You are on page 1of 10

Combustion and Flame 159 (2012) 2789–2798

Contents lists available at SciVerse ScienceDirect

Combustion and Flame


j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m b u s t fl a m e

Numerical investigation of soot formation mechanisms in partially-premixed


ethylene–air co-flow flames
Victor Chernov a, Qingan Zhang a,1, Murray John Thomson a,⇑, Seth Benjamin Dworkin b
a
Mechanical and Industrial Engineering, University of Toronto, 5 King’s College Circle, Toronto, Ontario, Canada M5S 2G8
b
Mechanical and Industrial Engineering, Ryerson University, 350 Victoria Street, Toronto, Ontario, Canada M5B 2K3

a r t i c l e i n f o a b s t r a c t

Article history: Recently, an improved chemical mechanism of PAH growth was developed and tested in soot computa-
Received 31 October 2011 tions for a laminar co-flow non-premixed ethylene–air diffusion flame [Dworkin et al., Combust. Flame
Received in revised form 22 February 2012 158(9) (2011) 1682–1695]. With the intention of testing the robustness of the solution methodology
Accepted 23 February 2012
on partially-premixed systems, this work used the same algorithm as that in the study of Dworkin
Available online 1 May 2012
et al. for computations of two sets of sooting partially-premixed co-flow laminar ethylene–air flames.
The results show very good qualitative and good quantitative agreement with the experimental results
Keywords:
for soot volume fractions and soot precursors, without any changes to the parameters of the model.
Soot
Laminar flame
The soot yield was found to initially increase with decreasing primary equivalence ratios, and then to
Diffusion flame decrease for U < 24, reaching levels lower than the non-premixed case for U < 10. On the flame center-
Partially-premixed flame line, both PAH and acetylene-related processes were found to be important for soot growth. The initial
Computational combustion increase in the soot yield was linked to higher inception rates. On the wings of the flame the dominant
soot growth process was found to be HACA growth. The initial increase in the soot yield was mostly due
to higher acetylene yield leading to faster surface growth. The primary air was also found to influence the
soot oxidation process by increasing OH radicals in both the centerline and the wings region.
Ó 2012 Published by Elsevier Inc. on behalf of The Combustion Institute.

1. Introduction The laminar co-flow diffusion flame is a common configuration


that is used to study soot behavior. The most popular configuration
Reducing the environmental impact of combustion is one of the for investigation of laminar co-flow flames is the annular burner
major goals in the design of energy systems. One of the most impor- [2]. This burner allows for the studying of convection–diffusion ef-
tant combustion-generated pollutants is soot. Soot consists of small, fects on the soot behavior without the additional complexities that
sub-micron particles that contain mostly carbon. The size of the par- are introduced when the flame is turbulent. Many experimental
ticles vary according to the flame conditions. Soot poses various and computational studies have been conducted using the Santoro
risks and influences many areas; it is a health hazard, it plays a role burner by various groups [3–11].
in climate change, it reduces the lifetime of energy producing com- The partially-premixed flame is an additional step towards a
ponents, and it is an esthetic nuisance. Understanding the mecha- more complete understanding of combustion processes. In this
nisms behind the formation and behavior of these particles can case, the co-flow configuration is used, but the central tube sup-
help create less polluting- and more effective combustion systems. plies a rich mixture of fuel and air. Partial premixing is important
Soot inception, growth, and oxidation are influenced both by for several reasons. Firstly, many practical combustion applica-
the chemistry and fluid dynamics of the system [1]. Many practical tions, such as gas-turbine combustors and domestic appliances
systems feature complex geometries and the combustion is usually are partially-premixed. In most cases, the aforementioned flames
non-premixed and turbulent. Unfortunately, a detailed soot emis- are turbulent, but understanding the laminar case is an important
sion model of such systems is not feasible, both due to the high step towards understanding those flames. Secondly, in turbulent
computational cost involved, and due to the lack of a complete flames, different degrees of mixing are present. Data gathered from
understanding of the processes behind soot behavior. partially-premixed laminar flames has the potential of providing
insights for turbulent flame analysis. Thirdly, the additional com-
⇑ Corresponding author. Fax: +1 416 978 7753. plexity of the partially-premixed configuration allows for better
E-mail addresses: victor@mie.utoronto.ca (V. Chernov), aqzhang@alumni. validation of the computational methods and approaches.
utoronto.ca (Q. Zhang), thomson@mie.utoronto.ca (M.J. Thomson), seth.dworkin@ Unfortunately, the amount of works that have examined soot
ryerson.ca (S.B. Dworkin). formation in the partially-premixed configuration is limited.
1
Has moved to Canadian Natural Resources Limited, 855-2nd St. SW Calgary, AB,
Works that include soot measurement either contain a spatial
Canada T2P 4J8.

