A Bayesian framework of inverse uncertainty quantification with principal component analysis and Kriging for the reliability analysis of passive safety systems

• DOI: 10.1016/j.nucengdes.2021.111230
• Publication Date: 2021-04-19T18:52:41Z
• Authors: Giovanni Roma; Francesco Di Maio; Andrea Bersano; Nicola Pedroni; Cristina Bertani; Fulvio Mascari; Enrico Zio
• Keywords: Surrogate modeling; Principal component analysis; 600; 02 engineering and technology; Passive safety systems Reliability analysis Inverse uncertainty quantification; Kriging; [SPI]Engineering Sciences [physics]; Passive safety systems; 0202 electrical engineering, electronic engineering, information engineering; [INFO]Computer Science [cs]; Inverse uncertainty quantification; Reliability analysis; Inverse uncertainty quantification; Kriging; Passive safety systems; Principal component analysis; Reliability analysis; Surrogate modeling

Abstract In this work, we propose an Inverse Uncertainty Quantification (IUQ) approach to assigning Probability Density Functions (PDFs) to uncertain input parameters of Thermal-Hydraulic (T-H) models used to assess the reliability of passive safety systems. The approach uses experimental data within a Bayesian framework. The application to a RELAP5-3D model of the PERSEO (In-Pool Energy Removal System for Emergency Operation) facility located at SIET laboratory (Piacenza, Italy) is demonstrated. Principal Component Analysis (PCA) is applied for output dimensionality reduction and Kriging meta-modeling is used to emulate the reduced set of RELAP5-3D code outputs. This is done to decrease the computational cost of the Markov Chain Monte Carlo (MCMC) posterior sampling of the uncertain input parameters, which requires a large number of model simulations.