Rapid calculation models for flow and heat transfer in tube bundle structures toward digital twin applications

In nuclear thermal systems, digital twin technology enhances intelligent monitoring and optimization by enabling real-time interaction with digital models. However, existing models are limited by scarce monitoring data and the computational burden of mechanistic simulations, restricting real-time applicability. Therefore, developing rapid calculation models has become a key focus. In this study, a tube bundle heat exchanger was investigated. An experimental rig was built to obtain flow and heat transfer data in air-cooled tube bundle channels, and a validated simulation model was established. Four rapid calculation models were then developed: two interpolation-based (bilinear interpolation, BI, and quadratic Lagrange interpolation) and two combining reduced-order modeling with machine learning (POD-SVR and POD-MLP). The effect of training sample size on accuracy and efficiency was evaluated, and the stability and uncertainty quantification of the models were compared. Results show that accuracy improves with increasing training samples, reaching the best performance with 36 conditions. Among the models, the BI-based model performed best, achieving R² = 0.999, RMSE = 0.276 °C, a prediction interval mean of 1.043 °C, and a computation time of 3.21 s. These findings indicate that the bilinear interpolation method, owing to its simplicity and low cost, can serve as a preferred approach for rapid calculation models in digital twin applications of flow and heat transfer. Furthermore, the BI-based model has been preliminarily applied in our digital twin system, enabling real-time interaction between the physical entity and virtual model. This work provides a foundation for future studies on more complex equipment.