Forecasting the lift of a randomly maneuvering airfoil under dynamic stall conditions, Re ∼ 10⁵

Abstract

Dynamic stall is the abrupt flow separation from airfoils rapidly changing their orientation. This phenomenon, characterized by a delayed stall followed by a sharp drop in lift, has prompted efforts to prevent or delay it. This study aims to predict the lift of an airfoil randomly maneuvering under dynamic stall conditions by utilizing sparse surface pressure measurements, which we believe can maximize the effectiveness of various dynamic stall suppression techniques. Using data from large eddy simulations, we demonstrate that a long short-term memory network, fed with raw surface pressures, delivers accurate predictions. Also, a new method introduced here, IdDM, conclusively links the characteristic frequency range of pressure fluctuations that emerges during the dynamic stall to the chord-lengthscale vortex dynamics. However, further analysis suggests that the forecast predominantly relies on the lower frequency components tied to the airfoil motion, possibly because the vortex dynamics are dependent on and sensitive to the airfoil motion. Meanwhile, specific sensor locations are proven to be more informative than others in this random, unsteady flow, and we show that optimal sensor placement can be quickly determined using mutual information alone. It reveals that two pressure sensors positioned near the leading edge, one on each side of the airfoil, capture most of the information needed to predict lift. The lift can be predicted with sparse sensors because surface pressures are strongly correlated across the airfoil, with large-scale flow structures dominating the forces.