The Development and Validation of a Simplified Soot Model ......Figure 4.1 – Schematic of the...
Transcript of The Development and Validation of a Simplified Soot Model ......Figure 4.1 – Schematic of the...
The Development and Validation of a Simplified Soot Model for use in Soot Emissions Prediction in Natural Gas Fuelled Engine Simulations
by
Justin Jeekee Shum
A thesis submitted in conformity with the requirements for the degree of Masters of Applied Science
Graduate Department of Mechanical & Industrial Engineering University of Toronto
© Copyright 2012 by Justin Jeekee Shum
ii
Abstract
The Development and Validation of a Simplified Soot Model for use in Soot Emissions Prediction in Natural Gas Fuelled Engine Simulations
Justin Jeekee Shum
Masters of Applied Science Graduate Department of Mechanical & Industrial Engineering
University of Toronto 2012
This study employs a novel approach in order to satisfy the need in industry for a computationally
inexpensive means to modelling soot formation in engines fuelled by natural gas. The complex geometries
found in practical combustion devices along with the requirement to solve turbulent, chemically reacting,
and multi-phase flows necessitates this goal. A two-equation model, which tracks soot mass and soot
number density, is employed. The goal is to apply this model in engine simulations at Westport Innovations,
an industry partner.
Experimental data is used to validate the model in various operating conditions. Numerical data
obtained from a detailed sectional soot model is also used to augment available validation data, especially
with respect to soot formation/oxidation mechanisms. The developed model shows good agreement
compared to experimental data and the detailed sectional soot model among all cases considered and will
be further tested and applied in Westport’s natural gas engine simulations.
iii
Acknowledgements First and foremost, the author would like to acknowledge his supervisor Professor M.J. Thomson for his
valued guidance, direction, and support as without it, this project would not have been possible. The author
would also like to acknowledge the assistance provided by Professor S.B. Dworkin as his expertise in soot
modelling was most helpful in many stages of this work. Special thanks are also given to Dr. Q. Zhang who
provided his prior work on the two equation soot model. The author also acknowledges the Natural
Sciences and Engineering Research Council of Canada, Westport Innovations, and Dr. B. Wasmund for
financial support. The author would also like to thank Dr. N. Slavinskaya and Professor U. Riedel of the
German Aerospace Centre (DLR) and Dr. J. Huang of Westport Innovations for providing the chemical
mechanisms, thermodynamic data, and transport data for methane/air combustion. The author would also
like to further recognize Dr. J. Huang's contributions to the development and validation of the simplified
soot model. Further acknowledgements are given to Dr. G. McTaggart-Cowan, Professor S. Rogak, and the
rest of the APC team for their feedback and advice. The SciNet HPC Consortium is also acknowledged for
providing the computational resources necessary to complete this project. SciNet is funded by: the Canada
Foundation for Innovation under the auspices of Compute Canada; the Government of Ontario; Ontario
Research Fund - Research Excellence; and the University of Toronto. Many thanks are also warranted to the
author's peers, colleagues, and friends at the Combustion Research Laboratory who all found the time to
provide support in too many ways to list. Finally, the author would like to acknowledge his family, friends,
and Varsity fencing teammates/coaches for their endless encouragement, helpfulness, and much needed
distractions.
iv
Contents
1. Introduction ................................................................................................................................................1
1.1. Motivation ...........................................................................................................................................2
1.2. Objectives............................................................................................................................................3
2. Background and Literature Review .............................................................................................................5
2.1. An Introduction to Soot ......................................................................................................................5
2.1.1. Soot Characteristics and overview ..............................................................................................5
2.1.2. Soot nucleation/inception ..........................................................................................................7
2.1.3. Soot particle surface growth .......................................................................................................9
2.1.4. Soot particle oxidation ............................................................................................................. 10
2.1.5. Soot particle agglomeration..................................................................................................... 11
2.2. Current Approaches to Soot Modelling ........................................................................................... 11
2.2.1. Empirical Models ...................................................................................................................... 12
2.2.2. Detailed Models ....................................................................................................................... 14
2.2.3. Semi-empirical models ............................................................................................................. 16
2.2.4. Particle size distribution and soot aerosol dynamics ............................................................... 19
2.3. Laminar Coflow Diffusion Flame ...................................................................................................... 21
2.4. Turbulent Combustion Modelling .................................................................................................... 23
3. Model Development Methodology ......................................................................................................... 28
3.1. Experimental Datasets for Model Validation ................................................................................... 28
3.2. Comparison to Sectional Detailed Soot Chemistry Model ............................................................... 29
3.3. Experimental Cases Considered ....................................................................................................... 31
3.4. Considerations for use in Westport Simulations ............................................................................. 32
3.4.1. Coupling of Soot Model to Gas Phase Species Consumption .................................................. 33
3.4.2. Coupling to Radiation Heat Transfer ........................................................................................ 33
4. Mathematical Formulation ...................................................................................................................... 34
4.1. Computational Domain .................................................................................................................... 34
4.2. Governing Equations ........................................................................................................................ 34
4.2.1. Conservation of mass ............................................................................................................... 35
4.2.2. Conservation of momentum .................................................................................................... 35
v
4.2.3. Conservation of species ........................................................................................................... 35
4.2.4. Conservation of energy ............................................................................................................ 36
4.3. Diffusivity of Gaseous Species ......................................................................................................... 36
4.4. Radiation Heat Transfer ................................................................................................................... 37
4.5. Simplified Soot Model ...................................................................................................................... 38
4.5.1. Soot transport equations ......................................................................................................... 42
4.6. Numerical Method ........................................................................................................................... 42
4.6.1. Mesh and boundary conditions ............................................................................................... 43
4.6.2. Parallel Computation ............................................................................................................... 45
4.7. Detailed Sectional Soot Model......................................................................................................... 46
5. Development and Validation of Soot Model ........................................................................................... 48
5.1. Chapter Outline ................................................................................................................................ 48
5.2. Sensitivity Analysis of Parameter Terms in Simplified Model.......................................................... 48
5.3. Model Development at 1 atmosphere using the Smooke et al. [61] data set ................................. 56
5.3.1. Comparisons to numerical data from detailed sectional soot model ..................................... 58
5.4. Model Development at 1 atmosphere using the Schittkowski et al. [76] dataset .......................... 64
5.4.1. Comparisons to numerical data from detailed sectional soot model ..................................... 69
5.5. Results at elevated pressures .......................................................................................................... 70
5.5.1. Comparisons to numerical data from detailed sectional soot model ..................................... 73
5.6. Preliminary Results in Engine Simulations ....................................................................................... 74
5.7. Model Improvements ...................................................................................................................... 74
5.7.1. Oxidation Mechanism of Soot Model ...................................................................................... 75
5.7.2. Updated Model Parameters and Improved Results ................................................................ 78
5.8. Effect of Coupling – Gas Phase Species Consumption and Radiation .............................................. 87
5.9. Computational Cost Comparison ..................................................................................................... 89
6. Concluding Remarks ................................................................................................................................. 91
6.1. Conclusion/Summary ....................................................................................................................... 91
6.2. Future Work ..................................................................................................................................... 92
vi
List of Figures
Figure 2.1 – Example of soot aggregate structure in diesel exhaust. Taken from [6]. .......................................6
Figure 2.2 – Representation of soot formation in premixed flames. Adapted from [19]. ..................................8
Figure 2.4 – Five major components of soot modelling. ................................................................................. 12
Figure 2.5 – Typical setup of a laminar coflow diffusion flame adapted from [61] ......................................... 22
Figure 2.6 – Soot formation zones in a coflow diffusion flame along with soot aggregate structure evolution.
Adapted from [10]. .......................................................................................................................................... 23
Figure 3.1 – Workflow diagram of project ....................................................................................................... 28
Figure 3.2 – Adapted workflow diagram of project ......................................................................................... 30
Figure 3.3 – Schematic of high pressure combustion rig taken from [78]. ...................................................... 32
Figure 3.4 – (a) Table summarizing burner geometry and operating conditions of experiments (b) Diagram of
coflow burner defining and .................................................................................................................... 32
Figure 4.1 – Schematic of the computational domain (greyed out area) super-imposed on a diagram of a
typical laminar coflow diffusion flame. Also illustrated is the orientation of the coordinate system used.
Note that the illustration is not to scale. Taken from [75]. ............................................................................. 34
Figure 4.2 – Diagram of typical non-uniform mesh employed in the simulations presented. Adapted from
[75]. .................................................................................................................................................................. 45
Figure 5.1 – Diagram of the “wing” and “centreline” regions of a typical flame. ........................................... 50
Figure 5.2 – Sensitivity of sooting behaviour to the pre-exponential value of soot inception, . Solid
lines show peak values at the wing of the flame and dashed lines show values in the centreline of the flame.
A logarithmic scale is applied to the Y-axis. ..................................................................................................... 50
Figure 5.3 – Sensitivity of sooting behaviour to the pre-exponential value of soot surface growth, .
Solid lines show peak values at the wing of the flame and dashed lines show values in the centreline of the
flame. A logarithmic scale is applied to the Y-axis. Legend is in Figure 5.2. .................................................... 52
Figure 5.4 – Sensitivity of sooting behaviour to the pre-exponential value of soot oxidation via O2, .
Solid lines show peak values at the wing of the flame and dashed lines show values in the centreline of the
flame. A logarithmic scale is applied to the Y-axis. Legend is in Figure 5.2. .................................................... 53
Figure 5.5 – Sensitivity of sooting behaviour to the pre-exponential value of soot oxidation via OH, .
Solid lines show peak values at the wing of the flame and dashed lines show values in the centreline of the
flame. A logarithmic scale is applied to the Y-axis. Legend is in Figure 5.2. .................................................... 54
Figure 5.6 – Sensitivity of sooting behaviour to the selected incipient particle diameter, with the default
value at 12 nm. Solid lines show peak values at the wing of the flame and dashed lines show values in the
centreline of the flame. A logarithmic scale is applied to the Y-axis. Legend is in Figure 5.2. ........................ 55
vii
Figure 5.7 – Sensitivity of sooting behaviour to the selected agglomeration rate, . Solid lines show peak
values at the wing of the flame and dashed lines show values in the centreline of the flame. A logarithmic
scale is applied to the Y-axis. Legend is in Figure 5.2. ..................................................................................... 56
Figure 5.8 – Soot volume fraction profiles at different axial heights above the burner. Z1, Z2, Z3, and Z4
correspond to heights of 2.0, 2.25, 2.5, 2.75 cm for experimental measurements and computations from the
simplified model for the Smooke et al. [61] flame. For computations from the sectional model, 0.6 cm was
added to each axial height to account for the delay in PAH formation. ......................................................... 58
Figure 5.9 – Contours of soot volume fraction (ppm) generated using the simplified model (left) and
detailed sectional model (right) for the Smooke et al. [61] flame. ................................................................. 59
Figure 5.10 – Contours of soot number density of aggregates (#/cc) generated using the simplified model
(left) and detailed sectional model (right) for the Smooke et al. [61] flame. .................................................. 60
Figure 5.11 – Contours of soot aggregate mass averaged diameters (nm) generated using the simplified
model (left) and detailed sectional model (right) for the Smooke et al. [61] flame. ...................................... 61
Figure 5.12 – Example of a pathline of maximum soot. .................................................................................. 62
Figure 5.13 – Diagram of methodology used to compare inception mechanisms between the simplified code
and the detailed code. ..................................................................................................................................... 63
Figure 5.14 – Calculated plots of soot inception rate, surface growth rate, and oxidation along the pathline
of maximum soot for the Smooke et al. [61] flame. The Y-axis is plotted on a logarithmic scale. ................. 63
Figure 5.15 – Contours of soot volume fraction side by side with experimental measurements in the
Schittwkowski et al. [76] flame. Experimental measurements are on the left side of the flame and
computations are on the right side of the flame. Results of the simplified model are shown in (a) and results
of the detailed model are shown in (b). .......................................................................................................... 65
Figure 5.16 – Contours of soot particle diameter in the Schittkowski et al. [76] flame. (a) shows the
experimental measurements of primary particle diameter on the left and the calculated contour of primary
particle diameter on the right from the detailed model. (b) shows the calculated contour of mass averaged
aggregate particle diameter from the detailed model on the left and the simplified model on the right. .... 67
Figure 5.17 – Contours of soot particle number density in the Schittkowski et al. [76] flame. (a) Shows the
experimental measurements of primary particle number density on the left and the calculated primary
particle number density on the right from the detailed model. Different scales are used for each half of the
flame. (b) Shows the calculated contour of aggregate particle number density from the detailed model on
the left and the simplified model on the right. ................................................................................................ 68
Figure 5.18 – Calculated plots of soot inception rate, surface growth rate, and oxidation along the pathline
of maximum soot for the Schittkowski et al. [76] flame. The Y-axis is plotted on a logarithmic scale. .......... 70
viii
Figure 5.19 – Contours of soot volume fraction (ppm) with experimental measurements made by Joo and
Gülder [79]. Experimental measurements are on the left and calculated contours from the simplified model
are on the right. Operating pressures are at 10 atm (a), 20 atm (b), and 40 atm (c). ..................................... 71
Figure 5.20 – Contours of soot volume fraction (ppm) with experimental measurements made by Joo and
Gülder [79]. Experimental measurements are on the leftt and calculated contours from the detailed model
are on the right. Operating pressures are at 10 atm (a), 20 atm (b), and 40 atm (c). ..................................... 72
Figure 5.21 – (a) Graph illustrating the correction factors employed. (b) The effect of the correction factors
on the rates predicted by the utilized oxidation models. ................................................................................ 76
Figure 5.22 – Integrated soot volume fraction (ppm · cm2) profiles of the Santoro et al. [74] flames of the F2
Non-smoking flame in (a) and the F4 smoking flame in (b). ............................................................................ 77
Figure 5.23 – Peak values of soot volume fraction predicted by the simplified model compared to
experimental results in the centreline (a) and the wing (b) of the Smooke et al. [61] and Schittkowski et al.
[76] flames. ...................................................................................................................................................... 80
Figure 5.24 – Peak values of mass averaged aggregate particle diameters predicted by the simplified model
compared to the detailed model in the centreline (a) and the wing (b) of the Smooke et al. [61] and
Schittkowski et al. [76] flames. ........................................................................................................................ 81
Figure 5.25 – Peak values of aggregate particle number density predicted by the simplified model compared
to the detailed model in the centreline (a) and the wing (b) of the Smooke et al. [61] and Schittkowski et al.
[76] flames. ...................................................................................................................................................... 83
Figure 5.26 – Peak values of soot volume fraction predicted by the simplified models and detailed model
compared to experimental results from Joo and Gülder [79] in the centreline (a) and wing (b) of the flame.
......................................................................................................................................................................... 84
Figure 5.27 – Peak values of mass averaged aggregate diameter as predicted by the simplified models and
detailed model in the centreline (a) and wing (b) of the Joo and Gülder [79] flame. An uncertainty of
is assumed for the detailed code calculations. ................................................................................................ 86
Figure 5.28 – The effect of coupling on the predicted peak soot volume fractions in the wing and centreline
(CL) of the Smooke et al. [61] flame and the Schittkowski et al. [76] flame. .................................................. 87
Figure 5.29 – Graph illustrating the factor of increase in calculated peak values of soot volume fraction in
the wings and centreline (CL) of the high pressure Joo and Gülder [79] flames. From left to right, the 10, 20,
and 40 atm cases are shown plotted with respect to the maximum soot volume fractions measured in the
experiment. A logarithmic scale is applied to the Y-axis. ................................................................................ 88
ix
List of Tables
Table 2.1 – List of reactions in the HACA surface growth sequence. Adapted from [13]. .............................. 10
Table 4.1 – Summary of reaction rate constants in the Arrhenius form , where units
are in g, cm, mol, s, K. ...................................................................................................................................... 40
Table 4.2 – Summary of geometric properties of meshes used in laminar coflow flame simulations. ........... 44
Table 4.3 – A summary of the major differences between the employed simplified soot model and a
previously developed detailed sectional model. ............................................................................................. 46
Table 5.1 – List of major parameters in simplified two equation soot model. ................................................ 49
The Smooke et al. [61] data set was used as a baseline case upon which to investigate the sensitivity of soot
volume fraction, soot aggregate averaged diameters, soot number density, soot inception rate, soot surface
growth rate, and soot oxidation rate (henceforth collectively referred to as “sooting behaviour”) to the
parameters listed in Table 5.1. Simulations were run where a single parameter was modified while all other
parameters were held constant. Then, the effect on the aforementioned soot details relative to the baseline
case with initial parameters listed in Table 4.1 and Section 4.5 was recorded. This process was repeated for
each parameter listed in Table 5.1. Peak values of sooting behaviour were recorded in both the wing and
the centreline of the flame (illustrated in Figure 5.1)...................................................................................... 49
Table 5.2 – Calculated peak mass averaged aggregate particle diameters (nm) in the Joo and Gülder [79]
flames. .............................................................................................................................................................. 73
Table 5.3 – Calculated peak aggregate particle number densities (#/cc) in the Joo and Gülder [79] flames. 73
Table 5.4 – Summary of parameters in simplified model. See Equation (5.3) for definition of correction
factor (identical for both O2 and OH). Changes are highlighted in red and underlined. .............................. 79
Table 5.5 – Representative comparison of computational costs of running the 2-D laminar flame code with
the detailed section soot model and the simplified two equation soot model. ............................................. 90
x
Nomenclature
Oxidation correction factor parameter
Agglomeration constant
Minimum number of carbon atoms found in a soot particle
Constant pressure specific heat of the mixture
Constant pressure specific heat of the species
Constant pressure specific heat of soot
Primary particle diameter
Mass averaged aggregate diameter of soot
Activation energy
Oxidation correction factor term
Soot volume fraction
Integrated soot volume fraction
Gravitational acceleration
Specific enthalpy of the species
Specific enthalpy of soot
Total number of gaseous species present in the chemical mechanism
Mass flow rate
Molar mass of carbon
Mass of soot
Soot number density
Aggregate number density
Primary particle number density
Avogadro’s number
Pressure
Total radiation heat transfer
Radial direction
Inner radius of fuel tube
Outer radius of fuel tube
Temperature
Oxidation correction factor parameter
Axial velocity
xi
Radial Velocity
Diffusive velocities in the Radial direction of the species
Diffusive velocities in the axial direction of the species
Radial thermophoretic velocities of soot particles
Axial thermophoretic velocities of soot particles
Molecular weight of the species
Molecular weight of soot
Mass fraction of the species
Soot mass fraction
Axial direction
Mixture fraction
Greek Symbols
Portion of available surface sites on soot particle for chemical reaction
Scalar dissipation
Thermal conductivity of the mixture
Boltzmann’s constant
Dynamic viscosity
Molar production rate of the species per unit volume
Molar production rate of soot per unit volume
Collisional efficiency of OH molecules for soot oxidation in simplified model
Mixture density
Density of soot
xii
Acronyms
CL Centreline (of flame)
CRL Combustion Research Laboratory (at the University of Toronto)
DOM Discrete Ordinate Method (Radiation Model)
HACA Hydrogen-Abstraction-Carbon-Addition
ILDM Instrinsic Low-Dimensional Manifold
NOx Nitric Oxide
PAH Polycyclic Aromatic Hydrocarbon
SLFM Steady/Stationary Laminar Flamelet Model
ULFM Unsteady Laminar Flamelet Model
1
1. Introduction
The work described in this thesis represents one component of a collaborative research project
between the Combustion Research Laboratory (CRL) supervised by Professor Thomson at the
University of Toronto, Professor Rogak and his research group at the University of British Columbia, and
Westport Innovations with the aim to aid in the research and development of natural gas fuelled heavy
duty compression ignition engines.
