Anda di halaman 1dari 9

A Critical Evaluation of

Turbulence Modeling
in a Model Combustor
Leiyong Jiang
Institute for Aerospace Research,
National Research Council Canada,
1200 Montreal Road, M-10,
Ottawa, ON K1A 0RG6, Canada
e-mail: leiyong.jiang@nrc-cnrc.gc.ca

Based on the previous benchmark studies on combustion, scalar transfer, and radiation
models, a critical evaluation of turbulence models in a propane-air diffusion flame combustor with interior and exterior conjugate heat transfers has been performed. Results
obtained from six turbulence models are presented and compared in detail with a comprehensive database obtained from a series of experimental measurements. It is found
that the Reynolds stress model (RSM), a second moment closure, is superior over the five
popular eddy-viscosity two-equation models. Although the main flow patterns are captured by all six turbulence models, only the RSM is able to successfully predict the
lengths of both recirculation zones and give fairly accurate predictions for mean velocity,
temperature, CO2 and CO mole fractions, as well as turbulence kinetic energy in the
combustor chamber. In addition, the realizable k-e (Rk-e) model illustrates better performance than four other two-equation models and can provide comparable results to
those from the RSM for the configuration and operating conditions considered in the
present study. [DOI: 10.1115/1.4023306]
Keywords: Reynolds stress turbulence model, eddy-viscosity turbulence models, combustor, flamelet combustion model

Introduction

Turbulence modeling is one major factor which affects the precision of current numerical simulations, particularly for reacting
flows. The random nature of turbulence involves a wide range of
time and length scales, and it is one of the principal unsolved
problems in physics today [1]. Despite the rapid development of
computing power, direct numerical simulations of turbulent flows
remain practical only at low Reynolds numbers, while large eddy
simulations are limited to benchmark cases with relatively simple
geometries [2]. This is particularly true for turbulent reacting
flows. Combustion, even without turbulence, is an intrinsically
complex process which can involve hundreds of species and thousands of element reactions and cause numerical difficulties [3].
For the above reasons, it is necessary to utilize turbulence models
in numerical simulations for the development of advanced practical combustion systems.
Much effort has been devoted to the development of turbulence
models in the last five decades. Progress has been reviewed by a
number of authors, and was brought up to date for reacting flows
by Jones [4]. Various algebraic, one- and two-equation turbulence
models were systematically evaluated by Wilcox [2] against a
number of well-documented nonreacting flows, including freeshear, boundary-layer, and separated flows. Some guidelines
regarding applications of these turbulence models were provided.
Six eddy-viscosity and two variants of Reynolds stress turbulence
models were critically assessed by Kim and Rhee [5]. In their
case, the flow field around a ship hull with strong cross-flows and
streamwise vortices was considered. It was found that the two
Reynolds stress models were able to reproduce all the salient
features of the flow field and the predicted Reynolds stresses and
turbulence kinetic energy were in good agreement with the experimental results. Turrell et al. [6] found that the RSM was able to
predict a vortex core in a high-swirl lean premixed gas turbine
combustor, which was qualitatively supported by the experimental
observations. In contrast, the standard k-e model failed to predict
this phenomenon.
Manuscript received February 11, 2012; final manuscript received September 11,
2012; published online June 24, 2013. Assoc. Editor: Srinath V. Ekkad.

There is a large number of publications on numerical simulations of practical combustion systems such as Refs. [69], and
tremendous contributions have been made for understanding
advantages and limitations of various turbulence models. However, systematic assessment and validation of turbulence models
in combustor flow fields against well-defined comprehensive
experimental results are rare.
To provide a benchmark database for evaluation and development of various physical models, a series of experiments has been
performed at the National Research Council of Canada. Measurements were made in a propane/air diffusion flame combustor
using advanced measurement techniques [10]. The combustor
geometry was relatively simple compared with practical combustion systems, but fundamentally similar and pertinent to the
modeling of other complex systems. A three-dimensional laser
Doppler anemometer (LDA) was used to measure three velocity
components in the downstream region of the combustion chamber.
Due to limited optical access, a two-dimensional LDA was used
to obtain axial and circumferential velocities in the upstream
region. Gas temperatures were acquired using an uncoated,
250-lm diameter, type S thermocouple mounted in a twin bore
ceramic tube. Gas species measurements were made with a
sampling probe connected to a Varian Model 3400 Gas Chromatograph. The major species measured included CO, CO2, C3H8,
and O2.
The present work is a continuation of the previous studies on
this combustor [1113], where different combustion models, radiation models, and the effect of Prandtl/Schmidt number were
assessed against the comprehensive experimental data. It was
found that the flamelet combustion model illustrated the best
performance among four combustion models, i.e., the eddydissipation, eddy-dissipation-finite-rate, probability density function, and laminar flamelet models [11]. Reference [12] reveals
that the turbulent Prandtl/Schmidt number has significant effects
on the predicted temperature and species fields in the combustor.
For accurate velocity and scalar field predictions, an optimized
Prandtl/Schmidt number of 0.5 is recommended for the present
configuration and flow conditions. As shown in Ref. [13], the
effect of radiation on the flow field is minor, particularly to the
velocity field. However, it has significant effect on the NO field.

