Вы находитесь на странице: 1из 15

Journal of

Fluids
Engineering Technical Briefs

Experimental Study of Turbulent x⫽O-plane. Besides as far as modeling is concerned, this con-
figuration will certainly lead to equations more amenable to solu-
Asymmetric Flow in a Flat tion than in the three-dimensional case. In order to contribute to
the understanding of this flow, measurements of pressure and lon-
Duct Symmetric Sudden Expansion gitudinal velocity downstream of this singularity have been car-
ried out 关11兴.
F. Aloui
2 Experimental and Measuring Device
Faculté des Sciences et des Techniques, Dpt. de Physique-
LGP 共UPRES-EA 1152兲 2, rue de la Houssinière, Experimental arrangement, as shown in Fig. 1, is composed of
a ‘‘closed’’ liquid 共water兲 circuit regulated at 20⫾0.5°C. This
BP. 92208-44322 Nantes Cédex 03, France

M. Souhar
LEMTA 共CNRS, UMR 7563兲-ENSEM-INPL-2, av. de la
Forêt de Haye, BP. 160 54504 Vandoeuvre lès
Nancy Cédex, France

This work presents an experimental study of recirculating flows


downstream of a flat duct sudden expansion. Results of measure-
ments of average and RMS quantities of the pressure P and the
axial velocity U exhibit the asymmetry of the flow behind the
sudden expansion as in the two-dimensional case.
关S0098-2202共00兲01401-2兴

1 Introduction
Transport circuits are mainly composed of various singularities.
The case of sudden expansion corresponds to a common situation
in practice. In the case of axisymmetrical sudden expansion, the
flow is largely treated in literature both theoretically and experi-
mentally. The works of Enzo and Tin-Tan 关1兴, Ha-Minh 关2兴, Bijay
et al. 关3兴, Teyssandier and Wilson 关4兴 can be cited as examples. In Fig. 1 Schematic diagram of experimental apparatus „dimen-
two-dimensional duct sudden expansion, several works are also sions in mm…
found in the literature: Durst et al. 关5兴, Restivo and Ẃhitelaw 关6兴,
Cherdron et al. 关7兴, Fearn et al. 关8兴, Gagnon et al. 关9兴, and re-
cently Alleborn et al. 关10兴. An examination of these different
works shows that the flow in a sudden expansion, which is ini-
tially symmetrical, becomes asymmetrical downstream of the sin-
gularity for Reynolds number greater than Re1⬇35 关8兴. This
asymmetry is characterized by the existence of recirculating zones
of very different sizes.
In the case of the sudden expansion of a flat duct of the type
encountered in plate heat exchangers, turbulent flow remains quite
unknown both theoretically and experimentally. In this particular
configuration 共see Fig. 1兲, the thickness of the flow ‘‘e’’ is very
small, and the Reynolds number Re is very high such as the ve-
locity profile U(x) can be considered as almost flat 共power law兲.
Therefore, most of our measurements have been carried out in the

Contributed by the Fluids Engineering Division of THE AMERICAN SOCIETY OF Fig. 2 Three-dimensional view downstream pressure distribu-
MECHANICAL ENGINEERS. Manuscript received by the Fluids Engineering Division tion under sudden expansion for U 1 Ä3.2 mÕs „Re1Ä32,000;
October 13, 1997; revised manuscript received October 13, 1999. Associate Techni- Re2Ä14208…; uncertainty estimates: on p : 7 percent, and ⌬ ␰
*
cal Editor: J.A.C. Humphrey. ÄÁ0.01

174 Õ Vol. 122, MARCH 2000 Copyright © 2000 by ASME Transactions of the ASME

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 3 The RMS pressure contours „ln Pa… downstream of the sudden expansion
for U 1 Ä3.2 mÕs, ⌬ p Ä50 Pa, „Re1Ä32,000; Re2Ä14,208…; uncertainty estimates: RMS
of p: 6 percent, and ⌬ y ÄÁ0.01
*

Fig. 4 Axial velocity profiles and the axial fluctuation rate be-
fore the sudden expansion for U 1 Ä3.2 mÕs at ␰ ÄÀ0.36; „Re1
Ä32,000; Re2Ä14,208…; uncertainty estimates: Ū : 2 percent,
and on u Õ U : 12 percent

Fig. 6 „a… Separation boundaries between principal and recir-


culation flow. „b… General reorganization of flow downstream of
the sudden expansion for U 1 Ä3.2 mÕs, „Re1Ä32,000; Re2
Ä14,208…; uncertainty estimates on Fig. 6„a…: 12 percent

