Tabulation of combustion chemistry via Artificial Neural Networks (ANNs): Methodology and application to LES-PDF simulation of Sydney flame L

摘要: Abstract In this work, a methodology for the tabulation of combustion mechanisms via Artificial Neural Networks (ANNs) is presented. The objective to train ANN using samples generated an abstract problem, such that they span composition space family problems. problem in case ensemble laminar flamelets with artificial pilot mixture fraction emulate ignition, varying strain rate up well into extinction range. thus covered anticipates regions visited typical simulation non-premixed flame. training consists two-stage process: clustering subdomains Self-Organising Map (SOM) and regression within each subdomain multilayer Perceptron (MLP). approach then employed tabulate mechanism CH4–air combustion, based on GRI 1.2 reduced Rate-Controlled Constrained Equilibrium (RCCE) Computational Singular Perturbation (CSP). applied simulate Sydney flame L, turbulent features significant levels local re-ignition. flow field resolved through Large Eddy Simulation (LES), while transported probability density function (PDF) modelling turbulence–chemistry interaction solved numerically stochastic fields method. Results demonstrate reasonable agreement experiments, indicating SOM-MLP provides good representation space, great savings CPU time allow be performed comprehensive model, as LES-PDF, modest resources workstation.