Journal of Thermal Science and Engineering Applications


C 2013 by ASME
Copyright V

SEPTEMBER 2013, Vol. 5 / 031002-1

Downloaded From: http://thermalscienceapplication.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

Presented and discussed in this paper are the numerical results


obtained from six turbulence models, i.e., the standard k-e model,
realizable k-e (Rk-e) model, renormalization group k-e (RNG)
model, k-x model, shear stress transport (SST) model, and RSM.
Except for the RSM, these are popular turbulence models in engineering applications: the first three have been extensively used in
internal reacting and nonreacting flows such as [Refs. 69,14],
while the k-x and SST models have been widely used in nonreacting flows such as aircraft, turbo-machinery, etc. such as
Refs. [1517]. By qualitative and quantitative comparisons
between the numerical results and experimental database, the
advantages and shortcomings of these turbulence models are
revealed.

Numerical Simulations

Axi-symmetric steady incompressible and turbulent reacting


flows were considered in the present study, and a commercial
software package, Fluent, was used for all numerical simulations.
The computation domain, selected physical models, boundary
conditions, and solution methods are described in the following
sub-sections.
2.1 Computational Domain. A schematic diagram of the
model combustor is shown in Fig. 1, including the fuel and air
inlets, disk flame-holder, combustion chamber, steel/insulation
walls, and contracted exhaust section (all dimensions are in mm).
Air entered the combustion chamber around a bluff body, while
fuel was fed through the center of the disk flame-holder.
The computational domain and mesh are shown in Fig. 2, where
for clarity only a small portion of nodes are displayed. Since the
flow field was axisymmetric, 2D quadrilateral meshes were generated over the flow region (from the fuel/air inlets to the exhaust
exit), the interior conjugate heat-transfer region (the flame holder
body within the inlet section) and the exterior conjugate heattransfer regions (the combustion chamber and insulation walls).
Fine grids were laid in the combustion chamber in order to properly resolve the flow recirculation and reacting regions. Fine grids
were also generated in the shear layers between the recirculation
region and the fuel and air jets. Coarse grids were used in the
exhaust section and solid wall regions. A total number of 74,100
elements were used for all the simulations. The skewness was less
than 0.2 in the flow-field domain and the aspect ratio was less
than 12 for 99.5% elements. Effort was made to keep the nondimensional wall-distances, y, to the desired value of 30. A number of meshes were tested to ensure the mesh independence of the
numerical solutions.

Fig. 2

Computational domain and mesh

2.2 Turbulence Models. The governing Reynolds-averaged


conservation equations for mass, species, momentum and energy
are not reproduced here. They can be readily found in the classic
literature, such as Ref. [18].
For closure of the governing equations, Reynolds stresses,
q u00i u00j , have to be modeled. Detailed description and formation of
each turbulence model are beyond the scope of the present paper.
However, the main concepts of these models are outlined below.
For the five eddy-viscosity two-equation turbulence models, the
Boussinesq hypothesis is adopted to model Reynolds stresses




@Ui @Uj
2
@Uk
 dij lt

qk
(1)
 q u00i u00j lt
3
@xj
@xi
@xk
With this approach, the turbulence viscosity of the k-e, Rk-e,
and RNG models, lt, for high Reynolds number flows is given by
the expression
lt qCl k2 =e

(2)

where Cl is a constant, and the values of the turbulence kinetic


energy, k, and dissipation rate, e, are calculated from a pair of
transport differential equations. The detailed descriptions of these
k-e models are given in Refs. [1921], respectively.
The turbulence viscosity of the k-x and SST models for high
Reynolds number flows is obtained by the following two equations, respectively
lt qk=x

(3)

qk
x

(4)

and
lt

Fig. 1

The model combustor

031002-2 / Vol. 5, SEPTEMBER 2013

1


SF
max 1;
ax

where the values of the turbulence kinetic energy and the specific
dissipation rate, x, are also determined from a pair of transport
Transactions of the ASME

