You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Acoustics-based Active Control of Unsteady Flow Dynamics using Reinforcement Learning Driven Synthetic Jets
Noise generated by vortices is a sensible measure of the strength of wakes generated in a flow field. This paper presents an innovative approach to actively control wakes and noise generated in a flow past a cylinder using deep reinforcement learning (DRL) as a continuously learning active control algorithm with noise levels recorded using acoustic level around the wake region. The two primary objectives of this study is to investigate the feasibility and effectiveness of use of acoustic level as controlling parameter and employing DRL algorithms to optimize the control of flow dynamics. A hydrophone array is utilised to capture the wake pattern along the flow field downstream of a circular cylinder in the form of acoustic signals. The collected data are used to create a real-time feedback loop for a DRL agent to adjust jet actuators strategically placed on the cylinder's surface. This approach enables the agent to learn and adapt its control strategy based on the observed acoustic feedback, resulting in a closed-loop control system. We demonstrate that the DRL-based flow control strategy can effectively reduce wake amplitude and the noise generated. The noise level reduces by an appreciable 6.9% and 9.5% in the given setting for two control configurations. Similarly, the drag coefficient reduces by a remarkable 15.9% and 23.8% respectively. Specifically, it reduces the oscillation amplitude in drag and noise. The proposed method offers promising results in terms of reducing flow-induced vibration. The paper highlights the potential for using DRL, jets, and hydrophone arrays in active flow control, opening new avenues for optimizing flow control in practical engineering applications.
Environment description:
Mesh:
Drag Coefficient through different stages of controller:
Sound Pressure Level through different stages of controller:
Mean fluctuation Pre(top)/Post(Bottom) Control:
Citation
@misc{ phan2023acousticsbased,
title={Acoustics-based Active Control of Unsteady Flow Dynamics using Reinforcement Learning Driven Synthetic Jets},
author={Khai Phan and Siddharth Rout and Chao-An Lin and Rajeev Jaiman},
year={2023},
eprint={2312.16376},
archivePrefix={arXiv},
primaryClass={physics.flu-dyn}
}
