You are on page 1of 18

LES modeling of premixed combustion using a thickened ame approach

coupled with FGM tabulated chemistry


G. Kuenne
a,b,
, A. Ketelheun
a
, J. Janicka
a,b
a
Institute of Energy and Power Plant Technology, Darmstadt University of Technology, Petersenstrasse 30, 64287 Darmstadt, Germany
b
Center of Smart Interfaces, Darmstadt University of Technology, Petersenstrasse 32, 64287 Darmstadt, Germany
a r t i c l e i n f o
Article history:
Received 23 September 2010
Received in revised form 17 December 2010
Accepted 8 January 2011
Available online 7 February 2011
Keywords:
Turbulent premixed combustion
Tabulated chemistry
Thickened ame
Large eddy simulation
Swirl burner
a b s t r a c t
Flamelet Generated Manifolds (FGM) tabulated chemistry is used in combination with a thickened ame
approach to perform Large Eddy Simulation (LES) of premixed combustion. Two-dimensional manifolds
are used to describe the chemistry by the mixture fraction and progress variable. Simulations of one-
dimensional ames have been used to verify the coupling of the tabulated chemistry and the LES solver
where important features like the grid dependence of ame propagation are carefully addressed. Finally,
the method is applied to the turbulent ame of a premixed swirl burner including the complex geometry
of the swirl nozzle. Results of the velocity, species and temperature are compared with experimental
data. Thereby different efciency functions are used to show the sensitivity related to this model param-
eter. Some aspects regarding dynamic thickening, numerical accuracy and computational efciency are
also addressed.
2011 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
1. Introduction
Lean premixed combustion is of increasing importance in many
industrial applications (such as land based gas turbines [1],
aero-engines [2] and automotive engines [3]) regarding the pollu-
tant formation due to the lower peak temperature when compared
to non-premixed systems. The occurrence of unsteady phenomena
like ashback and combustion instabilities makes these ames
difcult to control and hence it is desirable to predict their
behavior using LES. Since a ammable mixture exists before com-
bustion occurs the ame speed is a decisive parameter and needs
to be reproduced by the simulation. This is difcult because the
reaction can not be resolved on typical computational grids and
models are required to ensure the correct ame propagation. In
addition, the model needs to account for the interaction of
unresolved vortices with the ame. Therefore the ame speed
enters the model as an a priory known parameter, such as in the
G-equation or Flame Surface Density concept, or results on its
natural way from the integration of the chemical source term.
Regarding the latter strategy the use of a ltered chemical look-
up table has been proposed very recently by Vreman et al. [4]
which has been rened in the F-TACLES (ltered tabulated
chemistry for LES) model by Fiorina et al. [5] to allow a proper
description of the ltered ame structure and propagation. This
approach seems very attractive but not much experience has been
gained with it yet. Another well established strategy to resolve the
ame on LES meshes is to articially thicken the ame (Articially
Thickened Flame (ATF)-model).
The ATF-model is very attractive since its theoretical derivation
contains no restricting assumptions regarding the ame topology.
Due to its universal validity it has been applied to premixed ames
[6], lifted ames [7] as well as ignition sequences [8] where nite
rate chemistry effects are important. In order to minimize the
modeling effort introduced by the modied ame turbulence inter-
action the thickening factor is limited to resolve the length scales
of major species only. Therewith the complexity of the chemical
scheme is in general limited to strongly reduced mechanisms
(13 steps). Of course a small number of reactions is desirable
for computational efciency, but important ame characteristics
are difcult to reproduce. As mentioned by Selle et al. [6] and
Schmitt et al. [9] especially at higher equivalence ratios errors
occur regarding the ame speed and burnt gas temperature.
Boudier et al. [10] adjusted the preexponential constant to obtain
the correct ame propagation but the burnt gas temperature
remained to high in rich regions.
In the present work the ATF concept is combined with a tabula-
tion strategy to include detailed chemistry effects into the LES.
Regarding premixed ames two similar tabulation approaches
developed simultaneously called FPI (ame prolongation of
intrinsic low-dimensional manifolds, ILDM [11]) [12,13] and FGM
0010-2180/$ - see front matter 2011 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.combustame.2011.01.005

Corresponding author at: Institute of Energy and Power Plant Technology,


Darmstadt University of Technology, Petersenstrasse 30, 64287 Darmstadt,
Germany. Fax: +49 6151 16 6555.
E-mail address: kuenne@ekt.tu-darmstadt.de (G. Kuenne).
Combustion and Flame 158 (2011) 17501767
Contents lists available at ScienceDirect
Combustion and Flame
j our nal homepage: www. el sevi er . com/ l ocat e/ combust ame
(amelet generated manifolds) [14,15] proved to be well suited to
accurately describe the ame structure and propagation. It will be
demonstrated that important ame characteristics can be repro-
duced over the whole range of equivalence ratios by transporting
two controlling variables for mixing (mixture fraction) and reac-
tion progress (CO
2
mass fraction) combined with the FGM table.
Compared to reduced mechanisms this method is thought to be a
promising alternative when using articial thickening. Of course
more a priori knowledge about the reaction is required when using
tabulated chemistry. For example, if heat losses are important that
needs to be accounted for by an additional table dimension while it
naturally arises in the Arrhenius law of a reduced mechanism. In
addition, the freely propagating ame obtained from the one-
dimensional detailed chemistry solution is only an approximation
for the stretched and curved ame in a turbulent ow eld. But this
amelet assumption is not very restrictive as investigated by Poin-
sot et al. [16] by means of DNS where the domain of amelet mod-
eling has been found to be much larger than expected from
classical combustion diagrams. Further arguments related to the
mass burning rate of a stretched and curved ame in combination
with the Lewis number are given in Section 2.1.
The outline of this paper is as follows. In Section 2 the chemistry
reduction using FGM and the LES solver will be introduced. The
coupling of these will be investigated and the quantitative error
behavior of ame propagation on coarse grids will be explained
by means of theoretical derivations and 1-D numerical simula-
tions. In Section 3 the ATF concept will be summarized and the im-
pact of dynamic thickening based on a ame sensor on the ame
turbulence interaction is illustrated. Some interesting ndings
regarding the numerical treatment of the thickening procedure
are also discussed. Finally in Section 4 the model is applied to a
methane-air swirl burner at atmospheric pressure where the ame
is stabilized by a central recirculation zone which is common prac-
tice in gas turbine combustors. To allow for a fully developed tur-
bulent ow the upstream geometry including the bluff-body swirl
nozzle of full complexity was included in the computational
domain.
2. Coupling of tabulated chemistry with the LES solver
2.1. FGM tabulation
Like many other reduction strategies the technique of amelet
generated manifolds aims at describing detailed chemistry by only
a couple of controlling variables. This section is restricted to the
manifold construction of a methane-air ame used in this work.
A more fundamental description of the theory and its verication
can be found in [14,15]. For the table generation, rst a one-dimen-
sional freely propagating premixed ame at constant equivalence
ratio is simulated using the detailed chemistry one-dimensional
ame code Chem1D [17] with the GRI 3.0 reaction scheme [18].
This computation is repeated for different equivalence ratios to
span a two-dimensional manifold which can be parameterized by
the mixture faction and a reactive scalar as progress variable.
Within this work the non-normalized mass fraction of carbon
dioxide is used as progress variable even though it is not strictly
monotonic in rich regions (/ > 1.2). In the literature different pro-
gress variables (Y
pv
) built by a linear combination of several
(weighted) species mass fractions have been suggested to meet
the requirement of being monotonic for all equivalence ratios.
Two of them being often used are
Y
pv;1
= Y
CO
2
Y
CO
[19; 13[ (1)
Y
pv;2
=
Y
CO
2
W
CO
2

