Bayesian Calibration of UO₂ Creep Rates as a Tool for AFQ

Creep is an important deformation mechanism in UO₂ nuclear fuel used in Light Water Reactors (LWRs). Previously, using results from atomistic simulations, the Coble mechanism was found to be dominant in the diffusional regime. Here, we aim to address the inherent uncertainties from the lower length-scale data by building on a Bayesian inference calibration of uranium self-diffusion and UO_2±x thermochemistry to also include UO₂ thermal creep. Our goal is to calibrate the UO₂ thermal creep model and predict probability distributions on lower length-scale parameter values that would explain the experimental creep measurements. To reduce computational burden, we replace the creep model with a neural network surrogate model trained upon hundreds of thousands of model evaluations with randomized values of the most important underlying parameters which impact the creep results. These important parameters were identified by a sensitivity analysis study. Our calibrated results show meaningful uncertainties for the underlying lower length-scale parameters and uncertainty for the predicted creep rates. Furthermore, we infer the non-stoichiometric conditions that the experimental creep rates were most probably conducted under and show how this impacts the experimentally measured creep rates.