Deep learning-based fast prediction of flow field around multiple bluff bodies

Bluff bodies are widely employed in practical engineering, but they also introduce complex wind-induced flow characteristics, particularly due to interference effects arising from the interaction of multiple bluff bodies. Traditional wind tunnel experiments and numerical simulation methods are time-consuming and costly, posing significant challenges to design and construction. This study employed deep learning technique to rapidly predict two-dimensional flow fields around multiple bluff bodies. First, a physics-informed deep learning model is proposed, incorporating residual equations representing physical laws, i.e. the continuity and momentum equations, as loss functions. By enforcing the model training to converge in directions consistent with actual physical laws, the physics-informed deep learning model achieves rapid and high-precision predictions of the flow fields around multiple bluff bodies. Additionally, by utilizing spatially permutation invariance of point cloud method on the input of deep learning model, the data input no longer requires strict spatial constraints as in traditional methods, making it more representative of real-world data and expanding the engineering application scenarios of deep learning networks.