Application of Machine learning to particle laden turbulence

This statistical study carried out as part of a research project whose general objective is to initiate research activities through which students are placed under the responsibility of academic tutors. The main goal of this project is to use machine learning methods such as Tensorflow, KNN and DBSCAN in order to study the microphysical properties of particle-laden turbulence as well as the influence of two physical parameters (Reynolds number and Stokes number) in their interactions. The first characterizes the turbulence intensity of the flow and the second characterizes the behavior of a particle in a fluid. Both are dimensionless numbers. To apply these methods, a three-dimensional direct numerical simulation of particle-laden isotropic turbulence is used by solving the Navier-Stokes equations on supercomputers. The data have been kindly provided by Dr. Keigo Matsuda through the Mesocenter (Aix-Marseille Intensive Computing Center).