Y
H
2
O
W
H
2
O

Y
H
2
W
H
2
[20; 4; 21[ (2)
where W denotes the molar mass of the corresponding species.
Unfortunately these progress variables have a much thinner reac-
tion zone compared to Y
CO2
alone. This is illustrated on the left
of Fig. 1 where the thickness of the source term (d( _ x
pv
)) of the dif-
ferent progress variables dened by its full width at half maximum
is shown for various equivalence ratios. In the region around stoi-
chiometry the source term of Y
CO
2
can be resolved using only half
of the resolution compared to Eqs. (1) and (2) which is highly
desirable. Hence, we decided to neglect the decomposition of
CO
2
into CO at high temperature levels by using only the mono-
tonic part of the CO
2
mass fraction to allow for a unique relation
to the state variables extracted from the table. The resulting error
regarding the prediction of density and temperature dened to be
the maximum difference between the value at the turning point of
the CO
2
mass fraction (index CO
2
,tp which is the last monotonic
entry) and the correct value obtained in the amelet behind the
turning point
(q) =
max q
CO
2
;tp
q
flamelet
_ _
q
CO
2
;tp
(3)
is shown on the right of Fig. 1. As one can see Y
pv,1
and Y
pv,2
are the
more accurate progress variables to describe the complete ame
structure for all equivalence ratios whileas expecteddeviations
from the correct value in the rich region can be observed if Y
CO
2
is
used (the error in ame propagation speed is also below 10%). But
this error is small compared to the additional modeling uncertainty
introduced by increasing the amount of articial thickening (see
Section 3) to resolve the ame.
The resulting manifold is shown in Fig. 2 where the chemical
source term of carbon dioxide is shown as a function of the two
controlling variables (equivalence ratio has been used instead of
Fig. 1. Evaluation of different progress variables (see Eqs. (1) and (2)) using the results from 1-D simulations of a methane-air ame at T = 300 K, p = 101,325 Pa (GRI 3.0, 300
Gridpoints). Left: Source term thickness. Right: Corresponding error (Eq. (3)) when non-monotonic parts of the progress variable are neglected. Open symbols: error in
density q, lled symbols: error in temperature T.
G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767 1751
mixture fraction in the picture) which are transported by the LES
solver. Outside the ammability limits the extrapolation technique
given by Ketelheun et al. [22] has been applied to obtain a com-
plete manifold for all chemical states. The resulting table consists
of 901 101
1
entries for mixture fraction and progress variable
equidistantly distributed to allow a fast, non-searching table
access.
Throughout this work the Lewis number is assumed to be unity
although especially at higher equivalence ratios there are non-neg-
ligible differences in the laminar ame speed when compared to
complex transport [23]. Besides being a common assumption
[20] which simplies the thermo chemical description by mixture
fraction and progress variable there is good reason to do so regard-
ing the ame turbulence interaction. As mentioned by van Oijen
et al. [15] without the Lewis number being unity variations in
the element mass fractions and the enthalpy due to preferential
diffusion occur. As a result, the mass burning rate decreases further
than can be described by a single progress variable when the ame
is stretched by turbulence. Regarding this issue de Swart et al. [24]
have shown that the effective Lewis number of a methane-air mix-
ture is close to one due to partly canceling preferential diffusion ef-
fects. Using this assumption the effect of stretch and curvature on
the mass burning rate has been investigated for the corrugated
amelet regime [25] and the thin reaction zone regime [26] and
the resulting error has been found to be within 5% when using a
single progress variable.
2.2. The LES solver
The three-dimensional nite volume code FASTEST uses block-
structured, hexahedral, boundary tted grids to represent complex
geometries. Regarding the velocity, spatial discretization is based
on multi-dimensional Taylor series expansion [27] to ensure sec-
ond order accuracy on arbitrary grids. To assure boundedness of
scalar quantities the TVD-limiter suggested by Zhou et al. [28] is
used where the value on the cell face is obtained by its downwind
(index D) and upwind (index U) nodes by
2
U
face
= U
U

1
2
B(r)(U
U
U
UU
) (4)
with the limiter function
B(r) =
r 3r1 ( )
r1 ( )
2
: r > 0
0 : r 6 0
_
r =
U
D
U
U
U
U
U
UU
: (5)
An explicit RungeKutta scheme is used for the time advancement
of momentum and species mass fractions where the temperature
dependent transport coefcients are taken from the chemistry
table. The code solves the incompressible, variable density,
NavierStokes equations where an equation for the pressure correc-
tion is solved within each RungeKutta stage to satisfy continuity.
The solver is based on an ILU matrix decomposition and uses the
strongly implicit procedure [29] to take advantage of the block-
structure. Sub grid uxes of momentum are accounted for by the
eddy viscosity approach proposed by Smagorinsky [30] where the
model coefcient is obtained by the dynamic procedure of Germano
et al. [31] with a modication by Lilly [32]. A gradient approach has
been chosen for the sub-grid ux of scalar quantities with a
turbulent Schmidt number of 0.7.
3
2.3. Investigation of the coupling
For a quantitative assessment of the coupling, one-dimensional
ames have been computed with the LES solver. The aim is to ob-
tain the necessary conditions regarding the grid spacing which al-
low a proper representation of the ame structure and propagation
speed. Therefore several simulations over a range of equivalence
ratios have been carried out using successive coarsening starting
from a very ne grid. Figure 3 shows the grid dependence of the
ame structure for a equivalence ratio of / = 0.8 where the
E
q
u
i
v
a
l
e
n
c
e

r
a
t
i
o
0
0.5
1
1.5
2
CO2 mass fraction
0
0.05
0.1
C
O
2
s
o
u
r
c
e

t
e
r
m
0
50
100
Temperature
2200
1900
1600
1300
1000
700
400
Fig. 2. Manifold of a methane-air ame dened by two controlling variables
showing the source term of carbon dioxide (kg/m
3
s) colored with temperature (K).
Fig. 3. Flame structure represented by density (q), chemical source term ( _ xCO2
) and
CO
2
mass fraction (YCO2
) obtained on four different grids with FASTEST in
comparison with the detailed chemistry solution (Chem1D).
1
The memory requirements for the eight variables (source term, viscosity and
density plus ve post-processing variables like temperature) stored in the table are
6MB for each CPU.
2
for simplicity written for an equidistant grid here.
3
Except in the reaction layer where the unresolved scalar ux is treated by the
combustion model.
1752 G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767
Chem1D solution has been added as a reference. Here the density
(q), CO
2
source term ( _ x
CO
2
) and CO
2
mass fraction (Y
CO
2
, the trans-
ported control variable) through the ame front are shown for four
different grid sizes. One can see that above D
x
~ 0.17 mm strong
deviations occur and especially the source term is poorly predicted.
Since the propagation speed of the ame behaves as s
l

_

_ xdx
errors are expected which can be seen on the left in Fig. 4 where
the ame speed is shown as a function of the grid spacing. The
ame speed s
lD
introduced here will be used in the remainder of
this work to distinguish between the effective ame propagation
obtained on the numerical grid and the correct laminar ame
speed s
l
. It was obtained by three different methods leading to con-
sistent results. First, from the fulllment of continuity
s
lD
=
u
b
u
u
q
u
q
b
1
(6)
where u
b
and u
u
denote the burnt and unburnt gas velocity respec-
tively. Second, from spatial propagation
s
lD
= u
u

x
f
t
2
( ) x
f
t
1
( )
t
2
t
1
(7)
where x
f
denotes the ame position determined by Y
CO
2
= 0:5Y
eq
CO
2
at
time t
1
and t
2
, respectively. Third, from the scalar transport equa-
tion which yields in the ame reference frame
s
l
=
1
q
u
Y
CO
2
;b
_

_
x
CO
2
dx: (8)
To ensure that the results are independent of boundary conditions
different inlet velocitiesbelow as well as above the correct ame
speed of 24.8 cm/swere used. As expected, above D
x
~ 0.17 mm
the ame speed is neither independent of the grid size nor of the in-
let velocity. It is interesting to note that the ame speed converges
towards the unburnt gas velocity for very coarse grids. This is
emphasized on the right of Fig. 4 where the grid dependent devia-
tion from the correct ame speed has been normalized using the
unburnt gas velocity. The reason for this behavior is the increasing
error of the midpoint rule used to evaluate the source term in the
context of the nite volume method which is
s
l
=
1
q
u
Y
CO
2
;b
_

_ x
CO
2
dx ~ s
lD
=
1
q
u
Y
CO
2
;b

i
_ x
CO
2
;i
D
x;i
(9)
where the index i runs over all control volumes. For large
D
x
0.17 mm mainly one grid point at the ame front (index ff)
contributes to the sum leading to
Fig. 4. Left: Grid dependence of the ame speed s
lD
. Different unburnt gas velocities illustrate the numerical uncertainty if resolution requirements are not considered. Right:
Grid dependent deviation from the correct ame speed normalized using the unburnt gas velocity.
Fig. 5. Four time steps of chemical source term and CO
2
mass fraction to illustrate the stagnation of the ame when using insufcient resolution.
Fig. 6. Flame position and propagating speed over time showing the process
starting on the unstable part and nally stagnation on the stable part of the source
term (gray line: unburnt gas velocity).
G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767 1753
s
lD
~
1
q
u
Y
CO
2
;b
_ x
CO
2
;ff
D
x;ff
: (10)
Using this equation the behavior of unphysical ame stagnation will
be explained in the following. The 1-D simulation to supplement
the theory uses a typical LES grid size of D
x
= 0.63 mm. The unburnt
gas velocity is u
u
= 50 cm/s (above the correct ame speed of
24.8 cm/s). The initial eld to start the simulation is the correct
ame structure obtained on a sufciently ne grid. This eld was
then interpolated onto the coarse grid. The behavior of the rst time
steps now depends on the location of the contributing node to
Eq. (10). In this example the ame is positioned such that the node
with the major contribution is on the ascending part of the chemical
source term (Fig. 5, t = 0 ms, the area of the bars represents the sum
given by Eq. (9)). Since the distinction of ascending and descending
part is important in this illustration they are colored blue and red,
respectively, in Fig. 5. The corresponding ame position x
f
and ame
speed s
l
given by Eq. (9) are plotted in Fig. 6 over the simulation
time. Since the ame speed in the initial time step is below the un-
burnt gas velocity the ame starts to move backward which further
decreases the ame speed according to Eq. (10) in the next time
steps (see Fig. 6 and t = 0.02 ms in Fig. 5). Therefore, the ascending
part of the source term is unstable since it reduces the ame speed
if the ame is already belowthe unburnt gas velocity and vice versa.
At t = 0.1 ms the ame has moved further backwards such that the
major contribution of the source term is on the descending part one
grid point to the right. The descending part of the source term be-
haves inversely to the ascending part and therefore the ame stag-
nates for t ?as can be seen in Fig. 6. Since the maximum value of
_ x
CO
2
through the ame front given by the chemistry table will not
be exceeded in the simulation this derivation only holds for suf-
ciently coarse grids. Or, in other words, for every unburnt gas veloc-
ity there exists a minimum grid size D
x,min
to stagnate the ame
according to
s
lD
= u
u
=
1
q
u
Y
CO
2
;b
max( _ x
CO
2
)D
x;min
(11)
as observed on the right in Fig. 4.
However, the above derivation just makes clear how the ame
propagation is dominated by numerical uncertainties if the original
ame thickness is not resolved. Of course we need to stay below a
certain grid size but instead of judging from visual impression in
Fig. 4 we want to give a more precise denition of the acceptable
error. Due to the shape of the source term the evaluation of
Eq. (9) will oscillate when the ame passes a grid point where
the amplitude is only a function of the grid spacing. To illustrate
that, Fig. 7 shows the transient evolution of the ame speed
obtained during a 1-D simulation. On the left the denition of
the local error
D
has been added to be the largest deviation from
the correct ame speed s
l
. Using this denition several simulations
over a large range of equivalence ratios, grid spacings and unburnt
gas velocities have been carried out to obtain a universal relation of
this error to the grid spacing which is given in Fig. 8a. Here the rel-
ative error
D
/s
l
is a unique function of the grid spacing related to
characteristic ame length scales. Hence if we want the local error

