Vehicle Collision Surrogate Model Based On Numerical Simulation And Machine Learning

Vehicle collision is one of the important problems in vehicle passive safety.It is of great significance to improve the safety of the vehicle colliding and ensure the safe living space of the passengers.Traditional vehicle collision analysis uses vehicle collision test and numerical simulation,which has the disadvantages of high cost and low efficiency.Combined with the numerical simulation of vehicle collision and machine learning algorithm to build the surrogate model of vehicle collision,it can quickly predict the energy absorption characteristics of vehicle collision,provide a fast and effective basis for the safety and optimal design of vehicle,and has important engineering practical value.Firstly,this paper describes the nonlinear finite element method to simulate the dynamic response process of vehicle collision,including the finite element discretization of vehicle structure,Gauss integral,hourglass control,time integral method,collision contact algorithm and its implementation in ABAQUS finite element analysis software.Taking a pick-up truck as the research object,combined with Hyper Mesh and ABAQUS / CAE software,the three-dimensional finite element model of the truck is established,including geometry model,mesh division,mesh quality control,material attribute,part connection,contact setting,initial condition and boundary condition setting and result output setting,etc.Using ABAQUS finite element software to simulate the dynamic process of a truck colliding against a rigid wall and the frontal collision of two vehicles at five different initial speeds,the deformation process of the vehicle at different initial speeds is analyzed,and the collision force displacement curve and time acceleration curve of the vehicle under two kinds of collisions are obtained.Based on the results of finite element numerical simulation and BP neural network algorithm,a surrogate model is established to predict the vehicle collision characteristics.BP neural network modeling includes the determination of BP neural network layers,nodes in each layer,loss function and optimization algorithm.The finite element numerical simulation samples are divided into training set and test set.The test set samples are used to verify the accuracy of the model,establish the vehicle collision surrogate model in the full speed domain,and analyze the performance of the surrogate model.By using the surrogate model,the collision force displacement curve and time acceleration curve of the vehicle can be predicted rapidly at any collision speed,which provides the basis for the safety and optimal design of the vehicle.