Professional Documents
Culture Documents
2013intermittent Corner Separation in A Linear Compressor Cascade
2013intermittent Corner Separation in A Linear Compressor Cascade
DOI 10.1007/s00348-013-1546-y
RESEARCH ARTICLE
Leboeuf Francis
Received: 15 September 2012 / Revised: 27 March 2013 / Accepted: 23 May 2013 / Published online: 2 June 2013
Ó Springer-Verlag Berlin Heidelberg 2013
Abstract To gain a better knowledge of the mechanisms inherent flow feature in compressors and relates to the
of corner separation and to obtain a database that could be efficiency and stability of compressors (Cumpsty and
used to calibrate large eddy simulation, a detailed and Greitzer 2004). Generally speaking, corner separation is
accurate experimental investigation of the flow field in a caused mainly by the strong streamwise pressure gradient,
linear compressor cascade has been carried out at a Rey- the presence of secondary flow, and the merging of the wall
nolds number of 3.82 9 105 based on the blade chord and and blade boundary layers. A number of experimental and
the inlet free stream velocity. The experimental data numerical studies have been carried out to investigate the
include (1) the inlet flow conditions measured by one- impacts, the flow topology, the influencing factors, as well
dimensional hot-wire anemometry, (2) the static pressure as global consequence of corner separation (e.g., Schulz
on the blade and the endwall measured by pressure taps, (3) et al. 1990a, b; Gbadebo et al. 2005; Lei et al. 2008).
the outlet flow measured by a five-hole pressure probe, and However, until now, an effective control of these effects is
(4) the velocity field in the passage measured by two- still very difficult to achieve, because the mechanisms of
component laser Doppler anemometry. In this highly the corner separation are not clearly understood (Gbadebo
intermittent corner separation, the points with bimodal et al. 2005). In computational fluid dynamics (CFD), the
histograms of velocity exist and mainly appear around the Reynolds-averaged Navier–Stokes (RANS) cannot tackle
mean interface of separated flow and non-separated flow. accurately corner separations, especially their exact loca-
This bimodal phenomenon is due to two aperiodic modes tion and size, as well as the level of high losses that corner
in the flow field. separations generate in compressors. This is mainly
because corner separation is fully unsteady and highly
intermittent. The first objective of the present study is to
1 Introduction understand further the mechanisms of the corner separa-
tion, especially the intermittent feature of corner
Corner separation, the three-dimensional (3D) separation in separation.
the junction of the endwall and blade corner region, is an In some highly intermittent flow, a remarkable phe-
nomenon is the existence of bimodal histograms of velocity
M. Wei (&) O. Xavier L. Francis (referred as ‘‘bimodal histograms’’ later), which have two
Laboratoire de Mécanique des Fluides et d’Acoustique, peaks. In the literature, this phenomenon has been found
Ecole Centrale de Lyon, Ecully, France and discussed in different flow fields.
e-mail: maweigg@gmail.com
The most famous flow field is the junction flow firstly
Present Address: discussed by Devenport and Simpson (1990), in which
M. Wei bimodal histograms are mainly found in the vicinity of the
School of Aerospace, Tsinghua University, Beijing, China wing nose. Devenport and Simpson (1990) claimed that the
flow switched aperiodically between two different modes.
L. Lipeng
School of Jet Propulsion, Beijing University of Aeronautics Following there are numerous numerical simulations to
and Astronautics, Beijing, China understand further the physics of such flow. A recent
123
Page 2 of 17 Exp Fluids (2013) 54:1546
comprehensive review of both the experiments and simu- validation of LES in turbomachinery, is thus an obvious
lations can be found in Gand et al. (2010). Now, it is need.
