Training with multiple geometries

[LimitingFactor]

I would like to train a more general model that can predict flow over an arbitrary 2D airfoil. I have successfully trained models on a single geometry and a single geometry with parameterized boundary conditions.

Is it possible to train a model on a number of different nonparametric geometries?

In other data driven frameworks it is a matter of passing in a large number of example geometries. From the documentation it appears that modulus either requires the geometry to be parameterized or the model to be retrained using transfer learning for each geometry.

Any advice would be welcome. Thank you

[patterson]

Hi @LimitingFactor

If I’m using it correctly, which I’m still not sure about, then the answer is yes! They have an example in the geometry folder of their examples:

Gitlab Parameterized Tesselated Example

This appears to work for STL files and based on a quick look at their code for discrete geometry I think it would work for 2D geometry constructions. Disclaimer: I barely know what I’m doing with Modulus and python isn’t my first language so I could be wrong.