You are on page 1of 26

Combust. Sci. alld Tech., 1997, Vol. 128, pp.

2J-48
Reprints available directly from the publisher
Photocopying permitted by license only
.D 1997 OPA (Overseas Publishers Association)
Amsterdam B. V. Published under license
under the Gordon and Breach Science
Publishers imprint.
Printed in India.
Radiation Modelling in Non-Luminous
Nonpremixed Turbulent Flames
B. MARRACINO* and D. LENTINI!
Dipartimento di Meccanica e Aeronautica, Universita degli Studi di Roma
"La Sapienza", Via Eudossiana 18, 1-00184 Roma, Italy
(Received 15 August 1996; In final form 16 July 1997)
A finite-rate chemistry model for nonpremixed turbulent combustion. based on the stretched
laminar fiamelet (slf) approach, is extended to account for radiative heat transfer in non-
luminous flames, in the optically thin limit. The model is applied to a methane/air flame, and
results in greatly improved predictions of mean temperature, and of its probability density
function. Mean carbon monoxide concentration predictions are instead shown to be somewhat
deteriorated. The model is further tested to reproduce radiative heat flux levels; solutions
obtained by determining the average heat flux divergence by full convolution with the pdf. and
by the simpler but less fundamentally correct 'mean-property' method, are compared, and
show a significant gap. A formulation to deal with non-unity Lewis number flames is proposed;
the model holds potential for further extension.
Keywords: Combustion modelling; flame radiation; turbulent diffusion flames
1. INTRODUCTION
Thermal radiation in combusting systems is known to playa negligible
role only in flames of limited size. Lockwood, as quoted by Jones (1980),
quantifies this threshold in 0.2 m for luminous flames, and about I m for
non-luminous flames. However, radiation is quite often neglected in
models for turbulent combustion, though treatments are available, e.g. see
Faeth et al. (1989). As an example, out of 30 modelling or theoretical
papers on turbulent combustion presented at the latest (26th) Symposium
Present address: Riello RBL S.p.A., Via del Pontiere 17,1-37045 Legnago VR.
'Corresponding author.
23
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
24 B. MARRACINO AND D. LENTINI
(Int.) on Combustion, as few as three envisage account for radiation. Unfor-
tunately, even when radiation is accounted for, the simplifying assumptions
which are inevitably introduced to handle it can severely invalidate a quan-
titative account.
Neglect of radiation implies that the resulting predicted,temperature field is
overestimated, often substantially. Quotation of a few cases, somehow related
to the test cases discussed in the present paper, is indicative of the magnitude
of this overprediction. For instance, Rogg et al. (1986) apply a stretched
laminar flarnelet model, with turbulence treated by the k-8-g model and a
presumed probability density function (pdf), to a methane/air flame at a jet
Reynolds number Re =30000, investigated experimentally by Hassan et al.
(1980). They report a predicted centreline peak temperature about 300 K
above measurements; introducing some tentative account for the effect of
partial premixing does not substantially alter the picture. Methane/air flames
are also considered by Chen et al. (1989), but at jet Reynolds numbers
ranging from Re = 18000 to 21000; they use instead the approach of the
probability density function transport, with a second-moment model for the
fluid dynamical closure. If attention is restricted to the peak temperature in
the different radial profiles reported, predictions overshoot measurements up
to 600 K. Lentini and Puri (1995) consider instead a flame burning chloro-
methane (CH
3CI)
in air at Re = J 1700; incidentally, chloromethane is a heav-
ily sooting fuel. Adopting a stretched laminar flarnelet model, with quite
complex chemistry (38 species and 179 reversible reactions), and k-8-g, they
recover a peak centreline mean temperature as much as 500 K in excess with
respect to measurements (though the latter are obtained from uncompensated
thermocouples, which may however account for no more of 100-150 K of the
discrepancy). Such converging trends, obtained through totally different
modelling tools, and referring to flames fed with different fuels, are a strong
indication that the underlying reason for poor temperature prediction must
be sought in some mechanism not accounted for in all of the models under
consideration, rather than in the deficiencies of a particular model. A likely
candidate is of course thermal radiation, which is not considered by any of
the thermochemical models above.
In addition to the temperature field, neglect of radiation also affects the
predicted species concentration field. A word of caution is therefore in-
dicated when considering the marginal improvement brough about by 'ad-
vanced' turbulence combustion models, in the case they do no account for
radiation. In particular, heat losses are expected to remarkably affect NO
x
emissions in furnaces and combustors, due to the high temperature sensitiv-
ity of the thermal (Zel'dovich) pathway (Lentini and Puri, 1996).
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NON-LUMINOUS TURBULENT FLAMES 25
At any rate, radiation modelling, when adopted, is not at all free from
uncertainties, as modelling and computational assumptions can largely af-
fect the predicted results. E.g., the 'mean-property' method, commonly used
to describe the heat loss term in the mean .enthalpy equation, can lead to
grossly approximate estimates of the heat flux, and of the associated flow-
field.
The difficulty of introducing the effect of thermal radiation in turbulent
combustion models stems from its quite complex nature. In flames, thermal
radiation comes both from gas-phase and soot (when present). In most
hydrocarbons, soot radiation largely overwhelms radiation from gas bands;
methane is an exception, due to its relatively mildly sooting nature.
Thermochemical closure models based on the conserved scalar approach
introduce the assumption of adiabatic flow in order to reduce the conserva-
tion equations for energy and elemental mass fractions to the same form,
with the same boundary conditions. This enables to express the value of any
state quantity (with the additional assumptions of low-speed flow, unity
Lewis number for all species, and equilibrium chemistry) in terms of a single
quantity, the conserved scalar (also termed mixture fraction) Z. The equilib-
rium assumption is lifted in 'stretched laminar flamelet' (hereafter denoted
as 'slf"] models (Liew et al., 1984; Peters, 1984; 1986) by introducing a sec-
ond fundamental state quantity, the scalar dissipation rate Xc (conditioned
at the flame front, e.g., evaluated at stoichiometric or maximum tempera-
ture conditions), which accounts for finite-rate chemistry effects. A key ad-
vantage of the approach is that the main features of the joint probability
density function P(Z, xcl, needed to evaluate mean state quantities in turbulent
