Reconstruction of incomplete velocity field and prediction of pressure field of propeller wake based on deep learning

Efficiently obtaining the complete propeller wake flow field is significant for experimental fluid dynamics analysis and design optimization. This paper proposes a propeller wake flow field reconstruction method based on deep learning, which efficiently reconstructs the missing velocity information while predicting the pressure fields from incomplete velocity fields. The delayed detached eddy simulation model is used for numerical simulations of the propeller to generate large, high-fidelity datasets required for deep learning. The proposed complete propeller wake reconstruction convolutional neural network (CPWR) is established using residual convolution blocks to improve the nonlinear fitting capability, and the transformer encoder module is used to capture the multiscale characteristics of wakes. The sensitivity analysis assessed the impact of the training dataset size and neural network structure, revealing the robustness of the method in selecting these parameters. The results indicate that the CPWR can effectively fill in the missing portions of the propeller wake flow field, and the reconstructed complete wake flow field agrees well with the ground truth in spatial distribution variation. The prediction results of the pressure field also agree well with the ground truth, demonstrating the CPWR's ability to process the complex nonlinear relationship between the velocity and pressure fields. Furthermore, CPWR can reconstruct the complete wake flow field with reasonable accuracy under unseen operating conditions, further indicating the excellent generalizability of the proposed deep learning model in capturing the spatial relationships for missing propeller wake flow reconstruction.