Early Warning Method Of Abnormal Bearing Temperature Of High-speed Train Based On Structural Correlation And Complementarity

Bearing is a key component related to high-speed train operation safety.Bearing temperature is an important indicator that can reflect the bearing status of the train.The train monitors the bearing status through the temperature sensor and gives an alarm in time according to the status.In order to ensure the safety of the train during driving,once the axle temperature alarm occurs,the train needs to slow down or stop.If the alarm is axle temperature false alarm,it will seriously affect the train operation and bring unnecessary economic losses;If the alarm is not timely,the accident may occur.Therefore,it is of great significance to study the early warning technology for abnormal bearing temperature of high-speed train,improve the early warning amount and reduce the false alarm rate.However,there are many factors that affect the bearing temperature of high-speed trains,such as different manufacturing properties of bearings,ambient temperature,operating road conditions,operating speed,lubrication degree,etc.,which will affect the bearing temperature of high-speed trains,which brings great challenges to the abnormal detection and early warning of the axle temperature of high-speed trains.The main research work of this paper is as follows:(1)Aiming at the problem that the prediction model is easy to have fault following in the axle temperature prediction task,a new axle temperature anomaly early warning method for high-speed trains based on Long-Short Term Memory-Trend-Centralized Automatic Encoder is proposed.The correlation between the bearing temperature history data of high-speed train is obtained through the data correlation analysis method.The high correlation channel temperature data is regarded as the same temperature data,that is,the bearings with the same position and structure are the same bearings.The characteristics are processed based on the idea of comparison of similar bearings,that is,it is proposed to add a trend concentration layer to the multi input prediction network to improve the discrimination of the predicted axle temperature under normal and abnormal conditions,Then iterate the training network through a large number of resume data,and formulate the threshold according to the results after the network training.The example verification results show that the model can accurately distinguish the abnormal shaft temperature and has good early warning ability,but there is also a certain false alarm rate on some normal bearings.(2)Aiming at the problem that the bearing temperature prediction model lacks consideration of the influence relationship between bearing temperatures,according to the idea of transmission chain correlation prediction,a high-speed train axle temperature anomaly early warning method based on Gated Recurrent Unit-Gragh Attention Network is proposed.By analyzing the bearing temperature rise mechanism of high-speed train,the relationship between the bearing’s correlation in the transmission chain structure and its temperature is established.The relationship between the axle temperature channels is described with the help of graph neural network,which is used as the input of the prediction network together with the temperature data to realize the axle temperature prediction based on the mechanism and data drive.Then,the network is iteratively trained through a large number of resume data,and the threshold is established according to the results after the network training.The example verification results show that the model can accurately distinguish the abnormal shaft temperature,but while further improving the early warning ability,the false alarm rate on some normal bearings has also been improved to a certain extent.(3)In view of the different advantages of different structure association prediction in axle temperature early warning,a decision-making method for abnormal axle temperature of highspeed trains based on complementary evidence fusion rules is proposed.Firstly,the model threshold is improved according to the nonparametric estimation method,and the abnormal threshold of the above two models is used as the evidence of fusion decision.The complementary evidence combination rule is used for fusion,giving full play to the respective detection advantages of the two models,and making the final judgment on the fusion result through decision-making,so as to realize the abnormal discrimination and early warning of bearing temperature of high-speed train.The example verification results show that this method can accurately distinguish the axle temperature anomaly,compromise the early warning ability of the fused model,but greatly reduce the false alarm rate.(4)A set of early warning system for abnormal bearing temperature of high-speed train is developed.Based on the decision-making method of abnormal axle temperature of highspeed train based on complementary evidence fusion decision-making,analyze the relevant functional requirements and complete the system development.The system has the functions of off-line training,real-time axle temperature early warning,axle temperature data visualization,alarm log storage and so on,so as to provide reference for abnormal early warning of bearing temperature of high-speed train.