D
to be below 10% of the correct ame speed, the grid size should
not exceed
D
x;max
~ 0:7 d( _ x
CO
2
) ~ 0:3 d(Y
CO
2
) (12)
where d(Y
CO
2
) is the ame thickness obtained by the gradient of the
CO
2
mass fraction and the thickness of the chemical source term
d( _ x
CO
2
) is represented by its full width at half maximum. To assess
if the denition of the local error is a sufcient measure for the
resolution required we furthermore need an estimate how it affects
the absolute ame speed s
aD
which describes the movement of the
ame relative to the combustor geometry. Therefor some additional
information has been added to Fig. 7. The simulation shown here
uses a unburnt gas velocity twice the correct ame speed and a grid
0 1 2 3
t (ms)
20
22
24
26
28
30
32
0.0 0.2 0.4 0.6
x
f
(mm)
sl
s
l


(
c
m
/
s
)

Fig. 7. Oscillating behavior of the ame speed during a 1-D simulation with u
u
> s
l
as a function of ame position x
f
(left) and time (right). Blue: computed ame speed s
lD
(Eq.
(9)) used in the transport equation; Red: approximation of s
lD
using a sine function (see Eq. (13)). For visualization purpose a coarse grid producing a large error has been
choosen. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)
(a) (b)
Fig. 8. (a) Local error as dened in Fig. 7 as a function of the ratio of gridspacing to the characteristic ame length scales d(YCO2
) (dashed) and d( _ xCO2
) (solid line). (b) Global
error following from the deviation in absolute ame speed as a function of the unburnt gas velocity.
1754 G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767
spacing of D
x,max
(Eq. (12)) < D
x
< D
x,min
(Eq. (11)) and hence the
ame gets pushed back. While the oscillation is relatively symmet-
ric with respect to the ame position x
f
it gets distorted in time. This
happens because the larger contributions reduce the difference be-
tween the ame speed s
lD
and the unburnt gas velocity and will
therefore contribute a larger portion to the absolute ame speed
with respect to time (this is decisive since the ame position fol-
lows from the time integration of s
aD
). Obviously the frequency
and therewith the error in absolute ame speed (s
aD
s
a
)which
we call the global erroris a function of the unburnt gas velocity
which now needs to be quantied. Besides the 1-D simulations
we approximated the ame movement to follow
@x
f
@t
= s
aD
= u
u
s
lD
~ u
u
[s
l
o Asin(2pfx
f
)[ (13)
to derive an analytic relation between the local and the global error
(The offset o and amplitude A are dened in Fig. 7, note that the sine
function has been dened with respect to x
f
so the frequency f is no
additional unknown). As can be seen in Fig. 7 the approximation of
s
lD
does t very well to the real error behavior. The analytic solution
(taken from a standard handbook of mathematics, e.g. [33]) of Eq.
(13) is plotted together with simulation results in Fig. 8b. Here,
two regions can be identied. As outlined above for u
u
s
l
<
D
the ame will not move and hence the global error increases line-
arly. For higher unburnt gas velocities the ame will start to move
and the error decreases due to partly cancelling effects of the oscil-
lating integration error. With increasing u
u
the error becomes more
symmetric in time and hence the global error further decreases un-
til a constant value caused by the offset is reached. Hence, the max-
imum global error will not exceed the local error and therefore Eq.
(12) provides an appropriate criterion for the necessary resolution.
The corresponding maximum grid size is plotted in Fig. 9 for the
one-dimensional and three-dimensional casewhere the correct
ame propagation in a diagonal direction through the cell needs
also to be ensuredas a function of the equivalence ratio. If this grid
size is not exceeded the Chem1D solution can be reproduced very
accurately by the LES solver using transport equations for the con-
trolling variables. This has been veried for different equivalence
ratios as shown in Fig. 10. Here the ame speed obtained by
Chem1D and the LES solver are compared and a very good agree-
ment can be observed.
3. Combustion modeling
As shown in the previous section important properties of the
ame can only be captured by the LES solver below a certain grid
size (D
x
~ 0.1 mm), which is far below typical grid spacings used
in simulations of complex geometries (D
x
~ 1 mm). If these prop-
erties are important as in premixed combustion where the ame
position is determined by its propagation speed, additional model-
ing is required. In this work, the ATF-model is used in combination
with the FGM approach to ensure an accurate coupling on arbitrary
meshes. This approach is illustrated in the following. In addition,
some numerical aspects of the model will be discussed.
3.1. The ATF approach
The principle of the ATF-model is based on a coordinate trans-
formation that is applied to the scalar transport Eq. (14) to thicken
the ame front and therewith make it resolvable on coarse grids.
@qY
k
@t

@
@x
(qu
i
Y
k
) =
@
@x
qD
k
@Y
k
@x
_ _

_
x
k
(14)
This idea has been initiated by Butler and ORourke [34] and
ORourke and Bracco [35] where Eq. (14) is expressed in terms of
n
i
(x
i
) = T x
i
and s(t) = T t nally leading to
@qY
k
@s

@
@n
i
(qu
i
Y
k
) =
@
@n
i
qTD
k
@Y
k
@n
i
_ _

_ x
k
T
(15)
which is a pure mathematical operation without any assumptions.
If one just solves Eq. (15) with respect to t and x instead of s and
n, Y
k
gets thickened by T compared to the solution of Eq. (14).
Accordingly it is possible to resolve the ame and hence obtain
the correct propagation speed on a grid of D
x
if the thickening factor
satises T PD
x
=D
x;max
. To verify this, Fig. 11 represents the grid
dependency as introduced on the left in Fig. 4 for a ame thickened
by the factor ve where the ame speed s
lD
is now grid independent
up to a grid spacing ve times coarser compared to the unthickened
ame. The drawback of this modeling approach is the modied
interaction of the ame front with the turbulent ow eld. It has
been shown by Poinsot et al. [16] and Meneveau and Poinsot [36]
by means of DNS that the efciency of a vortex to wrinkle the ame
depends on the ratio of its radius r and the ame thickness d which
is obviously decreased by T. Therefore an efciency function S has
been derived [3739] to increase the ame speed and compensate
Fig. 9. Maximum grid size according to Eq. (12) to ensure the correct ame
propagation normal (1-D) and diagonal (3-D) through a cell as a function of
equivalence ratio (/).
Fig. 10. Flame speed as a function of equivalence ratio (/) obtained with the LES
solver FASTEST using two controlling variables in comparison with the detailed
chemistry solution (D
x
6 D
x,max
).
Fig. 11. Grid dependence of the ame speed s
lD
using a thickening factor of ve.
Filled symbols: constant thickening using T = 5, open symbols: dynamic thicken-
ing using Tmax = 5.
G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767 1755
for the lost ame surface caused by the thickening. Since the ame
speed and thickness follow
s
l

D
_
x
_
; d

D=
_
x
_
(16)
the diffusion as well as the source term are multiplied by S to keep
the thickness constant resulting in the nal transport equation
@qY
k
@t

