Effect of clustering on reactor network models

Predicting the formation of pollutants from combustion devices requires the combination of accurate CFD and detailed kinetic mechanisms. A possible approach to handle this problem consists in post-processing the related CFD results, obtained through a simplified kinetic scheme, with detailed mechanisms. This is done by the Kinetic Post Processor (KPP), a tool developed at Politecnico di Milano, able to estimate the emissions of compounds like NOx, CO and PAH with the required accuracy. Nevertheless, even by using a post-processing technique, this evaluation may be particularly heavy from a computational point of view, especially when dealing with realistic, full-scale models. In these cases, the corresponding reactor network model can be unburdened by grouping (or clustering) cells, which are similar from a fluid dynamic and kinetic standpoint. Literature shows that although in this way the original fields of motion and temperature are altered, its impact on the output can be negligible and the results close to experimental measures. The recent availability of a parallel version of the software has extended its potentials, thus making possible to work with reactor networks made up of 10^5÷10^6 reactors and kinetic schemes of 50÷100 species. Therefore, it has been possible to test models of devices without any kind of cells’ clustering. In particular, in this paper a sensitivity analysis to clustering of a case study (a low-NOx combustor) has been carried out and discussed; as a feedback, the output levels in terms of NO and CO are evaluated. The results provided important indications from two different point of views: the impact that clustering can have on pollutants when modeling complex devices, and the importance of the upstream CFD simulation when evaluating emissions through a post-processing strategy.