Speaker
Description
Turbulent flow is characterized by a wide spectrum of temporal and spatial scales, giving rise to complex dynamics. Turbulent flow often cannot be resolved fully in a direct numerical simulation (DNS) and coarsened approximate models for the dynamics at larger scales only, are considered. An important reduced-order model is so-called large-eddy simulation (LES), motivated through the application of a spatial convolution filter to the Navier-Stokes equations coupled to the combustion dynamics equations.
Filtering of the nonlinear terms of even the most basic combustion models gives rise to a complex closure problem involving many terms. This prompts the introduction of sub-filter models to represent the interactions among the various combustion and flow scales. In LES, several sub-filter models have been developed. A major issue concerns the specific uncertainties with which coarsened combustion can be predicted by a specific sub-filter model. This modeling uncertainty can be addressed by turning to self-contained modeling principles such as the dynamic procedure, approximate deconvolution modeling (ADM) or regularization such as Leray and NS-a modeling. The translation of an LES formulation into a computational model introduces further approximations of a numerical nature. These two sources of error, i.e., the sub-filter modelling error and the discretisation error, together induce a total simulation error that needs to be understood and quantified, to instill confidence in a simulation.
In the presentation, the LES closure problem and the decomposition of the total simulation error into modelling and discretisation components will be described. We first focus on the generic turbulence problem of homogeneous isotropic flow and compute the error landscape of eddy-viscosity models. From this, we infer optimal simulation conditions minimising the total simulation error. Subsequently, this approach is applied tolarge-eddy simulation of the turbulent non-premixed Sydney bluff-body flame. The error-landscape approach yields optimal parameter settings for accurate simulations at considerably coarsened resolutions compared to DNS requirements.
