The Research Of Intershaft Bearing Fault Diagnosis Based On Convolutional Neural Network

Intershaft bearing is one of critical components of aeroengine,and once a failure occurs,they will cause a significant threat to the normal operation of the engine.Therefore,it is of great significance to monitor the intershaft bearing in real time and make timely diagnosis.This article takes the intershaft bearing as the research object,and then the fault diagnosis method of the intershaft bearing based on the convolutional neural network is proposed.The research content of the paper is summarized as follows:(1)Expounds the research background and research significance of the subject,discusses the research situation of fault diagnosis of rolling bearing and aeroengine inters bearing at home and abroad,and discusses the application status of deep neural network in the field of bearing fault diagnosis.Describe the basic theories of three deep neural networks,and focus on the deep convolutional neural network model.Explore the training process of the model and introduce typical convolutional neural network models.(2)To deal with the problem of complex structure of typical convolutional neural network model and the difficulty of model parameter selection,a proportional nuclear convolutional neural network model is established.Using Adam optimization algorithm to improve network training speed,and the original time domain signal for typical failure of the intershaft bearing was used as input for verification.In order to solve the model over-fitting problem,add L2 regularization and dropout layer to the model.At the same time,discuss the influence of network structure hyperparameters on the model,and combine with the data of the intershaft bearing fault test bench,the proposed model has better recognition effect than the standard deep neural network and traditional diagnostic methods.(3)Aiming at a problem with excessive computational parameters for the proposed convolutional neural network model,Propose a diagnostic method for gray-scale transformation and optimization of proportional nuclear convolutional neural networks,Firstly,a three-layer proportional nuclear convolutional neural network is built.Secondly,the moving average model and the dropblock module are added to improve the robustness and generalization of the model.Finally,the pre-training sample size is increased by overlapping sampling,and the input data is preprocessed by using gray-scale transformation.The effects of linear gray-scale transformation method and nonlinear gray-scale transformation method indata preprocessing stage are studied respectively.Compared with other diagnostic methods,the results show that this method has faster diagnosis speed and higher diagnostic accuracy in the field of intershaft bearing fault identification.(4)Based on the LabVIEW platform to build an intershaft bearing condition monitoring and intelligent diagnosis system,and combine with the intershaft bearing fault test bench for verification.The system can not only realize online display of various signal indicators of the test bench,and complete signal acquisition and preservation,but also diagnose and analyze the state of the intershaft bearing through the deep neural network.