Active flow control of energy harvester via deep reinforcement learning to enhance energy harvesting with different Reynolds number

Flow-induced vibrations (FIVs) energy harvesters operating under variable inflow conditions often exhibit low and unstable power output. Active flow control (AFC) via deep reinforcement learning (DRL) is employed to amplify FIVs of the circular cylinder in the energy harvester at Re = 100-1000. Different from most existing DRL-based AFC studies that are conducted at a fixed Re and mainly focus on vibration suppression, the present study proposes a DRL-based AFC for variable Re conditions to improve its adaptability to varying inflow velocities. The agent is trained to manipulate four jets mounted on the cylinder to achieve real-time closed loop control and is further evaluated using a fluid-structure-electric interaction model to quantify its effects on cylinder's vibration and power generation. The results show that the proposed AFC significantly intensifies the fluid-induced force: the standard deviations of drag and lift coefficients increase by 136.9% and 34.2% at Re = 300, respectively. Consequently, the net average harvested power increases from 0.071 mW to 9.388 mW. After the agent's control, the lift force contains substantial energy over a wider range of frequency, rather than being concentrated only near the vortex-shedding frequency, which improves the adaptability of the energy harvester under fluctuating marine flow conditions.