A Fast and High-Fidelity Multi-Parameter Thermal-Field Prediction System Based on CFD and POD Coupling: Application to the RPV Insulation Structure
Yanjun Dai,Jie Zhao,Xiaoli Gui,Yungang Wang,Wenquan Tao,Zhenhai Zou,Fan Bai
DOI: https://doi.org/10.1016/j.ijheatmasstransfer.2024.125985
IF: 5.2
2024-01-01
International Journal of Heat and Mass Transfer
Abstract:The thermal performance of the reactor pressure vessel (RPV) insulation structure plays a crucial role in ensuring the operational safety of the reactor. In this study, a fast and high-fidelity prediction system is proposed, which is based on the snapshot ensemble generated by the computational fluid dynamics (CFD) method and incorporates proper orthogonal decomposition (POD) as its core component to predict the heat transfer performance of the RPV insulation structure. The proposed thermal-field prediction system is employed to examine the impacts of air inlet temperature, mass flow rate, and RPV outer wall temperature on the thermal performance parameters: average heat flux qout and temperature Tout of the RPV insulation outer wall, as well as the maximum local temperature Tmax of the concrete pit inner wall. The POD predictive results showcase that the maximum relative deviations of qout, Tout and Tmax for eight sets of off-design cases are 2.25 %, 3.04 %, and 2.37 %, respectively. Meanwhile, the maximum values of qout, Tout and Tmax are 90.13 W & sdot;m- 2, 44.69 degrees C, and 87.51 degrees C, respectively. They are all less than their prescribed limits. The field distributions of heat flux and temperature in off-design cases exhibit consistency between the CFD and POD methods. Notably, in comparison to the CFD simulation method, the POD method capturing more than 99.9 % energy demonstrates significant computational time savings for 99.87 %. Moreover, a multivariate linear regression (MLR) model for thermal performance parameters is established based on the POD prediction datasets. The results indicate that the maximum relative deviations between the POD and MLR methods in predicting qout, Tout and Tmax for the 72 sets of test cases are 5.1 %, 7.9 % and 7.1 %, respectively. Meanwhile, 99.99 % CPU time is reduced for MLR model compared with CFD method. Obviously, the predictive system showcases excellent accuracy and efficiency in forecasting.