The Research On Improving Surrogate Model Accuracy And Its Application In Lightweight Design Of Conceptual Aluminum Frame

With the development of computer hardware and finite element simulation technology,engineers can optimize the structure of automobiles at the early stage of design,which greatly shorten the development cycle.During automobile structure design,many performances such as mass,stiffness,strength,mode,impact,NVH and fatigue durability need to be considered,and these performances are connected and affect each other,which makes multi-disciplinary optimization widely used in solving engineering proble ms in recent years.However,the repeated iterations during optimization and the long calculation time of the single finite element model make the whole optimization design process needs to take a lot of time,which is difficult to meet the needs of modern body design.In order to reduce the amount of calculation and improve the efficiency of optimization,scholars at home and abroad have proposed replacing finite element model with surrogate model.Although surrogate model is considered as one of the effective ways to solve complex engineering problems,there are still some shortcomings in engineering practice,such as it takes a long time to sample and the precision of the surrogate model is not enough.Based on the principle of experimental design and surrogate model,the sequential sampling method and the ensemble surrogate model are proposed.Then they are applied to the lightweight design of conceptual aluminium frame,which improv e the building efficiency and accuracy of surrogate model.Finally,a conceptual body lightweight design method based on sequential sampling method and ensemble surrogate model is developed.The main research work and conclusions are as follows.(1)Research on sequential sampling method:The CVDA sequential sampling method is proposed to revise the exsiting problems in sequential sampling method.By comparing the effects of five sampling methods such as OLHD,MDA,MSDA,CVA and CVDA with two low-dimensional functions called Farhang Mehr and Multimodal and two high-dimensional functions called Extended Rosenbrock and Dixon-Price,it is concluded that CVDA sequential sampling methods has more obvious advantages in high-dimensional and strongly non-linear functions.(2)Research on assemble surrogate model:The EY assemble surrogate model is proposed to revise the exsiting problems in exsiting assemble surrogate mode.By comparing the effects of three surrogate model such as RSM,KRI,RBF and three assemble surrogate model such as EI,EG,EA with two low-dimensional functions called Branin-Hoo and Jin and two high-dimensional functions called Hock and Dixon-Price,it is concluded that EY assemble surrogate model has more obvious advant ages in high-dimensional,less samples and strongly non-linear functions.The CVDA sequential sampling methods and EY assemble surrogate model are applied to the lightweight design of conceptual aluminum frame.The surrogate models of bending stiffness,torsional stiffness,bending mode and torsional mode are established respectively.The final accuracy of bending stiffness is 0.984,which is 12.6% higher than the initial value.The final accuracy of torsional stiffness is 0.986,which is 17.9% higher than the initial value.The final accuracy of bending mode is 0.942,which is 18.8% higher than the initial value.The final accuracy of torsional mode is 0.947,which is 25.0% higher than the initial value.By using the combination strategy of global optimiza tion and gradient optimization,after 5997 iterations of global optimization and 15 iterations of gradient optimization,the results show that 6.98 kg weight reduction is achieved,with the bending stiffness,torsional stiffness,first order bending and torsional mode frequencies rise by 1.79%,2.67%,3.47% and 1.04% respectively.The effect of weight reduction is obvious,which provides guidance for subsequent detailed design.