DAta science, TrAnsition, Fluid instabiLity, contrOl, Turbulence (DATAFLOT)

Working towards mastery of turbulence is a major challenge that impacts a large number of applications in the engineering sciences. It is crucial to understand the mechanisms by which instabilities arise and grow, as well as the triggering and development of turbulence.

The main research themes are

Sergio Chibbaro, Didier Lucor, Lionel Mathelin, Onofrio Semeraro

Flow control remains one of the means of controlling the energy efficiency of systems and designing more efficient energy systems. Our activities in the field of flow control is a particularly strong and visible activity of the laboratory. In particular, it is supported by the Lidex ICODE (Université Paris-Saclay) on “decision support and control of complex dynamic dynamic processes”. Part of these activities is focused on instationnarity control. The other part focuses on non-linear closed-loop control techniques based on model-free methods model-free methods or reinforcement control..

In parallel with these activities, we are developing our know-how in data processing from numerical simulations and experiments in fluid mechanics and transfer mechanics. These developments are, on the one hand, useful for increasing our understanding of physical phenomena (modal decomposition, infinite-dimensional operator hollow sampling) and, on the other hand, necessary for the development of increasingly reliable representations or modeling (inference, assimilation, hollow representation), notably for application to control (Machine Learning, in particular). We are also working on the development of Uncertainty Quantification (UQ) techniques, which are a useful addition to the landscape of techniques for analyzing parametric sensitivity, particularly for dealing with complex model identification and inference problems.
In addition to methodological developments, the dissemination of UQ techniques should be intensified towards more applications (Bio-medical Engineering, Geosciences, Aerodynamics, …).
Our efforts will focus more specifically on :

  • data analysis for fluid mechanics estimation and assimilation:
    • dictionary learning, variety learning ;
    • infinite-dimensional operator hollow sampling (Koopman), function train
    • tensor approximation ;
    • random projection for model reduction;
    • data assimilation and optimization;
  • Uncertainty quantification (UQ):
    • efficient methodological developments ;
    • transfer to applications ;
    • Flow control :
    • control by reinforcement ;
    • model-free, machine-learning control;
    • setting up experimental demonstrators.

Yohann Duguet, Francois Lusseyran, Laurent Martin-Witkowski, Stéphanie Pellerin

Fluid flows can be classified into several regimes, such as laminar, transitional or turbulent, corresponding to significant differences from an energy point of view. The dynamic processes involved in moving from one regime to another, or in stabilizing one of these regimes, are still poorly understood. The instability of a given laminar flow in the face of arbitrary disturbances, of either infinitesimal or finite amplitude, gives rise to interesting and varied mathematical and numerical developments, depending on the type of flow considered. Transitions, often hysteretic, between different regimes also exist within turbulent flows. An original focus is placed on the analysis of spatial symmetries and their breaking by instability mechanisms. These are described qualitatively and quantified using innovative and efficient numerical algorithms, within the framework of three-dimensional unsteady simulations requiring considerable resources and specific methods for large-scale data. An experimental cell also enables the visualization and quantification of these same flows, in direct complementarity with numerical studies. Finally, a detailed understanding and modeling of the hydrodynamic mechanisms at work naturally leads to experimental and/or numerical control methods, enabling the system to be steered towards the desired regime.

Configurations

  • experimental and numerical study of the effects of ambient pollution on the stability of rotating flows with rigid or deformable free surfaces
  • numerical simulation, modeling and control of symmetry breaking in turbulent wakes
  • understanding the mechanisms of sub-critical transition to turbulence in sheared wall flows, from both a non-linear (description of the corresponding phase space) and spatio-temporal (analysis of intermittency) point of view
  • experimental and numerical study and closed-loop control of open shear flows

Coordination

Team members

Recent scientific publications

  • Pré-publication, Document de travail

    Edgar Jaber, Emmanuel Remy, Vincent Chabridon, Mathilde Mougeot, Didier Lucor. Fusion of heterogeneous data for robust degradation prognostics. 2025. ⟨hal-05091317⟩

