Generating synthetic turbulent flows using Machine Learning
F.X. Demoulin - B. Duret
The objective of this master internship is to research on a particular application of CFD in machine learning, namely investigation the modeling of a synthetic, single phase, turbulent flow.
We will first start by using Langevin’s equations and one hand to generate data and using machine learning on the other hand to finally compare both kind of results. In fact, modeling of turbulence signal using Langevin’s equations is well-known problem, and the main work will be mainly related to model the case using machine learning and to evaluate the results.
The neural network created can be further enriched by not only using the data from Langevin equation but also by using more advance model or experimental data and direct numerical simulations. This progressive way to introduce more and more realistic physics and to monitor how the network is able to adapt itself will be very instructive on further possibility of using deep learning in physics and more particularly in fluid mechanics.
Plus d’information dans le PDF joint.
Articles de ce dossier :