Downloaded From: http://thermalscienceapplication.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

differential equations [2,22,23], F is equal to one in the near-wall


region and zero in the free shear layer, S is the strain rate magnitude, and a is a constant.
The RSM is a second-moment closure. It is more general
than the above eddy-viscosity turbulence models, where the linear
relations between Reynolds stresses and mean rate of strain
and the isotropic eddy viscosity are assumed, as indicated in
Eqs. (1)(4). For the RSM, all Reynolds stress components in the
flow field are directly computed from their corresponding transport differential equations. In the present axisymmetric case, four
components of Reynolds stresses (q u00i u00j ) are resolved in the combustor flow field. In principle, this model is likely to have a wider
range of applicability than the eddy-viscosity concept models. To
convert Reynolds stress equations into a closed set of equations,
unknown terms must be modeled by mean flow variables and/or
Reynolds stresses. Details can be found in Refs. [24,25].
2.3 Other Physical Models. Other physical models were
selected based on the previous benchmark studies [1113]. For
combustion modeling, the laminar flamelet model [26] was used
to simulate chemical reactions in the flow. A major advantage of
the flamelet model over other combustion models, such as the
probability density function and eddy-dissipation, is that detailed
more realistic chemical kinetics can be incorporated into turbulent
reacting flows. In the present case, the chemical reaction mechanism for propane-air flames used by Stahl and Warnatz [27] was
employed. It consists of 228 chemical reactions and 31 species,
including O, O2, OH, H, H2, H2O, H2O2, HO2, N2, CO, CO2, CH,
CH2, CH3, CH4, CHO, CH2O, CH2CO, CH3CO, CH3CHO, C2H,
C2H2, C2H3, C2H4, C2H5, C2H6, C3H6, C3H8, N*C3H7, I*C3H7,
and C2HO.
To account for the radiation heat transfer between the gas mixture and flame-holder body and combustion chamber walls, the
discrete ordinates radiation model [28] was employed. The
absorption coefficient of gaseous mixture was determined from
the local species mass fractions in the mixture. For the turbulent
scalar transfer modeling [12], the optimized turbulent Prandtl/
Schmidt number of 0.5 was chosen for all numerical simulations.
Although the effort was made to keep y values around 30,
there were some local regions where y was above or below this
value. To account for this, an enhanced wall boundary treatment
was applied to all wall boundaries for the three k-e models and
RSM. In this approach, a two-layer model [29] is enhanced by
smoothly blending the linear (laminar) and logarithmic (turbulent)
laws-of-the-wall [30]. Consequently, it can apply to the entire
near-wall region [23].
Polynomials derived from the JANAF tables [31] were used to
calculate the specific heats of species as a function of temperature.
For other thermal properties of the mixture such as molecular
viscosity and thermal conductivity, the values for air at 900 K
were used.
2.4 Boundary Conditions. The fuel (propane) mass flow
rate was 16.2 g/s, the airflow rate was 550 g/s, and the corresponding overall equivalence ratio was 0.46. The Reynolds number
based on the air entry velocity and flame-holder diameter was
1.9  105. An estimated turbulence intensity of 10% and hydraulic
diameters of the fuel/air inlets (8.4 mm and 36.3 mm, respectively)
were used to estimate Reynolds stress components and turbulence
dissipation rate at the fuel and air inlets. For both air and fuel
flows, the inlet temperature was 293 K.
The exterior wall temperatures were defined based on the experimental measurements and observation, as shown in Fig. 3. A
room temperature of 293 K was assigned to the walls of the inlet
section, and the upstream edges of the combustion chamber and
insulation walls. A linear temperature profile from 293 K to 405 K
was specified along the outer boundary of the insulation wall,
which was a good approximation to the experimental measurements. The temperature of the exterior boundary of the exit
Journal of Thermal Science and Engineering Applications

Fig. 3 Exterior wall temperature profiles of the computational


domain

section was set to 960 K, as an estimation of the experimental


observation. Based on the thermal resistance and preliminary
results, a linear temperature profile from 960 K to 780 K was
assigned to the downstream edge of the combustion chamber wall,
while another linear profile from 780 K to 405 K was defined at
the downstream edge of the insulation wall. Finally, the pressure
at the combustor exit was set to the atmospheric value.
2.5 Solution Methods. A segregated (pressure-based) solver
with a second-order accurate scheme was used to resolve the flow
field. At convergence, the normalized residuals of flow variables
were about or less than 105 in all test cases. The monitored axial
velocities at two points in the shear layer downstream of the flame
holder remained unchanged at least for the first four digits. This
ensured that the flow field reached steady conditions. A 4-node
LINUX cluster providing 64-bit, 2.6 GHz, 8-Core, and 32 GB RAM
per node, was used to carry out all simulations.

