A data-driven approach for hydrogen premixed combustion modeling

Lean premixed hydrogen flames are characterized by the high mobility of the hydrogen molecule that yields mixtures with an effective sub-unity Lewis number. This causes formation of intrinsic instabilities as a consequence of the hydrodynamic and thermo-diffusive mechanisms. The interaction of these instabilities leads to complex and non-linear corrugations of the flame front, even in laminar flames, with characteristic length-scales falling under the subgrid level, requiring therefore dedicated modeling when approaching the flame to LES. In this contribution, a data-driven model for lean premixed hydrogen combustion is presented. It consists in a two-scalars manifold representation, obtained by the spatial filtering of a two-dimensional, fully resolved, unstable, self-wrinkling DNS hydrogen/air dataset. The method is validated a-priori and a-posteriori against a two-dimensional DNS.