CoPiT: Conditional Position-induced Transformer for High-Fidelity Airfoil Flow Prediction

Airfoil aerodynamic optimization is critical for improving lift-to-drag performance, reducing fuel consumption, and mitigating environmental impact in aerospace engineering. Recent advances in data-driven surrogate modeling offer efficient alternatives to costly CFD simulations. In this work, we introduce CoPiT (Conditional Position-induced Transformer), a novel neural operator that augments the Position-induced Transformer (PiT) with Feature-wise Linear Modulation (FiLM) layers to incorporate global conditioning information. Leveraging the high-fidelity AirfRANS dataset generated from steady-state RANS simulations over parameterized NACA airfoils, CoPiT is trained to predict velocity, pressure, and turbulent kinematic viscosity fields. Numerical experiments across full-data, scarce-data, Reynolds-number extrapolation, and angle-of-attack extrapolation tasks demonstrate that CoPiT consistently achieves lower prediction errors and stronger correlations with ground-truth quantities compared with state-of-the-art baselines (including Transolver). These results highlight the effectiveness of CoPiT for aerodynamic surrogate modeling and provide a solid foundation for subsequent integration into automated airfoil shape optimization frameworks.