You are on page 1of 14

Combustion and Flame 148 (2007) 6275

www.elsevier.com/locate/combustame
Dissipation length scales in turbulent nonpremixed
jet ames
Guanghua Wang
a
, Adonios N. Karpetis
b
, Robert S. Barlow
a,
a
Combustion Research Facility, Sandia National Laboratories, Livermore, CA 94550, USA
b
Department of Aerospace Engineering, Texas A&M University, College Station, TX 77843-3141, USA
Received 24 March 2006; received in revised form 30 August 2006; accepted 12 September 2006
Available online 28 November 2006
Abstract
Line imaging of Raman/Rayleigh/CO-LIF is used to investigate the energy and dissipation spectra of turbulent
uctuations in temperature and mixture fraction in several ames, including CH
4
/H
2
/N
2
jet ames at Reynolds
numbers of 15,200 and 22,800 (DLR-A and DLR-B) and piloted CH
4
/air jet ames at Reynolds numbers of
13,400, 22,400, and 33,600 (Sandia ames C, D, and E). The high signal-to-noise ratio of the 1D Rayleigh scat-
tering images enables determination of the turbulent cutoff wavenumber from 1D dissipation spectra. The local
length scale inferred from this cutoff is analogous to the Batchelor scale in nonreacting ows. The measured ther-
mal dissipation spectra in the turbulent ames are shown to be similar to the model spectrum of Pope for turbulent
kinetic energy dissipation. Furthermore, for ames with Lewis number near unity, the 1D dissipation spectra for
temperature and mixture fraction are shown to follow nearly the same rolloff in the high-wavenumber range, such
that the cutoff length scale for thermal dissipation is equal to or slightly smaller than the cutoff length scale for
mixture fraction dissipation. Measurements from the piloted CH
4
/air ames are used to demonstrate that a surro-
gate cutoff scale may be obtained from the dissipation spectrum of the inverse of the Rayleigh signal itself, even
when the Rayleigh scattering cross section varies through the ame. This suggests that the cutoff length scale
determined from Rayleigh scattering measurements may be used to dene the local resolution requirements and
optimal data processing procedures for accurate determination of the mean mixture fraction dissipation, based
upon Raman scattering measurements or other multiscalar imaging techniques.
2006 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
Keywords: Turbulent ames; Scalar dissipation; Dissipation length scales; Raman scattering; Rayleigh scattering
1. Introduction
The accurate measurement of scalar gradients is
important in a wide range of turbulent ow applica-
tions, including those that involve mixing, combus-
*
Corresponding author.
E-mail address: barlow@ca.sandia.gov (R.S. Barlow).
tion, and heat transfer [13]. For example, the mixture
fraction dissipation rate , dened as
(1) =2D

,
where is the mixture fraction and D

is the mix-
ture averaged mass diffusivity, is important in scalar
mixing because it is a measure of the rate at which
inhomogeneities in the scalar property are removed
by diffusion. Due to the importance of scalar gradi-
0010-2180/$ see front matter 2006 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.combustame.2006.09.005
G. Wang et al. / Combustion and Flame 148 (2007) 6275 63
ents and scalar dissipation in combustion, a great deal
of recent work has been directed toward their mea-
surement in turbulent ames using one-dimensional
(1D) [410] and two-dimensional (2D) [1115] imag-
ing techniques.
Such experiments are challenging because the
measurement of gradients at the smallest turbulent
scales requires both high precision and good spatial
resolution. Since scalar dissipation occurs mainly at
the nest scales of turbulence, it is very important to
quantify the errors imposed by the nite resolution of
the measurement system, particularly in the presence
of noise. Generally, spatial averaging and noise have
opposite effects, as limited resolution acts to reduce
scalar gradients, while noise tends to increase the
measured dissipation. These opposing effects make
the assessment of the experimental accuracy of dissi-
pation measurements in ames very difcult [4,9,16].
Most of the reported mixture fraction measure-
ments have not fully quantied these effects, espe-
cially in reacting ows. The effects of spatial reso-
lution on scalar dissipation measurements have been
more widely investigated in nonreacting turbulent
ows [1719], and it has been suggested [19] that
it is necessary to resolve approximately one or two
times the Batchelor scale, the nest scale over which a
scalar gradient can be sustained in a turbulent ow, in
order to obtain an accurate measurement of the mean
dissipation rate. The Batchelor scale [20] is dened as
(2)
B
=

D
2
/

1/4
=Sc
1/2
,
where =(
3
/)
1/4
is the Kolmogorov scale, the
kinematic viscosity, the mean rate of kinetic en-
ergy dissipation, and Sc =/D the Schmidt number.
In turbulent ames, the spatial averaging effect on the
measured scalar and its variance has been studied us-
ing a model spectrum from nonreacting turbulence
theory [21]. Similarly, scaling laws based on the non-
reacting theory have been used to estimate the resolu-
tion requirements for the scalar dissipation measure-
ments [16,22,23]. For example, using measurements
of in nonreacting circular jets [24,25], the Batch-
elor scale can be shown to be
(3)
B
=0.38(x x
0
)Re
3/4
d
Sc
1/2
,
where x is the downstream distance, x
0
is the virtual
origin, and Re
d
is the jet exit Reynolds number. Us-
ing the local Reynolds number Re

=U
C
/
C
, where
is the local full width at half maximum (FWHM)
thickness of the velocity prole, U
C
and
C
are the
jet centerline mean velocity and kinematic viscosity,
respectively, the Batchelor scale can be expressed as
(4)
B
=2.3Re
3/4

