BackgroundThe accuracy of the chosen flow model and near-wall treatment requires detailed experimental verification. It can affect the accuracy of the numerical results obtained. Thus, the selection of an appropriate flow model is important. However, a review of the literature found that few researchers have engaged in the inverse convection-conduction conjugate heat transfer problem. MethodsThis study uses an inverse computational fluid dynamics (CFD) method combined with the least squares method and overdetermined experimental data to predict the unknown heat transfer rate Q and the average Nusselt number Nuk on each tube. The conservation equations of the air outside and inside the hot box and the heat conduction equation of the hot box walls are also solved by coupling. Then, in order to select a suitable flow model, a comparison of the root mean square errors between the CFD results of temperature obtained from all selected flow models and the experimental data is required. Significant FindingsThe significant findings are that the RNG k-ε model is selected as the appropriate flow model for this study through the validation tests of various flow models. The vortex ring appears in the central low-velocity region of the four tubes with StD=2 and gradually disappears with the decrease of StD. The two thermal plumes above the upper two tubes with StD=4/3 are slightly inclined inward and merge with each other. But these two plumes above the upper tubes are inclined slightly outward as StD increases. The obtained velocity patterns and Nuk agree with the interferometric images and the modified correlation, respectively.