The development of combustion systems is a complex and time consuming endeavour – while the
basic operating principles of the technology used today have not changed significantly over the last few
decades, there is the ongoing need to address the growing demands of consumers while satisfying
stringent environmental policies. Despite recent advances in research and development made to move
away from combustion-based devices in an effort to mitigate the effects of global warming and limit
harmful pollutant emissions, alternative technologies are unlikely to become viable for widespread use
for many years to come [1]. This is especially true in long-distance transportation applications where
the use of conventional fossil fuels is necessitated by their high energy density relative to alternative
sources such as battery storage. As a result, combustion systems must be able to meet the ever
increasing legislated limits on harmful pollutant emissions such as nitric oxides (NOx), carbon
monoxide, volatile organic compounds (VOCs), and particulate matter (PM).
The design of a combustion system is often an iterative process from the early stage of conception,
its eventual design and manufacture, and finally, testing and validation. As such, projects can quickly
become both time consuming and expensive as individual prototypes and test rigs are either
manufactured or purchased along with the necessary data extraction and analysis tools. To mitigate
this, numerical models and predictive tools are often used to assist in the development of combustion
systems. Another advantage of numerical modelling is that it provides insight to combustion
2
phenomena, fluid flows, and other behaviour that is normally not directly observable by experimental
procedures. This type of insight can help to design more efficient and less polluting systems.
1.1. Motivation
The focus of this work lies in the development and validation of a numerical model that can
accurately predict the emissions of the combustion-generated particulate matter known as soot. The
aim is then to apply this model in compression-ignition natural gas fuelled engine simulations in order
to gain a better understanding of how and why soot is emitted from these engines. This knowledge will
facilitate the ability for rapid and iterative engine design in the difficult task of balancing performance
and pollutant emissions.
The formation of soot particulate matter in combustion and its subsequent release into the
atmosphere has received attention in research and industry due its adverse environmental and health
effects. Individual soot particles can often be formed on the scale of 100 nanometers or less. This poses
a significant health risk as inhalation of soot particles to the lungs can cause inflammation and cancer.
These fine particles are then able to migrate through the bloodstream and damage other vital organs
such as the heart or brain [2]. Soot is also known to play a major role in global warming phenomena
and is thought to be a major contributor to global warming effects behind CO2 [3]. From a design
standpoint, the deposition of soot particles on equipment can also pose a problem for issues such as
maintenance or thermal loading (due to the high thermal absorptivity of soot particles). Also important
to note is that the formation of soot in itself represents a degree of inefficiency in converting the
energy contained in a fuel to useful work. Thus, it is easy to see why accurately predicting soot
formation and emission in combustion devices is an important endeavour.
The mechanisms of soot production in combustion are very complex and not yet fully understood.
Modelling soot formation in various combustion applications is an ongoing area of research that has
received much attention of late [4],[5],[6],[7],[8],[9]. Due to the highly complex nature of soot
3
formation, these models are often computationally expensive as they attempt to recreate our current
understanding of fundamental soot formation/oxidation mechanisms. This study employs a simplified
approach in order to satisfy the need in industry for a computationally inexpensive approach to
modelling soot formation in engines fuelled by natural gas, such as those at Westport. As an engine
designer for natural gas fuelled compression ignition engines, Westport is interested in low
computational cost soot modelling techniques that can allow them to quickly evaluate new engine
concepts. The complex geometries found in practical combustion devices along with the requirement
to solve turbulent, chemically reacting, and multi-phase flows drives the goal of reducing the
computational cost of soot modelling.
The use of natural gas, which is mostly composed of methane, is rapidly becoming an important
alternative fuel in transportation applications. There is a growing interest in using natural gas as a fuel
due to its wider availability (and subsequently, affordability) and lower carbon footprint per unit of
energy. Potential greenhouse gas emissions reductions have been estimated to be over 20% per vehicle
in large scale industrial applications simply by switching from conventional fuels to natural gas [10].
Natural gas has also been demonstrated to be a cleaner fuel with less harmful pollutant emissions
compared to conventional fuels such as diesel or gasoline [11]. In addition, natural gas can be derived
sustainably from biomass and can also be collected from landfills, which is sometimes referred to as
synthetic natural gas or renewable natural gas. Thus, a switch to natural gas technology could also
allow for the use of sustainable fuels.
1.2. Objectives
The objectives of the present work are as follows:
1) Develop a simplified yet robust soot model that can be applied (i.e. is computationally
tractable) in more complex engine simulations, such as those being carried out at Westport
4
Innovations. Specifically, the aim is to apply this model in natural gas fueled compression
ignition engines.
2) Validate the developed model in a variety of operating conditions for key soot characteristics
such as soot volume fraction, soot number density, and soot particle diameters. If necessary,
improve model performance and document this process.
3) Investigate any key issues for soot model behaviour – especially when applied to engine
simulations at Westport. For example, can the model predict the observed experimental soot
emissions? If not, can qualitative trends in soot emissions with changing engine parameters
such as load condition be reproduced? Resolve these issues if possible; otherwise, identify
possible alternatives or methods for soot model application in engine simulations.
5
2. Background and Literature Review
This chapter explores some of the background information necessary to fully understand the
present work and its associated goals. Current understanding of soot and its formation/oxidation is
explored as well as the different approaches that are generally employed to model and predict soot
emissions.
2.1. An Introduction to Soot
Soot is best described as a type of airborne pollutant that is commonly created as a by-product of
hydrocarbon combustion. Unlike other common pollutants such as NOx or CO, which are emitted in the
gaseous phase, soot is unique in that it is emitted as a solid – usually consisting of a complex aggregate
structure composed of many smaller soot particles. These individual soot particles, which are
commonly referred to as “primary particles”, are roughly spherical, mainly composed of carbon, [12]
and are graphitic in nature [13]. Soot emissions from combustion applications such as diesel engines
are often recognized as dark-black exhaust plumes. The presence of soot within a flame is usually
characterized by the yellow-orange glow given off by the flame due to the luminosity of soot particles
within the flame.
2.1.1. Soot Characteristics and overview
Many experimental observations have shown that soot primary particles range in size from 10 – 50
nanometers (nm) [13],[14],[15]. Soot aggregate structure is fractal in nature and can consist of
anything from a few primary particles to hundreds of primary particles. An example of the soot
aggregate structure is shown in Figure 2.1. The macroscopic density of a soot particle is usually
considered to be between 1.8 [16] to 2.0 [12] grams/cm3. Finally, the overall characteristics of soot
particles do not seem to be strongly dependent on the type of flame, fuel, application, or other types of
operating conditions involved [17].
6
Soot in combustion can be characterized in a variety of ways, the most common being soot volume
fraction ( ), soot number density , and diameter of primary particles ( . Soot volume fraction,
which is a unit-less term, is defined as the volume of soot divided by the volume of gas. Soot number
density describes the number of soot particles per unit of volume (ex.
) or sometimes as
the number of soot particles per unit of mass (ex.
). In addition, soot number density can
refer to the total number of aggregates, denoted as or it can refer to the total number of primary
particles, denoted as (per volume/mass of the mixture). As seen in Figure 2.1, the number of
primary particles per aggregate can often vary by a large amount, even within the same type of
flame/application. This is one of the reasons accurately modelling soot can be very difficult, as one
needs to keep track of a population of soot particles that can vary in both diameter and aggregate
structure.
Figure 2.1 – Example of soot aggregate structure in diesel exhaust. Taken from [12].
7
Soot formation and oxidation processes have been reviewed by many authors; some of which
include the reviews by Glassman [17], Richter and Howard [13], Haynes and Wagner [18], Stanmore et
al. [12], and Appel et al. [19]. While the exact details of soot formation and oxidation is still an area of
debate, the overall process of soot formation and oxidation is generally agreed to start with precursor
formation, particle nucleation (or inception), followed by the parallel processes of soot particle surface
growth, particle agglomeration, and soot particle oxidation.
2.1.2. Soot nucleation/inception
Soot nucleation is one of the least understood mechanisms of soot formation, but there is a
general consensus that nucleation occurs due to the combination of polycyclic aromatic hydrocarbons
(PAHs) that transition to the solid phase [18],[19],[20]. This important step is actually preceded by the
pyrolysis of the fuel itself to give rise to various so-called “precursor species” that provide the input for
PAH species. Several precursor species have been identified, with acetylene (C2H2) typically receiving
the most attention [13]. The smallest aromatic species, benzene (C6H6), has also received much
attention as a precursor species due to the fact that PAHs can be considered to be grown from benzene
[13] via the addition of acetylene. It is thought that one of the bottlenecks to soot inception is the
formation of the first aromatic ring (such as benzene or phenyl) and not surprisingly, early work by
Glassman [17] showed that aromatic based fuels had a higher tendency to soot when compared to
alkanes, alkenes, and alkynes. Glassman [17] identified acetylene as an important precursor species to
soot formation and a later review by Richter and Howard [13] also reiterated the same point. Much
attention in recent research has been directed at determining exactly how PAHs form and grow from
their parent fuels [19],[20],[21]. In Figure 2.2, which illustrates a representation of soot formation in
premixed flames, one can observe an example of PAH growth in the “molecular zone” of the diagram.
There is still some debate over which chemical reaction pathways are the most important in forming
the first aromatic ring – some of which is outlined in [4],[21],[22],[23].
8
Regardless of the PAH formation route considered, once they are formed, they will continue to
grow due to subsequent chemical reactions with gaseous species and also collisions with other PAH
molecules to form PAH dimers, trimers, etc. At a certain size, the PAH molecules condense and
transition to a solid state – this results in a nascent soot particle, or in other words, soot particle
nucleation. The details of this process are poorly understood as experimental observations of this
phenomenon are difficult since these large PAH molecules still have relatively small diameters on the
order of 1 nm [13]. Soot inception/nucleation will ultimately add to both the number of soot particles
formed and the total mass of soot formed.
Figure 2.2 – Representation of soot formation in premixed flames. Adapted from [24].
9
2.1.3. Soot particle surface growth
Once the PAH molecules have transitioned into solid soot particles, they can continue to grow in
size due to heterogeneous chemical reactions with gaseous species on the surface of the soot particle.
It is generally agreed upon that acetylene plays a major role in contributing to soot surface growth, as
demonstrated in [19],[21],[25]. The amount of available soot surface area for reactions to occur also
plays a vital role in determining the amount of surface growth that occurs. Surface growth via
acetylene has been described by Frenklach et al. in [19],[21],[26] as the Hydrogen-Abstraction-Carbon-
Addition (HACA) reaction sequence, where the C-H bonds on the surface of soot particles can react
with gaseous species. The reactions contained in HACA are based off of analogous PAH gas phase
reactions and are shown in Table 2.1. represents a reaction site on the surface of the soot
particle where a carbon atom is bonded with a hydrogen atom; i.e. it is hydrogenated and non-reactive.
These hydrogenated sites can later become dehydrogenated through hydrogen abstraction via H atoms
or OH molecules as seen in S1 and S2 (in Table 2.1). represents a de-hydrogenated
(reactive/active) site on the surface of the soot particle that can accept acetylene molecules and
subsequently result in the growth of the soot particle (reaction S4). Alternatively, these de-
hydrogenated sites can be re-hydrogenated (S3) or even oxidized (S5 or S6), in which case the soot
particle becomes smaller. Technically speaking, reactions S5 and S6 are examples of a soot oxidation
mechanism, which will be explored further in Section 2.1.4. A phenomenon known as “ageing” has also
been observed in various experiments – in these experiments, the tendency of a soot particle to
undergo surface growth was observed to decline with increasing particle growth [21]. This can be
explained by the HACA mechanism as a reduction in the availability of active sites on a soot particle, a
decrease in H atom concentration, and/or an arrival at an equilibrium state for H atoms in the mixture.
10
Table 2.1 – List of reactions in the HACA surface growth sequence. Adapted from [19].
PAHs have also been proposed to contribute to surface growth in soot particles, in a mechanism
commonly referred to as PAH condensation [19],[27]. Similar to how PAH molecules can collide with
one another to form nascent soot particles; PAH molecules could also collide with existing soot
particles and condense on the surface of them. Macadam et al. [27] showed that in acetylene-lean
conditions, surface growth via PAH condensation was especially important. However, in acetylene-rich
conditions, surface growth via acetylene was dominant. Regardless of the avenue of soot growth, it is
generally agreed that soot surface growth is the dominant mechanism in forming additional soot mass
in a flame.
2.1.4. Soot particle oxidation
Soot particle oxidation is the mechanism by which soot particles can be oxidized and converted
back into gaseous species. As with soot surface growth, the amount of soot surface area available for
oxidizing agents to attack the soot particles plays a role in determining the rate of oxidation.
Competition between soot inception and surface growth mechanisms against soot oxidation
mechanisms ultimately determines whether or not a flame emits soot particles; i.e. if a flame is
smoking or non-smoking. Of the various species that can contribute to soot particle oxidation, O2 and
OH are generally regarded to be the most important [28],[29],[30],[31],[32]. O2 is generally a major
contributor under fuel-lean conditions, while OH is the dominant specie in fuel-rich conditions [33].
Oxidation by other species, such as the oxygen radical O has been investigated [28] and gasification of
soot via other species such as H2O and NO2 has also been shown to be possible [12]. Some studies have
11
also demonstrated that soot oxidation can lead to changes in the aggregate structure as well, such as
fragmentation of the aggregate into smaller structures [33],[34].
2.1.5. Soot particle agglomeration
Soot particle agglomeration is a unique mechanism of soot growth in that soot nucleation, surface
growth, and oxidation are mainly chemical processes where as agglomeration is mainly a physical
process. An exception to this is that PAH condensation is also a physical process. Agglomeration is best
described as the increase in soot particle size due to the collision of two or more soot particles. For
small (and newly formed particles), collisions may actually lead to a phenomenon referred to as
coalescence, where the two particles collide and merge into a larger spherical particle due to their
liquid-like behaviour [35]. Larger particles that collide can stick to one another and form complex
fractal-like aggregate structures as seen in Figure 2.1. Depending on the circumstances of the collision
and the particles involved, some intermediary result can occur, where the particles partially merge and
form a “bridge” or “neck” [16]. It is also worth noting that not all collisions will necessarily result in
merging or sticking of the particles involved as observed by Kellerer et al. [36]. D’Alessio et al. [37]
noted that under certain flame temperature conditions, particles might not stick due to a thermal
rebound effect – these observations contrast from an earlier belief that all collisions had a 100%
sticking efficiency. Ultimately, soot particle agglomeration will generally only affect the total number of
soot particles formed with negligible effect on the total mass of soot formed.
2.2. Current Approaches to Soot Modelling
Approaches to soot modelling can generally be categorized into three different types of
approaches: (i) empirical approaches, (ii) semi-empirical approaches, and (iii) detailed approaches. Soot
modelling was first extensively reviewed by Kennedy [38] and the continued development of new soot
modelling techniques remains an active area of research today [5],[6],[8],[20],[39]. Figure 2.3 illustrates
five major components of a soot model developed for a combustion system. A flow solver is needed to
calculate the solutions to the basic conservation equations to give the correct fluid flow field, while
12
combustion chemistry is needed to account for ongoing chemical reactions between species present in
the simulation. As soot formation is highly coupled to temperature, radiative effects from the flame
and soot particles often need to be considered. Finally, the components of the soot model itself must
also be included – this can be divided into two major parts: (1) Mechanisms which account for soot
formation/oxidation and (2) Mechanisms that consider the interactions between different soot
particles and also the population distribution of soot particle sizes. While all five components are
required for a full detailed approach, many empirical and semi-empirical approaches neglect one of
more of these components in the interest of reducing complexity and computational costs.
Figure 2.3 – Five major components of soot modelling.
2.2.1. Empirical Models
Empirical models are usually based solely on direct correlations between operating conditions and
the amount of soot that is emitted – i.e. all five components shown in Figure 2.3 could be neglected. In
the case where a flow solver is neglected, the soot model could be solely a function of combustion
input parameters, such as engine load or fuel input in the case of an empirical model for an engine.
They usually have very little computational cost and are quick to implement and run. As a result, they
are typically used in applications where it is not computationally feasible to include a more detailed
13
model, such as in a diesel engine or gas turbine. These types of combustion applications already use
significant computational resources to compute solutions for chemically reacting turbulent flows, and
the addition of a detailed soot model would make the calculations intractable. Due to the nature of
correlation, empirical models cannot be applied to applications or operating conditions that are
significantly different from the baseline from which the model was developed. They also fail to give any
insight on the specifics of soot formation; for example, one would not be able to determine precisely
where/when/why soot is formed in a diesel engine. For these reasons, empirical models may not be
practical for predictive purposes where engine geometries and operating parameters may change
radically from case to case. However, that is not to say that empirical models serve no useful purpose.
One application where empirical models are useful are in diagnostics systems where a user can be fed
real time information on how heavily their engine is sooting based on parameters such as combustion
temperatures. Since the engine operating conditions are not expected to change radically compared to
prescribed conditions for which the model is calibrated for, soot emissions can be accurately predicted
without the need for a complex measurement setup.
An example of an empirical model is the one developed by Khan et al. [40] for diesel engines. In
this model, Khan and co-workers assumed that the diameters of soot particles did not vary with respect
to operating speeds or loads. They also assumed that the overall formation rate of soot was only
dependent on inception, neglecting soot growth and oxidation, giving the equation:
(2.1)
where is the soot mass density [kg/m3], and are model parameters, is the activation energy of
soot formation set to 1.7 x 105 [kJ/kmol], is the volume of the soot formation zone [m3], is the
volume of the cylinder contents at normal temperature and pressure [m3], is the partial pressure of
unburned fuel [Pa], is the local unburned equivalence ratio, and is the local temperature. As
model parameters were adjusted until results fit the available experimental data, the model performed
14
reasonably well for the given conditions. However, one would expect that any significant departure
from the base set of calibrated data would result in poor performance, due in part to the neglect of
many fundamental soot formation/oxidation mechanisms.
Another example of an empirical model is the approach developed by Hiroyasu et al. [41]. Hiroyasu
and co-workers assumed that soot mass emissions were solely based on pressure, temperature, fuel
concentration, and O2 concentration – neglecting intermediary soot formation/oxidation mechanisms
and also ignoring the calculation of the number of soot particles. They defined the formation rate of
soot mass as:
(2.2)
where is the formation coefficient, is the oxidation coefficient, and
are the local mass
fractions for fuel and oxygen, and is the local mass fraction of soot. and are subsequently
defined as:
(2.3)
(2.4)
where and are model parameters, and is the pressure and operating pressure, and
are activation energies set to 6313 and 7070 [K-1] respectively. As with any empirical model, the
model performed relatively well as long as the conditions did not stray far from the conditions used to
calibrate the model parameters.
2.2.2. Detailed Models
On the opposite end of the spectrum, detailed soot models attempt to replicate the current
fundamental understanding of soot formation/growth/oxidation as described in Section 2.1. Detailed
soot models are typically based on fundamental combustion chemistry and make use of aerosol
15
dynamics theory. An ideal detailed soot model would work for any fuel, combustion application, and
operating condition; but in practice, all detailed soot models are still limited to a range of possible
scenarios for which the model was developed – albeit a much broader range than that of empirical
models. A well developed soot model can be applied to various fuels and applications and has better
general applicability. An example of this is the detailed model demonstrated in [42] which included
complex combustion chemistry to model the formation of PAHs and has since been applied with
success to both ethylene/air and methane/air flames under different operating conditions. Detailed
models are also capable of giving insight into the soot formation process and are also able to provide
information on the population size distribution of soot particles. The disadvantage of using detailed
soot models is that they tend to be very computationally expensive – most detailed soot models are
limited to simulations with simple geometry (1-D/2-D) and laminar flow conditions.
A commonly cited example of a detailed model is the one developed by Frenklach and Wang
[21],[26],[43] of which the chemical kinetic mechanism that describes everything from the pyrolysis of
fuel to the formation of PAHs is an integral component. Further details such as inception via PAH
molecules, growth by the HACA mechanism, oxidation, agglomeration, and aggregate structures were
also considered in this model. It is important to note that chemical kinetics play a major role in the
formation of soot at nearly every phase of soot production (inception, surface growth, and oxidation)
[38] and as such, detailed models almost ubiquitously employ some form of a PAH chemical kinetic
mechanism. Recent efforts such as those by Dworkin et al. [20] and Chernov et al. [44] have been made
in the application of improved PAH chemical mechanisms in detailed soot models.