Sc
1/2
,
which has been used for scale estimation in thermal
dissipation studies [16,23] in the downstream regions
of jet ames. However, it is not clear that laws for esti-
mating turbulent scales in nonreacting turbulent ows
are broadly applicable in turbulent ames, because
there can be signicant variations in local properties
and local Reynolds numbers are typically low [26].
The effect of spatial averaging on scalar dissipa-
tion has been investigated by applying spatial lters
to ensembles of instantaneous mixture fraction dissi-
pation proles at several representative locations in
piloted methane/air jet ames (Sandia ames C, D,
and E) [68]. Extrapolation to zero lter length was
then used to estimate the fully resolved scalar dissi-
pation and to determine a representative length scale
corresponding to 90% of the fully resolved dissipa-
tion. However, these estimates did not account for
the limiting behavior of turbulence at the dissipation
scales or the effects of noise on the extrapolated re-
sults.
The temperature eld in ames can be measured
with considerably higher signal-to-noise ratio (SNR)
and resolution than the mixture fraction eld. This
enables the decoupling of effects of resolution and
noise on the measured thermal dissipation. Recently,
imaging of Rayleigh scattering was used to measure
1D thermal dissipation spectra and determine the dis-
sipation cutoff length scale at selected locations in
turbulent jet ames of CH
4
/H
2
/N
2
[27]. The local
length scale inferred from this cutoff is analogous to
the Batchelor scale in nonreacting ows. Application
of locally optimized digital lters in combination with
a noise correction technique yielded results for mean
thermal dissipation that were fully resolved and with
signicantly reduced noise [27].
It has been argued that the length scales of tem-
perature and mixture fraction dissipation structures
are similar and length scale information regarding
the thermal dissipation may be directly applied to
the mixture fraction dissipation eld in ows with
Lewis number near unity [16,23,27]. Thermal dissi-
pation measurements in the downstream regions of
jet ames [27] showed that the experimentally deter-
mined dissipation cutoff length scale is close to the es-
timate based on scaling laws using the local Reynolds
number (Eq. (4)). However, as far as we are aware,
the relationship between dissipation scales for tem-
perature and mixture fraction has not been measured
directly, until now.
In the present paper, we compare dissipation spec-
tra for temperature and mixture fraction, based upon
results from simultaneous line imaging of Raman
scattering, Rayleigh scattering, and CO-LIF in turbu-
lent nonpremixed CH
4
/H
2
/N
2
jet ames at Reynolds
numbers of 15,200 (DLR-A) and 22,800 (DLR-B)
and in piloted CH
4
/air jet ames at Reynolds num-
64 G. Wang et al. / Combustion and Flame 148 (2007) 6275
bers of 13,400, 22,400, and 33,600 (Sandia ames
C, D, and E). These ames have been previously
studied [68,15,16,23,2732] and have been used as
benchmarks for the TNF Workshop [33]. The shapes
of these dissipation spectra in the high-wavenumber
rolloff region are shown to be similar, with the ther-
mal dissipation spectra from the reaction zone be-
ing shifted to slightly higher wavenumber relative to
the corresponding mixture fraction dissipation spec-
tra. Furthermore, it is shown that surrogate spectra
and cutoff length scales may be obtained from the in-
verse of the Rayleigh signal, even when the Rayleigh
cross section varies through the ame. This suggests
that local dissipation length scales and spatial resolu-
tion requirements in complex turbulent reacting ows
may be determined from Rayleigh imaging experi-
ments, which are relatively easy to conduct.
2. Experimental setup and data reduction
2.1. Experimental setup
The experimental setup has been described previ-
ously [68], and only a brief overview is presented
here. Four frequency-doubled Nd:YAG laser beams
were combined for the Raman/Rayleigh measure-
ments. The combined beam of roughly 1.6 J/pulse
was focused by a 500-mm lens to a waist diameter
of about 0.22 mm (1/e
2
) in the test section. Raman
scattered light was collected using a pair of 150-mm-
diameter achromats (f/2 and f/4). The f/4 lens
matched the entrance aperture of an imaging spec-
trometer, which was tted with a high-speed rotating
mechanical shutter [34]. A 16-bit, back-illuminated,
cryogenically cooled CCD camera (1300 1340 pix-
els) was used to detect the Raman spectrum. To re-
duce readout noise and hence improve the overall
SNR, pixels were binned on chip to form 28 su-
perpixels along the spatial dimension of the CCD,
corresponding to a spatial sampling resolution of
0.220 mm along the laser beam, and 14 superpixels
along the spectral dimension corresponding to the Ra-
man bands of the major species (CO
2
, O
2
, CO, N
2
,
CH
4
, H
2
O, H
2
) and interferences in between. The
effective experimental resolution for the combined
Raman/Rayleigh/CO-LIF measurements of mixture
fraction, including optical blur and data processing ef-
fects, is estimated at roughly 0.30 mm.
Rayleigh-scattered light was collected by two
matched achromats (82 mm diameter, 300 mm ,
1:1 imaging). The collected light was focused onto
a back-illuminated CCD detector (400 1340 pix-
els) through a 532-nm bandpass lter. Gating was
provided by a mechanical leaf shutter. The imaged
region was 6 mm in length along the laser beam
and 0.9 mm in height. For measurements in the
CH
4
/H
2
/N
2
ames the 0.020-mm pixels were binned
by 2 in the laser beam direction, giving a nominal
projected resolution of 0.040 mm. For the earlier
measurements in the piloted CH
4
/air ames the pix-
els were binned by 3, giving a nominal projected
resolution of 0.060 mm. The optical resolution of
the Rayleigh imaging system was approximately
0.050 mm, based upon ZEMAX calculations of the
point spread function as well as direct measurements.
The CO LIF imaging system used the same front
collection lens as the Rayleigh system. A dichroic
beam splitter in the collimated region reected CO
uorescence (484 nm) through another matched
achromat. An interference lter centered at 484 nm
(10 nm FWHM) passed the CO uorescence signal
onto an intensied CCD camera (512 512 pixels).
Typically 6000 samples were collected at each mea-
surement location.
2.2. Raman data reduction
Processing steps for the Raman, Rayleigh, and CO
LIF images included dark frame subtraction, correc-
tion for nonuniform throughput and response of each
imaging system, background subtraction, and normal-
ization by laser energy. For determination of tem-
perature and species mass fractions, the normalized
Rayleigh and CO-LIF line images were mapped lin-
early onto the Raman superpixels. The signals from
each superpixel were then converted to scalar quan-
tities using an iterative matrix inversion process [8].
The two photon LIF technique is superior to Raman
scattering for the measurement of CO in methane
ames [29], and LIF results for CO are used exclu-
sively in the current study.
The mixture fraction was calculated from the mea-
sured species mass fractions using Bilgers formula-
tion [35],
=

