the deep convolution neural network with cross connected modules are determined.Simulation results show that the accuracy and diagnosis time of PMSM fault diagnosis with deep convolutional neural network are better than those of stack noisereduction self coding network,adaptive probability neural network and shallow convolutional neural network.