You are on page 1of 8

Available online at www.sciencedirect.

com

Proceedings of the Combustion Institute 36 (2017) 1759–1766


www.elsevier.com/locate/proci

MMC-LES simulations of turbulent piloted flames with


varying levels of inlet inhomogeneity
S. Galindo∗, F. Salehi, M.J. Cleary, A.R. Masri
School of Aerospace, Mechanical and Mechatronic Engineering, The University of Sydney, NSW, 2006 Australia
Received 4 December 2015; accepted 13 July 2016
Available online 5 October 2016

Abstract

This paper presents simulations of piloted turbulent flames where compositional inlet conditions vary
from homogeneous to inhomogeneous. The combustion model is sparse-Lagrangian multiple mapping con-
ditioning (MMC) coupled with large eddy simulation (LES). This is a first attempt using MMC-LES to model
flows where multi-modes of combustion may occur. The studied burner has two concentric tubes surrounded
by an annular pilot. The central tube carrying methane can slide within the outer air tube to induce com-
positional inhomogeneity at the burner exit plane. This leads to enhanced flame stability at some optimal
inner fuel tube recess distance. For comparison purposes, model results are presented firstly for a case with
compositional homogeneity at the inlet plane and secondly for the inhomogeneous case. The MMC-LES
model shows a very good agreement with the experiment for the homogeneous-inlet case where the preva-
lence of non-premixed combustion is observed. The inhomogeneous-inlet case features a transition from a
premixed flame structure close to the burner to a diffusion flame further downstream. It represents an ex-
treme test of the chosen MMC model which incorporates the enforcement of mixing locally in an extended
space comprised of the LES reference mixture fraction and physical location. The computed results show a
fair agreement with the data close to the nozzle; closely approaching but not quite achieving the premixed
structure. The return to a broad, distributed diffusion flame away from the nozzle is, however, captured very
well.
© 2016 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

Keywords: Turbulent combustion; Partially premixed; Inhomogeneous inlets; MMC-LES

1. Introduction Navier Stokes (RANS)-based approaches to large


eddy simulations (LES) [2,3]. The employed com-
In recent years, there have been significant bustion models range from variants of the laminar
advances in computing the structure of both flamelet concept [4,5], to conditional moment
turbulent non-premixed and premixed flames [1]. closure (CMC) [6] and conditional source term
With the advent of fast computers, the models and estimation (CSE) models [7,8], and probability
codes have progressed from Reynolds-averaged density function (PDF) methods where either Eu-
lerian stochastic fields or Lagrangian Monte-Carlo
methods may be used [9,10]. While some of these
∗ Corresponding author. Fax: + 61 2 9345 0393. models, such as transported PDF methods, have
E-mail address: sgal8842@uni.sydney.edu.au been derived with generality to address turbulent
(S. Galindo). reacting flows independent of the combustion

http://dx.doi.org/10.1016/j.proci.2016.07.055
1540-7489 © 2016 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
1760 S. Galindo et al. / Proceedings of the Combustion Institute 36 (2017) 1759–1766

