You are on page 1of 38

Prog. Energy Combust. Sci. Vol. 23, pp.

95-132, 1997
Pergamon © 1997 Elsevier Science Ltd. All rights reserved
Printed in Great Britain
0360-1285/97 $29.00

PII:S0360-1285(97)00007-5

MODELS OF SOOT FORMATION A N D OXIDATION

IanM. Kennedy
D~artment of Mechan~al and Aeronautwal Engineering, Unive~i~ of Cal~ornm, Dav~, CA 95616, U.S.A.

Abstract--Advances in modeling soot formation and burnout in combustion systems are surveyed. The
types of models are divided up into three classes: empirical, semi-empiricaland detailed. Empirical models
use correlations of experimental data to predict trends in soot loadings. Semi-empiricalmodels solve rate
equations that are calibrated against experimentaldata. Detailed models seek to predict the concentrations
of all the important species in a flame, from fuel to polyaromatic hydrocarbons to soot. The three classes of
models have demonstrated success in predicting soot concentrations. However, our knowledge of some of
the fundamental underlying is still open to question and the success of the models has been obtained to
some degree by adjustments to fit measurements. © Elsevier Science Ltd.

CONTENTS

1. Introduction 95
2. Background to Soot Formation and Oxidation 95
3. Soot Models 99
3.1. Empirical Correlations 99
3.2. Semi-empiricalModels 102
3.3. Modelswith Detailed Chemistry 119
4. Concluding Remarks 129
References 129

1. I N T R O D U C T I O N into the controlling physical and chemical mechan-


isms can be distilled into models that may be
Great strides have been made over the last few formulated into either simple design guidelines,
decades in our understanding of the mechanisms and empirical correlations, semi-empirical computer
the phenomenology of soot formation and burnout in models or full-blown attempts to describe the
combustion systems. The attention that this problem detailed elementary chemical reactions and physics
has received reflects the very difficult problem that of soot formation. The latter two approaches require
soot formation poses for combustion system the use of computers that are used increasingly in the
designers and operators of combustion systems. design of advanced combustion systems.
Although progress has been achieved in understand- This review offers a critical appraisal of the models
ing the essential chemistry and physics, many of soot formation that are currently available. The
questions persist and some debate continues with models are grouped into three categories viz., (i)
regard to the details of soot formation, growth and purely empirical correlations, (ii) semi-empirical
oxidation. The relative importance of the various approaches that solve rate equations for soot
stages of soot formation is open to question and may formation with some input from experimental data
well be system-dependent. and (iii) detailed models that seek to solve the rate
Despite a continuing uncertainty in some aspects of equations for elementary reactions that lead to soot.
soot phenomenology, combustion engineers are faced While there is inevitably overlap between these
with a pressing need to design systems in which the categories, the classification is nevertheless useful in
amount of soot is controlled. Diesel engine designers bringing order to a large amount of material.
face a problem that is perhaps the most difficult of any
combustion based industry. They are confronted with
a severe regulatory limit on emissions of particulates 2. B A C K G R O U N D T O S O O T F O R M A T I O N A N D O X I D A T I O N

from their engines. At the same time, they must also


meet stringent limits on the emission of the oxides of Soot is mostly carbon; other elements such as
nitrogen. Conventional engineering solutions through hydrogen and oxygen are usually present in small
the development of numerous prototypes may no amounts. It is produced during the high temperature
longer be either economically attractive or, perhaps, pyrolysis or combustion of hydrocarbons. The
even feasible. The regulatory agenda require rapid emission of soot from a combustor or from a flame
responses on the part of combustion designers. is determined by the competition between soot
The formulation of soot models has been a less than formation and oxidation. A comprehensive model
rigorous science until recently. However, new insights of the soot process must include both phenomena.
95
96 I.M. Kennedy

A great deal of information has been obtained over processes are unimportant when modeling soot
the years with regard to the formation step of soot in formation. Alternatively, models presented by
shock tubes, laminar premixed flames and laminar Colket and Hall 3 for premixed flames indicated that
diffusion flames. The review by Haynes and Wagner l an approximately direct proportionality exists
remains one of the best sources for a full description between the soot inception rate and the total mass
of this phenomenon. As hydrocarbons pyrolyze, they fraction of soot, at least for lightly sooting flames. As
produce primarily smaller hydrocarbons, in particu- soot mass fractions increased for increasingly rich
lar acetylene. The initial step in the production of soot flames, this direct proportionality was lost as the
is the formation of the first aromatic species from importance of surface growth processes increased.
these aliphatic hydrocarbons. The details of this More recently, Hall e t al. 4 have modeled soot
process have been the subject of some considerable formation in a counterflow diffusion flame and
debate over the years; discussion is deferred until a later found a similar dependence on the inception process
section in this review. The aromatic species grow by the at very low sooting levels. However, as soot volume
addition of other aromatic and smaller alkyl species to fractions increased, this dependence rapidly decreased.
form larger polyaromatic hydrocarbons (PAH). Con- Hence, the earlier conclusions of Kennedy et al. 2
tinued growth of the PAH leads eventually to the smallest require revision so as to account for the importance of
identifiable soot particles with diameters of the order inception in lightly sooting flames.
of 1 nm and with masses of around 1000 amu. An excellent example of the importance of the
The production of soot particles in a flame is inception/nucleation process and its attendant rate to
inherently a chemically-controlled phenomenon. modeling soot formation under conditions of light
Low molecular weight (gaseous) hydrocarbons are soot loading is the thorough analysis performed by
converted to essentially solid carbon in just a few Markatou et al. 5 in their examination of sooting limits
milliseconds. Thermodynamics alone cannot describe in premixed flames. Using a detailed chemical kinetic
this process since soot is formed beyond regimes model of gas-phase chemistry, of soot particle
where it is thermodynamically stable relative to the nucleation and of surface growth, they were able to
oxides of carbon. Hence, chemical kinetics play an predict the thresholds (i.e. critical equivalence ratio)
important role in soot production. Chemistry occurs at which visible (radiant) soot first appeared over a
during nearly all phases of soot production: inception, range of flame temperatures and fuels. By showing
surface growth, aging, and surface oxidation. In that the rates of surface growth, oxidation, and
addition, it is now recognized that soot may coalescence varied slowly in the regime of the critical
participate chemically in the reduction of another sooting limit, while also showing a dramatic rise in the
undesirable pollutant, nitric oxide. computed concentration of the 'incepting' or 'nucle-
Particle inception initiates soot production in the ating' species (in this case, acepyrene), they were able
sense that the growth and oxidation of soot particles to conclude 'that soot particle inception and hence the
occur in some qualitatively different manner to the sooting limit is determined by particle nucleation.'
pre-particle, molecular chemistry. Some early soot Thermodynamics alone cannot adequately describe
formation models were based entirely upon the soot formation phenomena; it is inherently a kinetically
inception process. Conclusions drawn from these limited process. However, thermodynamics, in parti-
models indicated that the net amount of soot cular those related to detailed balancing of reactions,
production was linearly dependent on the rate of can dramatically affect the rates of initiation and
soot inception. This conclusion was based on the growth of soot. A simple example is the reversible
argument that the surface growth rates were linearly sequence
dependent upon the available surface area which was
C6H5 + C2H2 *-* (C6HsC2H2)* ~ C6HsC2H + H
dependent, in turn, upon the total number of particles
that were created during inception or nucleation. which interrelates the concentrations of phenyl
Hence, many research groups focused efforts over the radicals, vinylphenyl radicals, acetylene, phenyl
last fifteen years towards developing an understand- acetylene and H-atoms. In other words, the net
ing of these processes. On the other hand, Kennedy forward rate of this reaction is in general limited by
et al. 2 used a semi-empirical model with a nucleation reversible reactions. The kinetic expression, k[C2H2]
rate that was only a function of mixture fraction to [C6H5] , represents an upper limit for the net rate of
show that for a coflow diffusion flame, the total production of phenyl acetylene and vinylphenyl
amount of soot was not strongly dependent on the radicals. Predictions of PAH depend significantly on
soot inception rate; instead surface growth processes the selected thermodynamics.
controlled soot production and total soot volume Stein 6 identified a series of stable PAHs, represent-
fractions. They supported this contrary conclusion by ing islands of stability during the growth process. He
arguing that coalescence of a great number of predicted that PAH formation and soot formation
particles quickly hid the history related to the would proceed along this path. Frenklach et al. 7
inception process. This conclusion has significant identified similar species as key intermediates to the
implications since it implies that knowledge of the PAH formation and growth process and argued that
kinetic rates for initiation and the detailed associated these species essentially 'pulled' the chemistry along
Models of soot formation and oxidation 97

the soot formation path. Other less stable species (C2H2) , although PAHs may also play a role. 8 Rather
would be subject to decomposition and remain in more controversy has arisen over the role played by
equilibrium with lower molecular weight species. The the aerosol surface area. Some researchers proposed
PAH stable isomers, however, were much less suscep- that the surface area of the soot aerosol determined
tible to decomposition and hence, once formed, would rates of growth) 4'15 In this view, the rate of growth is
continue to grow towards larger species. A host of described by the aerosol surface area and the
experimental studies, together with the successes in concentration of growth species. However, experi-
modeling soot formation via a PAH route (see later ments by Wieschnowsky et al.16 indicated that surface
discussions), provide strong evidence that PAHs are area may not be the only relevant parameter. They
the key ingredients leading to soot production. The found that aerosols with the same apparent surface
article by Howard s is an excellent source for a survey areas exhibited different growth rates. They attributed
of growth and oxidation of PAH and soot from a the differences to the important role of active sites on
fundamental viewpoint in which phenomena are the surface of particles in determining growth rates.
interpreted in terms of active sites. The growth rate of soot was described in terms of a
Calcote and coworkers 9 have advocated the first order rate equation for the time rate of change of
importance of a link between ions in flames and soot volume (or mass ) fraction fv as ~ -- k(Jv* - f v )
soot formation. Excellent arguments and data have where k is a rate constant and fv* is the maximum soot
been presented showing a strong relationship between volume fraction that is attained. While this equation
soot and ions. Many of these arguments are similar to can represent the physical observations adequately, it
those for the PAHs, in that the appearance of ions is not useful as a predictive relationship in a model of
occurs just before visible soot appears. Advocates for soot formation since the maximum soot volume
a free radical mechanism of soot growth have fraction,fv*, is not known a priori.
countered the arguments of the ionic mechanism Soot formation in diffusion flames usually occurs
proponents, leading to an ongoing debate that is low in the flame and is followed by oxidation as soot is
beyond the scope of the present review. Suffice it to convected through the tip of the flame. Oxidation
say, the ionic viewpoint has not been implemented in occurs primarily as a result of attack by molecular
a predictive model of soot production in flames that oxygen, 02, and the OH radical. Other oxygenated
has met with wide acceptance. species such as the O atom, H20 and CO2 may be
Once soot particles are formed through the important under some conditions. Lee et al. 17
inception process, they can grow by two mechanisms measured the O2 dependence and the temperature
viz., collisional coagulation and surface growth. The dependence of the oxidation of soot in a laminar
former process is physical while the latter process is diffusion flame. Samples of soot and the gases were
chemical in nature. When particles are small and collected at different locations above the flame. The
surface growth is active, collisions between particles gas samples were analyzed on a gas chromatograph.
generally lead to the formation of a larger spheroid The size of the soot particles was determined with an
via the process of coalescence. Older particles electron microscope. The mass flow rate of soot was
undergo agglomeration in which the individual found from the samples. They assumed that the
spheroids are retained in long chains with a fractal particles were small enough that diffusion did not
like geometry. The rates of coagulation in the free limit the rate of oxidation; the kinetics of surface
molecular limit of Knudsen number Kn ~ c~ reactions were assumed to be the limiting mechanism.
(Kn = @ where the particle diameter is dp and the Temperatures in the experiments ranged from 1200 to
mean free path of the gas is A) and in the continuum almost 1700 K. The oxygen partial pressures ranged
regime ( Kn << 1 ) are fairly well understood. More from 4 × l0 -2 to 12 x 10-2atm. They arrived at the
difficulties arise in the transition regime in which an following expression for the oxidation rate of soot
interpolation formula such as Fuch's expression is over the ranges of temperatures and 02 partial
applied, l0 van der Waals forces may play some role in pressures that they considered:
increasing the rate of coagulation as Harris and
Kennedy showed) U2 In the absence of surface pd o dm
~Vo2 -
growth, an aerosol will tend to produce a so-called 6 m~/3 m 2/3 dt
self preserving size distribution under the impact (1)
of coagulation. 13 The self preserving size distribution
,As po2 ( 164,500
= 1.085 × lo ~T/ 2 e x p _ RT J
leads to useful relationships between various moments
of the distribution. The form of the distribution may where
be altered somewhat by van der Waals forces and by wo: is the surface specific oxidation rate (kg m -2
surface growth.I t sec- 1)
Surface growth of particles proceeds in conjunction do is the initial particle diameter (m)
with coagulation. Gas phase material is added p is the soot density (kg m -3)
through chemical reactions with active sites on the m0 is the initial mass flow rate of soot (kg sec -1)
soot particles. It is generally agreed that the major m is the mass flow rate of soot after time t (kg sec- I)
growth species in hydrocarbon flames is acetylene Po2 is the oxygen partial pressure (atm)
98 I.M. Kennedy

T is the temperature (K) formulation to improve the prediction of smoke


R is the universal gas constant (kJ kmol -l K - I ) levels from a gas turbine engine. Well-controlled
The most widely used model for 02 oxidation comes measurements under simpler conditions are desirable.
from Nagle and Strickland-Constable.lS They Modeling of OH attack on soot is currently on a
measured the oxidation of a pyro graphite over less fundamental level than 02 modeling, despite the
temperatures from 1000 to 2000°C and pressures from significant uncertainties that surround the latter issue.
0.1 to 0.6atm. The impact of mass transfer was Fenimore and Jones 21 pointed out the role of OH in
reduced by directing a high speed jet of 02 towards the burnout of soot in flames although they did not
a heated carbon rod. The measurements were have direct measurements of OH. Bradley et al. 22
interpreted in terms of a model that assumed two measured oxidation rates of graphite powder in a flat,
types of active sites were present on the carbon. Type laminar, methane-air premixed flame at sub-atmo-
A sites reacted with 02 to give another A site plus spheric pressure. They concluded that radicals such as
CO OH and O were the main oxidants of graphite at
high temperatures. Mor~ direct measurements have
A + 02 -* A + 2CO at a rate given b y ~ x indicated OH to be an important oxidant of soot,
especially on the fuel side of a diffusion flame23. The
where p is the pressure in atmospheres.
rate of OH attack on soot is usually taken from the
Type B sites reacted with 02 to give a Type A site measurements of Neoh et al. 24 in the absence of other
plus CO relevant data. They found that the efficiency of a
B + 02 --* A + 2CO at a rate of knp(1 - x). collision of OH with a soot particle is of the order of
Finally, Type A sites thermally rearranged to give 0.1. This result was confirmed by Roth et al. 2s For
Type B sites comparison, Bradley et al. 22 reported a collision
A --* B at a rate of kTx. efficiency of 0.28 for the reaction of OH with
powdered graphite. The difficulty in the prediction
The parameter x was defined as x -_ l+-7-(ff/ff~"
I The
of OH oxidation in diffusion flames is exacerbated by
rate constants were assigned values of the super-equilibrium levels of OH that are found to
k A = 2 x 104 e -130'000/Rr exist in these conditions. 23 The accurate prediction of
non-equilibrium concentrations of OH poses a
kB = 4.46 e -~'°°°/Rr challenge for modelers who hope to account
accurately for OH attack on soot.
kT = 1.51 x 108 e -4°6,°°°/Rr
Uncertainties in the modeling of soot formation
k z = 2.14 x 104 e +17'°°°/Rr. and oxidation are compounded by experimental
uncertainties in the determination of soot volume or
where R = 8.3144kJ kmo1-1 K - l . The units of the mass fractions in flames. The later stages of produc-
rate constants were (kg-atoms of carbon, m 2, seconds, tion are generally characterized by the transition from
atmospheres). The overall reaction rate was given by coalescent coagulation to agglomeration. Agglomera-
= ( kAp ~x+ksp(l_x ) (2) tion leads to the formation of chains of spherules
fi,°2 \ 1 + kz pJ in which the basic units have diameters of about
At low temperatures, the surface will be entirely 30-50 nm. Soot that is emitted from combustors is
Type A sites with the reaction rate first order in 02 usually in an agglomerated form. The morphology of
partial pressure for sub-atmospheric total pressures, the agglomerates may be important in predictions of
becoming independent of 02 concentration near one soot formation 26-3° and is certainly important in the
atmosphere. Despite the wide acceptance of the interpretation of light scattering data from the upper
Nagle-Strickland-Constable model, some reserva- reaches of laminar diffusion flames.
tions are in order. The measurements and rate Soot volume fractions and number densities have
constants were obtained under conditions that differ been measured typically by laser light extinction. It
significantly from the conditions under which soot is was usually assumed that the particle scattering was in
oxidized in flames. The composition of soot is not the the Rayleigh limit; in this case, light scattering was
same as the pyrographite that Nagle and Strickland- negligible in comparison with light absorption for
Constable used. Their maximum 02 partial pressure soot particles. However, Choi et alfl and Koylii and
was 0.6 atm. Soot burn out in diesel engines and in Faeth 2s'29 pointed out that soot aggregates in flames
gas turbines may occur with much higher 02 partial are not well described by the Rayleigh limit. In fact,
pressures although Radcliffe and Appleton 19 suggested the scattering cross section of soot that is emitted by
that the Nagle-Strickland-Constable formula could flames can be up to 41% of the absorption cross
be extended to higher pressures. A need clearly exists section. For example, Dobbins et al. 32 found that
to augment the data of Nagle and Strickland- when optical data were analyzed with an aggregate
Constable for conditions more appropriate to model instead of a Rayleigh model, the soot volume
hydrocarbon combustion in modern engines. For fraction was 30% less at the midpoint of a diffusion
example, Najjar and Goodger 2° modified the rate flame. The soot aerosol surface area was a factor of
constants in the Nagle-Strickland-Constable two to four greater when an aggregate model was
Models of soot formation and oxidation 99

//
1.2
-!1- Gravimetdc

-0- Optical
0

~- 0.8
._o

0.6-

>
"5
0
03
0.4-

0.2-
f
0 . . . . [ . . . . I' ' ' '1 .... I .... I ....

2.2 2.3 2.4 2.5 2.6 2.7 2.8


Equivalence ratio

Fig. 1. A comparison of soot volume fractions determined gravimetricatly and determined optically (from
Choi et al.31).

applied to the optical data in place of the usual tions of X¢ been proposed over the years. 36'37 In
Rayleigh analysis. The value of the complex refractive diffusion flames, the onset of soot emissions from a
index of soot is also open to question, particularly flame is gauged in terms of the sooting height, the
because it may change from fuel to fuel and even flame height at which soot is emitted. It is important
within a given flame. 33 to be able to estimate the sooting tendency of fuels
Choi et al. 31 determined soot volume fractions and mixtures of fuels that may be encountered in
gravimetrically and compared the results with laser practice. Calcote and Manos 3s proposed the use of a
extinction measurements of the soot volume fraction threshold sooting index (TSI) as a way of removing
(Fig. 1). The optical measurements used the refractive any system or burner dependence from measures of
index that was reported by Dalzell and Sarofim 34 and the sooting tendency of fuels. The TSI was defined as
that has been applied widely in the analysis of light
T S I = a - bXe (3)
extinction data to extract soot volume fractions.
The optical results typically overpredict the gravi- where the apparatus dependent constants a and b
metric results by about a factor of two. Therefore, one were determined by calibration. Gill and Olson 39
must exercise some caution in comparing modeling adopted this definition to predict soot thresholds for
with experiments in light of the apparent uncertainty fuel mixtures by using the individual TSIs of each
in the interpretation of the light extinction data. component. The mixing rule that they selected for
premixed flames was

3. S O O T M O D E L S (1.1) Tslm'' = EXj(1.1)TSb (4)


]
The successful modeling of the soot yield or the The T S I of each component of the fuel was T S I 2 and
critical C/O ratio of a premixed flame, or the smoke mole fraction of component j in the mixture was Xj.
point of a diffusion flame, requires in general an This correlation performed quite well for a ternary
accurate account of both formation and oxidation. fuel mixture of isooctane, decalin and l-methyl-
Unlike NOx formation, soot production is also naphthalene. Olson et al. 36 reported on the extension
inextricably linked to the flame structure through its of the threshold soot index (TSI) to laminar diffusion
impact on flame temperatures via radiative heat flames for a wide variety of fuels. The TSI approach
loss. 35 Predictions of soot yield or smoke point was able to account for the variation in measured
height remain amongst the greatest challenges that smoke point heights from different apparatus and
continue to face modelers. Advances will be ham- hence provided a useful means for assessing the
pered by uncertainties in the fundamental rates of sooting tendencies of fuels and mixtures.
processes such as 02 oxidation or surface growth. Many measurements have been made of the critical
Progress will depend on improvements in these areas. C/O ratio at the onset of sooting in premixed flames.
Nevertheless, some useful predictions of soot forma- Takahashi and Glassman 37 attempted to collapse
tion can be obtained from relatively crude models. premixed flame data by considering an effective
equivalence ratio, • = [C + H/2)]/O. Their analysis
3.1. E m p i r i c a l Correlations of the experimental results for the critical 9 at the
onset of soot led them to conclude that the sooting
Soot begins to form in a premixed flame at a critical tendency of a fuel was a function of its flame
or threshold equivalence ratio, Xc. Several correla- temperature, the C/H ratio and the number of C
100 I.M. Kennedy

1.4 • 1100 RPM


% 1.2--'1

11 . y
[] A [] • 2000 RPM
• 2700 RPM
• 1100 RPM
0.8 []
[] 2000 RPM
O 2700 RPM
07 .
,x High swirl
o.7 O Med swirl
[] Low swirl

0 ' '012' ' '014" '018' ' '018 ' ' '1.2
Predicted exhaust soot ( g rn"s )