widely accepted that the bimodal histograms in the junction In order to gain a better knowledge of the mechanisms
flow are consistent with large-scale low-frequency of corner separation especially their intermittent feature
unsteadiness of the instantaneous flow structure associated and to obtain an experimental database for calibrating LES,
with the horseshoe vortex. a detailed and accurate experiment on the flow field in a
In subsonic turbomachinery flows that are more com- linear compressor cascade has been carried out in the
plex than the junction flow mentioned above, bimodal Laboratoire de Mécanique des Fluides et d’Acoustique
histograms usually exist in two situations. One is in the (LMFA) at Ecole Centrale de Lyon. Experimental data
blade wakes that are similar to the Karman vortex street, were acquired for a Reynolds number of 3.82 9 105 based
due to the vortex shedding. The other one is in tip-clear- on the blade chord and the free stream velocity. Mea-
ance flows (e.g., Tian and Simpson 2007), due to the surements have been achieved using 1D hot-wire ane-
interaction between the tip-leakage vortex and tip-leakage mometry, pressure taps on blade, and endwall, a five-hole
jet. These interactions are periodic, and the frequencies of pressure probe and two-component LDA. Some details not
the vortex shedding and interaction can be fixed in the shown in this paper for keeping this paper more concise,
spectra of instantaneous velocities. This is the obvious such as the determinations of the uncertainties in the
difference with the bimodal phenomenon in the junction quantities measured by each technique and the results of oil
flow. visualization, can be found in Ma (2012).
Linear compressor cascade is a simplified model for
investigating the flow field in the compressor. In the pas-
sage of a linear subsonic compressor cascade without tip 2 Linear compressor cascade
clearance, bimodal histograms have also been found in the
flow field at mid-span measured by one-component LDA The investigation was carried out in a low-speed linear
carried out by Hobson et al. (1998). However, they only compressor cascade wind tunnel. Figure 1 shows the
associated the bimodal histograms with the backflow schematic of the experimental arrangement. The facility is
coefficients and did not interpret further its physics. One of powered by a 60 kW centrifugal blower that supplies air to
the objectives of this paper is to try to find bimodal his- a settling chamber terminated by a 5.14:1 contraction.
tograms in the passage of a linear compressor cascade From the contraction, flow enters the upstream part of the
without tip clearance and investigate further their test section, a rectangular duct with a cross-section of
mechanisms. 900 mm height by 370 mm wide. This cascade consists of
The insufficiency of RANS to simulate the corner sep- 13 blades. For the blade, its thickness distribution is the
aration becomes increasingly unable to meet the demand NACA 65-009 and its mean camber line is a circular arc
for efficient and accurate design tools, which is more and with a camber angle of 23.22°. The incidence angle is 4°, at
more important for designing aircraft engines with higher which no separation exists at mid-span and the extent of the
efficiency and better stability in a shorter design cycle. In corner separation is already large enough according to the
order to understand, predict, control, and eventually reduce results of oil visualization. The parameters of the cascade
the corner separations, it is paramount to extend novel are summarized in Table 1.
unsteady CFD techniques to these flows. LES is an alter- Sandpaper strips have been stuck on the whole span and
native, in which the largest turbulent eddies are represented at 6.0 mm downstream from the leading edge on the
directly and the smallest eddies are modelled by subgrid pressure and suction sides of all the blades in the cascade.
scale modelling (Sagaut 2001). LES has been used widely The width in streamwise direction and thickness of the trips
to analyze academic flows (e.g., Wu and Squires 1998; are 3.0 and 0.3 mm, respectively. The grid size of the
Laval et al. 2010). In turbomachinery flows, there are sandpaper is ISO P600, in which the average particle
already some applications of LES, for example, You et al. diameter is about 25.8 lm. From the results of oil visual-
(2007), Hah (2009), Boudet et al. (2010), and Gand et al. ization, the trips induce the transition and thus remove the
(2010). All of these researches have reflected the advan- transition bubbles both on the suction side and the pressure
tages of LES compared with RANS. However, LES still side. This enables to increase the incidence angle (and also
needs to be calibrated due to the use of subgrid models. the blade loading) without a separation on the suction side.
Unfortunately, in turbomachinery, available experimental And thus the authors can focus exclusively on the physics
data generally describe only the averaged quantities and associated with corner separation, without introducing
usually lack of detailed information on turbulence. A high- some problems related to the transition physics. At the
quality experimental database, specifically dedicated to the same time, it is tricky to simulate transition with CFD.
123
Exp Fluids (2013) 54:1546 Page 3 of 17
Table 1 Parameters of the cascade inlet flow velocity profiles in the streamwise direction were
Blade profile NACA 65
measured by 1D hot-wire anemometry at two stations,
which are located at 455 and 960 mm (i.e. 4.13ca and
Blade number 13
8.71ca) upstream of the leading edge of blade 6 (see in
Chord c = 150 mm
Fig. 1), respectively. The evolution of the inlet boundary
Camber angle 23.22°
layer provides detailed inlet flow conditions needed for
Stagger angle 42.70°
CFD simulations including LES.
