Wrong gradients in 3D geometry

[dmassegur]

Hi,
I have tried your MeshGradientPy package as it seems more useful than Pyvista, especially to integrate with ML packages. Impressive!
However, I have made a comparison of the gradients against Pyvista and I got wrong results on my test case. See picture attached:

Looking at the actual field, I believe pyvista being more correct than MeshGradientPy. The results are righton the upper surface of the wing (similar to XY plane) but wrong on the cylindrical fuselage. Also, your package gave me all 0 on the z gradient.

Are you able to provide any clue how to fix this, please?

Regards.

[DonsetPG]

Hey @dmassegur thanks for the message!
Regarding the z coordinates, I believe we did not implement the computation in 3 dimensions (at the time we were only experimenting in 2D).

Regarding the difference with PyVista, if I remember correctly we only ran test versus our internal solver, not pyvista. Would it be possible to get the test case you are using to see how we could improve?
Best

[dmassegur]

Hi @DonsetPG thanks for your reply!
Sorry for not getting back to you sooner but I was busy on other duties.
I'd be keen to send you my vtk file for you to investigate, if this is ok. How can I send it to you directly?
Many thanks.

[dmassegur]

Hi @DonsetPG
I wonder if you could assist with the gradient calcs on my test case?
Many thanks!

[dmassegur]

Dear @DonsetPG
How are you?
I wonder if we could work in order to make your code work for 3D surfaces. As I acknowledge your grad calculation is very useful to embed to pytorch ML models.