Pasar al contenido principal

HACIA UNA PRUEBA DE CONCEPTO PARA LA AVIACION SOSTENIBLE DE TECNICAS DE APRENDIZAJE PROFUNDO POR REFUERZO: DESDE ESCENARIOS ACADEMICOS HASTA SITUACIONES DEL MUNDO REAL.

PID2023-150408OB-C22

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal para Impulsar la Investigación Científico-Técnica y su Transferencia
Subprograma Subprograma Estatal de Generación de Conocimiento
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2023
Plan Estatal 2021-2023 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario UNIVERSITAT POLITECNICA DE CATALUNYA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 14
Encontrada(s) 1 página(s)

UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

Discovering flow separation control strategies in 3d wings via deep reinforcement learning

  • Montalà Sales, Ricard|||0009-0000-7911-7102
  • Font García, Bernat
  • Suárez Morales, Pol
  • Rabault, Jean
  • Lehmkuhl Barba, Oriol|||0000-0002-2670-1871
  • Vinuesa Motilva, Ricardo
  • Rodríguez Pérez, Ivette María|||0000-0002-3749-277X
In this work, deep reinforcement learning (DRL) is applied to active flow control (AFC) over a threedimensional SD7003 wing at a Reynolds number of Re = 60,000 and angle of attack of AoA = 14◦. In the uncontrolled baseline case, the flow exhibits massive separation and a fully turbulent wake. Using a GPU-accelerated CFD solver and multi-agent training, DRL discovers control strategies that enhance lift (79%), reduce drag (65%), and improve aerodynamic efficiency (408%). Flow visualizations confirm reattachment of the separated shear layer, demonstrating the potential of DRL for complex and turbulent flows., This research was partially funded by the Ministerio de Ciencia e Innovación of Spain (PID2023- 150408OB-C21/C22). Simulations were supported by the Red Española de Supercomputación (IM-2024-2-0004 and IM-2024-3-0005) and EuroHPC JU (EHPCREG-2024R01-038), granting access to MareNostrum V at the Barcelona Supercomputing Center. We also acknowledge AGAUR for supporting the LS/CFD (2021 SGR 00902) and TUAREG (2021 SGR 01051) research groups. R. M. thanks AGAUR for the FISDUR grant (2022 FISDU 00066) and R. V. acknowledges financial support from ERC grant no. 2021- CoG-101043998, DEEPCONTROL., Peer Reviewed, 9 - Indústria, Innovació i Infraestructura, 13 - Acció per al Clima
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

On the integration of geometry agnostic variational-autoencoders into large-scale SVD based models

  • Eiximeno Franch, Benet
  • Miró Jané, Arnau|||0000-0002-2772-6050
  • Kutz, Nathan
  • Rodríguez Pérez, Ivette María|||0000-0002-3749-277X
  • Lehmkuhl, Oriol
This manuscript introduces GAVI (Geometry Agnostic Variational-autoencoder Integration), a scalable technique for model order reduction of terabyte-level state spaces. Such methodology is agnostic to the spatial location of the data points, hence, it can handle any complex geometry regardless of the grid type in which it is represented. The dimensionality reduction of GAVI is performed on two separate steps. First, a parallel QR factorization is applied to the snapshot matrix to decompose, then a variational autoencoder learns a latent representation of the reduced R matrix. The final step can be done at laptop level regardless of the size of the original data. The methodology is used to compress the streamwise velocity component of the flow around a circular cylinder at 𝑅𝑒𝐷 = 𝑈∞𝐷∕𝜈 = 104 , the pressure coefficient of the flow around the Windsor body at 𝑅𝑒𝐿 = 𝑈∞𝐿∕𝜈 = 2.9 × 106 and the streamwise velocity of the realistic urban flow in the Zona Universitària neighborhood located in Barcelona. The latter example is the most demanding one as the 1032 snapshots, which are represented on an unstructured grid of 335 million points, have a total size of 2.05 Tb. GAVI can compress the full dataset into 6 latent vectors that recover up to the 95.29% of the energy., The research leading to this work has been partially funded by the project TIFON with reference PLEC2023-010251/ AEI/10.13039/ 501100011033 and by the Ministerio de Ciencia e Innovación of Spain (PID2023-150408OB-C21/C22). B. Eiximeno’s work was funded by a contract from the Subprograma de Ayudas Predoctorales given by the Ministerio de Ciencia e Innovación (PRE2021-096927). Oriol Lehmkuhl has been partially supported by a Ramon y Cajal postdoctoral contract (Ref: RYC2018-025949-I). The authors acknowledge the support of Departament de Recerca i Universitats de la Generalitat de Catalunya to the Research Group Large-scale Computational Fluid Dynamics (Code: 2021 SGR 00902) and the Turbulence and Aerodynamics Research Group (Code: 2021 SGR 01051). Nathan Kutz acknowledges the NSF AI Institute in Dynamics System., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

