Lionel Agostini

Researcher - CNRS (CR)

CNRS-Université de Poitiers-ISAE ENSMA
11 Boulevard Marie et Pierre Curie
Téléport 2, BP 30179
86962 Futuroscope Cédex

Office : Building H2 – Room 127

Research Team



Since October 2020 I am a researcher at Pprime institut. My research focuses mainly on problems related to the presence of turbulent boundary layers, such as the drag reduction. The main purpose of my work is to unveil the physics and deepen our knowledge in order to provide more accurate flow-dynamics modelling and innovative control strategies. In this regard, I have developed and used a variety of statistics algorithms and data-driven methods for mining large datasets mainly obtained by numerical simulations carried out on HPC.

In 2008, I started a thesis on compressible turbulent boundary layer submit to adverse pressure gradient. Subsequently, my Postdoctoral research at Imperial College London and Ohio State University allowed me to focus on both, compressibility and Reynolds number effects on boundary layer, and on a wide range of others research topics through the supervision of MSc and PhD students.

More recently, I turned my attention on possibilties offered by Machine Learning algorithms for flow modelling and control. In a “classical” approach, the first step is to understand, then predict and finally optimise. However, for the vast majority of flow configuration this “sequential” approach is untenable, as flows are highly dimensional, strongly non-linear and multiscale. The first two phases, understand and predict,  required an extremely high amount of hard work, this is especially true for wall-bounded flows. This model-driven approach has two major drawbacks: (i) it is difficult to interact in real time as the model is often complex, (ii) the low-dimensional dynamical model can not be easily generalisable. Machine Learning offers a shift of paradigm: the phases of understanding, prediction and optimisation are a data-driven iterative process (as depicted in Fig. 1). At a given level of understanding, a prediction model is determined from the data, then a first optimisation strategy is evaluated, deepening our understanding and improving the prediction. Data-driven algorithms offer a most efficient approach for unravelling mechanisms driving flow dynamics.

Figure 1- Shift of paradigm using data-driven algoritms


By  exploiting my knowledge and insight on near-wall turbulence combined with Pprime team expertise on : numerical simulations, experiments and data-driven methods,  we are in an good position to pursue the goal of deepen our knowledge of near-wall turbulence, introduce more universal model of turbulence and determine more efficient control device.