Graph neural networks for laminar flow prediction around random two-dimensional shapes

In recent years, the domain of fast flow field prediction has been vastly dominated by pixel-based convolutional neural networks. Yet, advent graph networks (GCNNs) attracted considerable attention in computational fluid dynamics (CFD) community. this contribution, we proposed a GCNN structure as surrogate model for laminar around two-dimensional (2D) obstacles. Unlike traditional convolution on image pixels, can be directly applied body-fitted triangular meshes, hence yielding an easy coupling with CFD solvers. The is trained over dataset composed CFD-computed flows 2000 random 2D shapes. Accuracy levels are assessed reconstructed velocity and pressure fields out-of-training obstacles compared that standard U-net architectures, especially boundary layer area.