Reduced-Order Model for Active Aerodynamics Prediction and Wing Control

Abstract

Active aerodynamics is an emerging technology that alters vehicle subsystems in real time to optimize as track conditions change. Repeated solver or computational fluid dynamics (CFD) evaluations inside a control loop are too slow for continuous optimization, often forcing fixed geometries and limiting performance. This paper proposes a physics-informed reduced-order model (PI-ROM) that replaces inefficient aerodynamic evaluations with fast, accurate coefficient predictions for active aero rear wing control. A training/testing dataset was generated using XFOIL CFD sweeps for NACA airfoils. The PI-ROM combines thin-airfoil and drag-polar physics as baselines with a neural network residual to improve accuracy for non-ideal conditions. The framework underwent three iterations: calibrating weights and preventing negative drag; improving drag prediction by leveraging the predicted lift within the residual; and revising the controller from curvature-only changes to an objective-driven strategy that prioritizes drag minimization on straights and downforce maximization in corners. The PI-ROM informed a closed-loop controller which was evaluated against a fixed-wing baseline in a vehicle dynamics track simulation described by ordinary differential equations. The PI-ROM achieved a 26.58x speedup relative to XFOIL prediction time while maintaining accuracy (Mean Absolute Error: 0.0298 for CL, 0.0034 for CD; Normalized Mean Absolute Percentage Error: 1.37% for CL and 8.73% for CD). In simulation, active control reduced lap time by 1.2 seconds and saved 5.61 kJ more energy versus a fixed wing on the same track. Future work would involve 3D wing implementation and hyper-realistic vehicle dynamics to match real-world motorsports.