Artificial Neural Networks Based Chemistry-Mixing Subgrid Model for LES

Abstract

Large Eddy Simulations (LES) of a non-premixed, planar, syngas/air flame is performed with special emphasis on speeding-up the sub-grid chemistry and mixing computations using the Artificial Neural Networks (ANN) approach. The focus of the current study is spent to generate multi-dimensional look-up tables based on the stand-alone 1-D Linear Eddy Mixing (LEM) model calculations of turbulent flames, and use them in ANN for the LES. The test case for LES validation is selected in accordance with a previous Direct Numerical Simulation (DNS) study, which was reported to exhibit considerable local extinction and re-ignition type of complex physics. It is shown that the thermo-chemical database created through LEM computations for ANN training is representative of the actual physial process since the turbulence information is fed through turbulent stirring. Hence, LES computations using ANN can detect the same physics pointed out by the DNS study on a computationally much affordable way, with reasonable accuracy.