Data Assimilation for Fuel Performance Code Development: Application to Oxide Fuel Thermal Properties

Data assimilation methods are suitable tools for handling experimental data and extracting synthetic results from them. In this work, the Gaussian Process (GP) method for making regressions on the thermal properties of fuel is presented. GP is a nonparametric supervised learning method used to solve regression and probabilistic classification problems. A GP regression model can make predictions incorporating prior knowledge (kernels) and provide uncertainty measures over predictions, with the advantage of not requiring assumptions about the shape of the correlation but relying only on the data set. The objective is to demonstrate the validity of this approach for choosing thermal property values within Fuel Performance Codes (FPCs) for the fuel pin thermo-mechanical analysis, for which two test cases are shown: melting temperature and thermal conductivity. For this purpose, experimental data regarding the MOX fuel are treated with Gaussian processes and the obtained synthetic results are provided to the TRANSURANUS FPC, without increasing the computational demand while running the FPC. These models were implemented as an external shell written in the Python language for SCIANTIX, an open-source meso-scale code to describe inert gas behaviour in nuclear fuel, that can be coupled with FPCs. The implementation is performed in a modular flexible way, to be extended to full set of models and properties.