Fig. 2. Comparison of measurements and predictions of soot in a diesel engine exhaust. 41 Points represent
predictions and measurements for the same engine conditions. (--) represents perfect agreement.

atoms but not the structure of the fuel. They that the diameters of soot particles from engines that
considered a wide range of aliphatic and aromatic were operated over very different speeds and loads did
fuels that were burned in a premixed flame a fixed not vary. As a result, they further assumed that the
value of the adiabtic flame temperature. The critical rate of soot formation is controlled entirely by the
was plotted as a function of the 'number of C - C formation of soot particles, i.e. by the soot inception
bonds' in which double carbon bonds were counted as rate. The rate of particle formation was assumed to be
two and triple bonds as three. This simple analysis a function of the pressure, the equivalence ratio of
yielded an excellent correlation of the experimental unburned gases and the temperature. They proposed
data. a general form of the soot equation as
Harris et aL 4° reconsidered the experiments of
n -E
Takahashi and Glassman. 37 They made detailed dCs/dt = c Vu Pu X (5)
measurements of temperature by sodium D line
reversal in a premixed flame of various aliphatie where Cs is the soot mass loading in (kg m-3), c is a
fuels. The measurements of this fundamental para- modeling parameter, Vu is the volume of the soot
meter permitted a realistic assessment to be made of formation zone in (m3), VNTP is the volume of the
the impact of fuel type and temperature on soot cylinder contents at normal temperature and pressure
thresholds without resort to the assumption of in (m3), Pu is the partial pressure of unburned fuel in
adiabatic flame temperatures. Harris et al. 4° assumed (Pa), X is the local unburned equivalence ratio, E is
that the onset of sooting was the manifestation of the activation energy and Tu is the local temperature.
competition between pyrolysis and oxidation, pri- The modeling parameters c, n and E were determined
marily OH oxidation. This allowed them to develop a by comparing the output of the model with experi-
satisfactory correlation of Xc for five aliphatic fuels ments on a direct injection diesel engine with a
with the inverse measured flame temperature as the displacement of 966cm 3 and compression ratio of
independent variable and Xc normalized by the OH 16:1. A value of n = 3 was found to give the best
concentration and C/H ratio of the fuel as the results although discrepancies of about a factor of 2
dependent variable. were apparent at injection timings of around 30 °
Most of the empirical modeling of soot formation BTDC. A similar exercise yielded a value of c of about
and emissions can be found in literature that is related 0.468 kg. N - l m -1 sec -l.The activation energy E was
to gas turbine and diesel engines. Both types of devices taken to be 1.7 x 105 kJ kmol - l . Two other important
can suffer from unacceptable emissions of smoke under parameters that needed to be considered were related
certain conditions. The complexity of the combustion to the mixing model in the cylinder. An entrainment
processes in engines has generally precluded, until ratio governed the rate of large scale mixing that
recently, the use of detailed or semi - detailed modeling varied with engine speed and with swirl ratio. A
of the basic soot formation mechanisms. On the other diffusivity constant determined the microscale mixing
hand, empirical models of soot formation that are of reactants and hence had an impact on heat release
based on a phenomenological description have found rates. The diffusivity constant was determined from
wide use in these communities. fitting predicted heat release rates to measured rates.
Khan et al. 41 proposed a model for soot emissions The entrainment ratio was found from a similar
from diesel engines that has been widely cited in the approach. The entrainment ratio had a strong impact
literature. Their scheme was based upon the notion on predicted soot levels.
Models of soot formation and oxidation 101

Khan et al. 42 compared the results of their model 0.9


with experiments for a direct injection diesel over a
range of speeds and injection timings. The solid line in
Fig. 2 indicates a perfect agreement with experiments;
the scatter demonstrates the error in the modeling.
"~ 0.7
8 1 / •

In general, the agreement was good over a fairly


wide range of speeds and injection rates. The authors
considered some of the details of their modeling. In
particular, they examined the trends in predicted
~ 0.5

0.t
1 ./•
/
exhaust soot with variations in the injection timing,
the injection period and the amount of fuel injected. 1 • f.
" •
They were able to demonstrate the correct trends with
these parameters by comparison with measurements. if_ 0 . 1 ~
A notable omission in the model, as the authors 0 '1 .... I .... I .... I .... I .... I''''1 ....

themselves acknowledged, related to oxidation of 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
soot. There was no account at all of this process. The Experimental soot concentration (g m"3 )
satisfactory agreement that they obtained must be
considered somewhat fortuitous in view of this Fig. 3. Predictions and measurements of exhaust soot
omission. The use of enough parameters in the concentrations from diesel engines.43 Points represent
predictions and measurements for the same engine
model to fit experimental data ensured some degree conditions. (--) represents perfect agreement.
of agreement. A similar agreement for other engine
types cannot be guaranteed by a model that omits a formation was given as
fundamental process.
Mehta and Das 43 summarized earlier correlation dCs/dt = A T a c b C ~ e x p ( - E / R T ) (6)
efforts for diesel engines and, in turn, proposed an where for kerosene combustion the model had the
improved correlation of their own. They concluded following specific form for soot formation
from their review of the literature that available dCs/dt =4.4 × lOl7 Cfexp(-121OO/T)
correlations included general operating parameters
such as the overall fuel-air ratio, the injection timing - 3 × 109Coexp(-7800/T)
and the engine speed but neglected the impact of fuel
[kmol m-3sec -1] (7)
injection and the in-cylinder air motion on soot
processes. Mehta and Das 43 proposed a seven wherein the second, negative term accounted for soot
parameter correlation that included an account of oxidation.
the spray mixing rate and the swirl mixing rate. The Soot production within gas turbines has been
other variables that they considered were the fueling modeled with empirical correlations that are coupled
rate, the compression ratio, the temperature at to detailed predictions of the flow field within the
injection and the engine speed. A linear regression combustor. The detailed calculations of the flow
analysis was applied to the results of tests on two within the combustor were used as inputs to
engines, a six cylinder and a single cylinder engine. correlations for CO, NO and soot formation and
The comparison of predicted soot emissions with oxidation. A three-dimensional code was tested on a
measurements (Fig. 3) showed about the same quality simulation of an Allison T56 combustor operated on
of agreement as shown in Fig. 2 from the results of the conventional aviation turbine fuels as well as high
computations of Khan et al. 41 The generality of the density fuels. 47'48 The combustor was divided into
correlation of Mehta and Das 43 is not likely to be sub-regions in which the rates of CO, NO and soot
good although the use of an empirical correlation formation were calculated. The authors used 800 of
certainly enjoys a significant advantage in economy these sub-volumes.
of calculations compared to the more detailed Soot formation and soot oxidation were handled
approaches that have been attempted. separately in each sub-volume. Soot formation was
The gas turbine community has been interested in described by the empirical expression
soot emissions and radiant heat transfer from soot
for many years. In a review of flame radiation in fi~soot(mg/kg) = CP2(18 - H) 1.5 [ (F/A)mb ]
kTmg~.2~j (8)
gas turbine engines, Lefebvre44 cited the work of
Khan and Greeves45 who applied Eq. (5) to the soot where C was a modeling constant, P was the liner inlet
formation problem. Soot oxidation was not consid- pressure (kPa), H was the weight percentage of
ered in this model. hydrogen in the fuel, F/A was the overall fuel-air
Edelman and Harsha 46 adopted a slightly more ratio, mb was the mass fraction of fuel that was
sophisticated description of soot formation; they burned in the computational volume, mA was the
included a separate first order dependence on mass flow rate of air, T was the gas temperature and
unburned molar fuel concentration (kmol m-3), Cf, Ck-e was a turbulence modeling constant that
and oxygen, Co. The general global equation for soot described the rate of turbulent mixing. The oxidation
102 I.M. Kennedy

Table 1. Comparison of predicted and measured SAE smoke numbers from modified gas turbine combustors (from Rizk and
Mongia4s )

Baseline Modification Modification Modification Low smoke


1 2 3 mod

Measured smoke number 38.0 38.0 32.0 28.0 11.0


Predicted smoke number 35.5 30.6 28.9 20.3 11.1

of soot was calculated from calculated simply from the kinetic theory of gases. As
, : rF/A Iv e0 lIT] a result, there was no activation energy for these
reactions; the only temperature dependence that
~b°x(mg/kg) = "~--V [,--T--J p z / m A - ( ~ - ~ / ( 1 8 - H) ''s
I . . I appeared was a 7 °.5 function that arose from the
(9) collision rate. In the first of their papers, they assumed
that the active species concentration was constant. A
where A was a modeling constant, Vwas volume (m a)
later paper 5° included a rate equation to account for
and subscript p z referred to the primary zone.
the pyrolysis of the original feedstock to yield the
The correlations accounted for the impact of
pressure (p2), residence time (through mA), tempera- active species, C. A major goal of the authors was the
determination of the ideal conditions for the deposi-
ture and fuel composition on soot emissions. The
tion of a high-density layer of graphite on the walls of
three-dimensional model was used to investigate
the reactor. A secondary interest was related to the
improvements to the performance of the T56
formation of gas phase carbon which they identified
combustor with various modifications to the liner
as soot when the particle exceeded a critical size, i.e.
and with the addition of a dome swirler. Measure-
about l0 a monomer units or about 3 nm dia. The rates
ments were obtained with a combustor inlet pressure
that were used by Heicklen and co-workers49'50 were
of 1110 kPa, an inlet temperature of 618 K, an air flow
based solely on kinetic theory collision rates; these
rate of 16.2kg sec -1 and an outlet temperature of
rates do not represent the slower chemical reactions
1491 K. The predictions of the soot emissions in terms
of the SAE Smoke number are compared with the that we know to be an essential feature of soot
formation. It is expected, therefore, that their model
measurements in Table 1.
Although the model tended to underpredict the would tend to grossly overpredict the production of
soot emissions, the agreement was surprisingly good. soot in a pyrolytic system.
Smith 51 proposed a model of soot that assumed that
The modeling constants were adjusted to give
the particle inception was entirely due to a physical
agreement with a set of baseline measurements so
nucleation process. Classical nucleation theory was
that it is uncertain, as with all empirical models,
used to calculate the rate of particle nucleation; the
whether the model offers any general validity for
value of properties such as the surface tension of a
other combustors of significantly different design.
carbon cluster were estimated. The size of a soot
particle was assumed to be limited by the depletion of
3.2. S e m i - e m p i r i c a l M o d e l s the growth species. However, evidence exists that soot
formation is not represented accurately, in fact, as a
The next level of soot modeling attempts to
physical condensation process. The particle inception
incorporate some aspects of the physics and chem-
process exhibits a relatively high activation energy
istry of the phenomenon, as opposed to a correlation
which is not consistent with physical condensation.
(albeit with physical insight) of experimental data.
Tesner and co-workers 52's3 presented a model of
This usually leads to the development of rate
soot formation in 1971 that has been widely used.
equations for reactions of soot percursors and soot
Tesner et al. 52 interpreted their measurements of soot
particles with a simple description of the chemistry.
particle formation in an acetylene-hydrogen flame in
The formation of carbon from the pyrolytic
terms of a simple kinetic model. They proposed a two
decomposition of gaseous carbon was modeled by
Hudson and Heicklen. 49 They considered an isother- step mechanism that described the formation rate of
soot nuclei n (radical nuclei as they were called)
mal flow reactor and neglected axial diffusion so as to
avoid the complications of solving elliptical differ- dn
ential equations. The pyrolysis of the feedstock was d~ = no + ( f - g ) n - g o N n (1 1)
assumed to yield an active species that they tentatively where
associated with either C2, C2H or C2H2. The formation n o is a temperature dependent rate of spontaneous
of the active species (referred to generically as C) was generation of nuclei
followed by a series of growth reactions such as f and g are branching and termination coefficients
C + Cn ~ Cn+l (10) go is the rate of loss of nuclei due to collisions with
They assumed that all reactions of C and C~ resulted soot particles
in the larger species, Cn+l. The rate constant was N is the number density of soot particles
Models of soot formation and oxidation 103

and the particle number density N as structures and in the surrounding fluid. The rates of
dN formation of soot nuclei and of soot mass were
d----{= (a - b N ) n (1 2) calculated with the model of Tesner et al. 52'53 Soot
oxidation was assumed to be proportional to the rate
This model has been applied widely over the years in a
range of combustion applications: 2'54'55 of combustion of fuel; no account was taken of OH or
02 concentrations. The complete model contained a
Surovikin 56 developed a more sophisticated model
large number of 'constants'. The turbulence model
that represented multiple steps for the production of
itself required the specification of 10 parameters,
carbon black. Oxidation was included in this
more than the usual requirement in turbulence models
formulation. The steps included: (1) the formation
due to the inclusion of the equations for soot nuclei
of radical nuclei; (2) the growth of the nuclei to a
number density and soot mass. In addition, the soot
critical diameter at which point they become incipient
model required the specification of a further six
particles with a physical surface; (3) the growth of the
parameters, most of which were taken from Tesner
incipient particles into carbon particles. The rate of
et al. 53
formation of radical nuclei was written as
Boundary layer forms of the governing equations
dn = V~ + f n - go nz - K (13) were solved. Two equations were solved for the soot
dt field viz., the mean number density of soot nuclei and
where V0 represented the rate of decomposition of the mean mass concentration of soot. The numerical
hydrocarbon molecules with an activation energy that results were compared with experimental data
corresponded to the energy of a C - C or C - H bond. obtained from light scattering. Reasonable agree-
The coefficient f represented the branching of the ment was demonstrated over a range of Reynolds
nuclei formation reaction. The loss of nuclei due to numbers. However, some caution is required in
the collisions of one with another was represented by interpreting the results as a confirmation of the eddy
the third term in Eq. (11) while the loss of nuclei on dissipation approach. The use of light scattering to
the surface of particles was represented by K. infer soot mass loadings in turbulent flames involved
The rate of change of the diameter of a radical uncertainty. Furthermore, the model contained many
nucleus was parameters. The physical basis for an eddy dissipation
model in a turbulent diffusion flame may also be
dDrn 1 { 6 ~1/2 -z/3dm
-- / - - J m - - (14) questioned.
dt 3 \Trprn,I dt Kyriakides et a l : 7 and Mehta et al. 58 used similar
where approaches to model soot emissions from diesel
Dm was the diameter of a radical nucleus engines. In both cases, they adopted the methodol-
p~ was the density of a radical nucleus ogy of Magnussen et al., 54 along with the soot model
m was the mass of a radical nucleus. of Tesner et al., 52 to calculate the soot formation in
The rate of increase of the mass of the nucleus was the engine cylinder. Kyriakides et al. 57 made use of
calculated from the rate of collisions of hydrocarbon the oxidation model of Lee et al. 17 to calculate soot
molecules with the nucleus and the mass of the burnout. Mehta et al. 58 employed a somewhat simpler
hydrocarbon molecule with an activation energy of model for soot burnout. In this case the local
1.3 x 105 kJ kmol -t. Finally, the particle nuclei (also oxidation rates of nuclei and soot particles were
referred to as seeds) grew as a result of surface related to the local rate of fuel consumption and the
addition of hydrocarbon molecules. The growth rate was mean concentration of nuclei and particles respec-
assumed to be proportional to the particle surface area. tively. This approach neglected the different tempera-
An activation energy of 1.3 x 105 kJ kmo1-1 was used. ture sensitivities of soot oxidation and fuel oxidation
The physical meaning of some terms in the formula- as well as the impact of OH on soot burnout.
tion of Tesner and Surovikin was not altogether clear, The agreement between predicted and measured
Nonetheless, many authors have made use of Tesner's soot emissions for these models was similar to the
model, including Magnussen et al. 54 They applied a results obtained by Khan et al. 42 that were shown in
generalized eddy dissipation model to a sooting, Fig. 2. Kyriakides et a l : 7 noted two important
turbulent acetylene flame. The eddy dissipation model shortcomings of their model. They assumed that all
treated chemical reactions as taking place in fine soot particles were the same size and that the size was
structures in the flow within which the reactants were a constant. Simulations with a particle size of 25 nm
assumed to be perfectly mixed. The fraction of the yielded soot levels that were too low so the particle
flow that was occupied by these structures was size was increased in an ad hoe manner to 50 nm.
determined through scaling arguments. However, Better agreement was achieved. Accurate predictions
not all the fine elements of fluid could react due to of spray mixing also presented a problem for soot
insufficiently high temperatures. Therefore, an addi- modeling in engines. The predictions of soot tended to
tional assumption was required to find the fraction of be rather sensitive to the details of the spray model.
fine structures that reacted. Mean rates of combustion Brown and Heywood 59 employed a stochastic
were determined on this basis. mixing model to account for inhomogeneous mixing
Soot was assumed to form in both the fine in a diesel engine cylinder. The stochastic model
104 I.M. Kennedy

considered the fluid to be made up of segregated measurements that were obtained with a smoke meter
elements that were chosen at random and allowed to over a range of loads and injection timings. The
mix on the molecular level following the approach of results showed the calculated soot concentrations in
Curl 6° and Pratt. 6~ The stochastic mixing model was the main chamber and in the pre-chamber for three
superimposed on a multi-dimensional calculation of different mean effective pressures, i.e. three different
the cylinder flow with the KIVA code. Their goal was loads. It was evident that most of the soot was
to predict NO and soot emissions from a direct formed in the pre-chamber because of the high overall
injection diesel engine. The model was compared with equivalence ratio and temperatures in the pre-chamber.
an engine that had a single cylinder with a 1173 cm -3 The agreement between measured and predicted
displacement volume. exhaust soot was good to within about 20% or
Finite rate soot formation was modeled after the better. Similar agreement was achieved with varia-
results of Wang e t al. 62 that were obtained in a shock tions in the injection timing. Predictions of NO
tube study of soot formation from toluene. Oxidation emissions from the engine were not as accurate as
was predicted with the Nagle-Strickland-Constable the soot predictions. The results demonstrated the
formula. Is The agreement of the model ranged from usefulness of a relatively simple model, at least over a
remarkably good in one case (without EGR and low limited range of conditions.
swirl) to poor in three other cases. In the latter cases, the Jensen 65 presented a model that had a firmer basis
model overpredicted the measured soot concentrations in the physics and chemistry of soot formation as they
by typically an order of magnitude. This was not are currently understood. The description included
altogether surprising in view of the fact that the soot most of the essential phenomena. He assumed that the
sub-model was based upon shock tube results in toluene, formation of soot was controlled by five processes.
conditions rather far from those in a diesel engine. The initial stage involved gas phase reactions that
The models that have been discussed so far were produced radical fragments which constituted nuclei.
applied to the simplest engine configuration, the Nucleation was followed by coagulation, growth and
direct injection diesel, although multi-zone calcula- oxidation.
tions were used in some cases. Indirect injection The initial nucleus was taken to be either C2 or
diesels introduce an additional complication due to C2H. An 11 step, reversible, gas phase mechanism was
the exchange of mass and energy between the proposed to account for the formation of these
prechamber and the main combustion chamber. species. Mass growth of the nuclei occurred as a
Kouremenos et al.63 developed a model for this result of the addition of C2, C2H, C3 or C2H2. The
more complex geometry to predict NO emissions and rate of growth was proportional to the particle
soot emissions from a single cylinder, swirl precham- surface area and the rate of arrival of growth species
ber Ricardo engine with displacement volume of at the surface. No activation energy was used for the
536cm 3. They used a two zone model in both growth reactions. Coagulation was treated by sub-
chambers to describe burned and unburned gases. dividing the soot aerosol into 15 size classes. The
Combustion chemistry was described by equilibrium coagulation rate was calculated in the free molecular
between 11 chemical species. The simple soot model limit. Oxidation was not considered in view of the fuel
of Hiroyasu et al.64was employed in their calculation. rich conditions that were of interest in rocket motors.
The net rate of soot mass formation was found as the The model was applied to a rocket chamber that
difference between the rates of formation and was fueled on isopropyl nitrate. Measurements were
oxidation viz., obtained of soot concentrations and gas composition
at the rocket nozzle exit plane. The observations
dins (dms~ _(dms'~ (15)
could be matched only if either (1) C2 were the nucleus
\ dt /formation \ dt /oxidation and C2H2 were the growth species or (2) C2H were the
where Arrhenius expressions were used for the rates nucleus and C2H2 were the growth species. Other
of soot formation and oxidation combinations of nucleating species and growth
species yielded soot concentrations that were much
( - ~ t s ) formation= AfmfvP°5 exp ( ~ T f) (16) too low compared to the measurements. He con-
cluded that for his conditions the controlling process
in soot formation was the rate of the gas phase
(-~)oxidation= Aomovpl'Sexp(~ff-~) (17) reactions, especially the rate of the initiation step

