Airfoil Aerodynamic Design & CFD Surrogate

Background

Airfoil aerodynamic design is central to aerospace engineering. Traditional CFD (RANS/LES) takes hours to days per run, blocking large-scale design-space exploration. Data-driven surrogates compress a single evaluation to milliseconds, enabling real-time scanning.

Approach

Dataset

• Parametric geometry of NACA 4/6-digit airfoils
• Batch CFD (OpenFOAM / SU2) at various angles of attack and Mach numbers
• Output: pressure distribution, Cp curves, lift/drag coefficients

Surrogate methods

• FNO (Fourier Neural Operator): excellent on airfoil-space mapping
• DeepONet: suited to multiphysics-parameter scenarios
• CNN-based: suited to structured grids

Key results

• RANS (CFD): 2-4 h per evaluation
• FNO: ~10 ms, Cd error < 5%
• DeepONet: ~5 ms, Cd error < 8%
• CNN: ~2 ms, Cd error < 10%

Use cases

• Fast airfoil screening (100+ candidates in early design)
• Aerodynamic module in Multidisciplinary Design Optimization (MDO)
• Rapid flight-envelope analysis

Resources

• AirfRANS (indexed here) is the starting benchmark for this direction
• NASA NACA 0012 public dataset as a first exercise