Multi-object aerodynamic design optimization using deep reinforcement learning
Aerodynamic design optimization is a key aspect in aircraft design. The further evolution of advanced aircraft derivatives requires a powerful optimization toolbox. Reinforcement learning (RL) is a powerful optimization tool but has rarely been utilized in the aerodynamic design. It can potentially...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
AIP Publishing LLC
2021-08-01
|
| Series: | AIP Advances |
| Online Access: | http://dx.doi.org/10.1063/5.0058088 |
| id |
doaj-2d06a9c65b5a46279493317440c2820f |
|---|---|
| record_format |
Article |
| spelling |
doaj-2d06a9c65b5a46279493317440c2820f2021-09-03T11:18:12ZengAIP Publishing LLCAIP Advances2158-32262021-08-01118085311085311-910.1063/5.0058088Multi-object aerodynamic design optimization using deep reinforcement learningXinyu Hui0Hui Wang1Wenqiang Li2Junqiang Bai3Fei Qin4Guoqiang He5School of Aeronautics, Northwestern Polytechnical University, Xi’an 710072, ChinaSchool of Aeronautics, Northwestern Polytechnical University, Xi’an 710072, ChinaKey Laboratory of Science and Technology on Combustion, Internal Flow and Thermal-Structure, Northwestern Polytechnical University, Xi’an 710072, ChinaSchool of Aeronautics, Northwestern Polytechnical University, Xi’an 710072, ChinaKey Laboratory of Science and Technology on Combustion, Internal Flow and Thermal-Structure, Northwestern Polytechnical University, Xi’an 710072, ChinaKey Laboratory of Science and Technology on Combustion, Internal Flow and Thermal-Structure, Northwestern Polytechnical University, Xi’an 710072, ChinaAerodynamic design optimization is a key aspect in aircraft design. The further evolution of advanced aircraft derivatives requires a powerful optimization toolbox. Reinforcement learning (RL) is a powerful optimization tool but has rarely been utilized in the aerodynamic design. It can potentially obtain results similar to those of a human designer, by accumulating experience from training. In this work, a popular RL method called proximal policy optimization (PPO) is proposed to investigate multi-object aerodynamic design optimization. By observing the aerodynamic performances of different airfoils, the PPO updates a reasonable policy to generate the optimal airfoils in a single step. In a Pareto optimization problem with constraints, the PPO requires only 15% of the computational time of the non-dominated sorted genetic algorithm (II) to achieve the same accuracy. The results from testing show that the agent learns a policy that can achieve ∼4.3%–10.1% improvements of the aerodynamic performance compared with the results of baseline.http://dx.doi.org/10.1063/5.0058088 |
| collection |
DOAJ |
| language |
English |
| format |
Article |
| sources |
DOAJ |
| author |
Xinyu Hui Hui Wang Wenqiang Li Junqiang Bai Fei Qin Guoqiang He |
| spellingShingle |
Xinyu Hui Hui Wang Wenqiang Li Junqiang Bai Fei Qin Guoqiang He Multi-object aerodynamic design optimization using deep reinforcement learning AIP Advances |
| author_facet |
Xinyu Hui Hui Wang Wenqiang Li Junqiang Bai Fei Qin Guoqiang He |
| author_sort |
Xinyu Hui |
| title |
Multi-object aerodynamic design optimization using deep reinforcement learning |
| title_short |
Multi-object aerodynamic design optimization using deep reinforcement learning |
| title_full |
Multi-object aerodynamic design optimization using deep reinforcement learning |
| title_fullStr |
Multi-object aerodynamic design optimization using deep reinforcement learning |
| title_full_unstemmed |
Multi-object aerodynamic design optimization using deep reinforcement learning |
| title_sort |
multi-object aerodynamic design optimization using deep reinforcement learning |
| publisher |
AIP Publishing LLC |
| series |
AIP Advances |
| issn |
2158-3226 |
| publishDate |
2021-08-01 |
| description |
Aerodynamic design optimization is a key aspect in aircraft design. The further evolution of advanced aircraft derivatives requires a powerful optimization toolbox. Reinforcement learning (RL) is a powerful optimization tool but has rarely been utilized in the aerodynamic design. It can potentially obtain results similar to those of a human designer, by accumulating experience from training. In this work, a popular RL method called proximal policy optimization (PPO) is proposed to investigate multi-object aerodynamic design optimization. By observing the aerodynamic performances of different airfoils, the PPO updates a reasonable policy to generate the optimal airfoils in a single step. In a Pareto optimization problem with constraints, the PPO requires only 15% of the computational time of the non-dominated sorted genetic algorithm (II) to achieve the same accuracy. The results from testing show that the agent learns a policy that can achieve ∼4.3%–10.1% improvements of the aerodynamic performance compared with the results of baseline. |
| url |
http://dx.doi.org/10.1063/5.0058088 |
| work_keys_str_mv |
AT xinyuhui multiobjectaerodynamicdesignoptimizationusingdeepreinforcementlearning AT huiwang multiobjectaerodynamicdesignoptimizationusingdeepreinforcementlearning AT wenqiangli multiobjectaerodynamicdesignoptimizationusingdeepreinforcementlearning AT junqiangbai multiobjectaerodynamicdesignoptimizationusingdeepreinforcementlearning AT feiqin multiobjectaerodynamicdesignoptimizationusingdeepreinforcementlearning AT guoqianghe multiobjectaerodynamicdesignoptimizationusingdeepreinforcementlearning |
| _version_ |
1717817416401027072 |