Kinetic data-driven approach to turbulence subgrid modeling

Recent advances in Machine Learning have opened up new perspectives for employing Artificial Neural Networks (ANNs) to enhance computational fluid dynamic solvers and develop data-driven turbulence models. In the context of Large Eddy Simulation (LES), ANNs have been used to establish subgrid scale (SGS) closure models from extensive datasets of fully resolved turbulent flows, leveraging their ability to handle high-dimensional and complex data.

This talk presents a data-driven kinetic approach to turbulence modeling, using Direct Numerical Simulation (DNS) data of homogenous isotropic turbluent flows to learn a surrogate collision operator for a lattice Boltzmann solver, which effectively acts as a SGS model. We show that by exploiting the extra degrees of freedom offered by the mesoscopic description the model allows for stable simulations on coarse grids, preserving the statistical properties of turbulent flows, correctly capturing the intermittency of high-order velocity correlations. This work highlights how ANN can be employed to embed new physics from data in the framework of kinetic models.