Research On Data Driven High-speed Train Bearing Fault Diagnosis Method

Bearing faults during high-speed train operation are major safety hazards that can cause delays and paralysis of train lines,even cause casualties.Therefore,the detection and locating of bearing faults during train operation is essential for train safety and health maintenance.The diagnosis method based on the vibration signal model has the problems of difficult mechanism modeling and the need for additional sensors.The current bearing alarm system normally applies rule-based method that cannot detect the fault before the bearing is heavily damaged,which causes problems of false alarm.Therefore,the data-driven fault detection and locating method for train bearings has important theoretical research and practical application value.This article relies on the National Natural Science Foundation’s major project "Theories and Methods for Fault Modeling of High-Speed Train Information Control System Based on Big Data and Knowledge".Taking the fault diagnosis of high-speed train bearings as the background and taking the auto-correlation and cross-correlation of multi-bearings running in similar environments and speeds into account,a data-driven bearing fault detection and locating method which is based on dynamic latent structure of multi-bearings temperature is proposed.The main research work includes the following contents:1.Taking the strong correlation between the temperature and time of multi-bearings during train operation into account,a method for fault detection of train bearings based on dynamic inner principal component analysis is proposed.First of all,the correlation and dynamics of the multi-bearings temperature of the same carriage and the same bearing temperature of different carriages are analyzed;Then,a dynamic intrinsic principal component analysis based dynamic latent structure modeling method of multi-beaings is proposed.Using the established dynamic latent structure model,the statistical indexes for the dynamic and static sections are established respectively,and a fault detection method of train bearings based on dynamic inner principal component analysis is proposed;Application results using bearing temperature data collected from the practical operation of a train demonstrate the effectiveness of the proposed method.2.Taking the influence of variance information when extracting dynamic latent variables by dynamic inner principal component analysis into account,a dynamic system fault detection method based on dynamic inner canonical correlation analysis is proposed.First,considering a dynamic latent variable reflecting the characteristics of the main change information of train bearing temperature,a dynamic system modeling method based on dynamic inner canonical correlation analysis is proposed.Then,taking the four unnecessary detection index problems of the proposed bearing fault detection method into account,a bearing fault detection method based on the DiCCA combined indexes is proposed.The principal component detection index of the dynamic latent variable prediction error is combined with the residual detection index to establish a dynamic combined indexes.The principal component detection index of the static residual is combined with the residual detection index to establish a static combined indexes.A dynamic system fault detection method based on DiCCA combined indexes is proposed,and the TE simulation verification results demonstrate the effectiveness of the proposed method.Taking the strong correlation between the temperature and time of multi-bearings during train operation into account,a fault detection method of train bearings based on DiCCA was proposed.The proposed method was successfully applied to the fault detection of actual train bearings,which has the advantage of early-stage compared with the fault detection rules of train systems.3.Considering the problem of difficult bearing identification of multiple faulty bearings,a train bearing fault locating method based on DiCCA multi-directional reconstruction contribution was proposed.This method establishes a fault candidate set based on the reconstruction of the combined indexes of DiCCA.Based on the established fault candidate set,the contribution of each bearing variable in the candidate set to the fault is obtained,and the fault is located based on the multi-directional reconstruction contribution plot.Application results using bearing temperature data collected from the practical operation of a train demonstrate the effectiveness of the proposed method.