Research On Rolling Bearing Fault Diagnosis Based On CEEMDAN And Optimized Neural Network

With the rapid development of industrial production,mechanical equipment is constantly changing,and plays an significant role in economic development.The rolling bearing is one of the important parts of mechanical equipment.Because it often operates in a bad environment,rolling bearing is prone to failure,which affects the normal operation of the equipment.Avoiding the loss caused by equipment failure,it is important of significance to carry out predictive maintenance of rolling bearings.Therefore,this paper regards rolling bearings as the research target,studies the faults of rolling bearings in various states,extracts characteristic information of vibration signals collected,and then identifies faults through neural networks.First,the feature extraction method of vibration signal.Generally,the vibration signals of rolling bearings will be affected by noise when they are collected,so the feature extraction of the signals will be incomplete because of the influence of noise.In view of this problem,in this paper,a method based on wavelet threshold and adaptive noise complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)is used to extract sample entropy features.The method firstly used wavelet threshold denoising to denoise the signal of rolling bearing,and a new threshold function is introduced,which overcomes the problems existing in the two threshold functions.Then the empirical mode decomposition(EMD)algorithm and its related methods are introduced.The intrinsic mode functions(IMF)were obtained by decomposition of the signals using CEEMDAN algorithm.The feature extraction of rolling bearing is accomplished by selecting appropriate component through correlation coefficient and then calculating sample entropy of selected IMF component as feature vector.Secondly,the rolling bearing fault diagnosis and identification model.After feature extraction of rolling bearing signals was completed,GA-BP neural network was used to classify feature information and complete fault identification.GA-BP network fault diagnosis model was established by optimizing BP neural network with genetic algorithm to ameliorate slow convergence speed and sinking into local extremum.Two groups of control experiments were designed.Through the comparative experiments,it was found that GA-BP network model was better than the other two methods in fault identification of rolling bearings.Therefore,it is concluded that the diagnosis model based on sample entropy feature extraction and GA-BP network combined with wavelet threshold and CEEMDAN denoising can effectively diagnose the fault of rolling bearings.