It should be noted that the divide between a “detailed model” and a “semi-empirical” model is not
clearly defined and many approaches saddle a grey area between the two. An example is the work by
Lindstedt [45] which employs a detailed chemical mechanism and simplified soot chemistry to model
soot formation in ethylene and propane counterflow diffusion flames. Soot nucleation was based on
16
the precursor species of acetylene and benzene, with some focus in the work spent on developing the
chemical kinetic mechanism to accurately predict benzene. Oxidation was modelled considering only
O2 as an oxidative species, using rates developed by Lee et al. [31]. Surface growth via acetylene was
considered – however, the dependence on the surface area of soot was modelled using four different
assumptions. One assumption was to assume that surface growth was linearly dependent on the
surface area of soot. A second assumption was that the surface growth was dependent on both the
surface area and the number of available reaction sites per unit area on the soot particle. The third
assumption was that surface growth was only dependent on the number of particles and the final
assumption was that surface growth was dependent only on the concentration of acetylene and
temperature. Results from Lindstedt’s work [45] showed that the third assumption actually produced
the best results, although the author conceded that it was in part due to the difficultly in modelling the
HACA sequence such that there was confidence in the number of available reaction sites and may have
also been a result of the other model parameters that were selected. Reasonable predictions for both
the ethylene and propane flame were obtained in terms of soot volume fractions and particle
diameters.
2.2.3. Semi-empirical models
Semi-empirical models represent a middle ground between empirical models and detailed models
and provide a compromise between computational costs and the ability to model fundamental soot
formation/oxidation behaviour. Semi-empirical models tend to incorporate many soot
formation/oxidation mechanisms but reduce computational costs by simplifying the chemistry
involved. Where detailed models typically require large chemical kinetic mechanisms detailing
hundreds of reactions necessary to account for the formation of PAHs, semi-empirical models typically
employ simplified chemistry to minimize computational costs.
Fairweather et al. [46] developed a model where nucleation of soot particles was solely based on
the precursor species acetylene, allowing for a reduced chemical mechanism without the need to
17
model PAH formation. This implementation differs from the Lindstedt [45] approach as a simplified
chemical mechanism was used. This model was applied to a turbulent diffusion natural gas/air flame
where chemistry was solved by using a flamelet library (i.e. species concentrations and temperatures
were linked to mixture fraction instead of being explicitly solved). Surface growth was considered only
to occur via C2H2 surface reactions and oxidation was considered to occur only via O2. Further
simplifications were made by neglecting soot aggregate structure and assuming all soot particles were
solid spheres without the fractal aggregate structure seen in Figure 2.1. Finally, it was assumed that
surface growth and oxidation rates were linearly related to the surface area of soot particles. Despite
these simplifications, the model performed satisfactorily and unlike a fully empirical model, it could still
provide some insight to soot formation/oxidation rates and also provide more detailed soot data such
as soot number density and diameters. The model was later updated by Woolley et al. [47] to include
inception via benzene molecules as well and also included additional oxidation via OH. It was applied to
a turbulent methane/air flame as well as a propane/air flame demonstrating good agreement with
experimental results and thereby showing fuel flexibility.
The model developed by Fairweather and coworkers represents a popular two-equation approach
to soot modelling – where one equation is used to track soot volume fraction and a second equation is
used to track soot number density. These two equations typically resemble the following form:
(2.5)
(2.6)
where represents the mass of soot, , and represent the mass of soot formed/destroyed
due to inception, growth, and oxidation, respectively, represents soot number density, and and
represent soot number density from inception and agglomeration, respectively.
represents the model specific constants that are usually calibrated based on the exact mechanisms
18
used to represent the aforementioned soot mechanisms and the application for which the model is
used.
A similar two-equation approach was used by Moss et al. [48] where the major difference was in
the rate equations used to represent inception, surface growth, oxidation, and agglomeration. Like
Fairweather et al. [46], Moss and co-workers [48] assumed that surface growth and oxidation were
linearly dependent on soot surface area. A flamelet library was again used to solve for combustion
chemistry. Unlike the Fairweather et al. [46] model, only OH oxidation was considered. The model was
able to predict reasonable soot volume fractions and soot number densities along the centre-line of an
ethylene laminar diffusion flame, but only after some model parameters were adjusted to match
experimental data. The model was later extended by Brookes and Moss [49] to a turbulent
methane/air jet diffusion flame and compared favourably to experimental results. However, it is
important to note that the parameters of the model were again adjusted to fit the experimental data. A
two equation approach utilizing a form of the Brookes and Moss [49] model was also recently applied
to predicting soot in an automotive diesel engine simulation in a study by Pang et al. [8]. Pang et al.
found that the values for constants in the Brookes and Moss model typically found in literature could
not reproduce satisfactory soot behaviour in the engine and henceforth needed to carefully calibrate
the constants such that the model reproduced experimental results.
Hong et al. [50] used aspects of the Fairweather et al. [46] model (namely the inception and
oxidation mechanisms) to model soot formation in a diesel engine. A skeletal form of a detailed
mechanism for n-heptane was used to calculate combustion chemistry. However, instead of using a
simplified acetylene-only based approach for surface growth, a series of surface growth reactions
based on available reaction sites was used. In addition, a method of moments was used in order to
allow for the tracking of particle size distributions by assuming a log normal distribution. While
19
quantitative results were under-predicted in the test cases and the diesel engine, the authors were
satisfied with the qualitative trends reproduced.
Besides the above-mentioned applications, a two equation soot model has also been applied in
laminar methane co-flow diffusion flames [51],[52], laminar methane opposed flow diffusion flames
[53], laminar ethylene diffusion co-flow flames [6],[54],[55], laminar ethylene opposed flow diffusion
flames [56], turbulent ethylene diffusion co-flow flames [9], laminar heptane opposed flow diffusion
flames [57], and laminar acetylene co-flow diffusion flames [58]. A smaller number of applications were
also at high pressure [6],[8],[51],[53],[57] as detailed measurements of soot at high pressure are not
readily available in the literature. In terms of handling combustion chemistry, some approaches used a
flamelet library [9],[47],[49],[52] approach while others calculated detailed chemistry
[6],[51],[53],[55],[56]. Other forms of reduced chemical kinetics were also used in [8],[46],[54],[58]. As
evidenced by the wide variety of applications, the two equation model shows promise in terms of
widespread applicability and low computational cost. A direct comparison of computational costs
between an implementation of the two equation model and a detailed soot model is shown in Section
5.9.
2.2.4. Particle size distribution and soot aerosol dynamics
One of the challenges of soot modelling, besides handling the complex soot chemistry, is how to
track the size and aggregate structure of every soot particle that is formed. In processes with
multiphase flow, additional equations are required to describe changes to the population of particles
which can evolve due to interactions with other particles or due to chemical reactions with species in
other phases.
In the case of soot modelling, the population of soot particles can be affected by the parallel
processes of nucleation, surface growth, agglomeration, particle fragmentation, and oxidation. These
processes can lead to a complex population of differing particles that vary in size, shape, and structure,
20
making it computationally expensive to model accurately. The approach to soot modelling can be said
to be split into two parts: the interaction between soot particles and the gas phase species (i.e. soot
kinetics detailed in the above sections) and the interaction between soot particles (i.e. soot aerosol
dynamics) [59].
There are currently a few major methods that are employed to handle soot particle dynamics [39].
One obvious approach would be to directly model each individual particle in the population of
particles, coined as a “continuous” model approach. While the continuous model is accurate, the
computational costs are large [60] and the implementation of such a model is also restricted to simple
zero to one dimensional cases [39]. A second approach is to model the population of soot particles into
discrete sections or “bins”. The discretization of the particle size distribution allows for computational
tractability in more complicated scenarios compared to a continuous approach. Sectional models still
provide good accuracy if an adequate number of sections is used to represent the particle size
distribution, but the drawback is that each additional variable used to describe the population (e.g.
volume, surface area, etc.) increases the number of needed equations exponentially [39]. As such, a
single parameter, such as the particle mass per section bin along with a spherical particle assumption is
used in most applications of the sectional model. A third approach is a stochastic approach, where the
population of particles is determined by using a stochastic algorithm such as the Monte Carlo method.
However, while this approach has shown success in laminar flames, this method is also very
computationally expensive and is not generally considered for turbulent cases [61]. Another approach
of interest is the Method of Moments (MOM) where evolution equations for moments of the
population distribution are solved instead of explicitly solving the population distribution. In the MOM,
a compromise is made between accuracy and computational costs. Instead of calculating the exact
distribution of a particle population, mean quantities (i.e. moments) are computed [39]. A moment can
be thought of as a measure of varying aspects in a distribution depending on the order of the moment.
Thus, knowledge of all moments from 0 to ∞ in essence fully describes the distribution function itself
21
[62]. It has been noted that 3-6 moments are generally sufficient for an accurate soot calculation [59].
A final “approach” is to neglect tracking the particle distribution altogether, as Fairweather et al. [46]
did in his implementation of a two equation soot model. In this model, it was assumed that all soot
particles were spheres of identical diameter within a control volume (i.e. a monodisperse spherical
particle assumption). While this approach can save computational cost, it introduces errors into the
model as the above-mentioned assumption is questionable. This can lead to some inaccuracy in
predicting the available soot surface area for soot kinetics such as surface growth and oxidation. To
handle this error, simplified models that neglect particle aerosol dynamics typically account for these
errors by adjusting model parameters and constants in order to compensate for the calculated soot
surface areas.
2.3. Laminar Coflow Diffusion Flame
In the development of our soot model, it is important to consider the need for a combustion
configuration that is amenable to iterative numerical experimentation, has extensive validation data,
and is representative of typical combustion conditions within compression ignition engines. While it
would be ideal to develop the model in a compression ignition engine configuration, the complicated
geometry, reciprocating motion, and turbulent flow would make numerical experimentation difficult
due to the associated high computational costs. In addition, due to the nature of combustion in
engines, detailed spatial measurements of soot characteristics are nearly non-existent. For these
reasons, a steady axisymmetric laminar coflow diffusion flame was selected. This combustion
configuration was chosen in part due to its simple laminar flow field which lends itself well to
conducting numerous detailed numerical experiments. In addition, the mixing of fuel and air in a
diffusion flame can be comparable to processes that occur in compression ignition engines; although
admittedly, the effects of turbulent mixing would not be present in a laminar flame. A schematic of a
laminar coflow diffusion flame can be seen in Figure 2.4.
22
Figure 2.4 – Typical setup of a laminar coflow diffusion flame adapted from [63]
The laminar coflow flame also provides an additional benefit in that soot formation/oxidation
mechanisms in the flame span a relatively wide region, which allow for multi-dimensional detailed
measurements of soot characteristics. Figure 2.5 illustrates the typical evolution of a soot particle as it
undergoes soot formation/oxidation mechanisms in a diffusion flame. In can be seen that inception
occurs in the early part of the flame, near the edge of the luminous envelope (the edge of the flame is
usually referred to as the “wing” of the flame). This is not surprising as in diffusion flames, soot is
generally formed in high temperature, fuel rich conditions between 1300 – 1600 K [64]. The soot
particles that are formed then continue to grow as they progress upwards in the flame, before being
oxidized as the temperature in the flame increases. In cases where there is insufficient oxidation in the
flame, some soot particles can escape the wings of the flame, causing a “smoking” flame. In cases
where there is enough oxidation to fully oxidize the soot particles formed, the flame is referred to as a
“non-smoking” flame.
23
Figure 2.5 – Soot formation zones in a coflow diffusion flame along with soot aggregate structure evolution. Adapted from [16].
2.4. Turbulent Combustion Modelling
While the work presented in this thesis focuses on laminar flame simulations, it is important to
consider the complexities and complications turbulent combustion modelling brings, as the eventual
goal of the work is to apply the soot model in turbulent engine simulations. Turbulent combustion
theory and modelling is a large field of research in its own right and as such, the details of turbulent
combustion modelling will only be briefly discussed. Extensive reviews of turbulent combustion theory
and modelling have been conducted by several researchers, such as the efforts by Bilger et al. [65],
Pitsch [66], and Buckmaster et al. [67].
Turbulent combustion modelling can be very difficult and computationally expensive as one needs
to simultaneously calculate the complex, evolving fluid fields as well as the ongoing chemical reactions.
24
While solving for the fluid field and chemistry simultaneously is computationally tractable for laminar,
steady flames, this is not the case for turbulent flames. Instead, chemistry is often de-coupled and/or
simplified in order to reduce computational costs. Some important methods used in non-premixed
turbulent combustion modelling are discussed below.
The first simplification is the assumption that combustion in the turbulent flame is entirely
dominated by mixing related phenomena. A variable referred to as mixture fraction ( ) is often used to
quantify the local degree of mixing. Mixture fraction can be defined as the local mass of material that
originated in the fuel stream divided by the local total mass of the mixture [68]. Thus in the case of a
coflow diffusion flame, will be equal to 1 in the fuel stream and equal to 0 in the oxidizer stream. The
mixture fraction will vary between 0 and 1 throughout the flame and is said to be a “conserved
scalar”, i.e. the scalar variable is conserved at every point in the flow and there is no
creation/consumption of . The general idea of the mixing dominated assumption is that the
instantaneous temperature and species composition could be related to the mixture fraction. In the
case of Reynolds/Favre-averaged approaches of modelling turbulent flow, the average reaction rate
can be obtained by weighting the instantaneous reaction rates related to the probability density
function (PDF) of mixture fraction. Unfortunately, soot (and many other pollutants) does not correlate
well with mixture fraction [38]. This is due to the fact that soot chemistry is relatively slow and is not in
chemical equilibrium. Furthermore, soot particles do not diffuse in the same manner as gaseous
species [38]. Therefore, while the assumption of assuming mixing-dominated combustion is
computationally feasible and realistic for many applications, it is problematic when attempting to
model soot formation/oxidation accurately. Later studies showed that with the assumption of fast
chemistry (i.e. when chemical reaction rates are fast compared to fluid mixing rates and species quickly
reach their chemical equilibrium levels) one could relate the reaction rates in flames to the rate of
scalar dissipation ( ) of the rate of molecular mixing ( ) [65]. The variable is used to describe the
instantaneous local departure from equilibrium observed in the diffusion flame – a high indicates a
25
high rate of removal of heat and species due to turbulent mixing. At a critical , often denoted the
rate of heat removal becomes too high and the flame is quenched.
A method that was developed and is closely related to the concept of mixing-dominated turbulent
combustion is the laminar flamelet model. The key assumption of this approach is that the flame
reaction zones are thinner than the smallest length scales of turbulence – if this is true, then the
turbulent flame can be said to be made up of a collection of smaller laminar flames (i.e. flamelets) [65].
The implication of this is that the local instantaneous species composition and temperature can be
derived from a laminar flame that has the same mixture fraction and scalar dissipation. In the
Stationary (or Steady) Laminar Flamelet Model (SLFM), the parameters of mixture fraction and scalar
dissipation are used to generate a library of values for temperature, species composition, and even
reaction rates [65]. This library is usually pre-calculated for a range of and in order to save time
during the simulation and is therefore a favourable approach to reducing computational costs. During
the CFD calculation, the mixture fraction and scalar dissipation (along with flow fields) would be solved
independently of temperatures and species compositions, which could then later be looked up in the
flamelet library. Unfortunately, this indirect nature of solving for species compositions and
temperatures presents a problem for soot modelling as it is not possible to include the effects of
species consumption due to soot formation/oxidation kinetics. A larger problem is that the SLFM has
difficulty accurately predicting slow-forming species like NOx despite a relaxed dependence on the
assumption of equilibrium chemistry. This is due to the fact that the flamelet structures cannot
respond instantaneously to changes in the scalar dissipation ( ). As a result, the SLFM is not
appropriate for applications where chemical time scales are comparable (i.e. on the same order of
magnitdue) to the flow time scale [69]. In other words, the SLFM is not appropriate for slow chemistry
applications (like soot) due to its fast chemistry assumptions.
26
The inability of the SLFM to handle transient, non-equilibrium effects has led to the development
of an Instationary (or Unsteady) Laminar Flamelet Model (ILFM/ULFM). In this approach, unsteady
flamelets are not pre-processed, but instead solved in conjunction with the CFD simulation. One
example of an unsteady flamelet model is the one developed by Pitsch et al. [70],[71] which was
successfully applied to both a turbulent ethylene jet diffusion flame and turbulent hydrogen jet
diffusion flame. Pitsch and coworkers accounted for the "history" of a flamelet by calculating a
"flamelet time" which was related to the distance to the fuel nozzle and the axial velocity of the flow
[71]. This calculated flamelet time could then be used to resolve the unsteady terms in the transport
equations for species mass fractions and temperature. By using this method, Pitsch and coworkers
were able to use the ULFM to account for transient affects such as radiative heat transfer [71].
The SLFM and ULFM are widely used in turbulent combustion modelling studies due to the
advantage of reduced computational costs and relatively good ability to predict species composition
and temperatures. However, it is important to remember the inherent assumption behind both
flamelet models – that is, that the reaction zone must be thinner than the smallest length scale of
turbulence. If this is not true, an alternative method must be used. One such alternative method is the
Intrinsic Low-Dimensional Manifold (ILDM), which was first developed by Maas and Pope [72]. Maas
and Pope showed that for a given chemical system with a given number of species and reactions, one
could decouple the chemical reactions with the fastest timescales, greatly reducing the number of
variables needed to describe the system [73]. This is a reasonable assumption as the range of chemical
timescales in a simulation can often extend across many orders of magnitude. If enough time passes,
the slower time scales begin to dominate the overall behaviour of the system and the faster reactions
can be assumed to be at equilibrium – the manifold represents a subspace of the entire reaction
system where this behaviour is observed [72]. Maas and Pope demonstrated that all reaction
trajectories tended to move towards this manifold regardless of the initial conditions specified. Thus,
the number of necessary calculations can be reduced by only solving for the slow chemistry that occurs
27
on this manifold and neglecting the relatively short amount of time it takes for the reaction trajectory
to reach the manifold [72]. As a result, one could feasibly create a low dimension tabulation of the
chemical system as a function of a smaller number of so-called “progress variables” (arbitrary
parameters) [69] instead of trying to account for every coupled species and reaction, which can result
in a table that would need over 30 dimensions for just a simple methane-air flame [73]. The dimension
of the ILDM is related to how much time must pass before the reaction trajectories reach the manifold
– the shorter the time, the higher the dimension required [74]. The ILDM approach has been applied
with some success in applications such as a turbulent methane diffusion jet under high pressure [75]
and also in compression ignition engine simulations at Westport Innovations. Compared to the flamelet
approach, the ILDM requires more computational resources to tabulate and to utilise within the
simulation (due to a higher number of variables in the table), but is more widely applicable to a variety
of flame conditions.
28
3. Model Development Methodology
This chapter outlines the overall methodology employed in developing a simplified soot model for
application in natural gas fuelled compression ignition engines. The goal is to begin model development
and validation with simple laboratory flame conditions and move to progressively more engine-like
conditions. This idealized workflow is illustrated in Figure 3.1.
Figure 3.1 – Workflow diagram of project
3.1. Experimental Datasets for Model Validation
A review of detailed soot measurements for methane and natural gas flames in the literature was
conducted in order to determine the availability of validation data. The full review can be found in
Appendix A of this thesis. While validation data is widely available for more heavily sooting simple fuels
such as ethylene, such as the study by Santoro et al. [76] and other studies detailed in [77], detailed
soot data for methane flames is relatively limited. This is due in large part to the lower tendency of
soot formation in methane combustion in atmospheric laboratory flame conditions. Nonetheless,
datasets with soot measurements were found for coflow laminar diffusion flames, coflow turbulent
diffusion flames, counterflow laminar diffusion flames, shocktubes, and laminar premixed flames. In
addition, soot measurement data was also available for natural gas fuelled engines, but limited to
exhaust measurements. Unfortunately, exhaust-only measurements are not particularly useful for
developing a robust soot model as the solution for a "correct" exhaust soot prediction is not a unique
one. For example, a model could have an inception rate that is too high, but is absolved by having an
29
oxidation rate that is also too high. For this reason, spatially resolved measurements of soot are more
useful – particularly in applications where there are clearly defined soot formation/oxidation regions.