@
@x
i
(qu
i
Y
k
) =
@
@x
i
qTSD
k
@Y
k
@x
i
_ _

S
T
_
x
k
(17)
which propagates a ame of thickness d
1
l
= Td
0
l
with the turbulent
ame speed s
T
= Ss
l
. In this work the efciency functions derived by
Colin et al. [38] as well as Charlette et al. [39] are used. Both formu-
lations utilize DNS of ame vortex interaction to characterize the
unresolved ame surface and have been implemented in the nite
volume code of Section 2.2. They are summarized in the following.
3.1.1. The original formulation by Colin et al
Colin et al. [38] dened the efciency function to be the ratio
between the wrinkling factor (i.e. the ame surface divided by its
Fig. 12. Comparison of a dynamic and constant thickened ame using Tmax = 5 and T = 5 respectively. An unmodied ame has been added. The ame sensor X (gray line)
is shown using the right y-axis. Solutions have been obtained using a very ne grid (lines) and a resolution according to Dx = Tmax Dx;max (points).
Fig. 13. Flame thickness obtained by the CO
2
mass fraction (YCO2
), temperature (T), density (q), and viscosity (l) for different thickening factors using constant (lines) and
dynamic (points) thickening. The thickness of the chemical source term ( _ xCO2
) is represented by its full width at half maximum.
1756 G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767
projection in the propagation direction) of the original ame of
thickness d
0
l
and the thickened ame of thickness d
1
l
(Eq. (18))
which is a natural consideration to account for the lost ame sur-
face. This wrinkling factor is based on the ratios of the test lter
D
e
~ 10D
x
and the ame thickness, and the velocity uctuation
u
/
De
acting on the test lter D
e
and the laminar ame speed s
0
l
.
S =
N d
0
l
_ _
N(d
1
l
)
=
1 aC
D
e
d
0
l
;
u
/
De
s
0
l
_ _
u
/
De
s
0
l
1 aC
D
e
d
1
l
;
u
/
De
s
0
l
_ _
u
/
De
s
0
l
(18)
with C
D
e
d
0;1
l
;
u
/
De
s
0
l
_ _
= 0:75exp 1:2=
u
/
De
s
0
l
_ _
0:3
_ _
D
e
d
0;1
l
_ _
2=3
(19)
The velocity uctuation u
/
De
is based on a similarity assumption
leading to u
/
De
= 2D
3
x
[\(\
2
~u)[ where the Laplace operator is eval-
uated on a 4D
x
stencil because the amplitude of its discrete Fourier
transform represents a good lter for the relevant length scales of
ame vortex interaction. The parameter a can be estimated with
the turbulent Reynolds number according to Eq. (20) where b is a
model constant in the order of unity.
a = b
2ln(2)
3c
ms
(Re
1=2
t
1)
; c
ms
= 0:28; Re
t
=
u
/
l
t
m
(20)
3.1.2. The power-law formulation by Charlette et al
The power law ame wrinkling model by Charlette et al. [39]
describes the unresolved ame surface (i.e. the efciency function)
in terms of an inner and outer cutoff scale which represent the
range of length scales that require modeling regarding the ame
turbulence interaction.
E
D
d
0
l
;
u
/
D
s
0
l
; Re
D
_ _
= 1 min
D
d
0
l
; C
u
/
D
s
0
l
_ _ _ _
c
(21)
Following the relationship between strain-rate and the energy spec-
trum the function C has been obtained by numerical integration
and the t
C
fit
D
d
0
l
;
u
/
D
s
0
l
; Re
D
_ _
= ((f
a
u
f
a
D
)
1=a
)
b
f
b
Re
_ _
1=b
(22)
is provided for practical use with:
f
u
= 4
27C
k
110
_ _
1=2
18C
k
55
_ _
u
/
D
s
0
l
_ _
2
(23)
f
D
=
27C
k
p
4=3
110
D
d
0
l
_ _
4=3
1
_
_
_
_
_
_
_
_
1=2
(24)
f
Re
=
9
55
exp
3
2
C
k
p
4=3
Re
1
D
_ _ _ _
1=2
Re
1=2
D
(25)
a = 0:6 0:2exp 0:1(u
/
D
=s
0
l
)
_
0:2exp 0:01(D=d
0
l
)
_
; b = 1:4
(26)
Herein C
k
= 1.5 is the Kolmogorov constant, D = Td
0
l
is the lter size
and Re
D
= 4(D=d
0
l
)(u
/
D
=s
0
l
) the corresponding sub-grid turbulent
Reynolds number. The velocity uctuation u
/
D
is obtained as
suggested by Colin et al. [38] and the exponent is set to c = 0.5
according to the non-dynamic formulation of the model.
3.2. Dynamic thickening
To ensure that pure mixing is accurately predicted the thicken-
ing should be limited to the ame front. Since the burner presented
in Section 4 includes the mixing of pure air with the premixed
reactants as well as with the burnt gas the use of a dynamic thick-
ening introduced by Legier et al. [7] is necessary. The idea of a spa-
tially varying thickening factor has already been introduced by
Butler and ORourke [34] and it is demonstrated in [35] that the
scalar transport equations remain invariant under the transforma-
tion to ensure the correct ame propagation. The procedure is to
use a more general denition of the new coordinate n(x) normal
to the ame according to
n(x) =
_
x
T(x
/
)dx
/
(27)
where the local thickening factor follows to be
@n
@x
= T(x). If Eq. (27)
is unique the spatial derivatives can be exchanged using
@
@x
=
@
@n
@n
@x
= T(x)
@
@n
: (28)
Further assuming that the spatial variation of species is dominating
in the ame normal direction the time derivative can be expressed
in terms of the absolute ame speed via
@
@t
=
@x
F
@t
..
uus
l
@
@x
: (29)
Inserting Eqs. (28) and (29) into Eq. (14) yields
0 0.01 0.02 0.03 0.04
0
0.005
0.01
0.015
0.02 400
350
300
250
200
150
100
50
0
-50
F=2 (Constant)
0 0.01 0.02 0.03 0.04
0
0.005
0.01
0.015
0.02 400
350
300
250
200
150
100
50
0
-50
F=5 (Constant)
0 0.01 0.02 0.03 0.04
0
0.005
0.01
0.015
0.02
400
350
300
250
200
150
100
50
0
-50
F=5 (Dynamic)
Fig. 14. Three snapshots of two-dimensional ame vortex interaction to illustrate
the reduced ame turbulence modication by the dynamic thickening. Contour of
the vorticity (1/s) to visualize the vortex and isolines of chemical source term to
mark the reaction zone. The snapshots correspond to t = 0.04 s in Fig. 15.
G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767 1757
@qY
k
@n
@x
F
@t

