Research On Key Technologies Of Hybrid Metamodeling For Lightweigt Design Of Autobody Structure

The development of automotive technology and increasingly strict emission regulations have determined that electrification is an irreversible trend in automobiles.Due to the low technology and the maturity of the electric drive itself,the quality of the electric vehicle under the same platform is often heavier than that of the fuel vehicle.In order to cross the end of life,the need for lightweight in the electrification era is even more urgent.Through the advanced optimization algorithm,the material,the optimal structure shape and size are used in the proper position of the automobile structure,so that each part of the material can exert its maximum load-bearing or energy-absorbing effect,which is an important way to realize the lightweight structure.In practical lightweight applications,for the strong nonlinear performance response of collision safety,vibration and noise,it is difficult to meet the requirements of modern body design and development through finite element analysis.The metamodel is used to fit and predict the structural performance response.It can greatly reduce the computational cost of the optimization process and improve the optimization efficiency.Although the metamodel based design optimization is considered as one of the most effective ways to solve complex structural design problems,there are still many shortcomings in engineering practice,which may cause the lightweight design scheme to fail.Based on the lightweight design parameters of the vehicle body structure,this paper studies the key technologies of hybrid metamodeling,and based on the design logic of “model construction-model update-bias correction-model based optimization”,it establishes a metamodeling framework for strong nonlinear performance of the body structure.The main research work is as follows:(1)For the existing hybrid metamodeling research,the problem of element metamodel selection is usually neglected,the model sifnificant index based on partial least squares regression is derived,and the strategy for unit model selection based on sifnificant index is established.An ideal set of element models is established under the conditions of quantity,prediction accuracy,and divergence between each element metamodel.In order to effectively avoid multiple correlations when calculating weight coefficients,a hybrid metamodeling process based on partial least squares regression is established.The effectiveness and robustness of the proposed method are verified by six,nine-dimensional,twelve-dimensional,sixteen-dimensional mathematical functions and two typical crashworthiness responses.(2)For the traditional sequential sampling algorithm,it is only suitable for the approximation metamodeling problem of single response.A dynamic sampling strategy based on multi-objective optimization and variable-weight TOPSIS is proposed,and an incremental learning process for multi-response approximation is established.By deep mining the knowledge information of the training sample and the element models,based on the NJV index,the sequential sample points are gradually increased in the region with large uncertainty of fitting to improve the global prediction accuracy of the hybrid model.Taking the response of acceleration and displacement involved in the lightweight design of the vehicle body as the research object,the influence of the initial sample point size and weight calculation method on the incremental learning effect of the integrated approximation model is analyzed.(3)For the inevitable existence of the prediction uncertainty of the metamodel,a Bayesian inference-based bias correction method for hybrid metamodeling is proposed,and the influence of the prior distribution on the effectiveness of the bias correction is discussed.Combined with the above-mentioned element metamodel selection strategy and incremental learning mechanism,an adaptive bias correction process of the hybrid n model is established.The research on the strong nonlinear responses of the vehicle body structure proves that the established flow can gradually reduce the prediction uncertainty of the hybrid metamodel.In addition,accurate quantification of prediction uncertainty can further guide the optimization design considering uncertainty.(4)For solving the optimization solution failure problem caused by ignoring the uncertainty of the approximate model in the traditional robustness optimization design,the robustness design based on the fusion of multi-source uncertainties considering the influence of the uncertainty of the design variables and the uncertainty of the hybrid metamodel is established.In order to further reduce the influence of the uncertainty of the hybrid metamodel on the robust optimization solution,an improved global constrained robustness optimization design algorithm is proposed.According to the established process,the lightweight crashworthiness design of the front structure of the car and the multi-disciplinary design are studied.The effectiveness and engineering feasibility of the strategy are verified,which provides effective optimization for the lightweight design of the vehicle body.The research aims for providing better surrogate models for lightweight design of autobody,as well as guiding and improving the process of lightweight autobody development,towards the end goal of improving the R&D capability of vehicle lightweight design.