Results and Discussion

The predicted distributions of velocity, temperature, and species inside the combustor chamber with combustion are presented
in the following sub-sections. By comparing the numerical results
with the experimental database, the advantages and short-comings
of the five combustion models are revealed.
3.1 Velocity Distributions. The upper halves of six plots in
Fig. 4 show the numerical results of axial velocity contours and
flow path-lines for six turbulence models, respectively, while the
lower halves are the experimental data with the zero axial velocity
SEPTEMBER 2013, Vol. 5 / 031002-3

Downloaded From: http://thermalscienceapplication.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

lines specified. For the experimental plots, due to the limited number of data points no flow path-lines are drawn. The main flow
features or patterns in the combustion chamber are captured by all
models. That is, there are two recirculation zones or vortices
behind the flame-holder. The central recirculation zone (CRZ)

Fig. 4

Axial velocity contours and flow path-lines

031002-4 / Vol. 5, SEPTEMBER 2013

induced by the fuel jet is completely confined within the annular


recirculation zone (ARZ) created by the annular air jet. This
means that the transportation of fuel into the main flow field is
realized by laminar and turbulent diffusion only through the ARZ.
The zero axial velocity lines divide the recirculation zones into
two parts: one with flow moving upstream and the other moving
downstream. As expected, another separated flow zone is
observed at the upper left corner of the combustion chamber.
In terms of predicting the reattachment points or lengths of the
ARZ and CRZ, various degrees of agreement with the experimental data are observed among the six turbulence models. For the
standard k-e model as shown in the first plot of Fig. 4, both ARZ
and CRZ lengths are significantly under-estimated. This deficiency is also observed by other investigators for nonreacting
flows [32,33]. The performance of the Rk-e model is superior to
the standard k-e model, where the ARZ length is correctly predicted although the CRZ length is under-predicted. In the case of
the RNG model, a slight under-prediction of the ARZ length and a
considerable under-prediction of the CRZ length are observed. It
is a little surprising that the performance of the two k-x models
is not as good as the Rk-e and RNG models. For the k-x model,
the ARZ length is under-estimated, while the CRZ length is
over-estimated. For the SST case, the ARZ length is significantly
over-estimated although the CRZ length is captured. As shown in
the last plot of Fig. 3, the RSM model demonstrates the best
performance, where both ARZ and CRZ lengths are properly
predicted.
Based on the results of the RSM, the ratio of the circulated
mass flow rate in the ARZ versus the airflow input is 5.5%, and
the ratio of the ARZ length and the flame-holder diameter is 1.7.
These parameters could be useful to experimentalists in combustion stability and emissions.
Figure 5 presents the axial velocity profiles along the combustor
centerline for the six turbulence models. The numerical results
are compared with the experimental measurements where the
negative axial velocity reaches its peak value (about 10 m/s) at
x  80 mm. Superimposed in the figure are error bars representing
an estimated measurement error of 4%. Large deviations are
observed in the upper stream region (1080 mm) for the k-e and
k-x models, and in the downstream region (80360 mm) for the ke and SST models. The centerline axial velocities are reasonably
well estimated by the RSM, Rk-e, and RNG models. Overall, the
RSM illustrates the best performance among the six turbulence
models. This is consistent with the fact that only the RSM can
adequately predict both recirculation zones. Among the five twoequation models, the Rk-e results show better agreement with the
experimental data than the other models.
Displayed in Fig. 6 are the axial velocity profiles at seven
cross-sections from x 20 to 240 mm, including three across the
recirculation zones, one near the annular stagnation point, and

Fig. 5

Axial velocities along the combustor centerline

Transactions of the ASME

Downloaded From: http://thermalscienceapplication.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

Fig. 7 Turbulence
x 5 60240 mm

kinetic

energy

profiles

at

sections,

reasonably well predicted in the central region, but over-predicted


away from the combustor centerline. This is consistent with the
fact that the SST model is able to predict the CRZ length, but fails
to predict the ARZ length as shown in Fig. 4. The cross-section of
x 60 mm cuts through both recirculation zones, and the two
peaks are where the centers of the two recirculation zones are
located. Unfortunately, none of the models captures the central
peak.
In short, in terms of velocity flow-field prediction, the RSM
is superior over the five two-equation models, and in general the
Rk-e model illustrates better performance than the other four
two-equation models.