@
@n
(quY
k
) =
@
@n
qT(x)D
k
@Y
k
@n
_ _

_ x
k
T(x)
(30)
which compared to the original equation
@qY
k
@x
@x
F
@t

@
@x
(quY
k
) =
@
@x
qD
k
@Y
k
@x
_ _
_ x
k
(31)
predicts the same absolute ame speed
@x
F
@t
in theame resolving
coordinate n(x). No physics have been changed so far. The step of
articial thickening is now to solve Eq. (30) with respect to x instead
of n which elongates the coordinate n onto the x-grid. Thereby the
ame gets thickened and, if T varies spatially, the ame structure
gets distorted as will be illustrated below. To dynamically deter-
mine the thickening factor we use the formulation for the ame
sensor X suggested by Durand and Polifke [40] to detect the ame
and apply the thickening according to:
X = 16[c(1 c)[
2
; c = Y
CO
2
=Y
eq:
CO2
; T = 1 (T
max
1)X: (32)
This local thickening ensures the correct ame propagation up
to a grid size similar to using T
max
everywhere (also shown in
Fig. 11) but since the thickening varies through the ame front
the ame structure is not identical. This is illustrated in Fig. 12
where the ame structures of a dynamically thickened ame using
T
max
= 5 and a constantly thickened ame with T = 5 are com-
pared. The preheating and oxidation zone of the dynamically thick-
ened ame are thinner while the sensor is almost one in the
reaction zone ensuring a correct ame propagation according to
Eq. (9). Two resolutions are shown in Fig. 12. To exclude numerical
uncertainties a very ne mesh (resolving a ame with T = 1) has
been used to assess the impact onto the ame structure when
Eq. (32) is used in Eq. (17). The second resolution satises
D
x
= T
max
D
x;max
which is the more relevant case for practical
use. Using a constant thickening both resolutions produce the
same result because all gradients are smooth compared to
the mesh. Differences occur when the dynamic thickening is
applied. While the source term is well resolved (as with constant
thickening) the preheating zone remains very sharp and gets
slightly attened because of the resolution limit. The impact of
the dynamic thickening procedure on the ame turbulence interac-
tion is twofold.
First, since the overall ame is thinner, the ame becomes more
sensitive to vortices again. An unthickened ame has been added
to Fig. 12 to illustrate that especially the change of density and
viscositywhich play a major role regarding the mechanism of
dissipating a vortexis not as modied as in a ame with
global thickening since it mostly occurs in the preheating zone.
Figure 13 shows the ame thickness represented by different vari-
ables. In the upper two graphs the gradient of the variables has
been used to obtain the thickness d
F
gradient
of the respective layer be-
tween the burnt and unburnt state. The two graphs below show
the corresponding thickness represented by the distance between
15% and 85% (d
F
15%85%
) of variation through the ame front. These
graphs have been added since the smooth s-shape of the variables
through the ame front is partly distorted by the dynamic thicken-
ing procedure and hence the gradient may not be the optimal
choice to characterize the ame thickness. In addition, the evalua-
tion of the gradient on the coarse mesh introduces an additional
error. The threshold of 1585% has been chosen since it leads to
a similar thickness as the gradient for the constantly thickened
ame. As expected the constantly thickened ame behaves linearly
with the thickening factor. This only holds for the dynamically
thickened ame when the thickness d
F
15%85%
is used. This demon-
strates that it is more appropriate in this case. Like in Fig. 12, two
grids have been considered to show its impact. Comparing the two
resolutions, the thickness is increased on the coarse mesh which is
consistent with the visual impression from Fig. 12. Nevertheless,
the dynamically thickened ame is still much thinner than the con-
stantly thickened ame. To illustrate the impact of this behavior,
two dimensional simulations of ame vortex interaction have been
carried out. Following previous works [38] the stream function
given by Poinsot et al. [16] has been used to initialize the velocity
eld. The comparison of three ames interacting with a vortex of
same strength is given in Figs. 14 and 15. Figure 14 shows snap-
shots taken at the same instant of time where the above mentioned
inuence of dynamic thickening can be observed. Regarding the
constant thickening the results are similar
4
to the simulations done
by Colin et al. [38] to derive the efciency function: The thicker
ame gets less distorted by the vortex. If the same thickening factor
is used with a dynamic thickening (last snapshot in Fig. 14) an
increased interaction of the ame with the vortex can be observed
which is quantied in Fig. 15. Here the total reaction rate x (inte-
grated over the computational domain x =
_
_ xdA) together with
Fig. 16. Increase of turbulent ame speed s
T
by the efciency function based on
constant and dynamic thickening.
Fig. 15. Reduced total reaction rate x (lines) and available ame surface k (symbols) during the ame vortex interaction shown in Fig. 14.
4
Since not all parameters of the simulations done by Colin et al. [38] are known the
results are not identical.
1758 G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767
the available ame surface k (obtained from the iso-surface
Y
CO
2
= 0:5Y
eq
CO
2
) normalized with the values of a freely propagating
ame are plotted. As expected by visual impression from Fig. 14,
the dynamically thickened ame with T
max
= 5 behaves like an
intermediately thickened ame between the simulations using a
constant thickening with T = 2 and T = 5. This observation is in
agreement with the study done very recently by Auzillon et al.
[41] where the effect of thickening on the ame structure and
dynamics was investigated in the context of the F-TACLES model
of Fiorina et al. [5]. They showed, that by ltering the ame by a
gaussian function the thermal thickness of the ame is much less
increased compared to the thickness of the chemical source term
which yields signicant improvements regarding the prediction of
ame turbulence interaction. As shown by Colin et al. [38] the
difference between different thickening factors in Fig. 15 is less pro-
nounced for higher ratios of vortex velocity to the laminar ame
speed.
Second, together with the thickening factor, both efciency
functions (Eqs. (18) and (21)) tend towards unity outside the ame
which is a necessary condition since otherwise Eq. (17) would still
be modied in regions of pure mixing. Hence, for a given velocity
uctuation u
/
De
, the efciency function is not constant through
the ame front
5
which inuences its impact on the propagation
speed. To quantify this effect one-dimensional simulations have
been carried out where the velocity uctuation u
/
De
has been set to
a constant value (of course u
/
De
would be zero in a 1-D ame if
u
/
De
= 2D
3
x
[\(\
2
~u)[ is used) yielding a given efciency S
max
=
S(X = 1). Figure 16 shows the results of these simulations
obtained with different values for S
max
(u
/
De
). As one can see the
dynamically thickened ame is slower than the ame with constant
thickening which uses S = S
max
everywhere. The results only slightly
depend on the thickening factor or the formulation for the efciency
function used, so only one graph is shown here.
The second of the effects mentioned above can be interpreted as
a reduction of the efciency function to account for the reduced
modication of ame turbulence interaction. For this work, it is as-
sumed that the efciency function arising from the dynamic proce-
dure yields appropriate values of turbulent ame propagation
corresponding to an effective ame thickness of the dynamically
thickened ame. Further knowledge about the impact of the
dynamic thickening procedure onto the efciency function is
desirable but necessitates DNS studies of ame turbulence interac-
tion which is beyond the scope of this work.
3.3. Numerical aspects of the thickening procedure
During this work several inuences of the thickening procedure
on the stability and accuracy of the scalar transport have been
Fig. 19. Substep strategy for the scalar transport applied within a 3-stage Runge
Kutta scheme.
(a) (b)
Fig. 17. Illustration of spatial interpolation: (a) Inuence of the thickening procedure on the limiter function B(r), centered scheme has been added as the optimum, (b) CO
2
mass fraction through the ame front at the computational nodes and the corresponding faces (T = 5).
Fig. 18. Left: Ratio of diffusion and convection number for the burnt/unburnt (red/blue) state with/without thickening (full/dashed). The molecular diffusion coefcient and
density have been taken from the burnt and unburnt values of a methane-air ame (/ = 0.8). Right: Corresponding theoretical speedup and number of substeps applied
within a 3-stage RungeKutta scheme for the ATF simulation. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of
this article.)
5
The preheating zone gets slightly compressed by this.
G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767 1759
observed. A positive feature of the model is the reduction of
gradients. Compared to the mesh the original ame is like a sharp
interface and hence gets attened by numerical inaccuracies as one
can see in Fig. 3. To keep the solution stable the limiter function
(Eq. (4)) reduces the order of spatial accuracy which increases
numerical diffusion to avoid unphysical oscillations. The introduc-
tion of an explicit thickening reduces this numerical diffusion and
therewith uncertainties inuencing the ame propagation. In
Fig. 17a the limiter function is shown with and without thickening
to illustrate the improved order of accuracy through the ame
front. Even on this relatively ne grid of D
x
= 0.3 mm the deviation
from a pure centered scheme is much smaller for the thickened
ame. In Fig. 17b the CO
2
mass fraction of the thickened ame is
shown together with the values given by Eq. (4) on the control vol-