flows, are clearly identified. In fact, Z and Xc turns out to be' substantially
(though not completely) statistically independent, and the scalar dissipation
rate pdf is unambiguously log-normal (Peters, 1984; Bilger, 1989). When
attention is extended to flows involving heat transfer by radiation, a
source/sink term appears in the energy equation, which must therefore be
considered explicitly. When convection to a solid boundary is considered,
the boundary conditions for the mean enthalpy equation differ from those
applicable to the mean conserved scalar defined in terms of elemental mass
fractions, with the consequence that the enthalpy equation must still be
treated separately. In models based instead on pdf transport, consideration
of heat transfer requires introducing the enthalpy as an additional indepen-
dent variable, an approach tested by Chen and Kollmann (1992).
Bray and Peters (1994) suggest a treatment for including the effect of heat
transfer in turbulent com busting flows, as an extension of the stretched
laminar flamelet formalism; the approach is therefore suited for implemen-
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
26 B. MARRACINO AND D. LENTINI
tation on application-oriented numerical codes, such as those using k-f.-g
modelling. It is based on introducing the concept of enthalpy defect, here
denoted as " defined as the difference between the actual enthalpy and the
enthalpy of an adiabatic, equidiffusive mixture characterized by the same
value of the conserved scalar. State variables can then be expressed as a
function of three fundamental variables, i.e., Z, X" ,.
In this paper the approach above is applied to a non premixed turbulent
flame in which heat transfer is dominated by gas-phase radiation. Attention
is therefore focused on methane flames, though the same treatment might be
applied to flames fed with non-sooting fuels such as hydrogen or carbon
monoxide. The model is carefully evaluated against measurements of the
temperature statistics, involving its mean, root-mean-square and pdf. The
effect of radiation on species concentration is also checked. Then, radiative
heat flux predictions are evaluated (using both full convolution with pdf,
and the simpler mean-property method) against measurement in meth-
ane/air flames at different Reynolds number. Finally, a further extension of
the approach by Bray and Peters is proposed in order to make it possible to
account for differential diffusion effects.
2. TREATMENT OF HEAT LOSS EFFECTS VIA
THE SLF APPROACH
In the context of slf modelling of adiabatic combustion, each state quantity
c/> in a laminar flame subjected to a given stretch (identified by the value of
the scalar dissipation rate Xc conditioned at the flame front) can be ex-
pressed as a function of the conserved scalar, or mixture fraction, Z; i.e.,
(I)
Average values of state quantities in turbulent flames can be recovered by
introducing a joint probability density function (pdf) of Z and Xc' denoted
as P (Z, Xc! (Note: the pdf's being considered are nor density-weighted).
Then, conventional averaging can be performed as
= f'" fl c/> (Z; Xc) P(Z, xJdZ dXc
o 0 .
(2)
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NON-LUMiNOUS TURBULENT FLAMES
and Favre (density-weighted) averaging as
- 11
0011
</J =-=- p(Z;xJ </J(Z, x,)P(Z,xJ dZ dx,
p 0 0
27
(3)
Owing to the observed statistical independence between Z and X" the joint
pdf can be split into a product of ordinary pdf's:
P(Z, X,) = P (Z) P(xJ (4)
This makes it possible to assign the pdf on presumed bases. A p-function
(which has been shown to work quite well, Lentini, 1990), or an equivalent,
can be assumed for P(Z), and a log-normal distribution for P(X,) (Kol-
mogorov, 1962; Liew et al., 1984; Peters, 1984).
Extension of the slf formalism to nonadiabatic combustion, as suggested
by Bray and Peters (1994), requires introducing the enthalpy defect (
(5)
which, accordingly, represents the difference between the actual enthalpy
and the enthalpy for an adiabatic flame (with h; and hf denoting the en-
thalpy of the oxidizer and fuel streams, respectively). Infact, in an adiabatic,
equidiffusive (i.e., with equal species diffusivities, and unity Lewis number)
flame, the conserved scalar can be defined indifferently as
Z= Zi-Zi,o
Zi,f-Zi.o
(6)
where Z, denotes the elemental mass fraction of any of the elements in-
volved. When non-adiabatic combustion is considered, only the definition
in terms of the Z:s can be applied. Once ( is introduced, state quantities in
laminar flames can therefore be expressed as
(7)
Then, average values of state quantities can, in principle, be recovered by
introducing a joint pdf of Z, X, and (:
(jJ=fOO roo r
t
</J(Z;x,;OP(Z,x"OdZdX,d(
-00 Jo J,
(8)
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
28 B. MARRACINO AND D LENTINI
where the integration range of ( is extended, for the sake of generality, to
include the effect of both heat loss and gain. A problem remains in the
identification of the joint pdf shape. At any rate, as any reasonable account
of heat losses represents an improvement with respect to the assumption of
adiabatic flow, Bray and Peters (1994) suggest adopting but the simplest
modelling assumption with respect to C i.e., they neglect the effect of its
fluctuations, which amounts to assuming
P(Z, x" () = P(Z) P(Xc) 0( - () (9)
Obviously the approach requires an additional equation for the mean en-
thalpy ii (the tilde denotes a Favre average, i.e., density-weighted), which
allows recovering' after the linear relationship (5):
(10)
The thermochemical data required as an input to the model are therefore
the libraries (7), i.e., a set of laminar flamelet profiles, giving the properties
of interest as a function of the conserved scalar (as mentioned, we neglect
differential diffusion at this stage for the sake of simplicity, but will take up
the issue further in Sec. 5). The profiles must be organized in 'shelves', where
a shelf contains entries referring to values of Xc ranging from equilibrium to
extinction, plus the inert (or pure-mixing) state, and each shelf refers to a
different value of (. Notice that the number of representative Xc's adopted in
the different shelves need not be the same for each value of Cand will likely
not be so, since the extinction (or quenching) limit X
q
becomes smaller as the
enthalpy defect is decreased (i.e., for increasing heat losses).
In the test cases considered in Sees. 3 and 4, heat loss is due to radiation
only. In these simple arrangements, the assumption of optically thin limit
can reasonably be adopted (e.g., see Hottel and Sarofim, 1967; Kuznetsov
and Sabel'nikov, 1990), though it is likely to overestimate radiation some-
what (as the gas is assumed to only give up heat, without re-absorbing
radiation). The heat loss term in the enthalpy equation is therefore modelled
as
div q, = - 4up [T(Z, x" 0]4 Lkj [T(Z, Xc' 0] Xi (Z, Xc' 0 (I I)
where q, is the radiative heat flux vector, a the Stefan-Boltzmann constant,
p the pressure, and the k/s are the absorption coefficients in terms of partial