0010-2180/$ - see front matter Ó 2012 Published by Elsevier Inc. on behalf of The Combustion Institute.
doi:10.1016/j.combustflame.2012.02.023
2790 V. Chernov et al. / Combustion and Flame 159 (2012) 2789–2798

measurement of a soot volume fraction without species measure- elliptical conservation equations for mass, energy, and species mass
ments [12–14], or soot volume fraction, temperature and species fraction are solved. Soot particle dynamics are described using a
concentration on the centerline only [15]. sectional model. Soot particle mass ranges are divided logarithmi-
The number of works that included computational analysis of cally into 35 discrete sections. Soot aggregates that are composed
the soot formation in partially-premixed laminar flames is even of primary particles of equal size with a constant fractal dimension
smaller. Domenico et al. [16] calculated the soot volume fraction of 1.8, are divided similarly to 35 sections for the number of primary
in the partially-premixed Santoro configuration. The authors used particles per aggregate. Conservation equations of soot aggregate
the sectional approach for PAH growth and a two-equation soot number densities and primary particle number densities are solved.
model. The results were compared with several experimental The nucleation of a soot particle is modeled as the collision of
works, including that of McEnally and Pfefferle [15]. Only one pre- two pyrene molecules [20]. Soot surface growth is implemented
mixing ratio was shown, though the authors say that all the other using the HACA mechanism and PAH condensation. The empirical
ratios measured by McEnally and Pfefferle [15] were tested. The re- correction parameter a, which reconciles the inaccuracies in treat-
sults showed good agreement with the experiments, but their use ing sites on the soot surface as corresponding sites on gaseous PAH
of a two-equation soot model prevents detailed soot analysis. Katta molecules, is chosen to be a = 0.078, the same value as that used by
et al. [14] used a different burner for a partially-premixed flame. Dworkin et al. [23]. Radiative heat transfer by soot, H2O, CO2, and
They too used a two-equation soot model. For the non-premixed CO is calculated using the discrete-ordinates method, and a statis-
case, the soot volume fractions that were calculated were higher tical narrow-band correlated-k-based model [24].
than measured values, and for low premixing the computations Two experimental datasets are modeled. The first dataset, la-
showed agreements with the measurements. Increasing premixing beled ‘‘Dataset 1’’ in this work, is modeled after the experiments
caused a major disparity between the predicted and measured pro- of Arana et al. [13]. The second dataset, labeled ‘‘Dataset 2’’ in this
files of soot volume fraction. work, is modeled after the experiments of McEnally and Pfefferle
The common approximation that the simple two-equation soot [15]. Both are based on a Santoro laminar coaxial annular burner
model uses is that of soot as monodispersed carbon spheres. This is [2]. For both datasets, a fuel–air mixture flows from a 0.55 cm-in-
far from physical reality [16]. Soot particles are complex aggre- ner-radius tube, and the secondary air flows from a 5.41 cm-radius
gates of carbon spheres. When performing a fundamental analysis, outer tube. The computational domain extends to 12.29 cm (192
this point should be taken into account. A common approach is to control volumes) in the axial direction, and to 5.41 cm (92 control
use the sectional representation [17,18]. Earlier works have volumes) in the radial direction. A non-uniform mesh with fine res-
adopted a sectional representation for the mass of a soot spherical olution in the flame region is used. The inlet velocities are assumed
particle only [3–5,19]. A more robust approach is to use the sec- to be uniform. The inlet temperature of the fuel mixture is set to
tional model for both the sizes of the spherical particles, and for 400 K and the temperature of the secondary air is set to 300 K.
the number of the primary particles per aggregates [9–11]. The lat- The fuel temperature inlet conditions are based on the experimen-
ter approach is used in the present work. tal data of Dataset 2 [15]. Dataset 1 [13] does not provide temper-
One of the main challenges in soot modeling is describing soot ature measurements at the inlet and since the experimental
inception. Most detailed computational studies model the nucle- conditions were similar, it was decided to use the same inlet tem-
ation of a soot particle as a result of a collision of two pyrene mol- peratures for both flames. Symmetry, no-slip (a chimney encloses
ecules [20]. Therefore it is vital to have a reliable pathway to the flame), and zero-gradient conditions are enforced at the center-
pyrene creation. The under-prediction by most aforementioned line, the outer radial boundary, and the outflow boundary condi-
works of the soot volume fraction on the centerline of the flame, tions, respectively. The flame is solved using distributed-memory
where the dominant process is inception, underlines this problem. parallel processing, with the computational domain divided uni-
The most common mechanism for pyrene creation is the formly into 192 subdomains, with the boundaries of each subdo-
hydrogen-abstraction-acetylene-addition (HACA) pathway [20,21]. main perpendicular to the z-axis.
However, this pathway was deemed to be incomplete, and re- The inlet velocities and the composition of the fuel mixtures
cently, a new mechanism that includes additional pyrene path- were different for the two datasets. The fuel mixture in Dataset 1
ways was suggested [22]. Dworkin et al. [23] implemented an [13] consisted of a mixture of ethylene and air. Five flames (with
updated reaction scheme from [22] for a laminar ethylene diffusion different primary equivalence ratios U) are investigated. U is de-
flame, and obtained a significant improvement of the soot volume fined as the ratio of the primary air flowrate that is required for
fractions compared to the HACA-dominated mechanisms. complete combustion, to the actual primary flow rate. The inlet
Our main interest in this work is to examine whether this im- velocity and the composition of the fuel–air mix for different val-
proved chemical mechanism of PAH growth, coupled with the sec- ues of U are given in Table 1. The composition is given in mole frac-
tional representation, can predict the main trends of a set of tions. The secondary air velocity is ua = 8.905 cm/s. Spatial soot
partially-premixed flames. Therefore, we continue the work of volume fractions were measured. No species concentrations were
Dworkin et al. [23], and implement the mechanism that was sug- reported in [13].
gested by Slavinskaya and Frank [22] in partially-premixed flames The fuel mixture of Dataset 2 [15] consists of ethylene, air,
of ethylene and air. The computational setup is shown in Section 2. nitrogen, and argon. Six flames are investigated. The inlet velocity
Experimental works of Arana et al. [13], called ‘‘Dataset 1’’, and of and the composition of the fuel–air mix for different values of U
McEnally and Pfefferle [15], called ‘‘Dataset 2’’, were used for mod-
el validation. The computed results for soot volume fractions, and
species, for both flames, are shown in Section 3, and a discussion is Table 1
held in Section 4. The inlet velocity and the composition of the fuel–air mix for different values of U for
Dataset 1 [13].