As discussed in Section 2.4 earlier, turbulent combustion modelling presents many of its own
challenges that would exceed the scope of this work and as such, development and validation for the
data found was not considered. Furthermore, the combustion behaviour of premixed flames and
shocktubes is fundamentally different from the combustion behaviour of the non-premixed
combustion found in direct injection compression ignition engines. Hong et al. [50] used a similar
approach in developing their soot model by using shock tube data from various experiments to
calibrate their model and then later applying it to a diesel engine simulation. However, while the model
gave reasonable qualitative results, the soot predictions quantitatively were under-predicted. Thus, it
was also determined that model development and validation for these cases would not be particularly
useful in serving our target application and were also neglected. Consequently, the focus for model
development and validation in the work presented focuses on the coflow laminar diffusion flame setup
(discussed in further detail in Section 2.3). Fortunately, the body of literature investigated for coflow
laminar diffusion flames provided the most volume of soot measurement data of all the types of
experimental investigations on soot in methane and natural gas combustion.
3.2. Comparison to Sectional Detailed Soot Chemistry Model
Despite the fact that there are several studies on methane/air coflow flames, the majority of the
measurements made in these studies are limited to spatially resolved measurements of soot volume
fraction only. Spatially resolved measurements of other important soot characteristics, such as number
density (primary/aggregate) or soot particle diameter (primary/aggregate) is limited to a study by
Schittkowski et al. [78] and is subject to a high degree of uncertainty due to the measurement
techniques used. In addition, there exists no data set where measurements were made on the rate of
soot formation/oxidation. In order to avoid creating a "curve-fitted" model that is only applicable in
30
one or two applications, it is desired to have some avenue of validation for the rates of soot
formation/oxidation as well.
One strategy to carry out this validation is to use a detailed sectional soot chemistry model that has
been developed in parallel in another study [20]. Since this model has been well-validated in a variety
of applications [5],[20],[42],[44] we can use this existing model to generate a rich dataset of numerical
"measurements" that can fill in the gaps found in literature. Thus, numerical experiments using the
sectional detailed soot chemistry model will be run and data from these numerical experiments can be
used to supplement existing measurement data in the development and validation of our simplified
soot model. This adapted workflow is outlined in Figure 3.2. However, it is important to note that the
employed sectional detailed soot model was developed and validated for ethylene/air flames, not
methane/air flames. Nonetheless, the similarity between the two fuels and their combustion behaviour
(chemical kinetic mechanisms for the two fuels are often interchangeable) should allow for reasonable
results from the model to be obtained.
Figure 3.2 – Adapted workflow diagram of project
31
3.3. Experimental Cases Considered
Data from studies by Smooke et al. [63], Thomson et al. [79] and Joo and Gülder [80], and
Schittkowski et al. [78] was used to validate the developed model over a variety of operating
conditions. In the study by Smooke et al. [63], spatially resolved measurements of soot volume fraction
were made in the laminar flame using a technique called thermocouple particle densitometry (TPD).
With TPD, soot volume fractions are determined by introducing a clean thermocouple into a sooting
area of the flame and subsequently relating the measured temperature history to an expected
temperature history using thermophoretic mass transfer formulations. The flame in the Smooke et al.
[63] case was placed inside a cylindrical chimney enclosure located at the edge of the coflow air inlet.
Thomson et al. [79] and Joo and Gülder [80] collected spatially resolved measurements of soot volume
fraction but at higher pressures using two non-intrusive techniques - Spectral Soot Emission diagnostics
(SSE) and Line-of-Sight-Attenuation (LOSA) - to make their measurements. The details and theory of SSE
and LOSA are discussed elsewhere in [81] and [82] respectively. The flame was surrounded with a
chimney composed of three flat windows to facilitate optical access and the burner assembly was
placed inside a high-pressure assembly which is illustrated in Figure 3.3. Thomson et al. [79] and Joo
and Gülder [80] used the same experimental setup but conducted their measurements in separate
facilities; however, Joo and Gülder [80] conducted some tests at higher pressures but also only
reported measurements using the SSE diagnostics technique.
32
Figure 3.3 – Schematic of high pressure combustion rig taken from [79]. Finally, in the study by Schittkowski et al. [78], spatially resolved measurements of soot volume
fraction, primary particle radius, and primary particle number density were made using a laser-induced
incandescence (LII) system, the details of which can be found within the aforementioned study. Figure
3.4 summarizes the burner geometry and operating conditions of the studies investigated in this work.
Data Set Geometry (cm)
Flow Rates (g/s) Operating Pressure (atm)
Smooke et al. [63]
Methane = 0.16 Air = 24.64
Argon = 0.004
1
Thomson et al. [79]
Methane = 0.00055 Air = 0.4
10 – 40
Joo and Gülder [80]
Methane = 0.00055 Air = 0.4
10 – 60
Schittkowski et al. [78]
Methane = 0.00326 Air = 1.172
1
(a) (b)
Figure 3.4 – (a) Table summarizing burner geometry and operating conditions of experiments (b) Diagram of coflow burner defining and
3.4. Considerations for use in Westport Simulations
In addition to ensuring the developed simplified model could well reproduce the sooting behaviour
observed in the aforementioned experiments, it was important to consider some implications for
future application in turbulent engine simulations such as those at Westport Innovations.
33
3.4.1. Coupling of Soot Model to Gas Phase Species Consumption
As discussed in Section 2.4, many of the current methods employed to calculate turbulent
combustion chemistry reduce computational costs by pre-calculating reaction rates, species
compositions, and temperatures in a table as a function of various parameters like mixture fraction ( )
or scalar dissipation ( . This approach is evident in the flamelet library approach (steady and
unsteady) as well in some aspects of the ILDM approach. While this method is an effective way to
include combustion chemistry without drastically reducing the chemical mechanism, it presents a
major problem for soot modelling. As the species conservation equations are not explicitly solved in
these approaches, it is not possible to couple the consumption of gas phases species to the
formation/oxidation of soot in these types of simulations. This can potentially cause some
overprediction in soot concentration levels as soot formation/growth can continue indefinitely. To
investigate the effect that this can have on the performance of the proposed soot model, simulations
with and without gas phase species consumption were run and their results were compared.
3.4.2. Coupling to Radiation Heat Transfer
Another important feedback loop that is difficult to implement with the flamelet library and ILDM
approach is the radiation heat loss due to soot particles in the flame. In these tabulated approaches to
solving for flame temperatures indirectly, it is difficult to incorporate a feedback mechanism by which
the presence of soot particles decreases the local temperature. Since a lowered temperature will
usually result in reduced reaction rates, neglecting the coupling of radiation heat transfer can result in
overpredicted flame temperatures and hence, overpredicted soot levels. Similar to the approach
described in Section 3.4.1, simulations with and without coupled radiation heat transfer were run and
their results were compared.
34
4. Mathematical Formulation
This chapter describes the theory and mathematical models employed within the presented work.
In particular, the conservation equations solved for mass, momentum, species, and energy for the
laminar coflow flame are described as well as the formulations of the simplified soot model.
4.1. Computational Domain
The flames considered in this study were all axi-symmetric co-flow laminar diffusion flames and as
such, the 3-dimensional flame is reduced to a 2-dimensional computational domain, which reduces
computational costs. A cylindrical coordinate system is constructed in the radial ( ) and axial ( )
directions and a representative diagram of the computational domain is shown in Figure 4.1.
Figure 4.1 – Schematic of the computational domain (greyed out area) super-imposed on a diagram of a typical laminar coflow diffusion flame. Also illustrated is the orientation of the coordinate system used. Note that the illustration is not to scale. Taken from [77].
4.2. Governing Equations
The 2-D laminar coflow diffusion flame code used in this thesis solves fully-coupled elliptical
conservation equations for mass, momentum, species, and energy. The laminar flame code used in this
study has been widely applied to other laminar coflow diffusion flame studies such as the recent efforts
by Guo et al. [83], Zhang et al. [42], Dworkin et al. [20], and Eaves et al. [5].
35
4.2.1. Conservation of mass
The equation for conservation of mass is as follows:
(4.1)
Where is the mixture density and and are the axial and radial velocities respectively.
4.2.2. Conservation of momentum
The equation for the conservation of axial momentum is as follows:
(4.2)
Here, represents the local pressure, is the dynamic viscosity of the gaseous component of the
mixture, and is the gravitational acceleration, assumed to be solely in the axial direction. Similarly,
the equation for the conservation of radial momentum is as follows:
(4.3)
4.2.3. Conservation of species
The conservation equations for gaseous species mass fractions are defined as:
(4.4)
is defined as the mass fraction of the species, and are the diffusive velocities in the axial
and radial directions respectively of the species, is the molecular weight of the species, and
is the molar production rate of the species per unit volume. is denoted as the total number
of gaseous species present in the chemical mechanism. An important point to note is that also
includes contributions/consumption due to soot formation/oxidation chemical processes. When
36
gaseous species consumption due to soot formation/oxidation were decoupled, only included
contributions from chemical reactions between gaseous species only.
4.2.4. Conservation of energy
The equation for the conservation of energy is formulated as:
(4.5)
Where is the constant pressure specific heat of the mixture, is the local temperature, is the
thermal conductivity of the mixture, is the constant pressure specific heat of the species, is
the specific enthalpy of the species, is the constant pressure specific heat of soot, is the
mass fraction of soot, and are the radial and axial thermophoretic velocities of soot particles,
is the specific enthalpy of soot, is the molecular weight of soot, which is considered to be the
same as carbon in this study, is the molar production rate of soot per unit volume, and is the
total radiation heat transfer from both gaseous species and soot. In specific simulations where
radiation heat transfer was decoupled, the term was neglected. Zhang [42] found that only H2O,
CO2, and CO contributed significantly to radiation heat transfer compared to the rest of the other
gaseous species. Thus, only these species were accounted for (along with radiation from soot particles)
in the term. Thermal properties of soot were assumed to be identical to that of graphite – the
thermal properties of which can be obtained from JANAF thermochemical tables [84].
4.3. Diffusivity of Gaseous Species
The diffusion velocity of gaseous species as noted in Equation (4.4) is calculated in the present
work using the following relation:
37
(4.6)
In the above equation, is the ordinary diffusion velocity of the species, and similarly, is
the thermal diffusion (i.e. thermophoretic) velocity of the species. is a correction diffusion
velocity term which becomes necessary due to the use of mixture-averaged diffusion coefficients. With
the use of a mixture-averaged diffusion coefficient, the net diffusion flux may not sum to zero, so the
correction term is applied [85].
and are in turn calculated by using an approximated mixture-averaged formulation
[77]:
(4.7)
(4.8)
Here is the mole fraction of the species, is the thermal diffusion ratio of the species, and
is the mixture diffusion coefficient of the species. is defined as:
(4.9)
Here is defined as the binary diffusion coefficient. In this study, thermal diffusion was neglected for
all species except for H2 and H.
4.4. Radiation Heat Transfer
Radiation heat transfer in the laminar coflow flame code is calculated using the Discrete Ordinate
Method (DOM) and is only briefly described here. The DOM radiation heat transfer model used in this
work was first developed by Liu et al. [86] and further details on the DOM model can be found in the
respective study. The DOM method is advantageous in that the accuracy of the model is comparable to
more computationally intensive approaches such as a Monte-Carlo approach and that it does not
require any a priori assumptions about the optical thickness of the flame being modelled [77].
38
Radiation Transfer Equations (RTEs) are numerically solved in an axisymmetric cylindrical coordinate
system and the angular and spatial coordinates are discretized using a T3 quadrature system [86]. This
was coupled with a statistical narrow-band-based correlated-absorptivity ( ) model in order to
determine the absorption coefficients of the gas phases species H2O, CO2, and CO. The spectral
absorption of soot was determined to be:
(4.10) Here, is the spectral absorption of soot and is the wavenumber of the spectral band. This
formulation was based on experimental measurements by Buckius and Tien [77],[87].
4.5. Simplified Soot Model
The semi-empirical soot model employed in this study is based on previous work done by
Fairweather et al. [46] and later updated by Woolley et al. [47]. It adds two equations that track soot
mass and soot number density and reduces computational costs by calculating averaged soot particle
diameters per control volume instead of tracking the soot particle size distributions and aggregate
structures. As a semi-empirical model, some simplifications have been made based on the current
understanding of the fundamentals of soot formation and oxidation (reviewed by Frenklach in [21]).
This two-equation approach has been shown to have some success in predicting soot in both laminar
and turbulent diffusion flames with the majority of work being done at lower pressures (See Section
2.2.3).
Soot inception (or nucleation), is typically understood in the literature to be the combination of
molecules known as polycyclic aromatic hydrocarbons (PAH) that condense to the solid phase [20].
However, the incipient species in the simplified model is instead chosen to be acetylene, as it is itself a
precursor to the formation of PAHs and removes the necessity to include large chemical kinetic
mechanisms, which can add to the computational costs of turbulent engine simulations. The chemical
formula for soot inception used in the present model is given as:
39
(4.11) Originally, Fairweather et al. [46] based inception solely on acetylene, while an update by Woolley et al.
[47] also included inception via benzene. Early efforts were made to use benzene in the present work
as well; however, as a relatively slow forming species, it was found to be problematic with the ILDM
approach of handling turbulent combustion. In addition, well validated chemical kinetic mechanisms
for methane combustion that included the formation of benzene were not available in the literature.
As Westport already employs a modified GRI-mechanism by Huang et al. [88] that works well at
predicting other pollutants (like CO), it was also not desired to change the chemical kinetic mechanism
to accommodate benzene formation. Thus, a soot inception mechanism solely through acetylene was
employed in this study.
As with inception, acetylene is assumed to be the only species that contributes to the surface
growth of existing soot particles and is represented by the following chemical reaction:
(4.12) As discussed earlier in Section 2.1.3, this is a simplification of the currently understood and accepted
mechanism of soot particle surface growth known as HACA. Oxidation in the simplified soot model is
considered via surface reactions with O2 and OH only and is represented as:
(4.13)
(4.14) While Fairweather et al. [46] originally neglected soot oxidation via OH, this was later included in the
two equation model in an update by Woolley et al. [47]. This better corresponds to the currently
accepted understanding that O2 and OH are dominant contributors to soot oxidation. The rate
constants for OH oxidation are taken from the study by Fenimore and Jones [28]. The soot chemical
reactions (4.11) through to (4.14) are governed by the following equations:
(4.15)
40
(4.16) (4.17)
(4.18) Where denotes a reaction rate determined by a typical Arrhenius rate expression, is the soot
particle surface area per unit volume of the mixture, and , , and represent the
concentrations of C2H2, O2, and OH respectively in units of
. represents the collision efficiency of
the OH molecules, which is set to 20% in this study. The initial values used for the Arrhenius reaction
rates in equations (4.15) to (4.18) are listed in Table 4.1. Rates for to are taken from Fairweather
et al. [46] while is derived from the rate determined by Fenimore and Jones [28].
Table 4.1 – Summary of reaction rate constants in the Arrhenius form
, where units are in g,
cm, mol, s, K.
1.35E6 0 20 365
5.0E4 0 12 080
1.78E6 0 19 630
864 0.5 0
As noted in equations (4.16), (4.17), and (4.18), there is a functional dependence on the soot particle
surface area for surface growth and oxidation. In addition, Fairweather et al. [46] suggested that this
functional dependence can simply be represented as a linear function such that:
(4.19) The linear dependence assumption is reasonable as it follows that as the soot particle surface area
increases, the tendency of surface reactions occurring would also increase. On the other hand, Liu et al.
[89] suggested that the functional dependence would be better represented as a proportional relation
to the square root of the soot particle surface area such that:
(4.20)
The rationale behind this change is related to the phenomenon of “soot surface ageing” (see Section
2.1.3), where the rate of surface growth for a soot particle tends to decline as it undergoes further
41
surface growth. With Liu and coworkers' implementation, the dependence of surface growth on
surface area is reduced as the soot particle grows larger. This in effect, according to Liu et al. [89],
reproduces an ageing effect of the soot particle. While both functional dependencies hold merit, Liu et
al. [89] noted that similar results for both assumptions could be attained if model constants for surface
growth were adjusted. Thus, the model in this work uses the linear dependence defined in equation
(4.19).
The soot aggregate structure in this simplified approach is neglected, and it is assumed that all soot
particles are spherical in shape. Thus, the available surface area of soot per unit volume of the mixture
is given by:
(4.21)
Here, is the soot aggregate number density defined as the number of soot particles per unit mass of
mixture. , which is the diameter of the representative sphere of soot calculated by using the relation:
(4.22)
where is the mass fraction of soot, and is the density of soot, taken to be 1.9 [g/cc].
In order to solve for the soot mass fraction and soot number density , two additional source
term equations are solved:
(4.23)
(4.24)
Where is the mass of soot, is the molar mass of carbon taken to be 12.011 [g/mol] in this study,
is Avogadro’s number and is Boltzmann’s constant. represents the minimum number of
carbon atoms found in a soot particle, which in a sense, determines the minimum diameter of a soot
particle. In this study, was set to 90,000 which translates into an incipient soot particle size of
42
approximately 12 nm. The 12 nm size was kept in part due to the minimum detection size of soot
particles in common experimental apparatuses. The rationale is that any soot particles smaller than this
would not be detected in experiments and therefore not play a role in model validation. Furthermore,
previous studies found that the predictions generated by the two equation model were relatively
insensitive to the value of selected [45]. is an agglomeration constant that determines the rate
at which smaller soot particles combine. For this study, was initially set to 3, but is known to vary
between 3 to 9 [45],[46].
4.5.1. Soot transport equations
The transport of soot in the work presented is calculated in a similar manner to the methods
presented by Zhang et al. [42], Eaves et al. [5], and Chernov et al. [44] in similar sooting laminar coflow
flame studies. If one accounts for normal diffusion, thermopohresis, and soot formation and oxidation,
the soot transport equations for soot mass and number density is as follows:
(4.25)
(4.26)
Thermophoretic velocity in this work is calculated according to the definition provide by Talbot et al.
[90] and is given by the equation:
(4.27)
4.6. Numerical Method
The numerical method used in this work is similar to the methods used in numerical studies of
other laminar flames [5],[20],[42],[44],[77]. As such, the reasoning and subsequent development of
these approaches is not part of the scope of this work. Further details can be found in [77].
Finite-volume method discretizations were used in order to solve the above-mentioned
conservation equations and soot equations. A staggered mesh was employed in order to avoid
43
calculated pressure gradients that are independent of the local control volume’s pressure. In order to
solve the discretized equations, a semi-implicit scheme was used to solve the coupled pressure and
velocity as well as the discretized governing equations [91]. A second order central difference scheme
is used to discretize the diffusive terms while a power law scheme was used to discretize the
convective terms [91]. Momentum and continuity (which is converted to a pressure correction
equation) equations are solved independently using a Tri-Diagonal Matrix Algorithm (TDMA). Next, the
gaseous species equations are solved simultaneously in each control volume in order to deal with the
overall stiffness of the system. Finally, the soot transport equations and energy equation are solved in a
segregated manner with the TDMA approach.
An arbitrary initial guess for the system is used (typically a temperature of 1900 K and ambient air)
and pseudo-time stepping is used to arrive at a converged steady state solution. Chemical reaction
rates for gaseous species as well as thermal properties are calculated using subroutines from the open-
source CHEMKIN-II [92],[93] libraries. Transport properties of gaseous species including mixture
averaged values for fluid viscosities, thermal conductivities, and diffusion coefficients were calculated
using TPLIB [85],[94]. Two different chemical mechanisms were used: a C1/C2 mechanism originally
developed by Slavinskaya and Frank [7] and a modified GRI-mechanism by Huang et al. [88] developed
for low temperature high pressure methane/air combustion. A modified version of the Slavinskaya and
Frank [7] mechanism with enhanced PAH growth was also employed in certain simulations. The
modifications to this mechanism are described in the work by Dworkin et al. [20] and Slavinskaya et al.
