A chemistry tabulation approach via Rate-Controlled Constrained Equilibrium (RCCE) and Artificial Neural Networks (ANNs), with application to turbulent non-premixed CH4/H2/N2 flames

Abstract

In this work we propose a chemistry tabulation approach based on Rate-Controlled Constrained Equilibrium (RCCE) and Artificial Neural Networks (ANNs) and apply it to two non-premixed and non-piloted, CH4/H2/N2 turbulent flames (DLR-A and DLR-B, with Re=15,200 and 22,800, respectively). The objective of this approach is to train the ANNs with an abstract problem in such a way that endows them with the predictive power to capture the structure of a real turbulent flame. The training approach involves simulating an ensemble of laminar flamelets to generate training samples. Reduced chemistry is obtained via the application of RCCE to a detailed C/H/O/N mechanism employing 17 constraints. Regarding the training, testing and simulation of ANNs, we make use of the Self-Organizing Map (SOM)–Multilayer Perceptron (MLP) concept to perform pattern recognition tasks and predict the temporal evolution of the leading species. To evaluate the proposed methodology, the accuracy of RCCE-ANNs results is compared to a real time application of RCCE and experimental data in the context of Reynolds Averaged Navier–Stokes (RANS) with a transported Probability Density Function (PDF) modelling of the combustion procedure. Comparison shows reasonable agreement with some discrepancies in the minor species, while a major speed-up of the chemical system integration is reported, indicating the potential of the RCCE-ANN tabulation methodology and ANN training approach for speeding up turbulent combustion computations.