Research On Fault Diagnosis Method Of Rolling Bearing Based On EEMD And CNN

Rolling bearings are an important component of many mechanical equipment,and establishing a complete bearing fault diagnosis system is the key to ensuring the normal operation of mechanical facilities.With the increase in the number and efficiency of data acquisition equipment,mechanical fault diagnosis ushered in the "big data" era.The feature extraction of fault signals and intelligent fault diagnosis algorithms designed by classifiers have become a hot research topic.First,this article preprocesses the data.The singular value difference spectrum method is used to reduce the noise of the data to reduce the high frequency components in the fault signal.Ensemble empirical mode decomposition(EEMD)is an effective fault signal analysis tool.Aiming at the problem that the intrinsic mode function(IMF)obtained by EEMD decomposition contains false components,this paper proposes a method based on the combination of energy proportion and correlation coefficient The method selects IMF to remove false components in the signal.The experimental results show that this method can obtain data characteristics better than the method of simply applying correlation coefficients.Then design the classifier model.In this paper,a convolutional neural network(CNN)model with multiple convolutional layers is designed for rolling bearing fault diagnosis,and the preprocessed data samples are used for model training.The model is diagnosed accurately on the Case Western Reserve University(CWRU)bearing database The rate is as high as 96.2%.At the same time,in order to verify the effect of data preprocessing,a comparative experiment has been added.Compared with the data samples without preprocessing,the data preprocessing can increase the accuracy of bearing fault diagnosis by about 2%.The application of overlapping sampling methods for data set enhancement has significant drawbacks,which can easily lead to overfitting problems in the training process of fault diagnosis models.This article focuses on the problems of CWRU bearing data samples being small and network model training difficult,and the generative confrontation network Based on(GAN),the generator and discriminator are designed to expand the one-dimensional sample data for the deep convolutional generative confrontation network(DCGAN)with four convolution layers.Experimental results prove that this method can overcome the defects of overlapping sampling and at the same time increase the accuracy of fault diagnosis to about 98%.Finally,in view of the translation invariance of the convolutional neural network and the loss of data characteristics in the pooling layer,following the frontiers of neural network research,a capsule neural network(Caps Net)model using dynamic routing algorithms is designed.Each capsule in the model is a vector,and each vector can contain multiple data features.Compared with the value obtained by the linear weighted summation of the convolutional neural network,the value of the capsule will be more fully utilized by the data.It is proved by experiments under the condition of using fewer data sets,the model can achieve a higher fault diagnosis accuracy rate.At the same time,quoting the model trained through the data set expanded by DCGAN can increase the accuracy of bearing fault diagnosis to about 99%.