Artificial Neural Net Chemistry Module for Large Eddy Simulations

Combustion Large Eddy Simulation (LES) is a vastly improved method of modeling turbulent-combustion in gas turbine combustors, and will provide improved emissions/instability prediction capability of UEET combustors. It is widely recognized that one of the best subgrid turbulence-combustion models is the Linear Eddy Mixing (LEM) model with multi-step reactions. However, this model is not commonly used in the design environment for one main reason: computational time. In this SBIR, we propose to develop a subgrid LEM module with multi-step Jet-A kinetics that is 100 times faster than the state-of-the-art direct LEM. This new, innovative method performs off-line LEM calculations over a range of turbulence levels and compositions to train an artificial neural net (ANN). The ANN is then used to determine the chemical source terms required in the combustion LES calculation. In Phase I of this SBIR, the feasibility of developing the LEM/ANN module will be demonstrated by training a one-step, five species Jet-A fuel oxidation mechanism. The preliminary LEM/ANN module will be implemented into the CFD-ACE+ code, validation performed, and speed-up demonstrated. At the end of Phase I, the module will be delivered to NASA for incorporation into the NCC code. In Phase II, the LEM/ANN approach will be extended to include multi-step Jet-A kinetic mechanisms, and applied to UEET combustor configurations.