Surface area deviation when converting from cell to point data about pyvista HOT 6 CLOSED

What is the purpose of doing this? Area is inherently a cell based quantity.

from pyvista.

It is used as part of the NVidia Modulus mesh graph net example focusing on aerodynamics:
Modulus - Ahmed Body Example

The areas are calculated following mesh loading:
Ahmed data loading and area calculation
and are used post-inference to determine drag forces/coefficients based on inferred pressure and shear stress data.

I found this when running my own models and the resulting forces were off by approximately the same factor that the area numbers were misaligned.

from pyvista.

A couple of points here, is it possible to use integrate_data filter to integrate the point data you are interested in over the area? This would be the way I would do this.

If you must use point data with area converted to point data, you will have to account for the number of points in each cell and the area of each cell that is connected to each point. This is doable, but not trivial. You cannot simply do a sum of the point data. This will give you the wrong answer as you obtained.

from pyvista.

Just to make sure I understand, you're suggesting:

1. Load mesh, don't save point area
2. Train and run inference
3. Add inferred pressure and shear stress data to original mesh as point data (multiplied by normals as needed for force directionality)
4. Run integrate_data on the mesh
5. Use integrated values for pressure and shear stress for forces

If that's correct I can test it out later today on a dataset I'm running now

from pyvista.

Sounds better to me

from pyvista.

Went well for the smaller data set, so I ran a much larger one. Results are within 2% between OpenFOAM and NVidia Modulus... I need to test with a lot more models but for now I'm going to take this as a win.

Thanks for the suggestion!

from pyvista.