Professional Documents
Culture Documents
a r t i c l e
i n f o
Article history:
Received 27 March 2012
Received in revised form 31 January 2013
Accepted 12 February 2013
Available online 7 March 2013
Keywords:
S-shaped diffuser
Aircraft inlet duct
a b s t r a c t
The S-shaped diffuser which connects the exit of the compressor to the inlet of the combustion chamber
of the Allison 250 gas turbine has been investigated using the Shear-Stress Transport turbulence model
(SST) and the commercial code ANSYS-CFX. The diffuser geometry includes an initial conical diffuser
which smoothly transitions into a constant cross-section S-duct. The numerical model and setup were
validated using both in-house processed experimental data and experimental data from the literature
on a similar geometry. The stream-wise velocity prole was observed to atten in the initial divergent
section, and then the region of the ow with the highest velocity is pushed toward the outer surface
of the rst bend, with a secondary-ow in the plane of the cross-section. This distortion of the streamwise velocity intensied when the inlet turbulence intensity was decreased or when the Reynolds number was increased. An increase of the Reynolds number also translated into higher static pressure recovery potential and lower wall friction coefcients. Six variations of the diffuser geometry were considered,
all having the same total cross-sectional area ratio and centreline offset. The qualitative results were the
same as those of the Allison 250 diffuser, but unlike the base geometry, all the considered variants
showed separated-ow regions (and reversed-ow regions in some cases) of different sizes and at different locations. The performance indicators for the Allison 250 S-shaped diffuser were the highest overall.
Most interestingly, the current duct geometry outperformed its variant with a cross-sectional area expansion extending over its entire length, which is the most common inlet duct conguration.
2013 Elsevier Inc. All rights reserved.
1. Introduction
Turbulent ows in passages with gradually varying cross-sectional area and bends are present in a large variety of applications and have long been the subject of interest from the
scientic community. 2D and 3D planar diffusers and curved
rectangular ducts have thus been investigated using both the
experimental and numerical approaches, evidencing occurrences
of separated-ow regions, the extents of which depend on the
passage area ratio and the severity of the duct curvature when
applicable, and the Reynolds number (Gullman-Strand et al.,
2004; Cherry et al., 2008; Schneider et al., 2010; Jakirlic et al.,
2010; Gopaliya et al., 2011). As far as axisymmetric geometries
are concerned, Azad (1996) authored a review of three-decades
of experimental research on an 8 conical diffuser at the University of Manitoba. According to the author, diffusers can be
viewed as devices that convert kinetic energy into potential energy, and the efciency of such conversion is the highest for
divergence angles between 6 and 8 for a conical diffuser, a
divergence angle of 11 for a rectangular 2D diffuser, and a
Corresponding author.
E-mail address: Goni.Boulama@rmc.ca (K. Goni Boulama).
0142-727X/$ - see front matter 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatuidow.2013.02.004
152
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
Nomenclature
a, A, B, C
Cf
CL
Cp
D
KE
NUI
Ps, PT
Re
r, z, x
R, Z, X
Tu
u, v, w
U
us
Y+
yp
sw
l
q
m
for the ow uniformity (Gopaliya et al., 2007). The latter study also
predicted that an increase in the Reynolds number does not have a
signicant effect on the pressure recovery. Abdellatif et al. (2008)
adopted the numerical approach in their investigation of the effect
of the area ratio on the performance of S-shaped diffusers. Considering an S-shaped diffuser of a constant turning angle and centreline
length, they changed the area ratio from 1 to 1.51, which resulted
in a static pressure recovery variation from 0.19 to 0.40. A further
increase of the area ratio to 1.9 did not signicantly improve the
pressure recovery, but the ow distortion increased. The authors
therefore concluded that for the given constraints, the area ratio of
1.51 constitutes an optimum. Whitelaw and Yu (1993) experimentally studied an S-shaped diffuser tube: the Royal Aerospace Establishment (RAE) series 2129, which has an area ratio of 1.4 and a nondimensionalized offset of 0.3. They used a Reynolds number of
400,000 and two different thicknesses of the boundary layer at the
inlet of the diffuser. In both cases, a region of separation was observed at the inside surface of the rst bend, just upstream of the
junction between the rst and second bends, and it was larger for
the thinner inlet boundary layer. A pair of counter-rotating vortices
was also observed at the exit plane, which increased the non-uniformity of the stream-wise velocity. No signicant pressure recovery
changes were observed for the two inlet boundary layer thicknesses
considered in the study. The RAE2129 was later reexamined by Lee
and Yu (1994) using a Reynolds Averaged NavierStokes (RANS)
method. Despite the simplicity of the method, the calculated mean
ow features reproduced the experimental data in Whitelaw and
Yu (1993) with sufcient accuracy.
