An experimental application of machine-learning methods to the active flow control of a circular cylinder wake via synthetic jets

This study reports on the experimental application of Machine Learning (ML) methods for the optimization of the open-loop Active Flow Control (AFC) via the Synthetic Jet (SJ) technology. Specifically, the AFC of the turbulent wake past a cylinder is addressed with the purposes of drag reduction and vortex shedding suppression. SJ actuators have been proven to be a promising tool in this direction, due to both their effectiveness in separation control and advantageous features, like compactness and small power requirements. In this work, two ML techniques, the Deep Reinforcement Learning and the Genetic Programming Control, are applied to seek for optimal control strategies in terms of the temporal law of the SJ actuator driving signal, thus overcoming the limitations inherent to conventional optimization methods. Experimental tests are carried out in a subsonic wind tunnel on a circular cylinder equipped with a SJ located at the rear stagnation point. An a-posteriori flow field analysis is also performed via particle image velocimetry. Therefore, the present investigation aims at both assessing feasibility of ML approaches for flow control on the experimental side and improving fundamental knowledge on the behavior of SJs and their interaction with the wake past bluff bodies.