Fig. 6

Axial velocity profiles at cross-sections, x 5 20240 mm

three further downstream. The predicted trends and magnitudes


are close to the measurement data, except for the k-e model at
x 40200 mm and the SST model at downstream sections,
x  160 mm, where significant deviations are observed. Since the
SST model properly predicts the central recirculation zone as
shown in Fig. 4, it illustrates best performance at the three
upstream sections. The results from the Rk-e and RNG models are
comparable with those from the RSM. The RSM shows improvement over the Rk-e model in the regions of x 120200 mm and
R 3040 mm.
Figure 7 shows quantitative comparisons between the numerical and experimental results for the turbulence kinetic energy at
four cross-sections from x 60 to 240 mm. Error bars in the figure
represent 8% of measurement accuracy. It is noted that only the
RSM gives encouraging results at all sections. The RNG model
substantially over-predicts the turbulence kinetic energy at all
sections, and this is also true for the k-e model at the upstream sections. The Rk-e and k-x models show reasonable agreement with
the experimental data, except for the k-x model at section
x 100 mm. For the SST model, the turbulence kinetic energy is
Journal of Thermal Science and Engineering Applications

3.2 Temperature Distributions. The temperature contour


results of the six turbulence models are shown in the upper halves
of Fig. 8. Superimposed is the stoichiometric line of the mean
mixture fraction (f~ 0:064), which passes through the middle of
the high-temperature region in the combustion chamber for all
cases. The flame starts at the edge of the flame-holder disk and
spreads downstream around the annular recirculation zone, where
the mixture of recirculated hot gases and fuel mixes with the fresh
air and burns. This agrees with the experimental observation that
a carbon deposit forms at the middle of the disk edge. In comparison with the experimental data in the lower halves, it is found
that for the RSM and Rk-e models, the size and location of the
high-temperature or flame region are in good agreement with the
experimental data, and the RSM results are slightly better than
those from the Rk-e model. For the k-e and RNG models, the
high-temperature region is under-predicted and shifted upstream.
In contrast, the high-temperature region is significantly overpredicted and shifted downstream by the k-x and SST models.
Note that the optimized turbulent Prandtl/Schmidt number of 0.5
is used in the simulations as stated earlier. If a default value of
0.85 is applied, the predicted high-temperature region will be
even greater. This is because the turbulent Prandtl/Schmidt number represents the ratio of the momentum eddy diffusivity to the
scalar eddy diffusivity (heat and mass), and the greater the
Prandtl/Schmidt number, the slower the turbulent scalar transportation in comparison with the momentum transfer [12].
The predicted temperature profiles along the combustor centerline are quantitatively compared with the experimental results in
SEPTEMBER 2013, Vol. 5 / 031002-5

Downloaded From: http://thermalscienceapplication.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

Fig. 9

Temperature profiles along the combustor centerline

flat. In general, the RSM and Rk-e models show better agreement
than other four models. For the k-e and RNG models, the temperature is over-estimated in the upstream region and under-estimated
in the downstream region. In contrast to this, it is under-predicted
in the upstream and significantly over-predicted in the downstream for the k-x and SST models.
In Figs. 8 and 9, it is found that the predicted temperatures are
higher than the measured values in the center region from x 140
to 250 mm. The maximum difference is about 150 K. Three possible reasons are expected. First, the temperature was measured by
a 0.25 mm diameter thermocouple, as mentioned earlier. Owing to
the radiation and conduction losses from the thermocouple, the
measurement error could exceed 100 K over regions where the
gas temperature was high and the flow velocity was low [34].
Second, the intrusion of the temperature probe could alter local
flow structure, thereby enhancing local turbulent mixing, and
resulting in an increase in local temperature [34]. The third possible reason is that the turbulence kinetic energy (Fig. 7) is not
accurately predicted. As a result, the combustion process and temperature prediction could be affected.
Figure 10 gives the temperature profiles at seven cross-sections
from x 52 to 353 mm for six turbulence models. The numerical
results agree reasonably well with the experimental results for the
RSM and Rk-e models, except for the most upstream section and
the region near the combustor wall. It is also true for the RNG
model at sections x 82293 mm. Poor agreement is observed for
the k-e model at upstream sections, x 52112 mm, the k-x
model at sections, x 82, 232353 mm, and the SST model at
most sections.
In brief, similar to the trends for velocity prediction, the temperature distributions obtained from the RSM and Rk-e models
are comparable and consistent with the experimental data in
general.

