Reliability-based Design Optimization For Vehicle Collision Based On Adaptive Gaussian Process Regression

Vehicle collision is subject to numerous uncertainties arising from manufacturing processes,human cognition,and external factors.These uncertainties include material properties,structural geometry,dimensions,and loads.When these uncertain factors are combined,traditional deterministic optimization methods for vehicle collision safety design may introduce significant analytical biases and fail to meet safety requirements.Furthermore,analyzing of car collisions is computationally intensive.It requires repeated calculations of the vehicle model to ensure reliability,which further exacerbates the computational workload and lowers the efficiency of the analysis.To improve the efficiency of vehicle collision reliability analysis,this paper proposes an adaptive dimension-reduction Gaussian process regression model.Based on this model,an optimization design method for a vehicle is developed.The specific research work is as follows:(1)Perform a frontal collision analysis on the vehicle model.A finite element model for vehicle collision is established,and corresponding collision parameters are set.The collision finite element model was calculated,and the energy change,B-pillar acceleration,and firewall deformation after the vehicle collision were obtained.(2)An adaptive dimension-reduction Gaussian process regression model is proposed.First,the Gaussian regression model is reduced to second order to simplify the original complex high-dimensional vehicle model.Then,based on an adaptive method,the lowcorrelation bivariable terms are eliminated.This method reduces the required calculation sample points while maintaining accuracy,improving the efficiency of model building.The adaptive dimension-reduction Gaussian process regression model is used to fit the deformation of the firewall,saving computation costs for research on collision reliability analysis.(3)Based on active learning,a method for reliability analyzing of vehicle collisions is proposed.The accuracy of reliability calculation mainly depends on the fitting accuracy of limit state function by metamodel.Active learning is used to ensure that the sampling positions for the vehicle collision model are closer to the limit state function in reliability analysis.This is to improve the accuracy of the model near the limit state function.Finally,Monte Carlo method and the adaptive dimension-reduction Gaussian process regression model are used to calculate the reliability of vehicle collisions.(4)Based on the whale optimization algorithm,a vehicle collision reliability optimization design is carried out.In this optimization design process,the thickness of 6vehicle parts is selected as design variables.The influence of uncertainty factors was considered,and a collision reliability optimization design model for vehicle was established with collision reliability as a constraint and minimizing vehicle weight as the optimization objective.