The activation energies were Ef = 8 x 104 kJ kmo1-1 CH 4 + M -~ CH 3 + H + M (18)


and E0 = 12 x 104 kJ kmol - l . It was a little surprising The rates of the growth reactions and the rate of
that they used an activation energy for oxidation that coagulation were found to be of secondary impor-
was greater than for formation, an approach that runs tance. However, caution should be used in drawing
counter to other observations. The values of the general conclusions from these results due to the
model parameters were chosen to match the results at rather specialized application.
one experimental condition. Graham 66 attempted to model the growth of soot
Predictions of exhaust soot were compared with that was formed by the pyrolysis of aromatic
Models of soot formation and oxidation 105

hydrocarbons in incident shock waves. 67 He assumed part soot and in part liquid polynuclear aromatic
that collisions of soot particles were coalescent, i.e. compounds. This conclusion was interesting in light
they were perfectly sticky and formed spheres. of the more recent observations of Dobbins et al.;68
Furthermore, he assumed that a self-preserving size they asserted that soot precursors were liquid droplets
distribution13 was maintained. Two fundamental that eventually underwent a graphitization process
models for soot growth were considered. The simpler to form mature soot particles. Graham further
model assumed that nucleation of new particles and concluded that this mechanism could account for
condensational growth onto existing particles gave the observation that the primary particles in soot
rise to zero net change in the mean volume of the soot agglomerates were of similar sizes. He believed that
particle size distribution, i.e. they exactly balanced the variation in the rate of graphitization with
each other with respect to mean particle size. The temperature was important in determining the final
somewhat more general model assumed nucleation properties of the soot aerosol. Ultimately, his
ceased at the beginning of shock heating and was modeling was not essentially predictive. Its purpose
negligible for all time after that. was to deduce mechanisms of soot formation from
The two models were expressed in terms of the rate light scattering measurements in shock tubes.
of change of the mean volume (I7") of the soot aerosol The mathematical formulation of soot formation
particles. The assumption of a self-preserving size models was continued by Mulholland.69 He proposed
distribution led to the formulation of the first model a simple scheme in which soot nuclei were generated
(coagulation only) as by the pyrolysis of a fuel. The soot nuclei were
allowed to add to the fuel to form larger nuclei. Rate
dF k fv~,l/6 (19) equations were written in an Arrhenius form for the
dt
production of the soot nuclei. The concentration of
where the rate constant, k, was expressed as
soot particles was assumed to be given by the sum of
1 ( 3 x 1/6 (6XT'~ ~/e the concentrations of all nuclei. The surface growth
k = ~ \~:) \~--/ o,C (20)
rate was assumed to be proportional to the particle
where ~ was Boltzmann's constant. The collision surface area. The growth species was the fuel itself so
integral was a and the enhancement of the collision the formulation was strictly correct only if the fuel
rate by long range dispersion forces 1132 was repre- were acetylene. For other fuels, additional pyrolysis
sented by G in Eq. (20). The coUisional enhancement steps that led to acetylene would be necessary. The
takes on values typically around 2. Equation (19) could time dependence of various parameters such as
be integrated to yield an expression for the mean soot moments of the soot particle size distribution, the
particle volume as a function of time, t number density, etc. were compared with experiments
under regimes in which the free radical growth of
"]6/5
g(t) = (go)5/6q-~kIi=tofv(t)dtj (21) nuclei or the surface growth or the coagulation
processes were switched off. It was found that a
where V0 was the initial particle volume. satisfactory time dependence could not be obtained
The second model included the effect of nucleation with any combination of processes for all the
and condensation and led to a formulation for the experimental conditions that were considered. This
rate of change of mean soot particle volume conclusion was not altogether surprising in view of
the simplified nature of the model.
d(r(t)-kfv(F')l/6+O~)dfVdt-~ (22) Some models of soot formation involve very little
mathematical formulation. Models such as that
which could not be integrated as simply as Eq. (19) proposed by Jones and Rosenfeld7° rely upon a
and had to be treated numerically. The scattered light purely physical hypothesis about soot formation and
intensities that would be expected from each of oxidation. Jones and Rosenfeld7° used Burke-
these models were compared with the results from Schumann theory 71 to describe the gross features of
the shock tube experiments by obtaining best fits to a laminar co-flow diffusion flame. They postulated
the data through varying parameters such as the soot that the opening of the stoichiometric surface (the
density, the particle refractive index and the colli- flame front) indicated the onset of sooting. Theore-
sional enhancement factor, G. Good fits to the tically, this would occur at an infinite flame height for
experiments were obtained with both models. Graham reactants that were supplied in stoichiometric propor-
concluded that his results showed that particle tions. The stoichiometry could be varied in practice
nucleation did not continue for a substantial time through variations in the fuel and oxidizer partial
following the shock. In other words, he concluded pressures in each stream of gas. The 02 partial
that nucleation was extremely rapid and substantial, pressure that lead to a theoretically infinite flame
leading him to the conclusion that soot particle height was used as a measure of the conditions at
nucleation was physical rather than chemical. which the flame opened. This partial pressure was
Graham also concluded that the collisions of soot compared with experimental values.
particles were sticky and perfectly coalescent. From The partial pressure of 02 in the outer flow was
this observation, he concluded that the particles were varied in a series of experiments. The maximum 02
106 I.M. Kennedy

partial pressure for which soot could be observed for mixture fractions ~ > ~* where ~* was the smallest
above the flame was taken to be a measure of the soot mixture fraction at which soot formed. The model
point for a number of fuels. The results showed a relied on the assumption that soot mass growth was
rather remarkable agreement between the predicted controlled solely by the availability of a gaseous
and the measured 02 partial pressures for a range of species such as acetylene. Growth in this view is
different fuels. They argued that the results indicated independent of the aerosol properties, including
the importance of the diffusion flame structure in surface area and number density of particles. There
determining sooting behavior. They relegated soot has been no confirmation via more detailed computa-
kinetics to a secondary role. It is not clear why the tions that this is an accurate approximation. Indeed,
theoretical conditions for an infinite flame height Lindstedt s3 found that a growth rate that was a
should relate so well to the experimentally observed function of the particle number density performed
soot point, particularly because the kinetics of soot very well. In that case, particle nucleation may be an
formation and oxidation are slow. The assumption of important feature of the flow, at least in moderately
diffusion control is not appropriate in this case. The loaded flames such as the counterflow flames that
amount of soot in a diffusion flame depends on the Lindstedt modeled.
time available to form soot, i.e. on the flame height. The maximum soot volume fraction was assumed
Therefore, one might expect some greater dependence to occur at the flame tip where ~ = ~st • The flow field
on fuel type. The results were used by the authors to was handled by noting that the centerline velocity, u¢,
point out the importance of controlling flame as a function of distance, x, in a laminar diffusion
characteristics in attempting to isolate chemical flame was controlled by buoyancy forces so that
effects in soot production. The simple model had no uo = ~ u v ~ (25)
real predictive capability.
By ignoring soot oxidation and by assuming that
The description of diffusion flames is facilitated by
temperature was a function of mixture fraction, Eq.
the use of the concept of a mixture fraction. Many
(23) could be integrated to yield a maximum soot
models of soot formation in diffusion flames have
volume f r a c t i o n f max of
made use of the mixture fraction to describe the gas
phase composition and temperature of the fvmax =Lf 1 (26)
flame. 2'35'72-79 The mixture fraction is an example of O~uV~f T
a Shvab-Zeldovich coupling function.7tAn essential where r was the time for soot formation that was
aspect of the Shvab-Zeldovich formulation is that all defined as
species and energy diffuse at the same rate. An
alternative way to express this requirement is that the
Lewis numbers for all species should be unity. The
r-l= Ii,,f
~,T)f~st'I/2p~xll)l----~st)
f"
-E
~--~st' exp(~)d~

mixture fraction provides a very useful method to (27)


describe the local thermochemical state of a diffusion and the flame length was Lf. Delichatsios went on to
flame because it is a so-called 'conserved' scalar, i.e. use the model to derive a scaling for the soot mass
the equation for its conservation contains no source flow rate in the Wolfhard-Parker flames of Kent and
term. Hence, the non-linear reaction rate terms in Wagner. s4 The model collapsed the data in different
combustion are eliminated. In order to relate the flames to a reasonable extent over two orders of soot
mixture fraction to the species mole fractions and mass flow rate but the functional relationship was not
temperature, some assumption is required with regard simple, particularly at large distances from the
to the thermochemistry. Chemical equilibrium is burner.
often used to obtain this relationship. An alternative Faeth and co-workers s° attempted to use the
method is to obtain the state relationships from mixture fraction concept to obtain a state relation-
measurements s° or calculations in strained laminar ship with soot volume fraction. In other words, soot
diffusion flames, sl A library of laminar flamelets is volume fraction was correlated with mixture fraction.
then developed over a range of strain rates. A calculation of a turbulent diffusion flame then
Delichatsioss2 formulated a theoretical model that involved the calculation of the mixture fraction and
developed a scaling of soot formation with flame the velocity field in the usual manner, s5
height and with pressure. He focused on the evolution Unfortunately, soot does not diffuse at the same
of soot along the axis of a laminar diffusion flame. rate as the molecular species. Furthermore, soot is the
The soot volume mass fraction was written as product of a relatively slow reaction and is not in
drs'X equilibrium. Hence, it cannot be expected on
pu--d-x-x) = if' (23) fundamental grounds that soot will be well corre-
lated with mixture fraction. For example, near the
where the reaction rate ffs was a function of the
base of the flame there is little soot although mixture
mixture fraction, ~ , temperature and the fuel mass
fractions cover the range from 0 in the free stream to
fraction at the nozzle exit
close to 1 on the axis of the flame. Significantly larger
Ws = CP2 YFO exp(-E/RT) (24) amounts of soot are found over a similar range of
mixture fractions further along the flame. The use of
Models of soot formation and oxidation 107

the mixture fraction state relationship is attractive in volume arose from surface growth. The surface growth
computational terms but it neglects the impact of rate was taken to be a function of mixture fraction and
chemistry and transport on the distribution of soot temperature. An activation temperature of 10,000K
throughout a flame. was assumed on the basis of the measurements of
A fundamentally more satisfactory approach uses Tesner et al. 53 The functional relationship between the
rate equations to account for the chemistry, albeit growth rate per unit surface area and the mixture
often in a very crude fashion. An entire class of fraction was obtained from measurements of soot
models are characterized by their attempt to solve a growth in a counterflow ethylene/air diffusion flame. As
small number of differential equations for the a result, the approach was specific to a given fuel,
conservation of quantities such as soot particle oxidizer and pressure at the conditions of the
number density and soot volume fraction. In their measurements. The modeling assumed that the surface
simplest manifestation, the models may include only growth rate depended on the soot aerosol surface area.
an account of the formation of soot nuclei. Kesten The model was initially applied to a laminar
et al. 86 used the model of Tesner et al. 52 to describe the ethylene diffusion flame in a co-flow configuration.
formation of soot around a fuel droplet. The only Predictions of soot volume fractions were obtained at
reference to the soot process in this simple model was a number of axial stations. In general, good agreement
to the production of' soot radical nuclei. Models was obtained within about a factor of two of the peak
developed by other researchers include rates for soot measured soot volume fractions. The authors did not
inception, surface growth, coagulation and soot attempt to model the burnout zone of the flame and
oxidation. Their common philosophy revolves hence did not subject the model to a full test.
around the prescription of a minimum set of rate The method of Kennedy et al. 2 was specific to
equations that can describe important soot phenom- ethylene fuel. Said et al. 88 attempted to remove this
ena. The ultimate goal is generally a model that can be restriction by postulating the existence of a hypothe-
used in codes for turbulent combustion in an tical species, intermediate between the fuel and soot.
industrial setting. Hence, the use of a full set of In reality, this imaginary species could be associated
detailed elementary reactions is impractical. The with acetylene although Said et al. did not need to
differences amongst the models in this category are make an explicit assignment of the intermediate to an
relatively minor. actual species. They followed the approach of
Kennedy et al. 2 attempted to use only one equation Kennedy et al. 2 in ignoring the dynamics of the soot
to describe soot formation and oxidation in a laminar aerosol, i.e. they did not solve an equation for the
diffusion flame as a forerunner to a model for particle number density. However, two rate equations
turbulent diffusion flames. The rationale called for were formulated in the model. The first equation
the minimum possible set of equations with a view to described the rate of formation and oxidation of the
ultimately applying the model in a PDF (probability intermediate species, denoted by '/'. The mass
density function) calculation.75 Measurements in fraction of I was found from
many flames have shown that the number density of ..... {- TAI"~
particles decays very quickly as a result of rapid ddtYl _ kA ifueI Io x exp ~ - - - )
coagulation near the particle inception zone. As a
result, Kennedy et al. 2 ignored the equation for -k.Y,Y~oxeXp(-~)-k,(T)r, (28)
particle number density in favor of an average
number density and hence they did not account for in which Yruel, Yox and Y1 were the mass fractions of
possible variations in nucleation rate with changing fuel, oxidizer and intermediate species respectively.
flame conditions. They used only an equation for The other parameters in the equation were deter-
conservation of the soot volume fraction throughout mined empirically. The production rate of soot was
the flame. As was noted earlier in this review, this found from the competition between the conversion
approach is most appropriate for flames that exhibit a of I into soot and the oxidation of soot as shown in
large soot loading. It is less likely to be satisfactory in the equation for soot mass fraction
flames that do not exhibit large soot volume fractions.
The equation forfv contained two source and two dt ~ YoxY s P T - V 2 exp (29)
sink terms. Soot volume fraction increased as a result
of particle nucleation. The size of the smallest particle The oxidation rate of soot in Eq. (29) was derived
was assumed to be I nm; the source of soot volume from the model of Lee et al. 17A constant value of soot
was equal to the product of the particle inception rate diameter, ds,was taken to be I nm in Eq. (29). This
with the volume of the individual nuclei. The particle value may be appropriate to freshly formed soot but it
inception rate was taken to be a function of mixture is not likely to be representative of soot at the tip of a
fraction that located the inception zone just to the fuel diffusion flame where much larger particles are found.
rich side of the flame. The inception rate was assigned a The formation rate was found from experiments. A
narrow Gaussian distribution in mixture fraction space value for ks was determined from examining experi-
with a peak rate that was derived from the measure- mental resultss9 in a pre-mixed laminar flame. The
ments of Wersborg et al. 87 T h e second source of soot post flame temperature was assumed to be constant
108 I.M. Kennedy

and the concentration of oxidizers was assumed to concentrations are spatially homogeneous. If surface
be zero. Equations (28) and (29) were simplified growth is proportional to surface area, then a t 3
considerably with these assumptions so that a simple dependence is expected for the same conditions.
expression for ~ t was obtained. Hence, a comparison However, temperature and concentration inhomo-
with the data of Kent et aL allowed ks to be found. geneities in a diffusion flame may overwhelm the
Values for TA1 and TA2 were obtained from the work impact of any assumptions regarding the functional
of Leung et al. 76 while ct and 3 were assigned form of surface growth on the temporal development
arbitrary, but not unreasonable, values of 0 and 1, of soot volume fraction.
respectively. The values of rate constants, kA and ka, The source terms (c~, 3, 7, 6) for the number density
were determined by adjustment in comparison with n and the volume fractionfv were written as functions
the experiments of Santoro et al. 9° through which of the mixture fraction ~ and the temperature T as
good agreement between the model and the measure- c~ = Cc,p2T1/2Xc e x p ( - T,~/T) = ~/144 (32)
ments was achieved. Oxygen mass fraction in Eq. (29)
was governed by an empirical relationship between /~ = C 3 T I/2 (33)
mixture fraction and oxygen mass fraction. 9~
The soot model was applied to the prediction of the
7 = C'rpT1/2Xc e x p ( - T T / T ) (34)
turbulent diffusion flame that was measured by Kent
and Honnery. 74 The agreement was very good,with where the Cis were constants that were determined,
the peak in the mean soot volume fractions being along with the activation temperatures T;, from
predicted accurately. The model included many measurements in laminar flames. The source terms
parameters, some of which were set abitrarily while were also functions of the fuel mole fraction, Xc.
others were adjusted by comparison with laminar Temperatures in real combustors, especially gas
flame experiments. The number of parameters was turbines, often exceed the peak temperatures that can
significantly greater than that used by Kennedy et al. 2 be achieved in laboratory flames burning hydrocar-
so that somewhat better agreement with measure- bons in air. Moss e t al. 77 applied the model to a
ments should be expected. Recalibration of the model laminar ethylene diffusion flame burning on a
would be necessary if the fuel or pressure were Wolfhard-Parker burner. The oxidizer was 52% 02
changed so it not clear that the inclusion of the psuedo and 48% Ar. Peak flame temperatures were around
intermediate, /, adds greatly to the general applic- 2600K. When the parameters that were obtained in
ability of the model in comparison to simpler semi the original study 92 were used for this case, unrealistic
empirical formulations. predictions of the mean particle size and number
Moss et al. 92 proposed a flamelet approach to density were obtained. Consequently, adjustments
modeling soot formation in diffusion flames. The were necessary to the original parameters (Ca, C 3 and
usual flow field equations were solved. In addition, C~) in order to achieve satisfactory agreement with
balance equations were solved for the number density the measured peak soot particle size and number
of soot particles, n, and for the soot volume fraction, density. The activation energies for soot nucleation
fv, in the following form and surface growth were unchanged. Oxidation by 02
and OH was included in this model. Radiation was
d / n\
(30) not included on the grounds that the flame was only
ai ~ o ) = a ( , ) - ~(,)(n/No) 2
moderately sooting.
The two-dimensional flow field was calculated with
dfv the thermochemistry derived from a flamelet
ps--d7 = 7(~)n + 6(~) (31)
approach. The flamelets were calculated with a one-
The density of soot, Ps, was taken to be 1800 kg m-a; dimensional code for a laminar diffusion flame.
No was Avagadro's number. The first term in Eq. (30) Generally good agreement with measurements was
represented the increase in soot particle number due demonstrated for soot volume fractions and soot
to particle inception. The second term accounted for particle diameters along the flame centerline. How-
the loss of particles as a result of coagulation. The first ever, this is not altogether unexpected, given the fact
term in Eq. (31) represented an increase in soot that the model parameters were adjusted to yield the
volume as a result of surface growth while the second appropriate peak values. The correct distribution of
term accounted for the increase in soot volume due to soot along the axis suggests that the essential physics
nucleation of new particles. It should be noted that have been included in this model although the
the first growth term in Eq. (31) is proportional to the requirement of adjusting the model parameters
number density, not a surface area. The number indicates that semi-empirical models such as these
density of an aerosol rapidly attains a nearly constant are not universal. Care must be taken when the model
value as a result of a balance between particle is applied to conditions that do not precisely match
inception and coagulation. Hence, if the number the conditions under which it was calibrated.
density, n, is approximately constant, Eq. (31) Syed e t al. 7s applied the flamelet modeling strategy
predicts that the soot volume fraction will increase of Moss e t al. 92 to a turbulent methane diffusion
linearly with time if the temperature and the flame. Their particular interest was in predictions of
Models of soot formation and oxidation 109

the thermal radiation from large buoyant flames. 6 30


Experiments were performed in a laminar methane
diffusion flame to verify the soot modeling. They ~ R 25 %
found that the surface growth model that Moss et
x
al. 92 used was incompatible with their observations in x
2O g
the methane flame. Syed et aL 7s implemented a
surface growth model that contained a linear
15 ~
dependence on aerosol surface area
d
Ps~tfv = "~(~)(Psfv)2/3n 1/3 (35)
0O
where -~was a modified value of the numerical constant
C7 and the terms withfv and n accounted for the soot
surface area. When the number density n approached o~ . . . . . . .~. . . . . . . . . . . ; ---~0
a constant value, this equation led to a predicted t3
0 1 2 3 4 5
Distance from centerline (mm)
variation in soot volume fraction as Kennedy et al. 2
pointed out. The comparison of measurements with Fig. 4. Soot volume fraction profiles in laminar Wolfhard-
the model predictions are shown in Fig. 4. Parker CH4-air flame at heights above burner of (a) 15 mm
(- - -) predictions, (m) measurements, (b) 30mm ( )
The agreement in the laminar diffusion flame was predictions, (&) measurements. From Syed et al. TM
generally quite satisfactory although the model
tended to underpredict the soot volume fraction on
the centerline of the flame. A similar problem existed heat loss; the latter issue was resolved in the manner
in the model of Kennedy et al. 2 Constraining particle described by Young et al. 96 The mixture fraction was
inception to small values of the mixture fraction decomposed into two components, one that was
worked reasonably well low in a diffusion flame but related to the gas phase only and the other that was
the correlation of inception with mixture fraction was related to the concentration of soot. The complexity
apparently less satisfactory higher in the flames. Saito of the flow field precluded the use of a more realistic
et al. 93 observed that benzene built up along the axis approach.
of a laminar diffusion flame. This evidence suggested The results showed a dramatic increase in soot
that the assumptions behind the use of a Shvab- volume fraction as pressure increased. The pressure
Zeldovich coupling function,71 such as the mixture effect was predominantly due to an increased rate of
fraction, were not appropriate in these circumstances, particle nucleation. However, the conditions of the
certainly as far as the pre-particle chemistry was simulation were well outside the range of calibration
concerned. of the soot model. The authors found that up to 75 %
Syed et al. 7s applied the soot model to a turbulent of the locally available carbon was converted to soot.
buoyant flame. A parabolic, k - e turbulence model In this case, it was important to account for the
was used along with laminar flamelet chemistry. The modification of the local mixture fraction due to the
temperature-mixture fraction relationship was per- carbon bound up as soot. The use of the modified
turbed to account for the impact of radiation. Mean mixture fraction restricted the production of soot to
predicted temperatures agreed quite well with the vicinity of the injector, inhibiting its production in
measurements that were obtained with a compen- regions close to the primary zone jets.
sated thermocouple. Less satisfactory agreement was The effect of pressure was explored by Young
97
et aL in a series of experiments that were designed to
obtained for narrow band radiation. An accurate
treatment of the oxidation of soot was believed to be yield property maps of the major variables viz.,
an important reason for the discrepancy. temperature, mixture fraction and soot volume
Alizadeh and MOSS94 evaluated the same model for fraction. Microprobe sampling was used to measure
soot in a comparison of measurements and computa- the mean mixture fraction in turbulent kerosene
tions in an idealized combustor of the same design as flames at pressures up to 6.4 bar. The mean soot
Heitor and Whitelaw. 95 State relationships were volume fraction was found by laser extinction and
obtained from both equilibrium and from a laminar temperatures were measured with a fine wire thermo-
flamelet approach. They also predicted NOx by post- couple. Isolating the effects of pressure in a turbulent
processing, i.e. by calculating the NO concentrations diffusion flame is difficult because the history of a fluid
after the flow field had been calculated. A similar element changes drastically as variables such as the
approach was used with the soot but the authors flame length change. Young et al. reduced their data
acknowledged that de-coupling the soot from the flow to a standardized form by normalizing distance from
field calculation was not likely to be successful due to the nozzle with the location of the maximum soot
two effects: (I) the impact of soot radiation on the volume fraction and by correcting the measured soot
temperature and density; (2) soot acting as a sink for volume fractions to a standard pressure. This analysis
carbon in the flame. suggested an approximately linear dependence of soot
They dealt with the first effect by ignoring radiative volume fraction and soot formation rate on pressure.
JPEC$ 23:2-B
110 I.M. Kennedy

