Kriging as an alternative for a more precise analysis of output parameters in nuclear safety-Large break LOCA calculation

• Published in: Applied stochastic models in business and industry Vol. 26; no. 5; pp. 565 - 576
• Main Authors: Roustant, Olivier, Joucla, Jérôme, Probst, Pierre
• Format: Journal Article
• Language: English
• Published: Chichester, UK John Wiley & Sons, Ltd 01.09.2010; Wiley
• Subjects: best estimate methods; computer experiments; Estimates; Irregularities; Kriging; kriging model; Mathematical models; meta-modelling; peak cladding temperature; Regression analysis; response surface methodology; Risk; Sensitivity analysis; Uncertainty; meta-modelling; sensitivity analysis; best estimate methods; peak cladding temperature; response surface methodology; kriging model; computer experiments
• ISSN: 1524-1904; 1526-4025; 1526-4025
• DOI: 10.1002/asmb.800

The methods using the best estimate codes are now applied in safety demonstration for nuclear power plants (NPP) to evaluate uncertainties of the relevant output parameters. Towards this objective, it is useful to further analyse the outputs, for example, to learn more about sensitivity to input parameters. In addition, this first analysis can be used to assess uncertainty. Such an analysis is difficult to obtain using the code itself because it is quite time‐consuming. One approach, called response surface methodology, consists in replacing the code by a simpler model, estimated with few runs. Linear regression is often used. In this paper, we propose kriging as introduced by Sacks et al. (Technometrics 1989; 31:41–47; Stat. Sci. 1989; 4(4):409–435) as an alternative. Kriging was applied to the Loss‐of‐Fluid Test (LOFT) loss of coolant experiment L2‐5, which was the subject of the former ISP 13 and the ongoing BEMUSE (ISP: International Standard Problem. BEMUSE: Best‐Estimate Methods–Uncertainty and Sensitivity Evaluation) international problem. The output is the second maximum peak cladding temperature (PCT) of the fuel. The best estimate code used is CATHARE2 V1.3L. We observe that kriging is more flexible and can handle irregularities. As a result, it gives more accurate predictions. In addition, sensitivity analysis is provided. This method offers complementary information and constructs a response surface more accurately, with a more realistic evaluation of risk. Copyright © 2009 John Wiley & Sons, Ltd.