A Scale Bridging Surrogate Model for UO2 Fuel Creep

Modeling the thermo-mechanical behavior of light water reactor fuel rods requires predictive models for fuel creep. Lower length scale modeling provides crucial insights into the microscopic mechanisms driving creep behavior, but these insights must be translated into forms that are usable at the engineering scale. We'll present our efforts in creating a scale-bridging surrogate model using artificial neural networks, trained on lower length scale simulation data. The surrogate model is integrated into the fuel performance code BISON and uses GPU acceleration to deliver a model suitable for large engineering scale calculations.
To evaluate the effectiveness and accuracy of our surrogate model, we conducted a series of scoping calculations over the parameter space. Furthermore, we applied the model to integral instrumented fuel assessment cases, providing a robust comparison against the current empirical model used in BISON.
We will discuss the specifics of the model implementation, focusing on the use of GPU acceleration through the coupling of BISON with the NEML2 material model library. Our results demonstrate significant step towards integrating lower length scale modeling data at the engineering scale at high computational efficiency, as a pathway to advance predictive modeling of the thermo-mechanical behavior of light water reactor fuel rods.