Decipher

Didier Lucor, Lionel Mathelin, Bérengère Podvin, Onofrio Semeraro

Flow control is one of the viable techniques proposed for the the design and maximization of the energy efficiency in engineering systems and it is one of our main axis of research. In particular, this activity is supported by Lidex ICODE (Université Paris-Saclay) on "l’aide à la décision et la maîtrise des processus dynamiques complexes" (decision support and process control complex dynamics).

These activities are both focussed on the practical and theoretical aspects of flow control, ranging from the control of instationary flow to the development of algorithms for the nonlinear control in closed loop, with or without physical models at hand, thus leveraging statistical learning (or reinforcement learning).

In parallel with the activities concerning flow control, we are developing our know-how in data-processing of large datasets obtained from numerical simulations, experiments in fluid mechanics and heat transfer. These developments are - on the one hand - useful for increasing our understanding of physical phenomena (modal decomposition, sampling of infinite-dimensional operators) and - on the other hand - necessary for the development of robust and high-fidelity modelling (inference, assimilation, sparse representation).

We are also working on the development of Uncertainty Quantification (UQ) techniques which usefully complement the landscape of techniques for the parametric analysis of sensitivity, especially to deal with inference and identification problems of complex models. Beyond methodological developments, the application of UQ techniques can be extend to numerous applications (Bio-medical Engineering, Geosciences, Aerodynamics, etc.).

Our efforts will specifically focus on

  • Data analysis for estimation and assimilation in fluid mechanics:
    • Dictionary learning, manifold learning;
    • Approximation of infinite-dimensions operators by sampling (Koopman operator);
    • Tensor-train decomposition;
    • Random algebra for model reduction;
    • Data assimilation and optimization;
  • Uncertainty quantification (UQ)
    • Methodological development;
    • Applications;
  • Flow control
    • Reinforcement learning for control;
    • Data-driven control using statistical learning;
    • Mise en place of experimental demonstrators.

LIMSI
Campus universitaire bât 507
Rue du Belvédère
F - 91405 Orsay cedex
Tél +33 (0) 1 69 15 80 15
Email

SCIENTIFIC REPORTS

Scientific report

LIMSI in numbers

8 Research Teams
100 Researchers
40 Technicians and Engineers
60 Doctoral Students
70 Trainees


Logo DataIA

 

© 2017 LIMSI CNRS