Axial chord ca = 110 mm
Pitch s = 134 mm
3.1 Hot-wire anemometry
Solidity 1.12
Span h = 370 mm
The sensors were operated by a Dantec 90C10 constant
Aspect ratio 2.47
temperature anemometer unit in Dantec Streamline. Square
Optimum incidence 0.18°
wave tests were used to optimize the bandwidth of the
Design upstream flow angle 54.31°
combined sensor/anemometer circuit. In the experiment,
Incidence angle 4° the bandwidth of the hot-wire probe/anemometer system
Inlet velocity U1 ¼ 40:0 m/s (or cutoff frequency) reached about 40 kHz. The low-pass
Reynolds number Re ¼ U1 c=m ¼ 3:82 105 filter was 100 kHz. The acquisition frequency was set to
200 kHz and the number of samples was set to one million.
The hot wire was calibrated by a fourth-order polynomial
The good states of periodicity and symmetry of the to correlate the wire output voltages with the cooling
cascade were validated by the outlet total pressure losses velocities. The Bearman correction (Bearman 1971) and
measured by a five-hole pressure probe (see in Ma 2012). Cox correction (Cox 1957) were used to correct for
The inlet reference conditions were measured at the ambient temperature and wall temperature proximity
reference point (see in Fig. 1) by a thermocouple type K effects, respectively. The active length l and the diameter
and a Pitot probe is used to normalize the experimental d of the boundary layer hot-wire probe were 0.82 mm and
data. The inlet reference conditions were always measured 4 lm, respectively. The corresponding value of l/d was
synchronously with each measurement, to reduce the larger than 200, so that conduction errors were avoided in
influence of the fluctuation in the inlet flow velocity even the experiment.
though this fluctuation was very small in this experiment. In the experiment, a lot of energy was used to improve
the accuracy of the distance from the hot-wire probe to the
endwall. The original position, i.e., the distance between
3 Inlet flow conditions the hot-wire probe and wall is zero, was sought and dem-
onstrated by the voltage change from an electric circuit.
First of all, the inlet main flow upstream of the three pas- The vibration of the hot wire in the experiment, which can
sages between the blade 5 and the blade 8 (see in Fig. 1) only be reduced and cannot be ruled out, was also con-
was measured by a five-hole pressure probe. The unifor- sidered concerning the distance. First of all, the original
mity of this inlet flow is more than 99.5 %. And then the position was sought without flow. And then the original
123
Page 4 of 17 Exp Fluids (2013) 54:1546
position was sought with flow. The difference between 3.3 Streamwise Reynolds normal stresses
these two original positions in the traverse system was
considered as the swinging distance of the hot wire. As a The relative streamwise Reynolds normal stresses, i.e.,
pffiffiffiffiffiffi
result, the swinging of the probe was \0.1 mm and not u0þ ¼ u02 =us , are shown in Fig. 3a. In this figure at each
significant. measurement station, a peak exists very close to the wall
The relative uncertainty in the mean velocity and where z? = 15. The value of this peak is about 2.7. In the
streamwise Reynolds normal stresses are \2 and 4 %, streamwise direction, u0þ increases slightly.
respectively. The absolute streamwise Reynolds normal stresses, i.e.,
pffiffiffiffiffiffi
0þ
3.2 Mean velocity u0 ¼ u02 =us0 , are shown in Fig. 3b, in which z? 0 = zus 0/m.
The friction velocity us 0 is the value at the first measure-
Figure 2a shows the mean velocity profiles. The thick- ment station. With the development in the streamwise
nesses of boundary layer, d, increases from 25.0 to direction, u0þ 0 decreases slightly in the inner part and
30.0 mm. Based on these mean velocity profiles, the dis- increases in the outer part.
placement thickness d*, the momentum thickness h, and the These developments are consistent with those in a tur-
energy thickness d3 are obtained and listed in Table 2. The bulent boundary layer with zero pressure gradient, which
corresponding two shape factors H are equal to 1.32, the have been discussed in detail by Ma et al. (2013). But it
typical value of fully developed turbulent boundary layers. needs to be noted that the increase of u0þ and decrease of
The Reynolds number based on the inlet velocity and the u0þ
0 in the inner part are not obvious if taking into account
momentum thickness, Reh, increases from 5,554 to 7,163. of the uncertainties in these two parameters.