ume faces indicating almost 2nd order interpolation.
Besides this expected behavior of spatial interpolation, instabil-
ities regarding the time integration occurred after the thickening
procedure was applied. Following classical stability analysis (e.g.
[42]) basically two parametersthe diffusion number T and the
Courant number carise, drawing restrictions for the explicit time
integration.
T =
DD
t
D
2
x
; c =
uD
t
D
x
(33)
The exact value of these numbers determining the largest possible
time step D
t
depends on the stability function of the corresponding
scheme as well as on the inuence of non-linearities (e.g.D(U)).
However, in this section we just want to point out how the ATF-
model may alter the scalar transport equation to be diffusion
dominated and how we dealt with it to keep the method computa-
tionally efcient. Without the ATF-model, the ow is in general
dominated by convection and hence the Courant number is decisive
for the time step size. But the diffusion coefcient D is increased by
the thickening factor and efciency function (Eq.(17)). Assuming
typical values of T ~ 10 and S ~ 3 this leads to T
ATF
= STT ~ 30T
indicating that the diffusion number may exceed the Courant num-
ber as is illustrated on the left of Fig.18. Here the ratio of the diffu-
sion number and Courant number (Eq.(33)) in the burnt and
unburnt state of a methane-air mixture (/ = 0.8) are shown with
and without application of the ATF-model. This plot has been ob-
tained assuming a velocity eld yielding a uctuation of
u
/
De
= 2 m=s at a reference grid spacing of 0.5 mm. The velocity to
obtain the Courant number has been set to 5 m/s. Hence the situa-
tion corresponds to the application of different grids for the same
ow conguration. A different velocity eld would shift the graphs
but the behavior would remain the same.
6
Below D
x
~ 0.1 mm the
ame is fully resolved and the function follows 1/D
x
while with
increasing grid size the ame needs to be thickened (T D
x
) and
hence the ratio remains constant. Further coarsening of the grid in-
creases the efciency function (through the increase of the thicken-
ing factor as well as the velocity uctuation u/
De
) which converges
towards S T
2=3
D
2=3
x
(when using Eq.(18)). This causes rising
Fig. 20. Geometry of the Tecam bluff-body swirl nozzle.
Coflow (pure air)
Inlet
(methane-air)
300mm
7
1
7
m
m
1
1
7
m
m
Measurement Planes
Monitoring
point
Fig. 21. Dimensions of the computational domain with a cut-out of the elliptically
smoothed mesh. For visualization purposes, the mesh is coarser than the one used
in the simulations.
6
Only for very high velocities or if the speed of sound is restrictive for the
numerical scheme the Courant number may dominate again.
1760 G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767
ratios of the diffusion and convection number with increasing grid
size which strongly deviates from the behavior without the ATF-
model where diffusion becomes less important. This inuence of
the combustion model on the scalar transport equation results in
stability restrictions on the maximal possible time step size, making
the simulation computational too expensive. Increasing the number
of stages of well established RungeKutta schemes does not seem to
be promising since the computational costs rise while the stability
region is only slightly extended (e.g.[43,44]). Since only the trans-
port equation for the scalars is modied, the momentum equation
alone would still be stable and diverges only as a result of the incor-
rect prediction of density and viscosity due to unstable scalar trans-
port when the time step is not sufciently decreased. Therefore,
multiple sub steps for the scalar transport equation assuming a con-
stant velocity in the convective term are used within a time step (in
this work within every RungeKutta stage) as illustrated in Fig. 19.
This procedure is very efcient since only for the correction of the
velocity a system needs to be solved to satisfy continuity. Since
the solver takes most of the computation time it is desirable to keep
the time step as large as the momentum equation allows, which
was possible with the substep strategy. The resulting speedup
dened by the ratio of computing time needed with and without
substeps to simulate the same pyhsical time is shown on the right
of Fig. 18. The graph has been obtained by extrapolating the speed-
up gained for the swirl burner of Section 4 using six substeps with a
grid size of D
x
= 0.5 mm in the reaction zone whereby the comput-
ing time for the individual code sections has been measured. Hence,
outside this region its a theoretical speedup that has not been ver-
ied. Several one and two-dimensional test cases of reacting and
non-reacting ows showed that the method is at least as accurate
as without substeps while allowing larger time steps when diffu-
sion becomes important.
4. Application to a premixed swirl burner
4.1. Conguration and numerical setup
The conguration investigated is the Tecamswirl burner of the
Institute of Energy and Power Plant Technology as shown in
Fig. 20a. Measurements of the velocity were performed by
Schneider et al. [45] using Laser Doppler Anemometry with mag-
nesium oxide as seeding material. Raman/Rayleigh scattering has
been used by Gregor et al. [46] to obtain the temperature and main
species distribution. The instantaneous data of this study also ver-
ied the amelet-like behavior of the ame structure. As illus-
trated in Fig. 20a, the air enters the conguration from the
bottom where methane is injected using a perforated ring line.
Then the methane-air mixture (temperature = 300 K, equivalence
ratio = 0.83) is deected by 90 to enter the radial and tangential
channels (see Fig. 20b) where the swirl is generated. Hereafter
the fuel moves upward again, passes the annulus around the bluff
0
2
4
6
8
0 1 2 3 4 5 6
r/R (-)
1mm
0
2
4
6
8
10mm
0
2
4
6
8
20mm
0
2
4
6
8
30mm
0
2
4
6
8
60mm
0
2
4
6
8
90mm
0
2
4
6
8
u
mean
(m/s)
120mm
0
1
2
3
0 1 2 3 4 5 6
r/R (-)
1
2
3
1
2
3
1
2
3
1
2
3
1
2
1
2
u
rms
(m/s)
0
2
4
6
8
0 1 2 3 4 5 6
r/R (-)
0
2
4
6
0
2
4
6
0
2
4
6
1
2
3
4
1
2
3
4
0
2
4
w
mean
(m/s)
0
1
2
3
0 1 2 3 4 5 6
r/R (-)
1
2
3
1
2
3
1
2
1
2
1
2
1
2
w
rms
(m/s)
Fig. 22. Isothermal ow. Mean and uctuating part of the axial (u) and azimuthal (w) velocity obtained by the LES (lines) in comparison with experimental data (dots).
R = 15 mm (radius of the buff-body).
G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767 1761
body and enters the unconned section. Here the Reynolds number
based on the bulk velocity (5 m/s) and the hydraulic diameter
(D
h
= 30 mm) is Re = 10,000. The swirl number which represents
the ratio of azimuthal and axial momentum at the nozzle exit is:
Fig. 23. Illustration of instantaneous ame turbulence interaction. (a) Isosurface of the chemical source term and a slice extracted at z = 0 mm showing the temperature eld
(K). (b) Slice extracted at x = 15 mm above the bluff body showing the axial velocity (m/s) and isolines of temperature (increasing to the centerline: T = 350, 1560 and 1840 K)
marking the preheating and the reaction zone.
0
2
4
6
8
0 1 2 3 4 5 6
r/R (-)
1mm
0
2
4
6
8
10mm
0
2
4
6
8
20mm
0
2
4
6
8
30mm
0
2
4
6
8
60mm
0
2
4
6
8
90mm
0
2
4
6
8
u
mean
(m/s)
120mm
0
1
2
3
0 1 2 3 4 5 6
r/R (-)
1
2
3
1
2
3
1
2
3
1
2
3
1
2
1
2
u
rms
(m/s)
0
1
2
3
0 1 2 3 4 5 6
r/R (-)
-1
0
1
2
3
0
1
2
3
4
0
1
2
3
4
0
1
2
3
0
1
2
0
1
2
v
mean
(m/s)
0
1
2
0 1 2 3 4 5 6
r/R (-)
1
2
3
1
2
3
1
2
3
1
2
3
1
2
1
2
v
rms
(m/s)
Fig. 24. Reacting ow. Mean and uctuating part of the axial (u) and radial (v) velocity obtained using the efciency function of Charlette et al. (solid lines) and Colin et al.
(dashed lines) in comparison with experimental data (dots). R = 15 mm (radius of the buff-body).
1762 G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767
S =
_
Ra
R
i
(uwu
/
w
/
)r
2
dr
D
h
_
Ra
R
i
(u
2
u
/2
)rdr
~ 0:7 (34)
As depicted in Fig. 21 the nozzle is placed concentrically inside a co-
ow of pure air issuing with 0.5 m/s. The dimensions and measure-
ment planes (velocity components have been measured in planes
ranging from 1 mm up to 120 mm above the swirler exit and spe-
cies data is available at 10 mm up to 60 mm) are also given in
Fig. 21.
The ame, which stabilizes by the recirculation of hot gases
above the bluff body, has a thermal power of 30kW and covers
the range of 1 < Ka < 4 and 4 < Da < 20 which leads to the regime
diagram classication of a thickened wrinkled ame.
The block-structured grid contains 3.2 million control volumes
and has been elliptically smoothed to obtain a better orthogonality.
The grid has been rened towards the near nozzle region whereas
it gets coarser with increasing distance to spare cells. Since the
ame is very compact it was possible to maintain almost cubic
cells with an edge length of 0.5 mm covering the reaction zone.
Hence a spatial constant thickening factor of T
max
= 5 was used
in combination with the dynamic thickening outlined in Section
3.2. A time step size of D
t
= 18ls has been chosen leading to a Cou-
rant number of c
3D
= 0:7
7
(temporal average of spatial maximum).
The diffusion number, increased by the thickening procedure, was
above T
3D
= 3 and therewith exceeding the stability limit of the em-
ployed RungeKutta scheme which is T
3D
< 0:63. Therefore, accord-
ing to Section 3.3, six substeps have been applied within every time
step of the momentum equation to keep the simulation stable. It
should be noted that the use of dynamic thickening restricts the
destabilizing effect of the ATF-model to a smaller spatial region
(i.e. the ame front) and due to the dependency of the diffusion
number on the grid size temporal variation will occur on non-
uniform grids. Nevertheless, despite this stabilizing effect the
simulation became unstable without using the substep strategy or
unaffordable small time steps. Previous studies of this burner
[4749] showed that the ow eld of the isothermal case can be pre-
