Use of Multi-Fidelity Training Data in Uncertainty Analysis of Nuclear Engineering Applications

TitleUse of Multi-Fidelity Training Data in Uncertainty Analysis of Nuclear Engineering Applications
Publication TypeConference Paper
Year of Publication2013
AuthorsRoderick, O, Anitescu, M
Conference NameAmerican Nuclear Society 2013; Atlanta, Georgia
Other NumbersANL/MCS-P4074-0513
Abstract

Uncertainty analysis of computational models plays an important role in nuclear engineering due to reliance on simulation codes to augment the rare and expensive experimental data. While development of hardware and computing techniques enables models of high fidelity and resolution, the data obtained by running such codes has significant computational cost. We resolve the contradiction between the need for data, and the cost of simulation by querying the simulation code at multiple levels of fidelity and relating multi-fidelity data by statistical models.

Previously, we presented elements of analysis on multi-fidelity training data. We are now demonstrating effectiveness on codes of higher complexity, with a fluid mechanics solver Nek5000 used as an example.