The mean velocity profile in inner variables is shown in
Fig. 2b, in which u? = U/us and z? = zus/m. The friction 3.4 Spectrum
velocity us is determined from a least-square fit to the law
of the wall (here van Driest formula was used) between At each measurement point, the frequency spectrum was
z? = 10–50. At the nearest measurement point to the wall obtained directly carrying out the fast Fourier transform
where z = 0.11 mm, z? is approximately 8, thus this point (FFT) from the unsteady velocity measured by hot-wire
is in the buffer layer. anemometry. Figure 4a shows a typical spectrum at the
0.8
20
0.7
U/U∞
0.6
u+
15
0.5
Van Driest profile
0.4 10
0.3 Visc. Buffer Log.
0.2 5
Station 1
0.1
Station 2 0
0 100 101 102 103 104
0 0.1 0.2 0.3 0.4 0.5 +
z
z/h
123
Exp Fluids (2013) 54:1546 Page 5 of 17
(a) 3 (b) 3
2.5 2.5
2 2
+
u’+
+ +
u’0
1.5 z =15 1.5 z =15
1 1
0.5 0.5
Station 1 Station 1
Station 2 Station 2
0 0
100 101 102 103 104 100 101 102 103 104
+ +
z z0
Fig. 3 Inlet streamwise Reynolds normal stress. a Relative development, b absolute development
E11(K) (m3/s2)
show the experimental spectra -2 f -2 K
10 10
E(f) (m2/s)
10-4 10-4
10-5 10-5
+
Reθ=7163, z =16.0
-6 -6 +
10 10 Reθ=2000, z =20.6
Reθ≈5×106, z+=16.9
-7 -7
10 10
0 1 2 3 4 5 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10 10 10 10
f (Hz) K (1/m)
point where z = 0.25 mm and z? = 16.0. The maximum spectra from Metzger et al. (2001). One is at z? = 20.6
frequency is 100 kHz, because the sampling rate was with low Reynolds number of 2,000. The other one is at
200 kHz in the experiment. Usually, energy contained at z? = 16.9 with very high Reynolds number of 5 9 106.
frequencies above the maximum frequency will be aliased These three spectra are similar, and the difference can be
into the range investigated, causing the estimated spectrum interpreted as the influence of Reynolds number.
to flatten out as it approaches the maximum frequency Therefore, the experimental results providing the inlet
selected. In Fig. 4a, only very little energy is present above conditions are reliable.
the maximum frequency and the effect of aliasing can thus
be ignored completely below 50 kHz.
Figure 4b shows the spatial spectrum at the same point 4 Mean static pressure on the blade surface
transformed from the frequency spectrum according to the and the endwall
Taylor hypothesis (e.g., the turbulent structures transfer by
the mean flow): The mean static pressure on the blade surface and the
endwall was measured by pressure taps. Forty pressure
U
E11 ðKÞ ¼ Eðf Þ ð1Þ taps are positioned on the blade 6. Twenty-five taps are
2p located on its suction side, while the other fifteen are
where the wave number K ¼ 2p=U f and U is the mean located on its pressure side. The pressure taps on the
velocity at the measurement point. This figure also blade are made by 1.6-mm-diameter alloy steel tubing
shows other two experimental near-wall turbulence embedded into the blade surface, with a port of 0.8 mm
123
Page 6 of 17 Exp Fluids (2013) 54:1546
Z
Z
X Mid-span Y
Cp Mid-span Cp
0.6 0.6
Y
0.5 0.5 X
0.4 0.4 PS
0.3 TE 0.3
0.2 0.2
0.1 SS 0.1 TE
0 0
-0.1 -0.1
-0.2 -0.2
LE
-0.3 -0.3
-0.4 -0.4
-0.5 -0.5
-0.6 -0.6
LE
Endwall
Endwall
Fig. 5 Mean static pressure on the blade surface and the endwall. Markers indicate the measurement positions
123
Exp Fluids (2013) 54:1546 Page 7 of 17
Z
consistent with that of small velocities. This is because
Mid-span the static pressure remains nearly constant; the losses
obviously induce a deficit of the total pressure and as a
Y consequence a velocity deficit. From the closest mea-
X
surement section to the furthest from the trailing edge,
ω
the extents of loss regions (where x [ 0.05) increase;
0.6
in contrast, the extents of the high loss regions (where
0.55 x [ 0.60) decrease. This obviously illustrates the
0.5 mixing process. A zone with high energy (small losses,
0.45
0.4
x \ 0.05) exists at each outlet section and is labelled
0.35 ‘‘zone A’’ in Fig. 8, which shows x at the second
0.3 outlet section. This zone is between the wake and the
Trailing edge
0.25
0.2
corner stall and it moves away from the endwall, when
0.15 going from the first measurement section to the third.