With measurements of the mean mixture fractions, this approach to calculate soot formation and
the results provided libraries of properties for use in a radiation in a turbulent ethylene diffusion flame.
flamelet approach to modeling soot formation and They were able to predict correlations between
radiation in high pressure flames. The maps of soot various important quantities such as the temperature
volume fraction as a function of mixture fraction over and the soot volume fraction.
a range of pressures could be used in modeling soot Young and M o s s 9s used a detailed calculation of
formation in gas turbine combustors. an adiabatic flame to obtain enthalpy and tempera-
Young and M o s s 98 have used an extended flamelet ture as functions of mixture fraction. They adopted a
approach to predict soot and radiation in turbulent somewhat cumbersome adhoc method to account for
diffusion flames. Their goal was to fully couple the the effect of radiative heat loss on the temperature/
soot formation and radiation processes to the flame mixture fraction relationship. They reduced the
calculation without a post-processing strategy. They enthalpy at a given mixture fraction by some arbitrary
adopted a conventional k - e turbulence model 85 to amount. The temperature was then obtained from the
which they added Favre averaged equations for the enthalpy. The application of the flamelet library in the
soot particle number density and the soot volume turbulent case involved computing a mean mixture
fraction enthalpy from a balance equation for energy with a
radiation sink term that accounted for soot luminosity.
0-~xj(P~J~") - 0-~xj\~-~¢~x-jx:]= ~ - fi2~d (36) The flamelet family was scanned for an enthalpy that
was close to the computed mean enthalpy and the
temperature of the mixture was obtained for that
0 _A(.t0cs value of the heat loss. It is not clear how this method
is implemented without an expensive iteration at each
~ + N o P ~ , - N 10 / 3P'~ ~1/3~/3 (37) grid point since the soot formation is temperature
= ox%n '~s
dependent and, in turn, the radiation heat loss that
where the tilde denoted a Favre average and where determines the enthalpy depends on the soot volume
~n = ~ and ffs = ~fv- The particle number density fraction. The authors applied their methodology to a
y~vo Y ,
was represented by n. No was Avagadro s Number, p~ turbulent ethylene diffusion flame. Trends in tem-
was the density of a soot particle, fv was the soot perature, mixture fraction and mean soot volume
volume fraction. The parameters a and 6 represented fraction were reproduced reasonably well although
the impact of particle nucleation on the number the quantitative agreement left room for improvement.
density n and on the soot volume fraction, fv. The Lindstedt and co-workers have developed soot models
processes of coagulation, surface growth and oxida- for laminar and turbulent diffusion flames. 72'73'76'83'99
tion were represented by the parameters ~, 7 and Wox Lindstedt76'99used a simple kinetic mechanism to predict
respectively. The explicit evaluation of a,/3 and % as the soot volume fractions in co-flow and counterflow
described by Moss et al., 92 was presented in Eqs (32), laminar diffusion flames. He assumed that acetylene
(33) and (34) above. was primarily responsible for the nucleation and the
As in most other models of turbulent combustion, growth of soot particles. The nucleation step was
the mean source terms presented the greatest closure treated simply as a one step process
difficulty as a result of their non-linear dependence on
C2H2 ~ 2Csoot + H2 (39)
temperature. Young and M o s s 98 attacked this
problem with their flamelet approach whereby the The reaction ignored the fact that soot was not
density, temperature and soot precursor/carbon composed purely of carbon but it is adequate for a
concentration were expressed as functions of a semi-empirical model. The rate of nucleation was
mixture fraction via extended flamelet relationships. assumed to be first order in acetylene concentration so
It was possible to find the mean source terms such as that the rate was calculated as
or ~'ox by averaging over a probability density I~n = 104e-21'l°°/T[c2H2] (40)
function for the mixture fraction so that the mean
source term was where [C2H2] was the concentration of acetylene and
the units were in kmol/ma/sec. The rate of surface
growth was also assumed to be first order in acetylene
concentration and dependent upon a function of the
Unfortunately, expressing the scalar quantities in aerosol surface area. Lindstedt did not assume that
terms of mixture fraction alone precluded the the growth rate was simply directly proportional to
evaluation of possible correlations. For example, in the aerosol surface area as some other authors
the case of surface growth, the term ~-~ was modeled have. 2'72"1°° Bockhorn et all °l showed that soot
as "~N0~ which accounted for the non-linear growth was not related only to surface area and
temperature dependence but neglected the effect of a could, in fact, be independent of area. Their results
correlation between 7 and n, i.e. 7~n'. Effects such as these could be interpreted in terms of a mass growth that
could be captured correctly by methods that solve for was governed by the presence of active sites. Lindstedt
joint probability density functions. Kollmarm et al. 75 used made the ad hoc assumption that the number of active
Models of soot formation and oxidation 111

sites was proportional to the square root of the total acetylene and hence of soot. Unfortunately, the
aerosol surface area. Hence, he wrote the soot growth experimental data included no information on gas
rate as phase species concentrations so that comparisons
were made solely with the soot volume fraction and
Wg = k(T)[f2n2][Cs]l/3[pg] I/6 (41)
number density. The agreement with the measured
where soot volume fractions over the range of reported
oxygen concentrations was quite good. Less satisfac-
k ( T ) = 6 x 10317r(6Mc/Trps)E/311/Ee-I2'l°°/r (42)
tory agreement was obtained with the particle number
Mc was the molecular weight of carbon, Ps was the density. In general, this is a difficult quantity for a
density of soot, N was the number of soot particles per model to reproduce with fidelity. Particle dynamics in
unit mass of soot/gas mixture and [Cs] was given by flames are complicated by variations in coagulation
p Ycs/Mc in which Ycs is the soot mass fraction in rates as a result of varying van der Waals forces, t2
the flow and Mc is the molar weight (12.011 kg sticking coefficients that may be less than one, and
kmol-l). The units were again in kmol/m/sec. significant uncertainties in the interpretation of light
Soot oxidation was modeled with the method of scattering data on the basis of Rayleigh scattering
Lee e t al. 17 Leung e t al. 76 argued that oxidation was from a monodisperse aerosol.
not a particularly important factor in soot formation Lindstodt 99 also applied his model to the prediction
in a counterflow flame. However, while 02 may not be of soot formation in the co-flowing, axisymmetric
important in the case of a counterflow flame, it is methane air flame of Garo et al. 1°2 The soot model
rather more likely that OH that diffuses back from the was unchanged. Again, generally good agreement was
reaction zone may be a significant oxidant in even the obtained for the distribution of the soot volume
counterflow geometry. The oxidation rate in the fraction. Lindstedt ascribed the discrepancies to the
model of Lindstedt 99 was adjusted by a factor of 14 problem of specifying the initial profile of soot
to give adequate agreement with the measurements of volume fraction near the nozzle where there were no
Garo et al. 102 This ad hoc adjustment is typical of the measurements. It is not clear why the model could not
semi-empirical models of soot formation in which it is be used from a location close to the nozzle where soot
usually found to be necessary to adjust one or two nucleation was not significant.
rates to achieve a reasonable agreement with data. In Fairweather e t al. 73 applied the model to the
this case, the adjustment was possibly necessitated by prediction of soot formation in a turbulent propane
the neglect of OH as an oxidant of soot. The rate of flame that issued into a co-flow of air. The approach
the soot oxidation reaction to the incorporation of the soot predictions into a
turbulence calculation was quite similar to the method of
1
Cs + ~O2 ~ CO (43) Syed et al. 7s A conventional k - e turbulence model was
used with additional equations for the soot mass fraction
was written as and the soot particle number density. The modeling by
71" [6Mc]2/3 Fairweather e t a / . 72'73 differed from the approach of
Wox = k( T) --~c
I 7rpc J [C02][Csl2/3[pN] 1/3 (44) Moss and co-workers7s'92 mostly through the use of
acetylene as the precursor species for soot nucleation
where [Co2] is the concentration of Oz and k(T) is and growth. Hence, the instantaneous rates of soot
given, in units of [kmol/m/sec] by nucleation and growth were proportional to the
k ( T ) = 1.25 x 105Tl/2e-19'68°/r (45) acetylene concentration rather than to the parent
fuel concentration. This concept is physically more
Particle number density was calculated assuming that plausible although it does complicate the modeling to
the aerosol was monodisperse. The equation for some extent by introducing the need to determine the
number density included the effects of particle acetylene mass fraction. This is not a significant
nucleation and particle coagulation. burden if the laminar flamelet method is used.
The gas phase reactions were treated in a fairly However, it is important to note that the modeling
simple manner, although with greater detail than the of Fairweather e t al. 72'73 and Syed e t al. 78 did not
simple equilibrium approach that was used by account for the action of OH as an oxidant of soot.
Kennedy et al. 2 An extended version of the global This is potentially an important omission.
mechanism of Jones and Lindstedt io3 was used with A counterflow flame code was used to generate
additional steps to model the formation and destruc- flamelet information for the acetylene concentration,
tion of acetylene. temperature, density and other scalars as functions of
The counterflow version of the code was used to the mixture fraction. The code was run at a low strain
predict the results of Vandsburger et al. 1°4 The rate (15sec -1) which yielded data for an effectively
experiments were performed at a fixed strain rate unstrained flame. The mixture fraction-scalar rela-
but with a variable composition of the oxidizer tionships did not change much if the strain rate were
stream. The 02 mole fractions ranged from 18 to lowered. The counterflow flame calculations were
28% in N2. The variation in Oz concentrations should conducted under adiabatic conditions. Fairweather
provide a good test of the accuracy of prediction of e t al. 73 found that it was necessary to account for
112 I.M. Kennedy

0.8~
radiation in a rather a d hoc manner by adjusting the 1m
flamelet temperature by some arbitrary amount to
match the measured peak temperatures. They found
that the required temperature decrease did not have a
significant impact on the calculation of the acetylene
mass fractions when the laminar counterflow calcula-
tions were repeated with a prescribed radiant energy
loss.
Turbulent flame predictions were compared with
the measurements of Nishida and Mukohara.105 The
Jl
predictions of the acetylene and 02 mole fractions 0.
were in good agreement with the measurements. Such
agreement is necessary, of course, if good agreement
is to be obtained with predictions of soot volume
fractions because soot formation and oxidation are 0 10 20 30 40
directly related to these species. Figure 5 shows the r (mm)
quality of the predictions of the radial distributions of Fig. 5. Radial measurements of (11)CEH2, (@) 02 and
C2H2 and 02 in a flame with the incoming air predictions (--) at x / D = 150 in a turbulent propane flame
preheated to 323 K. Figure 6 indicates the quality of (from Fairweather et al.73).
the predictions of the mean soot volume fraction
along the centerline of the same flame. 30
It can be seen that the gas phase species and the
soot volume fractions are predicted quite well.
Fairweather et al. 73 found that the soot oxidation
rate that was predicted with the expression of Lee
et al. 17 was apparently excessive. They repeated the
calculations with the oxidation rate reduced by a
i £
factor of two (the dashed line in Fig. 6). The
agreement between the predicted and the measured J
soot volume fractions near the end of the flame was
improved significantly. It might be expected, however,
that the inclusion of the action of OH as a soot
t
oxidant would cause the agreement to worsen.
Nucleation in this model was related entirely to
acetylene as an indicator species. Lindstedt s3 0 100 200 300 400 500
extended the model to improve the calculation of x/O
soot nucleation and mass growth. In particular, he
Fig. 6. Soot concentrations along the centerline of a
added a kinetic mechanism for the production of C3 turbulent propane diffusion flame: (11) measurements, (--)
and C4 species that were able to produce benzene normal oxidation rate, (- - -) oxidation rate/2 (from
through reactions such as Fairweather et al.73).
C3H 3 + C3H 2 ---, C6H 5 (46)
reactions. The first model (I) followed the approach
nC4H3 + C2H2 --~ C6H 5 (47) set out by Frenklach and Wang 1°7 in which surface
chemistry was attributed to hydrogen abstraction/
Soot nucleation was then assumed to result from not
acetylene addition reactions. A certain number of
only a first order acetylene reaction but also a benzene
active sites per unit area of aerosol was assumed.
reaction
Growth then depended on the surface area of the soot
C6H 6 --~ 6Cs + 3H2 (48) aerosol and a steady state approximation to the rate of
with a reaction rate given by the hydrogen abstraction/acetylene addition reactions.
His second growth model (II) removed the
(Vn = k ( T)[Cc6H6 ] (49)
hydrogen abstraction mechanism so that growth
where was proportional to the acetylene concentration and
k ( T ) = 7.5 x 104e-21'°°°/rin units of the soot surface area. This model for surface growth
has found wide application in semi-empirical models
[kmol/m/sec/K]
of soot formation. 2'14'72'92'1°° The third model (III)
Lindstedt 83 also paid some attention to the problem assumed that the growth rate of soot was propor-
of modeling soot mass growth, where growth has tional to the number density of soot particles but
been denoted as 'mass' growth in view of evidence independent of the surface area. This model is
that it may, in fact, be independent of surface area. 1°6 consistent with the observations of Wieschnowsky
Lindstedt considered four models for the soot growth et al. 16 in premixed flames. The fourth and crudest
Models of soot formation and oxidation 113

Table 2. Peak soot volume fractions predicted by the following soot growth models: I. growth proportional to surface area
plus the HACA mechanism; II. growth proportional" to surface area; III. growth proportional to number densitY,104independent
of surface area; IV. number density constant, compared to the experimental results of Vandsburger et al. for counterflow
ethylene diffusion flames with varying 02 mole fractions in the oxidizer stream. Taken from Lindstedts3

02 mole fraction Model I M o d e l II M o d e l III M o d e l IV E x p e r i m e n t a l results

0.18 2 . 3 8 - 10 -8 1.00- 10 -7 3.38. 10 -7 5.11 • 10 -7 3 . 7 - 10 -7


0.22 1 . 0 0 - 10 -6 1.03. 10 -6 1.02. 10 -6 9 . 7 8 . 10 -7 1.1 • 10 -6
0.28 1.11 • 10 -5 1.07. 10 -5 2.40. 10 -6 1 . 6 4 . 10 -6 2 . 2 . 10 -6

model (IV) assumed that the particle number density errors of the order of 40%. Lindstedt correctly
was constant throughout the flame. This model had pointed out that the model is attractive in simplified
the great advantage of removing the equation for models of soot formation (such as in turbulent flame
particle number density from the set of equations that calculations) but it is likely to be less generally valid.
needed to be solved. Hence, it offered some Its use will be restricted probably to fairly strongly
computational advantage, particularly in turbulent sooting flames in which particle coagulation is rapid.
flame calculations where computational cost is a The addition of the benzene route to soot nucleation
serious consideration. Kennedy e t al. 2 employed the was found to be important. A mechanism that
same approximation in their simplified modeling of incorporated only the acetylene nucleation step
soot formation in a diffusion flame. resulted in errors of up to 50% when the surface
The code was again compared to the experimental growth model (III), the most accurate model, was
results of Vandsburger e t al. 1°4 in a counterflow used.
ethylene diffusion flame with a varying oxygen Sivathanu and Gore 35 applied the soot model of
concentration in the oxidizer stream. Lindstedt Lindstedt, Jones and co-workers 72'76 to a laminar
compared his results with the peak soot volume acetylene air flame. They solved the axial momentum
fractions that Vandsburger e t al. 1°4 measured. The equation, the gas phase mixture fraction, the energy
comparison of the four growth models with experi- equation, the soot mass fraction and the soot particle
ments is shown in Table 2 number density in their boundary layer form. Because
Generally poor agreement was obtained with the up to 30% of the carbon in an acetylene flame can be
most sophisticated model (I). The good agreement in converted to soot, the specification of the gas phase
the case with 22% 02 was considered by Lindstedt s3 mixture fraction required special treatment. It was
to be fortuitous in view of the strong sensitivity of this defined as the fraction of the local fluid mass that
model to the modeling constant that related the originated from the nozzle and was in the gas phase.
benzene concentration to the number of active growth As a result of this definition, it was not a conserved
sites per unit area and the total soot surface area. scalar and was in fact increased by oxidation of soot
Further improvements to this model are apparently and reduced by the formation of soot. Hence, its
required in order for it to achieve greater accuracy. conservation equation required a source and sink
Lindstedt s3 proposed that the assumption of spherical term that also appear in the equation for the soot
particles may pose a problem with the model; K6ylu mass fraction. State relationships between the gas
e t al. 3° have suggested that particle morphology is an phase mixture fraction and the mole fractions of
important property of the soot that is not considered acetylene, 02 and chemical energy release were
adequately in models. It is possible that a shape obtained from the results of Gore l°s in an acetylene
dependent correction to surface areas is required flame. The energy equation contained three source
although one runs the risk of adding yet another level terms, one that accounted for the conversion of
of complexity and modeling uncertainty to the acetylene into soot, another that accounted for the
calculation. heat released through oxidation and a final term to
Very good agreement was obtained with model account for radiation. Oxidation was assumed to
(III) in which soot mass growth was simply a function occur only though the action of 02; OH was not
of particle number density and independent of aerosol included in calculations of soot oxidation.
surface area. The small discrepancies were attributed Radiation, Qrad, was considered to consist of two
by Lindstedt to uncertainties in the soot nucleation parts so that
step. However, given the uncertainties in all aspects of
Q r a d = Qabs - Qemi (50)
the flame modeling, the agreement should be
considered to be quite satisfactory and is likely to be The emission
difficult to improve with general validity. Model (II)
in which growth was proportional to surface area did
Oemi = CfvT5 (51)

not perform well. The results suggest that this model was calculated from the soot volume fractionfv = e_~
Ps
of soot growth may not have general validity. Model and the temperature T, where Ys was the soot mass
(IV), with a constant particle number density, actually fraction, p was the total density of both phases (gas
achieved better results than models (I) and (II) with and soot) and ps was the density of a soot particle,
114 I.M. Kennedy

taken in this case to be 2000 kg m -3 . The constant, C, in butene laminar diffusion flames. The measurements
was a function of the refractive index of soot. were corrected for background soot scattering and for
The flame was not considered to be optically thin. PAH fluorescence by detuning the laser off resonance.
Rather, the details of absorption were included at all Quenching corrections were applied along the axis;
points in the flame. This was achieved in an iterative approximate corrections were applied to radial
manner by first calculating an initial solution that profiles.
provided the radiation absorption terms without The profile of OH in relation to the other species in
coupling to the flame structure, i.e. adiabatically. the flame was worthy of note. The visible flame height
The absorption was calculated as was about 107 mm. The radial profiles of all the major
molecular species exhibited a maximum on the
foo r2~r rTr
centerline at 102ram above the nozzle exit. How-
Qabs : J0 J0 J0 axI:~ sind0 dO dA (52)
ever, the OH profile exhibited maxima off the axis.
The authors attributed this phenomenon to the loss of
where Ix was the incident spectral intensity and a~ was
OH on soot particles near the axis of the flame where
the Planck absorption coefficient
the soot was oxidized.
(1 - exp(-KxfvAs)) (53) The concentration of OH in flames has often been
a~ - As obtained from equilibrium calculations. 2 However,
the results of Puri et al. 23 showed that OH concentrations
The spatial discretization used in the calculations was can exceed equilibrium levels by one to two orders of
As; K~ was the specific absorption coefficient of soot. magnitude at atmospheric pressure. The presence of
After the initial estimate of the flame structure was superequilibrium concentrations poses a clear pro-
obtained, the entire calculation was repeated up to six blem for soot models that treat OH attack in a
times with a radiative sink term in the energy equation detailed manner. One approach may be to obtain the
to obtain convergence of the radiation source terms to thermochemical state relationships from a laminar
within 2%. Clearly, this is not an approach that could flamelet model.
be used easily with a detailed chemical model. The The analysis of the experiments of Puri et al. 23
results of the calculation were compared with showed that the contribution of 02 to the soot
experimental measurements of soot volume fraction oxidation rate was small. The majority of the soot
that were obtained with laser extinction and with oxidation was due to OH attack in their flames. Given
probe sampling measurements of the major species. the importance of OH as an oxidizer of soot, it is
Aspects of the sampling of gas phase species in soot essential that an accurate rate for OH attack on soot
laden flows necessitated the definition of an apparent be used in a model of soot in flames. Many models
gas phase mixture fraction have used the reaction efficiency that was first
reported by Neoh et al. 24 Puri et al. 23 found that the
is (54) collision efficiency of OH with soot was a function of
~ga - (1 - Ys)
the residence time of the soot in the flame, probably as
where ~g was the actual mixture fraction that was a result of a change in the soot reactivity. The
calculated. efficiency was as low as 0.05 at early times; it rose to
The predicted soot volume fractions were com- the value of 0.2 at later times.
pared with experiments as a function of the apparent Unfortunately, the estimation of the N a g l e -
gas phase mixture fraction, ~sa" The results were Strickland-Constable rate by Puri e t al. 23 in their
shown for two different Reynolds numbers at a range 1994 paper was based on an erroneous interpretation
of distances from the nozzle. The predictions of soot of the original rate expression for 02 attack. A
volume fraction were remarkably accurate, particu- subsequent short note 1°9 revised the estimate of the
larly in view of the fact that oxidation by OH was O2 rate upward by a factor of 12. Consequently, the
ignored. original comparison of OH oxidation rates with O2
The calculations of Sivathanu and Gore 35 high- rates overstated the role of OH. Furthermore, the case
lighted the importance of including radiation in a for a variation of the collision efficiency of OH with
coupled manner in calculations of sooting flames. The soot could not be sustained although the data from
results suggested that the soot model of Leung et al. 76 the CH4/C4HI0 flame continued to show a trend to
was fairly robust, although some caution was in order increasing collision efficiencies with height in the
in view of the absence of a satisfactory sub-model for flame. If the variation of collision efficiency with
oxidation that included the attack of OH on soot. One time were substantiated, it would pose an additional
may well expect a significant deterioration in the challenge to modelers of soot in flames. For example,
quality of the predictions using the same model Kennedy et al. 11o found that a variation of OH collision
parameters if OH attack were included. efficiency played a role in accurately capturing the
The importance and the difficulties of including an behavior of the soot aerosol during its formation
accurate model of OH oxidation were revealed in the and its oxidation.
paper by Puri et al. 23 They measured laser induced Miller et aL H1 attempted to calculate the growth of
fluorescence of OH in methane, methane-butane and PAHs, and hence soot, in a laminar diffusion flame by
Models of soot formation and oxidation I 15

