Active flow control in simulations of fluid flows based on deep reinforcement learning
Uloženo v:
| Název: | Active flow control in simulations of fluid flows based on deep reinforcement learning |
|---|---|
| Autoři: | Darshan Thummar |
| Informace o vydavateli: | Zenodo |
| Rok vydání: | 2021 |
| Sbírka: | Zenodo |
| Témata: | Active flow control, Flow around 2D cylinder, open-loop and closed-loop control, Deep reinforcement learning, Proximal policy optimization (PPO) algorithm |
| Popis: | For active flow control, flow around a 2D cylinder is considered a generic example. The von kármán vortices are generated for flow around the 2D cylinder. These vortices impose periodic drag and lift on the cylinder. The objective of this study is to reduce drag, fluctuation of and drag, and lift in order to stabilize the cylinder. Hence, open-loop control and closed-loop control strategies are implemented to control drag and lift. The optimal open-loop control strategy is achieved by parameter study. The optimal closed-loop control strategy is achieved by deep reinforcement learning using the PPO algorithm. |
| Druh dokumentu: | report |
| Jazyk: | English |
| Relation: | https://zenodo.org/records/4897961; oai:zenodo.org:4897961; https://doi.org/10.5281/zenodo.4897961 |
| DOI: | 10.5281/zenodo.4897961 |
| Dostupnost: | https://doi.org/10.5281/zenodo.4897961 https://zenodo.org/records/4897961 |
| Rights: | Creative Commons Attribution 4.0 International ; cc-by-4.0 ; https://creativecommons.org/licenses/by/4.0/legalcode |
| Přístupové číslo: | edsbas.D6E3503C |
| Databáze: | BASE |
Nájsť tento článok vo Web of Science | Abstrakt: | For active flow control, flow around a 2D cylinder is considered a generic example. The von kármán vortices are generated for flow around the 2D cylinder. These vortices impose periodic drag and lift on the cylinder. The objective of this study is to reduce drag, fluctuation of and drag, and lift in order to stabilize the cylinder. Hence, open-loop control and closed-loop control strategies are implemented to control drag and lift. The optimal open-loop control strategy is achieved by parameter study. The optimal closed-loop control strategy is achieved by deep reinforcement learning using the PPO algorithm. |
|---|---|
| DOI: | 10.5281/zenodo.4897961 |