dicted sufciently accurate with the RANS approach while large
deviations from experimental data occurred in the reacting case
even with tuning of model parameters. Especially the ame turbu-
lence interaction eludes a time averaged description and requires a
higher spatial and temporal resolution. For this reason LES is a prom-
ising tool to improve the prediction capabilities of CFD in such react-
ing ows.
4.2. Results
Since this work focuses on combustion, only a short summary of
the isothermal case will be given in order to show that the
non-reacting ow eld can be well predicted by the LES and the
7
The three-dimensional complement to c in Eq. (33).
0
2
4
6
8
0 1 2 3 4 5 6
r/R (-)
1mm
0
2
4
6
10mm
0
2
4
6
20mm
0
2
4
6
30mm
1
2
3
4
60mm
1
2
3
4
90mm
0
2
4
w
mean
(m/s)
120mm
0
1
2
3
0 1 2 3 4 5 6
r/R (-)
1
2
3
1
2
3
1
2
1
2
1
2
1
2
w
rms
(m/s)
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0 1 2 3 4 5 6
r/R (-)
10mm
0.00
0.01
0.02
0.03
0.04
0.05
0.06
20mm
0.00
0.01
0.02
0.03
0.04
0.05
0.06
30mm
0.00
0.01
0.02
0.03
0.04
0.05
0.06
z
mean
(-)
60mm
0.000
0.005
0.010
0.015
0.020
0 1 2 3 4 5 6
r/R (-)
0.000
0.005
0.010
0.015
0.020
0.000
0.005
0.010
0.015
0.020
0.000
0.005
0.010
0.015
0.020
z
rms
(-)
Fig. 25. Reacting ow. Mean and uctuating part of the azimuthal velocity (w) and mixture fraction (z) obtained using the efciency function of Charlette et al. (solid lines)
and Colin et al. (dashed lines) in comparison with experimental data (dots). R = 15 mm (radius of the buff-body).
G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767 1763
numerical setup used. A more extensive study regarding the
isothermal case of this conguration using RANS and LES can be
found in [48]. The main part of this section will be devoted to
assess the capability of the combustion model in predicting this
type of reactive ow. Therefore all three velocity components as
well as species mass fractions and temperature will be compared
with experimental data.
4.2.1. Isothermal case
In Fig. 22 the simulation results of axial and azimuthal veloci-
ties are compared with experimental data. As is typical for this
type of swirl ow, the velocity eld starts to expand right after
the nozzle exit, which produces a positive pressure gradient in
the axial and radial direction, high enough to form a central recir-
culation zone as intended for ame stabilization. In the rst mea-
surement plane (1 mm) directly at the nozzle exit the mean and
uctuating part of the velocity are well predicted. This indicates
that the inclusion of upstream geometry is sufcient to allow the
turbulent structures to form without articial forcing. Regarding
the time averaged quantities excellent agreement can be observed
in all planes further downstream. The uctuating part is also well
predicted even though it is slightly underestimated in most of the
measurement planes. In general, this is expected since only the re-
solved part of the uctuations is shown but since the amount of
sub-grid uctuations is unknown it would be too speculative to de-
vote the deviations to it. Overall the spreading of the turbulent
swirling ow and the size and intensity of the recirculation zone
are accurately predicted which is a necessary requirement to as-
sess the combustion model in the reacting case.
4.2.2. Reacting case
To give an illustration of the ame stabilization and instanta-
neous burning behavior two snapshots are shown in Fig. 23. As
one can see from the iso-surface of the chemical source term in
Fig. 23a the ame has stabilized above the bluff body where the
methane-air mixture issuing from the annulus is consumed. A slice
colored with the temperature has been added to illustrate the
burnt and unburnt state. As expected in a turbulent ow eld
the iso-surface is strongly distorted by vortices. Even though the
ame turbulence interaction acting on the smaller scales is sup-
pressed by the thickening procedure the increased ame surface
due to wrinkling is signicant. Figure 23b shows a slice extracted
normal to the axial direction (i.e. perpendicular to Fig. 23a) to illus-
trate how the ame is embedded in the turbulent ow eld. Going
from higher radii to the centerline the contour of the axial velocity
reects the almost laminar coow of 0.5m/s, the turbulent ow
issuing from the annulus around the bluff body and the central
recirculation zone. The isolines of temperature added in Fig. 23b
to mark the preheating and reaction zone show that the ame
has stabilized in the inner shear layer wherein a time averaged
sensethe turbulent ame speed matches the velocity component
normal to the ame front. As one can see the reaction zone is
Fig. 26. Reacting ow. Mean and uctuating part of the CO
2
mass fraction (YCO2
) and temperature (T) obtained using the efciency function of Charlette et al. (solid lines) and
Colin et al. (dashed lines) in comparison with experimental data (dots). R = 15 mm (radius of the buff-body).
1764 G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767
almost of constant shape while the thickness of the preheating
zone varies strongly. This should be viewed more as a qualitative
illustration rather than a quantitative assessment because some
of the deviations are devoted to three-dimensional effects, judging
from Fig. 23a, a non-negligible amount.
In Figs. 2427 the simulation results are compared with exper-
imental data. Since the efciency function is the main model
parameter, different formulations as proposed by Charlette et al.
[39] and Colin et al. [38] have been used in these simulations to
investigate its impact on the results. The model parameter a in
Eq. (20) has been set to a constant value of 0.08 based on measured
turbulent scales. In general, the velocity and scalar elds are well
predicted by both simulations. Compared to the isothermal case
all velocity components are increased by the thermal expansion
which is reproduced by the simulation very well. Except for the
overestimated radial velocity, only minor deviations from the mea-
surements exist which cannot be clearly attributed to the combus-
tion or turbulence model.
As in the experiment the mixture fraction (Fig. 25 right) drops
in the radial direction caused by the transition from the swirled
annular methane-air jet into co-owing air. The high standard
deviationwhich is almost identical in the simulation and experi-
mentindicates that the mixing layer is dominated by turbulent
diffusion. Since the mixture fraction is a non-reacting scalar the
slight deviation from the measurements in radial direction can
only be caused by the overestimated radial velocity. Spreading
and broadening of the mixing layer in downstream direction are
well predicted. It should be noted that most of the fuel consump-
tion occurs at constant equivalence ratio and only the outer edge
of the ame burns under stratied conditions by the dilution with
the co-owing air.
On the left of Fig. 26 the CO
2
mass fraction is compared with the
measurements. It is of major interest since it is the transported var-
iable to describe the reaction where the combustion model enters
explicitly. Since the ame is very sharp the distribution is almost
completely caused by the ame turbulence interaction resulting
in a wide turbulent ame brush. Indeed a bimodal distribution
with only very few intermediate states has been reported in [46].
Except that a small radial offset exists and the gradient at higher
axial positions is slightly overpredicted the ame brush was well
captured in the simulations. As expected from the thickening pro-
cedure the standard deviation is too low at 10 mm because more
intermediate states exist. The deviation from the experiments de-
creases at the higher axial positions since the ratio of the ame
brush thickness to the ame thickness increases. At higher radial
positions a second maximum of the CO
2
mass fraction exists, com-
ing along with a second peak in the standard deviation. This phe-
nomenon caused by the recirculation of hot gases over the outer
swirler plate has only been partly captured by the simulation.
However, the large scatter of the experiments indicates that it is
0.00
0.01
0.02
0.03
0.04
0.05
0 1 2 3 4 5 6
r/R (-)
10mm
0.00
0.01
0.02
0.03
0.04
0.05
20mm
0.00
0.01
0.02
0.03
0.04
0.05
30mm
0.00
0.01
0.02
0.03
0.04
0.05
Y
CH
4
,mean
(-)
60mm
0.00
0.01
0.02
0 1 2 3 4 5 6
r/R (-)
0.00
0.01
0.02
0.00
0.01
0.02
0.00
0.01
0.02
Y
CH
4
,rms
(-)
0.00
0.05
0.10
0.15
0.20
0.25
0 1 2 3 4 5 6
r/R (-)
0.00
0.05
0.10
0.15
0.20
0.25
0.00
0.05
0.10
0.15
0.20
0.25
0.00
0.05
0.10
0.15
0.20
0.25
Y
O
2
,mean
(-)
0.00
0.02
0.04
0.06
0.08
0 1 2 3 4 5 6
r/R (-)
0.00
0.02
0.04
0.06
0.08
0.00
0.02
0.04
0.06
0.08
0.00
0.02
0.04
0.06
0.08
Y
O
2
,rms
(-)
Fig. 27. Reacting ow. Mean and uctuating part of the CH
4
and O
2
mass fraction (YCH4
and YO2
) obtained using the efciency function of Charlette et al. (solid lines) and Colin
et al. (dashed lines) in comparison with experimental data (dots). R = 15 mm (radius of the buff-body).
G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767 1765
a rare event with large time scales introducing statistical uncer-
tainties (i.e. the samples available from the measurements as well
as the simulation may be insufcient to quantify it correctly).
On the right of Fig. 26 the temperature is compared with exper-
imental data. As outlined in Section 2.1 the temperature is only a
post-processing step and indicates whether the connection to the
two controlling variables can describe the physical situation. In
general, the agreement with the measurements is comparable to
the CO
2
mass fraction, but differences exist regarding the burnt
gas temperature. The main reason for this deviation is thought to
be the water-cooled bluff body with a wall temperature of 353 K.
The heat loss is obvious in the measurements at 10 mm where
the temperature drops to 1730 K on the centerline above the bluff
body. As expected, this cannot be captured by the simulation since
the zero-gradient boundary condition for the controlling variables