2(Y
C
Y
C,2
)/W
C
+(Y
H
Y
H,2
)/2W
H
(Y
O
Y
O,2
)/W
O

2(Y
C,1
Y
C,2
)/W
C
+(Y
H,1
Y
H,2
)/2W
H
(5) (Y
O,1
Y
O,2
)/W
O

,
where Ys are elemental mass fractions, Ws are atomic
weights, and subscripts 1 and 2 refer to the main
jet and coow air stream, respectively. To reduce
the measurement noise and interference effects on
the measured mixture fraction for the Sandia piloted
CH
4
/air ames C, D, and E, Bilgers denition was
modied as in previous work [8,36] to exclude the
oxygen terms:
(6) =
2(Y
C
Y
C,2
)/W
C
+(Y
H
Y
H,2
)/2W
H
2(Y
C,1
Y
C,2
)/W
C
+(Y
H,1
Y
H,2
)/2W
H
.
G. Wang et al. / Combustion and Flame 148 (2007) 6275 65
2.3. Gradient calculation
The radial derivatives of scalars were calcu-
lated using an implicit fourth-order compact nite-
difference scheme, which was developed as a higher-
order method for evaluating derivatives by Lele [37]
and is expressed as
(
,r
)
i2
+(
,r
)
i1
+(
,r
)
i
+(
,r
)
i+1
+(
,r
)
i+2
=c

i+3

i3
6
+b

i+2

i2
4
(7) +a

i+1

i1
2
,
where is the pixel spacing, parameters , , a,
b, and c are calculated by substituting Taylor se-
ries expansion coefcients, represents the values
of the scalar, and
,r
represents the implicitly de-
termined local spatial derivative of the scalar along
the radial direction. The parameters , , a, b, and c
are determined by the order of accuracy and number
of stencil points. In the current study a spectral-like
fourth-order, seven-point stencil, which has the high-
est resolving efciency, was used, with parameters
=0.5771439, =0.0896406, a =1.3025166, b =
0.9935500, and c = 0.03750245 determined through
an optimization process described in [37]. For the
compact high-order spectral-like stencil, the resolv-
ing efciency refers to the maximum normalized
wavenumber at which the error between the compact
and the ideal stencil is less than a certain tolerance
(10% for example). The boundary condition was han-
dled following the procedure in [37].
2.4. One-dimensional energy and dissipation spectra
The procedure for obtaining the 1D scalar energy
spectrum at each measurement location is to calcu-
late a discrete, one-sided spectrum from the sequence
of each measured scalar, , such as mixture fraction,
temperature, or inverted Rayleigh signal. The aver-
age scalar value is subtracted from the instantaneous
value,

= , before the power spectrum is cal-


culated. The 1D energy spectrum is then calculated
as
(8) E
1
(
l
) =
2
N
S

(l)

2
,
where
S
= 2/ is the sampling wavenumber, N
the number of samples in the data sequence,
l
=

S
l/N the discrete wavenumber, and l = 0, 1, . . . ,
N/2 1. (l) is the discrete Fourier transform of the
uctuating quantity

,
(9) (l) =
N1

n=0

(n)W
nl
N
,
Fig. 1. Normalized 1D energy and dissipation spectra calcu-
lated from the model spectrum for turbulent kinetic energy
from Pope [38] at Taylor Reynolds number Re

=100. The
shaded region indicates the wavenumber interval resolved
by the current 1D line Rayleigh imaging measurements, and
the large arrow indicates the 2% dissipation cutoff location,

1
=

1
=1.
where W
N
=e
i2/N
.
The ideal dissipation spectrum for turbulent uc-
tuations in scalar is generally dened as D
1
(
l
) =
2D

2
l
E
1
(
l
). In reacting ows, the scalar diffusiv-
ity D

is a function of species concentrations and a


strong function of temperature, which is not the case
in nonreacting ows where D

const. Therefore,
for the purposes of the present work, a surrogate of
the scalar dissipation spectrum is dened as
(10) D
1
(
l
) =2
2
l
E
1
(
l
).
This simplication may be justied in terms of scale
separation in turbulent reacting ow. The thermal and
mass diffusivities are dependent upon temperature
and composition, so uctuations in diffusivity occur
at length scales corresponding to the energy spectra
for uctuations in temperature and mixture fraction.
While there is some overlap of energy and dissipa-
tion spectra in the present ames, due to relatively low
local Reynolds number, the energy content of uctua-
tions in diffusivity within the dissipation range is very
small and will not inuence the high-wavenumber
rolloff of the dissipation spectrum. Therefore, while
diffusivity affects the mean dissipation rate, it will
have a negligible effect on the dissipation length scale
determined from the 1D spectrum. Separation of the
energy spectrum from the portion of the dissipation
spectrum of interest here is illustrated in Fig. 1 us-
ing the model spectrum of Pope [38] with turbulence
Reynolds number representative of the present ame
conditions.
For each instantaneous laser shot, the energy spec-
trumcan be calculated fromthe 1Dmeasurement as in
Eq. (8). Thousands of instantaneous laser shots were
taken at each measurement location to obtain good
statistics, and the ensemble average energy spectrum
66 G. Wang et al. / Combustion and Flame 148 (2007) 6275
was determined as
(11)

E
1
(
l
)

=
1
N
L
N
L

j=0
E
(j)
1
(
l
),
where j = 0, 1, . . . , N
L
indicates the jth laser shot,
N
L
is the total number of laser shots, and E
(j)
1
is
the instantaneous energy spectrum calculated at the
jth laser shot from Eq. (8). Integration in the spec-
tral domain of the complete scalar energy spectrum
gives the scalar variance, and the integral of the com-
plete power spectrum of the uctuating scalar gra-
dient gives the variance of the gradient, which is
just the mean scalar dissipation rate without diffusiv-
ity. The present experiments resolve only the high-
wavenumber portion of the scalar energy and dissipa-
tion spectra, as illustrated in Fig. 1. For such measure-
ments, the scalar variance and mean dissipation rate
can be determined from the ensemble averages of the
scalar and its gradient at each measurement location,
after proper ltering, as in [27].
If the noise from different image pixels is uncorre-
lated, there will be a at noise oor (NF) in the calcu-
lated energy spectrum. Using Parsevals theorem, the
noise oor can be determined as
(12) NF =2
2
n
/
S
,
where
2
n
is the variance of the noise averaged
across the line image. The measured energy spectrum
E
1
(
l
)
m
can then be corrected to get the noise-free
energy spectrum E
1
(
l
) as
(13)

