Data-driven Modeling Method For Train Crash Dynamics And Its Application

After continuous development,China's high railway has begun to take shape.With the development of high-speed railway,train safety is becoming more and more important.Train collision accident is one of the most important safety problems of trains.Once it occurs,it will bring destructive economic and human losses.So it is urgent to solve the collision safety problem of railway vehicles.It is of great significance to strengthen the research of train collision accidents,to design reasonable vehicle structure,to improve the collision resistance of vehicles,to reduce the injury of the occupants of secondary collisions and to avoid the recurrence of accidents.The traditional multi-body dynamic modeling method of train collision is usually based on the force-displacement curve to model the deformable structure at the end of the train.The numerical simulation results show that the train force-displacement curves are different under different initial velocities.Therefore,the traditional multi-rigid body simulation method can generate calculation error in an experiment simulation when a fixed velocity with the forcedisplacement curve is generalized to other velocity.Based on the stochastic forest algorithm in machine learning,a data-driven dynamics model of train collision is presented.The force-displacement curves of the end structure of the vehicle under given initial velocities are accurately predicted from the off-line finite element simulation results under different initial velocities.The results show that the predicted forcedisplacement curve satisfies the practical engineering application accuracy and the generalization ability of the model is strong when the source data is limited.The nonlinear spring damping model(NSDM)is applied to train collision simulation,which is implemented by parallel genetic algorithm and Newmark integral algorithm.In order to solve the parameters of NSDM model by genetic algorithm,120 groups of experiments have been carried out,and the optimal segments of spring and damping piecewise function have been selected by combining the fitness value and calculation efficiency.Taking the lead car collision experiment as an example,this paper gives a case study of the nonlinear spring damping model and force-displacement curve model respectively which are traditional multibody collision dynamic modeling methods.The numerical simulation performance of the three modeling methods under the initial velocity of 36km/h and the generalization ability of the three modeling methods under the initial velocity of 48km/h are compared and analyzed.The experimental results show that both the NSDM model and the DDTCM model can predict the dynamic response of a single vehicle crash scene well under the same initial velocity.As for the collision scenes where the fixed initial velocity is generalized to other velocities,the DDTCM model has the best prediction accuracy.In this paper,the dynamic responses of acceleration and velocity of DDTCM method and FEM simulation experiment data are compared and analyzed through the simulation experiments of 4-vehicle marshalling collision scenes.It is proved that the DDTCM method can be applied to the collision simulation of multi-vehicle marshalling with fast efficiency and that the model presented in this thesis has high engineering application value through the simulation cases of the whole train set collision and “secondary collision”.