Data-driven Subway Wheel Wear Prediction And Application Research

With rapidly increasing of subway lines and operating mileage,the requirements for vehicle maintenance and safety are getting serious.Wheels are the key component of vehicles.When the accumulated wear causes the wheel size to exceed the lower limit of the effective value,the wheel needs to be repaired to ensure the safety of the vehicle.China has many subway lines,large passenger volume,intense maintenance work but short working hours.If the wheel wear is not predicted,and the wheelset is repaired or replaced only when there is a problem,it is easy to cause untimely vehicle scheduling,resulting in lack of vehicles on the line and causing economic losses.Therefore,it is necessary to predict the wear state of the wheel and grasp the changing trend and law of the wheel size parameters.In this way wheels with problem can be repaired or replaced in time to keep the wheel parameter within a safe range.This is of great significance for improving the safety and economy of subway vehicle operation.Taking Shanghai Metro Line 17 as the research object,a vehicle dynamics model was established to study the influence of different factors on wheel wear,and the wear development law was studied based on the measured data.The wear state of wheel diameter is predicted based on neural network prediction model and probability prediction model,and a wheel diameter software prediction system is established according to the probability prediction model.The main research contents and conclusions of this paper include:1.The influence of different factors on wheel wear was studied.The metro vehicle dynamics model is established in UM and its dynamic performance is verified.The wear simulation is carried out based on the vehicle dynamics model,and the influence of wheel diameter difference,vehicle load,wheel-rail friction coefficient and wheel polygonal excitation on wheel wear is studied.2.Based on the measured data,the wear state of the subway wheels of Line 17 is analyzed.First complete the data cleaning,analyze the tread diameter,rim thickness,rim height and rim gradient data.Then the development law of wheel wear on this line is summarized.3.A wheel diameter prediction model based on the EEMD-LSTM model is established.Aiming at the prediction defects of a single model,the EEMD algorithm is used to stabilize the unstable data,and the prediction is based on the EEMD-LSTM network.The accuracy of the EEMD-LSTM model is verified by comparing with the prediction results of the EEMD-RNN,EEMD-BP and EEMD-GRU models.4.Two probabilistic prediction models are proposed based on mathematical statistics methods.In view of the disadvantage that the point prediction model cannot quantitatively describe the uncertainty of wheel diameter changes,the probability prediction of wheel diameter is carried out,and the fitting effect of different distribution models on wheel diameter data is studied.Based on the kernel density estimation method,a prediction model under different confidence levels is established to predict the change rule and probability of the wheel diameter of the new wheel of the same vehicle on the same route.Combining EEMD-LSTM point prediction and probability density prediction,probabilistic prediction of future wheel diameter changes.5.A software system for wheel diameter prediction is established.Using the Lab VIEW platform,based on the analysis results of wear influencing factors and the probability prediction model,the application of wheel diameter prediction is studied.