On the effects of the spanwise length on the flow past an airfoil in stall at moderate Reynolds numbers

  • Montalà Sales, Ricard|||0009-0000-7911-7102
  • Lehmkuhl Barba, Oriol|||0000-0002-2670-1871
  • Rodríguez Pérez, Ivette María|||0000-0002-3749-277X
This study investigates the influence of the spanwise periodic domain length Lz on the accuracy of numerical simulations of flow past an SD7003 airfoil (Selig–Donovan series) in post-stall conditions at an angle of attack a =14º and Reynolds number Rec = 60000. Large eddy simulations and direct numerical simulations are performed for spanwise extents ranging from Lz/c =0.2 to 2.0 to evaluate their impact on flow dynamics, aerodynamic forces, and wake characteristics. The results demonstrate a strong sensitivity of key flow features and aerodynamic coefficients to the chosen spanwise length. Domains that are too short artificially constrain three-dimensional flow structures, leading to significant overprediction of drag, lift, and unsteady lift fluctuations. This behavior is attributed to trailing-edge flow structures being forced into an unrealistically coherent spanwise configuration, which inhibits their natural three-dimensional breakdown in the wake. Analyses based on two-point correlations and flow statistics indicate that a minimum spanwise length of Lz/c ≥ 0:8 is necessary to capture the largest turbulent scales, reduce the influence of periodic boundary conditions, and obtain accurate predictions of unsteady aerodynamic forces. The findings further suggest that previous high-fidelity studies, which used Lz/c = 0.2 for computational efficiency, were insufficient to represent the full three-dimensional complexity of post-stall flow., This research has received financial support from the Ministerio de Ciencia e Innovación of Spain (Nos. PID2023-150408OB-C21 and PID2023-150408OB-C22). The authors also acknowledge the support of the Agència d’Ajuts Universitaris i Recerca of the Generalitat de Catalunya to the research groups Large-scale Computational Fluid Dynamics (No. 2021 SGR 00902) and Turbulence and Aerodynamics Research Group (No. 2021 SGR 01051). Furthermore, simulations were conducted with the assistance of the Red Española de Supercomputacióon (RES) and the European High Performance Computing Joint Undertaking (EuroHPC), who granted us computational resources at the HPC facilities of MareNostrum IV, at Barcelona Supercomputing Center (Nos. IM-2024-2-0004, IM-2024-3- 0005, and EHPC-REG-2024R01-038). Finally, R. Montalà would like to express gratitude to the Departament de Recerca i Universitats de la Generalitat de Catalunya for awarding the FI-SDUR grant (No. 2022 FISDU 00066), which supports his doctoral studies., Peer Reviewed, 9 - Indústria, Innovació i Infraestructura, 7 - Energia Assequible i No Contaminant
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

High-lift wing separation control via Bayesian optimization and deep reinforcement learning

