Simulations of methane-air flames using ISAT and ANN

Abstract

Large-eddy simulations (LES) of turbulent flames with detailed finite-rate kinetics is currently computationally infeasible due to the enormous cost associated with computation of reaction kinetics. Recently, an In-Situ Adaptive Tabulation (ISAT) methodology was shown to reduce the cost of direct integration considerably. However, ISAT tables require significant on-line storage in memory and may result in restriction on massivelly parallel systems. Furthermore, application of ISAT in LES requires re-evaluation of the tree structure and the access/retrieval process. Here, issues regarding the use of ISAT in a LES are discussed. Then, a storage-efficient Artificial Neural Network (ANN) is trained using ISAT data and used to simulate turbulent premixed flames in both the thin-reaction-zone and flamelet regimes. Finally, the issues to be addressed in order to apply this combined ISAT/ANN methodology for full-scale LES of reacting flows are discussed.

Artificial Neural Networks for Modeling of Chemical Source Terms in CFD Simulations of Turbulent Reactive Flows

“…The species -temperature profiles thus obtained were interpolated to provide a number of data points sufficiently large for an accurate ANN training and testing. To increase numerical accuracy [8], the realizable domain is divided into subsets defined in terms of limiting temperatures such that in each subset (or bin) every reaction rate profile maintains a unique elementary shape (such as line, parabola, circle, etc. ), a logarithmic transformation is applied to both the input and to the desired output, and a final normalization is applied.…”

“…More recently, an ANN approach combined with a modeled velocityscalar joint PDF was used to numerically simulate a H 2 -CO 2 premixed flame [6], and combined with In-Situ Adaptive Tabulation to study a partially stirred reactor [7]. ANN was also used to estimate the chemical source terms in Large Eddy Simulations (LES) of turbulent flames [8,9], or for chemistry representation, by replacing the flamelet chemistry tables with a set of off -line trained neural networks [10]. In different applications, but while still remaining in the field of turbulent combustion, ANNs were successfully employed to model the ignition delay in homogeneously charged compression ignition engines [11], or to create an LES sub-grid model for the near wall region [12].…”

Les Software for the Design of Low Emission Combustion Systems for Vision 21 Plants

“…That is, difficulties in locating nearest neighbor compositions and thus a duplication of records. Professor Menon at Georgia Tech has eluded to this potential problem in his recent work (Kapoor et al, 2001). Specifically, the following shortcomings with the original ISAT were identified:…”

“…Earlier studies (Kapoor et al, 2001) focused on a 19-species methane-air mechanism and demonstrated the feasibility of developing ANN for the thermo-chemical state. The ANN was developed and applied to a turbulent premixed flame (Flame F1) that was in the thin-reaction-zone regime.…”