ExternalFlowDataGeneration

Preliminary scripts generating external flow simulation examples for graph ML.

Overview

Here we provide a repository for generation of large amounts of flow simulations with fenics and oasis solvers. Currently cylinder geometries, ellipse geometries and channel nozzle geometries are supported. Geometries are generated and meshed with a gmsh kernel, and mesh formats are processed with meshio for adaptation to FEniCS.

Setup

To create a environment using conda:

conda update -n base -c defaults conda
conda env create --name fenicsproject --file=environment.yaml
conda activate fenicsproject

Quick start

Mesh generation

Mesh generation is completed in mesh.py. Change parameters in

if __name__ == "__main__":

part to create mesh of different sizes. Or write your own parameters and incorporate

from mesh import createFreeFlowMesh, createChannelFlowMesh

to use the [gmsh] kernel for your own mesh. Mesh file is stored by default in io_operations.

Run simulations

For a free flow dataset, an oasis simulation can be executed by:

python NSfracStep.py problem=Cylinder meshname=ellipse_1 meshdir=io_operations T=10 dt=0.001 output_timeseries_as_vector=False folder=data

To run multiple simulations, refer to simulation_series_oasis.py. Simply run the script to perform simulation on all samples generated. Run 'simulation_series.py' to perform the same simulation on a naive IPCS implementation by directly encoding all the functions into FEniCS.

Construct dataset