  • Montalà Sales, Ricard|||0009-0000-7911-7102
  • Font García, Bernat
  • Lehmkuhl Barba, Oriol|||0000-0002-2670-1871
  • Vinuesa Motilva, Ricardo
  • Rodríguez Pérez, Ivette María|||0000-0002-3749-277X
This study investigates active flow control (AFC) of a 30P30N high-lift wing at a Reynolds number Rec = 450,000 and angle of attack α = 23◦ using wallresolved large-eddy simulations (LES). Two optimization strategies are explored: openloop Bayesian optimization (BO) and closed-loop deep reinforcement learning (DRL), both targeting the mitigation of stall and the improvement of aerodynamic efficiency via synthetic jets on the slat, main, and flap elements. The uncontrolled configuration was validated against literature data, confirming the reliability of the LES setup. The BOframework successfully identified steady jet velocities that increased efficiency by +10.9% through a-9.7% drag reduction while maintaining lift. In contrast, the DRL agent, despite leveraging instantaneous flow information from distributed sensors, achieved only minor improvements in lift and drag, with negligible efficiency gain. Training analysis indicated that the penalty-dominated reward constrained exploration. These results highlight the need for carefully designed rewards and computational acceleration strategies in DRL-based flow control at high Reynolds numbers., This work was supported in part by the European Research Council under the Caust grant ERC-AdG101018287 during the 2025 Madrid Turbulence Workshop. This research was partially funded by the Ministerio de Ciencia e Innovación of Spain (PID2023-150408OB-C21/C22). Work has also been co-funded by the European High Performance Computing Joint Undertaking (EuroHPC JU) and Germany, Italy, Slovenia, Spain, Sweden, and France under grant agreement No 101092621; and under grant PCI2022-135085-2 funded by MICIU/AEI/10.13039/ 501100011033 and by the “European Union NextGenerationEU/PRTR”. Simulations were supported by the Red Española de Supercomputación (IM-2024-2-0004 and IM-2024-3-0005) and EuroHPC JU (EHPC-REG-2024R01-038), granting access to MN5 at the Barcelona Supercomputing Center. We also acknowledge AGAUR for supporting the LS/CFD (2021 SGR 00902) and TUAREG (2021 SGR 01051) research groups. R. M. thanks AGAUR for the FI-SDUR grant (2022 FISDU 00066), and R. V. acknowledges financial support from ERC grant no. 2021-CoG-101043998, DEEPCONTROL. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

Deep reinforcement learning for active flow control around a three-dimensional flow-separated wing at Re = 1, 000

  • Montalà Sales, Ricard|||0009-0000-7911-7102
  • Font García, Bernat
  • Suárez Morales, Pol
  • Rabault, Jean
  • Lehmkuhl Barba, Oriol|||0000-0002-2670-1871
  • Vinuesa Motilva, Ricardo
  • Rodríguez Pérez, Ivette María|||0000-0002-3749-277X
This study explores the use of deep reinforcement learning (DRL) for active flow control (AFC) to reduce flow separation on wings at high angles of attack. Concretely, here the DRL agent controls the flow over the three-dimensional NACA0012 wing section at the Reynolds number Re = 1, 000 and angle of attack AoA= 20◦, autonomously identifying optimal control actions through real-time flow data and a reward function focused on improving aerodynamic performance. The framework integrates the GPUaccelerated computational fluid dynamics (CFD) solver SOD2D with the TF-Agents DRL library via a Redis in-memory database, enabling rapid training. This work builds on previous DRL flow-control studies, demonstrating DRL’s potential to address complex aerodynamic challenges and push the boundaries of traditional AFC methods., This research has received financial support from the Ministerio de Ciencia e Innovacion of Spain (PID2023-150408OB-C21 and PID2023-150408OB-C22). Simulations were conducted with the assistance of the Red Española de Supercomputación (RES) and the EuroHPC JU, who granted us computational resources at the HPC facilities of MareNostrum V at Barcelona Supercomputing Cente (IM-2024-2-0004 and EHPC-REG-2024R01- 038, respectively). Authors also extend their gratitute to the Agència de Gestió d’Ajuts Universitaris i de Recerca (AGAUR) for supporting the research group Large-scale Computational Fluid Dynamics (2021 SGR 00902) and the Turbulence and Aerodynamics Research Group (2021 SGR 01051). Ricard Montalà express also gratitude to AGAUR for awarding the FI-SDUR grant (2022 FISDU 00066), which supports his doctoral studies. Finally, Ricardo Vinuesa acknowledges financial support from ERC grant no.2021-CoG101043998, DEEPCONTROL., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Flow separation control of an infinite wing section via multi-agent reinforcement learning

