New AI Model Improves Unsteady Aerofoil Flow Prediction

Accurately predicting unsteady separated flows, crucial for aerodynamic design and control, is a significant challenge due to the complex, non-linear dynamics of separation and vortex shedding. While high-fidelity Computational Fluid Dynamics (CFD) can resolve these mechanisms, its high cost limits its use in iterative design processes. Standard field-level surrogate models often fall short because they don't differentiate between flow regions that contribute most to aerodynamic loads. To address this, researchers developed VATO (Vortex-Force-Aware Transformer Operator), which integrates the Vortex Force Map (VFM) method into a geometry-aware neural operator. VATO employs two mechanisms: VATO-S adds training-only supervision of the local VFM force-contribution field, improving accuracy without increasing inference cost. VATO-A uses VFM contribution and sensitivity fields to prioritize force-relevant locations for residual cross-attention, further enhancing prediction. Evaluated on unsteady CFD data for aerofoils, VATO-S and VATO-A consistently reduced errors in velocity, pressure, and vorticity. VATO-A, in particular, showed substantial improvements in lift and drag error, and maintained significant vorticity error reduction even for prediction lead times extending 50% beyond the training range. This demonstrates VATO's ability to improve both flow-field prediction and critical aerodynamic functional accuracy.