轴向通流共转腔换热分析及神经网络参数预测

Traditional CFD simulation is time and cost consuming, and it is difficult to quickly predict heat transfer parameters under different structures and conditions. To address this, a data-driven method based on convolutional neural networks (CNN) to predict heat transfer parameters in rotating cavity is developed. Firstly, numerical simulations are conducted in a rotating cavity with axial throughflow based on Reynolds-Averaged Navier-Stokes (RANS) equations, and then heat transfer is analyzed to determine the main characteristic parameters which affect heat transfer. Finally, the prediction model is proposed to predict heat transfer parameters. The results show that the heat transfer state at high radius area is mainly influenced by free convection caused by rotation, while the heat transfer state at low radius area is caused by impact effect and reflux heat exchange. Circumferentially averaged disk Nusselt number distribution is similar at the same Rossby number (Ro). The heat transfer intensity is promoted by widening gap ratio of disk centers (G_c), which is more obvious at high rotational speed and high radius area. Shroud averaged Nusselt number (Nu_sh) increases with the increase of rotation Reynolds number (Re_ω), while it increases first and then decreases with the increase of Ro under different Re_ω , peaking at Ro=1.5. Shroud heat transfer intensity is also enhanced by widening G_c. The CNN network extracts input feature information of different conditions and structures, and predicts heat transfer parameters of disk and shroud. The prediction accuracy of the model is above 90%.