Session P04: Reacting Flows: Theory, Modeling, and Simulations (3:10pm - 3:55pm CST)

Simulations of shock-induced combustion using adaptive mesh refinement algorithm on mapped meshes

Shock-induced combustion (SIC) has gained interest as a promising combustion mode for hypersonic airbreathing propulsion devices. We use adaptive mesh refinement (AMR) on mapped meshes to resolve the complex interaction between the bow shock and reaction wave of SIC, where the body-fitted meshes are refined dynamically. This technique is applied to simulate Lehr's experiments, in which projectiles at a Mach number from 4.18 to 5.11 travel through hydrogen/air. The 2-D axisymmetric Euler equations with detailed chemical kinetics are solved within the AMROC finite volume framework. A second-order accurate MUSCL-Hancock scheme with Minmod limiter is used for the reconstruction. The inviscid flux on mapped meshes is evaluated by a rotating advection upstream splitting method (AUSM). Godunov splitting is adopted for the reactive source term and the Jachimowski hydrogen/air reaction mechanism is employed. The results show that the computed oscillation frequencies, observed in the stagnation point pressure, are in good agreement with the frequencies from Lehr's experiments when the inflow is subdetonative. The simulations predict an irregular and unstable oscillation when the inflow is superdetonative and a detonation occurs at the front of the bow shock occasionally.   

Leveraging In-Situ Adaptive Manifolds for computationally efficient simulations of turbulent combustion with multiple and/or inhomogeneous inlets

Modeling turbulent combustion with reduced-order manifold approaches involves projecting the thermochemical state onto a lower-dimensional space comprising just a few manifold variables, substantially reducing computational cost compared to brute-force modeling. The thermochemical state is traditionally precomputed by solving manifold equations and pretabulated in a thermochemical database. Describing complex turbulent combustion processes featuring multiple and/or inhomogeneous inlet streams requires multiple mixture fractions (manifold variables) and a corresponding increase in the number of precomputed manifold solutions. The memory required to store these solutions, many of which are not even utilized in a CFD calculation, precludes the use of pretabulation for configurations requiring more than a couple of mixture fractions. In this work, a recently-developed approach termed In-Situ Adaptive Manifolds (ISAM) overcomes this limitation by computing manifold solutions `on-the-fly' and reusing them with In-Situ Adaptive Tabulation (ISAT), enabling the utilization of manifold-based turbulent combustion models for complex inlet configurations. The performance of ISAM is evaluated via LES calculations of multi-stream turbulent flames.   

Closure modeling for the conditional momentum equation in turbulent premixed combustion

A manifold-based approach that relies on conditionally averaging the momentum equation against a progress variable has been recently proposed for modeling of combustion heat release effects on turbulence in turbulent premixed combustion. In this work, closure models for all relevant (conditional) unclosed terms in the conditional momentum equation are developed. Each of the models considers the relative contributions of combustion heat release and turbulent shear as distinct components to model flames at any finite Karlovitz number. The closure models are validated \textit{a priori} against two DNS databases of turbulent premixed hydrogen/air planar jet flames at low and high Karlovitz numbers, where the influence of combustion heat release on turbulence is significant at low Karlovitz number and insignificant at high Karlovitz number. As a preliminary step, \textit{a posteriori} validation of the models against the DNS databases is performed.   

Bayesian Neural Networks for Assimilation of Experimental Data into a G-equation Flame Model

In the G-equation flame model, a thin laminar flame propagates at a fixed speed into a premixed gas moving at a prescribed velocity. This model can simulate cusp formation and flame pinch-off, while remaining computationally cheap, making it attractive for design optimization. The G-equation model is sufficiently accurate to provide a qualitative description of a flame. In order to be quantitatively accurate, however, the model parameters need to be fitted to experimental or numerical data. In this study we use Bayesian neural networks (BNNs) to fit the model parameters and their uncertainties for a conical premixed Bunsen flame. The BNNs are trained on a library of reference flames created with G-equation simulation data and are then applied to experimental data. This is an extremely fast way to assimilate experimental data into a model. The assimilation is orders of magnitude faster than the current state of the art. The method itself is general and can assimilate data from any flame that can be modelled by the G-equation.   