[95].
4.6.1. Mesh and boundary conditions
The properties of the mesh used for finite-volume discretizations varied depending on the
experiment and the geometric properties of each mesh can be found in Table 4.2. A non-uniform mesh
in all cases is employed in order to resolve the large gradients of temperature, species, velocity, etc.
near the flame while reducing computational cost in areas farther away from the flame. Generally
44
speaking, the mesh size was kept at a constant small size until a distance of approximately three times
the flame height and flame radius upon which a stretching factor was employed to allow the mesh size
to grow. This effect can be observed in Figure 4.2.
Data Set Number of Control Volumes
Size of domain [cm x cm]
Initial (stretch start)
{stretch factor} [cm, cm, unitless]
Initial (stretch start)
{stretch factor} [cm, cm, unitless]
Smooke et al. [63] 192 x 92 12.29 x 5.41 0.02 (1.00)
{1.071}
0.05 (6.70)
{1.0205}
Schittkowski et al. [78]
224 x 130 19.23 x 4.37 0.02 (1.99)
{1.075}
0.05 (7.80) {1.03}
Thomson et al. [79] and Joo and
Gülder [80]
240 x 130 3.25 x 1.304 0.0035 (0.31)
{1.075}
0.0077 (1.30) {1.03}
Table 4.2 – Summary of geometric properties of meshes used in laminar coflow flame simulations.
A diagram of the non-uniform mesh is presented in Figure 4.2. As previously mentioned in Section
4.1, the computational domain considered is a 2-dimensional slice of the axisymmetric coflow flame. As
a result, in addition to the normal inlet and outflow boundary conditions employed, there are
additional constraints for the boundary conditions at the axis of symmetry and outer radial boundary of
the computational domain. The inlet condition in all cases was considered to be uniform in both
temperature and velocity. The outflow boundary was defined as a zero gradient condition given as:
(4.28)
Similarly, zero gradient conditions are also imposed on the axis of symmetry:
(4.29)
The outer radial boundary condition varied depending on whether or not it was an open air flame (i.e.
free-slip) or a flame within a chimney (i.e. no-slip). For a free-slip outer radial boundary the following
conditions were used:
45
(4.30)
On the other hand, for a no-slip outer radial boundary condition, the following conditions were used:
(4.31)
Figure 4.2 – Diagram of typical non-uniform mesh employed in the simulations presented. Adapted from [77].
4.6.2. Parallel Computation
The laminar coflow flame code employed in this work takes advantage of a distributed-memory
parallelization with strip-domain decomposition method in order to make calculations tractable and
complete within reasonable time limits. Further details on the development of the parallel flame code
can be found in [77]. The computational domain is divided uniformly by assigning each row of control
volumes perpendicular to the z-axis to an individual CPU. The Message Passing Interface (MPI) library
[96] is used to facilitate the communication and distribution of workload between CPUs. Calculations
were performed on the General Purpose Cluster (GPC) at the SciNet supercomputer centre using a 8-
core Intel Xeon E5540s with 2.53 GHz chip speeds and InfiniBand network connections.
46
4.7. Detailed Sectional Soot Model
As outlined in Section 3.2, a detailed sectional soot model will be employed to run numerical
experiments in order to complement the existing experimental data found in literature and facilitate
the validation of the simplified soot model presented in this study. The detailed sectional soot model is
the result of an ongoing parallel research program being conducted at the CRL at the University of
Toronto and more details on the model can be found in the work presented by Zhang et al. [42],[77].
The major differences between the detailed sectional model and the simplified model are illustrated in
Table 4.3. The biggest simplification in the simplified model is the absence of a soot particle size
distribution, followed by the simplified inception pathway through acetylene, and the surface growth
rate with a linear dependence on soot surface area.
Mechanism Simplified Model Detailed Sectional Model
Soot Inception Nucleation of soot particles based on acetylene
Nucleation of soot particles based on the PAH molecule pyrene (A4)
Surface Growth Reaction rate based on acetylene and a linear dependence on soot surface area
HACA surface reaction scheme with an empirical parameter to correct for deficiencies in model
Oxidation Reaction rates based on oxidation models from Lee et al. (O2) [31] and Fenimore and Jones (OH) [28] and a linear dependence on soot surface area
Oxidation via O2 and OH accounted for as part of the HACA surface reaction scheme. OH oxidation based on collisional frequency and O2 oxidation based on Frenklach and Wang [26] model based on Nagle and Strickland-Constable [30] rate
Soot aerosol dynamics Soot particle size distribution is neglected and all soot particles are assumed to agglomerate into spheres
Soot particle size distribution is modeled using a sectional approach with 35 sections to track primary particles and an additional 35 sections to track aggregate structures. The smallest bin size 0.86 nm and the largest bin size is 13867 nm
Table 4.3 – A summary of the major differences between the employed simplified soot model and a previously developed detailed sectional model.
47
The detailed sectional model was run using an identical flame code with almost the same governing
equations, the exception being the soot transport equations where each section required its own
transport equation. The DOM radiation model was used along with the same mesh, boundary
conditions, and solvers. An empirical parameter, , is used in the detailed model as a correction factor
to account for the actual number of reactive surface sites relative to the number predicted by the
current implementation of the HACA model. A value of is typically selected for a simulation such that
reasonable predictions of soot volume fractions are made. More details on can be found in [77] and
in recent investigations by Dworkin et al. [20] and Eaves et al. [5].
48
5. Development and Validation of Soot Model
The simplified soot model developed in this work was based on the two equation approach
demonstrated by Fairweather et al. [46]. The model was initially calibrated for a turbulent natural gas
flame and as such, the initial model constants are used as a basis for initial investigations on the
model’s performance in the previously mentioned experimental studies chosen for model validation (in
Section 3.3). However, the original Fairweather et al. [46] model lacked OH oxidation, so the Fenimore
and Jones [28] OH oxidation model was also included. The initial model constants have been listed
earlier in Section 4.5.
5.1. Chapter Outline
The work described in this chapter will begin with a sensitivity analysis of the model parameters in
the simplified soot model in Section 5.2. Calculations and results from the simplified model and
detailed sectional model are subsequently presented in Sections 5.3 - 5.6. Improvements and
modifications to the simplified model are discussed in Section 5.7 and updated simulation findings are
presented in Section 5.7.2. The effect of uncoupling gas phase species consumption and radiation heat
transfer from the rest of the model is then subsequently investigated in Section 5.8. Finally, the
computational cost between the simplified model and the detailed sectional model is compared in
Section 5.9.
5.2. Sensitivity Analysis of Parameter Terms in Simplified Model
The parameters used in the simplified two equation soot model consist of the pre-exponential term
and activation energy in each Arrhenius rate expression for equations (4.15) to (4.18) in Table 4.1.
Activation energies for each rate equation have been investigated in previous studies [45],[46] and are
considered to be constants in this study and hence kept at their original values listed in Table 4.1. This
appears to be a reasonable approach as Woolley et al. [47] also kept the original activation energies
when adapting and updating the two equation model for a different combustion application. Additional
parameters in the two equation model also consist of the and terms used in the soot number
49
density source term equation (4.24). The major parameters investigated in this study are summarized
in Table 5.1. The other parameters in the simplified model that were not investigated in this study were
the activation energies, , for the inception, growth, and O2 oxidation Arrhenius rate expressions in
equations (4.15) to (4.17).
Parameter Note
Pre-exponential term in Arrhenius rate expression for soot inception.
Pre-exponential term in Arrhenius rate expression for soot surface growth.
Pre-exponential term in Arrhenius rate
expression for soot oxidation via O2.
Pre-exponential term in Arrhenius rate expression for soot oxidation via OH.
Determines the minimum number of carbon atoms in an incipient soot particle and hence determines the minimum diameter of a soot particle.
Determines the rate at which soot particles agglomerate.
Table 5.1 – List of major parameters in simplified two equation soot model.
The Smooke et al. [63] data set was used as a baseline case upon which to investigate the
sensitivity of soot volume fraction, soot aggregate averaged diameters, soot number density, soot
inception rate, soot surface growth rate, and soot oxidation rate (henceforth collectively referred to as
“sooting behaviour”) to the parameters listed in Table 5.1. Simulations were run where a single
parameter was modified while all other parameters were held constant. Then, the effect on the
aforementioned soot details relative to the baseline case with initial parameters listed in Table 4.1 and
Section 4.5 was recorded. This process was repeated for each parameter listed in Table 5.2. Peak values
of sooting behaviour were recorded in both the wing and the centreline of the flame (illustrated in
Figure 5.1).
50
Figure 5.1 – Diagram of the “wing” and “centreline” regions of a typical flame.
Figure 5.2 – Sensitivity of sooting behaviour to the pre-exponential value of soot inception, . Solid
lines show peak values at the wing of the flame and dashed lines show values in the centreline of the flame. A logarithmic scale is applied to the Y-axis. Figure 5.2 shows the sensitivity of sooting behaviour on the pre-exponential value of . The
legend shown in Figure 5.2 is used for all subsequent graphs in 5.2. Here, “Pre-exponential Factor
Increase” is defined as:
Pre-exponential Factor Increase
(5.1)
51
A similar definition for subsequent “Pre-exponential Factor Increase” is used in Figure 5.3 to Figure 5.7.
Similarly, “Change relative to baseline” is defined as:
(5.2)
Not surprisingly, Figure 5.2 demonstrates that all sooting behaviour terms are nearly linearly
dependent on the inception rate (a log scale is applied since underpredictions share the same weight as
overpredictions, which is not obvious on a linear scale). An increase in inception rate leads to more
soot particles being created and hence a higher soot particle number density and soot volume fraction.
In addition, the increased soot particle surface area means that surface growth and oxidation
mechanisms become stronger as they are linearly related to the soot particle surface area. On the
other hand, the inception rate does not seem to largely affect the soot aggregate averaged diameters
as the additional soot mass gained due to a higher surface growth rate (due to a higher surface area
from a higher number of particles) that is effectively cancelled out by a higher oxidation rate.
52
Figure 5.3 – Sensitivity of sooting behaviour to the pre-exponential value of soot surface growth, . Solid lines show peak values at the wing of the flame and dashed lines show values in the
centreline of the flame. A logarithmic scale is applied to the Y-axis. Legend is in Figure 5.2. The surface growth rate sensitivity analysis, which is illustrated in Figure 5.3, shows that all sooting
behaviour aspects are very sensitive to the pre-exponential . The behaviour is less pronounced in
the centreline of the flame, and in fact, the soot inception rate and soot number density actually drops
in the centreline relative to the baseline case. This can be explained by the fact that inception and
surface growth both compete for acetylene, so the increased surface growth rate will reduce the
inception rate, thereby lowering the soot particle number density.
53
Figure 5.4 – Sensitivity of sooting behaviour to the pre-exponential value of soot oxidation via O2,
.
Solid lines show peak values at the wing of the flame and dashed lines show values in the centreline of the flame. A logarithmic scale is applied to the Y-axis. Legend is in Figure 5.2. An analysis of the sooting behaviour sensitivity on soot oxidation via O2, shown in Figure 5.4, revealed
that the O2 oxidation model by Lee et al. [31] has a limited effect on sooting behaviour in the flame
compared to surface growth and inception. Only in the extreme case of removing O2 oxidation entirely
does one start to observe significant changes to sooting behaviour characteristics in the wing of the
flame. Removing O2 oxidation creates the asymptote observed in Figure 5.4 as the logarithmic value of
zero is undefined. It is worth noting that the sooting behaviour in nearly all cases is unaffected by
changes in the O2 oxidation rate. One exception to this is the O2 oxidation rate itself, which actually
increases faster in the centreline than it does in the wing with increasing O2 oxidation rate.
54
Figure 5.5 – Sensitivity of sooting behaviour to the pre-exponential value of soot oxidation via OH, . Solid lines show peak values at the wing of the flame and dashed lines show values in the centreline of the flame. A logarithmic scale is applied to the Y-axis. Legend is in Figure 5.2. The sooting behaviour as affected by the OH oxidation rate, shown in Figure 5.5, is very similar to
effects observed with changes to the O2 oxidation rate described above. As before, the removal of OH
oxidation by setting the pre-exponential value of AOH to zero creates the asymptote observed. For the
most part, there are minimal changes to the soot volume fraction, soot number density, and soot
diameter predictions in the centreline and wing. This is in contrast to the changes to the O2 oxidation
rate where changes in the sooting behaviour in the wings of the flame were much more noticeable.
Thus, it appears that the O2 oxidation model is the dominant oxidizing mechanism in this particular
simplified model and flame setup. The competing behaviour between O2 oxidation and OH oxidation is
once again observed – as OH oxidation is increased, the O2 oxidation rate decreases and vice versa.
Interestingly, increasing the OH oxidation pre-exponential did not significantly increase the overall peak
OH oxidation rate. As the peak OH concentration in the cases investigated did not vary, this suggests
55
that in the particular flame investigated, OH concentration is the limiting factor, and not the pre-
exponential of the oxidation rate.
Figure 5.6 – Sensitivity of sooting behaviour to the selected incipient particle diameter, with the default value at 12 nm. Solid lines show peak values at the wing of the flame and dashed lines show values in the centreline of the flame. A logarithmic scale is applied to the Y-axis. Legend is in Figure 5.2. The sensitivity of sooting behaviour on the parameter, which affects the incipient particle’s
diameter, is shown in Figure 5.6. It can be observed that the soot particle number density is most
affected by the choice of incipient particle diameter, which is not surprising as the source term for soot
number density [Equation (4.24)] is heavily influenced by the choice of the parameter. The
smaller is, the less inception reactions [Equation (4.11)] are required to occur before a new soot
particle is created (and vice versa). This change in soot number density will in turn affect the available
soot surface area, and as such, affects the soot surface growth and oxidation rates in a similar manner.
However, the overall change in measureable sooting characteristics other than soot number density
(i.e. soot volume fraction and soot aggregate diameters) is minimal.
56
Figure 5.7 – Sensitivity of sooting behaviour to the selected agglomeration rate, . Solid lines show peak values at the wing of the flame and dashed lines show values in the centreline of the flame. A logarithmic scale is applied to the Y-axis. Legend is in Figure 5.2. Sooting behaviour was found to be largely unaffected by the agglomeration parameter, with the
exception of soot aggregate diameters and soot particle number density, as discerned in Figure 5.7.
This is not too surprising as the agglomeration parameter, only plays a role in the soot number
density source term in Equation (4.24). As increases, the tendency of soot particles to combine and
agglomerate increases, effectively decreasing the soot particle number density. As this does not affect
the soot mass fraction, it follows that the averaged soot particle diameter will also increase.
5.3. Model Development at 1 atmosphere using the Smooke et al. [63] data set
Computations were completed based on the experiments performed by Smooke et al. [63] as
described in Section 3.3. A modified GRI mechanism developed by Huang et al. [88] was used for
simulations that employed the simplified two equation soot model while a detailed mechanism with
PAH formation developed by Slavinskaya and coworkers [7],[20],[95] was used for simulations that
utilized the sectional soot model. An value of 1 was selected for the simulations conducted with the
57
detailed model for this flame. In the following discussion, the term “simplified model” refers to the
“base simplified model” with constants as outlined in Table 5.5 in Section 5.7.
The radial profiles of soot volume fraction measurements and calculated soot volume fractions
were compared at four different axial heights in the flame and are shown in Figure 5.8. The detailed
sectional code predicted a peak soot volume fraction of 0.49 ppm in the wing of the flame, which
matches the experimental peak measurement of 0.49 ppm in the wing of flame. However, the detailed
sectional code predicted the peak at a slightly higher axial position in the flame – approximately 0.6 cm
higher (of which the peak flame height was approximately 6 cm). This is postulated to be the result of
delayed PAH formation due to some deficiencies in the chemical mechanism used. Since the detailed
code uses PAHs as a direct precursor to soot inception, this delay in sooting behaviour is likely due to
the aforementioned slow PAH formation. The improvement of PAH formation behaviour in chemical
mechanisms is currently the focus of other ongoing studies and further investigation is outside the
scope of this work and instead, the delay was accounted for in the comparisons by simply making the
comparisons at an axial offset of 0.6 cm. On the other hand, the simplified model predicted a peak soot
volume fraction of 0.24 ppm, which underpredicts by slightly more than a factor of two, just over the
reported measurement uncertainty of . However, unlike the detailed sectional model, the peak
soot volume fraction is predicted at the correct axial height in the flame since the soot inception is
based on acetylene and does not suffer a delay. In both cases, the overall computed radial soot volume
fraction profiles capture the general trends found in the experimental measurements. However, at
higher axial heights (namely Z3 and Z4), the soot volume fraction predictions from both models are
shifted outward in the radial direction, with noticeable underprediction in the centreline. This
behaviour has been observed in other studies [5],[20],[63] and appears to be independent of the soot
model used. It is theorized that the centreline underprediction is related to deficiencies in the chemical
mechanisms' ability to predict the correct amount of soot precursor growth within the centre of the
flame. Nonetheless, the qualitative and quantitative results obtained are promising compared to the
58
available experimental data. Recent studies by Eaves and coworkers at CRL have also shown that this
behaviour might also be due to the fact that the model neglects pre-heating of the fuel tube caused by
the flame sitting near the inlet. At the time of writing, this work has not yet been published.
Figure 5.8 – Soot volume fraction profiles at different axial heights above the burner. Z1, Z2, Z3, and Z4 correspond to heights of 2.0, 2.25, 2.5, 2.75 cm for experimental measurements and computations from the simplified model for the Smooke et al. [63] flame. For computations from the sectional model, 0.6 cm was added to each axial height to account for the delay in PAH formation.
5.3.1. Comparisons to numerical data from detailed sectional soot model
While soot volume predictions from the simplified soot model are promising, the computed values
can be improved by adjusting some of the model parameters. However, as previously stated in Section
3.2, it is not desired to simply adjust the parameters at random since arriving at a reasonable
prediction is not a unique solution. Instead, the detailed sectional soot model has been employed to
produce a numerical data set upon which further details on sooting behaviour can be validated. In the
case of Smooke et al. [63], only a few soot volume fraction measurements were made at four heights
in the flame, which were not adequate to reproduce a reasonable looking contour of soot volume
59
fraction. Thus, contours of soot volume fraction could only be compared amongst the simplified code
and detailed sectional model, illustrated in Figure 5.9.
Figure 5.9 – Contours of soot volume fraction (ppm) generated using the simplified model (left) and detailed sectional model (right) for the Smooke et al. [63] flame. The contours of soot volume fraction compare favourably between the simplified model and the
detailed model, with the major difference being the peak soot volume fractions predicted and the
slight axial shift in soot as previously observed in Figure 5.8.
60
Figure 5.10 – Contours of soot number density of aggregates (#/cc) generated using the simplified model (left) and detailed sectional model (right) for the Smooke et al. [63] flame. Computed soot number density contours are shown in Figure 5.10 – unlike the contour of soot volume
fraction, there are significant differences in the predictions of soot number density between the two
models. Firstly, the peak soot number density is about 40% higher in the simplified model than in the
detailed model, with the peak in the wing instead of in the centreline of the flame. The axial shift in the
location of soot is once again observable in the contours. Based on the current implementation of the
simplified model, the higher soot number density can be attributed to either an inception rate that is
too high, or an agglomeration rate that is too low. It’s worth noting that in order to make the
comparison reasonable, particles smaller than 12 nm (the incipient particle size in the simplified model)
were neglected for the purposes of calculating soot number density in the detailed model. However,
particles smaller than 12 nm were still used in the calculation of mass averaged aggregate diameters in
the following comparison. Overall, the discrepancy between the two models can also likely be
attributed to the lack of soot aerosol dynamics in the simplified model (such as fragmentation of
particles) and is a given limitation of the simplified model as it neglects soot aggregate structure. The
soot number density is also sensitive to the incipient particle diameter; however, in order to reduce
61
soot number density, the incipient particle diameter would have to be increased, making it unrealistic
(as most experimental apparatuses start to detect soot particles over a size of ~10 nm).