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NON-LUMINOUS TURBULENT FLAMES 29
pressure of the relevant species (e.g., see Hottel and Sarofim, 1967), whose
molar fraction is Xi' Clearly, Eq. (II) assumes the form (7) and can be
averaged accordingly as in (8) for inclusion into the mean enthalpy equa-
tion. In the test cases considered, the immaterial reverse effect of radiation
from the laboratory environment to the flame is neglected. The k,'s are
reduced to a function of the temperature only due to the optically thin
medium assumption. The summation in (II) is extended to the radiating
species H
2
0 , CO
2
, CO and CH
4
. Soot radiation can be neglected for meth-
ane, owing to its very limited weight. The following approximating expres-
sions are derived for the absorption coefficients of the relevant species
(expressed in atm-I m- I, with T in K):
loglo k
H
0 =0.93567 - 5.5258.10-
4
T
,
(12)
loglok
co
=1.0738+4.9997' 10-
4
T- 5.769' 10-
7
T
'
+ 1.0674 10- 10 T
3
(14)
,
10g,ok
cH
= -0.28331 +3.58631O-
3T-3.757510- 6T2
.
+ 1.6037.10-
9
T
3
-2.5024'10-
13
T
4
(15)
Incidentally, other models for radiating flames based on the slf approach
have been developed, e.g., see Young and Moss (1995), which further ac-
count for soot radiation, but not yet for finite-rate chemistry, which is
important for concentration predictions in hydrocarbon-fed flames. The
approach followed here accounts for finite-rate chemistry, and holds poten-
tial for similar extension to sooting flames.
3. EFFECT OF RADIATION ON A TURBULENT
METHANE/AIR FLAME
The model outlined in the previous Section is applied here to a non-
premixed turbulent methane/air flame, investigated experimentally at the
Imperical College, University of London (Hassan et al., 1980; El-Banhawy
et al., 1983): several authors has also adopted it as a test case in numerical
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
30 B. MARRACINO AND D. LENTINI
simulations (Liew et al., 1984; Rogg et al., 1986; Lentini, 1994). We focus
here on a flame at Re = 15000, stabilized by a small hydrogen coflow, as
this is the case which is reported in greatest detail; this correspond to a jet
bulk velocity U = 39.9 m/s. The experimental data available include, inter
alia, centreline and radial profiles of mean temperature, the centreline pro-
file of root-mean-square temperature, and the probability density function
of the temperature at a discrete number of centreline locations in between
x] D = 36.2 and x] D = 204, the fuel nozzle diameter being D = 2R= 7.74
mm. The observed flame length is about 1 m, thus underlining the need to
account for radiation; however, the assumption of an optically thin medium
will likely result in overestimating the impact of radiation losses for a flame
of this size (Kuznetsov and Sabel'nikov, 1990).
The computational code used here is an extension of a code for para-
bolized axisymmetric flows, based on the k-c-g model, previously tested for
adiabatic combustion (Lentini, 1994). For such a first-moment closure
model, the modelled conservation equations are supplemented by the mean
enthalpy equation in the form
8 __ - 18 __ - 18 [(11 11,)8iiJ ._
-(puh)+--(rpvh)=-- r -+- - +dlvq
ax r ar r ar (1 (1, ar '
(16)
where II and II" (1 and (1, denote the molecular and turbulent viscosity and
Prandtl number, respectively.
The flamelet library adopted for the methane-air flame under consider-
ation is computed by means of a simplified procedure operating directly in
conserved scalar space (Lentini, 1993). It solves for the species equations
only, cast in Z-space thanks to the assumption of thin flamelet; local values
of X and Z across the flamelet are related through a similarity solution,
albeit for constant-density flow (Peters, 1984; 1986). The code inter alia
neglects differential diffusion, and overestimates the extinction limit, but has
been checked to produce results for turbulent flames that are in close agree-
ment to those obtained by adopting a fully detailed library, due to the
weight of other accompanying modelling uncertainties. The reaction scheme
considered is the 'full' one discussed by Peters (1985), involving 15 species
and 20 reaction steps. Setting up a flamelet library for a radiating flame is
made easier by adopting the assumption of equal diffusivities. If the same
value of the enthalpy defect' is enforced at both the fuel and oxidizer sides,
then this assumption warrants the enthalpy defect to be uniform across the
flamelet thickness. The minimum value of the mean enthalpy defect in the
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NON-LUMINOUS TURBULENT FLAMES 31
entire flow field turns out to be (= - 664 kJ /kg, Therefore, a flamelet
library is used with as many as 15 'shelves', with enthalpy defects (= 0,
- 50, - 100, .,. , - 650, - 700 kJ /kg. Though this may at first sight appear
as a quite fine representation, it will soon be shown that some small discon-
tinuities appear in the predicted profiles, due to the neglected smearing
effect of enthalpy fluctuations (as the enthalpy defect pdf is modelled as a
"-function). Each shelf comprises entries ranging from near-equilibrium to
near-extinction, plus the inert state. Figure 1 shows a small sample from the
library, reporting a few shelves with six to four burning entries each. It
focuses on temperature; clearly, the library includes similar results for den-
sity and species concentration. As expected, the temperature is reduced by
increasing both X, (i.e., the weight of finite-rate chemistry effects) and 1(1 (i.e.,
the weight of heat losses). Quite similar profiles are also derived for the
radiative heat loss term (II), and are shown in Figure 2.
The importance of accurately taking into account the effect of fluctu-
ations when evaluating the average of term (II) is stressed by Figure 3,
which refers to a near-equilibrium flame with no enthalpy defect; the shape
of the conserved scalar pdf is taken as specified by Lentini (1994). It is seen
that, when the level of scalar fluctations 2"2 is progressively increased from
zero towards the limit Z(1 - Z), the curve giving the mean divergence of the
radiative heat flux vector is increasingly smeared; in particular, the (abso-
lute) peak value is substantially reduced. Further smearing effects are ex-
pected due to the action of the scalar dissipation rate fluctuations.
With the value of the mean enthalpy h given by Eq. (16), and the corre-
sponding enthalpy defect recovered after Eq. (10), application of Eq. (9)
requires evaluating scalar properties for (=(. Given the finite number of
shelves available in the library, this is performed by means of an interpola-
tion. If m denotes the library shelf index, such that the value of ( at the
location under scrutiny turns out to be in the range (m';;; ( < (m-" then
interpolation of the generic state quantity is carried out as
}. ( - (m:i: (m- I - ( }.
'I' = r _ r 'l'm-' + r _ r 'I'm
~ m - 1 ':.m ':.m- I ~ m
where m stands for
and similarly for m-"
(17)
(18)
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
32 B. MARRACINO AND D. LENTINI
T (K)
2400
2100

