Surrogate Modeling for Active Drag Reduction in Turbulent Boundary Layer Flows Using Multitask Gaussian Process Regression

ABSTRACT Single‐ and multitask Gaussian process regression (GPR), termed sGPR and mGPR, are investigated as surrogate models to guide the design of actuation parameters for active drag reduction techniques in turbulent boundary layer (TBL) flows. The turbulent flow over a flat plate, which serves as a first step approximation to more realistic scenarios such as airfoil applications, is subjected to spanwise traveling transversal sinusoidal surface waves and simulated by wall‐resolved large‐eddy simulations (LESs). These simulation data are used to model the dependence of the two objective functions, that is, relative drag reduction and relative net power savings, on the actuation parameters, that is, the optimization variables. This work extends previous studies, which focus on single‐task surrogate modeling of relative drag reduction only, to enhance the prediction quality of relative net power savings to identify promising regions of actuation parameters by learning and leveraging mGPR‐based correlations between the two objective functions, that is, relative drag reduction and relative net power savings.