######################## Our method proposes a neural network architecture, AMGNET, which combines graph neural networks with the core ideas from the algebraic multigrid method. In essence, this network architecture consists of a series of coarsening layers (encoding), processing layers (processing) followed by recovery layers (decoding). The first layers perform a coarsening of the fine-scale input to different grid scales, from which features can be extracted through graph network blocks. Given this extracted information, the recovery layers reconstruct the predicted flow field through an upsampling procedure.
Airfoil dataset comes from Filipe de Avila Belbute-Peres, Thomas D. Economon, and J. Zico Kolter. Combining differentiable PDE solvers and graph neural networks for fluid flow prediction.
Cylinder dataset is generated by using Ansys Fluent
train_airfoil.py train on airfoil training dataset tarin_cylinder.py train on cylinder training dataset test_airfoil.py test on airfoil test dataset test_cylinder.py test on cylinder test dataset
utils.py Contains some method functions, such as writing tecplot files
environment:
torch 1.10.0+cu113
torch-cluster 1.5.9
torch-geometric 2.0.2
torch-scatter 2.0.9
torch-sparse 0.6.12