Fig. 8 Temperature contours and flow path-lines

Fig. 9, where the measurement error is about 5%. Along the combustor centerline (50350 mm), the predicted trends are close to
the experimental values although the numerical results show
peaks in the middle portion, while the measurements tend to be
031002-6 / Vol. 5, SEPTEMBER 2013

3.3 Species Distributions. As mentioned earlier, the detailed


chemical mechanism with 31 spices is employed in the present
study, where the intermediate species, reactions and dissociation
effect are considered. Therefore, the results are more relevant and
provide insight into the flow field than those from a simplified
one-equation combustion model. Figure 11 presents the CO2 mole
fraction profiles at five cross-sections, x 21171 mm, for six turbulence models, and compares them with the experimental data.
For species measurements, the estimated error is about 5%, as
indicated by error bars in the figure. To show the main features of
chemical reactions, the locations of the five cross-sections were
selected as: two across the central recirculation zone, two passing
through the annular recirculation zone, and the last one behind the
recirculation region (see Fig. 4).
Transactions of the ASME

Downloaded From: http://thermalscienceapplication.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

Fig. 11 CO2 profiles at cross-sections, x 5 21291 mm

Fig. 10

Temperature profiles at cross-sections, x 5 52353 mm

Carbon dioxide is one of the final major products of propane-air


combustion. As shown in Fig. 11, except for the central portion at
x 111 mm and the middle portion at x 81 mm, the Rk-e and
RSM results agree fairly well with the measured data. Poor performance is observed for the k-e model at x 51 mm, k-x model
at x 81 and 171 mm, and SST model at x 171 mm.
The radial profiles of CO mole fraction, one major immediate
species in hydrocarbon fuel combustion, are shown in Fig. 12, and
are quantitatively compared with the experimental results. Like
the CO2 case, the CO profile at the most upstream section is properly predicted by all models, except for the SST model which had
a small bump at r 38 mm. For the RSM and Rk-e models, the
results agree fairly with the experimental data at x 51 and
171 mm. At the other two sections, x 81 and 111 mm, it is interesting to notice that the CO mole fraction is over-estimated in the
central region, while the CO2 is under-estimated by these two
models. However, the sum of CO2 and CO of the numerical
results is close to the sum of the measured CO2 and CO. This
implies that the predicted oxidization of CO at these two sections
is slower than the reality. Again poor performance is observed for
Journal of Thermal Science and Engineering Applications

Fig. 12 CO profiles at cross-sections, x 5 21291 mm

SEPTEMBER 2013, Vol. 5 / 031002-7

Downloaded From: http://thermalscienceapplication.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

the k-e model at x 51 mm, k-x model at x 171 mm, and SST
model at x 111171 mm.
In general, the species predictions are encouraging. Except for
some local regions, the Rk-e and RSM results are consistent with
the measured data, and show better overall performance than the
other models.
From the above qualitative and quantitative comparisons of
velocity, temperature and species distributions inside the combustor between the numerical and experimental results, it is clear that
the second-moment closure model, RSM, is superior over the
eddy-viscosity models. This is consistent with the observations by
other authors, such as Kim and Rhee [5] for a nonreacting flow
and Turrell et al. [6] for a reacting flow as mentioned earlier.
Among the five two-equation models, the Rk-e model displays
better performance than others. Application of the Rk-e model to
numerical simulations of practical gas turbine combustors, instead
of the RSM, may eliminate some difficulties, such as timeconsuming and numerical stability issues.
The SST model has been successfully applied to many nonreacting flows, such as adverse pressure gradient, backwardfacing step, and NACA 4412 airfoil flows [22]. This may explain
why it is able to properly predict the central recirculation zone as
shown in Figs. 4, 6, and 7, since the temperature in this region is
low as displayed in Fig. 8. However, it significantly over-predicts
the annular recirculation zone and high-temperature region in the
combustor. The fact that the predicted flame region from the SST
model is significantly larger than that obtained from the Rk-e
model is also observed for the simulations of a practical gas
turbine combustor [35]. The reasons may be two fold. First, the
validation cases for the SST model [2,22] are isothermal or almost
isothermal flows and the effects of combustion and significant
thermal expansion may not be properly modeled. Second, the flow
with multirecirculation zones is a typical phenomenon in combustion systems, which is more complicated than the flow field with a
single backward-facing step. In particular, for the present case, the
central recirculation zone is completed confined by the annular
recirculation zone.

Conclusion

A propane-air diffusion flame combustor with the interior and


exterior conjugate heat transfers was numerically investigated
with six turbulence models, including the standard k-e, the realizable k-e, the renormalization group k-e, the standard k-x, the SST,
and the Reynolds stress models. Also used were the laminar
flamelet combustion model and the discrete ordinates radiation
model.
Although the flow features or patterns are captured by all turbulence models, in terms of quantitatively predicting the velocity,
temperature and species fields, various degrees of agreement
with the experimental data are observed among the six turbulence
models. The RSM model illustrates the best performance, and it is
the only one that can correctly predict the lengths of both recirculation zones and give reasonable prediction on the turbulence
kinetic energy distribution in the combustor. Although the RSM
model uses more computing time than other RANS models, this is
not a hurdle for current affordable parallel computing clusters.
The present study also indicates that the performance of the Rke model is better than other four two-equation models, and can
provide similar predictions as those from the RSM for the present
configuration and operating conditions considered. Further studies
on the k-x and SST models are planned to check the above findings, including simulations with y value 1 and at isothermal
conditions.

