Title:Prediction of Intermittent Fluctuations from Surface Pressure Measurements on a Turbulent Airfoil

Abstract:This work studies the effectiveness of several machine learning techniques for predicting extreme events occurring in the flow around an airfoil at low Reynolds. For certain Reynolds numbers the aerodynamic forces exhibit intermittent fluctuations caused by changes in the behavior of vortices in the airfoil wake. Such events are prototypical of the unsteady behavior observed in airfoils at low Reynolds and their prediction is extremely challenging due to their intermittency and the chaotic nature of the flow. We seek to forecast these fluctuations in advance of their occurrence by a specified length of time. We assume knowledge only of the pressure at a discrete set of points on the surface of the airfoil, as well as offline knowledge of the state of the flow. Methods include direct prediction from historical pressure measurements, flow reconstruction followed by forward integration using a full order solver, and data-driven dynamic models in various low dimensional quantities. Methods are compared using several criteria tailored for extreme event prediction. We show that methods using data-driven models of low order dynamic variables outperform those without dynamic models and that unlike previous works, low dimensional initializations do not accurately predict observables with extreme events such as drag.