large-eddy simulation of correlation moments in turbulent combustion and validation of the rans-som...
TRANSCRIPT
Large-eddy simulation of correlation moments in turbulent combustion
and validation of the RANS-SOM combustion model
F. Wang *, L.X. Zhou, C.X. Xu
Department of Engineering Mechanics, School of Aerospace, Tsinghua University, Beijing 100084, China
Received 7 June 2005; received in revised form 8 November 2005; accepted 10 November 2005
Available online 15 December 2005
Abstract
The three-dimensional large-eddy simulation (LES) is carried out for a piloted methane–air jet flame (Flame C), measured in Sandia National
Laboratory, and its statistical results are validated by experimental data. The LES statistically-averaged time-averaged temperature, the root mean
square (RMS) value of temperature, and time-averaged methane and oxygen concentration are compared with those obtained using the Reynolds-
averaged Navier–Stokes (RANS) equations with second–order moment (SOM) combustion model. Also, the cross-correlations in the time-
averaged reaction rate expression of SOM model are given by the LES statistics. It is found that there is a similarity between the distribution of the
correlation moments and the distribution of the products of corresponding averaged variable’s gradients, the closure assumptions made in the
algebraic second–order moment (ASOM) combustion model for RANS model is approximately valid.
q 2005 Elsevier Ltd. All rights reserved.
Keywords: Turbulent combustion; Large-eddy simulation; Correlation moments
1. Introduction
The turbulent combustion modeling in engineering is a
challenging problem at the present time. It is found that the
eddy-break-up (EBU), EBU-Arrhenius and simplified prob-
ability density function (PDF) models widely adopted in
commercial software frequently lead to remarkable errors in
the predictions. The PDF transport equation model is more
reasonable, but has not yet been widely used, since it needs
large computation time for complex engineering flows. The
laminar flame-let and conditional moment closure models
are encouraging, but are not yet well validated by
experiments for various cases. Recently, a second-order
moment (SOM) combustion model was proposed by the
present authors1.
The unconditional second-order moment turbulence-chem-
istry models, based on the idea of second-order moment
turbulence models, exhibit both the reasonability and the
economy, and hence are considered as perspective models in
engineering application. The early-developed second-order
moment model2 gives the time-averaged reaction rate by
0016-2361/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.fuel.2005.11.006
* Corresponding author. Tel.: C86 1062782231; fax: C86 1062855007.
E-mail address: [email protected] (F. Wang).
making an approximation of E/RT$T 0/T/1 and omitting the
higher order terms in the series expansion of the non-linear
exponential term. However, in many practical combustion
processes, in particular NO formation, E/RT[1, and T 0/T is
not much smaller than unity, therefore the series expansion
approximation gives serious errors. In recent years, we did
systematic studies on developing second-order moment
turbulence-chemistry models. A version of second-order
moment-PDF (SOM-PDF) model3 and a unified second-order
moment (USM) model4 were proposed. In the SOM-PDF
model, the concentration–concentration correlation is closed
using the second-order moment method, while the tempera-
ture–concentration correlation is closed using the presumed
PDF. In the USM model all of correlation moments, including
the concentration–concentration correlation and the correlation
of the reaction rate coefficient fluctuation with the concen-
tration fluctuation, are closed using the conservation equations
in the same form. The USM model was used to simulate
methane–air jet diffusion combustion and NO formation
(Flame D) measured by the Sandia National Laboratory, and
methane–air swirling combustion and NO formation measured
in our laboratory.1,5 Simulation results and the comparison
with experiments show that the proposed model is more
reasonable than the EBU-Arrhenius (E-A) and simplified PDF
model.