0.1 At the same time, its extent increases. This phenom-
0.05
enon is due the development of the flow in the middle
of the passage that has high momentum. The main
flow is displaced upwards due to the blockage of
corner stall and batters the edge of the zone with high
losses.
Figure 8 also shows the secondary flow vector, which
is defined as the 3D velocity vector projected on the
plane perpendicular to the corresponding mid-span flow
direction. In our experiment, the measurement sections
Endwall are in the y-z plan. The secondary flow vector Vs is
defined as
Fig. 7 Total pressure loss coefficient distributions at three outlet
sections Vs ¼ V ðVÞmidspan ð6Þ
100
0.35
averaged total pressure loss coefficient x*. The second one
0.3 is the mass-averaged total pressure loss coefficient x0 .
zoneA
123
Page 8 of 17 Exp Fluids (2013) 54:1546
0.12
0.3
z/h
ω’
0.2
0.11
0.1
0 0.1
0 0.1 0.2 0.3 0.4 0.5 1 2 3
ω* NO. of section
Fig. 9 a Pitchwise mass-averaged total pressure loss coefficient x*, b mass-averaged total pressure loss coefficient x0 at outlet sections. Lines
for visual aid only
123
Exp Fluids (2013) 54:1546 Page 9 of 17
6.2.2 In corner
123
Page 10 of 17 Exp Fluids (2013) 54:1546
In order to present more clearly the structure of the 6.3 Bimodal histograms
corner separation, especially the positions of the vortex
cores, the mean velocity field in the corner region in these 6.3.1 Illustration of bimodal histograms and their
five spanwise cross-sections is also shown in Fig. 14. The decompositions
connection of the vortex cores looks like a vortex ring. This
topology of the corner separation is very similar to that Because two velocity components ux and uy were mea-
proposed by Schulz et al. (1990a). sured synchronously in the experiment, the histograms of
123
Exp Fluids (2013) 54:1546 Page 11 of 17
(ux, uy) were obtained directly. Figure 15 shows the Pðus Þ ¼ G1 ðus ; a1 ; Us1 ; r2us 1 Þ þ G2 ðus ; a2 ; Us2 ; r2us2 Þ ð10aÞ
PDFs of (ux, uy) at 5 measurement points at the mea-
surement station where s* = 0.41 and z = 5.00 mm. This where
measurement station (see in Fig. 12) is located near the " #
endwall and just downstream of the mean separation a1 ðus Us1 Þ2
G1 ðus ; a1 ; Us1 ; r2us1 Þ ¼ pffiffiffiffiffiffi exp
point in its spanwise cross-section. The main feature 2p rus 1 2r2us 1
characterizing these PDFs is the presence of one or two ð10bÞ
peaks, depending on the locations of the measurement " #
points. Only one peak exists near the endwall as the a2 ðus Us2 Þ2
G2 ðus ; a2 ; Us2 ; r2us2 Þ ¼ pffiffiffiffiffiffi exp
measurement point where n = 2.00 mm. Getting further 2p rus 2 2r2us 2
from the wall, a second peak appears and becomes more
ð10cÞ
and more important compared with the first peak, as the
measurement points where n = 5.50, 6.50, and with the requirements
10.00 mm. At last, only the second peak exists at the 8
measurement point where n = 20.00 mm. >
> a1 ; Us1 ; rus1 ; a2 ; Us2 ; rus 2 2 R;
>
>
Figure 15 shows also the corresponding PDFs of four < 0\a1 ; a2 \1:0;
a1 þ a2 ¼ 1:0; ð10dÞ
velocity components, which are simpler that 2D PDFs. >
>
> usmin \Us1 \Us2 \usmax ;
>
When two peaks appear in the 2D PDFs, two peaks also :
rus 1 ; rus 2 [ 0:
appear in the PDFs of ux, uy, and us, but not in those of un.