circuit feeds a test section with rectangular cross section of 44


⫻5 mm2 and 100⫻5 mm2 upstream and downstream of the ex-
pansion, respectively. The liquid flow rates were measured by a
diaphragm flowmeter.
The experimental test section was made using transparent plexi-
glass and has been largely described in 关12兴. This test section has
a 460 mm long with a rectangular cross section 100⫻5 mm2
downstream of the singularity. The step height ‘‘h’’ is 28 mm and
the ratio of cross section, upstream and downstream, is ␴
⫽0.444. The upper wall of the test section, composed of a sliding
plate, is equipped with two openings that house the interchange-
able measuring blocks. With the aid of hot film probe support on
the measuring block that allows a transversal displacement 共along
y兲 and a rack-and-pinion, which guides the sliding plate longitu-
dinally 共along z兲, it is possible to reach every point in the expan-
sion. The pressure measuring block has 17 pressure tappings uni-
formly distributed along y and are connected to a differential
pressure sensor 共Validyne DP45兲. The measurement of axial ve-
locity was made by a hot film probe 共DISA 155M01 and 1210-20
W hot film兲. The signals obtained for the different measured quan-
Fig. 5 Axial velocity profiles downstream of the sudden ex- tities were processed on a UNIX-HP 9360 station. For each flow
pansion, for U 1 Ä3.2 mÕs „Re1Ä32,000; Re2Ä14,208…; uncer- rate, a fine grid consisting of 720 measurement points spread uni-
tainty estimates; Ū Õ U 1 4 percent, and ⌬ y ÄÁ0.2 mm formly downstream of the sudden expansion was used.

Journal of Fluids Engineering MARCH 2000, Vol. 122 Õ 175

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 7 RMS-u contours „in mÕs… downstream of the sudden expansion
„⌬ U Ä0.05 mÕs… for U 1 Ä3.2 mÕs; „Re1Ä32,000; Re2Ä14,208…; uncertainty estimates:
10 percent

3 Results and Discussions the whole flow up to end of the expansion ( ␰ ⫽15). In both sides
of the expansion, the recirculating zones have different sizes and
By injection of very little quantities of small gas bubbles, and
geometries. Using the different U-profiles, the different jet lines
by using a reflex camera, photographs of the flow were taken at
that separate the main flow from the recirculating zones are deter-
different liquid velocities 共3.2 to 7 m/s兲. An example of the pho- 1 /2 ⫺1 /2
tograph is given in 关12兴 and shows an asymmetry of the flow mined by: 兰 y 2 Ū(y).dy⫽0 and 兰 ⫺y 2 Ū(y).dy⫽0 where y j and
j ⬘j
downstream of the sudden expansion. Furthermore, a fast speed y ⬘j are the distances which separate the center line of the duct from
camera 共Ektapro 1000 frames/s兲 has been used to determine the
the jet line. The results are shown in Fig. 6共a兲. We note that the
direction of the flow by examining movements of bubbles in all
two principal jet lines on both sides of the axis of the expansion
the area explored. This direction will serve to determine the sign
are not symmetrical, and there is a very small stationary vortex
of U-longitudinal velocity measured by the one hot film compo-
located between ␰ ⫽6.5 and ␰ ⫽8.5 共at the side of the small prin-
nent. In the following, ␰ indicates the axial dimensionless length
cipal vortex兲. This small vortex is created by the takeoff of the jet
␰ ⫽z/h and Re the Reynolds number: Rei⫽UiDH /␯ where U i is
when it reflectes on the lateral wall at about ␰ ⫽5 to 6.5. The
the average velocity of the flow, D H the hydraulic duct diameter,
u-RMS fluctuations given in contour plots in Fig. 7 also exhibit
and ␯ the kinematic viscosity. The indices i 共i⫽1 or 2兲 refer to
the asymmetry of the flow, and pass from a low value in the core
upstream and downstream, respectively.
of flow to reach high values in regions of recirculating flow. The
The average pressure P̄ and the RMS value p were measured vizualization and the examination of the results of pressure and
and the results are presented in 3D form in Fig. 2. In this figure, axial velocity component enabled us to propose a diagram 共Fig.
the average dimensionless pressure p * is defined by: p * 6共b兲兲 of the general organization of the turbulent flow downstream
⫽( P̄⫺ P atm)/(0.5 ␳ U 21 ), where P atm is the atmospheric pressure of the flat duct sudden expansion. The regions of recirculating
and ␳ the density of liquid. By examination of Fig. 2, it can be flow are constituted by five vortices of different sizes. On one
seen that the distribution of the p * is not symmetrical about the side, the first vortex region contains two vortices of which one is
axis of the sudden expansion (x⫽y⫽0). Such situations had al- about twice the size of the other. The second region is composed
ready been observed by numerous authors on velocity profiles. In of three vortices with the smallest one located far from the other
addition, we can note that the distribution of pressure decreases two, outside the principal vortex region.
then increases to reach a plateau when one moves further away
from the singularity. The RMS of p were also measured, and the 4 Conclusion
equi RMS contour plots of the ⫺p are given in Fig. 3. This
contour plot reflect the time-averaged streamlines of the different To our knowledge, the flow in flat duct sudden expansion has
vortices and have an order of magnitude of the wall shear stress not made the object of study contrary to an axisymmetrical or
␶ W by using ␶ W ⬇C"p where C is approximatively 0.3 关13兴. two-dimensional case where many works can be found in the
The voltage signal delivered by hot film was recorded on the literature. This experimental study fulfills this gap and shows that
HP station and suitably linearized by a computer program to give the flow is asymmetric as in a two-dimensional case. It also brings
interesting data for numerical simulation for this configuration.
the average velocity Ū and the standard deviation u. The results of
These data concern the average and the RMS of the pressure and
Ū and u upstream of the expansion ( ␰ ⫽⫺0.36) are shown in Fig. the longitudinal velocity, and an order of magnitude of the wall
4. It can be seen that the main flow is symmetrical and the veloc- shear stress.
ity profile remain quasi-flat in the entire section. The fluctuations
rates are the same order of magnitude as the usual fluctuations
rates found in the literature. By integrating the Ū-profile and by References
using the power law U(x)/Ū⫽ 关关 (e/2)⫺x 兴 /(e/2) 兴 1/7, the operat- 关1兴 Enzo, O. M., and Tin-Tan, H. 1967, ‘‘Computational and Experimental Study
of a Captive Annular Eddy,’’ J. Fluid Mech., 28, Part 1, pp. 43–64.
ing liquid flow rate U 1 is obtained with a small deviation 共about 关2兴 Ha-Minh, H., 1970, ‘‘Decollement provoque d’un écoulement turbulent in-
5%兲. This constitutes a good control of the processing procedure compressible,’’ Thèse de Doctorat és-Sciences, Toulouse.
for the signal obtained from the hot film probe. 关3兴 Bijay, K. S., Paul, G. N., and Darryl, E. M., 1991, ‘‘Turbulent Flow Prediction
The axial component of the velocity downstream of the sudden in a Sudden Axisymmetric Expansion,’’ Turbulent Shear Flow 7, Seventh
International Symposium, Stanford University, pp. 655–663.
expansion was measured at every point 共720 points兲. These mea- 关4兴 Teyssandier, R. G., and Wilson, M. P. 1974, ‘‘An Analysis of Flow Through
surements are given in 3D form in Fig. 5. At the inlet of the Sudden Enlargements in Pipes,’’ J. Fluid Mech., 64, Part 1, pp. 85–95.
expansion ( ␰ ⫽0.07), the flow is perfectly symmetrical. This be- 关5兴 Durst, F., Melling, A., and Whitelaw, J. H. 1974, ‘‘Low Reynolds Number
comes progressively asymmetrical further away from the expan- Flow Over a Plane Symmetrical Sudden Expansion,’’ J. Fluid Mech., 64, Part
1, pp. 111–128.
sion. The axial velocity profiles are flat at the inlet and become 关6兴 Restivo, A., and Whitelaw, J. H., 1978, ‘‘Turbulence Characteristics of the
curved toward a lateral side of the sudden expansion, thereby Flow Downstream of a Symmetric Plane Sudden Expansion,’’ J. Heat Trans-
leading to asymmetry of the flow. The latter remains apparent in fer, 100, Sept., pp. 308–310.