1800
1500
1200
900
600
300
0.0 0.2
Z
T (K)
2400
2100

1800
1500
1200
900
600
300
0.0 0.2 0.4 0.6 1.0
Z
T (K)
2400
2100

1800
1500
1200
900
600
300
0.0 0.2 0.4 0.6 0.8 1.0
Z
FIG URE 1 Sample of the flamelet library: temperature us.conserved scalar, for different values
of the enthalpy defect (top C=0, middle C=- 300 k.l/kg, bottom C= - 600 kJ /kg); - 7.,'" 0.86
S-I, __ S-I, __- 12.5S-I, '_ X('=::::44 S-I,_"_"_ X('=X
9
-
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NON-LUMINOUS TURBULENT FLAMES 33
ME -1E6
~
d'
:6 -2E6
\
\
\
, , _ . ~ " ,
\
" \,
0.4
z
s=O
0.6 0.8 1.0
S=-300kJ/kg
z
ME -lE6
-
~
d'
. ~ -2E6
1J
S=-600kJ/kg
z
FIGURE 2 Sample of the flamelet library: divergence of the radiative heat flux vector vs.
conserved scalar, for different values of the enthalpy defect (top' =0, middle' = - 300 k.l/kg,
bottorn I = -600 k.l/kg); -7.,,,,0.86 s-', - - 7., '" 3 s-', - - -7.,'" 12.5 s-', -'-'-
Xc ~ 4 4 5-
1,
- .. -.,- Xc = X
q