This literature survey conrms that the research on turbulent
ows in curved and/or diffusing ducts is topical, and several of
its aspects have been studied using both the experimental and
numerical approaches. RANS models have been used in the literature with a wide variety of prediction accuracy and implementation cost (Gullman-Strand et al., 2004; Schneider et al., 2010;
Gopaliya et al., 2007, 2011; Sparrow et al., 2009; Yang and Hou,
1999; Patankar et al., 1975; Rudolf and Desova, 2007; Saha et al.,
2007; Lee and Yu, 1994; El-Behery and Hamed, 2011). Herbst
et al. (2007) mentioned that a suitable RANS model could result
in a reasonable agreement with experimental data, but cautioned
that the prediction capability is highly model-dependent. Their
own Large Eddy Simulation (LES) study predicted a slightly smaller
separated-ow region than the one measured, as well as delayed
separation and reattachment. Abe and Ohtsuka (2010) also used
the LES approach and observed a signicantly under-predicted
wall friction, resulting in an overly early separation. Much
improved prediction capability was obtained when a hybrid
LES/RANS method was used. On the other hand, Ohlsson et al.
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
2. Experimental setup
The geometry of the Allison 250 model diffuser duct (hereafter
referred to as A250 S-Diffuser) is shown in Fig. 1a. It has a total
length of 426 mm, which includes a 155.2 mm conical section followed by a 270.8 mm constant diameter S-shaped duct. The inlet
and outlet diameters are 46.83 mm and 65.11 mm, respectively,
giving an area ratio of 1.933. The offset between the inlet and
the outlet is 48.2 mm. The top edge of the duct in the view in
Fig. 1a is adjoining the combustion chamber and turbine, and will
herein be referred to as the inner surface of the rst bend, or the
wall at R = 0.5. Accordingly, the bottom edge will be referred to
as the outer surface of the rst bend, or the wall at R = 0.5.
A service-worn sample of the A250 S-Diffuser was obtained as a
test article. A exible traverse system with a 0.8 mm pitch,
equipped with a Pitot-static tube is designed to be placed at three
stream-wise locations (1: entrance; 2: end of conical section; 4:
exit plane of diffuser). Furthermore, pressure measurements could
be taken in the four directions indicated in Fig. 1b. An Omega
PX139 pressure transducer rated for 0.3 psi with an output voltage between 0.25 and 4.25 V is used, in conjunction with a National Instruments Analog/Digital converter (NI cRIO9201). At
each position, measurements are taken over a period of 10 s at
153
100 Hz, which allows for the calculation of a mean value along
with its standard deviation. A Westinghouse 1.5 hp, 3450 rpm motor, mated to a Canadian Blower & Forge Co. centrifugal blower, is
used to generate the desired ow. A 2 m long steel, drawn-overmandrel, tube of inside diameter equal to that of the inlet of the
A250 S-Diffuser is congured to the blower of outlet diameter
4.5 in via a plain reducer and a contraction cone. Honeycomb ow
straighteners are placed at the entrance of this initial tube in order
to minimize instabilities before the ow enters the test-section. In
addition, a 1 m steel extension tube, with an inside diameter equal
to that of the exit of the A250 S-Diffuser, is placed at the exit of the
diffuser in order to prevent outlet conditions from affecting the
ow in the region of interest. This practice is common in both
experimental and numerical approaches (Gullman-Strand et al.,
2004; Cherry et al., 2008; Schneider et al., 2010; Jakirlic et al.,
2010; Sparrow et al., 2009; Saha et al., 2007; Anand et al., 2003;
Gopaliya et al., 2007; Abdellatif et al., 2008; Whitelaw and Yu,
1993; Lee and Yu, 1994; El-Behery and Hamed, 2011; Ohlsson
et al., 2010). Under the considered conditions, a mass ow-rate
of 0.055 kg/s and a Reynolds number of 80,000 are obtained. The
area averaged turbulence intensity at the inlet plane of the component (Station 1) is estimated to be 1.6%. Note that the experimental
rig is, in the long-term, meant to be used in order to study the
propagation of ow instabilities from the compressor to the combustion chamber.