E
1
(
l
)

E
1
(
l
)

m
NF.
A similar procedure has been applied to time-
series density uctuation measurements in high-speed
jets [39] and to time-series OH uorescence and tem-
perature measurements in turbulent ames [31,40].
Equation (12) can also be used to evaluate the noise
level in the measured signal as
2
n
=NF
S
/2, which
has been used to correct the apparent dissipation in
thermal dissipation measurements [16]. Additional
discussion of the effect of noise on the measured
dissipation spectra is included in the following sec-
tion.
The
2
l
term in Eq. (10) is the ideal spectral dif-
ference stencil and can only be approximated when
differentiating discrete experimental data in the physi-
cal domain. An alternate procedure for calculating the
1D dissipation spectrum is to differentiate the spatial
proles rst and then calculate the power spectrum,
so that the dissipation spectrum is determined directly
from the gradient of the uctuating scalar. It should
be noted that the dissipation spectrum calculation is
a linear system, so switching the order of these cal-
culations will not affect the resulting 1D dissipation
spectrum, as long as the spectral and spatial stencils
are consistent. However, in practical data processing,
the radial prole is of nite length, and the boundary
effects were attenuated by applying Blackman win-
dowing [41], which will break the equality between
1D dissipation spectra calculated from Eq. (10) and
from using the gradient of the uctuating scalar. The
1D spectrum of the gradient of the uctuating scalar
was used in the current study since it includes effects
of nite record length and boundary condition directly
in the calculated dissipation spectrum, which is con-
sistent with the scalar energy spectrum. The compact
high-order spectral-like stencil also works as a digital
bandpass lter that will reduce the noise effects in the
high-wavenumber region. This is different from the
ideal difference stencil in Eq. (10), which is a digital
high-pass lter in nature.
3. Results and discussion
3.1. Dissipation spectra and cutoff length scale in
DLR CH
4
/H
2
/N
2
jet ames
Identifying the spatial length scale that corre-
sponds to the high-wavenumber cutoff in turbulent
uctuations is important in scalar gradient and dis-
sipation measurements, especially for laser diagnos-
tic techniques, where higher resolution will gener-
ally result in lower signal levels and lower SNR.
It is well known in nonreacting turbulence theory
[38,42,43] that the cutoff in the kinetic-energy dis-
sipation spectrum is at about =

= 1, where
is the wavenumber with units rad/m and is the
Kolmogorov scale. Note that the physical wavelength
corresponding to this cutoff is 2 6, which in-
dicates that scales smaller than 6 contribute little to
the total mean dissipation [38]. For Schmidt number
near unity we have
B
, where
B
is the Batchelor
scale, and we expect the cutoff to be nearly the same
for the scalar and velocity spectra at a given ow con-
dition.
Fig. 1 illustrates the region of the turbulence spec-
trum that is coherently accessed in the current mea-
surements. The cutoff wavenumber at
1
= 1 cor-
responds to 2% of the peak value in the 1D model
dissipation spectrum, as indicated by the arrow in
Fig. 1. The dissipation spectrum decays exponentially
after passing the peak value, and this is referred to
the rolloff region in the current study. As will be
shown, the full-resolution Rayleigh measurements,
which have higher SNR and better spatial sampling
resolution than the Raman measurements (0.040 or
0.060 mm vs 0.220 mm), have a spatial dynamic
range that includes the peak in the dissipation spec-
trum (in most cases) and extends beyond the dissipa-
G. Wang et al. / Combustion and Flame 148 (2007) 6275 67
Fig. 2. Comparison of the measured 1D dissipation spectra of mixture fraction, , temperature, T , and the inverse of the Rayleigh
signal, I , from various radial locations at x/d = 10 and 20 in ames DLR-A and DLR-B. The 1D dissipation spectra are
normalized by their own maximum values.
tion cutoff, such that the length scale corresponding
to the cutoff may be determined at each measurement
location.
As discussed above, the energy spectrum of the
mixture fraction gradient or temperature gradient is
calculated without diffusivity, and we refer to these
here as the mixture fraction and thermal dissipa-
tion spectra for brevity. Fig. 2 compares normal-
ized 1D dissipation spectra for three quantities (mix-
ture fraction, ; temperature, T ; and the inverse of
the Rayleigh signal, I ) measured in ames DLR-A
and DLR-B with the 6-mm probe on the centerline
(3 mm < r < 3 mm) and at radial locations corre-
sponding to maxima in the rms mixture fraction, mean
temperature, and rms temperature, as listed in Table 1
(only three positions at x/d =10).
Each measured spectrum in Fig. 2 is normalized
by its own maximum value. The cutoff wavenum-
ber and the corresponding length scale for each mea-
surement location was determined by considering the
68 G. Wang et al. / Combustion and Flame 148 (2007) 6275
Table 1
Experimentally determined cutoff length scales
I

[m] for
cases of x/d =10 and 20 in ames DLR-A and DLR-B
Flame Re
d
x/d
I

(r/d)
DLR-A 15,200 10 56 (0) 92 (1.0) 105 (1.5)
20 95 (0) 121 (1.4) 125 (2.0) 146 (2.6)
DLR-B 22,800 10 43 (0) 71 (1.0) 86 (1.5)
20 76 (0) 95 (1.4) 106 (2.0) 124 (2.6)
Note.
I

is from high-resolution 1D dissipation spectra of


the inverted Rayleigh scattering signal. Radial positions nor-
malized by the jet diameter are shown in parentheses. The
underscored values indicate cases that are underresolved by
the combined 1D Raman/Rayleigh/CO-LIF setup.
model spectrum of Pope [38] (see Fig. 1), which
shows that the cutoff (
1

