Computationally Efficient Large Eddy Simulation of Multi-Stream Partially Premixed Turbulent Combustion

Abstract

Modern combustion system designs increasingly incorporate multiple inhomogeneously mixed fuel and oxidizer inlet streams, leading to a partially premixed combustion environment. As a result, there may be regions of premixed-like, nonpremixed-like, and intermediate combustion modes that occur within a single system. The current computationally efficient state-of-the-art for turbulent combustion simulation is to couple Large Eddy Simulation (LES) of the turbulent flow field with reduced-order manifold combustion models that assume single-mode combustion with one or two homogeneous inlet streams. Therefore, multi-stream mixing and multi-modal combustion present significant modeling challenges that must be overcome to enable predictive simulations of engineering systems.

This dissertation addresses the modeling challenges for partially premixed turbulent combustion, with new models being motivated and validated in the context of the Sydney Inhomogeneous Inlet Burner, which emulates key features of practical systems, namely three-stream mixing and multi-modal combustion. The limitations of single-mode combustion models are addressed by developing a set of models that use two mixture fractions to account for inhomogeneous partial premixing in this three-stream system. Multi-modal combustion is accounted for using three approaches: two that use the mixing state to indicate the combustion mode and one that uses a regime-indicating parameter. Unlike single-mode models, using two mixture fractions enables qualitatively correct predictions of minor species including CO, but some challenges remain in predicting the transition between combustion modes in the flame of interest.

Modeling systems with three-stream mixing requires accounting for the unresolved distribution of the two mixture fractions, which is accomplished using the presumed probability density function (PDF) approach. A new physics-based method for PDF model selection has been developed, validated using Direct Numerical Simulations of non-reacting three-stream turbulent mixing, and applied in LES of the Sydney burner. The PDF models are a leading source of uncertainty in the LES calculations, but appropriate models can be selected using the method developed in this dissertation. The relationship between the models and high-performance computing resources is considered, and a novel ``convolution-on-the-fly" approach for the PDF models is implemented to better leverage the capabilities of current computing hardware, enabling faster computation and application of more complex models.