corresponds to an adiabatic wall. This could be improved by adding
enthalpy as a controlling parameter into the table and solving the
corresponding transport equation (e.g. [15,19]). As reported in [46]
the temperature drop can be observed at the higher axial position,
too due to the long residence time of the uid in the recirculation
zone.
To further assess the capability of the chemistry reduction to
predict the species distribution the mass fractions of CH
4
and O
2
representing the reactants of the system are compared with the
measurements in Fig. 27. The mean prole of the methane mass
fraction reects its consumption by the chemical reaction on the
left and the mixing with the co-owing air on the right coming
along with a double peak in the standard deviation. Thereby
the uctuation induced by the ame turbulence interaction
signicantly exceeds that of the pure mixing situation. At the high-
er axial positions both the CH
4
and O
2
mass fraction exceed the
measurements indicating insufcient fuel consumption which is
consistent with the CO
2
mass fraction being too low in this region.
Comparing the efciency functions, a clear trend regarding all
ow quantities can be observed when going over from the formu-
lation given by Colin et al. [38] to that of Charlette et al. [39]
(marked with arrows in Figs. 2427). As mentioned in [39] Eq.
(21) results in higher values than Eq. (18) when the model constant
b in Eq. (20) is not adjusted according to the turbulent Reynolds
number (which is not possible since the local turbulent Reynolds
number is unknown in LES). In agreement with this, the mean
ame position is shifted to a more forward position (e.g. position
30 mm in Fig. 26 and CH
4
mass fraction at 60 mm in Fig. 27). As
a result the axial and azimuthal velocity at 1030 mm increase
due to the thermal expansion through the ame front. In addition,
the radial velocity is slightly increased since the ow is directed
further outwards by the forwarded ame position. In general, the
differences between the results obtained with the two formula-
tions are small and hence the model uncertainty will not inuence
the results crucially.
Finally, measurements of the temporal autocovariance will be
used to further assess the simulations. These measurements have
been done by Schneider et al. [45] to quantify the precessing vortex
core (PVC) which rotates above the bluff body. Under isothermal
conditions, this coherent structure contributes a signicant part
to the uctuations while it vanishes in the reacting case. Hence it
is interesting if the simulations are able to capture this transient
phenomenon and its suppression by viscous forces. As in the
experiment, the axial velocity has been monitored at the swirler
exit (the position is given in Fig. 21) to construct the spectra based
on the autocovariance R
uu
= u(t)u(t + s)) ( ) denotes time averag-
ing, s is the temporal shift, the velocity has been monitored for 3 s
corresponding to 114 rotations of the PVC). Results are given in
Fig. 28 where an excellent agreement of the simulations and the
experiment can be observed. The magnitude of energy associated
with the corresponding frequency is well predicted. Regarding
the PVC, a distinct peak at 38 Hz, almost identical in the simulation
and experiment, can be observed in the isothermal case. Even
though the monitoring point is positioned in the cold ow up-
stream of the ame front, this peak does not exist in the reacting
case since the large coherent structure is not able to form under
this condition.
5. Conclusion
The well established thickened ame approach has been cou-
pled with two-dimensional tabulated chemistry to include detailed
chemistry effects into the simulation. The correct coupling
with the LES solver has been veried and a quantication of the
resulting error when neglecting resolution requirements revealed
unacceptable numerical inaccuracies. A separated treatment of
time integration for scalars and velocity improved the computa-
tional efciency to obtain results in an acceptable amount of time.
Findings regarding the dynamic ame thickening have been given
but its impact on the efciency function cannot be quantied with-
out three-dimensional DNS. The application of the model to a
swirled turbulent premixed ame showed good agreement with
experimental data.
Acknowledgments
Computations have been performed on the National Supercom-
puter of the Leibniz-Rechenzentrum der Bayerischen Akademie
der Wissenschaften (HLRB-II Project ID pr47ve) and on the
Hessian High Performance Computer (HHLR) in Darmstadt. We
Fig. 28. Frequency spectrum taken at x = 1 mm, r = 20 mm.
1766 G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767
gratefully acknowledge nancial support by the German Research
Council (DFG) and the German Science Council (WR) through the
Collaborative Research Center SFB 568 and the Cluster of Excel-
lence initiative Center of Smart Interfaces.
References
[1] F. Biagioli, Combust. Theor. Model. 10 (2006) 389412.
[2] W. Lazik, T. Doerr, S. Bake, R.v.d. Bank, L. Rackwitz, in: ASME Turbo Expo. Conf.
Proc., 2008, pp. 797807.
[3] F. Zhao, M.C. Lai, D.L. Harrington, Prog. Energy Combust. Sci. 25 (1999) 437
562.
[4] A.W. Vreman, J.A. van Oijen, L.P.H. de Goey, R.J.M. Bastiaans, Flow Turbul.
Combust. 82 (2009) 511535.
[5] B. Fiorina, R. Vicquelin, P. Auzillon, N. Darabiha, O. Gicquel, D. Veynante,
Combust. Flame 157 (2010) 465475.
[6] L. Selle, G. Lartigue, T. Poinsot, R. Koch, K.U. Schildmacher, W. Krebs, B. Prade, P.
Kaufmann, D. Veynante, Combust. Flame 137 (2004) 489505.
[7] J.P. Legier, T. Poinsot, D. Veynante, in: Proceedings of the Summer Program
2000, Center for Turbulence Research, 2000, pp. 157168.
[8] M. Boileau, G. Staffelbach, B. Cuenot, T. Poinsot, C. Brat, Combust. Flame 154
(2008) 222.
[9] P. Schmitt, T. Poinsot, B. Schuermans, K.P. Geigle, J. Fluid Mech. 570 (2007) 17
46.
[10] G. Boudier, L. Gicquel, T. Poinsot, Combust. Flame 155 (2008) 196214.
[11] U. Maas, S. Pope, Combust. Flame 88 (1992) 239264.
[12] O. Gicquel, N. Darabiha, D. Thvenin, Proc. Combust. Inst. 28 (2000) 19011908.
[13] B. Fiorina, O. Gicquel, L. Vervisch, S. Carpentier, N. Darabiha, Combust. Flame
140 (2005) 147160.
[14] J.A. van Oijen, L.P.H. de Goey, Combust. Sci. Technol. 161 (2000) 113137.
[15] J.A. van Oijen, F.A. Lammers, L.P.H. de Goey, Combust. Flame 127 (2001) 2124
2134.
[16] T. Poinsot, D. Veynante, S. Candel, J. Fluid Mech. 228 (1991) 561606.
[17] Chem1D, A one-dimensional laminar ame code, developed at Eindhoven
University of Technology, <www.combustion.tue.nl/chem1d>.
[18] G.P. Smith, D.M. Golden, M. Frenklach, N.W. Moriarty, B. Eiteneer, M.
Goldenberg, C.T. Bowman, R.K. Hanson, S. Song, WCG Jr., V.V. Lissianski, Z.
Qin, GRI-Mech 3.0, <www.me.berkeley.edu/gri_mech/>.
[19] B. Fiorina, R. Baron, O. Gicquel, D. Thevenin, S. Carpentier, N. Darabiha,
Combust. Theor. Model. 7 (2003) 449.
[20] A.W. Vreman, B. Albrecht, J.A. van Oijen, L.P.H. de Goey, R.J.M. Bastiaans,
Combust. Flame 153 (2008) 394416.
[21] W.J.S. Ramaekers, J.A. Oijen, L.P.H. Goey, Flow Turbul. Combust. 84 (2009)
439458.
[22] A. Ketelheun, C. Olbricht, F. Hahn, J. Janicka, in: ASME Turbo Expo Conf. Proc.,
2009, pp. 695705.
[23] T. Poinsot, D. Veynante, Theoretical and Numerical Combustion, second ed.,
R.T. Edwards, Inc., Philadelphia, USA, 2005.
[24] J. de Swart, G. Groot, J.A. van Oijen, J.H.M. ten Thije Boonkkamp, L.P.H. de Goey,
Combust. Flame 145 (2006) 245258.
[25] J.A. van Oijen, G. Groot, R.J.M. Bastiaans, L.P.H. de Goey, Proc. Combust. Inst. 30
(2005) 657664.
[26] J.A. van Oijen, R.J.M. Bastiaans, L.P.H. de Goey, Proc. Combust. Inst. 31 (2007)
13771384.
[27] T. Lehnhaeuser, M. Schaefer, Int. J. Numer. Methods Fluids 38 (2002) 625645.
[28] G. Zhou, L. Davidson, E. Olsson, in: Fourteenth International Conference on
Numerical Methods in Fluid Dynamics, 1995, pp. 372378.
[29] H.L. Stone, SIAM J. Numer. Anal. 5 (1968) 530558.
[30] J. Smagorinsky, Monthly Weather Rev. 91 (1963) 99164.
[31] M. Germano, U. Piomelli, P. Moin, W.H. Cabot, Phys. Fluids A 3 (1991) 1760
1765.
[32] D.K. Lilly, Phys. Fluids A 4 (1992) 633635.
[33] I.N. Bronstein, K.A. Semendjajew, Handbook of Mathematics, third ed.,
Springer-Verlag GmbH, Heidelberg Germany, 1997.
[34] T. Butler, P. ORourke, in: International Symposium on Combustion, vol. 16,
1977, pp. 15031515.
[35] P.J. ORourke, F.V. Bracco, J. Comput. Phys. 33 (1979) 185203.
[36] C. Meneveau, T. Poinsot, Combust. Flame 86 (1991) 311332.
[37] C. Angelberger, D. Veynante, F. Egolfopoulos, T. Poinsot, in: Proceedings of
the Summer Program 1998, Center for Turbulence Research, 1998, pp.
6182.
[38] O. Colin, F. Ducros, D. Veynante, T. Poinsot, Phys. Fluids 12 (2000) 1843
1863.
[39] F. Charlette, C. Meneveau, D. Veynante, Combust. Flame 131 (2002) 159
180.
[40] L. Durand, W. Polifke, in: ASME Turbo Expo Conf. Proc., 2007, pp. 869878.
[41] P. Auzillon, B. Fiorina, R. Vicquelin, N. Darabiha, O. Gicquel, D. Veynante, Proc.
Combust. Inst. 33 (2010).
[42] C. Hirsch, Numerical computation of internal and external ows,
Fundamentals of Numerical Discretization, vol. 1, John Wiley & Sons, Inc,
Hoboken, USA, 2001.
[43] J.D. Lambert, Numerical Methods for Ordinary Differential Systems: The Initial
Value Problem, John Wiley & Sons, Inc, Hoboken, USA, 1991.
[44] J.C. Butcher, Numerical Methods for Ordinary Differential Equations, second
ed., John Wiley & Sons, Inc, Hoboken, USA, 2008.
[45] C. Schneider, A. Dreizler, J. Janicka, Flow Turbul. Combust. 74 (2005) 103127.
[46] M. Gregor, F. Seffrin, F. Fuest, D. Geyer, A. Dreizler, Proc. Combust. Inst. 32
(2009) 17391746.
[47] E. Schneider, A. Maltsev, A. Sadiki, J. Janicka, Combust. Flame 152 (2008) 548
572.
[48] F. Hahn, C. Olbricht, C. Klewer, G. Kuenne, R. Ohnutek, J. Janicka, in: Proc. of the
ISTP19, 2008.
[49] G. Kuenne, C. Klewer, J. Janicka, in: ASME Turbo Expo Conf. Proc., 2009, pp.
369381.
G. Kuenne et al. / Combustion and Flame 158 (2011) 17501767 1767

You might also like