Using SolidWorks Meshes with DiffusionNet for Turbine Performance Prediction

[KedSpace]

Hello,

I’m working on a research project to optimise the aerodynamic performance of a hybrid Savonius-Darrieus vertical axis wind turbine under low wind speed with Geometric Deep Learning (GDL). I’ve modelled my geometry in SolidWorks and exported it as STL mesh files.

I would like to use DiffusionNet as a surrogate model as a GDL to predict scalar outputs such as Cp (power coefficient) based on the mesh geometry, trained on CFD simulation data.

Could you please clarify:

1. Can I use .STL meshes exported from SolidWorks directly, or are there any requirements for preprocessing?
2. Are there any specific mesh requirements or preprocessing steps I should follow before using these files with DiffusionNet?
3. For scalar regression (Cp), is it best to pool vertex features globally before output? Or is there a better recommended approach?
4. Can I pass the mesh directly as a face-based or vertex-based graph, and map to a global output?

Thank you for this powerful tool! Your earlier comment in Issue #13 was really helpful, and I’d love to apply this to wind turbine optimisation.

Warm regards,
KedSpace