Adaptive multi-point sequential sampling methodology for high-dimensional and highly nonlinear crashworthiness optimization problems

Abstract

Metamodel-based crashworthiness optimization is an expensive and highly nonlinear design problem. Due to the lack of finite element simulations, the responses fitted by different metamodeling methods are not fully equivalent to the real responses. The metamodel error may induce to find a local or an infeasible design solution. Compared to the traditional one-step sampling DOE method, the objective-oriented sequential sampling strategies have been demonstrated as a higher efficient way to find the true optimum design. However, existing infilling criteria of the sequential sampling methods are restricted to specify the number of the sequential samples obtained in each iteration. It is not practical for the real engineering optimization applications. In this paper, a new adaptive multi-point sequential sampling method is developed. The sequential samples obtained in each iteration are determined by the prediction states of the fitting metamodels. To demonstrate the benefits, the new proposed method is applied to a high-dimensional and highly nonlinear frontal crashworthiness optimization problem. Results show that the proposed method can mitigate the effect of the metamodel prediction error and more efficiently find the global design solution compared to the conventional approach.