2. Computational model and setup U uf (cm/s) X C2 H4 X O2 X N2

1 7.072 1 0 0
The computational approach is summarized here, however, for a 24 11.29 0.62647 0.07807 0.29546
more detailed description, the reader is directed to [9,10,23]. The 20 12.13 0.58292 0.08717 0.32991
10 17.19 0.4114 0.123 0.4656
axial symmetry of the burner permits use of a two-dimensional (z
5 27.32 0.2589 0.1549 0.5862
and r) coordinate system. For the gaseous phase, fully-coupled
V. Chernov et al. / Combustion and Flame 159 (2012) 2789–2798 2791

Table 2
The inlet velocity and the composition of the fuel–air mix for different values of U for
Dataset 2 [15].

U uf (cm/s) X C2 H4 X O2 X N2 X Ar

1 16.7 0.3025 0 0.6875 0.01


24 19.69 0.25662 0.03169 0.70317 0.00852
12 22.67 0.22283 0.05504 0.71474 0.00739
6 28.87 0.17498 0.0881 0.73111 0.00581
4 26.13 0.14499 0.10882 0.74138 0.00481
3 30.78 0.12309 0.12395 0.74887 0.00409

are given in Table 2. The composition is given in mole fractions.


The secondary air inlet is partially obstructed by a copper ring that
extends from a radius of 2.75 cm, to the outer chimney (i.e. at z = 0,
r P 2.75 cm, uair = 0). This led to the secondary air velocity at the
inlet being ua = 32.63 cm/s. Centerline soot volume fractions and
species concentrations were reported. The results are reported as
a function of normalized axial distance, where the normalizing fac-
tor is the height, HT at which the maximum temperature is
observed.