Ensemble Predictions of Wildfire Spread Through TPU-Compatible TensorFlow Acceleration

Wildfire spread is a complex process that is sensitive to numerous environmental factors, making it difficult to simulate from first principles. Percolation models offer a simple framework in which to implement physical relationships derived from observations. This research presents a percolation-based wildfire spread model, weighted by environmental parameters such as fuel characteristics, wind, and slope. It is implemented using the TensorFlow machine learning platform, which provides useful data structures as well as an interface with Tensor Processing Units (TPUs). Originally designed for machine learning, these processors feature linear algebra optimizations, making them highly efficient for ensemble simulations. By formulating the model as a discrete convolution, this implementation can efficiency simulate heat accumulation and fire-spread dynamics. This additionally allows for a simple extension to compute multiple wildfire cases simultaneously. When there is uncertainty in input parameters, aggregate spread paths may be generated using multiple cases which span this uncertainty, thereby increasing confidence in the result.   

Not Participating

One of the main limitations when deploying tabulated flamelet models for computational simulations of practical combustion systems is the excessive computational storage requirements associated with high-dimensional flamelet tables. This study introduces and validates an approach for circumventing this issue, by learning the table using an ensemble of deep neural networks to predict the species mass fractions and progress variable source term as functions of the control variables. In the proposed approach, a mixture of experts (MoE) technique is used, where multiple artificial neural networks, called experts, are trained concurrently. The artificial neural networks compete for training samples within the manifold, and another network, known as the gating network, rewards the experts that have superior training performance with stronger training signals. This leads to specialization of the experts in different portions of the table. The MoE framework is applied to unsteady flamelet progress variable (UFPV) modeling of an n-dodecane Spray A flame based on the Engine Combustion Network (ECN). It is demonstrated that the MoE-based tabulation can accurately predict global flame characteristics such as the lift-off length and ignition delay over a range of ambient conditions.   

The Influence of Turbulence on Mineral Scale Deposition

Scale deposition in pipelines for oil and gas production has proven a complex problem to predict. Therefore, a better understanding of the influence of flow characteristics on mixing is sought. This study examines the influence of turbulent transport of chemical species on surface reaction rates through experiments and numerical modelling. A Taylor-Couette flow cell with an axial inflow of two incompatible brines has been used for experimental studies. The outflow concentrations are measured to quantify the real-time mass deposition of mineral scale. Different turbulence intensities are obtained by adjusting the rotational speed and flow rate. A CFD model is implemented in STAR--CCM+ to model flow characteristics using a combination of LES and RANS. The LES model is used to calibrate the RANS model including the turbulent Schmidt number. The RANS model uses the Reynolds Stress Tensor model to simulate the slower transients of the experiments. A customized numerical chemical model for saturation, reaction and wall deposition has been implemented in the CFD models. The obtained results are used to describe the importance of flow properties on the prediction and prevention of scale buildup.   

A numerical study of the influence of pilot fuel injection timing on air-fuel mixing and combustion of a dual fuel, two-stroke marine engine

Three-dimensional computational fluid dynamic (CFD) simulations of a dual fuel marine engine are conducted where a small amount of pilot diesel fuel is injected near top dead center, follow by a direct injection of natural gas. A skeletal chemical n-heptane mechanism (56 species) which consists of the methane oxidation reaction pathways, is utilized to simulate the oxidation of diesel and methane as well as the formation of emissions in the dual-fuel combustion. The n-heptane and methane IDT were evaluated by performing homogeneous reactor calculations. A phenomenological soot model is also considered to investigate the formation of soot. The CFD model is evaluated using the experimental pressure measurement. Effects of pilot diesel fuel injection timing is studied with an emphasis on the associated combustion and emission characteristics. It is found that by retarding the pilot fuel injection timing, a larger unburned methane cloud is formed inside the combustion chamber prior to being ignited by the pilot fuel which leads to high heat release rate. Furthermore, retarding the pilot fuel injection timing leads to a higher amount of formed soot as well as higher unburned CH$_{4}$ and CO emissions. On the other hand, advancing the diesel injection timing increases the NO emission.   

A framework for simulating precipitate reactions in microfluidic devices

Chemical processes within flows are ubiquitous. There exists an important class of reactions that result in a phase change from liquid to solid: precipitation reactions. Inspired by recent microfluidic experiments, this talk describes a novel mathematical framework for handling such reactions occurring within a slow-moving fluid flow. A key challenge for precipitate reactions is that, in general, the location of the developed solid is unknown a priori. To model this situation, we use a multiphase framework with fluid and solid phases; the aqueous chemicals exist as scalar fields that react within the fluid to induce phase change. To demonstrate the functionality of this framework, we conduct full-scale simulations in a realistic microfluidic geometry. The framework can be applied to precipitate reactions where the precipitate greatly affects the surrounding flow, a situation appearing in many laboratory and geophysical contexts including the hydrothermal vent theory for the origin of life. More generally, this model can be used to address low Reynolds number fluid–structure interaction problems that feature the dynamic generation of solids.   