B
= 1) corresponds to a
power level of about 2% of the peak in the 1D dissi-
pation spectrum. For each measurement location the
2% dissipation cutoff was determined from the dissi-
pation spectrum for the inverse of the Rayleigh signal
(I -spectrum), which has the highest spatial resolu-
tion and lowest noise oor. The local length scale
I

determined from
I

=1 is analogous to the Batch-


elor scale
B
. If the peak was not observed in the
measured 1D spectrum, the highest value in the mea-
sured dissipation spectrum was taken as the reference.
This gives a conservative estimate of the spatial res-
olution requirement because the dissipation at higher
wavenumber is expected to contribute less than 1%
to the mean dissipation, based on integration of the
model spectrum. Due to the exponential rolloff of the
dissipation spectrum, the error in length scale deter-
mined from data that do not quite reach the true dissi-
pation peak will be very small. An alternative method
would be to t the measured spectrum to the model
spectrum, but differences in the resulting cutoff val-
ues would be unimportant in the present work.
It should be noted that a at noise oor (uncorre-
lated noise) in the 1D measurements of a given scalar
yields a slope of 2 in the corresponding dissipation
spectrum. The valley in each dissipation spectrum in
Fig. 2 is determined by competition between the true
dissipation rate and noise in the measurement. It is
clear that the highest wavenumber portion of each
measured spectrum is inuenced or even dominated
by noise. The noise oor for mixture fraction is much
higher than that for temperature. The noise oor for I ,
the inverse of the Rayleigh signal, is the lowest be-
cause it is unaffected by the subsampling and post-
processing steps applied in the combined data analy-
sis to obtain species mass fractions and temperature.
It is important to note that the fuel mixture in
the DLR ames has a Rayleigh scattering cross sec-
tion that matches air, and there is only a small vari-
ation (3%) in cross section throughout the ame
[28,29]. Therefore, the inverse of the Rayleigh sig-
nal is related to temperature by a constant, and the
normalized dissipation spectra for I in these special-
ized ames are essentially thermal dissipation spectra.
The spectra labeled T in Fig. 2 are based on temper-
atures obtained from the combined Raman/Rayleigh
processing, which involves subsampling of the high-
resolution Rayleigh image to match the Raman res-
olution. The T -spectra have lower resolution and
higher noise oor than the I -spectra, but they are oth-
erwise nearly identical in the DLR ames because
they represent the same scalar eld.
Using the above procedure, the estimated wave-
number for the dissipation cutoff measured on the
centerline at x/d = 20 in DLR-A is roughly
I

=
10,500 rad/m. The corresponding scale
I

= 0.095
mm is comparable to the value
B
= 0.070 mm
estimated from scaling laws using local Reynolds
number, as in [16,23,27]. Note that the physical
wavelength corresponding to the cutoff is 2
I

(0.6 mm) for this local ame condition. Table 1 lists


values of
I

for all measurement locations.


The dissipation spectra normalized by the cutoff
length scales from Table 1 are shown in Fig. 3 for
both Reynolds number cases. It can be seen that af-
ter normalization the dissipation spectra for , T ,
and I nearly collapse until the points in the and
T spectra where noise becomes important. The cut-
off wavenumber corresponding to
I

= 1 is close
to the end of the measured dissipation spectra for
and T , and at some locations the cutoff is smaller than
the experimental resolution for . Furthermore, it can
be seen that the normalized spectra follow nearly the
same rolloff as the model spectrum, which implies
that the measured 1D dissipation spectra follow ex-
ponential decay in the high-wavenumber region. This
indicates that turbulence structure at the dissipation
scales in jet ames, as represented by averaged spec-
tra, is similar to that in nonreacting turbulent ows,
which is an important experimental observation that
should be investigated in a wider range of ames.
The values in Table 1 show that the local dissi-
pation cutoff scale increases with radial distance at
x/d =10 and 20 in both DLR ames, which indicates
that the dissipation spectra shift to lower wavenumber
as one moves from the centerline to the outer edge
of the jet ame. The dissipation spectra are nearly
overlapped (length scale nearly the same) for the two
intermediate radial locations, i.e., at the maxima in the
rms mixture fraction and mean temperature, which
are both close to the mean reaction zone. This implies
that there are three mixing regimes at this streamwise
location in these two jet ames. Near the centerline
there is mixing between hot combustion products and
relatively cold, fuel-rich uid having high convec-
tive velocity, and this yields a relatively high effec-
G. Wang et al. / Combustion and Flame 148 (2007) 6275 69
Fig. 3. Comparison of the normalized 1D dissipation spectra of mixture fraction, , temperature, T , and the inverse of the
Rayleigh signal, I , from various radial locations at x/d =10 and 20 in ames DLR-A and DLR-B. The 1D model dissipation
spectrum is based on the model spectrum for turbulent kinetic energy from Pope [38] at Taylor Reynolds number Re