3. Comparisons of the computed results with the experimental


data

3.1. Centerline soot volume fraction

Figure 1 shows the soot volume fractions along the centerline


for Dataset 1 [13]. The authors in [13] do not report the experimen-
tal uncertainty, but judging from the scatter of the experimental
results on Fig. 1 alone, an uncertainty of 30% is a reasonable esti-
mate; the actual total uncertainty might be higher due to a lack
of reliable estimates of the optical parameters of soot. It can be
seen that the predictions have the correct order of magnitude
and follow the same general trend as the experimental data. The
plots in Fig. 1 also indicate a systematic under-prediction by a fac-
tor of two to three.
Figure 2 shows the soot volume fractions along the centerline
for Dataset 2 [15]. The fuel mixture in Dataset 2 is diluted with
N2 and Ar and the flame temperature is about 150 degrees lower
than the flame temperature in Dataset 1 [13]. It can be seen that
excellent qualitative agreement was obtained – the results capture
well the slight increase in soot when U is decreased from infinity
(non-premixed) to 24, and the subsequent decrease in soot con-
centration as U is further decreased. The overall shape and loca-
tion, relative to the flame height, of the region in which the soot
is present, is also captured well. Quantity-wise, the soot volume
fraction is under-predicted by a factor of six, which is in line with
centerline predictions made by Dworkin et al. [23]. The under-pre-
diction factor is similar for all primary equivalence ratios.
Neither Dataset 1 [13] nor Dataset 2 [15] show centerline soot
volume fractions varying significantly with U in the U > 10 range.

3.2. Radial soot volume fractions

Figure 3 shows experimental (left side) and numerical (right


side) comparisons for radial profiles of soot volume fraction, at var-
ious axial heights for Dataset 1 [13]. Here, the comparisons are Fig. 1. Centerline soot volume fractions of Dataset 1 for various primary equiva-
made lower in the flame at (a) z = 2 cm, (b) z = 3 cm, and (c) lence ratios: (a) U = 1, (b) U = 24, (c) U = 20, (d) U = 10, (e) U = 5. The line solid is
the model predictions, the symbols are the experimental results from Arana et al.
z = 4 cm. Similarly, Fig. 4 shows the same comparisons higher up
[13].
in the flame at (a) z = 5 cm and (b) z = 6 cm. Unfortunately, no ra-
dial measurements were reported for Dataset 2 [15]. The model
predicts well the trends for U = 5, U = 10, and U = 20, with the soot U = 24, whereas in the computed data, the largest soot concentra-
on the wings (in the r = 2 cm and r = 4 cm ranges) decreasing with tions are at U = 1. The model does not predict the slight soot vol-
increased flowrates of primary air. One discrepancy is that in the ume fraction increase for U = 24 over that of U = 1 seen on the
experimental data, the largest soot concentrations are seen at wings in Figs. 3 and 4. However, quantitatively, the predictions fall
2792 V. Chernov et al. / Combustion and Flame 159 (2012) 2789–2798

Fig. 2. Centerline soot volume fractions of Dataset 2 for different primary equivalence ratios (a) computed and (b) measured [15].

Fig. 3. Radial soot volume fractions for the lower part of Dataset 1 (a) z = 2 cm, (b) z = 3 cm, (c) z = 4 cm. The left hand side figures show experimental results [13], the right
hand side figures show computed results.
V. Chernov et al. / Combustion and Flame 159 (2012) 2789–2798 2793

within experimental errors for all primary equivalence ratios ex- the concentrations of acetylene, benzene and naphthalene, respec-
cept U = 5. The predicted flame is slightly higher than the mea- tively. In each of the Figs. 5–7, the computed concentration profiles
sured one, and therefore does not predict as well soot volume are plotted on the left, and the measured values appear on the right.
fractions for z = 6 cm (Fig. 4b). At this height soot has essentially The trends of initial increase and subsequent decrease of the species
burned out, but the model still predicts the existence of a signifi- is captured by the computations for each of these three species. The
cant amount of soot. It is important to mention that the qualitative quality of the comparisons differ among the species. Acetylene
agreement was obtained using the non-premixed modeling param- (Fig. 5) is predicted very well for the non-premixed case, but the var-
eters of Dworkin et al. [23], without any modifications, thereby iation in its peak concentration with increased primary air flow is
demonstrating model robustness. much smaller than that which was observed experimentally. Ben-
zene (Fig. 6) computations are remarkably accurate in the concen-
trations and in the shape of the profile. Naphthalene (Fig. 7) is
3.3. Centerline species
under-predicted by a factor of two for all primary equivalence ratios.
Despite significant experimental uncertainty in the measured
Species measurements were reported for Dataset 2 [15] only. The
values, the data suggests that the reason for the under-prediction
measurements were taken along the centerline. Figures 5–7 show

Fig. 4. Radial soot volume fractions for the higher part of Dataset 1 (a) z = 5 cm and (b) z = 6 cm. The left hand side figures show experimental results [13], the right hand side
figures show computed results.

