Artificial Intelligence Control of Turbulent Boundary Layer Based on Distributed Synthetic Jets

An artificial intelligence (AI) control system is developed to manipulate a turbulent boundary layer over a flat plate at a momentum-thickness-based Reynolds number Re_θ of 1450. The system comprises 11 distributed synthetic minijets (actuation unit), 9 wall wires (sensing unit), and a genetic-algorithm-based control unit for the unsupervised learning of optimal forcing. While the exit velocities and excitation frequencies of the synthetic jets are made the same, optimally determined from conventional uniform forcing, the optimal phase shifts between the jets are obtained from the learning process of the AI control experiments so that the cost function or spanwise averaged friction drag estimated from the 9 wall-wires is minimized. Two distinct optimal solutions, albeit yielding the same cost function, are found after the learning process is converged, both yielding a drag reduction (DR) substantially exceeding conventional uniform forcing. The DR mechanism is discussed. This study points to the great potential of AI in finding the optimal solution even when the control parameter number is large.