    DATAFLOT

    Year of publication

    Available in free access

  • Thèse

    Remy Hosseinkhan-Boucher. On Learning-Based Control of Dynamical Systems. Artificial Intelligence [cs.AI]. Université Paris-Saclay, 2025. English. ⟨NNT : 2025UPASG029⟩. ⟨tel-05061303⟩

    DATAFLOT

    Year of publication

    Available in free access

  • Thèse

    Michele Quattromini. Graph Neural Networks for fluid mechanics : data-assimilation and optimization. Machine Learning [cs.LG]. Université Paris-Saclay; Politecnico di Bari, 2024. English. ⟨NNT : 2024UPAST161⟩. ⟨tel-04964013⟩

    DATAFLOT

    Year of publication

    Available in free access

  • Thèse

    Thibault Monsel. Deep Learning for Partially Observed Dynamical Systems. Discrete Mathematics [cs.DM]. Université Paris-Saclay, 2024. English. ⟨NNT : 2024UPASG113⟩. ⟨tel-04952358⟩

    AO, AO, DATAFLOT, DATAFLOT

    Year of publication

    Available in free access

  • Article dans une revue

    Yohann Duguet. Intermittency in Transitional Shear Flows. Entropy, 2021, 23 (3), pp.280. ⟨10.3390/e23030280⟩. ⟨hal-04938938⟩

    DATAFLOT

    Year of publication

    Available in free access

  • Thèse

    Nemo Malhomme. Statistical learning for climate models. Machine Learning [stat.ML]. Université Paris-Saclay, 2024. English. ⟨NNT : 2024UPAST165⟩. ⟨tel-04939186⟩

    DATAFLOT, DATAFLOT

    Year of publication

    Available in free access

  • Communication dans un congrès

    Soufiane Mrini, Anne Sergent, Francesca Chillà, Julien Salort, Didier Lucor. Addressing turbulent convection experimental data challenges in PINNs with appropriate physical sampling. Joint event Euromech Colloquium on Data-Driven Fluid Dynamics/2nd ERCOFTAC Workshop on Machine Learning for Fluid Dynamics, Apr 2025, London, United Kingdom. ⟨hal-04924450⟩

    COMET, DATAFLOT

    Year of publication

    Available in free access

  • Communication dans un congrès

    Marie-Christine Volk, Didier Lucor, Anne Sergent, Michael Mommert, Christian Bauer, et al.. A PINN Methodology for Temperature Field Inference in the PIV Measurement Plane: Case of Rayleigh-Bénard Convection. Joint event Euromech Colloquium on Data-Driven Fluid Dynamics/2nd ERCOFTAC Workshop on Machine Learning for Fluid Dynamics, Apr 2025, London, United Kingdom. ⟨hal-04924426⟩

    COMET, DATAFLOT

    Year of publication

    Available in free access

  • Communication dans un congrès

    Jai Kumar, Anne Sergent, Francesca Chillà, Julien Salort, Didier Lucor. Bridging Experimental shadowgraphs and DNS in Turbulent Convection Using physically-informed U-Net. Joint event Euromech Colloquium on Data-Driven Fluid Dynamics/2nd ERCOFTAC Workshop on Machine Learning for Fluid Dynamics, Apr 2025, London, United Kingdom. ⟨hal-04924440⟩

    COMET, DATAFLOT

    Year of publication

    Available in free access

  • Communication dans un congrès

    Anne Sergent, Soufiane Mrini, Elian Bernard, Didier Lucor. Lagrangian Measurements and Physics-Informed Neural Network for Rayleigh-Bénard Flow Reconstruction. 26th International Congress of Theoretical and Applied Mechanics (ICTAM2024), Aug 2024, Daegu, South Korea. ⟨hal-04924332⟩

    COMET, DATAFLOT

    Year of publication

    Available in free access

  • Article dans une revue

    A. Gesla, Y. Duguet, P. Le Quéré, Laurent Martin Witkowski. On the origin of circular rolls in rotor-stator flow. Journal of Fluid Mechanics, 2024, 1000, pp.A47. ⟨10.1017/jfm.2024.1011⟩. ⟨hal-04902902⟩

