No abstract

No abstract

In this paper, we discuss how to replace a timedemanding computer code having both quantitative and qualitative variables with a surrogate model based on Kriging technique. The recourse to a predictive model is motivated by a challenging electromagnetic compatibility problem of maximum lightning-induced current in electronic equipment connected to a large industrial site. The technical aspects of the incorporation of qualitative variables via specific correlation functions are detailed. The accuracy of the built surrogate model is statistically validated. The application of the resulting metamodel to the uncertainty analysis of induced currents will be also investigated.

No abstract

Abstract-Several computer codes with varying accuracy from rigorous full-wave methods (highfidelity models) to less accurate Transmission Line (TL) approaches (low-fidelity model) have been proposed to solve EMC problems of interference between parasitic waves and wired communication systems. For solving engineering tasks, with a limited computational budget, we need to build surrogate models of high-fidelity (HF) computer codes. However, one of their main limitations is their expensive computational time. Rather than using only the computationally costly HF simulations, we apply another type of surrogate models, called Multifidelity (MF) metamodel which efficiently combines, within a Bayesian framework, high and low-fidelity (LF) evaluations to speed up the surrogate model building. The numerical results of combination of an expensive EMC simulator and a cheap TL code to solve a plane wave illumination problem, show that, compared to Kriging, a reliable Bayesian MF metamodel of equivalent or higher predictivity can be obtained within less simulation time.