A multi-fidelity multi-scale methodology to accelerate development of fuel performance codes

Multi-scale methodologies have been developed and applied successfully in the frame of nuclear fuel performance analyses, but the complexity of the tools involved hinders their extensive application. Gaps in modelling capabilities of specific input/outputs in particular limits code-to-code communication. In this work, we propose a multi-fidelity methodology to tackle this issue. The application presented here concerns the inclusion of a meso-scale module describing fission gas behaviour (SCIANTIX) in a fuel performance code (FRAPCON). A critical input parameter of the meso-scale module, the local hydro-static stress in the fuel, is not predicted by such fuel performance code, hence limiting this coupling. This gap is filled by using a second fuel performance code (TRANSURANUS) to construct a virtual dataset of local hydro-static stress values, on which an artificial neural network is trained and included in the FRAPCON/SCIANTIX coupled suite. This multi-fidelity methodology is demonstrated by simulating the Risø AN3 irradiation experiment. • A multi-fidelity methodology is proposed to enhance nuclear fuel rod performance calculations. • The meso-scale SCIANTIX module is integrated into the FRAPCON code to model fission gas behaviour. • An artificial neural network provides FRAPCON with local hydro-static stress data. • The methodology is validated against the Risø AN3 irradiation experiment.