Fig. 5. Centerline acetylene concentrations of Dataset 2 for different primary equivalence ratios (a) computed and (b) measured [15].
2794 V. Chernov et al. / Combustion and Flame 159 (2012) 2789–2798

Fig. 6. Centerline benzene concentrations of Dataset 2 for different primary equivalence ratios (a) computed and (b) measured [15].

Fig. 7. Centerline naphthalene concentrations of Dataset 2 for different primary equivalence ratios (a) computed and (b) measured [15].

of the soot volume fractions for Dataset 2 [15] is the gas-phase facing – the lack of accurate PAH species measurements in flames
reaction mechanism that under-predicts the growth of PAHs. For for model validation.
each level of premixing (U), the growth to benzene (the first aro-
matic ring) is very well predicted (Fig. 6). However, the growth
of the second ring to naphthalene is consistently under-predicted 4. Discussion
by about a factor of two (Fig. 7). This suggests that despite signif-
icant development, the pathways to PAH growth in the chemical One of the most interesting results that is observed in partially-
mechanism [22,23] may still be incomplete. This fact is further premixed flames is that the soot does not monotonically decrease
supported by the under-prediction of soot in the inception- with decreasing primary equivalence ratio. Figure 2 illustrates this
dominated flame centerline (Figs. 1 and 2) by a factor of two to behavior by showing that the soot volume fraction for U = 24 is
eight, which has been shown to be closely correlated to pyrene slightly higher than for the non-premixed case, and otherwise
(the fourth aromatic ring) [23]. As the size of the PAHs increases, the soot volume fraction decreases with decreasing U. However,
so too does the under-prediction factor. This remains an open area since soot volume fraction is influenced not only by the combus-
of investigation in chemical mechanism development. tion processes, but also by the initial fuel concentration, it is best
It should be noted that the PAH measurements in Figs. 6 and 7 to look at the local soot yields (i.e. the fraction of the fuel carbon
have a large uncertainty. McEnally and Pfefferle [15] report an converted to soot), in order to assess the nature of the soot behav-
uncertainty of 30% for the benzene measurements and a factor of ior in the partially-premixed flames. Figure 8a shows the com-
three for the naphthalene measurements. As a result, the only con- puted centerline local soot yield normalized by the maximum
crete conclusion that can be said about the computational results is centerline local soot yield, for the non-premixed case of Dataset
that they are consistent with the measurements. This difficulty 2 [15]. Figure 8b shows the computed local soot yield on the max-
highlights a general challenge that flame and soot modeling is imum soot path through the wings of the flame, normalized by the
V. Chernov et al. / Combustion and Flame 159 (2012) 2789–2798 2795

Fig. 8. The computed normalized soot yield (a) on the centerline in Dataset 2 [15] and (b) on the path of maximum soot in Dataset 1 [13].

maximum local soot yield for the non-premixed case of Dataset 1


[13]. The path of maximum soot is an axial collection of points,
where their radial location is determined according to the maxi-
mum soot point at that particular height. The maximum soot path
is often referred to as ‘‘wings’’, since in most parts of the flame that
region is located somewhat off the burner centerline. Wings are
usually characterized by fast soot growth and most of the soot
can be found in this region. It is located just inside of the flamefront
in the non-premixed case.
We see that in both flame regions, the centerline and the wings,
there is a trend of initially rising local soot yield peaking at U = 24
and subsequently, declining with decreasing U. However, we can
also see that a wide range of primary equivalence ratios exhibit lo-
cal soot yields higher than in the non-premixed case (U = 12, 24 on Fig. 9. The computed relative mass contribution for the centerline of Dataset 2 [15].
the centerline and U = 10, 20, 24 on the maximum soot path). The
local soot yields start to fall to levels below the non-premixed case
only for low values of primary equivalence ratio (U 6 6 for the cen-
terline of Dataset 2 [15] and U < 10 for the wings of Dataset 1 [13]).
The next sections explain this phenomenon by identifying the
dominant soot processes in both flame regions, and by finding their
relation to the presence of the primary air.