However, the closure assumptions made in the RANS-
SOM model, that is, for the production term in the transport
Fuel 85 (2006) 1242–1247
www.fuelfirst.com
Nomenclature
Cs empirical model constant
B pre-exponential factor
d distance to the closest wall
E activation energy
g SGS mass flux
h enthalpy
k turbulent kinetic energy
K reaction rate coefficient
Ls mixing length for sub-grid scales
P pressure
Pr Prandtl number
q SGS heat flux
R gas-law constant
Sc Schmidt number
S strain-rate tensor
t time
T temperature
u velocity component
V volume of the computational cell
w reaction rate
x space coordinate
Y mass fraction
Greek symbols
3 dissipation rate
k Von Karman constant
m dynamic viscosity
mt sub-grid scale turbulent viscodity
r density
t sub-grid scale value
Subscripts
Fu, fu fuel
i, j, k component
Ox, ox oxygen
s s species
Superscripts
w filtered value
F. Wang et al. / Fuel 85 (2006) 1242–1247 1243
equation model, or simply in the algebraic model, the
correlations of the reaction-rate coefficient fluctuation
with the concentration fluctuation is proportional to the
product of corresponding time-averaged values, need to be
theoretically justified. The large-eddy simulation (LES) of
turbulent combustion attracts more and more attention in
recent years, since it needs less computation requirements
for simulating high Re flows and can give the turbulence
structures in combustion, hence can help us to understand
the mechanism of turbulence-chemistry interaction. How-
ever, up to now it is less used to validate RANS
combustion models.
There are various sub-grid scale (SGS) combustion
models in LES, such as the EBU model, laminar flame-let
model, the G-equation model, linear-eddy model, PDF
equation model and so on. As for turbulent reaction jet
flow simulation, various flame-let models were mostly used.
Pitsch6 applied the Lagrangian flame-let model in a
methane–air jet flame and achieved good predictions.
While, Mattsson et al.7 used the SGS laminar flame-let
model to simulate a propane–air jet flame, but the predicted
statistical results near the inlet region are not in agreement
with the laser holography measurement results. DesJardin
et al.8 did the similar LES and found that the simulation
over-predicts the combustion products. Liu et al.9 used a
dynamic similarity sub-grid-scale SGS combustion model to
simulate a jet diffusion flame; the predicted temperature
distribution is in agreement with that obtained using direct
numerical simulation (DNS), but its comparison with
experimental results is not reported. Meanwhile, a SGS-
SOM combustion model is proposed by Hu et al.10 and is
used to simulate swirling combustion; the statistical results
are in good agreement with experimental results.
In this paper, the SGS-SOM combustion model is
adopted for LES of a piloted jet flame measured in the
Sandia National Laboratory, USA.11 The LES statistical
results will be validated by experiments and the statistical
results of various auto-correlations and cross-correlations,
such as the auto-correlation of temperature fluctuation,
the cross-correlation of concentration fluctuation and the
cross-correlation of the fluctuation of the reaction-rate
coefficient with the concentration fluctuation, will be
compared with those obtained using the RANS model,
including the SOM model, for verifying and improving the
RANS-SOM model.
2. Mathematical model and numerical procedure
The filtered conservation equations for LES are:
Continuity equation
vr
vtC
v
vxi
ðr ~uiÞ Z 0 (1)
Momentum equation
v
vtðr ~uiÞC
v
vxj
ðr ~ui ~ujÞ Zv
vxj
mv ~ui
vxj
� �K
v ~P
vxi
Kvtij
vxj
(2)
where the sub-grid-scale stress tij hrguiujKr ~ui ~uj is closed by
the Smagorinsky–Lilly eddy-viscosity model tij ZK2mt~SijC
1⁄3tkkdij, mt ZrL2Sj~Sj, LSZminðkd;CSV1=3Þ, CsZ0.1.
Fig. 1. The jet combustor.
F. Wang et al. / Fuel 85 (2006) 1242–12471244
Species equation
vr ~Ys
vtC
v
vxj
ðr ~uj~YsÞ Z
v
vxj
m
SCs
v ~Ys
vxj
� �K ~wsKwS;sK
vgSj;s
vxj
(3)
where ~ws and wS,s are the filtered and sub-grid scale reaction
rates, respectively. The sub-grid scale mass flux is
gSj;s Zmt
Sct;s
v ~Ys
vXj
:
Energy equation
vr ~h
vtC
v
vxj
ðr ~uj~hÞ Z
v
vxj
m
Pr
v ~h
vxj
� �K
vqSj
vxj
(4)
where the sub-grid scale heat flux is qSj Zmt
Prt
v ~Tvxj
� �The SGS mass flux and heat flux are also closed by the
Smagorinsky–Lilly model.
