Adaptive unscented transform for uncertainty quantification in EMC large-scale systems
Résumé
Purpose - The purpose of this paper is to introduce a novel methodology for uncertainty quantification in large-scale systems. It is a non-intrusive approach based on the unscented transform (UT) but it requires far less simulations from a EM solver for certain models. Design/methodology/approach - The methodology of uncertainty propagation is carried out adaptively instead of considering all input variables. First, a ranking of input variables is determined and after a classical UT is applied successively considering each time one more input variable. The convergence is reached once the most important variables were considered. Findings - The adaptive UT can be an efficient alternative of uncertainty propagation for large dimensional systems. Originality/value - The classical UT is unfeasible for large-scale systems. This paper presents one new possibility to use this stochastic collocation method for systems with large number of input dimensions.