Self-Organizing Maps for Fast LES Combustion Modeling

Tremendous advances have been made in the development of large and accurate detailed reaction chemistry models for hydrocarbon fuels. Comparable progress has also been achieved in CFD as an engineering design tool. Highly accurate hydrocarbon chemistry is now desired for simulating gas turbine combustors and automobile engines to better predict both performance and pollutant emissions. Newer and more accurate CFD techniques like Large Eddy Simulation (LES) are being used more as computational power increases along with the demand for better flow predictions. Unfortunately, using large, detailed chemical mechanisms to simulate real turbulent combustion devices is problematic due to the sheer computational burden of the added chemistry. As a result, chemistry mechanisms employing a large number of chemical species are currently only feasible to run in the simplest of flow geometries, and only the simplest and least accurate chemistry models are currently tractable to run in LES CFD codes. We propose using a unique neural network approach to create a fast and accurate species source term function that could alleviate both of these problems.