  • Montalà Sales, Ricard|||0009-0000-7911-7102
  • Font García, Bernat
  • Suárez Morales, Pol
  • Rabault, Jean
  • Lehmkuhl Barba, Oriol|||0000-0002-2670-1871
  • Vinuesa Motilva, Ricardo
  • Rodríguez Pérez, Ivette María|||0000-0002-3749-277X
This study investigates the application of deep reinforcement learning (DRL) for active flow control in a three-dimensional NACA0012 wing section with periodic spanwise boundary conditions at low Reynolds number (𝑅𝑒𝑐 =1000) and high angle of attack (AoA =20∘), where the flow exhibits massive separation, strong vortex shedding and a chaotic three-dimensional wake. The baseline configuration is validated against previous numerical studies, showing the accuracy of the computational set-up. Two DRL control policies are trained with different reward functions, leading to fundamentally different control strategies. The DRL set-up uses a multi-agent reinforcement learning framework to coordinate distributed actuators along the wing span. In the first case study, when the reward prioritises drag reduction, agents delay the leading-edge separation and stabilise the wake, reducing drag by 21.1 % and lift fluctuations by 54.1 % while maintaining baseline lift. In the second case, when the reward targets lift enhancement instead, agents synchronise actuations with vortex shedding, forming leading-edge instabilities that increase lift by 42.3 %, albeit at the cost of higher fluctuations and slightly increased drag. A comparison with classical open-loop control further demonstrates that DRL autonomously determines physically meaningful and interpretable strategies consistent with established mechanisms in the literature. To the best of the authors’ knowledge, this is the first study in which a fully three-dimensional computational fluid dynamics simulation has been employed to train a DRL model for active flow control in wings, suggesting promising directions for extending DRL-based strategies to higher Reynolds numbers and more complex wing configurations where prior physical knowledge may be limited., This research was partially funded by the Ministerio de Ciencia e Innovación of Spain (PID2023-150408OB-C21 and PID2023-150408OB-C22). Work has also been co-funded by the European High Performance Computing Joint Undertaking, together with Germany, Italy, Slovenia, Spain, Sweden and France under grant agreement No 101092621; and under grant PCI2022-135085-2, funded by MICIU/AEI/10.13039/501100011033 and by the ‘European Union NextGenerationEU/PRTR’. R. Montalà’s work was supported by a contract from the Agència d’Ajuts Universitaris i Recerca of the Generalitat de Catalunya through a FI-SDUR grant (2022 FISDU 00066). O. Lehmkuhl has been partially supported by a Ramón y Cajal postdoctoral contract (RYC2018-025949-I). The authors also acknowledge the support of the Agència d’Ajuts Universitaris i Recerca of the Generalitat de Catalunya to the research groups Large-scale Computational Fluid Dynamics (2021 SGR 00902) and Turbulence and Aerodynamics Research Group (2021 SGR 01051). R. Vinuesa acknowledges financial support from the European Research Council under grant no. 2021-CoG101043998, DEEPCONTROL. The views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union or the European Research Council., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Capítulo o parte de un libro (BookPart)

Towards active flow control strategies through deep reinforcement learning

  • Montalà Sales, Ricard|||0009-0000-7911-7102
  • Font García, Bernat
  • Suárez Morales, Pol
  • Rabault, Jean
  • Lehmkuhl Barba, Oriol|||0000-0002-2670-1871
  • Vinuesa Motilva, Ricardo
  • Rodríguez Pérez, Ivette María|||0000-0002-3749-277X
This paper presents a deep reinforcement learning (DRL) framework for active flow control (AFC) to reduce drag in aerodynamic bodies. Tested on a 3D cylin-der at Re=100 Re = 100, the DRL approach achieved a 9.32% 9.32% drag reduction and a 78.4% 78.4% decrease in lift oscillations by learning advanced actuation strategies. The method-ology integrates a CFD solver with a DRL model using an in-memory database for efficient communication between the two instances, making it scalable to more complex flows and higher Reynolds numbers., This research has received financial support from the Ministerio de Ciencia e Innovacion of Spain (PID2023-150408OB-C21 and PID2023-150408OB-C22). Authors also extend their gratitude to AGAUR for supporting the research group Large-scale Computational Fluid Dynamics (2021 SGR 00902) and the Turbulence and Aerodynamics Research Group (2021 SGR 01051). Ricard Montalà also expresses gratitude to AGAUR for awarding the FI-SDUR grant (2022 FISDU 00066). Ricardo Vinuesa acknowledges financial support from ERC grant no. 2021-CoG-101043998, DEEPCONTROL., Peer Reviewed, 13 - Acció per al Clima, 9 - Indústria, Innovació i Infraestructura
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Toward a GPU-enabled billionaire SVD in pyLOM

