PhD defence of Aakash Patil

Deep Learning-Assisted Modelling of Turbulence in Fluids

 

 

Aakash Patil conducted his PhD work in the CFL team under the supervision of Elie Hachem and Jonathan Viquerat. He will defend his PhD in Computational Mathematics, High Performance Computing and Data, on February 1st, 23 in front of the following jury:

 

Gianluigi Rozza, SISSA, Trieste, Italy, reviewer

Paola Cinnela, Sorbonne University

Ricardo Vinuesa, KTH Stockholm, Sweden

Elie Hachem, Mines Paris – PSL

Jonathan Viquerat, Mines Paris – PSL

 

 

Abstract:

 

Despite several advancements in experimental and computational resources, and despite progress in theoretical and mathematical procedures to address the closure of Navier-Stokes equations, turbulence remains an unsolved problem even after 200 years of continuous research. On the other hand, artificial machine intelligence and related technologies are making rapid advancements in several domains of science and engineering, helping us humans to efficiently solve modeling problems and discover new physics. Present work tries to combine these two branches and explore if computational machines can be used to efficiently study turbulence in fluids, and perhaps someday help us in the discovery of the missing universal laws.  Deep learning is employed to learn turbulence modeling and a patch-based method is proposed for robust learning. Learning of subgrid-scale turbulence from the resolved large scales is demonstrated along with investigation of effect of the coarse-graining methods and successive refinements.  Spatio-temporal learning of turbulent flows is proposed to learn the temporal snapshots and a posteriori simulations are performed.

 

Keywords: Turbulence, Modeling, Deep Learning