A project concerned with the prediction of (laminar) velocity and pressure fields from an input shape using different U-net architectures. If you exploit this work for your own research, please consider citing https://arxiv.org/abs/1910.13532. This work was produced during the PhD thesis of J. Chen.
The dataset contains 12.000 random shapes.
Each shape in the dataset is labeled with its velocity and pressure fields (directly as images):
Here are some examples of flow predictions at Re=10
. Top row is the U-net prediction, while bottom row is the CFD computation.
Different U-net architectures were exploited in this work: