Research On Fault Diagnosis Of High-speed Train Axle Box Bearings Based On Dynamics-data Driven Model

The axle box bearing is one of the important rotating parts in the travel section of high-speed trains,which plays the role of load transfer and motion conversion between axle and frame,and its operation status directly affects the safety,smoothness and reliability of train operation.During the driving process of high-speed trains,the complex excitation between wheels and rails has a significant impact on the motion change and load distribution of axle box bearings.The establishment of a scientific axle box bearing dynamics model and the analysis of bearing dynamics behavior and fault mechanism research under actual operating conditions are essential for accurate diagnosis of axle box bearing faults.Therefore,this thesis constructs a coupled vehicle-axle box bearing dynamics model containing typical faults,and carries out in-depth research on axle box bearing fault diagnosis of high-speed trains using data-driven models based on the simulation analysis of the dynamics model,with the main research contents as follows:(1)Using the multi-body dynamics simulation software UM to establish a high-speed train vehicle-track dynamics model,the model takes into account wheel-rail contact force,non-linear suspension parameters,track model,track unevenness excitation and other factors,and verifies the validity of the model through safety,smoothness and stability indicators.The measured Jingjin track spectrum is used as the input of track excitation to obtain the load impact on the axle box.On the linear track with uneven excitation,the box bearing is subjected to vertical,longitudinal and transverse dynamic loads,and the loads on the left and right box bearinges are basically the same for the same round.(2)The axle box bearing dynamics model is established by Solid Works software and ADAMS software,and the constraint relationship,contact force,drive and load of each element inside the axle box bearing are considered.According to the axle box bearing fault types,the bearing dynamics models of normal and containing typical faults were established.Rolling vibration experiments of single wheel pair axle box bearings of high-speed trains were carried out to obtain vibration acceleration signals of axle box bearings,analyze the relationship between experimental and simulated signals,as well as compare the simulated and theoretical values of cage and roller speed respectively to verify the validity of the built bearing models.(3)Analyze the change law of speed,mass center trajectory,contact force and vibration signal of each element inside the normal and raceway stripping fault axlebox bearing under the condition with/without wheel-track excitation.The effect of wheel track excitation on normal axle box bearings is relatively mild,while the axle box bearing with inner ring fault has a greater impact,which can increase the contact force between the rollers and other components,reduce the rotational speed of the rollers and cage,intensify the mass center fluctuation of each component of the bearing,and make the vibration signal of the axle box bearing more complex.The influence of wheel track excitation on the dynamic behavior of both sides of the inner ring fault bearing shows obvious differences,and the dynamic behavior of the fault side is especially obvious,and presents the characteristics of "the fault side is larger than the non-fault side,and the non-fault side is larger than the normal bearing".(4)By analyzing the commonality and difference between the simulated and actual signals,a deep neural network and migration learning based box bearing bearing fault diagnosis method is proposed.A large amount of simulation data is used to train the network model,and migration learning is used to transfer the trained network parameters from the simulation data to the bearing diagnosis network model of the actual data,freeze part of the model parameters,and use a small amount of actual data to fine-tune the parameters of the fully connected layer,and then complete the fault diagnosis of the bearing.Through experimental comparison,it is shown that the method can learn the fault characteristics of different failure states from the simulation data and apply them to the bearing diagnosis analysis of the actual data.This thesis realizes the migration between the simulation data and the experimental bench data,and provides a new feasible way for the problem of insufficient training samples in box bearing bearing fault diagnosis.