In the following, the radial coordinates z (AA direction) and r
(CC direction) are non-dimensionalized by the local duct diameter,
the axial coordinate x is non-dimensionalized by the total diffuser
length, and the velocity components u, v and w are non-dimensionalized by the local bulk velocity.
In this study, the sensitivity of the ow features and performance
characteristics of the A250 S-Diffuser are discussed by considering
several geometrical variations, see Fig. 2. All geometries in Fig. 2
have the same inlet and outlet diameters and centreline offset as
the A250 S-Diffuser. For the S-Diffuser 1/3 (Fig. 2a), the cross-sectional area expansion occurs between Sections 1 and 3. For the S-Diffuser 1/4 (Fig. 2b), the area expansion extends over the entire duct
length, as it is for the RAE2129 and most inlet ducts. For the S-Diffuser 2/3 (Fig. 2c), the area expansion occurs between Sections 2
and 3. The same nomenclature convention is adopted for the S-Diffuser 2/4 (Fig. 2d), S-Diffuser 3/4 (Fig. 2e) and S-Diffuser 1/23/4 (not
shown). Consistently, the A250 S-Diffuser (Fig. 1a) could be referred
to as S-Diffuser 1/2. Finally, an S-shaped duct (not shown) with a
constant cross-sectional area, a diameter equal to that of the inlet
of the A250 S-Diffuser, and the same centreline offset as the A250
S-Diffuser will also be considered for comparison purposes.
3. Numerical approach
The commercial CFD software ANSYS-CFX is adopted for this
study to solve the steady-state Cartesian coordinates Navier
Stokes equations using a fully conservative, control-volume-based,
Fig. 1. (a) Geometry of the A250 S-Diffuser and (b) pressure traverse paths.
154
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
Fig. 2. (a) S-Diffuser 1/3, (b) S-Diffuser 1/4, (c) S-Diffuser 2/3, (d) S-Diffuser 2/4 and (e) S-Diffuser 3/4.
@qU j k f
@
@k
P k b qkx
l rk lt
@xj
@xj
@xj
@U j x
@
@x
aqS2 bqx2
l rx lt
@xj
@xj
@xj
1 @k @ x
21 F 1 qrx2
x @xj @xj
where F1 is equal to zero away from the surface (je model) and
equal to one inside the boundary layer (jx model). The model also
includes a slight amendment to the denition of the eddy viscosity
for a better prediction of the turbulent shear stress. The calculation
of the distance from the wall is achieved by the solution of a Poisson
equation. More details on the SST model could be found in Sparrow
et al. (2009), and Menter (1994), and are not reproduced here for
conciseness. The suitability of the SST turbulence model is discussed in the next section. For each simulation, the convergence
155
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
is judged based on the variations of the scaled residuals of the governing equations (values in the order of or below 106 are desired)
and veried using an integral mass balance.
The wall boundary conditions are the standard no-slip and
impermeability conditions. At the inlet of the computational domain, a uniform velocity prole corresponding to the desired Reynolds number is prescribed. This results in fully-developed ow
conditions at the inlet of the component under study. At the other
end of the computational domain, stream-wise second derivatives
of velocities are assigned a zero value to represent fully-developed
ow conditions. Finally, all simulations are performed with standard air at the constant temperature of 25 C.
Table 1
Characteristics of some of the tested grid resolutions.
Grid resolution
Total
G1
G2
G3
G4
G5
G6
45.4
6.98
1.12
1.12
1.12
0.68
10
25
20
20
20
30
15
55
70
70
70
100
100
100
100
50
150
100
180
180
180
90
270
180
288,108
2,770,893
2,692,188
1,789,148
3,905,648
5,907,168
0.006
G1
G2
G3
0.004
G4
Cf
G5
G6
0.002
0.25
0.5
0.75
0.75
X*
(a) R* = 0.5
0.006
G1
G2
G3
0.004
G4
G5
G6
Cf
0.002
0.25
0.5
X*
(b) R* = -0.5
Fig. 4. Effects of the grid resolution on the calculated the friction coefcients.