extending the approach of Kennedy et al. 2 They when it was incorporated into the numerical model of
assumed that PAH did not diffuse and as a result, the flame.
these relatively large molecules followed streamlines Measurements were performed in laminar ethane
in the flame. A laminar flamelet concept was invoked and ethylene flames. Maps of the soot mass growth
to correlate the concentrations of species such as H, rates per unit aerosol surface area (assuming a
H2, C2H2, 02 and OH with the local mixture fraction monodisperse aerosol for the sake of measurement
at some point in the flame. Measurements by Smyth interpretation) were produced from data for a
et al)12 in a methane-air flame provided an estimate number of flames. The maps set out contours of
of the formation rate of benzene as a function of mass growth rates as functions of mixture fraction
mixture fraction. Benzene was considered to be the and temperature. When the map data for the range of
inception species for soot particles; the formation of flames of a given fuel type were combined, the root
benzene was followed by an irreversible growth of the mean square scatter was about 10-30% of the mean
PAH. The growth rates of the PAH were determined growth rate. This result suggests that it may reason-
from a modified version of Frenklach's HACA able to apply the growth function over a broader
scheme. Oxidation of PAH or soot by 02 and by range of conditions for a given fuel. The hypothesis
OH was considered. The rate of OH attack was based was tested by using the growth function as a source
on the reported rate of OH reaction with benzene113 term in the numerical model of the flame with an
while the rate of O2 reaction was based on the additional equation for the soot mass.
reported rate of reaction of oxygen with phenyl The correlation of measured soot volume fractions
radical. Soot was defined to be the sum of masses of with predicted mixture fraction showed a significant
all species larger than coronene; the density of soot variation with height in the flame. This provided clear
was assumed to be 1800kg m -3. proof that the use of mixture fraction alone is not
Initial calculations with the model yielded dis- sufficient to yield useful predictions of soot loading.
appointingly small soot volume fractions, with peak Honnery and Kent 116 and Kent and Honnery t 14 in a
values of the order of 10-21 . The sensitivity to growth later study attempted to improve on the mixture
rates and to oxidation rates was explored. The fraction approach by correlating measured soot
ultimate peak soot volume fraction was sensitive to volume fractions with two variables viz., mixture
the growth rate. Volume fractions of the order of 10-7 fraction and temperature. Honnery and Kent 1t6
were obtained by increasing growth rates by a factor analyzed the specific surface growth rate data of
of 200. The fastest growth of PAHs was along Honnery et a l ) 17 in C2H4 and C2H6 diffusion flames.
streamlines on which H atoms and C2H2 overlapped The rates, wg were correlated in terms of mixture
in the absence of OH or 02. This was true for mixture fraction and temperature, i.e. wg = A s k s ( T , ~) where
fractions between 0.08 and 0.11. As was the soot surface area, ks was the specific
It has been shown rather convincingly that simple growth rate and ~ was the mixture fraction. Equations
state relationships between soot volume fraction and were solved for the mixture fraction, enthalpy and
mixture fraction are not adequate. Kent and soot volume fraction. An approximation for radiative
Honnery74'114 sought to extend the utility of the transfer, suitable for use with a parabolic code, was
mixture fraction approach by mapping soot growth applied. The soot surface area As was calculated from
rates in diffusion flames as functions of mixture a correlation of optically determined surface area
fractions and temperature. The original work of Kent and soot volume fraction in a linear form viz.,
and HonneryTM used laser extinction measurements in A s = a + bfv. For an ethylene flame, a = 55.5 and
turbulent ethylene diffusion flames to obtain a b = 78.6 with units of [m] and [ppmv]. The use of the
correlation between soot volume fraction and mean empirical correlation of surface area with volume
mixture fraction. The mixture fraction at the location fraction obviates the need to calculate the particle
where the soot volume fraction was measured was number density. This approach is somewhat more
inferred from a numerical calculation of the turbulent sophisticated than the approach adopted by Kennedy
flame. et al. 2 who assumed a constant average number
The numerical model of the flame was written in density in which case the surface area varied with
terms of a stream function115 formulation of the fv 2/3. Oxidation was treated with the scheme of
axisymmetric boundary layer equations for a laminar Bradley et al. 22 and included 02 and OH attack. The
diffusion flame. Equations were solved for momen- concentrations of these species were obtained from an
tum and energy. The energy equation contained a equilibrium calculation.
term to account for the loss of enthalpy via radiation. The model was used to predict the soot mass flow
The radiation flux was found by solving the radiative rate, soot volume fractions and temperatures in
transfer equation in the horizontal plane. The ethane and ethylene flames of various flame heights.
numerical simulation of the flame served two Generally impressive agreement with the measure-
purposes viz., to determine the velocity field and the ments of soot mass flow rates was found in the
mixture fraction field so that soot mass growth rates formation zones of the flames although the agreement
could be deduced along soot trajectories and also to was rather less satisfactory at large distances from the
evaluate the accuracy of the model of soot growth nozzle, in the burnout zone. The measured soot mass
116 I.M. Kennedy

flow rates were overpredicted, possibly as a result of the inner tube; air flowed through the outer tube. The
uncertainties in the oxidation model. Honnery and Reynolds numbers were about 50. Kaplan et al. l~s
Kent 116 reaffirmed the sensitivity of soot volume found that their computations indicated that the
fraction predictions to the specific surface growth flame was unsteady, despite the relatively low
rates that was reported by Kennedy et al. 2 Reynolds number. The flicker at 16Hz was not
Results in the later study 114 in ethane and ethylene attributed to a numerical instability, rather to a
flames were encouraging, particularly in the more buoyancy driven instability of the flame itself. The
heavily sooting ethylene flames. Fuel flow rates authors compared their calculations with the experi-
covered an order of magnitude for the ethylene mental results of Gore and Faeth s° by analyzing their
flames, from non-sooting to heavily sooting. The results in terms of state relationships between mixture
rate map model performed reasonably well in fraction (or fuel equivalence ratio in this case) and
reproducing trends of soot mass flow rates as species mass fractions and soot volume fractions.
functions of height above the burner. The greatest They reported a reasonable agreement for the major
difficulty was in capturing the burnout of soot at the species such as C2H4, 02, CO 2 and H20. It is very
end of the flame with the low flow rate. The model difficult to ascertain the accuracy of the soot
predicted a sooting flame, whereas the experiments calculation from the state relationship interpretation
showed that the flame was non-sooting. The authors of the results, particularly since the numerical results
ascribed the discrepancy to superequilibrium concen- exhibited some non-systematic scatter. The scatter
trations of O and OH. The approach of Kent and may be due to the inadequacy of the soot-mixture
Honnery114 was not greatly different to that proposed fraction state relationship.
and tested by Kennedy et al. 2 and Villasenor and A second calculation was carried out for a larger
Kennedy. v9 Both groups used measured rates of soot Reynolds number of about 5000. The results of this
growth as inputs to models of flames. Kent and study were used to illustrate the impact of radiation
Honnery114 chose to work with a correlation between on the structure of the diffusion flame. Radiative heat
growth rates and temperatures while Villasenor and losses reduced temperatures and chemical heat release
Kennedy 79 incorporated the effect of temperature via rates. As a result, the flame was shorter than the
an assumed activation energy. The quality of the corresponding adiabatic flame. The calculations were
predictions was similar. Burnout of soot at the end of computationally intensive, largely as a result of the
a diffusion flame was difficult to capture accurately use of the DOM radiation model on the fine grid that
due to uncertainties in the kinetic rates of oxidation was required to resolve the flame structure.
and more importantly, due to uncertainties in Kennedy et aL ~l° modeled a sooting and a non-
predicted OH and O concentrations. This point was sooting flame for the conditions of Santoro's
emphasized by Kennedy et al.U° experiments. 9° The boundary layer forms of the
Some authors have used a semi-empirical model of governing equations for a round laminar jet flame
soot formation to investigate the impact of soot were solved in primitive variable form with zero axial
radiation on flame structure in detail. Kaplan et al.118 pressure gradient. The energy equation was written in
used a rather complete radiation model in a terms of the temperature with the radiation repre-
calculation of a laminar ethylene diffusion flame. sented by the assumption of an optically thin flame
They employed the Discrete Ordinate Method 119 to without gas phase radiation, only soot radiation.
calculate the radiation from CO2, H20 and soot. Equations were solved for N-I species mass
Experimental data were used to estimate the absorp- fractions
tion coefficient of these species.
They used the soot model of Moss et al. 92 Hence, OYn OYn _ 1 0 {rpYnV,} +Muff' n (55)
pu Oxx + Try Or r Or
two equations were solved for the soot viz., the soot
volume fraction and the soot number density. with the Ntb species being N2. The N-1 equations
Oxidation was attributed entirely to the action of O2 included one for the soot mass fraction, Ys
through the use of the Nagle-Strickland-Constable
formulaJ s As a result, the rate of oxidation was ors oYs 1o
PU-~x +P'V Or rOr { r p V T Y s } + p S ( Y s ' T ' f )
probably underestimated to some extent by ignoring
the important effect of OH on soot. A detailed (56)
description of the flow field was used with equations
for continuity, momentum, energy and species In this equation, the thermophoretic velocity was
number densities. However, the chemical kinetics found from the appropriate expression for a free
were described in very simple terms with a one step molecular aerosol ~2°
reaction of fuel to COz and H20.
A 'benchmark' calculation was undertaken to VT = -0.55T00T (57)
compare the code with the measurements of Gore
and Faeth s° in a laminar, co-flow, ethylene diffusion
flame. The burner consisted of a 14.3 mm dia round The source term in Eq. (56) included the contributions
tube within a 102 mm outer tube. Fuel flowed through of soot nucleation (~nuc0, soot surface growth 0bs)
Models of soot formation and oxidation 117

and soot oxidation (Wo) A

1.8
E
S(Ys, T,f) = Wnucl+ fig - fro (58) 1.6-
The latter term had two contributions viz., oxidation × 1.4-
of soot by 02 and by OH to form CO.
Earlier modeling studies of soot formation by --~ 1.2'
Kennedy et al. 2 used an average number density to
obtain an estimate of the soot aerosol surface area.
The more recent study followed the approach of
Fairweather et al. 72 in solving an equation for the o.6
number density of particles. It should be noted that
0.41
their equation for the particle number density did not
incorporate the effect of thermophoresis. mO.2 ~
The model of Fairweather et al. 72 was adopted in its
_~ o .... I .... I .... I .... I .... I .... I .... I ....
entirety with the following important difference. The 0 1 2 3 4 5 6 7 8
rate of soot surface growth was increased by a factor x/D
of two to yield an adequate representation of the
measured soot volume fractions in the flame. This a d Fig. 7. Integrated soot volume fractions in the non-sooting
flame of Santoro et al. 90 (--) Predictions from Kennedy
hoc action was adopted because Fairweather et al. 72 et al. n°
did not incorporate the impact of OH in their model.
In their case, the oxidation of soot was attributed in a discrete form. The non-linear terms (convective
entirely to 02. The parameters in their model were and rate terms) were linearized by using Newton's
calibrated against experiments on that basis. As a method. The resulting Jacobian matrices were
result, their soot growth model underpredicted the determined numerically in the manner described by
measured soot volume fractions when it was used by Smooke and Giovangigli.123 Grid adaptation was
Kennedy et a l . l l ° with OH oxidation. used in order to resolve the high gradient region in the
Particular attention was given to the correct flow field with a minimum number of grid points.
formulation of the oxidation mechanism in the Most of the results were obtained in the non-
model. The Nagle-Strickland-Constable18 formula sooting flame of Santoro et al. 90 Predictions of
was used for 02 oxidation of soot. Oxidation of soot temperature profiles and velocities were quite good.
by OH was generally more important in most of the Predicted profiles of major species compared quite
flame and, hence, it required greater attention than well with the measurements that were obtained by
the 02 mechanism. Calculations were originally probe sampling and GC/MS analysis. A measure of
performed with the approach that was adopted by the accuracy of the soot model is given by the
Villasenor and Kennedy,79 i.e. a collision efficiency of calculation of the integrated soot volume fraction
0.1 was used for the reaction of OH with soot. (Fig. 7) in a non-sooting flame. This quantity was
However, it was found that a constant value for the obtained by integrating the soot volume fraction
collision efficiency, regardless of its magnitude, could across a plane of the jet. It indicates the total amount
not faithfully reproduce the characteristics of the soot of soot at a given axial location in the flame. The
distribution in the flame. Consequently, reference was agreement between measurements and calculations
made to the results of Puri et al. 23 who found that the was good. The model was able to reproduce the peak
apparent collision efficiency of OH with soot changed integrated soot volume fraction as well as predict the
with time and/or temperature. The collision efficiency burnout of soot at the flame tip. Detailed compar-
was assumed to vary linearly with distance from the isons of the radial distribution of soot in the flame are
fuel nozzle, from a minimum of 0.05 to a maximum of shown in Figs 8 and 9 at 15 and 50mm from the
0.2 at the flame tip. This model of OH reactivity nozzle, respectively. In both cases, the maximum soot
achieved reasonable results although it was quite volume fraction was captured quite well by the model.
crude. However, the presence of significant amounts of soot
A simple mechanism was used to describe concisely on the centerline of the flame at a position 50 mm
the Cl and the C2 reactions for ethylene oxidation. from the nozzle was not predicted. In order to
The methane mechanism of Smooke et al. 121 was reproduce this behavior, a more detailed soot model
implemented in the code with the rates for all the C2 is required, one that accounts for soot formation from
chemistry from Frenklach et al. 122 The C2 chemistry PAH that are known to form on the axis of diffusion
was described by a series of reactions that converted flames.
C2H4 to C2H 3 and C2H2. The entire reaction Hydroxyl mole fractions were measured with laser
mechanism was composed of 62 reactions involving induced fluorescence. The calculated hydroxyl mole
24 species. The important oxidation steps by O, 02 fractions were compared with the measurements at 7
and OH were included. The CO oxidation steps and and 70 mm. The model tended to overpredict the OH
the H2-O2 steps were also included. mole fractions throughout the flame. However, it
The governing differential equations were written should be borne in mind that LIF measurements
118 I.M. Kennedy

3.5 greatest for the minor, radical species such as O and


OH. These species are of particular interest in the
3 formation and oxidation of pollutants such as NO
% and soot. The ratio of computed OH mole
x 2.5 fractions to equilibrium mole fractions at the
C
O
local computed temperature were calculated at
2
40mm from the nozzle. At a radius of r / D = 0.745,
°Ht~'.~=' = 22 while ~ = 478. This location
~ 1.5
was ~n the air side of t'~e reaction zone at a mixture
fraction of around 0.03 (stoichiometric mixture
fraction is 0.064). It was apparent that large super-
o3
0.5 equilibrium concentrations could be achieved as a
result of the depressed temperatures in this radiating
0 t,,, i,,,, i,,,, i,,,, i,,,, i,,,,
flame. Puri et al. 124 found that the OH super-
0 0.1 0.2 0.3 0.4 0.5 0.6 equilibrium ratio was 9.4 at the peak OH location in
r/D this flame at a height of 70 mm. The results suggested
Fig. 8. Soot volume fractions at 15mm from nozzle. ( - - ) is that the assumption of equilibrium chemistry was
predictions from Kennedy et al.; 11° measurements from grossly inadequate for the prediction of the minor
Santoro et al. 9° species. This finding is particularly troublesome in
terms of the calculation of OH oxidation of soot.
12 One of the most difficult tests for any model of soot
behavior in diffusion flames is the prediction of the
% 1o • transition from a non-sooting to a sooting condition
as the fuel flow rate is increased. The correct
X reproduction of the transitional behavior requires
i8 a
° not only the accurate calculation of soot formation
but also the correct calculation of the soot burnout at
® 6- the tip of the flame. Experimental conditions for the
sooting flame of Santoro e t al. 9° were used in the
4, • calculation of Kennedy e t al. 11° The increased
production of soot with an increased flow rate of
09 2 fuel was captured quite well by the code, particularly
in regard to the peak soot volume fraction. The code
did a reasonable job of predicting the peak soot
0 '''' I'''' I'''' I'''' I'''' I''''
0 0.05 0.1 0.15 0.2 0.25 0.3 volume fraction in the flame. However, it was not able
r/D to correctly predict the emission of soot, as is evident
in Fig. 10 that shows the integrated soot volume
Fig. 9. Soot volume fractions at 50 mm from nozzle. (--) is
predictions from Kennedy et al.; 11° measurements from fractions in a sooting flame. This defect was attributed
Santoro et al. 9° to an excessively vigorous oxidation of soot that
proceeded too rapidly and persisted too long in the
involve uncertainties that are related to quenching cooling post flame gases. Kennedy et al. 12° believed
corrections and other errors. The errors were that the problem lay with the oxidation rate that was
estimated to be +50%. TM The numerical model calculated from the Nagle-Strickland-Constable
indicated the measured trend in OH mole fractions formula. 2s Evidence is available in the literature that
with distance from the nozzle although absolute the Nagle-Strickland-Constable formula overesti-
amounts were overpredicted. The concentrations of mates the oxidation rate of soot and char by
OH decreased as the flame tip was approached due to 02225'126 at temperatures below about 1800K;
the drop in temperature and due to the loss of OH by temperatures in postflame gases are typically about
reaction with soot particles and CO. 1600K or less. Evidently, the accurate prediction of
The use of a detailed kinetics model with a model the transition from non-sooting to sooting flames may
for soot formation and radiation permits the require an improved model for the low temperature
examination of questions regarding the impact of oxidation of soot by 02.
the soot process on flame chemistry in a manner that Kennedy et al.22° found that the inclusion of soot in
experiments alone cannot attempt. For example, the description of a diffusion flame could have a
calculations of turbulent flames require a model for significant impact on the structure of the flame. The
the thermochemistry. A convenient and very common rate of loss of OH on soot particles was comparable to
approach is to use chemical equilibrium. It is possible the rate of formation and destruction by the gas phase
to examine the accuracy of this assumption when chemistry, as shown in Fig. 11 at a location near the
there is substantial radiative heat loss from a tip of the non-sooting flame of Santoro e t al. 9° It was
hydrocarbon flame. The non-equilibrium effect is apparent from these results that soot is a significant
Models of soot formation and oxidation 119

2 • phases of development was directed to modeling

fh
soot formation in laminar, premixed flames. More
recently, Yoshihara et al. 12s applied the model to the
industrially important problem of soot formation and
burnout in a diesel engine.