  • Miró Jané, Arnau|||0000-0002-2772-6050
  • Eiximeno Franch, Benet
  • Silva, Lucas Gasparino Ferreira da
  • Kutz, Nathan
  • Rodríguez Pérez, Ivette María|||0000-0002-3749-277X
  • Lehmkuhl Barba, Oriol|||0000-0002-2670-1871
We develop and implement an accelerated high-performance and open-source computing environment for model order reduction in fluid dynamics called pyLOM. It contains singular value decompositionbased algorithms implemented for massively parallel GPU architectures. The library is profiled in detail under the MareNostrum V supercomputer. The largest case has been computed under 20s with 100 GPUs and consisted of a billion nodes by a thousand snapshots matrix. A hybrid CPU-GPU parallel randomized QR factorization has been found to be able to leverage such large matrices. The largest speedup factor of 83 has been found on the QR factorization, while the matrix–matrix multiplication has shown a speedup factor of about 2. Additionally, two examples of application are provided in the flow around a cylinder and the Windsor body, whose PODiscomputed under 3s with 100 GPUs. This showcases the efficiency of GPUs, resulting in a 97% reduction in energy to solution and a reduction of 0.11 kg of CO2 emissions. The scalability and efficiency achieved suggest that this framework can play a key role in reducing the energy demands and environmental impact of large-scale data analysis and model order reduction across a wide range of applications., Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature. The research leading to this work has been partially funded by project TIFON with reference PLEC2023-010251/ AEI/10.13039/501100011033 and by the Ministerio de Ciencia e Innovación of Spain (PID2023-150408OB-C21/C22). Benet Eiximeno’s work was funded by a contract from the Subprograma de Ayudas Predoctorales given by the Ministerio de Ciencia e Innovación (PRE2021-096927). Oriol Lehmkuhl has been partially supported by a Ramon y Cajal postdoctoral contract (Ref: RYC2018-025949-I). The authors acknowledge the support of Departament de Recerca i Universitats de la Generalitat de Catalunya to the Research Group Largescale Computational Fluid Dynamics (Code: 2021 SGR 00902) and the Turbulence and Aerodynamics Research Group (Code: 2021 SGR 01051). Nathan Kutz acknowledges the NSF AI Institute in Dynamics System grant 2112085. We also acknowledge the Barcelona Supercomputing Center for awarding us access to the MareNostrum V machine based in Barcelona, Spain., Peer Reviewed
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Flow separation control of an infinite wing section via multi-agent reinforcement learning

  • Montalà Sales, Ricard
  • Font García, Bernat
  • Suárez Morales, Pol
  • Rabault, Jean
  • Lehmkuhl Barba, Oriol
  • Vinuesa Motilva, Ricardo
  • Rodríguez Pérez, Ivette María
This study investigates the application of deep reinforcement learning (DRL) for active flow control in a three-dimensional NACA0012 wing section with periodic spanwise boundary conditions at low Reynolds number (𝑅𝑒𝑐 =1000) and high angle of attack (AoA =20∘), where the flow exhibits massive separation, strong vortex shedding and a chaotic three-dimensional wake. The baseline configuration is validated against previous numerical studies, showing the accuracy of the computational set-up. Two DRL control policies are trained with different reward functions, leading to fundamentally different control strategies. The DRL set-up uses a multi-agent reinforcement learning framework to coordinate distributed actuators along the wing span. In the first case study, when the reward prioritises drag reduction, agents delay the leading-edge separation and stabilise the wake, reducing drag by 21.1 % and lift fluctuations by 54.1 % while maintaining baseline lift. In the second case, when the reward targets lift enhancement instead, agents synchronise actuations with vortex shedding, forming leading-edge instabilities that increase lift by 42.3 %, albeit at the cost of higher fluctuations and slightly increased drag. A comparison with classical open-loop control further demonstrates that DRL autonomously determines physically meaningful and interpretable strategies consistent with established mechanisms in the literature. To the best of the authors’ knowledge, this is the first study in which a fully three-dimensional computational fluid dynamics simulation has been employed to train a DRL model for active flow control in wings, suggesting promising directions for extending DRL-based strategies to higher Reynolds numbers and more complex wing configurations where prior physical knowledge may be limited., This research was partially funded by the Ministerio de Ciencia e Innovación of Spain (PID2023-150408OB-C21 and PID2023-150408OB-C22). Work has also been co-funded by the European High Performance Computing Joint Undertaking, together with Germany, Italy, Slovenia, Spain, Sweden and France under grant agreement No 101092621; and under grant PCI2022-135085-2, funded by MICIU/AEI/10.13039/501100011033 and by the ‘European Union NextGenerationEU/PRTR’. R. Montalà’s work was supported by a contract from the Agència d’Ajuts Universitaris i Recerca of the Generalitat de Catalunya through a FI-SDUR grant (2022 FISDU 00066). O. Lehmkuhl has been partially supported by a Ramón y Cajal postdoctoral contract (RYC2018-025949-I). The authors also acknowledge the support of the Agència d’Ajuts Universitaris i Recerca of the Generalitat de Catalunya to the research groups Large-scale Computational Fluid Dynamics (2021 SGR 00902) and Turbulence and Aerodynamics Research Group (2021 SGR 01051). R. Vinuesa acknowledges financial support from the European Research Council under grant no. 2021-CoG101043998, DEEPCONTROL. The views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union or the European Research Council., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Contribución de congreso (ConferenceOutput)

