2021 AdG GRANT Smart-TURB
A Physics-Informed Machine-Learning Platform for
Smart Lagrangian Harness and Control of TURBulence
is it difficult to control, predict and model a flowing system? to
search and navigate inside it? to be prepared against extreme events?
to tame them? It is in turbulent flows.
Turbulence is ubiquitous
and unsolved from the point of view of out-of-equilibrium fundamental
physics, uncontrollable from the engineering aspects, and a deadlock
for brute-force numerical and experimental investigations. Indeed,
progress by using conventional methods has been slow.
In this project, I
propose to explore new avenues crossing the boundaries between
Theoretical Engineering and Applied Physics using algorithms from
Artificial Intelligence (AI) to study and control turbulence in
an innovative way using smart Lagrangian objects in a vast array of
flows. I am committed to: (i) develop original applications of AI
algorithms to track and harness moving coherent structures and/or
statistical turbulent fluctuations, (ii) optimise flow navigation of
buoyant objects and active surface drifter, (iii) invent collective
search protocols to locate emissions from fixed or floating
sources, (iv) minimise turbulent dispersion of a swarm of
autonomous underwater explorer and (v) perform new in-silico
experiments for data-assimilation, to predict extreme-events, or to
control turbulent fluctuations by novel Lagrangian injection/adsorption
unifying fil-rouge of my project is to gain a Deep Understanding of
turbulence by performing cutting-edge Lagrangian numerical studies. The
project is both methodology oriented, with the grand challenge of
developing fully unconventional applications of (Deep) Reinforcement
Learning for fluid dynamics, and problem driven, delivering a series of
specific efficient control strategies for important realistic flowset-ups and applications to the geophysical fields .