4.1. Centerline soot behavior

Soot growth is influenced by three main processes: inception,


acetylene HACA surface growth, and PAH condensation. In this
work, we choose to divide these processes into two parts for the
analysis; PAH related processes, and acetylene related processes.
In practice it means that inception and PAH condensation are
Fig. 10. Computed relative values of the inception rate and local soot yield on the
examined together, and the HACA surface growth is examined sep- centerline of Dataset 2 [15].
arately. The reason for this treatment is that currently the incep-
tion model of a collision of two pyrene molecules [20] is not a
comprehensive way to couple the PAH gas-phase mechanism with by inception – changing the values of a had a negligible influence
the soot growth mechanism, and the PAH condensation model on the soot volume fraction, but a mechanism that predicted pyr-
when considering incipient soot particles is akin to an inception ene concentration more accurately on the centerline played a vital
process. role in obtaining accurate soot volume fractions. The present work
Figure 9 shows the relative contribution to the soot mass of shows that inception, even though it is modeled in a representative
growth species, as a function of the primary equivalence ratio U, manner, plays a very important role in explaining the higher cen-
on the centerline of flames, in Dataset 2 [15]. It can be seen that terline local soot yield profiles for U > 6 that are seen in Fig. 8a. Fig-
with increased primary air flow, the contribution of acetylene ure 10 shows the values of the maximum inception rate and local
along the centerline is increased, while the contribution of PAHs soot yield normalized by the non-premixed flame values for Data-
is decreased. It can be seen that both processes contribute signifi- set 2 [15]. It can be seen that inception correlates closely with the
cantly, and both are important as far as the soot mass is concerned. local soot yield, suggesting that indeed inception is an important
Figure 9 does not tell the whole story. In our group’s previous mechanism converting carbon to soot on the centerline, and that
work [23], it was found that for the ethylene diffusion flame, the the current inception model is able to describe qualitatively the
correct prediction of centerline soot volume fraction is dominated physics of the particle nucleation. Therefore, the explanation for
2796 V. Chernov et al. / Combustion and Flame 159 (2012) 2789–2798

Fig. 11. Computed centerline soot volume fraction of Dataset 2 [15] for (a) U = 24 and (b) U = 3 for three different oxidation modes.

Fig. 12. Computed mole fractions along the centerline of Dataset 2 [15] (a) O2 and (b) OH.

the soot yield behavior seen in Fig. 8a lies in the gas phase PAH oxidation has a negligible effect. Here it can be concluded that in
chemistry that is responsible for the changes in inception rates the flame region, the dominant oxidative species is OH. As premix-
with changing primary equivalence ratios. ing is increased, OH oxidation of soot increases and lowers peak
Let us examine the soot oxidation processes. Soot oxidation is levels. O2 oxidation is only important in the post-flame region of
modeled using two species, O2 and OH. To better understand their the less-premixed flames.
respective roles in partially-premixed flames, Fig. 11 shows the The oxidative species can originate from two sources; from the
computed centerline soot volume fraction for Dataset 2 [15], with primary or the secondary air. Figure 12a shows the concentration
three different oxidative models; the first model considers both O2 of O2 along the centerline of Dataset 2 [15]. It can be seen that
and OH to be participating in soot surface oxidation, the second O2 from the primary air is consumed before z/HT = 0.5, for all values
model is with OH participating only, and the third model is with of U. In the context of Fig. 11a, from which it is noted that O2 oxi-
O2 oxidation only. It can be seen from Fig. 11a that O2 oxidation dation only becomes significant after z/HT = 0.8 for U = 24, it can be
does not affect peak soot levels. However for the case of little pre- concluded that in the cases where O2 participates in the soot oxi-
mixing (U = 24) when O2 oxidation is removed, the soot is not dation, it comes from the secondary air. Figure 12b shows that
completely oxidized at the end of the computational domain and the concentration of OH increases with decreasing U, and that its
shows characteristics of a smoking flame. When OH oxidation is re- concentration starts to rise approximately at z/HT = 0.6. Figure
moved (O2 oxidation only), the centerline soot volume fraction in- 11b shows that the soot oxidation for U = 3 starts at around the
creases slightly and the effect of OH oxidation on peak soot levels same height. According to our calculations, the differences in OH
increases with further decreasing primary equivalence ratio. From fractions cannot be explained by temperature differences only.
Fig. 10b, it can be seen that in the case of heavy premixing (U = 3), Therefore we can conclude that the OH that participates in soot
when oxidation by OH is removed, the soot volume fraction center- oxidation on the centerline originates from the reactions of the pri-
line profile increases by a factor of two, whereas removing O2 mary air with the fuel.
V. Chernov et al. / Combustion and Flame 159 (2012) 2789–2798 2797