A familiar histogram of velocity usually has a bell- where usmin and usmax are the minimum and maximum
shaped curve and is close to the normal distribution. A values of us, respectively. The parameter a1 is the proba-
bimodal histogram often reflects that two different uni- bility of the first Gaussian distribution. Us1 and Us2 are the
modal processes are ‘‘mixed’’ in the velocity. In order to mean velocities of the first and second Gaussian distribu-
analyze this, a better solution is to decompose the bimodal tion, respectively. After the convergence of the fitting
histogram into two Gaussian distributions. Taking us as an process, the correlation coefficients for the sum of two
example, Gaussian distributions of the original histograms are above
123
Page 12 of 17 Exp Fluids (2013) 54:1546
(a) n= 2.00 mm (b) n= 5.50 mm (c) n= 6.50 mm (d) n=10.00 mm (e) n=20.00 mm
35 12 9 12 12
30 10 8 10 10
35 25 12 9 7 12 12
8 8 6 8 8
30 20 10 6 5 10 6 10 6
15 7 4
25
PDF
PDF
10 8 4 6 3 8 4 8 4
20 2 5 2 2 2
5 6 1 6 6
15 0 0 4 0 0 0
10 4 3 4 4
2
5 0.8 2 0.8 0.8 2 0.8 2 0.8
1
0 0.4 0 0.4 0 0.4 0 0.4 0 0.4
-0.4
0 0 uy/U∞ -0.4
0 0 uy/U∞ -0.4
0 0 uy/U∞ -0.4
0 0 uy/U∞ -0.4
0 0 uy/U∞
0.4 0.4 0.4 0.4 0.4
ux/U∞ 0.8 -0.4 ux/U∞ 0.8 -0.4 ux/U∞ 0.8 -0.4 ux/U∞ 0.8 -0.4 ux/U∞ 0.8 -0.4
uy/U∞
uy/U∞
uy/U∞
uy/U∞
0.4 20 0.4 0.4 5 0.4 0.4
6 6 6
0.2 15 0.2 0.2 4 0.2 0.2
4 3 4 4
0 10 0 0 0 0
2
-0.2 5 -0.2 2 -0.2 -0.2 2 -0.2 2
1
-0.4 0 -0.4 0 -0.4 0 -0.4 0 -0.4 0
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
ux/U∞ ux/U∞ ux/U∞ ux/U∞ ux/U∞
5 3 3 4 4
exp
G1
4 G2 3 3
G1+G2
2 2
3
PDF
PDF
2 2
2
1 1
1 1
1
0 0 0 0 0
-0.4 0 0.4 0.8 -0.4 0 0.4 0.8 -0.4 0 0.4 0.8 -0.4 0 0.4 0.8 -0.4 0 0.4 0.8
ux/U∞ ux/U∞ ux/U∞ ux/U∞ ux/U∞
6 3 3 3 4
5
3
4 2 2 2
PDF
PDF
3 2
2 1 1 1
1
1
0 0 0 0 0
-0.4 0 0.4 0.8 -0.4 0 0.4 0.8 -0.4 0 0.4 0.8 -0.4 0 0.4 0.8 -0.4 0 0.4 0.8
uy/U∞ uy/U∞ uy/U∞ uy/U∞ uy/U∞
4 3 3 3 3
3
2 2 2 2
PDF
PDF
2
1 1 1 1
1
0 0 0 0 0
-0.4 0 0.4 0.8 -0.4 0 0.4 0.8 -0.4 0 0.4 0.8 -0.4 0 0.4 0.8 0 0.4 0.8
us/U∞ us/U∞ us/U∞ us/U∞ us/U∞
8 5 5 5 5
7
4 4 4 4
6
5 3 3 3 3
PDF
4
3 2 2 2 2
2
1 1 1 1
1
0 0 0 0 0
-0.4 0 0.4 0.8 -0.4 0 0.4 0.8 0 0.4 0.8 0 0.4 0.8 0 0.4 0.8
un/U∞ un/U∞ un/U∞ un/U∞ un/U∞
Fig. 15 Development of the PDFs of velocity at the measurement station where z = 5.00 mm and s* = 0.41
0.995. Because the fitting depends on the assumption that confirmed by the decreases of a1 with n, as shown in
the bimodal histograms in the experiment are caused by the Fig. 16. Figure 16 also summarizes the decomposition
two modes in the flow field, the success of the fitting means parameters of velocity at this measurement station. At a
that this assumption is reasonable. measurement point, the obvious relationhips between
Figure 15 shows also the decomposition results at the velocities are Us1 \ Us \ Us2 and Un1 \ Un \ Un2, where
five measurement points at the measurement station where Us and Un are mean velocities of us and un, respectively.