The filtered reaction rate ~ws is ~KgYOXfYFu , where
~KZBr2Ð
expðKE=R ~TÞpð ~TÞd ~T .
For the SGS-SOM model, the sub-grid scale reaction rate
wS,s is
wS;Fu Z ~KðgYOXfYFu KgYOX
fYFu ÞC gYOXð gKYFu K gKYFu Þ
C fYFu ð gKYOX K ~KgYOXÞ (5)
Fig. 2. Time-averag
It expresses the effect of small-scale turbulence on the
reaction rate. The correlations are closed using the gradient
modeling. For example, the correlation of the fluctuation of the
reaction-rate coefficient with the concentration fluctuation is
given by
ð gKYOX K ~KgYOX Þ Z CK;YOXL2
S
v ~K
vxj
vgYOX
vxj
(6)
The RANS-SOM model is given in Ref. [1]. The laminar
chemical kinetics of methane–air combustion is given by
Ref. [12]
wfu Z 2:119!1011Y1:3ox Y0:2
fu expðK2:027!108=RTÞ (7)
For the numerical procedure in LES, the grid size is taken as
0.5 mm near the inlet and 2 mm in other regions; the time steps
are 1 ms for non-reacting flows and 0.1 ms for reacting flows.
The second-order difference scheme is adopted, and PISO
algorithm is adopted in numerical solution. A random
fluctuation of Gaussian distribution is superposed to the inlet
velocity. For 2D RANS model the grid sizes are 2–5 mm. For
3D LES running a case on the PC with a 3.0G Intel Xeon CPU
takes about 72 h for non-reacting flows and 480 h for reacting
flows. The 2D RANS-SOM model on the same PC takes about
0.5 h.
3. Results and discussion
Fig. 1 gives the geometrical configuration and sizes of the
piloted methane–air jet flame, measured in Sandia National
Laboratory, USA.11 The central jet consists of 25% methane
and 75% air in volume, and its inlet velocity is 29.7 m/s. The
annular jet is hydrogen–air combustion products with an inlet
temperature of 1880 K and inlet velocity of 6.8 m/s. The co-
flow air velocity is 0.9 m/s.
Fig. 2 shows the comparison of predicted time-averaged
temperature by LES-SOM and RANS-SOM model with the
experimental results. In most regions, the LES-SOM results
ed temperature.
Fig. 5. Time-averaged oxygen concentration.
Fig. 3. RMS values of temperature fluctuation.
Fig. 4. Time-averaged methane concentration.
F. Wang et al. / Fuel 85 (2006) 1242–1247 1245
Fig. 6. Correlation and product of gradients of time-averaged values.
F. Wang et al. / Fuel 85 (2006) 1242–12471246
and RANS-SOM results are in agreement with the
experimental results. Comparatively, the LES-SOM model
is better than the RANS-SOM model predictions, in
particular at the cross-sections of xZ7.5, 15 and 30, d.
The RANS-SOM model over-predicts the temperature at
these positions.
Fig. 3 gives the RMS value of temperature fluctuation. The
LES-SOM model is in good agreement with the experimental
results, and is better than the RANS-SOM. The RANS-SOM
model over-predicts the temperature fluctuation in most cross-
sections. However, in predicting the time-averaged tempera-
ture, which is more interested in engineering application, the
RANS-SOM model is preferred, considering that it can save
much more time than the LES.
Figs. 4 and 5 show the time-averaged methane and oxygen
concentration, respectively. In most regions, both LES-SOM
and RANS-SOM model give good results.
