Robust transonic flow prediction via physics-informed operator learning and geometry-driven passive flow control

Accurate and efficient prediction of transonic flow fields remains a significant challenge in fan blade geometry optimization using computational fluid dynamics, particularly under changing operating conditions. This paper introduces POLANet, a physics-aware operator learning framework that integrates geometry-driven encoding, multi-head attention mechanisms, and physics-informed loss functions to robustly predict transonic flow fields. Traditional operator learning models struggle to capture complex flow features, including high-pressure regions, shock waves and wake regions. To address this limitation, POLANet integrates adaptive geometry encoding, multi-head attention, and physics-informed loss functions, enabling accurate and robust prediction across complex flow conditions. The proposed framework effectively captures high-gradient regions and wake zones and enhances the ability to generalize across different flow regimes, while maintaining physical consistency through physics-informed constraints embedded in the training loss. Simulation results on diverse flow conditions, including cases under different inlet conditions and geometry-induced operating variation, show that POLANet dramatically reduces the prediction mistakes seen in baseline methods, while baseline methods tend to produce oscillations and spurious multi-wake. Instead of training multiple models for each condition, which is computationally inefficient, lacks generalization, and compromises physical consistency, POLANet learns a unified mapping across diverse conditions, offering a scalable and physically grounded solution. Optimization results show that the proposed framework improves aerodynamic performance, enhancing system stability and efficiency without compromising compression capabilities. The proposed framework advances operator learning methods by introducing a geometry-driven, physics-aware framework for robust transonic field prediction and passive flow control.