AI Framework Bridges Gap Between Aerodynamic Simulations and Wind-Tunnel Data

Researchers have introduced an experimentally grounded data fusion framework designed to align computational fluid dynamics (CFD) surrogate models with physical wind-tunnel experiments. While deep learning models trained on high-fidelity CFD data reproduce numerical outputs efficiently, they often diverge from actual physical observations due to systematic discrepancies.

To bridge this gap without retraining the base Geotransolver surrogate model, the team developed an auxiliary correction network. The primary model was trained on 2,300 high-fidelity CFD simulations of the NASA Common Research Model (CRM) wing-body configuration across Mach numbers 0.70 to 0.85 and angles of attack from 0 to 4 degrees.

By leveraging pressure-sensitive paint (PSP) wind-tunnel measurements, the correction network learned the differences in surface-pressure distributions. On held-out test conditions, the grounded model reduced prediction errors to within 2.3–2.7% of the measured pressure coefficient range, improving fidelity at suction peaks and shock locations while preserving overall computational efficiency.