UPCommons. Portal del coneixement obert de la UPC
Publication
Contribución de congreso (ConferenceOutput)
Deep reinforcement learning strategies for optimizing flow control in wings
A computational fluid dynamics (CFD) framework to combine deep reinforcement learning (DRL) with active flow control (AFC) simulations is presented. This framework is first validated in a canonical geometry, the flow past a cylinder at REYNOLDS number 𝑅𝑒 = 100, achieving a drag reduction of roughly 10%. Then, the methodology is applied to a more complex case in order to exploit the DRL capabilities to control massive separations of turbulent flows. To do this, the flow past the SD7003 wing at 𝑅𝑒 = 60,000 and angle of attack 𝐴𝑜𝐴 = 14◦ is analyzed, where the DRL control achieves an increase in the aerodynamic efficiency of approximately 157%., This research has received financial support from the Ministerio de Ciencia e Innovación of Spain (PID2020116937RB-C21 and PID2020-116937RB-C22).Simulations were conducted with the assistance of the Red Española de Supercomputación, who granted computational resources to the HPC facility of MareNostrum 5, at the Barcelona Supercomputing Center (IM-2024-2-0004). Ricard Montalà work is funded by a FI-SDUR grant (2022 FISDU 00066) from the Departament de Recerca i Universitats de la Generalitat de Catalunya. Oriol Lehmkuhl has been partially supported by a Ramón y Cajal postdoctoral contract (RYC2018-025949-I). The authors also acknowledge the support of the Departament de Recerca i Universitats de la Generalitat de Catalunya through the research group Largescale Computational Fluid Dynamics (2021 SGR 00902) and the Turbulence and Aerodynamics Research Group (2021 SGR 01051)., Peer Reviewed