Research On Motor Rolling Bearing Fault Diagnosis Algorithm Based On VMD And Convolutional Neural Network

With the rise of the new energy field,the new energy electric vehicles develop rapidly.As the core component of pure electric vehicles,the performance of the motor is directly related to the power performance and energy conversion efficiency of electric vehicles.Therefore,the establishment of a reliable fault detection system is the key to ensure the smooth operation of the electric vehicle drive motor.With the advent of the era of "big data",the method of manually extracting fault features and then classifying them in the past is inefficient and has been unable to meet the current requirements of intelligent diagnosis.In order to solve this problem,this paper takes the rolling bearing of the most important component in the drive motor as the research object,and the variational mode decomposition algorithm(VMD)and adaptive convolution neural network(ADCNN)are applied to the fault diagnosis to process and identify the vibration signal generated by the bearing.This paper first introduces the bearing on the drive motor,expounds the basic structure and failure form of the bearing,and the specific differences between the drive motor bearing and other mechanical equipment bearings,according to the characteristics of the rolling bearing,puts forward the fault diagnosis idea.First,chaos initialization and Gaussian variation are introduced into the Sparrow search algorithm(SSA)for improvement.The improved Sparrow search algorithm(ISSA)is used to optimize the parameters of VMD,and the advantages and disadvantages of the optimized VMD are compared with the EMD algorithm.The optimized VMD algorithm is applied to the vibration signal processing of the motor rolling bearing,extracting the characteristic value,and providing data support for the subsequent fault diagnosis.Based on the traditional convolutional neural network,this paper proposes an adaptive convolutional neural network model(ADCNN)for bearing fault diagnosis.To solve the problem of difficult fault signal feature extraction,the first layer convolution kernel is designed as a large convolution kernel to extract short-time features,which are small except for the first layer,to reduce the computational amount.Adam algorithm and batch normalization algorithm are used to optimize the network model to improve the recognition rate and speed of the model.The hyperparameter design criterion of ADCNN model is given,which reduces the design difficulty of fault diagnosis algorithm.Experiments show that ADCNN can achieve 90%recognition rate of data set.Under complex working conditions,the identification rate of the system for bearing fault diagnosis will decrease.The ADCNN model designed in this paper will be tested in different environments to obtain specific test results,and the test results of the model will be compared with the results of other fault diagnosis models,so as to judge whether the ADCNN model can ensure good diagnostic performance in different working environments.Later,the visualization technology(T-SEN)was used to display the classification process of the model,which reduced the analysis difficulty of the neural network model.The experimental platform and experimental process of collecting the bearing vibration signal of the drive motor are introduced,and the VMD algorithm and ADCNN model are applied to the actual data set.It ensures that the convolutional neural network module has a high recognition rate and diagnosis speed in the actual fault diagnosis,which verifies the reliability of ADCNN model.