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
0.4
0.4
0.2
0.2
Z*
Z*
156
-0.2
SST
-0.2
Station 1
Station 2
Station 4
-0.4
-0.4
0
0.5
1.5
0.5
U*
(a) AA line
0.4
0.4
0.2
0.2
R*
R*
(a) AA line
0
Station 1
Station 2
Station 4
-0.2
1.5
U*
SST
-0.2
-0.4
-0.4
0
0.5
1.5
0.5
U*
U*
(b) CC line
(b) CC line
1.5
time) may be explained by the fact that the former two models account for some near-wall turbulence anisotropy, which is not the
case for the je model. Menter (1994) also noted the very little
sensitivity of the je model to the pressure gradient, and attributed the better performance of the SST model compared to the
jx model to the presence of an additional cross-diffusion term
in the dissipation rate transport equation (last term on the right
hand side of Eq. (2)).
In the absence of published data on the A250 S-Diffuser, the
numerical code has nally also been tested against available experimental data on the RAE2129 diffuser from the literature (Whitelaw and Yu, 1993). The RAE2129 S-shaped inlet duct differs from
the A250 S-Diffuser in that it has a smaller area ratio (1.4), a larger
offset (0.3), and its cross-sectional area expands over its entire
length. The test is conducted for the fully-developed ow inlet condition of Whitelaw and Yu (1993) at a Reynolds number of
400,000. Fig. 7 shows that the development of the stream-wise
velocity prole is well predicted by the model (present calculated
results are shown in the form of velocity vectors), while Fig. 8 also
shows a reasonably good agreement between the experimental
and numerical data; this agreement is by all means at least as good
as the one obtained by the same authors when they adopted the
numerical approach (Lee and Yu, 1994).
In view of the grid resolution sensitivity analysis (Fig. 4), the
experimental validation cases (Figs. 5, 7 and 8) and the turbulence
157
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
0.4
0.2
Cp
(a) AA line
0
-0.2
-0.4
-0.1
0.1
0.2
(b) CC line
0.3
X
Fig. 8. Development of the pressure coefcient in the RAE2129 (Experimental data
are from Whitelaw and Yu, 1993).
model discussion (Fig. 6), the numerical model (the SST turbulence
model as imbedded in ANSYS-CFX) and the implementation
adopted in this study are considered validated and appropriate
for the investigation of the problem at hand.
4. Results
4.1. Development of the stream-wise velocity prole in the A250 SDiffuser
Fig. 9 shows the variations of the stream-wise velocity prole
along the A250 S-Diffuser at a Re = 80,000. In the AA direction
(Fig. 9a), the velocity prole is symmetrical and gradually attens
downstream, until the centreline velocity becomes smaller than
the velocity in the wall region, with the emergence of an increasingly pronounced overshoot. Along the CC line (Fig. 9b), the symmetrical prole at the inlet gets increasingly distorted towards
the wall at R = 0.5, while the ow near the wall at R = 0.5 is
decelerated. This is an indication of the presence of a secondaryow in the plane of the cross-section as it will be discussed later
in this report.
4.2. Effects of the turbulence intensity at the inlet on the ow and
performance of the A250 S-Diffuser
While rarely investigated in the literature, different operating
conditions could result in signicantly different inlet turbulence
levels, considerably affecting the ow and performance of the diffuser (Sullerey et al., 1983). Three turbulence intensity values are
therefore tested here: Tu = 1%, Tu = 5% and Tu = 10%. The Reynolds
number has been kept constant at 80,000 for all three simulations,
and Figs. 1012 present calculated data at the exit plane of the diffuser (Station 4). The qualitative trends are similar, but increasing
the inlet turbulence intensity (equivalent to enhancing the mixing)
results in a markedly less pronounced overshoot of the velocity
prole in the AA direction (Fig. 10a), and also a less pronounced
distortion of the velocity prole along the CC direction (Fig. 10b).