D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
34 B. MARRACINO AND D. LENTINI
OEO
I
r. \
.1 \.
\', \.,
~ I -c,
l')
-1 E6
I
E
\ ,
--
\ '
~
\ '-
~
,
0-
. ~ -2E6
"0
---
~ = 0, z, = 0.88 5.
1
-
Z
1.0
FIGURE 3 Effect of conserved scalar fluctuations on the divergence of the mean radiative
. heat flux vector VS. mean conserved scalar, for a near-equilibrium flame with no enthalpy
defect; - T'=O, - - T'=o.t2(1-2), - - - T'=O.4 2(1-2), -'-'-
7' = 0.72(1 - 2).
Comparisons of predictions, worked out under both assumptions of
adiabatic and radiating flow, and experiments are now considered; inciden-
tally, predictions have been checked to be substantially grid-independent.
Before going into the details of the effect of radiation on the scalar field,
evidence is shown in Figure 4 that the computational model substantially
reproduces the mean velocity field, as measured by EI-Banhawy et al. (1983),
with a limited effect of radiation. As far as the temperature field is
concerned, Figure 5 compares centreline predicted and measured mean tem-
peratures; it also reports predictions worked out for an adiabatic, near-
equilibrium flame, so as to assess the weight of finite-rate chemistry
compared to radiative heat transfer. It is clearly seen that in this flame the
effects due to heat loss overwhelm those brought about by finite-rate
chemistry; in the following, the less accurate assumption of near-equilibrium
chemistry will not be considered any longer. As far as radiative effects are
concerned, Figure 5 shows that, whereas adiabatic predictions grossly
overestimate the mean temperature, predictions accounting for radiation
according to the proposed model give a reasonably close match to the
experimental data. In particular, the peak mean temperature, originally
overpredicted by as much as 270 K, is underestimated by 130 K only when
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NONLUMINOUS TURBULENT FLAMES 35
1.4
1.2
~
I
I
1.0
,
\ p
I
0.8
I
=>
,
Ie
--
\
1:::J
0.6f-
\ e
\ c
\
"
0.4
0.2
e
e
30
,
60 90 120 150 180 21 0
x/D
FIGURE 4 Centerline profile of dimensionless mean axial velocity; D measurements (EI
Banhawy et al., 1983), -=-- - - predictions without radiation, - predictions with radiation.
II-
1500
1200
900
600
FIGURE 5 Centerline mean temperature: D measurements (Hassan et al., 1980), -'-'-
predictions without radiation and with near-equilibrium chemistry, - --predictions without
radiation, - predictions with radiation.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
36 B. MARRACINO AND D. LENTINI
thermal radiation is included. The fact that the proposed radiation model
underpredicts the centreline peak temperature can be ascribed (within the
framework of turbulence and combustion modelling uncertainties) to the
overestimated heat losses involved in the optically thin medium assumption.
Radial profiles of mean temperature at four axial stations are reported in
Figure 6. At the first measuring station, x] D = 36.2, the entity of heat losses
is still relatively small; predictions without radiation overestimate the peak
temperature by about 130 K, whereas it is correctly predicted (within 25 K)
by the proposed radiation model. The agreement is closer away from the
peak, too. At x] D = 75, the model without radiation overshoots the peak
mean temperature by about 225 K; with radiation, it is underestimated by
35 K. It is noted that at this station the centreline too is now remarkably
affected by heat losses. The radiation model greatly improves matching with
measurements up to the radial location r I R "" 15, but appearently leads to
somewhat deteriorated predictions at the outer edge of the flame. However,
it is observed that in this outer region the adiabatic model too under-
predicts the flame temperature, therefore the effect must probably not be
ascribed to deficiencies of the combustion model, but rather to an inaccur-
ate prediction of scalar mixing by the model for the fluid dynamical closure.
The trend is replicated at x] D = 113.7, with the peak discrepancy between
the two solutions (305. K in excess without radiation, 105 in defect with)
now at the centreline, indicating that the mean flame front is close the axis.
Finally, at the most downstream station, x] D = 204, the discrepancy
between predicted temperatures is close to 300 K, with the prediction ac-
counting for radiation giving better results all over the measured range. On
the whole, a uniform improvement of predictions can be clearly appreciated.
Figure 6, though giving a fairly satisfactory agreement with respect to
measurements, features apparent discontinuities, though of quite limited
size. It can be checked that discontinuities occur at locations where the
mean enthalpy crosses the value corresponding to a particular shelf, say the
m-th shelf. Then, whereas on one side of the location under scrutiny
the average value of state quantities is obtained by interpolating between
shelves m- I and m, on the other side interpolation is carried out between
shelves m and m+ I, and the limited number of shelves built in the library
results in the observed, slightly discontinuous, behaviour. Notice however
that the size of the discontinuities is much smaller than the expected accu-
racy of the model, therefore they do not alter the reported picture of re-
markably improved agreement.