L \ Frenklach et al. 7 considered the detailed modeling


of soot formation during the shock tube pyrolysis of
acetylene. They extended the acetylene pyrolysis
mechanism of Tanzawa and Gardiner 129 to describe
the early hydrocarbon chemistry. The dominant route
8o. "i \"
\" for the formation of the first aromatic ring from an
atiphatic fuel was found to occur via vinyl addition to
acetylene, leading eventually to cyclization via the
o~,,,,,,,,,,,,,,,,,,,' "%,,. interaction of an unpaired electron with a triple
0 2 4 6 8 10 12 carbon bond. The aromatic ring then grew through
x/D the so-called HACA route (H abstraction-C2H2
Fig. 10. Integrated soot volume fractions in the sooting flame addition) in which thermodynamically stable species
of Santoro et al. 9° (--) predictions,U°points: measurements. such as acenaphthalene and coronene were formed
and were subsequently reactivated by H atom
3 1800 abstraction. Reactive coagulation of PAH could
also be important, i.e. the fusing of one or more
PAH to form a larger entity. This mechanism may be
1400 particularly important when the parent fuel is an
~tlJ
~, 1200 aromatic such as benzene. Frenklach et al. 7 found
E that the major 'bottleneck' in soot formation from
1000 -~
acetylene pyrolysis was at the step that led to
80O formation of the first aromatic ring. Their computa-
®
- I :
-60o E tional results showed reasonable agreement with
-2 II experiments in the shock tube, particularly if
8 II --14°° allowance were made for uncertainties in the inter-
ca -3- t -200 pretation of optical measurements of soot.
The growth of larger PAHs through many
-4 ,,i,,,~, ,),,,~,,,i,,,;,,,t,,,),,,),,, 0
0 0.2 0.4 0.6 (:).8 1 1.2 1.4 1.6 1.8 2 elementary reactions was accounted for via the
r/D method of linear lumping in which equations were
derived for the change in moments of the distributions
Fig. 11. Source and sink terms for OH at 70 mm from the nozzle
in the non-sooting flame of Santoro eta/. 9° and temperature
of concentration and size. 127The method of moments
profile. ( - - ) gas phase chemistry source of OH, (- - -) loss rate of is potentially more computationally efficient than a
OH on soot particles. From Kennedy et al. t=° sectional description 13° in which an aerosol distribu-
tion is divided into finite size groups that then
sink of OH near the end of the flame in comparison to represent a group of particles. The kth moment of
the gas phase chemistry. concentration was defined as
oo

3.3. Models with Detailed Chemistry Mk = Z mki Ni (59)


i=1

Models that rely on empirical inputs for soot where M k was the kth concentration moment, m~ was
nucleation, growth and oxidation rates are limited mass of PAH species of class i and Ni was the number
inherently to specific', conditions. They cannot be concentration of the same size class. The size
applied to conditions far from those under which the moments were defined as normalized concentration
rates were measured. Hence, they cannot be extended moments as follows
easily to different fuels or different pressures. A
complete description of the kinetics of PAH and soot
k mk
-- M~ (60)
growth is required for generality.
Frenklach and co-workers 7A°TA27have worked on Linear lumping was also applied to the evolution of
the development of detailed soot models over the last the size distribution of soot particles. A soot particle
decade or so. Their model has reached the greatest was taken to be formed when two PAH molecules
level of sophistication of any of the extant models. combined to form a dimer. Rate equations were
Frenklach treats the full panoply of phenomena, from written for the change in the moments of the soot
the initial pyrolysis of fuel to the nucleation of soot distribution; these equations included nucleation,
particles, their growth and coagulation and their coagulation and surface reactions in a form
ultimate oxidation. Most of the attention in the early compatible with the moment formulation.
120 I.M. Kennedy

Growth of soot particles by surface reaction was products of soot oxidation in a heavily laden diffusion
described by the addition of acetylene to a radical flame may participate in the overall flame chemistry in
'arm chair' site on a particle) 27 The radical site was important ways. Our present understanding of the
formed by H atom abstraction as shown below details of OH reactions with soot is deficient.
Frenklach and Wang 1°7 compared their modeling
C r H + H ~ Ci • +H2 (61)
results with the measurements of Wieschnowsky et
Ci • + H -~ Ci-H (62) al. 16 and Harris et al. 131 in premixed flames. The
irreversible coalescence of PAHs was enhanced by a
Ci • +C2H2 --' Ci+2-H + H (63) factor of 2.2, following the numerical results of Harris
and Kennedy ~1 for van der Waals forces between soot
Oxidation by 02 was also included
particles. This may be an overestimate of the
Ci • + 0 2 --, products (64) collisional enhancement by dispersion forces because
The rate of reaction (64) was found by analogy with the value of 2.2 was appropriate to small, like-sized
the gas phase reaction of an aryl radical with 02 particles, not to the full panoply of collisional pairs.
Frenklach and Wang employed a value of a of 0.1 for
aryl • + 0 2 -~ products modeling Flame 1 of Wieschnowsky et a l ) 6 The
The rates of reactions (61), (62) and (63) were found agreement with measurements of particle diameter
by analogy with the gas phase reactions of H atoms was good although predictions of the particle number
and C2H2 with aryl radicals under the assumption that density were less satisfactory. The parameter, a , was
the collision efficiencies were the same per active site. increased to 0.7 in order to achieve satisfactory
The rate of growth of a particle was formulated as agreement with the measurements of Harris et a l l 3~
Several interesting conclusions were drawn from the
fig = kCgak~Csoot • Sn (65)
modeling efforts. Frenklach and Wang 1°7 were able to
in which k was the per-site rate coefficient, C s was the reconcile the observations of Wieschnowsky et al. 16
concentration of the growth species, a was an with the measurements of Harris et al. ~4:5 Specifi-
empirical steric factor, S was the surface area of a cally, Wieschnowsky et al. 16 found that it was
spherical particle and n was the number density of possible to modify the growth of soot by inhibiting
particles of given size class. It assumed that the soot coagulation through seeding flames with CsCI. They
particle could be represented as a sphere with concluded that the results of their experiments
diameter that was based on its mass and the density indicated that soot growth was independent of
of bulk soot (taken as 1800kg m-a). Surface area of surface area. On the other hand, Harris et al. 14'15A31
the particle was multiplied by a factor, kVc~oot• , to claimed that soot growth rates were first order in
account for the number of surface radicals per unit aerosol surface area. Furthermore, it should be noted
surface area. This factor was, in turn, derived from a that under the conditions of the Wieschnowsky
steady state assumption applied to reactions (61)-(64) experiment, the surface growth was dominated by
so that the addition of PAHs whereas growth in the
k61 [H] experiments of Harris et al. was predominantly due
~/Csoot • = to acetylene. The model of Frenklach and Wang 127
k-61 [n2] + k62[n] + k63[C2H2] + k64[O2]
described soot surface reactions in terms of elemen-
X ~ItCsoot_H (66) tary reactions at active sites. The density of active sites
The number of c i - a sites per unit area of soot on soot particles was controlled by the gas phase
particle (kVCsoot_H) was estimated on the basis of environment. For relatively low temperature flames,
dimensions of PAH rings and distance between PAH such as those studied by Harris and co-workers, the
layers in soot to have a value of about 2.3 x 1019 to denominator of Eq. (66) was dominated by the term
2.9 x 1019 m -2. k63 [C2H2] so that the surface growth rate due to
Soot was oxidized due to the reaction of 02 with a acetylene addition became
radical site ~g = 2Otff'Csoot_Hk61[H]Sn (69)
Ci • + 0 2 --* products (67) The rate of surface growth was then controlled by the
and by the attack of OH on the soot surface rate of H atom abstraction and was first order in
aerosol surface area. Frenklach and Wang 127 were
C i - H + OH ~ products (68)
able to reproduce the experimental results of
The model employed the experimental results of Neoh Wieschnowsky et a l ) 6 in a higher temperature flame
et al. 24 and hence used a reaction probability of OH in which growth rates were not a function of surface
with soot of 0.13. The products of 02 and OH area. Hence, the apparent contradiction was resolved
reaction with soot were not given explicitly because through a reasonable hypothesis that invoked a
they did not play an important role in the chemistry mechanism of soot surface reactions in terms of
for the conditions that were studied. However, this active sites.
issue can be important in terms of the structure of the Frenklach and Wang 127 gave attention to the
flame as Kennedy et al)1o pointed out. The gas phase optical properties of the soot aerosol by using an
Models of soot formation and oxidation 121

expansion of the Mie scattering coefficient and 2.2


rewriting it in terms of the soot aerosol moments, #
2.1'
and M, that were defined above. This approach O
allowed the authors to compare their predictions ._o 2. •
directly with measured scattering and extinction
coefficients. ~
,.-° 1.9-

Calculations were performed for the same condi- _e

tions as those used by Frenklach and W a n g ) °7 ._> 1.8.


Attention was focused on Flame 1 of Wieschnowsky 1.7:
e t al.; 16 it was a low pressure premixed flame of ~
t..)
•.~.

acetylene, oxygen and argon. By choosing a value of c~ :~ 1.6- "-


0 "
of 0.35, the authors were able to achieve impressive
1.5: •
agreement with the light scattering/extinction data. A
value ofc~ = 0.1 gave rise to discrepancies of an order 1.4 . ,, , . . . ,, . . , , , . , , . . , ,.,
of magnitude. It should be noted that the authors had 5.2 5.4 5.6 5.8 6 6.2 6.4
used a value of ~ of 0.1 in an earlier modeling effort t27 104/T ( K "1 )

for the same flame in which the optical data had been
Fig. 12. Critical equivalence ratios for N2 diluted flames.
interpreted in a simple fashion. The need to revise
Points are computed by Markatou et al. 5 Corresponding
the value of c~ highlighted the uncertainty in the lines are curve fits to tile experimental data of Harris
interpretation of optical data without an accurate et al.4°'134(- - -) • C2H2, (--) • C2H4, (- . -) • C2H6.
estimate of scattering and extinction cross sections.
An important test of a soot model for laminar temperature dependence
premixed flames is its ability to predict the critical
1t /12180 1]
equivalence ratio, X¢, at which soot first appears. = [tanh~- 7.19) + (70)
Markatou e t al. 5 examined the laminar premixed
flames of Harris e t al. 40A32 who studied flames of This function was obtained by fitting values of a at
acetylene, ethylene and ethane. The temperatures of two temperatures to values determined in an earlier
the flames were controlled through the addition of N2. modeling effort j°7 and by noting that the hyperbolic
The early flame chemistry was computed in two steps: tangent function exhibited a physically reasonable
the Ci to C4 chemistry was calculated with 32 species behavior in the limits of low and high temperatures.
and 139 reactions for a free (non-burner stabilized) In order to compare numerical results with experi-
flame. This yielded the temperature and the flow field. ments, the authors computed luminous radiation
The production of PAH up to coronene was then intensities with approximations to Mie theory. The
calculated with a 'full' mechanism of 70 species and critical equivalence ratio was defined as the intercept
337 reactions. Splitting the calculation in this manner on the X axis by a tangent to the curve of luminous
saved considerable computational time and yet it radiation flux. The relationship to visual observations
yielded acceptable answers. The SANDIA burner of the onset of soot luminosity in experiments is
code 133 was used to perform the calculations. somewhat uncertain. The definition did provide,
The formation of the first aromatic ring was however, a consistent manner in which to deal with
described by reactions of acetylene with Ca species the numerical results.
A comparison of the predicted critical equivalence
n -- C4H 3 + C2H 2 ~ phenyl
ratios with a linear fit to the measurements of Harris
n - C4H 5 + C2H2 ~ benzene + H e t al. 4°A34 is shown in Fig. 12. The agreement is
remarkably good in view of the uncertainties in the
as well as the thermal route through
definition of critical equivalence ratio, both experi-
n - C4H 3 + C2H 2 ~ n - C6H 5 ~ phenyl mental and computational, as well as the host of other
factors that are included in a complex model of soot
n - C4H 5 + C2H 2 ~ n - C6H 7 ~ benzene + H
formation. Markatou e t al. 5 went on to use their
The combination of propargyl (C3H 3 + C3H3) to model of the premixed flames to investigate the
form benzene or phenyl was not included by relative importance of nucleation, coagulation, sur-
Markatou e t al. 5 because they calculated propargyl face growth and oxidation in controlling the produc-
mole fractions that were considered to be too low to tion of soot in these flames. They concluded that the
have an impact on the formation of benzene in their formation of PAH and hence soot was controlled
flames. largely by the relative abundance of acetylene and by
The growth of PAH and soot particles followed the the thermal decomposition or growth of PAH species.
route set out by Frenklach and Wang 127 with the Oxidation of soot particles, p e r se, was not predicted
following notable exception: the steric factor that to be an important contributor to the overall response
was used by Frenklach and Wang 1°7 was not a of the soot process to equivalence ratio. The
constant value in the calculations of Markatou e t aL 5 major role for oxidation was ascribed to the early
They assumed that the steric factor, a, exhibited a chemistry of hydrocarbon oxidation and pyrolysis
122 I.M. Kennedy

that leads to the formation of the first aromatic unnecessary to consider larger PAHs in view of the
ring. authors' observations that the size of the ultimate
In a somewhat ambitious departure from modeling PAH had a minor impact, less than a factor of two, on
simple laminar flames, Yoshihara et a l l 2s employed a predictions of soot concentrations. The soot model
modified version of the soot model of Frenklach and itself followed closely the earlier work by Frenklach
Wang ~27to calculate fuel oxidation, PAH growth and and his co-workersJ °7'127 Growth of soot was
soot formation in a natural gas-fueled, direct injection assumed to follow the H A C A model. The aerosol
diesel engine. The level of detail of the modeling was dynamics were treated by the method of moments.
significantly greater than previous attempts to model The only significant departure from the earlier models
soot formation in diesel engines. 43'57'63 The calcula- was related to the model of OH oxidation of soot. A
tions were intended to match experimental conditions temperature dependent rate of OH attack on soot was
for a single cylinder, naturally aspirated engine with a developed to accommodate the experimental findings
compression ratio of 16 running at a speed of of Puri et al. 23 The collision efficiency for OH with
1800 rpm. soot (70~) was written as 12s
A simple model of the turbulent mixing of fuel and
air within the engine cylinder was used. The contents 7oH = 0.42 [tanh ( - ~ - 2 . 8 ) + 1] (71)
of the cylinder were divided into 197 computational
cells of equal mass, 190 cells of air and 7 of fuel. The
cells were allowed to mix and coalesce in the manner It was found that 197 cells were sufficient to yield
described by Curl. 6° Two cells that underwent mixing statistically similar results. The only data point for
were replaced by two cells of a homogeneous mixture comparison with experiments consisted of a measure-
determined by the state of the two cells that coalesced. ment of soot mass fraction at the engine exhaust,
After mixing, the gases in the new cells reacted at a 5 x 10-5. The model achieved a remarkably good
finite rate according to a reduced chemical mechan- agreement with the measurement.
ism. The collision frequency of cells was estimated by Some attention was given to the details of the OH
fitting calculated heat release rates to measurements. collision efficiency. The temperature dependent
The model ignored entirely the complexities of the efficiency shown in Eq. (71) was replaced with a
fluid flow within the cylinder and apparently constant OH collision efficiency of 7oH = 0.13. The
considered the gases within the cylinder to constitute soot mass fraction at the exhaust crank angle in this
an unmixed but macroscopically uniform distribution case plummeted to about 10-7 , well below the
of fuel and air. Hence, details of fuel spray dynamics measured value. The validity of Eq. (71) is open to
did not enter the calculation. question in light of the discovery by Puri e t al. 1°9 of an
The chemical kinetics were treated in a logical error in their original estimation of the rate of 02
fashion by dividing the chemistry into three groups: oxidation of soot and, hence, in their estimation of the
hydrocarbon oxidation kinetics, PAH growth and rate of OH attack. However, Kennedy et alJ l° also
soot formation and oxidation. The comprehensive found that it was necessary to include a temperature
mechanism of Frenklach et alJ 22 for methane dependence of the OH collision efficiency to match the
oxidation was reduced systematically135 to 21 species observed experimental data. The very good agree-
and 70 reactions. Reactions were added to account for ment that Yoshihara et al.12s obtained with measured
the reactions of O atoms with acetylene, dissociation exhaust soot levels must be considered rather
of propane, formation of C3Hx species and finally fortuitous in view of the uncertainty that surrounds
formation of benzene via C3H3 + C3H3 --} benzene. Eq. (71) and the rate of OH attack on soot.
The more comprehensive PAH mechanisms of Kazakov e t al. 136 returned to the issue of assigning
Frenklach and co-workers 1°7'127 were subjected to a a value to the factor, a, in a study of soot formation in
reaction flux analysis to identify 12 key reactions that high pressure, 10 bar, premixed ethene-air flamesJ 37
formed PAHs up to pyrene. When steady state and The equivalence ratio of the flames was adjusted in
partial equilibrium assumptions were applied to this the experiments to yield a range of temperatures from
reduced reaction set, an analytical expression for 1711 to 2017 K. The soot modeling of Frenklach and
pyrene concentrations was obtained in terms of Wang 1°7'127 was used for the high pressure flames.
benzene concentrations. The results obtained with Very good agreement with the experiments at 10 bar
this approximation were compared with a calculation was obtained by adjusting only one parameter,a. It
that used the complete reaction set for typical engine was argued that the mobility of the structural units of
conditions. The steady state/partial equilibrium soot would increase with temperature and hence a
assumptions generally led to overpredicted pyrene would decrease with increasing temperature. This
concentrations compared to the more complete hypothesis was tested by finding the value of a
kinetics calculations. For example, the approximated that gave the best fit to experiments for a range of
pyrene concentration was an order of magnitude conditions. The results in Fig. 13 appear to confirm
greater than the kinetically derived concentration at a the physical reasoning. The functional form of a(T)
time of 0.2msec after initiation of reaction. PAH that was presented in Eq. (70) was revised by Kazakov
growth was followed up to pyrene. It was deemed et al. 136 on the basis of the data shown in Fig. 13 to
Models of soot formation and oxidation 123

0.7 although it was argued that it was energetically


possible. The detailed growth modeling supported
0.6' earlier findings that growth, at least for the conditions
of the study, was proportional to the C2H2 concen-
0.5. tration. The use of Monte Carlo simulations of
surface growth was the most ambitious fundamental
0.4 modeling effort that has been applied to soot
• & processes. However, it is quite possible that rather
0.3-
different conclusions may have been reached if some
adjustments to the rate constants were implemented.
0.2-
A simplified version of Frenklach's early soot
0.1- modeling7 was used by Nickerson and Johnson 14° to
predict the formation of soot in a rocket gas generator
0 fueled on liquid oxygen/hydrocarbon (LOX/HC).
1600 1¢00 ts'oo 19'00 20bo 2100 The problem was similar to the situation that
Maximum flame temperature (K) motivated Makel and Kennedyl°° to model a
laminar inverse diffusion flame. However, Nickerson
Fig. 13. Fraction of radical sites available on soot surfaces,
a, as a function of the maximum flame temperature. From and Johnson attempted to predict the behavior of a
Kazakov et aL 136 (m) ethene-air flameat 1 MPa, (O) ethene- full scale rocket gas generator. The flow field in the
air flame at 101 kPa, (&) ethene-air flame at 12kPa. 127 combustion chamber was calculated with a one-
dimensional code that is typical of the rocket
read as industry. Much greater detail was used to model the
fuel oxidation and the soot formation processes.
Jachiowski's mechanism 141 for propane oxidation
was used to deal with the fuel combustion.
It is also worth noting that an obvious pressure The soot mechanism consisted of a series of
dependence does not. appear in Fig. 13. Unfortu- elementary steps that lead from acetylene and
nately, the model is very sensitive to the chosen value hydrogen to the formation of an aromatic ring via
of ~ and hence the validity of a p r i o r i calculations of the production of C4 species. Growth of the aromatic
soot volume fractions for unmeasured conditions rings proceeded by acetylene addition and hydrogen
must be called somewhat into doubt. abstraction. The formation of a four ring compound
Frenklach 138 returned to the issue of reactivity of was taken to signify the onset of sooting. The details
soot and surface growth rate in a paper that of soot growth and oxidation were not considered,
developed a detailed description of surface reac- although the latter effect was probably unimportant
tions. A Monte Carlo simulation was run to simulate under the very fuel rich conditions of a LOX/HC gas
the collisions of reactive gas species (H, HE and C2H2) generator. The predictions yielded qualitative agree-
with a soot surface. A 13 step reaction scheme was ment with observations of the threshold equivalence
presented to represent the details of hydrogen ratio for the onset of sooting. However, the experi-
abstraction, C2H2 addition and cyclization that was mental data did not lend themselves easily to detailed
handled in a rudimentary manner by the original comparisons between predictions and computations
HACA scheme [reactions (61)-(64)]. Of course, the so that the results of the paper must be considered to
rate constants had to be estimated from analogous have limited use. They certainly did not afford a good
gas phase PAH reactions with large uncertainties. In test of the soot model. More recent developments
some cases, reaction probabilities greater than unity from Frenklach's group 1°7'122 could be used to
were obtained with this procedure. The Monte Carlo supersede the early soot model that Nickerson and
simulation showed that surface growth rates Johnson used.
increased with temperature, even if reversibility of Mau8 e t al. 142 applied the soot model of Maul3
reactions (62) and (63) were included. Hence, e t al. 143 (with a close relationship to the modeling
Frenklach concluded that reversibility of the acety- approach of Frenklach and co-workers 1°7) to a
lene addition reaction and cyclization was not laminar premixed flame with a particular interest in
important for soot particles and could not explain the role of aerosol surface area and acetylene
the observed decay in soot reactivity. The growth of concentration in controlling soot growth. Experi-
PAH gaseous species was found to result in 'zig zag' ments were conducted on a flat flame burner with
edges that were inconsistent with 'arm chair' sites that mixtures of Ar, 02 and C2H2 at low pressures from 9
were assumed in the HACA model. The apparent to 18 kPa. The model used a 250 reaction mechanism
inconsistency could be accommodated by the inclu- to calculate the aliphatic chemistry. The growth
sion of an additional graphitization step 6s'139 or by reactions of benzene to small PAH were described
assuming that five membered rings were formed that in the manner of Frenklach and Warnatz 144 although
participated in subsequent growth. There was no the growth of PAH was assumed to be a fast
direct physical evidence for the latter assumption polymerization process. Consequently, the PAH were
124 I.M. Kennedy

assumed to be in a quasi-steady state and their growth model 146) were used with an activation energy
concentrations were found from algebraic expres- of 1.33 x l0 s kJ kmo1-1 . Various modified versions of
sions. Comparisons of predicted quantities with the the growth model of Frenklach and Wang 1°7 (referred
experiments showed generally excellent agreement. to as the Frenklach/Wang model) were also used. The
They concluded, in agreement with Frenklach and so-called 'modified Frenklach/Wang model' was
Wang, 1°7 that the decrease in temperature, along with described in some detail by Colket and Hall; 3 they
the depletion of gaseous radicals and active sites on included possible elimination of acetylene from the
the soot surface, could account for the observed soot radical. In addition, they separated the acetylene
decrease in soot growth rate without resort to an addition process into a reversible formation of the
additional model of soot aging. The approach of radical adduct and a cyclization process. In both
MauB et al.142"143 was applied to the prediction of the cases, the mass growth rate of soot was proportional
reactivation of soot growth in the results reported by to the soot aerosol surface area. Predictions with the
Marquardt e t a / . 145 who measured the properties of five growth models were compared to the measured
soot that was formed in 'back to back' premixed specific surface growth rates, i.e. the rate of increase in
counterflow flames of acetylene/oxygen/argon. soot mass normalized by the soot aerosol surface
Colket and Hall developed a model of soot area. In general, the models underpredicted the
formation and oxidation in a laminar, premixed measured mass growth rates by factors of two or
flame and applied it to the measurements of Harris three. It must be recalled, however, that the
and Weiner 146 and Bockhorn et al. 106 The Sandia measurements invoked assumptions in the interpreta-
burner c o d e 133 w a s used to calculate the flame structure. tion of light scattering data that may have led to
The widely used aerosol program MAEROS 147 was errors in the estimation of the aerosol surface area
coupled to the output of the Sandia code to calculate and hence in the specific growth rates. For example,
the growth, coagulation and oxidation of soot Dobbins et al. 32 estimated that the interpretation of
particles. Pre-particle gas phase chemistry was optical scattering measurements on the basis of
described with detailed kinetics. In the case of the Rayleigh theory yielded surface areas that were two
atmospheric ethylene flame of Harris and Weiner, 146 to four times smaller than the areas found from a
the mechanism of Harris et al.148 was used along with more sophisticated analysis of aggregate scattering.
an additional reaction to describe benzene formation Larger surface areas would tend to shift the
from the recombination of the propargyl radical viz., experimental results for specific surface growth rates
to lower values and hence would improve the
C3H 3 + C3H 3 *-* C6H 5 + H (73) agreement with the predictions.
A modified version of the mechanism of Miller and The apparent underestimation of the growth rates
Bowman 149 with propane reactions from Westbrook was offset in the predictions of the soot volume
and Dryer 15° was used to model the low pressure fractions by the use of benzene as the inception
flames of acetylene and propane of Bockhorn e t al.1°1 species. This may have led to an artificially high
This scheme included the C3 chemistry [reaction (63)] inception rate. Of perhaps greater concern than
that was believed to be important to benzene absolute magnitudes of the specific growth rates was
production. the trend with time in the growth rate. None of the
Colket and Hall 3 followed a rather different growth models performed particularly well in this
approach to that of Frenklach and co-workers with regard. They did not reproduce the drop in growth
regard to the issue of particle inception. They rates with time that the experimental data indicated.
considered that the calculation of the formation of This led Colket and Hall 3 to advocate the use of an
multi-ring compounds was fraught with considerable aging model in the belief that the drop in H atom
uncertainty, as well as being quite expensive compu- population did not provide an adequate account of
tationally. Furthermore, they correctly stated that the the decrease in apparent soot reactivity with time. It is
choice of the so-called inception species was rather worth noting the contrast in this conclusion with the
arbitrary. In fact, the growth of PAH is a continuum conclusions drawn by MauB et al. 142 in a numerical
and the distinction between a soot particle and a large study of a premixed flame. They determined that the
molecule is uncertain. Hence, they defined the change in the temperature and concentrations of
inception species to have the same mass as a benzene radicals accounted for the loss of soot reactivity
molecule and the inception rate to correspond to the without the need for an aging model. The contra-
rate of production of benzene itself. It was obvious diction serves to highlight the uncertainty that continues
that the nucleation flux must have been overestimated to surround fundamental aspects of soot modeling.
to some extent by this assumption and the irreversible Soot oxidation was included in the model through
kinetics of surface growth invoked too early in the the use of the Nagle and Strickland-Constable
development of the soot. formula is for 0 2 attack and the use of a collision
Particle inception was followed by growth due to efficiency of 0.13 for OH oxidation. 24 Oxidation was
acetylene addition and coalescence. Five surface not important in the Harris/Weiner flames but was
growth models were considered. The results of important in the flames that were studied by Bockhorn
Harris and Weiner (referred to as the Harris/Weiner et a l ) °l
Models of soot formation and oxidation 125

