An accuracy-enhanced transonic flow prediction method fusing deep learning and a reduced-order model

It is difficult to accurately predict the flow field over an aircraft in the presence of shock waves due to its strong nonlinear characteristics. In this study, we developed an accuracy-enhanced flow prediction method that fuses deep learning and a reduced-order model to achieve accurate flow field prediction for various aerodynamic shapes. Herein, we establish a convolutional neural network/proper orthogonal decomposition (CNN-POD) model for mapping geometries to the overall flow field. Then, local flow regions containing nonlinear flow structures can be identified by the POD reconstruction to build the enhanced model. A CNN model is established to map geometries to the local flow field. The proposed method was applied to two cases involving the prediction of transonic flow over airfoils. The results indicate that the proposed accuracy-enhanced flow prediction method can reduce the prediction error for flow properties in regions with nonlinear flow structures by values ranging from 13% to 66.27%. Additionally, the proposed method demonstrates better efficiency and robustness in comparison to existing methods, and it can also address the prediction problem of complex transonic flow with multiple strong nonlinear structures.