Hybrid Physics-Based Data-Driven Methods For Wheel Health Management Of Emus

Wheel–rail dynamics is a central issue in rail transport systems.Wheels are core components of electric multiple units(EMUs)and also the focus of EMU operation and maintenance.In China,there are a huge number of EMU wheels.Due to the high operational speeds and complex operational conditions,these wheels require frequent inspection and maintenance,resulting in huge life cycle costs.The condition of wheels has a direct impact on the safety,reliability,and operational quality of EMUs.For the huge number of in-service EMUs in China,wheel health management is necessary for the operation and maintenance of EMUs.It is found from a literature review that pure physics-driven or data-driven approaches cannot meet the needs of wheel health management and maintenance decision-making.In this context,this thesis combines the physics of vehicle dynamics with big data approaches to build up a wheel health management framework for the key degradation processes,including condition monitoring,degradation prediction,and maintenance decision-making.The major content and conclusions of this thesis are summarized below.(1)Wheelset hunting monitoring and stability assessment.This thesis investigates the characteristics of wheelset hunting motions and stability.It is found that,due to the complex influence of many factors,wheelset hunting is time-varying with significant epistemic uncertainties in its frequency and amplitude.A phase trajectory periodicity index is proposed to quantify the periodicity of a time series from a nonlinear system,based on which a strategy for hunting monitoring and vehicle stability assessment is further proposed.Case studies from simulations and field measurements demonstrate the effectiveness of the proposed method in detecting wheelset hunting motions,including slight hunting motions.Moreover,the proposed method is effective in assessing vehicle stability in spatial and temporal domains,i.e.,identifying faulty or degraded wheelsets or bogies and characterizing the trend of stability degradation.(2)Wheel size degradation modeling and life prediction.This thesis studies the laws governing the influence of wheel size on tread wear and flange wear through theoretical analysis,simulations,and statistics.It shows that,for the selected EMU type,a smaller wheel diameter leads to faster tread wear,while a smaller wheel diameter or flange thickness leads to slower flange wear.Based on these findings,a wear model is established considering uncertainties.Then,this thesis investigates the relationship between the cutting depth in wheel diameter and the variation in flange thickness during wheel reprofiling and develops a physicsbased data-driven model for wheel reprofiling.Further,a Monte Carlo-based method is proposed to predict wheel size degradation,based on which the reliability of wheel life can be assessed.The accuracy and effectiveness of the proposed method are verified by using reallife wheel degradation data.(3)Risk analysis and prediction of wheel polygonization.This thesis studies the effect of polygonal wheels on vehicle dynamics through vehicle dynamics simulations and vehicle vibration monitoring.Factors correlated with wheel polygonization are investigated based on historical data,indicating that,for the selected EMU type,the rate of wheel polygonization is higher for wheels operating in summer,wheels with smaller diameters,wheels on motor vehicles,or wheels that were previously polygonal.Further,a predictive model for wheel polygonization risk is developed based on an array of Bayesian networks.It is demonstrated by using real-life wheel polygonization data that the proposed model can predict the distribution of polygonization risks at different depths and identify high-risk wheels,which can support risk-based maintenance.(4)Risk analysis and prediction of wheel rolling contact fatigue(RCF).This thesis analyzes the influencing factors of RCF and their intrinsic mechanism based on the theory of wheel RCF.Then,dimension reduction methods are applied to historical damage data to extract eight RCF predictors,describing the influence of wheel out-of-roundness,wheel wear,wheelset conicity,wheel diameter,month of operation,EMU fleet,historical state,and running mileage on wheel RCF,respectively.Each predictor is positively correlated with wheel damage rate.Further,a predictive model for wheel RCF risk is developed based on logistic regression.Validation based on real-life wheel damage data shows that the proposed model has the advantages of high accuracy,low computational cost,and good interpretability to support riskbased maintenance.(5)Optimization of maintenance decision-making in wheel life cycle.Based on the established models of wheel wear,reprofiling,and damage,this thesis models the key degradation processes in the form of discrete probability distributions.Then,based on the theory of Markov decision process,all the key processes are integrated into a wheel life cycle degradation model considering sequential maintenance decisions and life cycle costs.Further,a two-loop optimization scheme is proposed to achieve the global optimization of wheel reprofiling strategy and overhaul limits minimizing life cycle costs.Case studies show that the optimized reprofiling strategy can control wheel degradation processes through the principles of prolonging,protecting,and abandoning.Besides,the optimized overhaul limits can avoid cost wastage due to unreasonable maintenance limits.The models and methods proposed in this thesis are of high value for engineering applications.Some of them have been programmed into software,which is tested and applied in a number of EMU fleets in China.It not only improves the safety and reliability of EMU operation but also improves the efficiency and cost savings of EMU maintenance,yielding significant economic benefits.