Deep reinforcement learning for active flow control around a three-dimensional flow-separated wing at Re = 1, 000

  • Montalà Sales, Ricard
  • Font García, Bernat
  • Suárez Morales, Pol
  • Rabault, Jean
  • Lehmkuhl Barba, Oriol
  • Vinuesa Motilva, Ricardo
  • Rodríguez Pérez, Ivette María
This study explores the use of deep reinforcement learning (DRL) for active flow control (AFC) to reduce flow separation on wings at high angles of attack. Concretely, here the DRL agent controls the flow over the three-dimensional NACA0012 wing section at the Reynolds number Re = 1, 000 and angle of attack AoA= 20◦, autonomously identifying optimal control actions through real-time flow data and a reward function focused on improving aerodynamic performance. The framework integrates the GPUaccelerated computational fluid dynamics (CFD) solver SOD2D with the TF-Agents DRL library via a Redis in-memory database, enabling rapid training. This work builds on previous DRL flow-control studies, demonstrating DRL’s potential to address complex aerodynamic challenges and push the boundaries of traditional AFC methods., This research has received financial support from the Ministerio de Ciencia e Innovacion of Spain (PID2023-150408OB-C21 and PID2023-150408OB-C22). Simulations were conducted with the assistance of the Red Española de Supercomputación (RES) and the EuroHPC JU, who granted us computational resources at the HPC facilities of MareNostrum V at Barcelona Supercomputing Cente (IM-2024-2-0004 and EHPC-REG-2024R01- 038, respectively). Authors also extend their gratitute to the Agència de Gestió d’Ajuts Universitaris i de Recerca (AGAUR) for supporting the research group Large-scale Computational Fluid Dynamics (2021 SGR 00902) and the Turbulence and Aerodynamics Research Group (2021 SGR 01051). Ricard Montalà express also gratitude to AGAUR for awarding the FI-SDUR grant (2022 FISDU 00066), which supports his doctoral studies. Finally, Ricardo Vinuesa acknowledges financial support from ERC grant no.2021-CoG101043998, DEEPCONTROL., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Contribución de congreso (ConferenceOutput)

Discovering flow separation control strategies in 3d wings via deep reinforcement learning