=100.
tive Reynolds number and small dissipation length
scale. Close to the mean reaction zone, ow turbu-
lence is damped by viscosity at high temperature,
which results in a smaller effective Reynolds num-
ber and larger dissipation length scales. Toward the
edge of the ame, products mix with cold coowing
air, reducing viscosity. However, the dissipation spec-
trum continues to shift to lower wavenumber (larger
cutoff scale), indicating a further decrease in the lo-
cal effective Reynolds number toward the edge of the
ame, where the shear rate is low. This behavior is
qualitatively consistent with observations regarding
the length scale associated with turbulent uctuations
of the mixture fraction in piloted jet ames [8].
Figs. 2 and 3 show that the 1D dissipation spec-
tra of mixture fraction from line Raman imaging have
70 G. Wang et al. / Combustion and Flame 148 (2007) 6275
Fig. 4. Measured instantaneous 1D radial proles of (a) tem-
perature and mixture fraction; (b) squared gradients of tem-
perature and mixture fraction at x/d = 20, r/d = 1.4. T
,r
and
,r
are the radial gradients of T and , respectively.
lower SNR and reduced spatial resolution compared
to spectra for the inverse Rayleigh signal. Typically,
the wavenumber for 2% of the peak level cannot be
determined directly from the measured 1D mixture
fraction dissipation spectra. Comparing the 1D ther-
mal and mixture fraction dissipation spectra in Fig. 3,
it can be seen that the 1D thermal dissipation spectra
follow nearly the same rolloff as the mixture fraction
dissipation spectra at the centerline. When moving
toward the edge of the jet ame, the 1D mixture frac-
tion dissipation spectra roll off faster. The difference
is greatest at the radial location corresponding to the
maximum mean temperature.
This is believed to result from the fact that ther-
mal dissipation has a double-peak structure whenever
maximum temperature is crossed (near stoichiometric
mixture). This is illustrated in Fig. 4 using a sin-
gle measurement realization as an example. Fig. 4a
includes instantaneous radial proles of mixture frac-
tion and temperature measured near the reaction zone
at x/d = 20. The corresponding radial proles of
square gradients are shown in Fig. 4b. The stoi-
chiometric mixture fraction is
st
= 0.167 for the
DLR fuel combination. It can be seen that there are
two thermal and mixture fraction dissipation struc-
tures within the fuel-rich region (8 < r < 11 mm) in
Fig. 4b, and these structures have about the same
thickness. However, when crossing the maximum
temperature condition near stoichiometric mixture
(11 < r < 14 mm), there are two thermal dissipation
structures for a single-mixture-fraction dissipation
structure. Frequent occurrence of such double-peaked
structures is expected to contribute to a shift in the
thermal dissipation spectrum relative to the corre-
sponding spectrum for mixture-fraction dissipation,
so that the cutoff length scale for thermal dissipation
will be smaller than that for mixture-fraction dissipa-
tion.
Based upon these results, it appears that the dis-
sipation length scale determined from the 1D dissi-
pation spectrum of the uctuating temperature will
always be equal to or slightly smaller than the cor-
responding length scale for mixture-fraction dissipa-
tion. Therefore, in the context of determining spa-
tial resolution requirements for measuring the mean
mixture-fraction dissipation, a local cutoff length
scale determined from the thermal dissipation spec-
trum (or the inverse Rayleigh dissipation spectrum)
will lead to a conservative estimate of the required
experimental resolution.
3.2. Dissipation spectra and cutoff length scale in
the piloted CH
4
/air jet ames
Results from the DLR ames demonstrated that
dissipation spectra and the corresponding cutoff
length scale could be obtained directly from the in-
verse of the Rayleigh signal, which has higher SNR
and better spatial resolution than the fully processed
temperature data from the combined Raman/Rayleigh
measurements. In this section, measurements from
two piloted CH
4
/air jet ames [68] are used to
demonstrate that this approach is not limited to ames
having nearly constant Rayleigh cross section. These
piloted ames have a fuel jet composition of 25%
CH
4
and 75% air, and the Rayleigh cross section
of the fuel is roughly 1.3 times that of air. The Ra-
man/Rayleigh/LIF data sets used here were described
previously [68]. The Rayleigh line images in these
earlier experiments were acquired with resolution
0.060 mm in the laser beam direction, using on-chip
binning by three pixels, and these high-resolution 1D
Rayleigh data were processed in the present work to
obtain surrogate dissipation spectra and local cutoff
length scales based on the inverse Rayleigh signal.
Fig. 5 shows the measured 1D dissipation spectra
of the uctuating mixture fraction, temperature, and
the inverse Rayleigh signal. The measurement loca-
tions are at x/d = 7.5, 15, 30 and radial locations
of the probe center corresponding to the maximum
mixture fraction variance [68]. As described above,
each spectrum is normalized by its maximum value.
Similar to the DLR-A and DLR-B results, the 1D
uctuating mixture fraction dissipation spectra show
a much higher noise oor than those of temperature
and inverted Rayleigh signal. With the peak of the
1D dissipation spectra resolved, the mixture fraction
spectrum does not have high enough SNR to reach the
2% cutoff wavenumber. It can also be seen that the
G. Wang et al. / Combustion and Flame 148 (2007) 6275 71
Fig. 5. Comparison of the measured 1D dissipation spectra of mixture fraction, , temperature, T , and the inverse of the Rayleigh
signal, I , at x/d = 30, 15, and 7.5 in Sandia ames C, D, and E. The 1D dissipation spectra are normalized by their own
maximum values.
1D dissipation spectra of the uctuating mixture frac-
tion, temperature, and inverted Rayleigh signal follow
nearly the same rolloff before the different noise lev-
els cause them to diverge. This shows that the dissipa-
tion length scale determined from the 1D dissipation
spectra of the inverted laser Rayleigh scattering signal
can be used to approximate both the mixture fraction
and thermal dissipation cutoff length scales. Using the
same procedure as for the DLR ames, Table 2 also
lists values of
I

for the pilot jet ames C, D, and E


at downstream locations of x/d =7.5, 15, 30 and ra-
dial locations of the probe center corresponding to the
maximum mixture fraction variance [68].
The normalized 1D dissipation spectra are also
compared in Fig. 6 with the 1D model spectrum
from Pope [38]. Here, the wavenumbers of the mea-
sured spectra are normalized by the cutoff length scale

determined from the 2% level in the dissipation


spectrum for the inverse Rayleigh signal. As for the
DLR-A and DLR-B cases, the measured and model
dissipation spectra follow nearly the same exponen-
tial rolloff. It is clear from Fig. 5 that the dissipa-
tion spectra from the inverse Rayleigh signal may be
used in place of the thermal dissipation spectra, even
though the Rayleigh cross section is not constant in
these piloted jet ames. This result may also be ex-
72 G. Wang et al. / Combustion and Flame 148 (2007) 6275
Table 2
Experimentally determined cutoff length scales
I

[m] for
cases of x/d =7.5, 15, and 30 in Sandia ames C, D, and E
Flame Re
d
Streamwise location x/d
Probe center (r/d)
7.5 15 30
(0.83) (1.11) (1.39)
C 13,400 108 123 140
D 22,400 81 86 106
E 33,600 65 71 89
Note.
I

is from high-resolution 1D dissipation spectra of


the inverted Rayleigh scattering signal. Radial positions nor-
malized by the jet diameter are shown in parentheses. The
underscored values indicate cases that are underresolved by
the combined 1D Raman/Rayleigh/CO-LIF setup.
plained on the basis of scale separation because the
Rayleigh cross section depends on composition, and
so uctuations in the Rayleigh cross section occur at
length scales corresponding to the energy spectrum
for mixture fraction uctuations. There is negligible
energy at length scales corresponding to the portion
of the dissipation spectra considered here, especially
in the rolloff region, so the cutoff length scale for
thermal dissipation is not inuenced by the fact that
the Rayleigh cross section is not constant across the
ame. It is clear that the cutoff length scale obtained
from the inverse Rayleigh spectra may be used to
determine local cutoff length scale for thermal dissi-
pation. Furthermore, this local surrogate length scale
may be used to estimate the corresponding length
scale for mixture-fraction dissipation and resolution
requirements for accurate measurement of the mean
mixture-fraction dissipation.
3.3. Discussion
The two sets of ames considered here cover mod-
est ranges in Reynolds number (see Tables 1, 2).
Therefore, it is of interest to compare the measured
scale ratios in these ames with those predicted from
the Reynolds number dependence of the scaling laws
(Eqs. (3), (4)) for nonreacting jets. At each down-
stream location in the piloted ames, the dissipation
cutoff scales satisfy the relation (
I

