Study On The Diagnosis And Early Warning Of High-speed EMU Axle Box Bearing Abnormal Temperature Rise

Axle box bearing is an important component to ensure the safe operation of high-speed EMU.It is a vulnerable part,because it works in tough conditions,suffers a large load,and runs fast.Once it breaks down,which will greatly threaten the safety of passengers.Therefore,the research on the condition monitoring of axle box bearing of high-speed EMU has great practical value.The existing high-speed EMU axle box bearing condition monitoring methods are divided into the onboard warning system and the remote warning system.These two warning systems only use the threshold value of temperature signal at a certain time to determine whether the bearing is faulty.There are not only a large number of false warnings to increase the workload of manual inspection,but also the untimely warnings cause train late which brings economic loss,even causes a serious adverse social impact.In order to improve the accuracy of high-speed EMU axle box bearing condition monitoring and realize axle box bearing fault early warning,this thesis carried out the following research work:(1)To improve the diagnostic accuracy of abnormal temperature rise of axle box bearing,a multi-dimensional isolation forest abnormal temperature rise diagnosis model for axle box bearing is proposed.Modeling from the spatial dimension based on the isolation forest algorithm,a multidimensional isolation forest model based on ensemble learning is proposed,which uses the same bogie,the same coach,and the same side axle box bearing data to build the isolation forest model respectively.The output of the three models adopts the ensemble learning strategy to realize the abnormal temperature rise diagnosis of axle box bearing.Experiments show that the model can effectively improve the accuracy of abnormal temperature rise diagnosis and reduce the false warnings.(2)To accurately identify the axle box bearing fault and bearing temperature sensor fault,an abnormal temperature rise type identification model of axle box bearing based on sequential temperature signals is proposed.In the proposed model,the historical time series temperature signal of the axle box bearing is used as the model input to train the model,and the continuous change characteristics of the bearing temperature in time dimension are considered.Firstly,the mean value of the temperature data on the same side is calculated as the temperature of each bearing on the same side to eliminate the influence of the individual difference of the bearing.Then,the mean value of the temperature input into the Multilayer Long Short-Term Memory network.Finally,the output of the network is added with a slip window to eliminate the influence of abnormal disturbance on the bearing temperature sensor.The experiment shows that the proposed model can effectively distinguish the bearing fault and the sensor fault to reduce the manual inspection work.(3)To realize the fault early warning of axle box bearing,a fault early warning model of axle box bearing based on multi-source data fusion is proposed.The model uses the spatiotemporal dimension data related to the bearing temperature as the input of the model to predict the changing trend of axle box bearing temperature in the next 6 minutes.Firstly,the Multilayer Long Short-Term Memory network is used to predict the changing trend of the axle box bearing temperature,and then the deviation index is calculated.Finally,the deviation index input into isolation forest algorithm to realize unsupervised classification.The experimental results show that the error between the predicted temperature and the actual temperature is small,which meets the requirements of the engineering application.Compared with the experiment results,the model can give early warning effectively,to reserve more time for emergency treatment in case of bearing failure.The model is an unsupervised deep learning model,that is,it does not need to label the sample data manually,which can greatly reduce the impact of subjective factors on sample categories.More importantly,the model can also effectively reduce miss warnings and false warnings.