Simulations of the Harris/Weiner flames were full coupling of the soot formation with gas phase
carried out at C/O ratios of 0.8, 0.92 and 0.96. chemistry and the temperature field with radiation.
Results were presented using the growth model The C1 and C2 chemistry was described by the
obtained from Harris and Weiner, 15'146 from Frenk- reaction set of Smooke et al. 15~ with an additional
lach and Wang 1°7 and with a modified version of the 19 reactions to account for the formation benzene.
growth model of Frenklach and Wang. In general, the The dominant reaction was the propargyl recombina-
simulations overpredicted the measured soot volume tion [reaction (73)]. The governing equations for
fractions close to the flame. The use of the Harris/ momentum, energy and species were cast in a
Weiner growth model and the modified Frenklach/ similarity form and solved as a system of ordinary
Wang model achieved reasonable predictions of the differential equations. The energy equation included
trend in the ultimate soot volume fraction with the divergence of the radiation heat flux vector. With
changes in the C/O ratio. On the other hand, the the assumption of optical thinness, the radiation term
unmodified Frenklach/Wang growth model did not was written as
reproduce the correct variation in soot loading with
changes in C/O ratio. Colket and Hall 3 attributed N N
dqR
their lack of success with the Frenklach/Wang (FW) - Cry T 5 + 41r ~ ~_, czijpilij (74)
-dfy i=I j=1
model 1°7 to the observation that 'The F W model has
little or no dependence on acetylene pressure (under in which C was a parameter that depended on the
the Harris and Weiner flame conditions), and cannot optical properties of the soot. Gas radiation was
therefore reproduce the relative stoichiometric depen- calculated by the second term in Eq. (74) in which the
dence with this inception model'. Colket and Hall first summation in i was over all N species with
went on to argue that the relative failure of the concentrations p; and the second summation in j was
Frenklach/Wang growth model to account for the C/ over the active radiation bands of species i. Values of
O ratio effect could not be ascribed entirely to their C and the gas phase integrated band radiation
inception model. They believed that a growth model intensities were obtained from Grosshandler.152
with a greater sensitivity to acetylene partial pressures The approach of Colket and Hall 3 to the issue of
was required; the impact of their inception model, particle inception was extended by Hall et al. 4 They
based on benzene, remains a contentious issue within included reactions of benzene with acetylene and of
the soot modeling community. Furthermore, they benzene with the phenyl radical to yield CloH7 • and
partly attributed their apparent lack of success with Cl4H10, respectively. The inception rate was taken to
this model, relative to the success achieved by be the sum of the rates of production of these species.
Frenklach and Wang, ~°7 to differences in the predic- Although this approach appears to be more realistic
tions of benzene concentrations. A comparison of the than assuming benzene to be the sole inception
numerical predictions with the experimental data of species, the authors were nevertheless faced with the
Bockhorn et al.l°l showed that the various versions of necessity of increasing the inception rate by an
the Frenklach/Wang growth models overpredicted arbitrary factor of 8. Surface growth and oxidation
the soot volume fractions high above the burner by followed essentially the same routes as set out by
about a factor of two. The predictions using the Colket and Hall. 3 MAEROS was used again to
Harris/Weiner growth model overpredicted the soot calculate the development of the aerosol size
volume fractions above the burner by about a factor distribution.
of 40. Coiket and Hall 3 attributed the failure of the The authors demonstrated the sensitivity of soot
Harris/Weiner model to its extrapolation to the high formation and oxidation to energy loss and tempera-
temperature conditions of the Bockhorn acetylene ture reduction by radiation. They also showed the
flame with the same activation energy (1.33 x 105 kJ potential importance of 'scrubbing' of gas phase
kmol -~) that was obtained from lower temperature reactants by the soot aerosol, indicating the impor-
measurements. They claimed that the use of a 'steric tant coupling between the aerosol phase and the gas
factor' of 0.1 with the Harris/Weiner model yielded phase. The evaluation of different soot growth models
very good results for the Bockhorn flame. However, was of particular interest. Hall et al. presented a
this was simply a curve fitting exercise with little comparison of the Harris/Weiner, the Frenklach/
justification, although the same criticism could be Wang and the modified Frenklach models for the
leveled at any model that invokes the use of an counter flow flame. A comparison of the models with
empirical steric factor. The results of Colket and Hall 3 the calculated oxidation rate is shown in Fig. 14.
in the premixed flame served to point out the danger It is apparent that the growth models yielded quite
in extrapolating an apparently successful model ( the different results. Hall et al. 4 pointed out that the two
Harris/Weiner model for ethylene flames ) to other Frenklach models did not yield any net growth rate;
conditions (such as the acetylene flames of Bockhorn the peak in the Frenklach/Wang model was close to
et al)°l). the flame itself where OH and 02 concentrations were
Hall et al. 4 applied essentially the same model to the relatively high. The oxidation rate exceeded the
calculation of soot formation in a counter flow growth rates in this case. The Harris/Weiner growth
diffusion flame of methane and air. They included model succeeded in providing a positive net growth
126 I.M. Kennedy

4 5

-,= 2-
"o
E
~0-
0
"

i1 °
21
I.k
X -2" ID

•m . 4 . O

t/)
0
-8 I I I ' ' ' I ' ' ' t ' ' ' I ' ' ' I ' '

0 0.2 0.4 0.6 0.8 1 1.2 0 10 20 30 40 50 60


Normalized c o o r d i n a t e Axial Distance (ram)
Fig. 14. Calculated specific surface growth and oxidation Fig. 15. Predictions of soot volume fractions in a laminar
rates in a counter flow methane-air flame with three growth acetylene-N2 diffusion flame compared to the measurements of
models. (I-q) Harris/Weiner growth model, (O) Frenklach/ Gomez et al. TM (from Balthasar et a/.153~. (0) Measurements,TM
Wang growth model, (<>) modified Frenklach model, (x) (--) predictions.153
oxidation rate. The normalized coordinate in a direction
normal to the flame is 1 at the flame front and is 0 at the
stagnation plane. From Hall et al. 4 An equation for the mass fraction of soot in the
laminar diffusion flame was written as
because it was distributed over a wider region on the
fuel side of the flame. Given the inherent uncertainty
in the growth rates, a series of numerical experiments
--
o
=
o r ..drsl a[ lot]
0.54r/7 ~x~xj
were conducted that used growth rates enhanced by
powers of 2. The results showed that the sensitivity to + [~t (pJ~)] (75)
the inception rate decreased as growth rates increased,
providing some confirmation of the arguments and wherein the first term on the right hand side accounts
approach advocated by Kennedy e t aL 2 wherein for soot diffusion, the second accounts for thermo-
particle inception was ignored in a moderately phoresis and the final term is the net source term that
sooting ethylene diffusion flame. is derived from the flamelet library. Because the
The incorporation of detailed soot models into source term for soot volume fraction that appeared in
multi-dimensional flow codes, particularly with Eq. (75) required information about other moments
turbulence, is not feasible at present. However, of the aerosol size distribution, such as the aerosol
Balthasar e t al. 153 have attempted to incorporate surface area, the flamelet calculations required
elements of the detailed kinetics approach into a estimation of the rate of formation of at least two
simpler flamelet model for a laminar, acetylene air moments of the distribution, the first moment being
diffusion flame that was investigated experimentally the soot volume fraction itself. Fractional moments
by Gomez e t a l ) 54 The notion of flamelet modeling such as the soot aerosol surface area were found by
was retained, i.e. the thermochemical state could be interpolation. This may have introduced unknown
specified by the local mixture fraction and scalar errors into the procedure.
dissipation rate. Soot volume fractions are not well Predicted soot volume fractions were compared
correlated by these variables and hence it is necessary with measurements (Fig. 15). Although the measure-
to solve rate equations for the soot volume fraction. ments and predictions disagreed by as much as a
The flamelet method has been used for the gas phase factor of two low in the flame, a slight axial shift in the
thermochemistry by Moss and co-workers 77'78'sl who predicted profile would improve the agreement
developed rate equations for soot particle number significantly. It was possible that small errors in the
density and mass fraction. The parameters in these velocity field could contribute to such a shift.
equations included source and sink terms with Accurate predictions of soot volume fraction require
appropriate temperature dependence. The para- an accurate solution for the velocity, temperature and
meters were determined by comparison, or 'calibra- species fields, all of which are coupled through density
tion', against experiments. Hence, discussion of the and the impact of buoyancy forces. The approach
modeling of Moss e t al. was included under the that has been taken by Balthasar e t aL is3 suggests a
section devoted to semi-empirial methods. Balthasar potentially useful means for incorporating detailed
e t al. is3 departed from this approach by finding the kinetics of PAH and soot formation into a turbulent
source term for the soot mass fraction from a detailed flame calculation although the method cannot handle
calculation of soot formation and oxidation in a strongly sooting flames in which the presence of soot
counterflow diffusion flame. The detailed model can alter the flame chemistry and heat release profile.
followed the method of Frenklach and Wang TM and Soot can be a major sink for important species such as
its subsequent modificationsJ 4a OH and acetylene and it may generate significant
Models of soot formation and oxidation 127

Table 3. Summary of modeling approaches

Reference Variables solved Fuel Chemistry model Application


Empirical models
Hudson and Concentration of -- -- Pyrolysis in isothermal
Heicklen49 carbon clusters flow reactor
Edelman and Molar soot Kerosene Arrhenius expressions Gas turbine engine
Harsha 46 concentration for soot formation
and oxidation
Khan et al. 4t Exhaust soot Diesel - - Diesel engine
concentration
Mehta and Das 43 ]Exhaust soot Diesel Diesel engine
concentration
Rizk and Smoke number/soot Aviation turbine Gas turbine engine
Mongia 47'48 mass concentration fuel

Semi-empirical models
Fairweather et al. 72 Soot mass fraction Natural gas Flamelet library based Turbulent diffusion
and particle number on detailed chemistry flame
density
Fairweather et al. 73 Soot mass fraction Propane Flamelet library based Turbulent diffusion
and particle number on detailed chemistry flame
density
Honnery and Soot mass fraction Ethane and ethylene Equilibrium for gas Laminar diffusion
Kent ~~6 phase; empirical flames
rates for soot
formation and
oxidation
Jensen 155 Concentrations of Isopropyl nitrate 12 step gas phase Rocket motor
15 sizes of carbon mechanism
clusters
Kennedy et al. 2 Soot volume fraction Ethylene-air Constrained Laminar ethylene/air
equilibrium (scalars diffusion flame (1 atm)
function of mixture
fraction and enthalpy-
radiation heat loss
included)
Kennedy et al. i10 Soot mass fraction Ethylene-air Short mechanism for Laminar ethylene/air
and particle number C1 and C2 chemistry diffusion flame (1 atm)
density
Kollmann et al. 75 Mean soot volume Ethylene-air Constrained Turbulent diffusion
fraction equilibrium (scalars flame (1 atm)
function of mixture
fraction and
enthalpy-radiation
heat loss included)
Kouremenos et al. 63 ]Massof soot with Diesel Equilibrium Diesel engine
model of
Hiroyasu 64
Kyriakides et al. 42 ]Nuclei and soot Diesel Direct injection diesel
particle number engines
densities (particle
mass assumed)
Leung et al. 76 Soot mass fraction C2H4/O2-N2 111 step C1-C3 Counterflow laminar
and number density hydrocarbon diffusion flame
mechanism
Makel and Soot volume fraction Ethylene-air Equilibrium Inverse laminar diffusion
Kennedy 1°° flame (air into fuel)
Moss et al. 77 Soot volume C2H4-O2-Ar Flamelet library Laminar diffusion flame
tYaetion and particle
number density
Said et al. ss Soot precursor mass Ethylene-air Single step fuel Laminar and turbulent
fraction; soot mass oxidation diffusion flame
fraction
Continued overleaf
128 I. M. Kennedy

Table 3 continued

Reference Variables solved Fuel Chemistry model Application


Syed et al. 7s Soot volume Methane/air Flamelet library Turbulent diffusion
fraction and particle flames
number density
Tesner et al. 52 Concentration of Acetylene/air Laminar diffusion flame
radical nuclei and
soot particles
Young and MOSS98 Soot particle number Ethylene-air Fiamelet library Turbulent ethylene
density and volume for density, diffusion flame (1 atm)
fraction temperature and
soot precursor
concentrations

Detailed models
Balthasar et a1.153 Soot volume fraction 32% acetylene- Flamelet model for Laminar diffusion flame
68% N2 in air thermoebemistry as
well as source terms
for soot volume
fraction
Colket and Hall 3 Species and soot C2H4-O2-Ar Detailed kinetics up Laminar, premixed
mass fractions (1 atm); C2H2 to benzene; various flames
(12 kPa); C3H8 soot growth models
(15 kPa)
Frenklach et al. 7 Soot yield 1.09% acetylene Detailed kinetics of Shock tube
-Ar acetylene pyrolysis;
PAH growth
Frenklach and Species mass C2H2-O2-Ar Detailed kinetics of Laminar, premixed
Wang 1°7 fractions; soot (12kPa); acetylene pyrolysis; flame
number, size and C2H4-O2-Ar PAH growth
volume fraction (101kPa)
Frenklach and Species mass CEH2-O2-Ar Detailed kinetics of Laminar, premixed
Wang 127 fractions; soot (12kPa) acetylene pyrolysis; flame
number, size and PAH growth
volume fraction
Hall et al. 4 Species and soot Methane-air 102 reactions for Laminar, counterflow
mass fractions aliphatic chemistry diffusion flames (1 atm)
and benzene
formation
Kazakov et al. 136 Species mass Ethylene-air Detailed gas phase Laminar, premixed
fractions; soot kinetics; Frenklach- flames (10 bar)
volume fractions Wang for PAH and
soot
Lindstedt s3 Soot mass fraction C2H4-O2-N2; 292 step kinetics Laminar counterflow
and particle number methane-air and co-flow flames
density
Markatou et aL s Mole fractions of Ethane, ethylene Detailed kinetics for Laminar, premixed
gas phase species and acetylene-air aliphatics and PAH flames (1 atm)
and soot; critical
equivalence ratio
MauB eta/. 143 Gas phase mass Acetylene-air-O2 Detailed gas phase Laminar, premixed
fractions; moments kinetics; modified flame; post flame soot
of soot aerosol Frenklach-Wang directed against
distribution; soot PAH growth and counterflow of pure
volume fraction soot growth oxygen (1 atm)
MauB et al. j42 Gas phase mass C2H2-O2-Ar Detailed gas phase Laminar, premixed
fractions; moments kinetics; modified flames at pressures of 9-
of soot aerosol Frenklach-Wang 18kPa
distribution; soot PAH growth and
volume fraction soot growth
Yoshihara et al.12s Exhaust soot mass 88% CH4-6% Reduced mechanism Direct injection diesel
fraction C2H6-6% C3Hs for fuel/air reaction; engine
Frenklach and Wang
for soot
Models of soot formation and oxidation 129

amounts of CO or other species as it is oxidized. The Nonetheless, advances have been made in our
zone of heat release in the flame may be altered. understanding of the underlying chemical and
Important interactions of soot with the flame physical processes of soot formation. Modeling has
structure are not handled at present by the flamelet been useful in determining the areas in which we are
approach. currently still uncertain. More information is required
about some fundamental aspects of soot processes
such as the low temperature oxidation of soot by
4. CONCLUDINGREMARKS oxygen in post flame regions, the potential necessity
of including another step for graphitization of liquid
Modeling of complex phenomena such as chemi- soot precursors, the details of OH oxidation of soot,
cally reacting flows or turbulence requires approx- the inclusion of the results of fractal analysis for soot
imations that can be formulated at varying degrees of particle morphology, accurate modeling of radiative
sophistication. For example, turbulence can be heat loss in order to capture flame temperatures
modeled very simply with a mixing length approach, correctly, etc. At present, semi-empirical models of
with somewhat more difficulty by a k-e model or soot formation are useful design tools for a restricted
higher order closure, with a Large Eddy Simulation or range of combustion conditions; detailed models of
finally, with a Direct Numerical Simulation. The soot formation hold out the prospect of predicting
complexity of the flow, and the Reynolds number to soot formation over an unrestricted range of fuels,
which these methods can be applied successfully, is temperatures and pressures. The challenge will be to
limited by current computational power. However, in reduce them to a scope that can be handled by codes
the case of Direct Numerical Simulations, the accuracy of that treat turbulent reacting flow.
the solution is not compromised by uncertainties in the In the course of reviewing the soot models that were
underlying Navier-Stokes equations. available, it became clear that in most cases authors
Modeling of soot formation in flames bears some had chosen to consider unique experimental condi-
similarities to turbulence modeling in that a hierarchy tions that differed from others in the literature. In
of methods may be applied. A summary of the extant many cases, models were calibrated with and
methods is provided in Table 3. The simplest compared against measurements by the authors
approach is that taken by engineers in the engine themselves. All the models contained either implicit
community who adopt empirical correlations of soot or explicit parameters that could be adjusted to
production as a function of engine operating condi- enhance agreement with experiments. Good agree-
tions. As with all correlations, the method is only ment with experimental data was achieved in almost
useful as an extrapolation for conditions close to all cases after some adjustment. Hence, it was very
those under which the original data were first difficult to evaluate the inherent superiority of any
obtained. Somewhat more sophisticated numerical particular approach. It is apparent that it may be
methods solve rate equations for the formation of worthwhile to emulate the turbulent reacting flow
soot nuclei, soot particle growth and oxidation with community in attempting to develop a set of
simple models for these processes that are derived universally acceptable experimental data sets.
from experiments. Again, this approach can only be Models could then be required to predict a subset of
used for conditions under which the original nuclea- these data or the full range over a variety of flame
tion, growth and oxidation results were obtained or types, fuels and pressures, depending on their claim to
calibrated. Recent years have seen a surge of interest generality. Only then would it be possible to evaluate
in attacking soot modeling with very detailed and compare the efficacy of models.
descriptions of the soot physics and chemistry. An
analogy might be drawn in this case with the Acknowledgements--The input of Dr M. Colket of UTRC to
application of Direct Numerical Simulations in this paper is gratefully acknowledged. Conversations with
Dr D. Honnery of Monash University, Australia during the
turbulence modeling. However, much greater reliance early stages of planning the paper were also most helpful.
can be placed in the Navier-Stokes equations of fluid The support of the NIEHS Superfund Basic Research
flow than in the detailed chemical schemes that are program P42ES04699 for some of the author's research
used to describe soot formation because very few into particulate formation in flames is appreciated.
elementary rate constants have been measured.
It is remarkable that all the methods that have been
REFERENCES
reviewed in this survey have demonstrated the
potential for accurate predictions of soot formation
1. Haynes, B. S. and Wagner, H. G., Prog. Energy
over a wide range of conditions. A healthy skepticism Combust. Sci. 7, 229-273 (1981).
is, of course, in order. All the results were compared 2. Kennedy, I. M., Kollmann, W. and Chen, J.-Y.,
to measurements after some adjustment of para- Combust. Flame 81, 73-85 (1990).
meters, whether it be an activation energy of soot 3. Colket, M. B. and Hall, R. J., Successes and
uncertainties in modeling soot formation in laminar,
surface growth or a steric parameter or some other premixed flames, in Soot Formation in Combustion, H.
less obvious modeling variable. To some extent, our Bockhorn (Ed.), pp. 442-468, Springer-Verlag, Berlin
models of soot have been adjusted to fit the data. (1994).
130 I.M. Kennedy