)
E
< (
I

)
D
<
(
I

)
C
. This is consistent with the jet exit Reynolds
number, (Re
d
)
E
> (Re
d
)
D
> (Re
d
)
C
. Similar results
are also found in the DLR-A and DLR-B ames.
The mean ratios of these cutoff lengths, (
I

)
E
/(
I

)
D
and (
I

)
D
/(
I

)
C
, are 0.82 and 0.74, respectively.
The corresponding ratio, (
I

)
B
/(
I

)
A
, for the DLR
ames is between 0.77 near the centerline, 0.85 at
larger radius, and the mean value is about 0.81. These
values are close to, but consistently higher than, the
predicted ratios fromthe Re
3/4
scaling for nonreact-
ing jets. Note that there is a signicant degree of lo-
calized extinction in ame E and also some extinction
in ame DLR-B. Furthermore, the local turbulence
Reynolds numbers in these ames are relatively low,
and the measurement locations are in the upstream
portion of these jet ames and in a region of develop-
ing turbulence. Therefore, one would not necessarily
expect the Re
3/4
scaling to be followed. Measure-
ments in ames of higher Reynolds number would be
useful. These considerations reinforce the importance
of using direct measurements of the local dissipation
scale, as applied here, as opposed to estimates from
scaling laws.
As discussed earlier and following Pope [38],

is the nominal cutoff scale determined from the


measured dissipation spectrum as
I

= 1/
I

, and
the physical wavelength corresponding to the cutoff
wavenumber
I

is 2
I

. The experimental resolu-


tion required to resolve this length scale is
I

ac-
cording to the Nyquist criteria. It should be empha-
sized that the experimental resolution includes not
only the sampling resolution but also any spatial av-
eraging effects from optical blur and data processing.
For the current combined Raman/Rayleigh/CO-LIF
setup, the sampling resolution is 220 m. This im-
plies that the current experimental system combined
with a sharp-spectral stencil can resolve the mean dis-
sipation at ame conditions corresponding to a cutoff
length scale,
I

, as small as 70 m.
Cases that are underresolved by the current 1D
Raman/Rayleigh/CO-LIF experimental setup are in-
dicated in Tables 1 and 2, based on the dissipation cut-
off length scale determined from the inverse Rayleigh
signal. The experimentally determined cutoff scale
generally increases with downstream distance in each
ame. At each streamwise location, the cutoff scale is
smallest on the centerline and increases with radial
distance. This suggests that the resolution require-
ment is most stringent in the jet ame near-eld re-
gion. For example, the dissipation cutoff length scale
is 43 m at the centerline of x/d = 10 in DLR-B
ame, so the required experimental resolution (
I

)
for accurate measurement of the mean dissipation is
roughly 135 m, which is beyond the current spatial
resolution of 220 m. This results in deviation be-
tween the 1D dissipation spectra of temperature and
inverse Rayleigh signal, as seen in Figs. 2 and 3. Res-
olution of the dissipation cutoff scale will be even
more challenging in higher-Reynolds-number ames.
However, if one assumes that the dissipation rolloff
follows the form of the model spectrum, as is indi-
cated by the present results, then the cutoff scale may
be estimated in any ame where the peak in the dis-
sipation spectrum can be resolved using the methods
G. Wang et al. / Combustion and Flame 148 (2007) 6275 73
Fig. 6. Comparison of the normalized 1D dissipation spectra of mixture fraction, , temperature, T , and the inverse of the
Rayleigh signal, I , at x/d =30, 15, and 7.5 in Sandia ames C, D, and E. The 1D model dissipation spectrum is based on the
model spectrum for turbulent kinetic energy from Pope [38] at Taylor Reynolds number Re

