Reduced interaction between numerical and model errors through anisotropic filtering

Abstract

The work addresses to a better comprehension of the error assessment in LES due to the coupling between the model and the numerical discretisation.

The possibility to reduce the interactions between the error sources is investigated through the use of an algebraic function correlating the characteristics length of the subgrid model Δ _e to the subgrid scale Δ related to the grid discretization, incrementing the ratio between the two where the scales are poorly resolved.

The analysis, considering a range of grid resolutions as well as subfilter models, has been performed starting from database sets which have been reconstructed with ordinary kriging to estimate the sensitiveness of the strategy with respect to the simulation parameters. The results indicate that a reduction in the error cost function can be achieved for most subfilter models and that the approach looks quite stable for a moderate range of the grid resolution.