2026 Annual Technical Meeting of CCP/SIG Nuclear Thermal Hydraulics

This talk introduced a neural-gated mixture of engineering models approach for data-driven boiling predictions. Building on lessons from XBOIL, the presentation explored how existing empirical, semi-empirical, and mechanistic boiling models can be blended using a learned neural gate rather than replaced by a black-box model. Applications included bubble departure diameter prediction, boiling curves, and critical heat flux, highlighting the potential for improved agreement with experimental data while retaining interpretability through explicit model weights.

2026 Annual Technical Meeting of CCP/SIG Nuclear Thermal Hydraulics