4.2. Soot behavior on the path of maximum soot (wings)

It is common to assume that HACA growth is the main contrib-


utor to soot mass on the wings of the flame [23]. Figure 13 shows
the relative contribution to the soot mass of growth species as a
function of the primary equivalence ratio U on the wings of flames
in Dataset 1 [13], at the point of maximum soot. It can be seen that
HACA growth is the dominant process of soot growth for the full
range of U.
Examining the behavior of acetylene on the wings can explain
the reason for the initial increase in the soot yield with decreasing
U. Figure 14 shows the local yields of acetylene on the wings. It can
be seen that the local yield of acetylene (i.e., the relative amount of Fig. 15. Calculated OH mole fractions along the maximum soot path of Dataset 1
fuel converted to acetylene) increases monotonically with decreas- [13].
ing primary equivalence ratio.
The total mass gain is also a function of the oxidation, which
inhibits the growth of the soot particles. Figure 15 shows the mole in acetylene and pyrene. At high premixing (U 6 5), soot yields are
fraction of OH along the maximum soot path of Dataset 1 [13], for decreased, suggesting that the effects of increased OH dominate
each of the primary equivalence ratios. O2 is not shown since in the over the increases in acetylene and pyrene.
growth region it is negligible. It can be seen that the concentration
of OH increased by a factor of seven between the non-premixed 5. Conclusions
case and U = 5, which leads to faster soot oxidation. Therefore it
can be concluded that the partial premixing influences the soot A detailed soot model including PAH growth, soot inception,
on the wings in two competing ways; firstly, it provides relative in- and aerosol dynamics, was applied for the first time to a set of
creases in acetylene concentration, which contributes to the fast partially-premixed laminar flames of ethylene and air. The goals
soot growth on the wings, and tends to increase soot formation were to see whether the computations are able to track the trends
as partial premixing is increased, and secondly, it provides more of soot formation, to assess the accuracy of the model, and to draw
OH to the wings, which increases the soot oxidation, and tends conclusions about main effects on soot processes caused by partial
to decrease soot formation as partial premixing is increased. For premixing. The computed results were tested against two experi-
a wide range of primary equivalence ratios (24 6 U 6 10), soot mental sets of data. Several main conclusions can be drawn.
yields are slightly higher than in the non-premixed case, and do All of the computed results were within reasonable accuracy of
not vary significantly between the different values of U, suggesting soot prediction, and some results exhibited excellent quantitative
that the effects of the increases in OH are balanced by the increases agreement with the experiments. Major trends were also predicted
well in most cases. All this was done without changing any param-
eter in the soot model proposed by Dworkin et al. [23] for non-
premixed flames, which demonstrates the model’s robustness.
It was found that the soot yield initially increases with the
introduction of the primary air. When the flowrate of the primary
air is further increased, the soot yield decreases and only falls be-
low the soot yield of the non-premixed flame for U < 10. The rea-
son for this lies mainly in the gas phase chemistry, but specifics
differ according to the flame region.
The centerline soot growth was found to be related to soot
inception rates. The non-monotonic increase in soot yield from
U = 1 to U = 24, followed by subsequently decreasing soot yields
for successively lowering values of U, was found to be attributable
to the gas phase PAH chemistry that controls soot inception. Both
the acetylene addition through the HACA mechanism, and the PAH
Fig. 13. The computed relative mass contribution for the wings of Dataset 1 [13]. induced growth through inception and condensation were found to
be important factors in the soot mass accumulation. On the wings,
the soot mass yield was dominated by acetylene addition. The non-
monotonic increase in soot yield from U = 1 to U = 24, the plateau
for 24 6 U 6 10, and the eventual decrease for U = 5 was found to
be attributable to competing factors; increasing OH for soot oxida-
tion, and increasing acetylene for growth with decreasing U. The
initial increase in acetylene caused higher soot yields for U = 24.
For lower primary equivalence ratios, these increases were bal-
anced by the increases in OH, which eventually became dominant
at U = 5.
The soot oxidation was found to be OH dominated, both on the
centerline and on the wings. The OH effect increased with lower
primary equivalence ratios (higher levels of premixing). For both
datasets, increases in the flowrate of the primary air led to a rapid
and full consumption of the O2 in the fuel stream, and to increased
Fig. 14. Computed local yields of acetylene along the maximum soot path of supply of OH radicals, both to the centerline and in the wings
Dataset 1 [13]. region.
2798 V. Chernov et al. / Combustion and Flame 159 (2012) 2789–2798

