Session Z11: Reacting Flows: Extinction and Ignition and Chemical Kinetics (12:15pm - 1:00pm CST)

Predicting the evolution of chemical species using Neural ODEs

While running computational fluid dynamics simulations of combustion devices, one of the main bottlenecks remains the computation of the chemical species' source terms and integrating them. In recent years, deep learning-based methods have emerged as a promising approach to solve this problem, where artificial neural networks have been used to predict the source terms of chemical species as a function of the chemical state of the system. One main drawback of machine learning-based methods introduced in previous studies is that they minimize the \textit{a priori} error (i.e., error during training), but do not guarantee that the errors during deployment in CFD will be small. In this work, we propose to address this issue by applying a recent class of neural networks, Neural ODEs, to learn to predict chemical source terms as functions of the current state. This approach is applied to the problem of a perfectly stirred reactor (PSR), and subsequently, to the simulation of a turbulent non-premixed flame in a mixing layer. It is shown that even when the dimensionality of the thermochemical manifold is trimmed to remove redundant species, the proposed approach accurately reproduces the results obtained with detailed chemical mechanisms, at a fraction of the computational cost.   

Skeletal Reaction Model Generation with Optimally Time Dependent Modes

Sensitivity analysis framework based on optimally time dependent (OTD) modes is introduced and demonstrated for generating skeletal kinetic models. This framework expands the sensitivity matrix into a finite-dimensional, time dependent, orthonormal basis which captures directions of the phase space associated with transient instabilities. These directions highlight the active reaction paths and active species at each time instance. Evolution equations for the orthonormal basis and the projections of sensitivity matrix into the basis are derived. For demonstration, sensitivity analysis is conducted of constant pressure hydrogen-air and ethylene-air burning in a zero-dimensional reactor and new skeletal models are generated. The flame speed, ignition delay and extinction curve of resulted skeletal models are compared with the same results from reaction networks.   

Uncertainty quantification for LES of auto-igniting flames

In combustion studies, state of the art reactive Large Eddy Simulation, or LES, is praised for its high-fidelity, due to its high resolution in both space and time. However, LES still carries uncertainties which can significantly impact the quantities of interest. These uncertainties can come, for example, from model calibration, operating conditions, exact composition of complex fuels, etc. In this context, the propagation of these uncertainties through the simulation is essential to gauge the result's confidence. However, as LES is expensive, sample-efficient methods are needed to make uncertainty propagation affordable.\\ This presentation will focus on proposing a method to propagate chemical and experimental uncertainties in the LES of the $H_2$ Cabra flame, which is an auto-igniting flame. A proper uncertain dimension reduction will be inferred from a reduced - yet representative - physical problem. Several surrogate modelling methods and their associated design of experiments will then be compared on this same reduced physical problem to put forward a suitable method for uncertainty propagation through LES.   

Light-round Simulations of an Annular Spray Combustor with Ambient Temperature Walls

Numerical simulations of ignition in annular aeronautical combustors have progressed thanks to experimental data on the burner-to-burner flame propagation. This last phase of ignition is known as light-round. Large-Eddy simulations of the liquid-fueled annular combustor MICCA-spray featuring sixteen swirled injectors have followed the trends observed experimentally. Recent studies have proven that the wall temperatures strongly affect the light-round duration, which has not been satisfactorily retrieved in LES so far. Indeed, a priori studies suggest that variable thermodynamic properties of the boundary layer must be taken into account to improve the prediction of wall heat losses. Furthermore, the combustion model previously relied on the assumption of a constant flame wrinkling parameter. This appears to be inappropriate as shown by dedicated simulations in which the wrinkling parameter is computed dynamically. Both issues, which have only been studied separately before, are addressed in a light-round simulation of the MICCA-spray combustor using an Euler-Lagrange formalism for the liquid phase, a dynamic evaluation of the wrinkling parameter and a novel approach for wall modeling. The impact on the light-round duration is compared to available experimental data.   

Ignition of a fuel--oxidizer interface by laser-induced breakdown

Laser-induced breakdown of a gas produces a high-temperature plasma kernel that expands rapidly and ejects hot gas that can travel several times the kernel size to ignite fuel. Using detailed simulations, we show that the post-breakdown flow and ignition dynamics of such a configuration are highly sensitive to breakdown proximity to the fuel--oxidizer interface and molecular weight disparity. The breakdown-induced flow can enhance, delay, or completely suppress ignition depending on these factors and even subtle alterations to the plasma kernel geometry. This is especially pronounced for a hydrogen--oxygen system. The heat of radical recombination is also surprisingly important as relaxing kernel remnants advect towards fuel. The induced-flow sensitivity can be leveraged in a homogeneous-mixture dual-pulse configuration, where pulse timing and position can enhance dispersal of hot gas and increase the burning rate of nascent flames. Hydrodynamic coupling with the nonequilibrium plasma is also assessed, and it is found that electron recombination can enhance the plasma kernel expansion.   

Lewis number effects on extinction of premixed flames under oscillating strain rates.

Flame extinction due to transient and oscillating flow strain is common in combustion devices, such as IC engines and gas turbines. To evaluate the role of differential diffusion on such extinction, we present an experimental study of a counterflow premixed twin-flame under oscillating strain with varied Lewis numbers (Le), defined as the ratio of the thermal diffusivity of the mixture to the mass diffusivity of the deficient species. By using methane and propane as fuels, the measured instantaneous strain rates required for extinction at various oscillation frequencies and mean strain rates were recorded and compared for both rich and lean mixtures. At low mean strain rates, the extinction was found to be controlled by flow reversal at nozzle. However, for relatively large mean strain rates, normal extinction was achieved for all flames considered. The maximum strain rate for Le \textgreater 1, was found to be greater than the steady state extinction strain rate. While for Le \textless 1 flames, the maximum strain rate at extinction was found to be insensitive to oscillations and approximately equal to the steady state extinction strain rate. This distinctively different behavior depending on the non-unity nature of the Lewis number is analyzed using a time-scale analysis.   

Pareto-optimal modeling of autoignition in a turbulent methane jet-in-hot-coflow configuration

Autoignition plays a key role in several industrial applications. Modeling such flame regimes is challenging as they exhibit multi-mode combustion. The Pareto-efficient Combustion (PEC) framework was developed to model multi-mode combustion by assessing the compliance of combustion models to the underlying flow physics by means of a drift term. To model this transient ignition regime, the PEC-framework is extended by considering the ignition as a process quantity. The resulting PEC-formulation is applied to simulations of ignition in a jet-in-hot-coflow burner and two combustion sub-models are considered: Finite-Rate Chemistry (FRC) and steady flamelet (FPV). It is shown that a monolithic FPV-model significantly mispredicts the flame lift-off and ignition time. The novel PEC-formulation is able to accurately capture the flame stabilization and autoignition time, resulting in comparable accuracy to that of a monolithic FRC simulation; however, as reduction of the computational cost in excess of 60%.