Fig. 11a shows an initial rapid decrease of the friction coefcients
at the wall at R = 0.5 in the conical section of the A250 S-Diffuser,
followed by a brief increase at the convex bend, and then a gradual
decrease until about X = 0.8 where the boundary layer seems to
detach from the surface though a ow reversal does not actually
occur, corresponding to what is referred to as a boundary layer collision in Talbot and Wong (1982). At the second bend, the friction
coefcient starts increasing again. Near the outer surface of the
rst bend (R = 0.5), the friction coefcients for all three inlet turbulent intensities decrease as the ow passage enlarges, closely
approaching zero, then sharply increase after the concave bend
as a consequence of the ow acceleration in that region of the passage (see also Fig. 9b). This increase is sustained until the convex
bend, at which the friction coefcients drop sharply. The effect of
the inlet turbulence intensity on the quantitative Cf variations is
limited overall, except along the R = 0.5 wall far downstream
where the highest turbulence level correlates with the least friction losses (Fig. 11b). Finally, Fig. 12 shows that the pressure coefcient variations for the three inlet turbulence intensity values
considered are qualitatively similar, indicating a gradual pressure
158
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
0.006
0.4
Tu=1%
Tu=5%
Tu=10%
Tu = 1%
Tu = 5%
Tu = 10%
0.2
Cf
Z*
0.004
-0.2
0.002
-0.4
0
0.5
1.5
U*
0.25
0.5
0.75
0.75
X*
(a) AA line
(a) R* = 0.5
0.4
Tu=1%
Tu=5%
Tu=10%
0.006
Tu = 1%
Tu = 5%
Tu = 10%
0.004
Cf
R*
0.2
-0.2
0.002
-0.4
0
0.5
1.5
U*
(b) CC line
Fig. 10. Effects of the inlet turbulence intensity on the stream-wise velocity prole
at the exit plane of the A250 S-Diffuser.
0.25
0.5
X*
(b) R* = -0.5
Fig. 11. Effects of the inlet turbulence intensity on the calculated friction
coefcients along the wall of the A250 S-Diffuser.
recovery, one of the most important diffuser performance parameters. Quantitatively however, the highest inlet turbulence intensity
returns slightly higher pressure coefcients, while the lowest turbulence intensity value consistently results in lower pressure coefcient values. These results may be explained by the fact that high
turbulence intensity is synonymous to more important mixing, and
are consistent with published experimental observations from the
literature (Sullerey et al., 1983).
the proles along the AA line (Fig. 13a) reveal an increasingly pronounced overshoot in the near wall region, and monotonically
decreasing stream-wise velocities in the centreline when the Reynolds number is increased. Along the CC line (Fig. 13b), the distortion of the velocity prole toward the R = 0.5 wall is also
observed to intensify as the Reynolds number is increased. It is
worth noting that the slope of the velocity prole at about
R = 0.3 suggests that a second overshoot may appear next to the
inner surface of the rst bend should the Reynolds number be further increased. The effect of the Reynolds number on the friction
coefcient variations is illustrated in Fig. 14. The qualitative trends
are again similar for all four Reynolds numbers tested, and the differences in the calculated Cf values are small along the inner surface of the rst bend (slight decrease of the friction coefcients
as the Reynolds number is increased). Along the R = 0.5 wall
however, the friction coefcient variations are extremely sensitive
to the Reynolds number, the simulation at the lowest Reynolds
number returning the highest Cf values over most of the length
of the duct, and most importantly downstream its conical section.
The effect of the Reynolds number on the variations of the pressure
coefcients is shown in Fig. 15, which displays qualitatively
similar results in all cases; however the highest Reynolds number
159
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
0.8
0.4
0.6
0.2
Z*
Cp
25K
0.4
80K
116K
209K
Tu = 1%
-0.2
Tu = 5%
0.2
Tu = 10%
-0.4
0.25
0.5
0.75
0.5
1.5
U*
X*
(a) AA line
(a) R* = 0.5
0.8
0.4
25K
80K
116K
0.2
0.6
Cp
R*
209K
0
0.4
Tu = 1%
-0.2
Tu = 5%
0.2
Tu = 10%
-0.4
0
0
0.25
0.5
0.75
X*
0.5
1.5
U*
(b) CC line
(b) R* = -0.5
Fig. 12. Effects of the inlet turbulence intensity on the calculated pressure
coefcients along the wall of the A250 S-Diffuser.
