DIGITAL.CSIC. Repositorio Institucional del CSIC
Publication
Artículo científico (JournalArticle)
Transfer learning-enhanced deep reinforcement learning for aerodynamic airfoil optimization subject to structural constraints
© 2025 Author(s). Published under an exclusive license by AIP Publishing., The main objective of this paper is to introduce a transfer learning-enhanced deep reinforcement learning (DRL) methodology that is able to optimize the geometry of any airfoil based on concomitant aerodynamic and structural integrity criteria. To showcase the method, we aim to maximize the lift-to-drag ratio C L / C D while preserving the structural integrity of the airfoil—as modeled by its maximum thickness—and train the DRL agent using a list of different transfer learning (TL) strategies. The performance of the DRL agent is compared with Particle Swarm Optimization (PSO), a traditional gradient-free optimization method. Results indicate that DRL agents are able to perform purely aerodynamic and hybrid aerodynamic/structural shape optimization that the DRL approach outperforms PSO in terms of computational efficiency and aerodynamic improvement and that the TL-enhanced DRL agent achieves performance comparable to the DRL one, while further saving substantial computational resources., The authors thank useful discussions within the NUMATH group. D.R., E.V., and G.R. acknowledge funding from Project TIFON (No. PLEC2023-010251) funded by MCIN/AEI/10.13039/501100011033, Spain, funding from the European Union (Project HERFUSE) under GA No. 101140567. Views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union or Clean Aviation Joint Undertaking. Neither the European Union nor Clean Aviation JU can be held responsible for them. L.L. acknowledges partial financial support from via Project MISLAND (No. PID2020-114324GB-C22), CSxAI (PID2024-157526NB-I00) and the Maria de Maeztu Program for units of Excellence (Grant No. CEX2021-001164-M), the three of them funded by MICIU/AEI/10.13039/501100011033, and funding from the European Commission Chips Joint Undertaking Project No. 101194363 (NEHIL). G.R. acknowledges partial financial support received by the Grant DeepCFD (Project No. PID2022-137899OB-I00) funded by MICIU/AEI/10.13039/501100011033 and by ERDF, EU. Finally, all authors gratefully acknowledge the Universidad Politécnica de Madrid (www.upm.es) for providing computing resources on Magerit Supercomputer., With funding from the Spanish government through the "María de Maeztu Unit of Excellence" accreditation (CEX2021-001164-M), Peer reviewed
Project:
AEI / Plan Estatal de Investigación Científica y Técnica y de Innovación 2021-2023 PLEC2023-010251
,
EC / HE 101140567
,
AEI / Plan Estatal de Investigación Científica y Técnica y de Innovación 2017-2020 PID2020-114324GB-C22
,
AEI / Plan Estatal de Investigación Científica y Técnica y de Innovación 2021-2023 PID2024-157526NB-I00
,
AEI / HE CEX2021-001164-M
,
EC / Plan Estatal de Investigación Científica y Técnica y de Innovación 2021-2023 101194363
,
AEI PID2022-137899OB-I00
DOI: http://hdl.handle.net/10261/400958, https://api.elsevier.com/content/abstract/scopus_id/105013566803
HANDLE: http://hdl.handle.net/10261/400958, https://api.elsevier.com/content/abstract/scopus_id/105013566803