Research On Fault Diagnosis Method Of Rolling Bearing Based On Full Vector DCNN

Rolling bearing is a key part in mechanical equipment,and it is also the most vulnerable part of the equipment to fault.With the rapid development of digital factory,the method of data acquisition has become more and more abundant.The traditional fault diagnosis methods use manual feature extraction and feature selection,which has high work complexity and low stability of the model.In this paper,rolling bearing is taken as the research object,and the deep convolutional neural network is used to carry out the adaptive extraction of rolling bearing fault features.Aiming at the fault judgment caused by information omission in fault diagnosis of single channel signal,the full vector spectrum information technology is used.Usually,the establishment of fault diagnosis model using convolution neural network,which structure parameters are usually optimized by experience and reference,genetic algorithm is introduced to optimize the model structure Aiming at the above problems,this paper presents a fault diagnosis model of rolling bearing based on the full vector convolutional neural network and genetic algorithm.The main research work of this paper is summarized as follows:(1)A fault diagnosis method for rolling bearing based on full vector spectrum technology and deep convolution neural network(full-vector DCNN)was established and verified by experiments.First full vector spectrum technology was used to carry on the homologous dual channel information fusion,a relatively complete sample data to obtain information and then convolution neural network model is established,using technologies such as dropout for inhibition network fitting,Adam optimization algorithm was used to adjust the weights and bias parameters for the optimization of model.Then,completing the training of the network model.The model test is performed on the test data set.Compared with the single-channel DCNN model and the full-vector DNN model,the experiment results show that the full-vector DCNN model has higher fault recognition accuracy than the single-channel DCNN model and the full-vector DNN model.After adding the same noise signal,the DCNN model,which has the convolution structure,has a higher accuracy than the fully connected DNN model,and the accuracy of the model is further improved when combined with the full vector spectrum technology.(2)To solve the problem of structural parameters of the network model designed according to experience,genetic algorithm is adopted to optimize the convolution structure of the full-vector DCNN model,and a bearing fault diagnosis model based on GA optimization full-vector DCNN is proposed.Firstly,comparing the network models with different convolution structures,the object of parameters to be optimized is determined.Then,the genetic algorithm was designed according to the characteristics and relevance of parameters,and the optimal structural parameters of the model were determined through iterative training.The experimental results showed that the optimized full-vector DCNN model was more accurate in fault diagnosis of rolling bearings,and the distribution boundary of feature data was clearer.