Figure 5.11 – Contours of soot aggregate mass averaged diameters (nm) generated using the simplified model (left) and detailed sectional model (right) for the Smooke et al. [63] flame. The contours of mass averaged diameters of soot aggregates, shown in Figure 5.11, were compared to
get a better idea of which soot kinetic mechanisms (i.e. inception, surface growth, oxidation,
agglomeration) needed improvement and also to evaluate the ability of the model to predict soot
diameters. As the minimum size of a soot particle was prescribed to be 12 nm, it is not surprising to see
a relatively flat distribution of particle diameters predicted by the simplified model. This is due to the
fact that even a miniscule amount of soot calculated by the model outside the core of the flame will be
reported to have a diameter of 12 nm. The peak size of diameters between the two models differ by
about a factor of two, which suggests the simplified model needs a higher surface growth rate and/or a
higher agglomeration rate to produce particles of a larger size. It is also possible the oxidation rates
may be too high in the model, which could also result in smaller than expected particles.
62
In order to better understand how the model parameters in the simplified model should be
modified in order to better reflect fundamental sooting behaviour as calculated by our detailed model,
inception, surface growth, and oxidation rates were plotted along the pathline of maximum soot, which
can be seen in Figure 5.14. This tracks the evolution of a soot particle that passes through the point of
peak soot volume fraction as it travels through the flame. An example of a soot pathline, which is
calculated considering both the bulk fluid velocity and the soot thermophoretic velocity, is illustrated
below in Figure 5.12.
Figure 5.12 – Example of a pathline of maximum soot. In order to be comparable to the simplified code, the inception rate for the sectional model was
calculated by summing the cumulative contributions to soot mass until it reached the incipient soot
particle size of the simplified model, which is highlighted in Figure 5.13. Essentially, the inception rate
in the simplified code has all the particle growth and agglomeration “built-in” from the gaseous phase
to its incipient particle size of 12 nm. As such, the inception rate in the sectional code needs to be
combined with the particle growth and agglomeration rates that occur between its incipient size of
about 0.9 nm to a size of approximately 12 nm.
63
Figure 5.13 – Diagram of methodology used to compare inception mechanisms between the simplified code and the detailed code.
Figure 5.14 – Calculated plots of soot inception rate, surface growth rate, and oxidation along the pathline of maximum soot for the Smooke et al. [63] flame. The Y-axis is plotted on a logarithmic scale. The cumulative inception rate in the detailed model is similar in terms of its peak value compared to
the inception rate calculated by the soot model, as seen in Figure 5.14. While the peak inception rate is
higher in the simplified model, the major difference is the fact that significant amounts of soot
inception happen earlier in the flame than in the sectional model. This can likely be attributed to earlier
64
abundance of acetylene in the simplified model and the aforementioned delay in PAH formation in the
detailed model. In fact, the shape of the inception rate curves for the detailed and simplified models
closely mirror plots of pyrene (PAH) and acetylene concentration, respectively (not shown here). The
rates of surface growth and oxidation were observed to be higher in the detailed model compared to
the simplified model. However, the simplified model again showed higher rates of surface growth and
oxidation lower in the flame, which is not surprising since inception is calculated to occur lower in the
flame as well. Similarly, the integrated sum of the rates (i.e. the area under each rate curve) also shared
the same comparisons: the integrated inception rate was higher in the simplified model and the
surface and oxidation integrated rates were higher in the detailed model.
Overall, initial results from the simplified and detailed model of the Smooke et al. [63] flame show
that although the simplified model is nearly within the experimental error of measured peak soot
volume fraction, modifications to the surface growth rate and agglomeration rate of the simplified
model may be warranted. This conclusion arises from the fact that soot number densities appear to be
overpredicted in the simplified model while soot aggregate mass averaged diameters appear to be
underpredicted, relative to the detailed soot model. In addition, an analysis of the inception, surface
growth, and oxidation rates along the pathline of maximum soot reinforce the idea that the current
surface growth rate of the simplified model is insufficient.
5.4. Model Development at 1 atmosphere using the Schittkowski et al. [78] dataset
Nearly identical computations and analyses were performed using the experimental configuration
and data provided by Schittkowski et al. [78] as those described in Section 5.3. Calculations for the
simplified model were completed with a modified version of the GRI mechanism by Huang et al. [88].
The detailed model employed a recently developed improved PAH mechanism that is largely the same
as the mechanism by Slavinskaya and Frank [7] with a few modifications to some chemical reaction
rates. This new PAH mechanism attempts to address some of the deficiencies in PAH formation
observed earlier in Section 5.3 and in other ongoing studies. An value of 0.65 was used in the
65
detailed model in order to match the calculated peak soot volume fraction to the experimental peak
soot volume fraction in the wing of the flame. As before, “simplified model” refers to the “base
simplified model” with constants as outlined in Table 5.5 in Section 5.7.
(a)
(b)
Figure 5.15 – Contours of soot volume fraction side by side with experimental measurements in the Schittwkowski et al. [78] flame. Experimental measurements are on the left side of the flame and computations are on the right side of the flame. Results of the simplified model are shown in (a) and results of the detailed model are shown in (b). Computed soot volume fraction contours from the simplified and detailed models were compared to
the experimental measurements made by Schittwkowski et al. [78]. Unfortunately, the raw data from
the Schittwkowski et al. [78] flame is no longer available and only the black and white images of the
contours from the paper could be used for comparisons. In addition, the method of using an LII
diagnostic technique introduces many uncertainties into the measurements made as the LII
66
measurement technique currently requires several assumptions about flame and soot properties as an
input – further details on this process can be found in a study by Will et al. [97]. Since no measurement
uncertainty was reported by Schittwkowski et al. [78] for their techniques, an uncertainty of
was assumed for the purposes of comparison with calculated results. As the parameter in the
detailed model was adjusted to fit the experimental results, it is not surprising that the calculated peak
of 0.50 ppm closely matches the apparent measured peak of 0.55 ppm, both in the wings of the flame.
However, the location of the calculated peak soot volume fraction is once again slightly shifted upward
in the axial direction. On the other hand, the simplified model underpredicted the peak soot volume
fraction by about a factor of 4, giving a calculated maximum soot volume fraction of 0.14 ppm. The
predicted tip of the flame (indicated by the location of the soot volume fraction contour) also does not
reach the same height in the simplified model as observed in the experimental measurements. This
suggests that the soot oxidation rates in the flame may be too high, or that inception and growth rates
in the model are too low.
67
(a)
(b)
Figure 5.16 – Contours of soot particle diameter in the Schittkowski et al. [78] flame. (a) shows the experimental measurements of primary particle diameter on the left and the calculated contour of primary particle diameter on the right from the detailed model. (b) shows the calculated contour of mass averaged aggregate particle diameter from the detailed model on the left and the simplified model on the right. Due to the limited experimental measurements made, only primary particle size of soot particles were
reported by Schittkowski et al. [78]. Since primary particles in the simplified model are not tracked (due
to the lack of a soot aerosol dynamics model), only a comparison to the detailed sectional model could
be made, which is shown in Figure 5.16. Experimental measurements show a predicted peak primary
particle size of approximately 20 nm in the flame, which compares favourably with the detailed model's
prediction of a peak primary particle size of 13 nm. In addition, the spatial distribution of primary
particle size predicted by the detailed model is a good match to the experimental results. In order to
have some insight on the particle sizes being predicted by the simplified model, the mass averaged
aggregate diameters are compared amongst the two models. A peak of about 42 nm was predicted by
the detailed model and a peak of 20 nm. This reinforces the earlier findings from Section 5.3.1 that
68
suggest the growth rate and/or the agglomeration rate predicted by the simplified model is too low
and/or the oxidation rates are too high.
(a)
(b)
Figure 5.17 – Contours of soot particle number density in the Schittkowski et al. [78] flame. (a) Shows the experimental measurements of primary particle number density on the left and the calculated primary particle number density on the right from the detailed model. Different scales are used for each half of the flame. (b) Shows the calculated contour of aggregate particle number density from the detailed model on the left and the simplified model on the right. Similar comparisons were made with respect to particle number density, as seen in Figure 5.17. For the
same reasons described previously, only the detailed model could be used to compare with the
experimental data. Here, we observe the first real problem that the detailed model exhibits as the peak
value of soot primary particle density is overpredicted by an order of magnitude. This is likely due in
part to the detailed model's handling of coalescence, which describes the process in which small,
liquid-like soot particles combine into a new primary particle sphere instead of forming a typical soot
aggregate structure. Without proper consideration of coalescence, the predicted primary particle
number densities can be expected to be overpredicted, as observed. The improvement of this
69
coalescence model in the detailed code is ongoing in a parallel research program and is not part of this
study. On the other hand, the predicted aggregate particle number densities appear to compare
favourably across the simplified and detailed model. Once again, the simplified model predicts a higher
peak value by about a factor of 1.5, suggesting again that either the inception rate is too high, or the
agglomeration rate is too low.
5.4.1. Comparisons to numerical data from detailed sectional soot model
Using a similar process outlined earlier in Section 5.3.1, soot inception, surface growth, and
oxidation rates were compared along the pathline maximum soot volume fraction between the
simplified model and the detailed model. The resulting plots, shown in Figure 5.18, illustrate once again
that although the peak inception rates predicted by both models are comparable, the peak growth rate
and subsequent oxidation rates are too low in the simplified model. This is not too surprising as the
earlier comparisons to experimental and numerical data from the detailed model showed that the
simplified model underpredicted soot volume fractions and particle diameters. As before, the
integrated sum of the rates (i.e. the area under each rate curve) also shared the same results with the
resulting integrated inception rate calculated higher in the simplified model and the resulting surface
and oxidation integrated rates calculated higher in the detailed model.
70
Figure 5.18 – Calculated plots of soot inception rate, surface growth rate, and oxidation along the pathline of maximum soot for the Schittkowski et al. [78] flame. The Y-axis is plotted on a logarithmic scale.
5.5. Results at elevated pressures
The flames of Thomson et al. [79] and Joo and Gülder [80] were simulated at pressures of 10 atm to
40 atm in order to investigate the performance of the simplified model at higher pressures.
Calculations for the simplified model were completed with a modified version of the GRI mechanism by
Huang et al. [88]. The detailed model employed the improved PAH mechanism that is based on the
mechanism by Slavinskaya and Frank [7] with a few modifications to some chemical reaction rates. An
value of 0.10 was used in the detailed model at all pressures investigated as it gave reasonable
predictions of peak soot volume fraction for each case investigated. Even better predictions could have
been obtained by specifically adjusting the value for each pressure, but it was deemed an
unnecessary use of computational resources due to the limited benefits and scientific merit it would to
the study. Finally, “simplified model” refers to the “base simplified model” with constants as outlined in
Table 5.5 in Section 5.7.
71
(a) (b)
(c)
Figure 5.19 – Contours of soot volume fraction (ppm) with experimental measurements made by Joo and Gülder [80]. Experimental measurements are on the left and calculated contours from the simplified model are on the right. Operating pressures are at 10 atm (a), 20 atm (b), and 40 atm (c). Calculated contours of soot volume fraction from the simplified model are compared with
experimental measurements in Figure 5.19. For the most part, differences in measurements made
between LOSA and SSE in Thomson et al. [79] and by SSE in Joo and Gülder [80] were found to be
negligible, with the largest differences occurring between LOSA and SSE measurements in Thomson et
al. [79] at higher pressures. Further details on the discussions of the cause of these discrepancies can
be found in their respective papers, but in the interest of saving the reader the trouble of looking at
three nearly identical measurement data sets, only the most recent results from Joo and Gülder [80]
are discussed. At all pressures, the spatial distribution of soot volume fraction is well reproduced in the
simplified model; however, there is some overall underprediction in the peak soot volume fraction
values. Notably, there is significant underprediction on the centreline of the flame, but the values in
the wing of the flame are all within experimental uncertainty. This can be attributed in part to the large
degree of uncertainty of SSE measurements in the core of the flame, where accuracy is limited due to
uncertainty of the temperature measurements and optical limitations due to the thin flame [79]. In
72
addition, the general trends observed in the experiment as pressure increases (thinning of flame,
higher soot volume fractions) are also well reproduced by the simplified model.
(a)
(b)
(c)
Figure 5.20 – Contours of soot volume fraction (ppm) with experimental measurements made by Joo and Gülder [80]. Experimental measurements are on the leftt and calculated contours from the detailed model are on the right. Operating pressures are at 10 atm (a), 20 atm (b), and 40 atm (c). The results of the detailed sectional soot model are compared with experimental results in Figure 5.20.
The delayed upward axial shift in soot volume fraction predictions are once again observed, but the
predicted values in the wing are well within experimental uncertainty. The centreline soot volume
fraction is within experimental uncertainty at 10 atm, but slightly over-predicted at higher pressures.
As mentioned earlier, there is some additional uncertainty of the measurements made in the core of
the flame. Another possible source of discrepancy is the performance of the refined PAH mechanism,
to which the sooting behaviour in the centreline is highly sensitive. The continued development of the
PAH mechanism, including its validation at higher pressures, is unfortunately outside the scope of this
work.
73
5.5.1. Comparisons to numerical data from detailed sectional soot model
While the detailed sectional model has been extensively investigated at atmospheric conditions, it
has not been as thoroughly validated at higher pressures – especially with respect to particle number
densities and soot particle diameters since there is no available experimental data for these soot
characteristics. Nonetheless, a brief comparison between the two models can provide some insight to
the computed sooting behaviour.
10 atmospheres 20 atmospheres 40 atmospheres
Detailed Model 131 254 309
Simplified Model 63 106 176
Table 5.3 – Calculated peak mass averaged aggregate particle diameters (nm) in the Joo and Gülder [80] flames.
10 atmospheres 20 atmospheres 40 atmospheres
Detailed Model 4.9E10 1.3E11 2.8E11
Simplified Model 1.3E11 1.6E11 2.0E11
Table 5.4 – Calculated peak aggregate particle number densities (#/cc) in the Joo and Gülder [80] flames. Table 5.3 and Table 5.4 list the calculated peak values of soot particle diameters and number densities
from the investigated high pressure flames. Of note is that the detailed model appears to indicate that
the soot diameters calculated by the simplified model should be larger, which is consistent with the
results in Section 5.4 and 5.5. The calculated peak soot number densities are similar, with the simplified
model having a higher soot number density at 10 atmospheres, and the detailed model having a higher
soot number density at 40 atmospheres. It is unclear why this change in behaviour is observed, but it is
likely due to the change in aggregation and coalescence behaviour of soot as pressure increases. Since
the employed detailed sectional soot model uses aggregation and coalescence models originally
developed for one atmosphere, it is possible that these models need to be adjusted for increased
pressures. As aggregation/coalescence behaviour can greatly affect the calculated number densities
[98], it is not surprising that there are some observed inconsistencies in the particle number densities.
Nonetheless, it is still interesting that the detailed model suggests that predicted particle diameters
74
should be higher in the simplified code, which once again implies that the growth rate and/or
agglomeration rate and/or oxidation rate should be adjusted.
5.6. Preliminary Results in Engine Simulations
The developed simplified model and some of its iterations were applied to an engine code at
Westport. The details of the code and the engine are unfortunately proprietary and cannot be
disclosed, but some general comments about the performance of the developed soot model can be
made. The simplified soot model was used in a turbulent reacting flow simulation with a moving mesh
that attempts to model the combustion behaviour in a reciprocating compression ignition engine. The
results from using the simplified model with the initial parameters showed that soot mass fractions
being predicted were about an order of magnitude too low, as well as soot diameters being an order of
magnitude too low, compared to experimental results. A possible reason for this is an inception rate
that is too low, a growth/agglomeration rate that is too low, and/or an oxidation rate that is too high.
As seen in the sensitivity analysis in Section 5.2, an increase in any of the above mentioned parameters
can remedy the underpredictions in soot mass and diameters. An odd result from initial runs of the two
equation soot model showed that the calculated soot diameters in the exhaust were smaller than the
incipient diameter of 12 nm, peaking at 5 nm. This can happen in a scenario where soot particles
undergo so much soot oxidation that they shrink to sizes below their incipient size. This suggests that
the soot oxidation mechanisms in the simplified model are too strong and that they result in oxidation
of far more soot than is realistic. In addition, the results from Section 5.3 to 5.5 would seem to suggest
that the growth rate and agglomeration rate needs to be improved.
5.7. Model Improvements
The cumulative simulation results from Section 5.3 to 5.6 were used to direct the modification of
parameters in the simplified model, with the goal of having a single set of parameters that performed
reasonably well in all conditions investigated.
75
5.7.1. Oxidation Mechanism of Soot Model
The oxidation mechanisms of soot, in comparison to the mechanisms of inception and surface
growth, have received far less attention in the literature. Most of the existing oxidation models have
been employed in non-smoking flames; however, this is not a thorough test of the accuracy of the
model as correct soot volume fractions can still be obtained as long as all the soot has been oxidized. In
other words, the soot oxidation models can overpredict the level of soot oxidation occurring in the
model and still correctly predict the peak soot volume fraction within the flame. In order to properly
test the performance of a soot oxidation model, it is desired to test its performance on both a smoking
and non-smoking flame. The improvement of oxidation models has been investigated before by Liu et
al. [29] and is also the subject of current investigation for the further development of the detailed
sectional model. As the refinement of oxidation mechanisms in soot modelling is a project of its own
right, only a brief look at utilizing some of the suggestions at improving current oxidation mechanisms
are investigated.
Unfortunately, there exists only one smoking/non-smoking flame dataset in the literature where
detailed measurements of soot were taken. This investigation was done by Santoro et al. [76] for
several ethylene/air flames, the details of which can be found in their respective paper. Of interest is
the flame denoted as “F2”, which is the non-smoking flame, and “F4”, which is the smoking flame. Liu
et al. [29] found that their employed oxidation model (Nagle and Strickland-Constable [30] for O2 and
Fenimore and Jones [28] for OH) combined with a semi-empirical two equation model could not
properly replicate the non-smoking and smoking behaviour of the F2/F4 flames simultaneously. To
rectify this, Liu et al. [29] introduced a correction factor of the form:
(5.3)
where and are parameters of the correction factor formula. The purpose of this correction
factor, which ranges from 0 to 1, is to introduce a sharp cutoff in the oxidation rates predicted by the
76
employed models in regions where the temperature is too low to facilitate oxidation. Liu et al. [29]
found that many experimental measurements showed that the cutoff for soot oxidation behaviour is
around 1300-1400 K and that the models employed overpredicted oxidation behaviour above these
temperatures. By multiplying the correction factor to the oxidation models used in their soot model,
Liu et al. [29] was able to reproduce the observed experimental non-smoking and smoking behaviour of
the flames.
This process was repeated for the simplified soot model in this study. An identical mesh to the one
used for the Smooke et al. [63] simulations was employed and the modified GRI mechanism by Huang
et al.[88] was utilized. , which determines the threshold at which the correction factor drops off
was set to 1525 K and , which controls the steepness of the drop-off was set to 60 K. These differ
slightly from the parameters Liu et al. [29] proposed as his parameters were specifically tuned to their
model. The employed correction factor alongside the original rates is shown in Figure 5.21 along with
an illustration of its effects on the oxidation rates calculated by the simplified model.
(a)
(b)
Figure 5.21 – (a) Graph illustrating the correction factors employed. (b) The effect of the correction factors on the rates predicted by the utilized oxidation models. By using the modified correction factors, the F2 and F4 flame conditions from Santoro et al. [76] were
simulated using the simplified model. As we were only interested in matching the soot volume
fractions predicted by the model and replicating non-smoking and smoking behaviour, the growth rate
of the simplified model was tuned (lowered to 70% of the initial pre-exponential value in Table 4.1) to
77
give the correct peak soot volume fractions. The results of these simulations are summarized in Figure
5.22.