z = 5.00 mm and s* = 0.41. The evolution of peaks at this This is reasonable according to the decomposition equa-
measurement station described above can also be tions (i.e., Eq. 10). When n increases, Us decreases from
123
Exp Fluids (2013) 54:1546 Page 13 of 17
14
12
n (mm)
10
bimodal
8
6
4
2
0
0 0.2 0.4 0.6 0.8 1 -0.4 -0.2 0 0.2 0.4 -0.2 0 0.2 0.5 1 1.5 2
a1 Us/U∞ Un/U∞ α
0 .9
0 .9
0 .9
0 .9
9
9
0 .8
0
0 .8
u’su’s/U ∞
2
0
0 .7
0 .7
0.
0
0
0.
60
60
0.
z=185.00 mm
0.
50
50
0.
z=40.00 mm 0.09
41
0.
31
0.08 0.
21
0.07
0.06
0.05
0.04
0.03
0.02
0 .9
0 .9
0.01
0 .9
0 .9
9
0 .8
0
0
0
0 .7
0.
0
60
z=70.00 mm
0.
50
z=20.00 mm
0 .9
0 .9
0 .9
0 .9
9
9
0 .8
0 .8
0
0
0 .7
0
0 .7
0
0
0.
60
0.
50
0.
0 .3 z=60.00 mm
41
0. 5 z=5.00 mm
0 .2 3 0
0 .2 5
1
Trips
zero on the suction side until reaches its minimum value Us2 are similar to that of Us, except that Us1 stays nearly
near n = 1.50 mm and then increases gradually. It is equal constant between 1:50 mm 6 n 6 5:00 mm. When n
to zero around n = 4.50 mm. The evolutions of Us1 and increases, Un nearly stays zero from the endwall to
123
Page 14 of 17 Exp Fluids (2013) 54:1546
0 .9
0 .9
0 .9
0 .9
9
9
0 .8
0
0 .8
0
u’nu’n/U ∞
2
0
0 .7
0 .7
0.
0
0
0.
60
60
0.
z=185.00 mm
0.
50
50
0.
z=40.00 mm
41
0.024 0.
31
0.022 0.
21
0.02
0.018
0.016
0.014
0.012
0.01
0 .9
0 .9
0.008
0 .9
0 .9
9
0 .8
0.006
0
0
0
0 .7
0.004
0.
0
60 0.002 z=70.00 mm
0.
50
z=20.00 mm
0 .9
0 .9
0 .9
0 .9
9
9
0 .8
0 .8
0
0
0 .7
0
0 .7
0
0
0.
60
0.
50
0.
0 .3 z=60.00 mm
41
0. 5 z=5.00 mm
0 .2 3 0
0 .2 5
1
Trips
n = 4.50 mm and then increases gradually. The differences critical value of a is very difficult to fix, because the
between Un1 and Un2 are always very small. As a result, the parameter may be also very large at some non-bimodal
relative positions of the two Gaussian distributions in the points, for example a = 1.24 (a very large value) at the
PDFs of un are very close, and thus the PDFs of un are not point where n = 20.00 mm, an obviously non-bimodal
bimodal. In this measurement station, bimodal points are point.
mainly located at 4:50 mm 6 n 6 10:00 mm according to
the definition of bimodal points which will be introduced 6.3.2 Positions of bimodal points
below.
