Crashworthiness optimization using a surrogate approach by stochastic response surface
Résumé
In the automotive passive safety field, numerical simulations gradually replace experimental crash-tests, and allow through parametric studies an improved definition of the architecture and the sizing of vehicles. In this context, this paper is focused on a methodology for crashworthiness optimization. After a review of difficulties inherent to the numerical modeling, we propose a global optimization strategy based on a surrogate approach: the resolution of the real optimization problem is replaced by a sequence of resolutions of approximate problems. An interpolation model is adopted in order to smoothen the objective function and constraints and to enable the analytical calculations of their gradients. The response surface model is build by a stochastic process. Unlike traditional techniques of construction of polynomial response surfaces by least squares regression, the approach developed, based on Sph (smooth particle hydrodynamics) methods, makes it possible to reproduce strong nonlinearities of the objective functions and limiting constraints. Moreover, the flexibility of these models allows the updating of the approximation during the optimization process, which makes it possible to improve locally the quality of the approximations. We compare the quality of the approximations for various types of optimal design of experiments.