Kinetic combustion neural modelling integrated into computational fluid dynamics

Abstract

The attempt to replace traditional chemical kinetics model calculations with new ones based on neural networks (NNs) has been successfully carried out. The paper deals with the methodology that has been followed to replace traditional model calculations with neural models (NMs) for methane/air combustion. The reacting flowfield has been described with account taken of the detailed chemical reaction mechanism. Convective and turbulent diffusive transport of species has been taken into consideration by means of a well-known finite volume computational fluid dynamics (CFD) code. Two versions of such a mechanism have been developed. The first one is based on traditional differential equations representing the species production rates. Such equations are integrated over the time intervals related to the cell volumes and local volumetric flows. The second version is based on neural models which can extract and store knowledge from the data presented to them. The neural model capability of connecting output to input quantities by means of the stored knowledge leads to very fast calculations. A reduced combustion mechanism involving 20 species and 68 reactions has been developed both for the traditional calculation and for the neural model calculations. It can be concluded that calculations using chemical kinetics neural models show a 42 times shorter CPU time than that of the traditional procedures, with a comparable solution accuracy of the combustion flowfields.