Drag reduction of a circular cylinder using a flexible splitter controlled by a reinforcement learning algorithm

We numerically investigate the drag reduction mechanism of a circular cylinder using an actively heaving flexible splitter controlled by its heaving frequency. An immersed boundary method (IBM) is used to capture the fluid–structure interaction. The heaving motion is imposed at the splitter’s leading-edge, and its posterior portions passively oscillate due to its flexibility. The vortices generated by the splitter interact with the shear layer shed by the cylinder in either a constructive or destructive mode, depending on the splitter’s heaving frequency. These two interaction modes have opposite effects on the drag force acting on the cylinder: the constructive mode increases the drag, whereas the destructive mode reduces it. To identify the optimal heaving frequency, we perform reinforcement learning based on a Q-learning algorithm. The Q-learning algorithm is found to identify the splitter’s optimal heaving frequency across different Reynolds numbers, demonstrating its potential as an adaptive controller that can select the optimal behavior under diverse flow conditions.