Sensitivity to Modeling Parameters in Bluff Body Stabilized Flames

Abstract

This study investigates the sensitivity of turbulent-reacting large-eddy simulations (LESs) to seven ubiquitous modeling parameters. A forward-propagating uncertainty-quantification technique is applied to chemistry–turbulence interaction and subgrid modeling parameters through the solution of a bluff-body stabilized, premixed flame. Both implicit- and explicit-filtering approaches are used for the LES, providing insight into the role of modeling parameters in both approaches. Uniformly distributed ranges are provided to the seven modeling parameters, and various levels of Smolyak sparse grid parameter sets are simulated via LES to statistical convergence. From these sparse grid LES realizations, a surrogate model of mean flow quantities (temperature, velocity, etc.) is computed for calculation of parameter space statistics. This study shows that the explicit-filtering simulations exhibit a magnified sensitivity to modeling parameters relative to the implicit-filtering simulations. These increased sensitivities manifest most prominently in the temperature field, with certain parameter sets quenching the flame in explicit-filtering simulations but not in the implicit-filtering simulations. Lastly, it is found that three of the seven modeling parameters (namely, the dynamically thickened flame efficiency function exponent , turbulent Prantdl number , and eddy-viscosity coefficient ) dominate the solution for both implicit- and explicit-filtering approaches.