In order to indicate the relative position of the two Before presenting the positions of the points with bimodal
Gaussian distributions, we define a parameter histograms, a criterion for bimodal histogram is needed. In
Us2 Us1 the literature, the description of bimodal histograms refers to
a¼ ð11Þ histograms with two peaks. This description is qualitative
rus 1 þ rus 2
and very empirical. In order to minimize the subjective
This parameter at the measurement station where z = 5.00 effects in determining the positions of the bimodal points, a
mm and s* = 0.41 is shown in Fig. 16. At bimodal points, new criterion is used here. For a histogram of velocity
the values of a are larger than 1.0. The origin objective of component, if its fitting line (i.e., the sum of the two
the definition of a is to find a parameter to indicate whether Gaussian distributions, introduced in Sect. 6.3.1) has a local
a histogram is bimodal or not (e.g., when a is larger than a minimum value, this histogram is bimodal. A point is
critical value, the histogram is bimodal). However, the denoted as bimodal if the histogram of any velocity
123
Exp Fluids (2013) 54:1546 Page 15 of 17
0 .9
0 .9
0 .9
0 .9
9
9
0 .8
0
0 .8
0
u’su’n/U 2∞
0
0 .7
0 .7
0.
0
0
0.
60
60
0.
z=185.00 mm
0.
50
50
0.
z=40.00 mm
41
0.006 0.
31
0.005 0.
21
0.004
0.003
0.002
0.001
0
-0.001
0 .9
0 .9
-0.002
0 .9
0 .9
9
0 .8
-0.003
0
0
0
0 .7
-0.004
0.
0
60
0.
50
-0.005 z=70.00 mm
z=20.00 mm -0.006
-0.007
-0.008
-0.009
0 .9
0 .9
0 .9
0 .9
9
9
0 .8
0 .8
0
0
0 .7
0
0 .7
0
0
0.
60
0.
50
0.
0 .3 z=60.00 mm
41
0. 5 z=5.00 mm
0 .2 3 0
0 .2 5
1
Trips
Us/U∞ 1
0 0.5 0 0.5 0 0.5 0 0.5 10
60
50
10
0
f-5/3
n (mm)
40
Us
E(f) (m /s)
30 Us1 -1
10
2
20 Us2
10 10
-2
ux
0 uy
s*=0.35 s*=0.41 s*=0.50 s*=0.60 us
-3 un
10
Fig. 20 Distributions of US, Us1, and Us2 at four measurement
stations s* = 0.35, 0.40, 0.50, and 0.60 in the plane z = 5.00 mm.
-4
Pink points, bimodal points 10
10-2 10-1 100 101 102 103
123
Page 16 of 17 Exp Fluids (2013) 54:1546
The positions of the bimodal points are shown with the consistent with that in the junction flow firstly investigated
mean velocity flow field in Figs. 12 and 14. These bimodal by Devenport and Simpson (1990).
points mainly appear around the mean interface of sepa- In the junction flow, the physics of bimodal phenome-
rated flow and non-separated flow. This phenomenon non is associated with the horseshoe vortex. In this
mainly reflects that the intermittence of the interface of the experiment, the horseshoe vortex is very small due to the
3D corner separation. low thickness of the blade leading edge. Additionally, the
The positions of the bimodal points are also shown for bimodal points also exist in the vicinity of the suction side
the Reynolds stresses in Figs. 17, 18, and 19. In these near the outlet. Therefore, the authors consider that the
figures, the contour lines in the spanwise cross-section physical mechanism of bimodal phenomenon in this
nearest to the endwall where z = 5.00 mm are less smooth experiment is not associated with the horseshoe vortex.
than those in other spanwise cross-sections. This is because Because the velocity field was measured by LDA point
the flow field in this spanwise cross-section is in the by point in our experiment, we cannot obtain an instant
boundary layer of the endwall, and thus, the uncertainties velocity field of the whole field from the LDA results. For
in the Reynolds stresses are larger relatively. Bimodal interpreting more accurately the physics of bimodal
points are mainly located in the region with high u0s u0s rather behavior in this experiment, extensive instantaneous results
than with u0n u0n and u0s u0n . This is because at bimodal points, are needed. The authors are currently working on the
the PDFs of us are bimodal but not those of un. numerical results of LES that could be used to investigate
further this physics.
123
Exp Fluids (2013) 54:1546 Page 17 of 17
123