Study On Wheel/rail Matching And Hunting Characteristics Of A High-speed Train

During the operation of high-speed trains in China,the low-frequency swaying and high-frequency shaking phenomenon caused by the lack of hunting stability due to poor wheel/rail matching occur occasionally,which seriously reduce the ride comfort and running safety.Limited by the strong nonlinearity of vehicle system,the general law of hunting phenomenon of high-speed trains and the exact relationship between wheel/rail matching and hunting motion have not been found.In this thesis,a random simulation method is adopted to carry out the nonlinear dynamic simulation of a high-speed trailer car,and the hunting characteristics under wide-area wheel/rail matching is obtained.Based on machine learning algorithms,the state of hunting is identified according to the wheel/rail contact parameters.The specific researches are as follows:(1)An anomaly detection method combining wavelet analysis and short-time energy is used to analyze the measured wheel/rail profile with distortion areas or outliers,which can accurately locate and eliminate the outliers of the wheel/rail profiles.The effects of profile discretization with equal arc-length and with equal x-distance on the vehicle’s small radius curve passing performance and hunting stability are compared,and the equal arc-length discretization method of 0.5 mm is recommended to avoid the interference of the profile resampling on the vehicle running performance.(2)The effects of the wavelet denoising with sliding-window algorithm and the spline approximation algorithm on the smoothing effect of the measured wheel/rail profiles,wheel/rail contact geometry parameters and vehicle hunting are studied.The wavelet denoising algorithm can adaptively eliminate the noise of the measured profiles,and the processed profiles fit the original one better.The proper spline approximation parameters are difficult to determine,and the improper parameters would result in underfitting or overfitting of the measured profiles.Reasonable profile smoothing algorithm can revise the positions of wheel/rail contact points,smooth the curvature of profile,reduce the deviation of equivalent conicity,improve the accuracy and efficiency of vehicle dynamics simulation,and the calculation results of hunting motion are more consistent with the actual operation.(3)An approximation algorithm for fast calculating the hunting bifurcation diagram and hunting frequency of high-speed trains is studied.The hunting bifurcation diagram and critical speed is approximately obtained by the wheelset lateral displacement peaks which are calculated by removing the outliers with a sliding-window,and the hunting frequency is determined by the zero-crossing search.The results show that the algorithm is robust to typical operating conditions and the obtained hunting bifurcation diagram and frequency curve are smooth,which meet the requirements of engineering applications.(4)The deceleration-speed method is used to calculate the hunting response of a high-speed train under wide-area random wheel/rail matching.The extensive dynamics simulations show that the inherent hunting characteristics of the vehicle consists of two main categories: one in which the vehicle starts to undergo hunting at a lower speed,with large wheelset lateral displacement and low hunting frequencies;and another in which the vehicle starts to undergo hunting only at a higher speed,with smaller wheelset lateral displacement and higher hunting frequencies.The former is mainly due to the fact that the equivalent conicity is less than the normal range,and the wheelset equilibrium position tends to deviate from the track centerline when the equivalent conicity is less than or close to 0.The latter is mainly caused by the excessive large equivalent conicity due to wheel wear,and the wheelset lateral displacement curve is prone to burr,local concave-convex phenomenon near the peak points,and complex hunting motion such as quasiperiodicity and chaos at high speed.(5)According to the critical speed,the hunting state is classified as "no hunting","hunting with low frequency",and "hunting with high frequency and small amplitude".The wheel/rail contact parameters and vehicle-track operation parameters are used to identify the hunting state based on machine learning algorithms,and the F1-values of the three hunting categories are 0.96,0.78 and 0.91 respectively.The results of the feature contribution of the ensemble learning algorithm shows that the equivalent conicity at the wheelset lateral displacement of 1 mm rather than 3 mm has the greatest effect on the hunting characteristics of the vehicle.