Investigating the effect of in-cylinder gas compositions on sulfuric acid formation and condensation using CFD modeling under large two-stroke marine engine-like conditions

A computational fluid dynamic simulation is utilized to model the formation and condensation of sulfuric acid (H$_{2}$SO$_{4}$) under large two-stroke marine diesel engine like conditions. A skeletal chemical mechanism coupled with a sulfur subset is used to simulate the combustion process and the formation of sulfur oxides (SO$_{x}$) and H$_{2}$SO$_{4}$. A fluid film model coupled with the Eulerian in-cylinder gas phase describes the condensation of H$_{2}$SO$_{4}$. Exhaust gas recirculation (EGR) is a well-known method to decrease the nitrogen oxides (NO$_{x}$) emission. However, one of the side-effects of EGR may be an increase in sulfuric acid condensation which leads to cold corrosion of liner. In this study the initial in-cylinder gas compositions are varied to imitate different EGR compositions (wet and dry) and the associated effects on the formation and condensation of H$_{2}$SO$_{4}$ are investigated. It is found that the amount of SO$_{x}$ formation is similar for these two kinds of EGR which is lower than base case (without EGR). The interesting finding is that the H$_{2}$SO$_{4}$ vapor formation for wet and dry EGR is higher and lower than the base case, respectively. The current CFD results show that applying EGR does not increase the H$_{2}$SO$_{4}$ condensation.   

Not Participating

Moisture content in thermally thin fuels has significant impact on the intensity and rate of spread of wildland fire. Fuel moisture within coupled CFD - wildland fire behaviour models can be handled using different techniques when multiple different fuels are present, such as a mix of live canopy foliage and dead forest floor material. One commonly-used method is a cell-averaged moisture content, which can lead to a loss of model sensitivity to fine fuel moisture content, especially near the ground. Furthermore, averaging moisture content without explicitly tracking wet and dry fuels and their inter-exchange of energy and mass before and during combustion can also lead to inaccuracies in predicted wildfire behaviour. To explore these effects, we use FIRETEC, a coupled LES-wildland fire behaviour model. We simulate a well-documented experimental burn that was conducted in May 2019, in Alberta, Canada. Simulations are completed using both a single averaged fuels version of FIRETEC and a multiple fuels version of FIRETEC, which allows for separate fuels with separate fuel attributes. Comparisons in fire behaviour metrics between these simulations are made.   

Investigation of the laser-induced ignition in a prototype rocket combustor

Accurate prediction of the reliability of laser-induced ignition of liquid propellants is crucial for aerospace propulsion, and has been challenging due to the short time scale and complex nature of ignition events in turbulence. In this exploratory study, we numerically analyze the ignition of high-speed, co-flowing jets of gaseous CH$_4$ and O$_2$ in a prototype rocket combustor. We model the deposition of pulsed-laser through point-source heating at downstream locations under conditions of a companion experiment. For specific sets of flow parameters and heating locations, mixing and reaction of the jets are simulated using the HTR solver, an open-source compressible reacting flow solver. The solver employs the Legion runtime system and is optimized for GPU-based, heterogeneous supercomputers. We address various combinations of the duration and the energy of heating, on the orders of $O$(0.1) us and $O$(10-100) mJ, respectively. Obtained data sets are used to draw a probability map of ignition on parameter space, whose success is defined in terms of reaction-rate. Through this map, we assess the posterior probability of ignition with respect to input variables. Finally, we discuss the sensitivity of this inference to upstream conditions of the jets.   

Reacting Simulations of Multi-mode Combustion in a Rapid Compression Machine under HCCI and SICI Modes

Multi-mode combustion is of practical relevance due to the possibility of improving engine efficiency, emissions, and performance over single-mode combustion. In particular, Spark Induced Compression Ignition (SICI) can improve the load range and timing consistency of Homogeneous Charge Combustion Ignition (HCCI) while retaining HCCI’s advantage in low NOx emission and efficiency. We investigate multi-mode combustion dynamics by performing reacting simulations of HCCI and SICI modes in a Rapid Compression Machine (RCM) configuration of Strozzi et al. (2019). The compression phase is simulated using a dynamic mesh approach to resolve the flow field’s heterogeneity at Top-Dead-Center (TDC). The impact of this heterogeneity on combustion dynamics after TDC through hot-spot autoignition is investigated.   