mode, most implementations and associated sub-


models have focused on premixed or non-premixed
combustion alone. However, flames in practical
combustors typically transition across a broad
mixture fraction range leading to multiple modes
of combustion in the same flame. Development
of models for partially premixed and mixed-mode
flames is underway [8,11–17], and the present
paper is part of that process. A good platform for
validating such models is the Sydney piloted burner
with compositionally inhomogeneous inlets, which
can stabilise multi-mode flames co-existing within Fig. 1. a) Inhomogeneous inlets burner cutout and
the same jet [18,19]. This is a simple variant of b) blow-off velocity versus recess distance (Lr) (figure
adapted from [26]).
the standard piloted burner [20] which has already
formed a very effective platform to advance the
modelling of turbulence-chemistry interaction in
non-premixed flames [21]. tions [26]. The experiments were performed using
The current study aims to test the validity of the the well-known Sydney piloted jet burner [20] mod-
sparse-Lagrangian multiple mapping conditioning ified to include two concentric tubes surrounded by
model in the LES framework (MMC-LES) for the pilot stream. The inner tube may be recessed
flames with inhomogeneous inlets and multi-mode within the outer tubes to vary the degree of pre-
flame structures. MMC-LES has proven to be a mixing between the two streams. A schematic of the
very affordable and accurate combustion model burner is shown in Fig. 1a and more details can be
for a range of non-premixed flames [22,23]. It uses found in Ref. [18,27]. The burner assembly is cen-
a Monte Carlo approach to solve the subfilter tred in a wind tunnel providing a uniform air co-
PDF. Owing to the small number of stochastic flow which is fixed at Uc = 15 m/s. Changes in the
Pope particles used in the simulations, MMC-LES conditions at the burner exit are achieved by vary-
requires a significantly reduced computational ing the recess distance, Lr . At sufficiently large val-
effort relative to PDF methods with conventional ues of Lr , homogeneous, partially premixed condi-
mixing models [22,24,25]. The model offers a tions are produced, while the non-premixed con-
controlled mixing process with enforcement of dition prevails when both tubes are flush at the
localness in an extended space comprised of a burner exit plane. The burning mode for the flames
reference mixture fraction and physical location. with inhomogeneous composition at the exit plane
While MMC-LES has been successfully applied will depend upon the equivalence ratio at the jet
to various non-premixed flames, this is a first edge encountered by the pilot. Figure 1b) presents
attempt at modelling multi-mode conditions. A the blow-off velocity (Ubo ) versus recess distance at
homogenous case and an inhomogeneous case conditions where the volumetric air to fuel ratio in
are considered. In the latter the compositional the inner tube and the annulus is VA /VF = 2 [26].
inhomogeneity is such that the mixture at the Abbreviation FJ refers to the condition where the
interface between the jet and the pilot is close to fuel is issued from the inner tube whereas the air is
the stoichiometric mixture fraction so the flame carried by the outer tube. The reverse condition is
features a premixed structure close to the nozzle labelled FA. A five-gas pilot, labelled as 5GP, is em-
and a transition to a non-premixed structure at ployed in the experiments. The pilot fuel is a stoi-
downstream locations. In generalised MMC, fluc- chiometric mixture of acetylene, hydrogen, oxygen,
tuations relative to the reference mixture fraction nitrogen and carbon dioxide. The pilot has the same
manifold allow for modelling of conditional fluc- C/H ratio and adiabatic flame temperature as the
tuations and some extent of premixing. We test the methane/air mixture. As can be seen, the FJ and FA
extent to which this model can approach the fully cases have very different stability characteristics.
premixed structure close to the nozzle and capture For the FJ configuration the optimal stability (max.
the return to non-premixed burning downstream. blow-off velocity) is obtained for Lr = 75 mm. In
contrast, the stability of the FA flames increases
slowly and monotonically with increasing Lr .
2. Experimental conditions and the computational Two cases, FJ200-5GP-Lr300-59 and FJ200-
model 5GP-Lr75-80, are selected for simulation. Both are
indicated by the star symbols in Fig. 1b. The first
2.1. Sydney piloted burner with inhomogeneous case with Lr = 300 mm has a compositionally ho-
inlet mogeneous mixture at the burner exit whereas the
second case with Lr = 75 mm is inhomogeneous.
A complete experimental data set has recently The bulk jet exit velocities are 59 and 80 m/s, re-
been provided by the University of Sydney for a spectively, such that both flames are at 70% of the
series of piloted flames with variable inlet condi- blow-off velocity at that value of Lr . Herein, for
S. Galindo et al. / Proceedings of the Combustion Institute 36 (2017) 1759–1766 1761