Figure 7 reports the centreline profile of root-mean-square temperature,
indicating a fair agreement overall. Clearly, second-moments are more
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NON-LUMINOUS TURBULENT FLAMES 37
xlD = 36.2
30 25 20
2100
1800
1500
~
~
1200
~
II-
900
600
300
0
1800
~
~
1200
~
II-
900 -
600
300
0
_------------, xlD = 75
,
,
,
,
,
0',
\
\
\
0\
\
\
\
\
0\
\
\
\
~
\
\
\0
\
\\ 0
\
\
30
FIGURE 6 Radial profiles of mean temperature at four axial stations: 0 measurements
(Hassan et at.. 1980), - - - predictions without radiation, - predictions with radiation.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
38 B. MARRACINO AND D. LENTINI
2100
--,
"
1800 ---
-
xlD =113.7
-
,
D D D
D
-,
,
D
,
,
D',
,
19,
~
,
,
~
1200 '"
~
,
-,
II-
''0
,
,
900
'.p
,
,
'..E'
-
-
600
300
0 15 20 25 30
r / R
2100
1800 xlD =204
1500
-
----
~
-
1200
I-
900 D
600
300
0 30
FIGURE6 (Continued).
difficult to reproduce than first-moments (average quantities), especially
when using the popular first-moment closure turbulence models. The over-
all trend is basically reproduced, though the predicted peak occuring shortly
downstream of the location where the turbulent flow reaches the axis is
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NON-LUMINOUS TURBULENT FLAMES
[J
39
30 60
[J
[J
[J [J
[J
[J
[J
90 120 150 180 210
xI 0
FIGURE 7 Centreline rms temperature: 0 measurements (Hassan er al., 1980), - - -
predictionswithout radiation, - predictionswith radiation.
instead barely perceivable in the experiments. This may be due both to
limitations of the turbulence model, and to smearing effects present in
experiments (e.g., finite thickness of the fuel nozzle) which cannot be repro-
duced in the present parabolized computation. Away from this zone, the
radiating model seems to perform appreciably better than the adiabatic one.
Figure 8 compares instead the temperature pdf measured at four centre-
line locations to that predicted without, or with, the effect of radiation. An
improvement is manifest at all locations. At x] D = 36.2 the improvement is
marginal, due to the still relatively small predicted impact of radiation, but
anyway in the right direction. Consistently with the trend for mean tem-
peratures, the predicted pdf peak is located at temperatures higher than
measured, under both the assumptions of adiabatic and radiating flow. A
marked improvement is instead observed at x] D =75; while the adiabatic
pdf prediction peaks at about 1735 K, the radiating model peaks around
1535 K, then approaching the measured peak at 1430 K. Notice that at
both locations x] D = 36.2 and x] D = 75 the width of the distribution is also
quite well captured. At locations xlD = 113.7 and x l D = 204 the most
probable temperature is fairly correctly captured by the radiating model;
the width of the distribution is instead less satisfactory, as it appears to be
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
40 B. MARRACINO AND D. LENTINI
0.006
x/D=36.2
0.005
0.004
~
~
0.003 q;P
~
I-
D
~
C-
0.002
D
D
0.001 D
D
O.oogoo
0.006
xlD =75
0.005
0.004
~
~
DO
0.003
~
I-
~
C-
0.002
0.001
O.oogoo
FIGURE 8 Probability density function or the temperature at four centreline locations:
o measurements (Hassan et ol., 1980),---predictions without radiation,- predictions with
radiation.
underpredicted at the high temperature side at x ]D = 113.7, and over-
predicted at x] D = 204. The improvement with respect to the adiabatic
calculation is anyway remarkable at all locations, and is particularly
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NON-LUMINOUS TURBULENT FLAMES 41
0.006
I
I
XI D =113.7
I
I
0.005 I
I
I
I
I
, I
I I
~ " I
~
, /IV
0.003
II
~ II
I-
Ir
~ I
n, I
0.002
I
DD
01
II
oDD
dl
oD
0.001
/ Do
I D
/
D
O.oogoo
- ~ /
600 1200 1500 1800 2100 2400
T (K)
0.006
xl D =204
0.005
0.004
~
~
~
0.003
t:-
o,
0.002
0.001
O.oog
FIGURE8 (Continued),
significant as the pdf is the most difficult temperature-related characteristic
to reproduce.
It is as well to remark that, while the radiation model gives remarkably
improved predictions of the temperature field, it fails to explain disagree-
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
42 B. MARRACINO AND D. LENTINI
ment between predictions and measurements of species concentration, thus
calling for further investigation. As an example, Figure 9 reports radial
profiles of mean carbon monoxide molar fraction at x] D = 75. It is seen
that the predicted peak concentration level, when making allowance for
radiation, compare with measurements even worse than adiabatic predic-
tions. This suggests that, in general, caution is indicated when considering
concentration predictions worked out by adiabatic models; further, other
effects must be investigated to explain the reported discrepancies.
4. PREDICTIONS OF RADIATIVE HEAT FLUX
In order to directly test the radiation model here proposed, in this Section
predictions are worked out for the radiative heat flux from turbulent meth-
ane/air flames at three different jet Reynolds numbers, corresponding to the
experiments by Jeng and Faeth (1984). The test rig in this case involves a
vertical fuel nozzle of diameter D = 5 mm; the radiation sensor can slide on
a vertical line at a distance of 0.575 m from the axis, and measurements are
X
eo
0.06
0.05 -
o
c
0.04
c
0.03
0.02
0.Q1
0.000
10
r / R
x/ D =75
30