  • Montalà Sales, Ricard
  • Font García, Bernat
  • Suárez Morales, Pol
  • Rabault, Jean
  • Lehmkuhl Barba, Oriol
  • Vinuesa Motilva, Ricardo
  • Rodríguez Pérez, Ivette María
In this work, deep reinforcement learning (DRL) is applied to active flow control (AFC) over a threedimensional SD7003 wing at a Reynolds number of Re = 60,000 and angle of attack of AoA = 14◦. In the uncontrolled baseline case, the flow exhibits massive separation and a fully turbulent wake. Using a GPU-accelerated CFD solver and multi-agent training, DRL discovers control strategies that enhance lift (79%), reduce drag (65%), and improve aerodynamic efficiency (408%). Flow visualizations confirm reattachment of the separated shear layer, demonstrating the potential of DRL for complex and turbulent flows., This research was partially funded by the Ministerio de Ciencia e Innovación of Spain (PID2023- 150408OB-C21/C22). Simulations were supported by the Red Española de Supercomputación (IM-2024-2-0004 and IM-2024-3-0005) and EuroHPC JU (EHPCREG-2024R01-038), granting access to MareNostrum V at the Barcelona Supercomputing Center. We also acknowledge AGAUR for supporting the LS/CFD (2021 SGR 00902) and TUAREG (2021 SGR 01051) research groups. R. M. thanks AGAUR for the FISDUR grant (2022 FISDU 00066) and R. V. acknowledges financial support from ERC grant no. 2021- CoG-101043998, DEEPCONTROL., Peer Reviewed, 9 - Indústria, Innovació i Infraestructura, 13 - Acció per al Clima, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

On the integration of geometry agnostic variational-autoencoders into large-scale SVD based models

  • Eiximeno Franch, Benet
  • Miró Jané, Arnau
  • Kutz, Nathan
  • Rodríguez Pérez, Ivette María
  • Lehmkuhl, Oriol
This manuscript introduces GAVI (Geometry Agnostic Variational-autoencoder Integration), a scalable technique for model order reduction of terabyte-level state spaces. Such methodology is agnostic to the spatial location of the data points, hence, it can handle any complex geometry regardless of the grid type in which it is represented. The dimensionality reduction of GAVI is performed on two separate steps. First, a parallel QR factorization is applied to the snapshot matrix to decompose, then a variational autoencoder learns a latent representation of the reduced R matrix. The final step can be done at laptop level regardless of the size of the original data. The methodology is used to compress the streamwise velocity component of the flow around a circular cylinder at 𝑅𝑒𝐷 = 𝑈∞𝐷∕𝜈 = 104 , the pressure coefficient of the flow around the Windsor body at 𝑅𝑒𝐿 = 𝑈∞𝐿∕𝜈 = 2.9 × 106 and the streamwise velocity of the realistic urban flow in the Zona Universitària neighborhood located in Barcelona. The latter example is the most demanding one as the 1032 snapshots, which are represented on an unstructured grid of 335 million points, have a total size of 2.05 Tb. GAVI can compress the full dataset into 6 latent vectors that recover up to the 95.29% of the energy., The research leading to this work has been partially funded by the project TIFON with reference PLEC2023-010251/ AEI/10.13039/ 501100011033 and by the Ministerio de Ciencia e Innovación of Spain (PID2023-150408OB-C21/C22). B. Eiximeno’s work was funded by a contract from the Subprograma de Ayudas Predoctorales given by the Ministerio de Ciencia e Innovación (PRE2021-096927). Oriol Lehmkuhl has been partially supported by a Ramon y Cajal postdoctoral contract (Ref: RYC2018-025949-I). The authors acknowledge the support of Departament de Recerca i Universitats de la Generalitat de Catalunya to the Research Group Large-scale Computational Fluid Dynamics (Code: 2021 SGR 00902) and the Turbulence and Aerodynamics Research Group (Code: 2021 SGR 01051). Nathan Kutz acknowledges the NSF AI Institute in Dynamics System., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Toward a GPU-enabled billionaire SVD in pyLOM

