Research On Deterioration State Identification Method And Its Application Of Urban Rail Train Rolling Bearing Based On Feature Selection

As a key component of urban rail train,rolling bearing is widely used in many train systems.It has a poor operating condition and easy to wear.Once it fails,the safe operation of the train will be seriously affected,even cause fatal accident.Therefore,it is of great significance to accurately identify the working state of rolling bearing to promote the development of efficient operation and maintenance of train and ensure the safety of train operation.This paper takes the urban rail train rolling bearing as the object and studies the identification method and application of the deterioration state of the bearing,including extracting the deterioration characteristics,selecting the deterioration characteristics,dividing the deterioration state,constructing a variety of deterioration state identification models and verifying the models with experimental data.The specific research contents are as follows:(1)Feature extraction of rolling bearing deterioration state.Firstly,the basic structure and deterioration process of rolling bearing are analyzed.On this basis,aiming at the problem that a single feature can not fully reflect the deterioration state of rolling bearing,a variety of statistical features and entropy features in time domain,frequency domain and time-frequency domain of rolling bearing life cycle vibration signal are extracted.And the representation of each feature parameter on the deterioration state of rolling bearing is analyzed.(2)Selecting the deterioration characteristics and dividing the deterioration state.In this paper,a comprehensive evaluation criterion of degradation characteristic performance is proposed,which includes three performance evaluation indexes and removes useless characteristic parameters.Based on this,the feature vector set is constructed,which can comprehensively and accurately characterize the bearing deterioration state.Then,the principal component analysis(PCA)is used to reduce the dimension of the deterioration feature vector set,and the first principal component is used to make a comparative analysis of a variety of clustering methods.Through the classification coefficient,the superiority of Gath Geva(GG)clustering algorithm is proved quantitatively.The GG clustering algorithm is used to cluster the first principal component of the unselected deterioration feature set,which further proves the necessity and feasibility of using the comprehensive performance evaluation index to optimize the deterioration feature.Finally,according to GG algorithm and optimal feature set,the bearing deterioration state is divided.(3)Building the identification model of rolling bearing.The deterioration state recognition model based on back propagation(BP)neural network and self-organizing feature map(SOM)neural network are constructed respectively.The performance of each model is verified by the degradation feature data set.Then,on the basis of integrating the advantages of the two models,a deterioration state recognition model based on SOM-BP series neural network is constructed.Under the same deterioration feature data set,the recognition effects of the three models are compared and analyzed,and the advantages of the series neural network state recognition model are proved.Finally,compared with the recognition results of support vector machine(SVM),it further proves the applicability of SOM-BP neural network model in rolling bearing degradation state recognition.(4)Research on the application of deterioration state identification method based on SOM-BP series neural network.Under the same condition and the same fault type,one rolling bearing deterioration feature data set is taken as the historical state data set,and the other two data sets are taken as the current state data set.After training the state recognition model based on SOM-BP series neural network with historical data,the current bearing state is recognized.According to the accuracy of the identification results,it is proved that under certain circumstances,on the basis of historical bearing deterioration data,the deterioration state identification model constructed in this paper can be used to identify the current bearing state.The usability of the model is verified.