Airfoil design at the speed of inference.
AeroSurrogate Studio replaces hours of RANS simulation with a MeshGraphNet surrogate trained on AirfRANS. Shape an airfoil, read the full flow field and aerodynamic coefficients right away, then let an optimizer search thousands of designs.
From five numbers to a full RANS flow field
The surrogate learned the mapping from mesh geometry and inflow to the converged k-ω SST solution. For a new design it needs only a mesh, and one comes from morphing the closest training case.
- 01
Parameterize
NACA 4- or 5-digit parameters, angle of attack and inflow speed define a design.
- 02
Morph a mesh
The nearest AirfRANS training mesh (~180k cells) is morphed onto the new shape and quality-checked.
- 03
Predict the flow
A MeshGraphNet passes messages across the mesh graph and predicts Ux, Uy, p and νt at every node.
- 04
Integrate forces
Surface pressure and wall shear integrate to Cl, Cd and Cm, with ensemble uncertainty.
- 05
Optimize
Evolutionary search calls the surrogate thousands of times to find better shapes.
A complete aerodynamic design loop
Design Studio
Drag NACA parameters and see pressure, velocity and turbulence fields update, with Cl, Cd and L/D.
OpenPolar analysis
Lift curves, drag polars, L/D and pitching moment across angle of attack, overlaid for several designs.
OpenOptimizer
Genetic algorithm and NSGA-II over the design space, with constraints and live convergence.
OpenCompare
Up to four designs side by side: shapes, Cp distributions and coefficient deltas.
OpenTrust signals
Ensemble uncertainty, training-envelope checks and mesh-quality reports on every prediction.
OpenFull-field output
Not just coefficients: the whole flow field, to see where separation and suction peaks happen.
OpenA surrogate that tells you when not to trust it
Learned models are fast but can be confidently wrong outside their training data. Every prediction here carries the signals an engineer needs to judge it.
- Ensemble uncertaintySeveral independently trained models; their spread becomes the ±σ on each coefficient, and optimization can penalize it.
- Training-envelope checksEach input is checked against the training distribution, with explicit warnings when a prediction extrapolates.
- Mesh-quality gateMorphed meshes are checked for inverted cells and near-wall resolution before inference.
- Physics constraintsNo-slip is enforced exactly at the wall; physics-residual diagnostics accompany each prediction.
Start designing
Pick a classic NACA section, or start from scratch.