convenience, flames H and I are used to refer 2.3. Numerical implementation


to the homogeneous and inhomogeneous cases,
respectively. The MMC-LES model has been implemented
into a new solver called mmcFoam that is com-
patible with the open source, object orienteded
2.2. MMC-LES model C++ OpenFOAM suite of solvers [30]. The new
sparse-Lagrangian MMC classes are coupled with
The main MMC-LES concepts are presented OpenFOAM’s existing low-Mach, compressible
here and complete details, including derivation LES solver with the standard Smagorinsky subgrid
of the model equations, can be found in [22]. The model [31]. There is two-way coupling between the
filtered continuity, velocity, pressure and reference LES and MMC so that the fields remain consistent.
mixture fraction fields are simulated using the The velocity, reference mixture fraction, pressure
Eulerian LES, whereas a stochastic Lagrangian and turbulence parameters computed by LES,
scheme is employed for the reactive scalar fields. In are interpolated at particle locations and used to
the latter, stochastic differential equations (SDEs) integrate particle movement followed by particle
are solved for an ensemble of Pope particles which mixing and reaction substeps. Density coupling is
represent the local sub-filter state of the composi- achieved through an adaptation [22] of the equiv-
tion, otherwise known as a filtered density function alent enthalpy method [32]. Additional equivalent
(FDF) [1]. A mixing model is required to emulate major species and sensible enthalpy transport
the unclosed conditional dissipation of subfilter equations are solved on the LES solely for the
scalar variances. Subramaniam and Pope [28], in purpose of determining the filtered density. The
discussing the principles of a good mixing model, source terms in those equations are conditional
highlight localness and linearity and independence averages obtained from a mean-square minimisa-
as key requirements. Here, as in all previous MMC- tion of a local ensemble of particle mass fractions
LES applications to date, the localness is enforced relative to a library of precomputed flamelets.
in an extended space comprised of the reference Here, the library contains six flamelets computed
mixture fraction (simulated by the Eulerian LES) from laminar calculations with varying strain rates.
and physical location. In non-premixed flames The ensemble is limited to those particles within
localness in mixture fraction space implies a high a so-called supercell containing multiple LES grid
degree of localness in composition space. An cells (see below). Once the flamelet curve is selected
MMC model was recently developed specifically locally, the equivalent species conditional source
for premixed combustion based on a distance-like term is the flamelet source term corresponding to
reference variable [29] but it has not yet been im- the LES filtered mixture fraction. A relaxation time
plemented within LES. In the present flame cases, is applied to the source term to ensure stability.
we continue to use the mixture fraction reference More details can be found in Ref. [22]. The multi-
variable even for Flame I which exhibits premixed mode flame produced above the inhomogeneous
burning near the pilot and a non-premixed struc- burner (Flame I) can lead to numerical instabilities
ture over the remainder of the domain. Particles in the LES due to jumping between the library
are mixed in pairs with a timescale that is de- curves. Stable simulations were achieved using a
termined from the local turbulence parameters, CFL number of 0.4 combined with a density cou-
determining the level of fluctuations relative to the pling using a relaxation time equal to 10 time steps
reference mixture fraction manifold. Thus, MMC- [22]. Simulations of Flame I were first run with
LES has significant differences to both flamelet a non-premixed library curve and statistics were
and first order CMC models for which conditional obtained. Following that a premixed curve was
fluctuations are not directly considered. As is added to the library and no substantial changes in
shown below, this allows the MMC-LES model to the mixing field were observed.
closely approximate the premixed structure. The computational domain consists of a 3D
Localness between mixing particle pairs is con- rectangular mesh that extends 50 jet diameters
trolled by the selection of characteristic physical axially and 10 jet diameters in both transverse di-
and reference mixture fraction scales, denoted as rm rections. The mesh consists of 1.4 million cells and
and f m , respectively. In the present modelling both it is refined in the jet and pilot regions resulting in
parameters have global values. The optimal value a mesh size of 0.4 mm cubed near the centreline. In
for f m is likely to be chemistry dependent with the sparse-Lagrangian MMC, 1 particle per 8 Eu-
different fuels exhibiting different reaction zone lerian LES cells (1 L/8E) is used. Particle number
thickness. For pure methane, f m = 0.02 is suggested control is associated with supercells consisting of
[22] while it was found that f m = 0.03 can provide 4 × 4 × 8 LES cells with an average of 16 particles
a better prediction for broader reaction zones such in each. The (1 L/8E) configuration is used based on
as found in partially premixed methane-air flames extensive validation in [22]. The supercells used for
[23]. Here, we use f m = 0.03, while rm is calculated particle control are also used for density coupling
according to the fractal gradient model of Cleary as discussed above. For both flame cases, zero-
and Klimenko [22]. gradient pressure boundary conditions are applied
1762 S. Galindo et al. / Proceedings of the Combustion Institute 36 (2017) 1759–1766

Fig. 2. Radial profiles of mean and rms of mixture fraction, temperature and CO mass fraction for flame H. Axial location
is indicated at the top (x/D = 1, 5, 15, 20). Solid lines: simulations; squares: experiments.