    COMET, DATAFLOT

    Year of publication

    Available in free access

  • Communication dans un congrès

    Vincent Blot, Alexandra Lorenzo de Brionne, Ines Sellami, Olivier Trassard, Isabelle Beau, et al.. Efficient Precision Control in Object Detection Models for Enhanced and Reliable Ovarian Follicle Counting. Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, Carole H. Sudre (University College London); Raghav Mehta (Imperial College London); Cheng Ouyang (Oxford University); Chen Qin (Imperial College London); Marianne Rakic (Massachusetts Institute of Technology); William M. Wells (Harvard Medical School), 2024, Marrakesh, Morocco. pp.183-193, ⟨10.1007/978-3-031-73158-7_17⟩. ⟨hal-04895227⟩

    DATAFLOT

    Year of publication

  • Article dans une revue

    Yohann Duguet. Puffing along. Nature Physics, 2024, 20 (8), pp.1227-1227. ⟨10.1038/s41567-024-02565-2⟩. ⟨hal-04795934⟩

    DATAFLOT

    Year of publication

    Available in free access

  • Poster de conférence

    Thibault Monsel, Onofrio Semeraro, Lionel Mathelin, Guillaume Charpiat. Time and State Dependent Neural Delay Differential Equations. ML-DE@ECAI 2024 : Machine Learning Meets Differential Equations: From Theory to Applications, Sep 2024, Santiago de compostela, Galicia, Spain. ⟨hal-04794800⟩

    AO, DATAFLOT

    Year of publication

    Available in free access

  • Article dans une revue

    Thomas Boeck, Mattias Brynjell-Rahkola, Yohann Duguet. Energy stability of magnetohydrodynamic flow in channels and ducts. Journal of Fluid Mechanics, 2024, 987, pp.A33. ⟨10.1017/jfm.2024.393⟩. ⟨hal-04795942⟩

    DATAFLOT

    Year of publication

    Available in free access

  • Article dans une revue

    Mattias Brynjell-Rahkola, Yohann Duguet, Thomas Boeck. Chaotic edge regimes in magnetohydrodynamic channel flow: An alternative path towards the tipping point. Physical Review Research, 2024, 6 (3), pp.033066. ⟨10.1103/PhysRevResearch.6.033066⟩. ⟨hal-04795955⟩

    DATAFLOT

    Year of publication

    Available in free access

  • Article dans une revue

    Thomas Boeck, Mattias Brynjell-Rahkola, Yohann Duguet. Energy stability analysis of MHDMagnétohydrodynamique flow in a rectangular duct. PAMM, 2024, 24 (2), pp.e202400041. ⟨10.1002/pamm.202400041⟩. ⟨hal-04795978⟩

    DATAFLOT

    Year of publication

    Available in free access

  • Article dans une revue

    Yanis Zatout, Adrien Toutant, Onofrio Semeraro, Lionel Mathelin, Françoise Bataille. A Priori Reconstruction of Thermal-Large Eddy Simulation (T-LES) by Deep Learning Reconstruction a Priori de Champs de Simulations Des Grandes Echelles Thermiques Par Apprentissage Profond. Entropie : thermodynamique – énergie – environnement – économie, 2023, 4 (3), ⟨10.21494/ISTE.OP.2023.1015⟩. ⟨hal-04751095⟩

    DATAFLOT

    Year of publication

    Available in free access

  • Article dans une revue

    P. Kashyap, Y. Duguet, Olivier Dauchot. Linear stability of turbulent channel flow with one-point closure. Physical Review Fluids, 2024, 9 (6), pp.063906. ⟨10.1103/PhysRevFluids.9.063906⟩. ⟨hal-04749262⟩

    DATAFLOT

    Year of publication

    Available in free access

  • Communication dans un congrès

    Siham El Garroussi, Sophie Ricci, M. de Lozzo, N. Goutal, D. Lucor. Towards the reduction of uncertainty in hydraulic modles for better flood forecasting. Simhydro 21, Jun 2021, Nice, France. ⟨hal-04739507⟩

    DATAFLOT

    Year of publication