Fig. 13. Effects of the Reynolds number on the stream-wise velocity prole at the
exit plane of the A250 S-Diffuser.
and the region of the ow where the stream-wise velocity is maximum gradually shifts toward the concave surface under the double effect of centrifugal forces and radial pressure gradients. For the
S-Diffuser 1/4, a separated-ow region is visible at X = 0.81. A separated-ow region is also observed for the S-Diffuser 2/3 at
X = 0.51; it is bigger at X = 0.66, smaller at X = 0.81, and seems
to have disappeared by the exit of the duct. As for the S-Diffuser
3/4, the separated-ow region is rst observed at X = 0.81, and it
is bigger at the exit plane. One single pair of recirculation vortices
is predicted in most cases. Most of these observations are in very
close agreement with those by previous authors on similar geometries (Dean and Hurst, 1959; Patankar et al., 1975; Lee and Yu,
1994). In order to quantify the intensity of the secondary-ow, Table 2 gives the calculated non-uniformity index (NUI) at the exit
plane for the eight geometries considered in this study. The nondiffusing duct (S-Duct) is seen to correspond to the lowest NUI,
agreeing with conclusions in a previous LES study (Abdellatif
et al., 2008). The NUI for the A250 S-Diffuser is relatively high,
but it is the S-Diffuser 2/3 geometry that corresponds to the highest NUI. It is also interesting to note that the normalized maximum
secondary-ow velocity at the exit plane is smaller for all the
geometries in this study than for the RAE2129 in Whitelaw and
Yu (1993). While the most uniform ow possible is desired at
the exit plane of inlet ducts (NUI < 10% qualies as uniform ow)
160
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
0.006
0.8
25K
80K
0.6
116K
0.004
Cp
Cf
209K
25K
0.4
80K
116K
0.002
209K
0.2
0.25
0.5
0.75
X*
0.25
0.5
0.75
X*
(a) R* = 0.5
(a) R* = 0.5
0.006
0.8
25K
80K
116K
0.6
209K
Cp
Cf
0.004
25K
0.4
80K
116K
0.002
209K
0.2
0.25
0.5
0.75
X*
(b) R* = -0.5
0.25
0.5
0.75
X*
(b) R* = -0.5
Fig. 14. Effects of the Reynolds number on the development of the friction
coefcients along the wall of the A250 S-Diffuser.
Fig. 15. Effects of the Reynolds number on the development of the pressure
coefcients along the wall of the A250 S-Diffuser.
the earliest, is the biggest, and extends over more than half the
length of the duct, while the ow separates but does not reattach
(within the component) for three other geometrical congurations.
Overall, Fig. 17 and Table 3 suggest that the A250 S-Diffuser is the
most efcient geometry, as it produces the least friction losses, and
it resists ow separation.
The static pressure recovery potentials for each of the geometries considered are shown in Fig. 18. It is seen that the geometries
that present cross-sectional area expansions starting at the inlet of
the duct are the ones that allow for the highest performances.
These geometries also correspond to the shortest lengths of separated-ow region, if any. It is also worth noting that the superior
performance of the A250 S-Diffuser compared to the S-Diffuser
1/4 suggests that the current most common inlet duct design (i.e.
cross-sectional area variation over the entire length of the duct)
is not optimal, at least as far as the static pressure recovery potential is concerned. The S-Diffuser 2/3 and the S-Diffuser 2/4 also
possess decent static pressure recovery potentials, but they are
penalized by the pressure loss between Sections 1 and 2. Limiting
the cross-sectional area expansion to the last section of the duct is
obviously counterproductive. It is also noted that the value of the
static pressure recovery coefcient for the constant cross-sectional
area S-shaped duct is exactly equal to that predicted by the LES
161
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
A250 S-Diffuser
S-Diffuser 1/4
S-Diffuser 2/3
S-Diffuser 3/4
X* = 0
X* = 0.35
X* = 0.51
X* = 0.66
X* = 0.81
X* = 1
Fig. 16. Sensitivity of the stream-wise and cross-ow velocity to the geometry of the duct.
sure loss coefcients are compared in Fig. 19, conrming the better
performance of geometries with cross-sectional area expansions
starting at the inlet. At the other end of the spectrum, the
162
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
Table 2
Sensitivity of the non-uniformity index (NUI) to the geometry of the duct.