  • Miró Jané, Arnau
  • Eiximeno Franch, Benet
  • Silva, Lucas Gasparino Ferreira da
  • Kutz, Nathan
  • Rodríguez Pérez, Ivette María
  • Lehmkuhl Barba, Oriol
We develop and implement an accelerated high-performance and open-source computing environment for model order reduction in fluid dynamics called pyLOM. It contains singular value decompositionbased algorithms implemented for massively parallel GPU architectures. The library is profiled in detail under the MareNostrum V supercomputer. The largest case has been computed under 20s with 100 GPUs and consisted of a billion nodes by a thousand snapshots matrix. A hybrid CPU-GPU parallel randomized QR factorization has been found to be able to leverage such large matrices. The largest speedup factor of 83 has been found on the QR factorization, while the matrix–matrix multiplication has shown a speedup factor of about 2. Additionally, two examples of application are provided in the flow around a cylinder and the Windsor body, whose PODiscomputed under 3s with 100 GPUs. This showcases the efficiency of GPUs, resulting in a 97% reduction in energy to solution and a reduction of 0.11 kg of CO2 emissions. The scalability and efficiency achieved suggest that this framework can play a key role in reducing the energy demands and environmental impact of large-scale data analysis and model order reduction across a wide range of applications., Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature. The research leading to this work has been partially funded by project TIFON with reference PLEC2023-010251/ AEI/10.13039/501100011033 and by the Ministerio de Ciencia e Innovación of Spain (PID2023-150408OB-C21/C22). Benet Eiximeno’s work was funded by a contract from the Subprograma de Ayudas Predoctorales given by the Ministerio de Ciencia e Innovación (PRE2021-096927). Oriol Lehmkuhl has been partially supported by a Ramon y Cajal postdoctoral contract (Ref: RYC2018-025949-I). The authors acknowledge the support of Departament de Recerca i Universitats de la Generalitat de Catalunya to the Research Group Largescale Computational Fluid Dynamics (Code: 2021 SGR 00902) and the Turbulence and Aerodynamics Research Group (Code: 2021 SGR 01051). Nathan Kutz acknowledges the NSF AI Institute in Dynamics System grant 2112085. We also acknowledge the Barcelona Supercomputing Center for awarding us access to the MareNostrum V machine based in Barcelona, Spain., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

On the effects of the spanwise length on the flow past an airfoil in stall at moderate Reynolds numbers

  • Montalà Sales, Ricard
  • Lehmkuhl Barba, Oriol
  • Rodríguez Pérez, Ivette María
This study investigates the influence of the spanwise periodic domain length Lz on the accuracy of numerical simulations of flow past an SD7003 airfoil (Selig–Donovan series) in post-stall conditions at an angle of attack a =14º and Reynolds number Rec = 60000. Large eddy simulations and direct numerical simulations are performed for spanwise extents ranging from Lz/c =0.2 to 2.0 to evaluate their impact on flow dynamics, aerodynamic forces, and wake characteristics. The results demonstrate a strong sensitivity of key flow features and aerodynamic coefficients to the chosen spanwise length. Domains that are too short artificially constrain three-dimensional flow structures, leading to significant overprediction of drag, lift, and unsteady lift fluctuations. This behavior is attributed to trailing-edge flow structures being forced into an unrealistically coherent spanwise configuration, which inhibits their natural three-dimensional breakdown in the wake. Analyses based on two-point correlations and flow statistics indicate that a minimum spanwise length of Lz/c ≥ 0:8 is necessary to capture the largest turbulent scales, reduce the influence of periodic boundary conditions, and obtain accurate predictions of unsteady aerodynamic forces. The findings further suggest that previous high-fidelity studies, which used Lz/c = 0.2 for computational efficiency, were insufficient to represent the full three-dimensional complexity of post-stall flow., This research has received financial support from the Ministerio de Ciencia e Innovación of Spain (Nos. PID2023-150408OB-C21 and PID2023-150408OB-C22). The authors also acknowledge the support of the Agència d’Ajuts Universitaris i Recerca of the Generalitat de Catalunya to the research groups Large-scale Computational Fluid Dynamics (No. 2021 SGR 00902) and Turbulence and Aerodynamics Research Group (No. 2021 SGR 01051). Furthermore, simulations were conducted with the assistance of the Red Española de Supercomputacióon (RES) and the European High Performance Computing Joint Undertaking (EuroHPC), who granted us computational resources at the HPC facilities of MareNostrum IV, at Barcelona Supercomputing Center (Nos. IM-2024-2-0004, IM-2024-3- 0005, and EHPC-REG-2024R01-038). Finally, R. Montalà would like to express gratitude to the Departament de Recerca i Universitats de la Generalitat de Catalunya for awarding the FI-SDUR grant (No. 2022 FISDU 00066), which supports his doctoral studies., Peer Reviewed, 9 - Indústria, Innovació i Infraestructura, 7 - Energia Assequible i No Contaminant, Postprint (published version)