The correlation of reaction-rate coefficient fluctuation with
the methane concentration fluctuation and the product of
Fig. 7. Time-averaged temper
Fig. 8. Instantaneous temper
gradients of corresponding time-averaged values given by
LES, are shown in Fig. 6. It is found that although these
correlation moments are not exactly proportional to the
products of gradients of corresponding averaged variables,
but there is a similarity between the distribution of the
correlation moments and the distribution of the products of
corresponding averaged variable’s gradients, the general trends
of their distributions are similar to each other: their values go
up and down at the same regions and their peak values are
located at the same locations. Thus, the gradient modeling
assumptions made in the algebraic second-order moment
(ASOM) combustion model for RANS model is approximately
valid. Furthermore, the empirical constants in RANS-SOM
model can be determined using the LES results.
Figs. 7 and 8 are time-averaged and instantaneous
temperature maps obtained by RANS-SOM and LES-
SOM, respectively. The physical features of an outer
premixed flame and an inner diffusion flame can be seen
in Fig. 7. The instantaneous temperature map in Fig. 8 gives
the coherent structures of turbulence in the jet flame,
showing the wrinkled flame front.
4. Conclusions
(1) There is a similarity between the distribution of the
correlation moments and the distribution of the products
of corresponding averaged variable’s gradients, the
closure assumptions made in the algebraic second
order moment (ASOM) combustion model is approxi-
mately valid.
(2) For time-averaged temperature and concentration the
LES-SOM results are near to RANS-SOM results.
(3) For RMS values of turbulence properties, LES is
obvious better than RANS modeling.
(4) Considering acceptable accuracy in simulation of time-
averaged variables and much less computation time, the
ature map (RANS-SOM).
ature map (LES-SOM).
F. Wang et al. / Fuel 85 (2006) 1242–1247 1247
RANS-SOM model is preferred to be used in
engineering.
Acknowledgements
This study was sponsored by the 985 Project of Tsinghua
University.
References
[1] Zhou LX, Wang F, Zhang J. Simulation of swirling combustion and NO
formation using a USM turbulence-chemistry model. Fuel 2003;82:
1579–86.
[2] Zhou LX. Dynamics of multi-phase turbulent reacting flows. Beijing:
Defense Industry Press; 2002.
[3] Zhou LX, Chen XL, Zheng CG, Yin J. Second-order moment turbulence-
chemistry models for simulating NOx formation in gas combustion. Fuel
2000;79:1289–301.
[4] Zhou LX, Qiao L, Zhang J. A unified second-order moment turbulence-
chemistry model for simulating turbulent combustion and NOx formation.
Fuel 2002;81:1703–9.
[5] Zhou LX, Chen XL, Zhang J. Studies on the effect of swirl on NO
formation in methane–air turbulent combustion. Proc Combust Inst 2003;
29:561–8 [Part 1].
[6] Pitsch H, Steiner H. Large eddy simulation of a turbulent piloted
methane/air diffusion flame (Sandia flame D). Phys Fluids 2000;12:
2541–54.
[7] Mattsson R, Kupiainen M, Gren P, Wahlin A, Carlsson TE, Fureby C.
Pulsed TV holography and schlieren studies, and large eddy
simulations of a turbulent jet diffusion flame. Combust Flame 2004;
139:1–15.
[8] DesJardin PE, Frankel SH. Large eddy simulation of a non-premixed
reacting jet: application and assessment of sub-grid scale combustion
models. Phys Fluids 1998;10:2298–314.
[9] Liu Y, Lau KS, Chan CK, Guo YC, Lin WY. Structures of scalar transport
in 2D transitional jet diffusion flames by LES. Int J Heat Mass Transfer
2003;46:3841–51.
[10] Hu LY, Zhou LX, Zhang J. Comparison between LES and RANS
modeling of turbulent swirling flows and swirling diffusion combustion.
Chin J Chem Eng 2005;13(3):313–7.
[11] http://www.ca.sandia.gov/TNF/pilotedjet.html, 2003.
[12] Westbrook CK, Dryer FL. Simplified reaction mechanisms for the
oxidation of hydrocarbon fuels in flames. Combust Sci Technol 1981;27:
31–43.