State Recognition Of Railway Wagon Brake Shoes Based On Wheel Temperature Rise Data

With the rapid development of high-speed and heavy-duty railway wagons,the reliability of railway wagon components has attracted much attention.As a key component of the basic braking system,the state of the brake shoe has a great impact on the operation of the railway wagons.At present,the main method of judging the state of brake shoes is on-site inspection and repair,which is a serious waste of manpower and material resources.With the development of rail-side sensing equipments,monitoring data related to the wear state of brake shoes has been widely collected.Through the use of monitoring data in the online monitoring system,the connection between the brake shoe status and the condition monitoring data is established.It is very necessary to monitor and early warning of the state of the brake shoe.This thesis takes the brake shoes of railway wagon as the research object,and carries out research on the recognition of brake shoes based on the temperature rise of wheels.Firstly,the wear state of brake shoe is defined,and the maintenance information reflecting the wear state is extracted.In the wear state of brake shoe division stage,a ensemble clustering algorithm based on the weighted co-correlation matrix is used to divide the wear state of brake shoe,considering different characteristics and applicability of clustering algorithms.Clustering methods such as K-Means and hierarchical clustering are used to generate base clusters;information entropy method is used to measure the stability of base clusters;ensemble-driven clustering index is used as an index to evaluate cluster instability;the integrated driving clustering indexs are calculated and the hierarchical clustering algorithm is used obtain the final state division result of the brake shoe.Secondly,this thesis proposes to use the joint method of adaptive integrated oversampling method(ADASYN)and convolutional neural network(CNN)for the wear state of brake shoe recognition.ADASYN is used to deal with the unbalanced dataset.The processed data is used to train a convolutional neural network(CNN)to obtain the state recognition model of the brake shoe.Then test data set are used to test the brake shoe state recognition model to verify the validity of the model.After analysis and comparison with unbalanced data,the effectiveness of the ADASYN algorithm is verified.The balanced data is used to train classification algorithms such as Naive Bayes,Knearest neighbor algorithm,logistic regression,decision tree,and support vector machine.The classification results are compared with the ADASYN-CNN method,showing that ADASYN-CNN has a higher accuracy rate.Finally,the trained brake shoe state recognition model is migrated,and the data sets of different samples are used to realize the state recognition of the brake shoe state recognition model for different samples.The migrated model has a higher accuracy rate on different sample data sets.The results shows the feasibility of brake shoe state recognition based on the model migration method.This thesis verifies the feasibility of technical route of identifying the brake shoe wearing status based on indirect condition monitoring data of wheel temperature rise,and establishes a connection between the state of the brake shoe and the condition monitoring data to realize the monitoring and early warning of the brake shoe status.