Modeling uncertainties in primary zone soot predictions for a rich-quench-lean combustion system

A comprehensive uncertainty quantification (UQ) of soot volume fraction (SVF) predictions in the primary zone of a Rich-Quench-Lean (RQL) combustor is presented, with particular emphasis on the modeling uncertainties in the rates of nucleation, growth, oxidation, condensation, and coagulation. Large-eddy simulations (LES) of soot formation in a realistic single-sector RQL combustor are first performed, and the local thermochemical data, along with the volumes of the zones containing the respective finite volume cells, are then used to construct a chemical reactor network (CRN) model. The CRN model, coupled with the UQ tool DAKOTA, is used to conduct forward UQ, sensitivity analysis, and inverse UQ. The forward UQ indicates variability ranging from 28 % to 89% around the mean soot prediction, depending on the location within the combustor. The local and global sensitivity analyses highlight the contributions of nucleation and condensation processes near the fuel injection zone, while growth and oxidation processes predominantly influence soot predictions in the primary zone. Since the baseline model underpredicts soot compared to experimental measurements, a Bayesian inference-based inverse UQ analysis is performed to identify sensitive input rate uncertainties that can improve the quantitative agreement with experimental soot levels. Thus, the overall strategy identifies the most influential aspects of the soot model, their relevant sensitivity to local zones within the combustor, and further refinements to the baseline rates that can provide valuable insights for future model developments.