L-SplineFNO

Moves features from an unstructured engineering mesh onto a latent Cartesian grid with a sparse, learnable tensor-product B-spline scatter, processes them with an FNO, and maps back with the analytic spline. Design and operating parameters condition the model, so it maps geometry and parameters to the steady field. Because the decoder is a spline, spatial derivatives and guaranteed bounds on the prediction come in closed form.

Share on LinkedIn, X or Bluesky, or copy the link or the citation.