• A surrogate framework integrating CFD and DNN is established for lead–bismuth reactor analysis. • The DNN model enables fast predictions in 3–5 s, drastically reducing thermal–hydraulic simulation time. • High accuracy is achieved for interpolation, while extrapolation clarifies the model's application limits. Lead-cooled Fast Reactors (LFR), recognized for their high efficiency and inherent safety, hold significant promise among Generation IV nuclear systems. Nevertheless, the low Prandtl number (Pr) of lead–bismuth eutectic (LBE) coolant results in thermal–hydraulic behaviors distinct from conventional fluids. While Computational Fluid Dynamics (CFD) has advanced the understanding of LBE heat transfer mechanisms, its prohibitive computational cost remains challenging for complex geometries and multiple operating conditions.This study establishes a validated CFD model to systematically generate experimental LBE heat transfer data across diverse operational regimes. These data train a Deep Neural Network (DNN) to develop a rapid-prediction surrogate model. Results demonstrate that the model maintains high predictive accuracy within the sampled parameter space, with deviations observed only for extrapolated conditions. Compared to direct CFD calculations which take 20 min, the proposed DNN-CFD framework achieves a near-instantaneous response of roughly 5 s. The proposed CFD-DNN coupled framework drastically curtails the computational overhead and time-to-solution required for the thermal–hydraulic analysis of lead–bismuth eutectic (LBE) systems, offering a computationally efficient alternative to high-fidelity CFD simulations.
Sheng et al. (Fri,) studied this question.