0.6
Non-uniformity index
0.085
0.065
0.073
0.064
0.112
0.071
0.059
0.027
Pressure Recovery
Duct geometry
0.4
0.2
0.01
A250 S-Diffuser
S-Duct
S-Diffuser 1/2-3/4
S-Diffuser 1/3
S-Diffuser 1/4
S-Diffuser 2/3
S-Diffuser 2/4
S-Diffuser 3/4
0.008
Cf
0.006
-0.2
0.25
0.5
0.75
X*
Fig. 18. Sensitivity of the static pressure recovery to the geometry of the duct (same
legend as in Figs. 17 and 19).
0.004
0.2
A250 S-Diffuser
S-Duct
S-Diffuser 1/2-3/4
S-Diffuser 1/3
S-Diffuser 1/4
S-Diffuser 2/3
S-Diffuser 2/4
S-Diffuser 3/4
0.002
0.15
0.25
0.5
0.75
CL
0.1
X*
(a) R* = 0.5
0.05
A250 S-Diffuser
S-Duct
S-Diffuser 1/2-3/4
S-Diffuser 1/3
S-Diffuser 1/4
S-Diffuser 2/3
S-Diffuser 2/4
S-Diffuser 3/4
0.008
0.25
0.5
0.75
X*
Fig. 19. Sensitivity of the total pressure loss coefcient to the geometry of the duct.
Cf
0.006
0.004
S-Diffuser 2/3 and the S-Diffuser 3/4 produce total pressure losses
equivalent to those of the constant cross-sectional area, S-shaped
duct.
0.002
5. Conclusions
0
0.25
0.5
0.75
X*
(b) R* = -0.5
Fig. 17. Sensitivity of the friction coefcients to the geometry of the duct.
Table 3
Separated ow regions for the S-shaped diffuser geometries investigated.
Separation Reattachment Separated
point
point
ow
length
A250 S-Diffuser (S-Diffuser 1/2)
S-Diffuser 1/23/4
S-Diffuser 1/3
S-Diffuser 1/4
S-Diffuser 2/3
S-Diffuser 2/4
S-Diffuser 3/4
S-Duct
n/a
0.765
0.569
0.747
0.443
0.618
0.716
n/a
n/a
n/a
0.744
0.989
0.969
n/a
n/a
n/a
n/a
n/a
0.175
0.242
0.526
n/a
n/a
n/a
An investigation has been conducted on the Allison 250 gas turbine S-shaped diffuser which connects the exit of the compressor
to the inlet of the combustion chamber using the standard je
and jx models, as well as the Shear-Stress Transport (SST) model
implemented in the commercial code ANSYS-CFX. Data measured
using an in-house purpose-designed test rig and other experimental data taken from the literature were used for validation. The
standard jx model was shown to qualitatively reproduce the
general trends, but the quantitative agreement with the experimental data was poor. The standard je model performed even
worse. Conversely, the SST model successfully passed the validation tests and was therefore adopted for the remainder of the analysis. Among other calculated results, the stream-wise velocity
prole was observed to atten in the initial divergent section,
experience a slight shift toward the inner surface at the rst bend,
before a gradually intensifying secondary-ow is initiated, pushing
the region of the ow with the highest velocity toward the concave
surface of the rst bend. This was accompanied by an increase of
G.G. Lee et al. / International Journal of Heat and Fluid Flow 42 (2013) 151163
163
Cherry, E.M., Elkins, C.J., Eaton, J.K., 2008. Geometric sensitivity of threedimensional separated ows. Int. J. Heat Fluid Flow 29, 803811.
Dean, W.R., Hurst, J.M., 1959. Note on the motion of uid in a curved pipe.
Mathematika 6, 7785.
El-Behery, S.M., Hamed, M.H., 2011. A comparative study of turbulence models
performance for separating ow in a planar asymmetric diffuser. Comput.
Fluids 44, 248257.
Gopaliya, M.K., Kumar, M., Kumar, S., Gopaliya, S.M., 2007. Analysis of performance
characteristics of S-shaped diffuser with offset. Aerospace Sci. Technol. 11, 130
135.
