Limitations of traditional Bayesian calibration techniques for fuel performance codes

This paper examines standard Bayesian calibration methods for reactor physics, specifically for fuel performance modeling, focusing on model inadequacy. We use a combination of Gaussian Processes and Markov Chain Monte Carlo for sampling from posterior distributions in computationally expensive OFFBEAT simulations. Sensitivity analysis identifies key parameters influencing the fuel rod cross-sectional average temperature at End-Of-Life, reducing the problem’s dimensionality. Our objective is to assess the success conditions of those techniques and their limitations in estimating the prediction uncertainties when dealing with model inadequacy. Future efforts will address shortcomings by combining uncertainty inflation with dimensionality reduction to refine uncertainty quantification in transient simulations.