Bayesian Optimization and DRL for High-Lift Wing Flow Control
A new study on arXiv (2605.11981) looks into active flow control applied to a 30P30N high-lift wing, tested at a Reynolds number of 450,000 and an angle of attack of 23°. The researchers explored two optimization strategies: open-loop Bayesian optimization (BO) and closed-loop deep reinforcement learning (DRL). Both methods aimed to improve aerodynamic performance and minimize stall using synthetic jets on various wing components. The results from the uncontrolled setup matched previous research. BO was effective, achieving a 10.9% boost in efficiency and a 9.7% decrease in drag while maintaining lift. On the other hand, the DRL approach, which used live flow data, showed only slight improvements in lift and drag, leading to minimal efficiency gains.
Key facts
- Study uses wall-resolved LES on 30P30N high-lift wing
- Reynolds number Re_c = 450,000
- Angle of attack α = 23°
- Compares Bayesian optimization and deep reinforcement learning
- BO achieved +10.9% efficiency via -9.7% drag reduction
- DRL showed minor improvements with negligible efficiency gain
- Synthetic jets placed on slat, main, and flap elements
- Uncontrolled configuration validated against literature
Entities
Institutions
- arXiv