4. Hail, R. J., Smooke, M. D. and Colket, M. B., 33. Charalampopoulos, T. T., Chang, H. and Stagg, B.,
Predictions of soot dynamics in opposed jet diffusion Fuel 68, 1173-1179 (1989).
flames, Special Issue o f C S T for L Glussman, F. L. 34. Dalzell, W. H. and Sarofim, A. F., J. Heat Trnsfr. -
Dryer and R. Sawyer (Eds), Gordon & Breach (1996). Trans. A S M E 91, 100-104 (1969).
5. Markatou, P., Wang, H. and Frenklach, M., Combust. 35. Sivathanu, Y. R. and Gore, J. P., Combust. Flame 97,
Flame 93, 467-482 (1993). 161-172 (1994).
6. Stein, S. E., J. Phys. Chem. 82, 566-571 (1978). 36. Olson, D. B., Pickens, J. C. and Gill, R. J., Combust.
7. Frenklach, M., Clary, D. W., Gardiner, J. and Stein, S. E., Flume 62, 43-60 (1985).
Twentieth Symposium (International) on Combustion, pp. 37. Takahashi, F. and Giassman, I., Combust. Sci. Tech.
887-901, The Combustion Institute, Pittsburgh, PA 37, 1-19 (1984).
(1984). 38. Calcote, H. F. and Manos, D. M., Combust. Flame 49,
8. Howard, J. B., Twenty-Third Symposium (Inter- 289 (1983).
national) on Combustion, pp. 1107-1127, The Com- 39. Gill, R. J. and OIson, D. B., Combust. Sci. Tech. 40,
bustion Institute, Pittsburgh, PA (1990). 307-315 (1984).
9. Calcote, H. F. and Gill, R. J., Comparison of the ionic 40. Harris, M. M., King, G. B. and Laurendau, N. M.,
mechanism of soot formation with a free radical Combust. Flume 64, 99-112 (1986).
mechanism, Soot Formation in Combustion, H. Bockhorn 41. Khan, I. M., Greeves, G. and Probert, D. M., Air
(Ed.), pp. 471-484, Springer-Verlag, Berlin (1994). Pollution Control in Transport Engines, Vol. C142/71,
10. Hidy, G. M. and Brock, J. R., The Dynamics o f pp. 205-217, The Institution of Mechanical Engineers,
Aerocolloidal Systems, Pergamon Press, Oxford (1970). London (1971).
11. Harris, S. and Kennedy, I. M., Combust. Sci. Tech. 59, 42. Kyriakides, S. C., Dent, J. C. and Mehta, P. S., SAE
443-454 (1988). Technical Paper Series 860330 0986).
12. Kennedy, I. M. and Harris, S. J., J. Coll. Inter. Sci. 130, 43. Mehta, P. S. and Das, S., Fuel 71, 689-692 (1992).
489-497 (1989). 44. Lefebvre, A., Int. J. Heat Muss Trnsfr. 27, 1493-1510
13. Lai, F. S., Friedlander, S. K., Pith, J. and Hidy, G. M., (1984).
J. Coll. Inter. Sci. 39, 395-405 (1972). 45. Khan, I. M. and Greeves, G., International Seminar,
14. Harris, S. J. and Weiner, A. M., Combust. Sci. Tech. Trogir, Yugoslavia (1973).
38, 75-87 (1984). 46. Edelman, R. B. and Harsha, P. T., Prog. Energy
15. Harris, S. J. and Weiner, A. M., Combust. Sci. Tech. Combust. Sci. 4, 1-62 (1978).
32, 267-275 (1983). 47. Rizk, N. K. and Mongia, H. C., J. Propuls. 6, 660-667
16. Wieschnowsky, U., Bockhorn, H. and Fetting, F., (1990).
Twenty-Second Symposium (International) on Combus- 48. Rizk, N. K. and Mongia, H. C., J. Propuls. 7, 445-451
tion, pp. 343-352, The Combustion Institute, Pitts- (1991).
burgh, PA (1988). 49. Hudson, J. L. and Heicklen, J., Carbon 6, 405-418
17. Lee, K. B., Thring, M. W. and Beer, J. M., Combust. (1968).
Flame 6, 137-145 (1962). 50. Heicklen, J., Hudson, J. L. and Armi, L., Carbon 7,
18. Nagle, J. and Strickland-Constable, R. F., Fifth 365-372 (1969).
Carbon Conference, Voi. 1, pp. 154-164, Pergamon, 5t. Smith, G., Combust. Flame 48, 265-272 (1982).
Oxford (1962). 52. Tesner, P. A., Snegiriova, T. D. and Knorre, V. G.,
19. Radcliffe, S. A. and Appleton, J. P., Combust. Sci. Combust. Flame 17, 253-260 (1971).
Tech. 4, 171-175 (1971). 53. Tesner, P. A., Tsygankova, E. I., Guilazetdinov, E. I.,
20. Najjar, Y. S. H. and Goodger, E. M., Fuel 60, 987-990 Zuyev, V. P. and Loshakova, G. V., Combust. Flume
(1981). 17, 279 (1971).
21. Fenimore, C. P. and Jones, G. W., Combust. Flame 13, 54. Magnussen, B. F., Hjertager, B. H., Olsen, J. G. and
303-310 (1969). Bhaduri, D., Seventeenth Symposium (International)
22. Bradley, D., Dixon-Lewis, G., Habik, S. E.-D. and on Combustion, pp. 1383-1393, The Combustion
Mushi, E. M. J., Twentieth Symposium (International) Institute, Pittsburgh, PA (1978).
on Combustion, pp. 931-940, The Combustion Insti- 55. Gordiets, B. F., Shelepin, L. A. and Shmotkin, Y. S.,
tute, Pittsburgh, PA (1984). Fizika Goreniya i Vzryva 18, 71-76 (1982).
23. Puri, R., Santoro, R. J. and Smyth, K. C., Combust. 56. Surovikin, V. F., Khimiya Tverdogo Topliva 10, 111-122
Flame 97, 125-144 (1994). (1976).
24. Neoh, K. G., Howard, J. B. and Sarofim, A. F., 57. Mehta, P. S., Gupta, A. K. and Gupta, C. P., SAE
Particulate Carbon Formation During Combustion, pp. Technical Paper Series 881251 (1988).
261-282, Plenum, New York (1981). 58. Ermolin, N. E., Korobeinichev, O. P., Tereshchenko,
25. Roth, P., Brandt, O. and Gersum, S. V., Twenty-Third A. G. and Fomin, V. M., Combust., Explosion Shock
Symposhan (International) on Combustion, pp. 1485- Waves 18, 36-38 (1982).
1491, The Combustion Institute, Pittsburgh, PA (1990). 59. Brown, A. J. and Heywood, J. B., Combust. Sci. Tech.
26. Dobbins, R. A., MulhoUand, G. W. and Bryner, N. P., 58, 195-207 (1988).
Atmos. Environ. 28, 889-897 (1994). 60. Curl, R. L., A.I.Ch.E.J. 9, 175-181 (1963).
27. Koylu, U. O. and Faeth, G. M., J. Heat Trnsfr. - Trans. 61. Pratt, D. T., Prog. Energy Combust. Sci. 1, 73 (1975).
A S M E 115, 409-417 (1993). 62. Wang, T. S., Matula, R. A. and Farmer, R. C.,
28. Koylu, U. O. and Faeth, G. M., J. Heat Trnsfr. - Trans. Twentieth Symposium (International) on Combustion,
A S M E 116, 152-159 (1994). pp. 1149, The Combustion Institute, Pittsburgh, PA
29. Koylu, U. O. and Faeth, G. M., J. Heat Trnsfr. - Trans. (1981).
A S M E 116, 971-979 (1994). 63. Kouremenos, D. A., Rakopoulos, C. D., Hountalas,
30. Koylu, U. O., Faeth, G. M., Farias, T. L. and D. and Kotsiopoulos, P., Forschung im Ingenieurwesen
Carvalho, M. G., Combust. Flame 100, 621-633 (1995). 56, 22-32 (1990).
31. Choi, M. Y., Mulholland, G. W., Hamins, A. and 64. Hiroyasu, H., Kadota, T. and Arai, M., Bull. J S M E
Kashiwagi, T., Combust. Flame 102, 161-169 (1995). 26, 569-575 (1983).
32. Dobbins, R. A., Santoro, R. J. and Semerjian, H. G., 65. Jensen, D. E., Proc. Royal Soc. Lond. A 338, 375-396
Twenty-Third Symposium (International) on Combus- (1974).
tion, pp. 1525-1532, The Combustion Institute, 66. Graham, S. C., Proc. RoyaISoc. Lond. A 377, 119-145
Pittsburgh, PA (1990). (1981).
Models of soot formation and oxidation 131

67. Graham, S. C., Howard, J. B. and Rosenfeld, J. L. J., 93. Saito, K., Williams, F. A. and Gordon, A. S., A S M E
Proc. Royal Soc. London A 344, 259-285 (1975). Transaction Heat Transfer 108, 640-648 (1985).
68. Dobbins, R. A., Fletcher, R. A. and Lu, W., Combust. 94. Alizadeh, S. and Moss, J. B., AGARD Conference
Flame 108, 301-309 (1995). Proceedings 536. Fuels and Combustion Technology for
69. Mulholland, G. W., Global soot growth model, Fire Advanced Aircraft Engines, 7-1 to 7-20, AGARD
Safety Science, C. E. Grant and P. J. Pagni (Eds), pp. Fiuggi, Italy (1993).
709-718, Hemisphere Publishing Corp., Washington 95. Heitor, M. V. and Whitelaw, J. H., Combust. Flame 64,
(1986). 1-32 (1986).
70. Jones, J. M. and Rosenfeld, J. L. J., Combust. Flume 96. Young, K. J., Stewart, C. D., Syed, K. J. and Moss, J. B.,
19, 427-434 (1972). Proceedings of the Tenth International Symposium on
71. Williams, F. A., Combustion Theory, Benjamin/Cum- Air Breathing Engines (1991).
mings Menlo Park, CA (1985). 97. Young, K. J., Stewart, C. D. and Moss, J. B., Twenty-
72. Fairweather, M., Jones, W. P. and Lindstedt, R. P., Fifth Symposium (International) on Combustion, The
Combust. Flume 89, 45-64 (1992). Combustion Institute, Pittsburgh, PA (1994).
73. Fairweather, M., Jones, W. P., Ledin, H. S. and 98. Young, K. J. and Moss, J. B., Combust. Sci. Tech. 105,
Lindstedt, R. P., Twenty-Fourth Symposium (Inter- 33-53 (1995).
national) on Combustion, pp. 1067-1074, The Com- 99. Lindstedt, P., A simple reaction mechanism for soot
bustion Institute, Pittsburgh, PA (1992). formation in non-premixed flames, Aerothermody-
74. Kent, J. H. and Honnery, D., Combust. Sci. Tech. 54, namics in Combustors, R. S. Lee, J. H. Whitelaw and
383-397 (1987). T. S. Wung (Eds), pp. 145-156, Springer-Verlag,
75. KoUmann, W., Kennedy, I. M., Metternich, M. and Berlin (1992).
Chen, J.-Y., Application of a soot model to a turbulent 100. Makel, D. B. and Kennedy, I. M., Combust. Sci. Tech.
ethylene diffusion flame, Soot Formation in Combus- 97, 303-314 (1994).
tion, H. Bockhorn (Ed.), pp. 503-526, Springer-Verlag, 101. Bockhorn, H., Moser, A., Wenz, H. and Warnatz, J.,
Berlin (1994). Nineteenth Symposium (International) on Combustion,
76. Leung, K. M., Lindstedt, R. P. and Jones, W. P., pp. 197-209, The Combustion Institute, Pittsburgh,
Combust. Flame 87, 289-305 (1991). PA (1982).
77. Moss, J. B., Stewart, C. D. and Young, K. J.,Combust. 102. Garo, A., Prado, G. and Lahaye, J., Combust. Flame
Flame 101, 491-500 (1995). 79, 226-233 (1990).
78. Syed, K. J., Stewart, C. D. and Moss, J. B., Twenty- 103. Jones, W. P. and Lindstedt, R. P., Combust. Flame 73,
Third Symposium (International) on Combustion, pp. 233-249 (1988).
1533-1541, The Combustion Institute, Pittsburgh, PA 104. Vandsburger, U., Kennedy, I. M. and Glassman, I.,
(1990). Combust. Sci. Teeh. 39, 263-285 (1984).
79. Villasenor, R. and Kennedy, I. M., Twenty-Fourth 105. Nishida, O. and Mukohara, S., Combust. Flame 47,
Symposium (International) on Combustion, pp. 269-279 (1982).
1023-1030, The Combustion Institute, Pittsburgh, 106. Bockhorn, H., Fetting, F. and Wenz, H. W., Ber.
PA (1992). Bunsen Gesell. Phys. Chem. 87, 1067-1073 (1983).
80. Gore, J. P. and Faeth, G. M., Twenty-First Symposium 107. Frenklach, M. and Wang, H., Twenty-Third Sympo-
(International) on Combustion, pp. 1521-1531, The sium (International) on Combustion, pp. 1559-1566,
Combustion Institute, Pittsburgh, PA (1986). The Combustion Institute, Pittsburgh, PA (1990).
81. Stewart, C. D., Syed, K. J. and Moss, J. B., Combust. 108. Gore, J. P., Ph.D., The Pennsylvania State University,
Sci. Tech. 75, 211-226 (1991). University Park (1986).
82. Delichatsios, M. A., Combust. Sci. Tech. 108, 283-298 109. Purl, R., Santoro, R. J. and Smyth, K. C., Combust.
(1994). Flame 102, 226-228 (1995).
83. Lindstedt, P. R., Simplified soot nucleation and surface 110. Kennedy, I. M., Rapp, D. R., Santoro, R. J. and Yam,
growth steps for non-premixed flames, Soot Formation C., Combust. Flame 107, 368-382 (1996).
in Combustion, H. Bockhorn (Ed.) , pp. 417-439, 111. Miller, J. H., Honnery, D. R. and Kent, J. H., Twenty-
Springer-Verlag, Berlin (1994). Fourth Symposium (International) on Combustion, pp.
84. Kent, J. H. and Wagner, H. G., Combust. Flame 47, 1031-1039, The Combustion Institute, Pittsburgh, PA
53--65 (1982). (1992).
85. Jones, W. P., Models for turbulent flows with variable 112. Smyth, K. C., Miller, J. H., Dorfman, R. C., Mallard,
density and combustion, Prediction Methods for W. G. and Santoro, R. J., Combust. Flame 62, 157-181
Turbulent Flows, W. Kollmann (Ed.), pp. 379-421, (1985).
Hemisphere, Washington (1980). 113. He, Y. Z., Mallard, W. G. and Tsang, W., J. Phys.
86. Kesten, A. S., Sangiovanni, J. J. and Goldberg, P., Chem. 92, 2196-2201 (1988).
ASME 79-GT-lsr-12 (1979). 114. Kent, J. H. and Honnery, D. R., Soot mass growth in
87. Wersborg, B. L., Yeung, A. C. and Howard, J. B., laminar diffusion flames--parametric modelling, Soot
Fifteenth Symposium (International) on Combustion, Formation in Combustion, H. Bockhorn (Ed.), pp.
pp. 1439-1448, The Combustion Insitute, Pittsburgh, 199-218, Springer - Verlag, Berlin (1994).
PA (1975). 115. Kent, J. H. and Honnery, D. R., Combust. Flame 79,
88. Said, R., Garo, A. and Borghi, R., Combust. Flame 287-298 (1990).
108, 71-86 (1997). 116. Honnery, D. R. and Kent, J. H., Twenty-Fourth
89. Kent, J. H., Jander, H. and Wagner, H. G., Eighteenth Symposium (International) on Combustion, pp.
Symposium (International) on Combustion, pp. 1041-1047, The Combustion Institute, Pittsburgh,
I 117-1126, The Combustion Institute (1980). PA (1992).
90. Santoro, R. J., Yeh, T. T., Horvath, J. J. and 117. Honnery, D. R., Tappe, M. and Kent, J. H., Combust.
Semerjian, H. G., Combust. Sci. Tech. 53, 89-115 (1987). Sci. Tech. 83, 305-321 (1992).
91. Sivathanu, Y. R. and Faeth, G. M., Combust. Flame 118. Kaplan, C. R., Back, S. W., Oran, E. S. and EIIzey, J. L.,
82, 211-230 (1990). Combust. Flame 96, 1-21 (1994).
92. Moss, J. B., Stewart, C. D. and Syed, K. J., Twenty- 119. Chandrasekhar, S., Radiative Transfer, Dover, New
Second Symposium (International) on Combustion, pp. York (1960).
413-423, The Combustion Institute, Pittsburgh, PA 120. Talbot, L., Cheng, R. K., Schefer, R. W. and Willis, D. R.,
(1988). J. Fluid Mech. 101, 737-758 (1980).
132 I.M. Kennedy

121. Smooke, M. D., Purl, I. K. and Seshadri, K., Twenty- 139. D'Anna, A., D'Alessio, A. and Minutolo, P., Spectro-
First Symposium (International) on Combustion, pp. scopic and chemical characterization of soot inception
1783-1792, The Combustion Institute, Pittsburgh, PA processes in premixed laminar flames at atmosphereic
(1986). pressure, Soot Formation in Combustion, H. Bockhorn
122. Frenklach, M., Wang, H. and Rabinowitz, M. J., Prog. (Ed.), pp. 83-101, Springer-Verlag, Berlin (1994).
Energy Combust. Sci. 18, 47-73 (1992). 140. Nickerson, G. R. and Johnson, C. W., AIAA/SAE/
123. Smooke, M. D. and Giovangigli,V., Impact Comp. Sci. ASME/ASEE 28th Joint Propulsion Conference, Paper
Engr. 4, 46-79 (1992). AIAA 92-3391, Nashville, TN (1992).
124. Purl, R., Moser, M., Santoro, R. J. and Smyth, K. C., 141. Jachimowski, C. J., Combust. Flame 55, 213-224
Twenty-Fourth Symposium (International) on Combus- (1984).
tion, pp. 1015-1022, The Combustion Institute, 142. MauB, F., Sch~ifer, T. and Bockhorn, H., Combust.
Pittsburgh, PA (1992). Flame 99, 697-705 (1994).
125. Levendis, Y. A., Flagan, R. C. and Gavalas, G. R., 143. Maufl, F., Trilken, B., Breitbach, H. and Peters, N.,
Combust. Flame 76, 221-241 (1989). Soot formation in partially premixed diffusion flames,
126. Chan, M.-L., Moody, K. N., Mullins, J. R. and Soot Formation in Combustion, H. Bockhorn (Ed.), pp.
Williams, A., Fuel 66, 1694-1698 (1987). 325-349, Sprlnger-Vedag, Berlin (1994).
127. Frenklach, M. and Wang, H., Detailed mechanism and 144. Frenklach, M. and Warnatz, J., Combust. Sci. Tech. 51,
modeling of soot particle formation, Soot Formation in 265-283 (1987).
Combustion , H. Bockhorn (Ed.), pp. 162-190, 145. Marquardt, M., MauB, F., Jungfleisch, B., Suntz, R.
Springer-Verlag, Berlin (1994). and Bockhorn, H., Twenty-Sixth Symposium (Inter-
128. Yoshihara, Y., Kazakov, A., Wang, H. and Frenklach, national) on Combustion, pp. 2343-2350, The Com-
M., Twenty-Fifth Symposium (International) on Com- bustion Institute, Pittsburgh, PA 0996).
bustion, pp. 941-948, The Combustion Institute, 146. Harris, S. J. and Weiner, A. M., Combust. Sci. Tech.
Pittsburgh, PA (1994). 31, 155-167 (1983).
129. Tanzawa, T. and W. C. Gardiner, J., J. Phys. Chem. 84, 147. Gelbard, F., MAEROS User Manual, Sandia
236-239 (1980). SAND80-0822 (1982).
130. Gelbard, F. and Seinfeld, J. H., J. Coll. Inter. Sci. 78, 148. Harris, S. J., Weiner, A. M. and Blint, R. J., Combust.
485-501 (1980). Flame 72, 91-109 (1988).
131. Harris, S. J., Weiner, A. M. and Ashcroft, C. C., 149. Miller, J. A. and Bowman, C. T., Prog. Energy
Combust. Flame 64, 65-81 (1986). Combust. Sci. 15, 287-338 (1989).
132. Harris, M. M., King, G. B. and Laurendau, N. M., 150. Westbrook, C. K. and Dryer, F. L., Prog. Energy
Combust. Flame 67, 269-272 (1987). Combust. Sci. 10, 1-57 (1984).
133. Kee, R. J., Grcar, J. F., Smooke, M. D. and Miller, J. A., 151. Smooke, M. D.,Xu, Y.,Zurn, R.M.,Lin, P., Frank, J.H.
Sandia National Laboratory, SAND85-8240 (1985). and Long, M. B., Twenty-Fourth Symposium (Inter-
134. Harris, S. J. and Frenklach, M., J. Coll. Inter. Sci. 118, national) on Combustion, pp. 813-821, The Combus-
252-261 (1987). tion Institute, Pittsburgh, PA (1992).
135. Wang, H. and Frenklach, M., Combust. Flame 87, 152. Grosshandler, W. L., RADCAL: A Narrow-Band
365-370 (1991). Model for Radiation Calculations in a Combustion
136. Kazakov, A., Wang, H. and Frenklach, M., Combust. Environment, NIST 1402 (1993).
Flame 100, 111-120 (1995). 153. Balthasar, M., Heyl, A., Mau6, F., Schmitt, F. and
137. B6nig, M., Feldermann, C., Jander, H., Lfiers, B., Bockhorn, H., Twenty-Sixth Symposium (Interna-
Rudolph, G. and Wagner, H. G., Twenty-Third tional) on Combustion, pp. 2369-2378, The Combus-
Symposium (International) on Combustion, pp. tion Institute, Pittsburgh, PA (1996).
1581-1587, The Combustion Institute, Pittsburgh, 154. Gomez, A., Littman, M. G. and Glassman, I.,
PA (1991). Combust. Flame 70, 225 (1987).
138. Frenklach, M., Twenty-Sixth Symposium (Interna- 155. Jensen, D. E. and Wilson, A. S., Combust. Flame 25,
tional) on Combustion, pp. 2285-2294, The Combus- 43-55 (1975).
tion Institute, Pittsburgh, PA (1996).

You might also like