Acknowledgment
The author is grateful to Dr. Ian Campbell for his permission to
use the valuable experimental database for this benchmarking
work.
031002-8 / Vol. 5, SEPTEMBER 2013

Nomenclature
a
Cl
f or f~
k
r
S
T
U
Ui
ui00
xi
x

constant
constant
mean mixture fraction
turbulence kinetic energy, (J or m2/s2 per unit mass)
radial coordinate, (mm or m)
strain rate magnitude, (/s)
temperature, (K)
mean axial velocity, (m/s)
ith mean velocity component, (m/s)
ith fluctuating velocity component, (m/s)
ith Cartesian coordinate, (mm or m)
coordinate along the combustor axis of symmetry,
(mm or m)
y distance to the wall boundary, (mm or m)
y nondimensional
q distance of the first node away from a
wall, sw =qf y=t

sw

e
x
q
q u00i u00j
lt
dij

wall shear stress, (kg/s2/m)


kinematic viscosity, (m2/s)
turbulence dissipation rate per unit mass, (m2/s3)
specific turbulence dissipation rate per unit mass, (/s)
density, (kg/m3)
Reynolds stresses, (kg/m/s2)
turbulence viscosity, (N s/m2)
Kronecker delta function (dij 1 if i j; dij 0 if i = j)

References
[1] Tennekes, H., and Lumley, J. L., 1983, A First Course in Turbulence, MIT,
Cambridge, Mass.
[2] Wilcox, D. C., 2001, Turbulence Modeling: An Overview, AIAA Paper No.
2001-0724.
[3] Dagaut, P., Reuillon, M., Boettner, J. C., and Cathonnet, M., 1994, Kerosene
Combustion at Pressures up to 40 Atm: Experimental Study and Detailed
Chemical Kinetic Modeling, 25th Symposium (International) on Combustion,
pp. 919926.
[4] Jones, W. P., 1994, Turbulence Modeling and Numerical Solution Methods for
Variable Density and Combustion Flows, Turbulent Reacting Flows, P. A.
Libby and F. A. Williams, eds., Academic, New York.
[5] Kim, S. E., and Rhee, S. H., 2002, Assessment of Eight Turbulence Models
for a Three-Dimensional Boundary Layer Involving Crossflow and Streamwise
Vortices, AIAA Paper No. 2002-0852.
[6] Turrell, M. D., Stopfod, P. J., Syed, K. J., and Buchanan, E., 2004, CFD Simulation of the Flow Within and Downstream of High-Swirl Lean Premixed Gas
Turbine Combustor, ASME Paper No. GT-2004-53112.
[7] Mongia, H. C., 1998, Aero-Thermal Design and Analysis of Gas Turbine
Combustion Systems: Current Status and Future Direction, AIAA Paper No.
98-3982.
[8] Alkabie, H., 2000, Design Methods of the ABB Alstom Power Gas Turbine
Dry Low Emission Combustion System, Proc. Inst. Mech. Eng., Part A,
214(4), pp. 293315.
[9] Cadorin, M., Pinelli, M., Vaccari, A., Calabria, R., Chiariello, F., Massoli, P.,
and Bianchi, E., 2011, Analysis of a Micro Gas Turbine Fed by Natural Gas
and Synthesis Gas: Test Bench and Combustor CFD Analysis, ASME Paper
No. GT-2011-46090.
[10] Campbell, I., and Logan, D. L., 1997, An Experimental Study of a Combusting
Flow Past a Confined Bluff Body, Combustion InstituteCanadian Section,
Halifax, Canada.
[11] Jiang, L. Y., and Campbell, I., 2007, Combustion Modeling in a Model
Combustor, J. Aerosp. Power, 22(5), pp. 694703.
[12] Jiang, L. Y., and Campbell, I., 2008, Reynolds Analogy in Combustor Modeling, Int. J. Heat Mass Transfer, 51(56), pp. 12511263.
[13] Jiang, L. Y., and Campbell, I., 2009, Radiation Benchmarking in a Model
Combustor, J. Eng. Gas Turbines Power, 131, p. 011501.
[14] Hu, T. C. J., Sze, R. M. L., and Sampath, P., 1998, Design and Development
of Advanced Combustion System for PW150 Turboprop Engine, Proceedings
of the CASI 45th Annual Conference, Calgary, Alberta, Canada.
[15] Rudnik, R., Melber, S., Ronzheimer, A., and Brodersen, O., 2001, ThreeDimensional NavierStokes Simulations for Transport Aircraft High-Lift Configurations, J. Aircr., 38(5), pp. 895903.
[16] El-Gabry, L. A., 2011, Comparison of Steady and Unsteady RANS Heat
Transfer Simulations of Hub and Endwall of a Turbine Blade Passage, J. Turbomach., 133(3), p. 031010.
[17] Sadiq, S., 2011, Cavity Acoustics AnalysisAn Extensive Comparison of
Turbulence Model Coefficients, 2011 Workshop on Scientific Use of Submarine Cables and Related Technologies.
[18] Poinsot, T., and Veynante, D., 2001, Theoretical and Numerical Combustion,
R. T. Edwards, Inc., PA.