(a)
(b)
Figure 5.22 – Integrated soot volume fraction (ppm · cm2) profiles of the Santoro et al. [76] flames of the F2 Non-smoking flame in (a) and the F4 smoking flame in (b).
78
Integrated soot volume fraction, which considers the total amount of soot at each axial height
throughout the flame, is defined as:
(5.4)
Here, represents the integrated soot volume fraction at a given height in the flame. As one can
observe from Figure 5.22, without using the correction factors, the smoking behaviour is not
reproduced by the simplified model as integrated soot volume fraction drops to zero after an axial
height of 10 cm in Figure 5.22b. However, with the correction factors employed, some soot “escapes”
the flame and the smoking behaviour is thus observed. In addition, the overall peak integrated soot
volume fractions remain the same in both the non-smoking and smoking cases. This correction factor
was then applied to the other methane/air flame simulations and negligible changes to predicted soot
volume fraction, soot diameter, and soot number density was observed. Thus, it was determined that
the correction factors for the oxidation models should be employed in future iterations of the
simplified model.
5.7.2. Updated Model Parameters and Improved Results
Model parameters were adjusted using a few differing strategies. One early strategy was to
attempt to match the global peak inception, surface growth, and oxidation rates predicted by the
detailed model to rates predicted the simplified model. While this proved successful for the specific
operating condition in which the constants were tuned, results were not favourable when the model
was applied to other operating conditions. A better strategy was to attempt to match the peak rates
along the pathlines of maximum soot by considering the peak rate predicted and the integrated sum of
the rates (i.e. the area under rate curves). This method, along with the conclusions made in Sections
5.3 to 5.6 helped guide the iterative process of improving the model predictions across all investigated
conditions. The effect of changing certain parameters on sooting behaviour could also be predicted by
using the results of the sensitivity analysis in Section 5.2. The final version of the proposed parameters
79
along with the original parameters in the simplified model is shown in Table 5.5. The author concedes
that while the parameters shown are improvements over the base simplified model, further
refinement is possible (especially with respect to model performance in Westport's engine simulations)
and is still being investigated at the time of writing. Further details on the approach used can be found
in Appendix B.
Parameter Fairweather et al. [46] Model
Base Simplified Model Improved Simplified Model
1.35E6 1.35E6 1.35E6
5.0E4 5.0E4 1.0E5
1.78E6 1.78E6 1.78E6
No OH oxidation model 864 864
90 000 90 000 90 000
3 3 9
Table 5.5 – Summary of parameters in simplified model. See Equation (5.3) for definition of correction factor (identical for both O2 and OH). Changes are highlighted in red and underlined. The results of the improved simplified model are summarized in the following figures. The rationale for
these changes were outlined in Sections 5.3 to 5.6 and Section 5.7.1 but are repeated briefly again for
the reader. The observed inception behaviour between the simplified model and the detailed model
were similar, so no changes were made. However, analysis indicated that both soot volume fraction
and mass averaged aggregate diameters needed to be increased, so the surface growth rate was
increased. Increasing the agglomeration rate ( ) also increased soot diameter while reducing .
Finally, a correction factor was applied to the oxidation rates in order to account for smoking and non-
smoking behaviour observed in the Santoro et al. [76] flame.
As the spatial distribution of predicted sooting behaviour did not change significantly in any of the
investigated conditions, only the peak values in the wing of the flame and the centreline of the flame
are reported for brevity. The unmodified model from Fairweather et al. [46] is also shown for
comparison. As one can observe, the behaviour of the unmodified model is very similar to the base
simplified model.
80
(a)
(b)
Figure 5.23 – Peak values of soot volume fraction predicted by the simplified model compared to experimental results in the centreline (a) and the wing (b) of the Smooke et al. [63] and Schittkowski et al. [78] flames. Figure 5.23 shows the changes in soot volume fraction predictions with the results of the improved
model compared to the original parameters and experimental measurements at atmospheric
conditions. In both cases the original model was underpredicting soot volume fractions in the wing of
the flame, but with the improved model, the predictions are now within experimental error. An
81
improvement in the unmodified Fairweather et al. [46] model is also seen in the prediction for peak
soot volume fraction in the wing of the Schittkowski et al. [78] flame.
(a)
(b)
Figure 5.24 – Peak values of mass averaged aggregate particle diameters predicted by the simplified model compared to the detailed model in the centreline (a) and the wing (b) of the Smooke et al. [63] and Schittkowski et al. [78] flames. The mass averaged aggregate particle diameters calculated by the simplified model are compared to
the diameters calculated by the detailed model in Figure 5.24 since no experimental results are
82
available for this particular soot characteristic. A numerical uncertainty of was assumed for the
detailed model. The improved model increases the diameter predictions in both cases in the wing of
the flame; however, there is a slight overprediction in the centreline of the Smooke et al. [63] flame.
This is not too concerning since there is great uncertainty in the behaviour of all soot models in the
centreline of the flame to begin with.
83
(a)
(b)
Figure 5.25 – Peak values of aggregate particle number density predicted by the simplified model compared to the detailed model in the centreline (a) and the wing (b) of the Smooke et al. [63] and Schittkowski et al. [78] flames. Similarly, the aggregate particle number densities computed by the simplified model are compared to
those calculated by the detailed model in Figure 5.25. The improved model reduces the overprediction
of soot number density in the Schittkowski et al. [78] flame while maintaining a reasonable prediction
84
in the Smooke et al. [63] flame in both the centreline and wing of the flames. Overall, the improved
simplified model is shown to rectify the deficiencies observed in Section 5.3 and 5.4 with regards to
soot volume fraction, particle diameter, and number density predictions.
(a)
(b)
Figure 5.26 – Peak values of soot volume fraction predicted by the simplified models and detailed model compared to experimental results from Joo and Gülder [80] in the centreline (a) and wing (b) of the flame.
Results for the high pressure cases are also presented, beginning with the calculated peak soot
volume fractions in Figure 5.26. The improved simplified model was still able to predict soot volume
85
fractions in the higher pressure flames within the prescribed experimental uncertainty. No significant
change in centreline peak soot volume fractions were observed, but predicted soot volume fractions in
the wing of the flame increased throughout all pressures. This reinforces the observation so far that the
centreline sooting behaviour is relatively insensitive to changes in the soot model. Also once again
observed is the overprediction in soot volume fraction in the centreline of the flame by the detailed
model. As stated earlier in Section 5.5, it is difficult to ascertain the significance of this overprediction
due to the many uncertainties regarding both the experimental measurements made in the centreline
and the performance of the soot model in the core of the flame.
86
(a)
(b)
Figure 5.27 – Peak values of mass averaged aggregate diameter as predicted by the simplified models and detailed model in the centreline (a) and wing (b) of the Joo and Gülder [80] flame. An uncertainty of is assumed for the detailed code calculations. While the particle diameters predicted by the detailed model are not validated at higher pressures,
they appear to reinforce the findings at lower pressures that the initial simplified model underpredicts
particle diameter size. The improved model, as seen in Figure 5.27, improves particle diameter
predictions in both the centreline and wing of the flame relative to the detailed code. When combined
87
with the earlier demonstrated improvements at one atmosphere for particle diameter, this does seem
to indicate the improved simplified model better predicts particle diameters.
5.8. Effect of Coupling – Gas Phase Species Consumption and Radiation
As previously mentioned in Section 3.4, the effect of gas phase species coupling and radiation
coupling to soot formation and oxidation was investigated. Computations using the original model
parameters were repeated with coupling disabled for all conditions in order to investigate the
importance of coupling soot formation/oxidation to the depletion of their respective gas phase species
and also the importance of the DOM radiation model on soot production with the results presented.
Figure 5.28 – The effect of coupling on the predicted peak soot volume fractions in the wing and centreline (CL) of the Smooke et al. [63] flame and the Schittkowski et al. [78] flame. The results of simulations, as seen in Figure 5.28, show that neglecting coupling at the one atmosphere
low sooting conditions seems to have minimal effect on predicted soot volume fractions. The largest
degree of increase is noted to be approximately a factor of 2.2 – 2.5 in the wing and centreline of the
flames when the model is fully uncoupled. At low sooting conditions (i.e. a peak of ~0.50 ppm),
radiation coupling appears to be more important than species coupling. Species coupling appears to be
88
nearly negligible – this is is not surprising since such a low amount of soot is formed, barely any species
relative to the total concentration is consumed in the first place.
Figure 5.29 – Graph illustrating the factor of increase in calculated peak values of soot volume fraction in the wings and centreline (CL) of the high pressure Joo and Gülder [80] flames. From left to right, the 10, 20, and 40 atm cases are shown plotted with respect to the maximum soot volume fractions measured in the experiment. A logarithmic scale is applied to the Y-axis.
Similar calculations were made for the high pressure cases, with the effect on soot volume
fractions shown in Figure 5.29. Unlike the earlier comparisons at low sooting conditions, coupling
effects are very significant at these highly sooting, high pressure conditions. With both radiation and
gas phase species coupling removed, soot volume fractions are observed to jump up significantly -
more than 15 times higher in the wing, and 12 times higher in the centreline in the 40 atm case.
However, the removal of the DOM radiation model on its own does not appear to affect the soot levels
significantly (approximately a factor of 1.2 overprediction at 140 ppm). On the other hand, the absence
of species coupling on its own has a noticeable effect (approx. a factor of 4.5 overprediction at 140
ppm), but not nearly as significant as the case where both mechanisms are removed (approx. a factor
of 14.5 overprediction at 140 ppm). Another important result that should be highlighted is that the
effect of coupling on overprediction of soot appears to increase non-linearly as the sooting behaviour
of the flame increases. As the majority of soot modeling work for turbulent flames has been done at
89
lower pressures with low peak soot volume fractions, it is possible that these previous studies would
not have demonstrated the significance of coupling effects on soot predictions. This result has major
implications for future turbulent soot modeling studies at higher sooting levels (ex. high pressure
combustion devices) as many current methods in turbulent combustion modeling are incompatible
with the need to track the depletion of species due to soot formation/oxidation. The need to include
energy feedback due to radiation is also evidently important. An important caveat of this conclusion is
that the residence time of typical laboratory scale flames is usually an order of magnitude or more
larger than the typical residence time in a combustion application such as an ignition compression
engine. In this scenario, the shortened residence time may reduce the effect of uninhibited soot
inception and growth as soot particles will escape the fuel-rich high sooting areas more quickly.
Originally, calculations were attempted to determine the importance of coupling species and
radiation in the detailed sectional model as well. While radiation effects were similar, it was found that
the detailed sectional model could not converge (within a reasonable timeframe) when species
coupling was disabled. This is due to the fact that inception (and some aspects of surface growth) in the
detailed sectional model is dependent on the concentration of pyrene, which is typically orders of
magnitude smaller than acetylene. The relative effect of consuming pyrene is therefore much higher
since there is a smaller amount of it to begin with. As such, when species coupling is disabled, the
amount of soot formed jumps up to unrealistic levels, causing the simulation to become unstable. A
similar problem could also be expected for any other soot model where soot chemistry is based on
species that are relatively low in concentration compared to major species. This was one of the primary
reasons an inception rate incorporating benzene in the developed simplified soot model was not
considered.
5.9. Computational Cost Comparison
Ultimately, the goal of developing the simplified model in the first place was to reduce the
computational costs involved with soot modelling in order to make it feasible to include in more
90
complex simulations. Table 5.6 shows a representative comparison of the computational costs of
running the two differing models for the high pressure methane air flame by Joo and Gülder [80]. Both
models were run using an initial guess of 1900 K with air populating the entire computational domain.
Pseudo-timestepping for the calculations started at an interval of 1E-07 seconds and increased by a
factor of two every 500 iterations. If the code diverged, then the time step was decreased by a factor of
two and restarted. In the case of the simplified model, the timestepping interval could be increased up
to 2E-04 seconds, whereas in the detailed model, the interval could only be increased to 4E-05
seconds.
Model Average Time per iteration (sec)
Iterations needed for convergence
Total Simulation Time (min)
Detailed Model 5.3 40000 – 50000 3562 – 4454
Simplified Model 3.6 15000 – 20000 893 – 1191
Table 5.6 – Representative comparison of computational costs of running the 2-D laminar flame code with the detailed section soot model and the simplified two equation soot model. As one can see, the detailed model can require upwards of four times the computational time relative
to the simplified model. Another important distinction that is lost in the simplified comparison above is
the fact that the detailed model also requires the use of a more complicated chemical kinetic
mechanism with PAH formation whereas the simplified model works reasonably well with the much
simpler GRI-mechanism. The difference in computational cost can be explained by the fewer equations
needed to solve the less complicated chemical mechanism as well as the few equations needed to
solve soot (two for the simplified model versus 70 for the detailed sectional code).
91
6. Concluding Remarks
6.1. Conclusion/Summary
A two equation simplified soot model was developed and applied to simulations of laminar, co-flow
methane-air diffusion flames. Calculations were made for three different data sets at 1 atm and 10, 20,
and 40 atm. Initial comparisons to available experimental data demonstrated relatively good success at
predicting peak soot volume fractions and also replicating spatial distributions, with some
underprediction of peak soot volume fractions in the 1 atm cases. A detailed sectional soot model,
which is well validated at atmospheric conditions, was used to augment the available experimental
data as detailed measurements of particle diameter and number density are rare in the literature. It
was found that at atmospheric conditions, soot volume fractions and particle diameters were being
underpredicted by the simplified model relative to the detailed model. Particle number densities were
also slightly overpredicted relative to the detailed model. At higher pressures, soot volume fraction
predictions in the wing of the flame were predicted within experimental uncertainty, but soot
diameters were too small relative to predictions made by the detailed model. Along with comparisons
of the soot inception, growth, and oxidation rates between the two models, this seemed to support the
idea that the simplified model had a growth rate that was too low and/or an agglomeration rate that
was too low and/or oxidation rates that were too high. Preliminary application to Westport's engine
simulations reiterated many of the above mentioned problems – namely, total soot mass was
underpredicted as were the soot particle diameters.
An improved simplified soot model incorporating modifications including an increased growth rate,
agglomeration rate, and oxidation factors was developed and applied once again to the investigated
laboratory flames. Predictions of soot volume fraction and particle diameters at 1 atm were improved,
while particle number densities remained within experimental uncertainty. At higher pressure
conditions, the soot volume fraction predictions remained within experimental uncertainty and the
soot particle diameter calculations were also improved. Overall, the developed simplified model was
92
able to reasonably predict soot volume fraction, aggregate particle number densities, and mass-
averaged aggregate particle diameters across a wide range of operating conditions using a single set of
parameters. The thesis also provides the framework for a methodology in developing a simplified soot
model using augmented datasets through a more advanced detailed soot model. Unique in this
approach is the wide range of validation and different aspects of soot morphology, which also includes
the use of numerical data from a more detailed soot model. This is a considerable advancement over
past exercises in simply "tuning" a model to a specific set of conditions to give reasonable results.
The effects of coupling soot formation/oxidation to the depletion of gas phase species was also
investigated. The lack of coupling contributes to overprediction in soot levels, with increasing
overprediction as the sooting tendency of the flame increases. If radiation modeling is also neglected,
the amount of overprediction more than doubles, even though the lack of radiation modeling on its
own appears to have little consequence. This need for coupling has major implications for future
studies of turbulent highly sooting flame applications, as steady flamelet approaches (or similar
turbulent combustion modeling approaches) cannot properly take into account species depletion
effects due to soot formation/oxidation.
6.2. Future Work
The developed simplified model has been demonstrated to be a viable candidate for use in engine
simulations at Westport but it is still expected that further development will be necessary before full-
scale use is possible. Based on the work completed in this thesis, the following future work is
recommended:
1) Extensive testing in engine simulations at Westport is needed to verify the simplified model’s
ability to reproduce quantitatively the observed soot measurements from experiments as well
as the simplified model’s ability to reproduce trends observed throughout varying engine
conditions.
93
2) Very little work in the literature has focused on the development of improved soot oxidation
models and the brief investigation conducted in this study has already demonstrated a need for
improving the soot oxidation mechanisms employed. Preliminary tests in engine simulations
have also alluded to the poor performance of existing oxidation models. The difficulty in
properly validating soot oxidation models is largely due to the propensity to develop soot
models for non-smoking flames. Future development of soot models should be validated for
smoking and non-smoking flames, but this is hindered by a lack of experimental data for
smoking flames.
3) Investigate methods or approaches to address the importance of coupling species and
radiation heat transfer in highly sooting applications. In addition to its importance at highly
sooting conditions, it is possible that there is a threshold residence time before which the
effect coupling species and radiation heat transfer is less significant. While there are no
detailed soot measurements for flames with respect to residence time, it is possible that
numerical simulations can determine when/if there is a threshold residence time where
coupling can be neglected.
4) Detailed sectional model improvements, especially at higher pressures, are needed in order to
provide more valuable input to the development of the simplified model. Some of the
necessary improvements include: validation of soot particle diameters and number densities at
high pressures, improvement of coalescence model to properly track primary particles and
aggregate structures, improved HACA surface growth mechanism so that does not need to be
tuned for specific cases, validated PAH growth in chemical mechanism at higher pressures, and
incorporation of improved oxidation models. However, the further development of detailed
soot models are hindered by the lack of experimental data, especially with respect to soot
diameter and aggregate structure studies.
94
References
[1] S.R. Turns, An Introduction to Combustion - Concepts and Applications.: McGraw Hill, 2006, ch. 1, pp. 1-8.
[2] K. Donaldson, L. Tran, L. A. Jimenez, R. Duffin, D. E. Newby, N. Mills, W. MacNee, and V. Stone, "Combustion-derived nanoparticles: A review of their toxicology following inhalation exposure," Particle and Fibre Toxicology, vol. 2, no. 10, October 2005.
[3] M.Z. Jacobson, "Strong radiative heating due to the mixing state of black carbon in atmospheric aerosols," Nature, vol. 409, pp. 695-697, Feburary 2001.
[4] A. D'Anna, A. Violi, A. D'Alessio, and A. F. Sarofim, "A reaction pathway for nanoparticle formation in rich premixed flames," Combustion and Flame, vol. 127, no. 1-2, pp. 1995-2003, 2001.
[5] N.A. Eaves, A. Veshkini, C. Riese, Q. Zhang, S.B. Dworkin, and M.J. Thomson, "A numerical study of high pressure, laminar, sooting, ethane-air coflow diffusion flames," Combustion and Flame, 2012, Article in press.
[6] M. R.J. Charest, C. P.T. Groth, and Ö. L. Gülder, "A numerical study on the effects of pressure and gravity in laminar ethylene diffusion flames," Combustion and Flame, vol. 158, pp. 1933-1945, 2011.
[7] N.A. Slavinskaya and P. Frank, "A modelling study of aromatic soot precursors formation in laminar methane and ethene flames," Combustion and Flame, vol. 156, no. 9, pp. 1705-1722, 2009.
[8] K. M. Pang, H. K. Ng, and S. Gan, "Simulation of temporal and spatial soot evolution in an automotive diesel engine using the Moss-Brookes soot model," Energy Conversion and Management, vol. 58, pp. 171-184, June 2012.
[9] A. J. Ricks, J. C. Hewson, A. R. Kerstein, J. P. Gore, S. R. Tieszen, and W. T. Ashurt, "A Spatially Developing One-Dimensional Turbulence (ODT) Study of Soot and Enthalpy Evolution in Meter-Scale Buoyant Turbulent Flames," Combustion Science and Technology, vol. 182, pp. 60-101, 2010.
[10] A Ishida, A Nishimura, M Uranishi, R Kihara, A. Nakamura, and P Newman, "The development of the ECOS-DDF natural gas engine for medium-duty trucks," JSAE Review, vol. 22, pp. 237-243, 2001.
[11] R. G. Papagiannakis and D. T. Hountalas, "Experimental investigation concerning the effect of natural gas percentage on performance and emissions of a DI dual fuel diesel engine," Applied Thermal Engineering, vol. 23, no. 3, pp. 353-365, February 2003.
