Shape Optimization using Geodesic Convolutional Networks

We published on Arxiv a paper about a new approach to Aerodynamic shape optimization. It improves upon previously used Kriging-based surrogate models for CFD optimization. This new contribution uses the Geometric Deep Learning methods that recently appeared in computer vision.

Arxiv Paper.

Showcase

Pressure prediction on TEST set. Pressure prediction on TEST set. Pressure prediction on CARS-TEST set.

Online Pressure and Drag prediction with GCNN.

Optimization Results

Optimization of a foil with control points.

Optimization of a bicycle shell in 3D.

Optimization of a Foil in 3D.

Finding the most aerodynamic shape around a sphere.