Research On Rollin Bearing Fault Diagnosis Baswd On EWT And Convolutional Neural Network

Rolling bearings are one of the most frequently used mechanical components in mechanical equipment,and are widely used in transportation,metallurgy,aerospace and other fields.As a core component in mechanical equipment,the health status of rolling bearings will directly affect the safety,productivity and life of the entire mechanical equipment.Therefore,the research on rolling bearing fault diagnosis is of great significance to guarantee the normal operation of mechanical equipment.Aiming at the problem of rolling bearing fault diagnosis,this paper combines the time-frequency signal processing method of empirical wavelet transform with the feature extraction and classification method of convolutional neural network to realize the intelligent diagnosis and recognition of rolling bearings.The main research contents are as follows:First of all,in view of the problems of non-stationary,non-linear vibration signals of rolling bearings and the difficulty of extracting their characteristics,the related principles of empirical wavelet transform(EWT)are studied.On this basis,in order to prove the performance of the method in signal processing,a simulation signal is constructed for the problems of false components and modal aliasing in signal decomposition,and empirical mode decomposition(EMD)is introduced to compare it.At the same time,in order to show the effect of this method in rolling bearing fault diagnosis more intuitively,a fault diagnosis method based on the combination of EWT,singular value entropy and t-SNE is proposed,and it is related to EMD,singular value entropy and t-SNE Combined methods for comparison.The double comparison between the simulated signal and the actual diagnosis shows the advantages of the empirical wavelet transform method.Secondly,in order to solve the problem that traditional fault diagnosis cannot be effectively diagnosed with massive data and no fault mechanism,the related principles of Convolutional Neural Network(CNN)are studied.Based on this,a shallow convolutional neural network model is designed,and the SE module is embedded in the model to improve the convolutional neural network.Dropout and data enhancement are used to prevent the network from overfitting.The rolling bearing vibration signal is used as the input of the improved convolutional neural network and the traditional convolutional neural network,and the experimental results show that the proposed network model has a certain degree of accuracy and training speed improvement compared to the traditional network model.Finally,in view of the insufficient feature extraction capability of shallow convolutional neural networks,an intelligent recognition method based on empirical wavelet transform and deep convolutional neural networks is proposed.The empirical wavelet transform is used to decompose the bearing vibration signal,and the inherent modal components obtained by the decomposition are used to construct multi-channel samples and create a data set.Based on the receptive field idea,the proposed shallow network model is improved and deepened,and batch normalization is introduced to improve the network training speed and reduce the impact of the size and distribution of training data on the model.A deep convolutional neural network model is used to diagnose and identify the created data set,and at the same time analyze the impact of various parameters in the proposed method on the experimental results.The experimental results show that the method has good generalization performance while achieving high accuracy of fault identification.