FIGURE 9 Radial profiles of mean carbon monoxide molar fraction at axial station x tD = 75:
D measurements (Hassan el al., 1980),---predictionswithout radiation,- predictions with
radiation.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NON-LUMINOUS TURBULENT FLAMES 43
taken at several stations along this line. The radiative heat flux is computed
by the 'discrete transfer' approach by Lockwood and Shah (1981), under
two different approximations to evaluate the average of expression (11) in
the radiative transfer equation. The most accurate procedure is a full con-
volution of Eq. (II) with the pdf (9), i.e.,
(19)
Incidentally, the same treatment of the term is applied in Eq. (16). A second,
widely used alternative is the so-called 'mean-property method', which esti-
mates the mean divergence as
divq,= -4apf4fOO roo r'Iki[T(Z,x"mX,(Z,x"OdZdt/J,d( (20)
-CQJOJOl
where r is the mean temperature, recovered similarly to Eq. (8). Clearly,
this approach is fundamentally less correct. The absorption coefficient itself
is often evaluated on the basis of mean temperature and concentrations,
with an accompanying further degradation of the approximation.
Results are reported in Figure 10. It is seen that both approaches correctly
reproduce the overall trends, and capture the location of the peaks, though
this appears to be more precisely identified when adopting the full convolu-
tion (19). The peak heat flux levels are instead better reproduced by the
mean-property method. However, this closer agreement with experimental
data cannot be adopted as an argument in favour of the mean-property
method, due to its inherently approximate nature. On the contrary, it de-
monstrates that the latter approach can give deceptively close agreement
because approximating the mean divergence in terms of the mean tempera-
ture compensates for other model deficiencies. In addition, the assumption
of an optically thin medium is bound to overestimate radiation, and predic-
tions adopting full convolution do reproduce this expected trend. The
mean-property method, due to its uncomplete weighting of the divergence
term, introduces a compensating effect which lowers the computed radiative
flux. Therefore, the present results indicate that predictions worked out by
the mean-property method should in general be taken with due caution.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
44
1250
1000
-
'" 750
E
......
3:
---- 500
....,
B. MARRACINO AND D. LENTINI
Re = 2920
Re =5850
400
FIGURE 10 Radiative heat flux at a distance of 0.575 m from the axis at three jet Reynolds
numbers. 0 measurements (Jeng and Faeth, 1984), - - - predictions with mean-property
method, - predictions by full convolution with pdf.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
--.
'"
'E
500
250
c
NON-LUMINOUS TURBULENT FLAMES
Re =11700
FIGURE 10 (Continued),
400
45
z
FIGURE II Enthalpy VS, conserved scalar in laminar flamelets, for two entries with the same
enthalpy defect enforced at the extrema: a) equidiffusive mixture, b) hypothetical picture with
differential diffusion accounted for - adiabatic flarnelet, - - flamelet with enthalpy defect.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
B. MARRACINO AND D. LENTINI 46
OEO
-1 E6
-2E6
~
Ol
~
.....
-3E6 .,
~
s:
-4E6
-5E6 S1 *' S2 *' S3
-6E6 S3
0.0 1.0
Z
FIGURE II (Continued).
5. EXTENSION TO NON-UNITY LEWIS NUMBERS
The slf approach for adiabatic flows allows making allowance for the different
diffusivities of species and enthalpy; this requires appropriately re-defining
2, usually as a weighted sum of scaled elemental mass fractions (Dixon-
Lewis and Missaghi, 1988; Bilger, 1988), as a replacement of (6). When heat
losses are considered in non-unity Lewis number flames, formula (5) no
longer applies, as assigning the same enthalpy defect at the extrema does
not warrant a uniform ( across the given laminar flamelet; a purely hypo-
thetical pictorial view of the situation is reported in Figure II. In this case
the approach proposed by Bray and Peters (1994) can still be used, but the
interpolation is slightly more complex than indicated in Eq. (17). Now each
shelf In can still be computed by enforcing a given enthalpy defect (mat both
sides, but the enthalpy defect in a flarnelet belonging to shelf m, rather than
being uniform, is itself a function of both 2 and 'I.,:
[( (2, X,)]m = [h (2, 'I.,)]m - [h (2, '1.,)]0 (21)
where index 0 refers to the shelf with associated zero enthalpy defect. This
replaces definition (5); interpolation can still be performed similarly to Eq. (17).
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
NONLUMINOUS TURBULENT FLAMES
6. CONCLUSIONS
47
An extension of the stretched laminar flamelet approach is used to account
for radiation effects in non premixed turbulent flames. In the particular
methane/air flame considered, it results in greatly improved predictions of
mean temperature, and of the temperature pdf. Mean carbon monoxide
concentration predictions are instead adversely affected by making allow-
ance for radiation. This suggests that concentration predictions by adiabatic
models must be considered with appropriate caution. Predictions of
radiative heat flux reproduce the experimental trends; an appropriate cau-
tion is again suggested when evaluating results worked out through the
'mean-property' method, which is shown to give closer agreement with