Acknowledgments [5] M.D. Smooke, M.B. Long, B.C. Connelly, M.B. Colket, R.J. Hall, Combust. Flame
143 (2005) 613–628.
[6] A. D’Anna, J.H. Kent, Combust. Flame 144 (2006) 249–260.
The authors acknowledge Dr. Nadezhda A. Slavinskaya, Prof. Pe- [7] A. D’Anna, J.H. Kent, R.J. Santoro, Combust. Sci. Technol. 179 (2007) 355–369.
ter Frank, and Prof. Uwe Riedel of the German Aerospace Centre [8] A. D’Anna, J.H. Kent, Combust. Flame 152 (2008) 573–587.
[9] Q. Zhang, M.J. Thomson, H. Guo, F. Liu, G.J. Smallwood, Combust. Flame 156
(DLR) for providing the DLR reaction mechanism. The authors
(2009) 697–705.
acknowledge Mr. Lyon Sachs of Montreal, Canada for financial sup- [10] Q. Zhang, H. Guo, F. Liu, G.J. Smallwood, M.J. Thomson, Proc. Combust. Inst. 32
port. The authors acknowledge the Natural Sciences and Engineer- (2009) 761–768.
[11] M. Saffaripour, P. Zabeti, S.B. Dworkin, Q. Zhang, H. Guo, F. Liu, G.J. Smallwood,
ing Research Council of Canada and the Ontario Ministry of
M.J. Thomson, Proc. Combust. Inst. 33 (2011) 601–608.
Research and Innovation for financial support. Computations were [12] A. Mitrovic, T.W. Lee, Combust. Flame 115 (1998) 437–442.
performed on the GPC supercomputer at the SciNet HPC Consor- [13] C.P. Arana, M. Pontoni, S. Sen, I.K. Puri, Combust. Flame 138 (2004) 362–372.
tium. SciNet is funded by: the Canada Foundation for Innovation [14] V.R. Katta, R.A. Forlines, W.M. Roquemore, W.S. Anderson, J. Zelina, J.R. Gord,
S.D. Stouffer, S. Roy, Combust. Flame 158 (2011) 511–524.
under the auspices of Compute Canada; the Government of Ontar- [15] C.S. McEnally, L.D. Pfefferle, Combust. Flame 121 (2000) 575–592.
io; Ontario Research Fund – Research Excellence; and the Univer- [16] M.D. Domenico, P. Gerlinger, M. Aigner, Combust. Flame 157 (2010) 246–258.
sity of Toronto. [17] F. Gelbard, J.H. Seinfeld, J. Colloid Interface Sci. 78 (1980) 485–501.
[18] F. Gelbard, Y. Tambour, J.H. Seinfeld, J. Colloid Interface Sci. 76 (1980) 541–
556.
[19] S.B. Dworkin, M.D. Smooke, V. Giovangigli, Proc. Combust. Inst. 32 (2009)
References 1165–1172.
[20] J. Appel, H. Bockhorn, M. Frenklach, Combust. Flame 121 (2000) 122–136.
[1] I. Glassman, Combustion, fourth ed., Elsevier, Boston, 2008, p. 458. [21] N.M. Marinov, W.J. Pitz, C.K. Westbrook, A.M. Vincitore, M.J. Castaldi, S.M.
[2] R.J. Santoro, H.G. Semerjian, R.A. Dobbins, Combust. Flame 51 (1983) 203–218. Senkan, C.F. Melius, Combust. Flame 114 (1998) 192–213.
[3] C.S. McEnally, A.M. Schaffer, M.B. Long, L.D. Pfefferle, M.D. Smooke, M.B. Colket, [22] N.A. Slavinskaya, P. Frank, Combust. Flame 156 (2009) 1705–1722.
R.J. Hall, Symp. Int. Combust. 27 (1998) 1497–1505. [23] S.B. Dworkin, Q. Zhang, M.J. Thomson, N.A. Slavinskaya, U. Riedel, Combust.
[4] M.D. Smooke, R.J. Hall, M.B. Colket, J. Fielding, M.B. Long, C.S. McEnally, L.D. Flame 158 (9) (2011) 1682–1695.
Pfefferle, Combust. Theor. Model. 8 (2004) 593–606. [24] F. Liu, H. Guo, G.J. Smallwood, Combust. Flame 138 (2004) 136–154.

You might also like