Gopaliya, M.K., Goel, P., Prashar, S., Dutt, A., 2011. CFD analysis of performance
characteristics of S-shaped diffusers with combined horizontal and vertical
offsets. Comput. Fluids 40, 280290.
Gullman-Strand, J., Tornblom, O., Lindgren, B., Amberg, G., Johansson, A.V., 2004.
Numerical and experimental study of separated ow in a plane asymmetric
diffuser. Int. J. Heat Fluid Flow 25, 451460.
Herbst, A.H., Schlatter, P., Henningson, D.S., 2007. Simulations of turbulent ow in a
plane asymmetric diffuser. Flow Turbul. Combust. 79, 275306.
Jakirlic, S., Kadavelil, G., Kornhaas, M., Shafer, M., Sternel, D.C., 2010. Tropea:
numerical and physical aspects in LES and hybrid LES/RANS of turbulent ow
separation in a 3-D diffuser. Int. J. Heat Fluid Flow 31, 820832.
Lee, K.M., Yu, S.C.M., 1994. Computational studies of ows in the RAE 2129 S-shaped
diffusing duct. In: 32nd AIAA Aerospace Sciences Conf., Reno, NV, Paper AIAA940658.
Lee, G.G.W., Allan, W.D.E., Goni Boulama, K., 2012. Numerical and experimental
analysis of the airow inside an A250 diffuser tube. In: Int. Gas Turbine Institute
Conf., Copenhagen, Denmark, Paper GT201269708.
McDill, P.L., 1989. An Experimental Evaluation of S-Duct Inlet-Diffuser
Congurations for Turboprop Offset Gearbox Applications. NASA, Report,
NASA-CR-179454.
Menter, F.R., 1994. Two-equation eddy-viscosity turbulence models for engineering
applications. AIAA J. 32, 15981605.
Ohlsson, J., Schlatter, P., Fischer, P.F., Henningson, D.S., 2010. Direct numerical
simulation of separated ow in a three dimensional diffuser. J. Fluid Mech. 650,
307318.
Patankar, S.V., Pratap, V.S., Spalding, D.B., 1975. Prediction of turbulent ow in
curved pipes. J. Fluid Mech. 67, 583595.
Rhie, C.M., Chow, W.L., 1983. Numerical study of the turbulent ow past an airfoil
with trailing edge separation. AIAA J. 21, 15251532.
Rudolf, P., Desova, M., 2007. Flow characteristics of curved ducts. Appl. Comput.
Mech. 1, 255264.
Saha, K., Singh, S.N., Seshadri, V., 2007. Computational analysis on ow through
transition S-diffusers: effect of inlet shape. J. Aircraft 44, 187193.
Schneider, H., von Terzi, D., Bauer, H.J., Rodi, W., 2010. Reliable and accurate
prediction of three-dimensional separation in asymmetric diffusers using
Large-Eddy Simulation. J. Fluids Eng. 132, 031101-1031101-8.
Sparrow, E.M., Abraham, J.P., Minkowycz, W.J., 2009. Flow separation in a diverging
conical duct: effect of Reynolds number and divergence angle. Int. J. Heat Mass
Transfer 52, 30793083.
Sullerey, R.K., Chandra, B., Muralidhar, V., 1983. Performance comparison of straight
and curved diffusers. Def. Sci. J. 33, 195203.
Talbot, L., Wong, S.J., 1982. A note on boundary-layer collision in a curved pipe. J.
Fluid Mech. 122, 505510.
Towne, C.E., 1984. Computation of viscous ow in curved ducts and comparison
with experimental data. In: AIAA Aerospace Sc. Meeting, Reno, NV, Paper AIAA84-0531.
Whitelaw, J.H., Yu, S.C.M., 1993. Turbulent ow characteristics in an S-shaped
diffusing duct. Flow Meas. Instrum. 4, 171179.
Wu, X., Schluter, J., Moin, P., Pitsch, H., Iaccarino, G., Ham, F., 2006. Computational
study on the internal layer in a diffuser. J. Fluid Mech. 550, 391412.
Yang, Y.T., Hou, C.F., 1999. Numerical calculation of turbulent ow in symmetric
two-dimensional diffusers. Acta Mech. 137, 4354.