experiments than a method based on full convolution with the pdf, but is
fundamentally less correct. An extension of the proposed approach to
flames with non-unity Lewis numbers is formulated. Further, it holds
potential for extension to sooting flames, which requires a model for soot
emission and radiation; both are currently available for different fuels (e.g.,
Moss et al., 1988; Syed et al., 1990; Gore and Jang, 1992).
Acknowledgements
This research has been supported by grant MURST.
References
Bilger, R. W. (1988) The structure of turbulent nonpremixed flames. In 22nd Symp. (Int.) on
Combust., The Combustion Institute, Pittsburgh, pp. 475-488.
Bilger, R. W. (1989) Turbulent diffusion flames. Ann. Rev. Fluid Mech., 21, 101.
Bray, K. N. C and Peters, N. (1994) In Turbulent Reacting Flows (Libby, P. A. and Williams,
F. A. Eds.), Academic Press, London, pp. 78-84.
Chen, J.- Y. and Kollmann, W. (1992) PDF modeling and analysis of thermal NO formation in
turbulent non premixed hydrogen-air jet flames. Combust. Flame, 88, 397.
Chen. J-Y.. Kollmann, W. and Dibble, R. W. (1989) Pdf modeling of turbulent nonpremixed
methane jet flames. Combust. Sci. Tech., 64, 315.
Dixon-Lewis, G. and Missaghi, M. (1988) Structure and extinction limits of counterflow diffu-
sion flames of hydrogen-nitrogen mixtures in air. In nnd Symp. (Int.) on Combust .. The
Combustion Institute, Pittsburgh, pp. 1461-1470.
EI-Banhawy, Y., Hassan, M. A., Lockwood, F. C and Moneib, H. A. (1983) Velocity and un-
burned hydrocarbon measurements in a vertical turbulent free jet diffusion flame. Com-
bust. Flame. 53, 145.
Faeth, G. M.. Gore, J. P., Cuech, S. G. and Jeng, S.-M. (1989) Radiation from turbulent diffu-
sion flames. In Ann. Rev. Num. Fluid Mech. Heat Transf (Tien, C L. and Chawla, 1. C,
Eds.], Hemisphere, New York, pp. 1-38.
Gore, J P. and Jang, J H. (1992) Transient radiation properties of a subgrid scale eddy. J.
Heat Transfer, 114,234.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0
48 B. MARRACINO AND D. LENTINI
Hassan, M. M. A., Lockwood, F. C. and Moneib, H. A. (1980) Fluctuating temperature and
mean concentration measurements in a vertical free jet diffusion flame. Rio. Combustibili,
34. 357.
Hottel, H. C. and Sarofim, A. F. (1967) Radiative Transfer, McGraw-Hili, New York, pp.
199-242.
Jeng, S.-M. and Faeth, G. M. (1984) Radiative heat fluxes near turbulent buoyant methane
diffusion flames. J. Heat Tmns.; 106,886.
Jones, W. P. (1980) Models for turbulent flows with variable density and combustion. In
Prediction methods for tIIrbulellt flows (W. Kollmann, Ed.), Hemisphere, Washington,
pp. 380- 421.
Kolmogorov, A. N. (1962) A refinement of previous hypotheses concerning the local structure of
. turbulence in a viscous incompressible fluid at high Reynolds number. J. Fluid Mech.; 13, 82.
Kuznetsov, V. R. and Sabel'nikov, V. A. (1990) Turbulence and Combustion, Hemisphere, New
York, pp. 205-210.
Lentini, D. (1990) Numerical prediction of nonpremixed turbulent flames. AIAA paper
90-0730.
Lentini, D. (1993) Validation of a formulation in conserved scalar space for stretched laminar
flamelet profiles. AIAA paper 93-2200.
Lentini, D. (1994) Assessment of the stretched laminar flamelet approach for non premixed
turbulent combustion. Combust. Sci. Tecn.. 100,95.
Lentini, D. and Puri, I. K. (1995) Stretched laminar flarnelet modeling of turbulent chloro-
methane-air nonpremixed jet flames. Combust. Flame, 103,328.
Lentini, D. and Puri, I. K. (1996) Radiation and NO, pathways in non premixed turbulent
flames. (Submitted).
Liew, S. K., Bray, K. N. C. and Moss, J. B. (1984) A stretched laminar flarnelet model of turbu-
lent nonpremixcd combustion. Combust. Flame, 56, 199.
Lockwood, F. C. and Shah, N. G. (1981) A new radiation solution method for incorporation in
general combustion prediction procedures. In 18th Symp. (lnt.; Oil Combust., The Combus-
tion Institute, Pittsburgh, pp. 1405-1414.
Moss, J. B., Stewart, C. D. and Syed, K. Y. (1988) Flowfield modelling of soot formation at
elevated pressure. In 22"d Symp. (Jnr.} on Combust, The Combustion Institute, Pittsburgh,
pp. 413-423.
Peters, N. (1984) Laminar diffusion flamelet models in nonpremixed turbulent combustion.
Prog. Enerqy Combust. Sci., 10,319.
Peters, N. (1985) Numerical and asymptotic analysis of systematically reduced reaction
schemes for hydrocarbon flames. In Lecture Notes in Physics, 241, Springer-Verlag, Berlin,
pp.90-109.
Peters, N. (1986) Laminar flarnclet concepts in turbulent combustion. In 21st Symp. (lilt.) 011
Combust., The Combustion Institute, Pittsburgh, pp. 1231-1250.
Rogg, B., Behrendt. F. and Warnatz, J. (1986) Turbulent non-premixed combustion in partially
premixed diffusion flarnelets with detailed chemistry. In 21st Symp. (lilt.) 011 Combust., The
Combustion Institute, Pittsburgh, pp. 1535-1541.
Syed, K. Y., Stewart, C. D. and Moss, J. B. (1990) Modelling soot formation and thermal radia-
tion in buoyant turbulent diffusion flames. In 23rd Symp. (lilt.) 011 Combust., The Combus-
tion Institute, Pittsburgh. pp. 1533-1541.
Young, K. J. and Moss, J. B. (1995) Modelling sooting turbulent jet flames using an extended
flarnelet technique. Combust. Sci. Tech., 105. 33.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
U
n
i
v
e
r
s
i
t


d
e
g
l
i

S
t
u
d
i

d
i

R
o
m
a

L
a

S
a
p
i
e
n
z
a
]

A
t
:

1
0
:
3
9

4

J
u
n
e

2
0
1
0

You might also like