The thermal-hydraulic characteristics of reactor cores is crucial for ensuring reactor safety analysis. Although Computational Fluid Dynamics (CFD) simulation can provide high-fidelity data for the core flow field, it consumes enormous computational resources. Conversely, solely using the physics-informed neural networks (PINN) is insufficient to fully describe the mathematical and physical equations involved in the complex phenomena within the reactor cores. This paper proposed a method for constructing high-fidelity flow field data for reactor cores: using low-fidelity CFD data to initialize PINN and encoding certain mathematical and physical equations into the PINN. The method is tested using a basic cylinder flow disturbance problem and the obtained velocities and temperatures show significantly improved accuracy compared to the low-fidelity data. Then, the method is applied to construct the core flow field data.
 

CFD Analysis,Constructing High-Fidelity Flow Field Data,Physics-Informed Neural Networks