at the inlets and fixed total pressure conditions at 3. Results


the domain sides and outflows. Composition and
velocity boundary conditions differ between the Figure 2 shows radial profiles of the steady-state
two cases. For Flame H, the composition at the mean and rms of mixture fraction, temperature
burner exit is homogeneous and the time-varying and CO mass fraction at various axial locations
velocity has a turbulence intensity of 8% and an in- for Flame H with the homogeneous inlet condi-
tegral length scale of 1/8th of the jet diameter. Both tions. There is a very good agreement between
values are found to give the correct jet breakup the predicted and measured quantities near the
location and subsequent scalar profiles along the burner and up to x/D = 15. Although not shown
centreline. On the other hand, Flame I has an inho- due to space limitations, predictions of CH4 ,
mogeneous composition at the exit plane. The cor- O2 , CO2 and H2 O show similar accuracy as the
related turbulent composition and velocity bound- temperature, and predictions of minor species
ary data set is generated a priori by performing including H2 show similar accuracy as CO. The
an LES simulation of the internal pipe flow in the effect of the pilot in the form of double-peak rms
burner. Within the flames, chemical reaction rates profiles is evident in the near-burner region. At
are calculated according to the DRM-22 reduced x/D = 30, the simulations appear to under-predict
kinetics scheme [33] which includes 22 reactive the jet spreading with the mean mixture fraction
species plus argon and nitrogen and 104 reactions. at the centreline failing to decay as quickly as was
All simulations were started and run until experimentally observed. The mixture fraction rms
they reached a statistically steady state and then is, however, reasonably well predicted throughout
statistics were collected over a period of five flow the domain. Consistent with the mixture fraction
through times. After this point we continued the field, the model under-predicts the mean values
simulations for another two flow through times of temperature and CO mass fraction at x/D= 30
observing no significant difference to the results. while their fluctuations are reasonably accurate.
As a comparison of computational cost on an Intel Comparisons of mean measured and computed
Core i7-4790 K @ 4 GHz x 8, for a simulated time radial profiles are shown in Fig. 3 for the more
of 6.3 ms which approximately corresponds to one challenging case Flame I with the inhomogeneous
flow through time for the homogeneous inlet case, inlet compositions. The computed mean mixture
320 CPU hours were required for computation of fraction and temperature agreed well with the
reacting flows while 94 CPU hours were required experiment near the burner and up to x/D = 15.
for an equivalent non-reacting cold flow. Further downstream, the mean profile of mixture
S. Galindo et al. / Proceedings of the Combustion Institute 36 (2017) 1759–1766 1763

Fig. 3. Radial profiles of mean and rms of mixture fraction, temperature and CO mass fraction for flame I. Axial location
is indicated at the top (x/D = 1, 5, 15, 20). Solid lines: simulations; squares: experiments.

fraction is well predicted showing good decay of the flame returns to a fully-burning state. MMC-
centreline values as well as jet spread however the LES captures this quite well although the exact
level of extinction is over-predicted resulting in location of reignition is a little further downstream
lower mean temperatures. It is important to note than in the experiments. As an aside, the dual rich-
that although the peak values of mixture fraction side branch observed in the experimental data at
rms are under-predicted, the location is well cap- x/D = 1 is due to asymmetry caused by difficulties
tured and this location correspond approximately in perfectly aligning the inner tube in the burner.
to location of the maximum gradient of the mean The scatter plots for Flame I shown in Fig. 4b
profile. It can be seen that mean CO values are reveal a very different flame structure, especially
under-predicted in the most part of the flame. in the near-nozzle region where the inhomogeneity
In order to examine the multi-mode flame struc- of the burner exit composition produces mixtures
ture in detail, we now consider instantaneous and close to the stoichiometric mixture fraction. A pre-
mean and rms quantities in mixture fraction space. mixed mode is observed in the experimental data
Figure 4 shows measured and computed scatter at the interface between the main jet and the pilot
plots for temperature versus mixture fraction for characterised by a steep increase in temperature
both flames at x/D = 1, 5 and 15. For Flame H from cold unburned to hot burned conditions.
shown in Fig. 4a, the dominant diffusion flame On the lean side of stoichiometric at x/D = 1,
structure is clearly evident and as expected based corresponding to the outer pilot region, a non-
on past experience with similar flames [23] the premixed structure occurs. The MMC-LES closely
MMC-LES performs very well. At x/D = 1 and 5 approaches the premixed mode of combustion at
the flame appears to be close to chemical equilib- x/D = 1 although the finite slope in the temperature
rium, displaying a flamelet-like structure which is scatter indicates that the non-premixed structure
easily captured by the model even with a sparse remains albeit with most of the heat release oc-
distribution of Pope particles due to the enforce- curring in a very narrow band on the rich side of
ment of mixing localness in the reference mixture stoichiometry. This behaviour is a consequence
fraction space. A small degree of local extinction of the mixture fraction dependence of the MMC
is detected downstream of the exit plane between mixing model. Downstream of x/D = 1, the flame
x/D = 10 and x/D = 15 and the MMC-LES model transitions from the premixed to the diffusion
also reproduces these features very well. Further mode and by x/D = 5 diffusion burning is evident
downstream (not shown due to space limitations) in the data although there are some remnants of
1764 S. Galindo et al. / Proceedings of the Combustion Institute 36 (2017) 1759–1766