=100.
applied here. Under these same assumptions it should
be possible to estimate the unresolved portion of the
mean thermal or mixture fraction dissipation, pro-
vided that the dissipation peak is resolved in a given
experiment, and this is an area of ongoing research.
Direct measurement of the local scalar spectrum
is also expected to be an effective approach to deter-
mining local turbulence scales and experimental reso-
lution requirements in complex ames, such as bluff-
body or swirl-stabilized ames, where simple scal-
ing laws for estimation of turbulent length scales are
not expected to be accurate. The fact that the inverse
of the Rayleigh signal may be used directly in this
analysis, without the need to know the local Rayleigh
scattering cross section, is a signicant experimen-
tal nding. It suggests that turbulence length scales
may be determined in complex reacting ows using
relatively simple laser diagnostics. Length scales de-
termined from experiments such as these may also be
useful for the design of grids for large eddy simulation
(LES) or for the evaluation of the relative resolution
of LES results.
4. Conclusions
1D imaging of Raman/Rayleigh/CO-LIF was used
to study scalar dissipation spectra in turbulent non-
74 G. Wang et al. / Combustion and Flame 148 (2007) 6275
premixed CH
4
/H
2
/N
2
jet ames at Reynolds num-
bers of 15,200 and 22,800 (DLR-A and DLR-B) and
piloted CH
4
/air jet ames at Reynolds numbers of
13,400, 22,400, and 33,600 (Sandia ames C, D, and
E). The local length scale corresponding to the cutoff
in the measured 1D dissipation spectrum was deter-
mined based on consideration of the model turbulent
kinetic energy dissipation spectrum of Pope. This lo-
cal length scale is analogous to the Batchelor scale in
nonreacting ows. The local thermal dissipation cut-
off length scale was shown to be equal to or slightly
smaller than the corresponding mixture fraction dis-
sipation cutoff length scale. The measured dissipa-
tion spectra for temperature and the inverse of the
Rayleigh signal were shown to follow nearly the same
exponential rolloff, even when the Rayleigh cross sec-
tion varies across the ame. The exponential rolloff
is similar to the model spectrum by Pope for nonre-
acting turbulent ows. A surrogate dissipation length
scale determined from the 1D dissipation spectra of
the inverted Rayleigh signal may be used to estimate
the mixture fraction dissipation cutoff length scale.
Acknowledgments
The U.S. Department of Energy, Ofce of Ba-
sic Energy Sciences, Division of Chemical Sciences,
Geosciences, and Biosciences supported this work.
Sandia National Laboratories is a multiprogram lab-
oratory operated by Sandia Corporation, a Lockheed
Martin Company, for the United States Department
of Energy under Contract DE-AC04-94-AL85000.
Excellent technical support by R.A. Harmon and
M. Boisselle in operating the Turbulent Combustion
Laboratory is gratefully acknowledged.
References
[1] R.W. Bilger, Prog. Energy Combust. Sci. 1 (1976) 87
109.
[2] N. Peters, Turbulent Combustion, Cambridge Univ.
Press, Cambridge, UK, 2000.
[3] D. Veynante, L. Vervisch, Prog. Energy Combust. Sci.
28 (2002) 193266.
[4] S.P. Nandula, T.M. Brown, R.W. Pitz, Combust. Fla-
me 99 (1994) 775783.
[5] A. Brockhinke, S. Haufe, K. Kohse-Hinghaus, Com-
bust. Flame 121 (2000) 367377.
[6] A.N. Karpetis, R.S. Barlow, Proc. Combust. Inst. 29
(2002) 19291936.
[7] R.S. Barlow, A.N. Karpetis, Proc. Combust. Inst. 30
(2005) 673680.
[8] R.S. Barlow, A.N. Karpetis, Flow Turb. Combust. 72
(2004) 427448.
[9] D. Geyer, 1D-Raman/Rayleigh Experiments in a Tur-
bulent Opposed-Jet, Ph.D. thesis, Technical University
of Darmstadt, 2004.
[10] D. Geyer, A. Kempf, A. Dreizler, J. Janicka, Combust.
Flame 143 (2005) 524548.
[11] S.H. Strner, R.W. Bilger, M.B. Long, J.H. Frank,
D.F. Marran, Combust. Sci. Technol. 129 (1997) 141
163.
[12] J.B. Kelman, A.R. Masri, Appl. Opt. 36 (1997) 3506
3514.
[13] J. Fielding, A.M. Schaffer, M.B. Long, Proc. Combust.
Inst. 27 (1998) 10071014.
[14] S.A. Sutton, J.F. Driscoll, Proc. Combust. Inst. 29
(2002) 27272734.
[15] J.H. Frank, S.A. Kaiser, M.B. Long, Combust. Fla-
me 143 (2005) 507523.
[16] G.-H. Wang, N.T. Clemens, P.L. Varghese, Appl. Opt.
44 (2005) 67416751.
[17] J.C. Wyngaard, Phys. Fluids 14 (1971) 20522054.
[18] R.A. Antonia, J. Mi, Exp. Fluids 14 (1993) 203
208.
[19] J. Mi, G.J. Nathan, Exp. Fluids 34 (2003) 687
696.
[20] G.K. Batchelor, J. Fluid Mech. 5 (1959) 113133.
[21] M.S. Mansour, R.W. Bilger, R.W. Dibble, Combust.
Flame 82 (1990) 411425.
[22] L.L. Smith, R.W. Dibble, L. Talbot, R.S. Barlow, C.D.
Carter, Combust. Flame 100 (1995) 153160.
[23] G.-H. Wang, N.T. Clemens, P.L. Varghese, Proc. Com-
bust. Inst. 30 (2005) 691699.
[24] C.A. Friehe, C.W. Van Atta, C.H. Gibson, Jet turbu-
lence: Dissipation rate measurements and correlations,
AGARD-CP-93, NATO Advisory Group for Aerospace
Research and Development (AGARD), 1971.
[25] R.A. Antonia, B.R. Satyaprakash, A.K.M.F. Hussain,
Phys. Fluids 23 (1980) 695700.
[26] L. Muniz, M.G. Mungal, Combust. Flame 126 (2001)
14021420.
[27] G.-H. Wang, R.S. Barlow, N.T. Clemens, Proc.
Combust. Inst. 31 (2006), in press, doi:10.1016/
j.proci.2006.07.242.
[28] V. Bergmann, W. Meier, D. Wolff, W. Stricker, Appl.
Phys. B 66 (1998) 489502.
[29] W. Meier, R.S. Barlow, Y.-L. Chen, J.-Y. Chen, Com-
bust. Flame 123 (2000) 326343.
[30] C. Schneider, A. Dreizler, J. Janicka, E.P. Hassel, Com-
bust. Flame 135 (2003) 185190.
[31] M.W. Renfro, J.P. Gore, N.M. Laurendeau, Combust.
Flame 129 (2002) 120135.
[32] J. Hult, U. Meier, W. Meier, A. Harvey, C.F. Kaminski,
Proc. Combust. Inst. 30 (2005) 701709.
[33] International Workshop on Measurement and Com-
putation of Turbulent Nonpremixed Flames (TNF),
http://www.ca.sandia.gov/TNF/, accessed 10 Novem-
ber 2006.
[34] R.S. Barlow, P.C. Miles, Proc. Combust. Inst. 28 (2000)
269276.
[35] R.W. Bilger, S.H. Strner, R.J. Kee, Combust. Flame 80
(1990) 135149.
[36] R.S. Barlow, A.N. Karpetis, J.H. Frank, J.-Y. Chen,
Combust. Flame 127 (2001) 21022118.
G. Wang et al. / Combustion and Flame 148 (2007) 6275 75
[37] S.K. Lele, J. Comput. Phys. 103 (1992) 1642.
[38] S.B. Pope, Turbulent Flows, Cambridge Univ. Press,
Cambridge, UK, 2000.
[39] J. Panda, R.G. Seasholtz, J. Fluid Mech. 450 (2002) 97
130.
[40] M.W. Renfro, S.D. Pack, G.B. King, N.M. Laurendeau,
Appl. Phys. B 69 (1999) 137146.
[41] A.V. Oppenheim, R.W. Schafer, J.R. Buck, Discrete-
Time Signal Processing, second ed., Prentice Hall, Up-
per Saddle River, NJ, 1999.
[42] H. Tennekes, J.L. Lumley, A First Course in Turbu-
lence, MIT Press, Cambridge, MA, 1972.
[43] J.O. Hinze, Turbulence, second ed., McGrawHill,
New York, 1975.

You might also like