Influence of Thermal Radiation on Layered Coal-Dust Explosions

There is significant debate on the role of thermal radiation on the propagation of coal-dust explosions. To date, numerical simulations of dust explosions either completely neglect radiation or use crude approximations. Here, we use numerical simulations to explore the role of radiation on coal-dust explosions by solving the radiation transport equation (RTE). The multiphase model couples multidimensional kinetic theory-based granular Eulerian multiphase model to a compressible reactive gas. The RTE is coupled to the flow using the FP3 approximation. Radiation is assumed to be gray with cold, black boundaries. Global reaction models are used to describe coal devolatilization, char combustion, and volatile (methane) combustion. Results from the simulations show that radiation may have a significant influence on the dust explosion. In some cases, radiation enhances flame propagation by transferring additional heat to cold reactants, in other cases radiation losses enhance quenching of the flame. Radiation also has significant impact on the structure of the flame and peak flame temperature. Our ongoing work is exploring the influence of modeling choices such as chemical kinetics and spectral accuracy.   

Numerical simulation of a triple flame in a swirling flow

A triple (tribrachial) flame consists of a rich premixed flame, a lean premixed flame, and, in between, a diffusion flame, all of which merge at a single point. Recent three-dimensional, unsteady simulations have shown that the blue whirl has a triple-flame structure, in which the merging point of the triple flame forms as a ring surrounding the bubble mode of vortex breakdown within a swirling flow. Now, in order to study how triple flames interact with swirling flows, we examine a configuration in which a swirling flow first evolves into the bubble mode of vortex breakdown with premixed fuel and air injected into the vortex core. After a quasi-steady state is established, the fuel-air mixture is ignited in the upstream portion of the bubble region and a triple flame forms within the vortex core. The flame is lifted above the inlet plane and remains in a stable position throughout the computation. We present and discuss the flame and flow structure and compare it with the blue whirl.   

The physics of mixing and reaction of co-axial and cross-flow jets with disparate viscosity

Mixing of miscible fast reacting liquids with disparate viscosity in a co-axial jet or jet in cross-flow is of great interest in several important industrial applications. Higher reaction rates turn the problem into mixing limited reactions. Computational analysis and experimental diagnostics of such a system are both challenging due to spatial and temporal scale separation. In this study, we use Particle-image Velocimetry (PIV) and Planar Laser induced Fluorescence (PLIF) imaging to measure the velocity field and mixture fraction, in conjunction with large-eddy simulation (LES) to investigate the details of the effect of large viscosity ratio on reaction yields. At high viscosity ratios the inner jet is turbulent, the outer jet is laminar and the downstream flow in the pipe appears to be 'fully developed' laminar flow. In this case, it is shown that folded segregated patterns in mixture fraction persist far downstream despite the flow being fully developed where mixing is primarily through diffusion and not convection. The experimental and LES results for the mixture fraction and reaction yields in the range of viscosity ratios from 0.3 to 250 will be presented. Furthermore, the unique mixing structure in this flow will be explained with gradient diffusion hypothesis.   

Machine learning assisted modeling of mixing timescale for LES/PDF of high-Ka turbulent premixed combustion

A power-law scaling mixing model has been developed recently for modeling mixing in the large-eddy simulation (LES)/probability density function (PDF) modeling of high-Ka turbulent premixed combustion. In the power-law scaling model, two model parameters need to be specified and empirical models have been developed to specify the model parameters. It is found that the empirical models are limited to accurately represent the model parameters found in the DNS data for a turbulent premixed round jet flame from Sandia. In this work, we explore the feasibility of using machine learning for specifying the model parameters in the mixing timescale model. The Sandia DNS data are used as the training data and machine learning models are constructed by using the random forest algorithm. Different input parameters are examined for machine learning. The excellent performance of the machine learning model for specifying the model parameters in the power-law mixing timescale is demonstrated in the DNS flame through a priori analysis.   

Large-eddy simulation/probability density function modeling of a high Karlovitz number turbulent premixed jet DNS flame

Simulation studies of a high Karlovitz number turbulent premixed jet DNS flame are conducted by using the combined large-eddy simulation (LES) and transported probability density function (PDF) methods to assess the model's predictive capability for the relevant premixed combustion regime. The interaction by exchange with the mean (IEM) model is used for mixing modeling with a recently develop mixing timescale model based on a power-law scaling. Two model parameters in the power-law scaling model are specified by a machine learning approach. The performance of the machine learning model is compared with an empirical model for the specification of the mixing model parameters. A consistent Eulerian Monte Carlo field method is used for solving the transported PDF equation efficiently and consistently. The impact of the model inconsistency in the traditional Eulerian Monte Carlo field method is examined. The DNS data are used to validate the model predictions. The predictive performance of the employed is assessed in detail.