Fig. 4. Scatter plots of temperature versus mixture fraction for cases a) Flame H and b) Flame I.

premixed fluid samples. Calculations reproduce Fig. 5b. Consistent with the scatter plots, these
these trends but slightly over-predict the rate of conditional profiles reveal that the MMC-LES
broadening of the reaction zone at this location. with a mixture fraction reference variable cannot
At x/D = 15 where the flame has completed the quite capture the premixed structure at x/D = 1,
transition to a non-premixed structure, the simula- instead showing a broader non-premixed zone of
tions are in good agreement with the experiments reaction at near stoichiometry. However, accuracy
albeit with a bit less conditional scatter. improves further downstream where the burning
Figure 5a presents conditional means and rms mode transitions to non-premixed combustion.
for the temperature and the mass fraction of CO at The computed rms of temperature and CO near
various axial locations for flame H. The computed stoichiometric mixture fraction at x/D = 1 are
conditional mean and conditional rms of tempera- under predicted and this is consistent with the
ture are in excellent agreement with the experiment observed scatter were fluid samples do not extend
while the MMC-LES model under-predicts condi- from fresh gases to burnt gases. This also reflects
tional mean mass fraction of CO close to the exit that in the mixing operation Pope particles could
plane. Although not shown, a similar discrepancy be mixing without passing through the premixed
is observed for hydrogen while the agreement for front. The slightly over-predicted mean tempera-
major species such as carbon dioxide and water is ture with under-precdicted rms values at x/D = 15
adequate. A simulation performed using the GRI-3 are consistent with the reduced scatter in the model
mechanism containing 34 species and 219 reactions results at that location as shown in Fig. 4b.
(NOx excluded) [34] has shown that the more de- It is evident that a specific premixed MMC
tailed chemistry only marginally improved the com- model is needed to reproduce the extent of premix-
puted CO mass fraction. It is worth noting that the ing observed in the inhomogeneous case Flame I.
thickness of mixing layers is slightly over-predicted The model by Sundaram et al. [29], which replaces
by the MMC-LES at the axial location close to the the reference mixture fraction by a distance-like ref-
exit plane. It is evident from the profiles shown in erence variable should capture the premixed flame
Fig. 2 that the computed mixture fraction in the structure, although LES requires a method for
shear layer is gently decreased while the experiment locating the flame surface. Alternative approaches
shows a very steep gradient of mixture fraction. based on a reaction progress variable have been
Profiles of conditional mean and rms for considered in MMC of homogeneous turbulence
Flame I with inhomogeneous inlet are presented in [35] and may be used in the current flames as well,
S. Galindo et al. / Proceedings of the Combustion Institute 36 (2017) 1759–1766 1765

Fig. 5. Conditional mean and rms of temperature and CO mass fraction for a) Flame H and b) Flame I. Solid lines:
computed values; squares: experiments.

although the need to model the progress variable Department of Science, Technology and Innova-
source term may be a complicating factor. tion – COLCIENCIAS.