Transactions of the ASME

Downloaded From: http://thermalscienceapplication.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

[19] Launder, B. E., and Spalding, D. B., 1972, Lectures in Mathematical Models of
Turbulence, Academic, London, UK.
[20] Shih, T. H., Liou, W. W., Shabbir, A., Yang, Z., and Zhu, J., 1995, A
New k-e Eddy-Viscosity Model for High Reynolds Number Turbulent
FlowsModel Development and Validation, Comput. Fluids, 24(3), pp.
227238.
[21] Yakhot, V., and Orszag, S. A., 1986, Renormalization Group Analysis of Turbulence: I. Basic Theory, J. Sci. Comput., 1(1), pp. 151.
[22] Menter, F. R., 1994, Two-Equation Eddy-Viscosity Turbulence Models for
Engineering Applications, AIAA J., 32(8), pp. 15981605.
[23] ANSYS Inc., 2010, Fluent 13.0 Documentation, 10 Cavendish Court, Lebanon,
NH 03766, USA.
[24] Launder, B. E., Reece, G. J., and Rodi, W., 1975, Progress in the
Development of a Reynolds-Stress Turbulence Closure, J. Fluid Mech., 68, pp.
537566.
[25] Moore, J. G., and Moore, J., 2006, Functional Reynolds Stress Modeling, Pocahontas Press, Inc., Blacksburg, VA.
[26] Peters, N., 1984, Laminar Diffusion Flamelet Models in Non-Premixed Turbulent Combustion, Prog. Energy Combust. Sci., 10, pp. 319339.
[27] Stahl, G., and Warnatz, J., 1991, Numerical Investigation of Time-Dependent
Properties and Extinction of Strained Methane- and Propane-Air Flamelets,
Combust. Flame, 85, pp. 285299.

Journal of Thermal Science and Engineering Applications

[28] Raithby, G. D., and Chui, E. H., 1990, A Finite-Volume Method for Predicting
a Radiant Heat Transfer in Enclosures With Participating Media, J. Heat
Transfer, 112, pp. 415423.
[29] Jongen, T., 1992, Simulation and Modeling of Turbulent Incompressible
Flows, Ph.D. thesis, EPF Lausanne, Lausanne, Switzerland.
[30] Kader, B., 1993, Temperature and Concentration Profiles in Fully Turbulent
Boundary Layers, Int. J. Heat Mass Transfer, 24(9), pp. 15411544.
[31] NIST, 1998, NIST-JANAF Thermochemical Tables, 4th ed., the National Institute for Standards and Technology, Washington, DC.
[32] Widmann, J. F., Charagundla, S. R., and Presser, C., 2000, Aerodynamic
Study of a Vane-Cascade Swirl Generator, Chem. Eng. Sci., 55, pp.
53115320.
[33] Xu, D., and Khoo, B. C., 1998, Numerical Simulation of Turbulent Flow in an
Axisymmetric Diffuser With a Curved Surface Center-Body, Int. J. Numer.
Meth. Heat Fluid Flow, 8(2), pp. 245255.
[34] Sislian, J. P., Jiang, L. Y., and Cusworth R. A., 1988, Laser Doppler Velocimetry Investigation of the Turbulence Structure of Axisymmetric Diffusion
Flames, Prog. Energy Combust. Sci., 14(2), pp. 99146.
[35] Jiang, L. Y., and Corber, A., 2011, Benchmark Modeling of T56 Gas Turbine
CombustorPhase I, CFD Model, Flow Features, Air Distribution and Combustor Can Temperature Distribution, IAR-LTR-GTL-2010-0088, the National
Research Council of Canada.

SEPTEMBER 2013, Vol. 5 / 031002-9

Downloaded From: http://thermalscienceapplication.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

Anda mungkin juga menyukai