Application of data driven POD reduced order model to helical cruciform fuel rod bundle

To enable efficient and accurate prediction of coolant thermal–hydraulic characteristics in helical cruciform fuel rod bundle channels, a data-driven rapid prediction framework is developed by combining proper orthogonal decomposition (POD) with surrogate models. Optimal Latin hypercube sampling (OLHS) was used to generate 34 operating conditions, and high-fidelity data are obtained through computational fluid dynamics (CFD) simulations. The modal bases and coefficients of temperature, axial and two lateral velocity fields were extracted. With 99.99% energy retention, the minimum reconstructed modal numbers were determined as k T = 2, k w = 2, k u = 3, k v = 4, respectively. Multi-section reconstruction results demonstrate that the temperature error is within ± 2 K, while those of axial and lateral velocity fields are within ± 0.08 m/s and ± 0.001 m/s, respectively. The reconstruction accuracy is insensitive to axial position and rotation angle. Furthermore, the prediction performances of Kriging and radial basis function (RBF) surrogate models are compared. The Kriging model exhibited local high errors in regions with strong gradients, whereas the RBF model showed more uniform error distributions, with prediction errors below 0.5 K for temperature and below 0.002 m/s and 0.010 m/s for lateral and axial velocity fields, respectively. The maximum relative error is only 2.02%. Moreover, the POD-RBF framework reduced the computational time of a single case from 7200 s to 7.328 s, achieving a speed-up factor of 982.533. These results demonstrate that the POD-RBF model provides an efficient and accurate approach for rapid thermal–hydraulic prediction of helical cruciform fuel rod bundles.