4. Conclusions
References
Turbulent piloted flames with both homoge- [1] H. Pitsch, Annu. Rev. Fluid Mech. 38 (2006) 453–482.
neous and inhomogeneous inlet compositions have [2] T. Poinsot, D. Veynante, Theoretical and Numerical
been simulated using MMC-LES. The predictions Combustion, R.T. Edwards, Inc., 2005.
for the homogenous inlet case, where non-premixed [3] R.W. Bilger, S.B. Pope, K.N.C. Bray, J.F. Driscoll,
flame structures are dominant, are in good agree- Proc. Combust. Inst. 30 (2005) 21–42.
ment with the experiments although the jet spread [4] N. Peters, Prog. Energy Combust. Sci. 10 (1984)
is under-predicted far downstream. The inhomo- 319–339.
geneous inlet case with mixed-mode combustion [5] H. Pitsch, M. Chen, N. Peters, Symp. (Int.) Combust.
27 (1998) 1057–1064.
represents an extreme test for the model. The in-
[6] A.Y. Klimenko, R.W. Bilger, Prog. Energy Combust.
homogeneity is evident in the calculations, and Sci. 25 (1999) 595–687.
while the premixed structure near the nozzle is ap- [7] H. Steiner, W.K. Bushe, Phys. Fluids 13 (2001)
proached it is not fully captured with a narrow non- 754–769.
premixed flame persisting. The return to a broad [8] D. Dovizio, J.W. Labahn, C.B. Devaud, Combust.
non-premixed flame structure at downstream lo- Flame 162 (2015) 1976–1986.
cations is very well captured. These results, while [9] S.B. Pope, Prog. Energy Combust. Sci. 11 (1985)
promising, reflect the need to incorporate a specific 119–192.
premixed MMC model. A number of options were [10] W.P. Jones, S. Navarro-Martinez, Combust. Flame
150 (2007) 170–187.
briefly discussed and these will be pursued in the
[11] M. Chen, M. Herrmann, N. Peters, Proc. Combust.
future. Inst. 28 (2000) 167–174.
[12] P. Domingo, L. Vervisch, K. Bray, Combust. Theory
Model. 6 (2002) 529–551.
Acknowledgements [13] Y.C. See, M. Ihme, Proc. Combust. Inst. 35 (2015)
1225–1234.
This work is supported by the Australian Re- [14] P.-D. Nguyen, L. Vervisch, V. Subramanian,
search Council and the Colombian Administrative P. Domingo, Combust. Flame 157 (2010) 43–61.
1766 S. Galindo et al. / Proceedings of the Combustion Institute 36 (2017) 1759–1766

[15] M. Hegetschweiler, C. Handwerk, P. Jenny, Combust. [25] R.R. Cao, S.B. Pope, A.R. Masri, Combust. Flame
Sci. Technol. 182 (2010) 480–490. 142 (2005) 438–453.
[16] E. Knudsen, H. Pitsch, Combust. Flame 159 (2012) [26] A.R. Masri, Available at http://web.aeromech.usyd.
242–264. edu.au/thermofluids/database.php.
[17] F. Cavallo Marincola, T. Ma, A.M. Kempf, Proc. [27] R.S. Barlow, S. Meares, G. Magnotti, H. Cutcher,
Combust. Inst. 34 (2013) 1307–1315. A.R. Masri, Combust. Flame 162 (2015) 3516–3540.
[18] S. Meares, A.R. Masri, Combust. Flame 161 (2014) [28] S. Subramaniam, S.B. Pope, Combust. Flame 115
484–495. (1998) 487–514.
[19] S. Meares, V.N. Prasad, G. Magnotti, R.S. Bar- [29] B. Sundaram, A.Y. Klimenko, M.J. Cleary, U. Maas,
low, A.R. Masri, Proc. Combust. Inst. 35 (2015) Proc. Combust. Inst. 35 (2015) 1517–1525.
1477–1484. [30] H.G. Weller, G. Tabor, H. Jasak, C. Fureby, Comput.
[20] A.R. Masri, R.W. Dibble, R.S. Barlow, Prog. Energy Phys. 12 (1998) 620–631.
Combust. Sci. 22 (1996) 307–362. [31] J. Smagorinsky, Mon. Weather Rev. 91 (1963)
[21] TNF Workshop, Available at http://www.sandia.gov/ 99–164.
TNF/abstract. [32] M. Muradoglu, S.B. Pope, D.A. Caughey, J. Comput.
[22] M.J. Cleary, A.Y. Klimenko, Phys. Fluids 23 (2011) Phys. 172 (2001) 841–878.
115102. [33] A. Kazakov, M. Frenklach, DRM22, Available at
[23] Y. Ge, M.J. Cleary, A.Y. Klimenko, Proc. Combust. http://www.me.berkeley.edu/drm/.
Inst. 34 (2013) 1325–1332. [34] G.P. Smith, D.M. Golden, M. Franklach, et al.,
[24] J. Xu, S.B. Pope, Combust. Flame 123 (2000) Available at http://www.me.berkeley.edu/gri_mech/.
281–307. [35] A. Kronenburg, M.J. Cleary, Combust. Flame 155
(2008) 215–231.

You might also like