Construction of a metamodel based on partially converged data in the context of assembly design
Résumé
This paper is concerned with the use of multi-fidelity strategies to compute an appropriate metamodel (or surrogate model) faster than with the traditional methods. In this paper, what we call multi-fidelity is the possibility to compute data with several level of precision. Multi-fidelity metamodeling refers to the enhanced prediction of the output of a complex system by incorporating auxiliary fast-to-obtain data of lower fidelity. The techniques presented in this paper are available for several level of fidelity but only two levels are used here. The LATIN algorithm (an iterative algorithm) paired with a "multiparametric" strategy is used to compute the different data and their different levels of precision by using an error indicator. Five methods to compute an accurate metamodel with low and high fidelity data are compared using two mechanical examples. © Civil-Comp Press, 2014.