[12] B.R. Stanmore, J.F. Brilhac, and Gilot P., "The oxidation of soot: a review of experiments, mechanisms and models," Carbon, vol. 39, pp. 2247-2268, 2001.
[13] H. Richter and J.B. Howard, "Formation of polycyclic aromatic hydrocarbons and their growth to soot—a review of chemical reaction pathways," Progress in Energy and Combustion Science, vol. 26, pp. 565–608, 2000.
[14] O.B. Popovitcheva, N.M. Persiantseva, Trukhin M.E., G.B. Rulev, and Shonija N., "Experimental chracterization of aircraft combustor soot: microstructure, surface area, porosity and water adsorption," Physical Chemistry, Chemical Physics, vol. 2, pp. 4421-4426, 2000.
[15] U. O. Köylü, G. M. Faeth, T. L. Farias, and M. G. Carvalho, "Fractal and projected structure properties of soot aggregates," Combustion and Flame, vol. 100, pp. 621-633, 1995.
[16] C. M. Megaridis and R. A. Dobbins, "Soot aerosol dynamics in a laminar ethylene diffusion flame," in Twenty-Second Symposium (International) on Combustion, Seattle, 1988, pp. 353-362.
[17] I Glassman, "Soot formation in combustion processes," in Twenty-Second Symposium (International) on Combustion, Seattle, 1988, pp. 295-311.
[18] B. S. Haynes and Wagner G. H., "Soot formation," Progress in Energy and Combustion Science, vol.
95
7, no. 4, pp. 229-273, 1981.
[19] J. Appel, H. Bockhorn, and M. Frenklach, "Kinetic Modeling of Soot Formation with Detailed Chemistry and Physics: Laminar Premixed Flames of C2 Hydrocarbons," Combustion and Flame, vol. 121, pp. 122-136, 2000.
[20] S.B. Dworkin, Q. Zhang, M.J. Thomson, N.A. Slavinskaya, and U. Riedel, "Application of an enhanced PAH growth model to soot formation in a laminar coflow ethylene/air diffusion flame," Combustion and Flame, vol. 158, no. 9, pp. 1682-1695, 2011.
[21] M. Frenklach, "Reaction mechanism of soot formation in flames," Physical Chemistry Chemical Physics, vol. 4, pp. 2028-2037, 2002.
[22] J. A. Miller and C. F. Melius, "Kinetic and Thermodynamic Issues in the Formation of Aromatic Compounds in Flames of Aliphatic Fuels," Combustion and Flame, vol. 91, pp. 21-39, 1992.
[23] C. F. Melius, M. E. Colvin, N. M. Marinov, W. J. Pit, and S. M. Senkan, "Reaction mechanisms in aromatic hydrocarbon formation involving the C5H5 cyclopentadienyl moiety," Proceedings of the Combustion Institute, vol. 26, pp. 685-692, 1996.
[24] H. Bockhorn, Ed., Soot formation in combustion: mechanisms and models. Berlin: Springer, 1994.
[25] S. J. Harris and A. M. Weiner, "Surface Growth of Soot Particles in Premixed Ethylene/Air Flames," Combustion Science and Technology, vol. 31, no. 3-4, pp. 155-167, 1983.
[26] M. Frenklach and Wang. H, "Detailed Mechanism and Modeling of Soot Particle Formation," in Soot Formation on Combustion, Mechanism and Models, H. Bockhorn, Ed. Berlin: Springer-Verlag, 1994, pp. 165-192.
[27] S. Macadam, J. M. Beér, A. F. Safofim, and A. B. Hoffmann, "Soot surface growth by polycyclic aromatic hydrocarbon and acetylene addition," in 26th Symposium (International) on Combustion, vol. 26, 1996, pp. 2295-2302.
[28] C. P. Fenimore and C. W. Jones, "Oxidation of Soot by Hydroxyl Radicals," The Journal of Physical Chemistry, vol. 71, no. 3, pp. 593-597, 1967.
[29] F. Liu, H. Guo, G. J. Smallwood, and Ö. Gülder, "Numerical modelling of soot formation and oxidation in laminar coflow non-smoking and smoking ethylene diffusion flames," Combustion Theory and Modelling, vol. 7, no. 2, pp. 301-315, 2003.
[30] J. Nagle and R. F. Strickland-Constable, "Oxidation of Carbon between 1000-2000 °C," in Proceedings of the Fifth Carbon Conference, New York, 1962, pp. 154-164.
[31] K. B. Lee, M. W. Thring, and J. M. Beér, "On the rate of combustion of soot in a laminar soot flame," Combustion and Flame, vol. 6, pp. 137-145, 1962.
[32] J. R. Walls and R. F. Strickland-Constable, "Oxidation of carbon between 1000–2400°C," Carbon, vol. 1, no. 3, pp. 333-338, 1964.
[33] K. G. Neoh, J. B. Howard, and A. F. Sarofim, "Effect of oxidation on the physical structure of soot," in Twentieth Symposium (International) on Combustion, 1984, pp. 951-957.
[34] F. Xu, A. M. El-Leathy, C. H. Kim, and G. M. Faeth, "Soot surface oxidation in hydrocarbon/air diffusion flames at atmospheric pressure," Combustion and Flame, vol. 132, pp. 43-57, 2003.
[35] R. A. Dobbins, R. A. Fletcher, and W. Lu, "Laser microprobe analysis of soot precursor particles and carbonaceous soot," Combustion and Flame, vol. 100, pp. 301-309, 1995.
[36] H. Kellerer, R. Koch, and S. Wittig, "Measurements of the growth and coagulation of soot particles in a high-pressure shock tube," Combustion and Flame, vol. 120, pp. 188-199, 2000.
[37] A. D'Alessio, C. Barone, R. Cau, A. D'Anna, and P. Minutolo, "Surface deposition and coagulation efficiency of combustion generated nanoparticles in the size range from 1 to 10 nm," in
96
Proceedings of the Combustion Institute, vol. 30, 2005, pp. 2595–2603.
[38] I. M. Kennedy, "Models of soot formation and oxidation," Progress in Energy and Combustion Science, vol. 23, no. 2, pp. 95-132, 1997.
[39] B. Blanquart, H. Pitsch, and M.E. Mueller, "A joint volume-surface model of soot aggregation with the method of moments," Proceedings of the Combustion Institute, vol. 32, pp. 785-792, 2009.
[40] I. M. Khan, G. Greeves, and D. M. Probert, "Air Pollution Control in Transport Engines," in The Institution of Mechanical Engineers, London, 1971, pp. 205-217.
[41] H. Hiroyasu, T. Kadota, and M. Arai, "Development and Use of a Spray Combustion Modeling to Predict Diesel Engine Efficiency and Pollutant Emissions," Bulletin of the JSME, vol. 26, no. 214, pp. 569-575, April 1983.
[42] Q. Zhang, H. Guo, F. Liu, G. J. Smallwood, and M. J. Thomson, "Modeling of soot aggregate formation and size distribution in a laminar ethylene/air coflow diffusion flame with detailed PAH chemistry and an advanced sectional aerosol dynamics model," in Proceedings of the Combustion Institute, vol. 32, 2009, pp. 761-768.
[43] M. Frenklach and H. Wang, "Detailed modeling of soot particle nucleation and growth," in Twenty-Third Symposium (Interational) on Combustion, 1990, pp. 1559-1566.
[44] V. Chernov, Q. Zhang, M. J. Thomson, and S. B. Dworkin, "Numerical investigation of soot formation mechanisms in partially-premixed ethylene–air co-flow flames," Combustion and Flame, 2012, Article in Press.
[45] P. R. Lindstedt, "Simplified soot nucleation and surface growth steps for non-premixed flames," in Soot Formation in Combustion, H. Bockhorn, Ed. Berlin: Springer-Verlag, 1994, pp. 417-439.
[46] M. Fairweather, W. P. Jones, and R. P. Lindstedt, "Predictions of Radiative Transfer from a Turbulent Reacting Jet in a Cross-Wind," Combustion and Flame, vol. 89, pp. 45-63, 1992.
[47] R. M. Woolley, M. Fairweather, and Yunardi, "Conditional moment closure modelling of soot formation in turbulent, non-premixed methane and propane flames," Fuel, vol. 88, pp. 393-407, 2009.
[48] J. B. Moss, C. D. Stewart, and K. J. Young, "Modeling Soot Formation and Burnout in a High Temperature Laminar Diffusion Flame Burning under Oxygen-Enriched Conditions," Combustion and Flame, vol. 101, pp. 491-500, 1995.
[49] S. J. Brookes and J. B. Moss, "Predictions of Soot and Thermal Radiation Properties in Confined Turbulent Jet Diffusion Flames," Combustion and Flame, vol. 116, pp. 486-503, 1999.
[50] S. Hong, M.S. Wooldridge, H.G. Im, D.N. Assanis, and H. Pitsch, "Development and application of a comprehensive soot model for 3D CFD reacting flow studies in a diesel engine," Combustion and Flame, vol. 143, pp. 11-16, 2005.
[51] M. R.J. Charest, C. P.T. Groth, and Ö. L. Gülder, "Effects of gravity and pressure on laminar coflow methane–air diffusion flames at pressures from 1 to 60 atmospheres," Combustion and Flame, vol. 158, pp. 860-875, 2011.
[52] D. Carbonell, A. Oliva, and Perez-Segarra C. D., "Implementation of two-equation soot flamelet models for laminar diffusion flames," Combustion and Flame, vol. 156, no. 3, pp. 621-632, March 2009.
[53] X. L. Zhu and J. P. Gore, "Radiation effects on combustion and pollutant emissions of high-pressure opposed flow methane/air diffusion flames," Combustion and Flame, vol. 141, pp. 118-130, 2005.
[54] C.B. Saji, C. Balaji, and T. Sundararajan, "Investigation of soot transport and radiative heat transfer in an ethylene jet diffusion flame," International Journal of Heat and Mass Transfer, vol. 51, pp. 4287-4299, 2008.
97
[55] H. Guo and Smallwood G. J., "The interaction between soot and NO formation in a laminar axisymmetric coflow ethylene/air diffusion flame," Combustion and Flame, vol. 149, pp. 225-233, 2007.
[56] H. Guo, F. Liu, and G. J. Smallwood, "Soot and NO formation in counterflow ethylene/oxygen/nitrogen diffusion flames," Combustion Theory and Modelling, vol. 8, no. 3, pp. 475-489, 2004.
[57] X. Zhu and J. P. Gore, "Study of flame structure and soot formation on heptane/air diffusion flame," AIAA Journal, vol. 42, no. 7, pp. 1491-1495, 2004.
[58] Y. R. Sivathanu and J. P. Gore, "Coupled radiation and soot kinetics calculations in laminar acetylene/air diffusion flames," Combustion and Flame, vol. 97, no. 2, pp. 161-172, May 1994.
[59] R. S. Mehta, D. C. Haworth, and M. F. Modest, "An assessment of gas-phase reaction mechanisms and soot models for laminar atmospheric-pressure ethylene-air flames," Proceedings of the Combustion Institute, vol. 32, no. 1, pp. 1327-1334, 2009.
[60] M. Frenklach and S. J. Harris, "Aerosol Dynamics Modeling Using the Method of Moments," Journal of Colloid and Interface Science, vol. 118, no. 1, pp. 252-261, July 1987.
[61] H. El-Asrag, T. Lu, C. K. Law, and S. Menon, "Simulation of soot formation in turbulent premixed flames," Combustion and Flame, vol. 150, no. 1-2, pp. 108-126, 2007.
[62] M. Frenklach, "Method of Moments with interpolative closure," Chemical Engineering Science, vol. 57, pp. 2229-2239, 2002.
[63] M. D. Smooke, C. S. McEnally, L. D. Pfefferle, R. J. Hall, and M. B. Colket, "Computational and experimental study of soot formation in a coflow, laminar diffusion flame," Combustion and Flame, vol. 117, no. 2, pp. 117-139, April 1999.
[64] S.R. Turns, "Soot Formation and Destruction," in An Introduction to Combustion - Concepts and Applications.: McGraw-Hill, 2006, ch. 9, pp. 343-346.
[65] R.W. Bilger, S B. Pope, K.N.C. Bray, and J.F. Driscoll, "Paradigms in turbulent combustion research," Proceedings of the Combustion Institute, vol. 30, pp. 21-42, 2005.
[66] H. Pitsch, "Large-Eddy Simulation of Turbulent Combustion," Annu. Rev. Fluid Mech., vol. 38, pp. 453-482, 2006.
[67] J. Buckmaster, P. Clavin, A. Liñán, M. Matalon, N. Peters, G. Sivashinsky, and F.A. Williams, "Combustion theory and modeling," Proceedings of the Combustion Institute, vol. 30, no. 1, pp. 1-19, 2005.
[68] S.R. Turns, "The Concept of a Conserved Scalar," in An Introduction to Combustion - Concepts and Applications.: McGraw-Hill, 2006, ch. 7, pp. 241-246.
[69] J. Huang, "Natural Gas Combustion Under Engine-Relevant Conditions," University of British Columbia, Vancouver, PhD Thesis 2006.
[70] H. Pitsch, E. Riesmeier, and N. Peters, "Unsteady flamelet modeling of soot formation in turbulent diffusion flames," Combustion Science and Technology, vol. 158, pp. 389-406, 2000.
[71] H. Pitsch, M. Chen, and N. Peters, "Unsteady flamelet modeling of turbulent hydrogen-air diffusion flames," Symposium (International) on Combustion, vol. 1, pp. 1057-1064, 1998.
[72] U. Maas and S.B. Pope, "Implementation of Simplified Chemical Kinetics Based on Intrinsic Low-Dimensional Manifolds," 24th Symposium (International) on Combustion, pp. 103-112, 1992.
[73] U. Maas and S.B. Pope, "Simplifying Chemical Kinetics: Intrinsic Low-Dimensional Manifolds in Composition Space," Combustion and Flame, vol. 88, pp. 239-264, 1992.
[74] N.J. Glassmaker, "Intrinsic Low-Dimensional Manifold Method for Rational Simplification of
98
Chemical Kinetics," University of Notre Dame, Notre Dame, Undergraduate Research Report 1999.
[75] J. Huang and W.K. Bushe, "Simulation of Transient Turbulent Methane Jet Ignition and Combustion under Engine-relevant Conditions Using Conditional Source-term Estimation with Detailed Chemistry," Combustion Theory and Modelling, vol. 11, no. 6, pp. 977-1008, December 2007.
[76] R. J. Santoro, T. T. Yeh, J. J. Horvath, and H. G. Semerjian, "The Transport and Growth of Soot Particles in Laminar Diffusion Flames," Combustion Science and Technology, vol. 53, no. 2-3, pp. 89-115, 1987.
[77] Q. Zhang, "Detailed Modeling of Soot Formation/Oxidation in Laminar Coflow Diffusion Flames," University of Toronto, Toronto, PhD Thesis 2009.
[78] T. Schittkowski, B. Mewes, and D. Brüggemann, "Laser-induced incandescence and Raman measurements in sooting methane and ethylene flames," Phys. Chem. Chem. Phys., vol. 4, pp. 2063-2071, 2002.
[79] K. A. Thomson, Gülder Ö. L., E. J. Weckman, R. A. Fraser, G. J. Smallwood, and D. R. Snelling, "Soot concentration and temperature measurements in co-annular, nonpremixed CH4/air laminar flames at pressures up to 4 MPa," Combustion and Flame, vol. 140, pp. 222-232, 2005.
[80] H. I. Joo and Ö. L. Gülder, "Soot formation and temperature field structure in co-flow laminar methane–air diffusion flames at pressures from 10 to 60 atm," Proceedings of the Combustion Institute, vol. 32, pp. 769-775, 2009.
[81] D. R. Snelling, K. A. Thomson, G. J. Smallwood, Ö. L. Gülder, E. J. Weckman, and R. A. Fraser, "Spectrally resolved measurement of flame radiation to determine soot temperature and concentration," AIAA Journal, vol. 40, no. 9, pp. 1789-1795, 2002.
[82] D. R. Snelling, K. A. Thomson, G. J. Smallwood, and Ö. L. Gülder, "Two-dimensional imaging of soot volume fraction in laminar diffusion flames," Applied Optics, vol. 38, no. 12, pp. 2478-2485, 1999.
[83] H. Guo, F. Liu, G.J. Smallwood, and Ö.L. Gülder, "Numerical study on the influence of hydrogen addition on soot formation in a laminar ethylene-air diffusion flame," Combustion and Flame, vol. 145, no. 1-2, pp. 324-338, 2006.
[84] M. W. Chase, C.A. Davies, J.R. Downey, D.J. Frurlp, R.A. McDonald, and A.N. Syverud, JANAF Thermochemical Tables, 3rd ed.: Springer, 1985.
[85] R. Kee, J. Dixon-Lewis, J. Warnatz, M. Coltrin, and J. Miller, "A fortran computer code package for the evaluation of gas-phase multicomponent transport properties," Sandia, Technical Report SAN86-8246, 1986.
[86] F. Liu, H. Guo, and G.J. Smallwood, "Effects of radiation model on the modeling of a laminar coflow methane/air diffusion flame," Combustion and Flame, vol. 138, pp. 136-154, 2004.
[87] R.O. Buckius and C.L. Tien, "Infrared flame radiation," International Journal of Heat and Mass Transfer, vol. 20, pp. 93-106, 1977.
[88] J. Huang, P.G. Hill, W.K. Bushe, and S.R. Munshi, "Shock-tube study of methane ignition under engine-relevant conditions: experiments and modeling," Combustion and Flame, vol. 136, pp. 25-42, 2004.
[89] F. Liu, K.A. Thomson, H. Guo, and G.J. Smallwood, "Numerical and experimental study of an axisymmetric coflow laminar methane–air diffusion flame at pressures between 5 and 40 atmospheres," Combustion and Flame, vol. 146, pp. 456-471, 2006.
[90] L. Talbot, R.K. Cheng, R.W. Schefer, and D.R. Willis, "Thermophoresis of particles in a heated boundary layer," Journal of Fluid Mechanics, vol. 101, pp. 737-758, 1980.
[91] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, 1st ed.: Hemisphere, 1980.
99
[92] R. Kee, J. Miller, and T. Jefferson, "Chemkin: A general purpose, problem independent, transportable, fortran chemical kinetics code package," Sandia, Technical Report SAN80-8003, 1980.
[93] R. Kee, F. Rupley, and J. Miller, "A fortran chemical kinetics package for the analysis of gas-phase chemical kinetics," Sandia, Technical Report SAN89-8009, 1989.
[94] R. Kee and J., Miller, J. Warnatz, "A fortran computer code package for the evaluation of gas-phase viscosities, conductivities, and diffusion coefficients," Sandia, Technical Report SAN82-8209, 1983.
[95] N.A. Slavinskaya, U. Riedel, S.B. Dworkin, and M.J. Thomson, "Detailed numerical modeling of PAH formation and growth in non-premixed ethylene and ethane flames," Combustion and Flame, vol. 159, no. 3, pp. 979-995, March 2012.
[96] W. Gropp, E. Lusk, and R. Thakur, Using MPI-2: Advanced Features of the Message Passing Interface, 1st ed.: The MIT Press, 1999.
[97] S. Will, S. Schraml, and A. Leipertz, "Comprehensive Two-Dimensional Soot Diagnostics Based on Laser-Induced Incandescence (LII)," Twenty-Sixth Symposium (International) on Combustion, pp. 2277-2284, 1996.
[98] Q. Zhang, M.J. Thomson, H. Guo, F Liu, and G.J. Smallwood, "A numerical study of soot aggregate formation in a laminar coflow diffusion flame," Combustion and Flame, vol. 156, no. 3, pp. 697-705, 2009.
[99] S. Hong, D. Assanis, and M. Wooldrige, "Multi-Dimensional Modeling of NO and Soot Emissions with Detailed Chemistry and Mixing in a Direct Injection Natural Gas Engine," SAE Technical Papers, no. 2002-01-112, January 2002.