176 Õ Vol. 122, MARCH 2000 Transactions of the ASME

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
关7兴 Cherdron, W., Durst, F., and Whitelaw, J. H., 1978, ‘‘Asymmetric Flows and regions of low energy flow. For example, boundary layer suction
Instabilities in Symmetric Ducts with Sudden Expansions,’’ J. Fluid Mech.,
84, Part 1, pp. 13–31.
or blowing, vortex generator, vanes or flow control rails installed
关8兴 Fearn, R. M., Mullin, T., and Kliffe, K. A. 1990, ‘‘Nonlinear Flow Phenomena on the surface, introduction of swirl or turbulence at the inlet,
in a Symmetric Sudden Expansion,’’ J. Fluid Mech., 211, pp. 595–608. control of wall contour and passive boundary layer control with a
关9兴 Gagnon, Y., Giovannini, A., and Hebrard, P., 1993, ‘‘Numerical Simulation combination of both suction and blowing.
and Physical Analysis of High Reynolds Number Recirculating Flows Behind
Sudden Expansions,’’ Phys. Fluids A, Fluid Dynamics, 5, No. 10, pp. 2377–
The concept of passive boundary layer control method was
2389. originally developed for drag reduction and buffet alleviation in
关10兴 Alleborn, N., Nandakumar, K., Raszillier, H., and Durst, F., 1997, ‘‘Further transonic flow 关7兴. It was based on simultaneous suction and
Contributions on the Two-Dimensional Flow in a Sudden Expansion,’’ J. Fluid blowing achieved by passive means by linking a low-pressure
Mech., 330, pp. 169–188.
关11兴 Aloui, F., 1994, ‘‘Etude des écoulements monophasiques et diphasiques dans
region with a high pressure region by a porous surface and a
les élargissements brusques axisymétrique et bidimensionnel,’’ Thèse de Doc- plenum underneath. It has been shown that the passive control can
torat de l’INPL, Nancy. also alleviate buffet excitation.
关12兴 Aloui, F., and Souhar, M., 1996, ‘‘Experimental Study of Two-Phase Bubbly In a diffuser there exists a natural pressure rise in the direction
Flow in a flat duct symmetric sudden expansion-Part 1 : Visualization, pres-
sure and void fraction,’’ Int. J. Multiphase Flow, 22, No. 4, pp. 651–665.
of flow, which could be used to achieve simultaneous suction and
关13兴 Hinze, J. O., 1975, Turbulence, McGraw-Hill, New York. blowing. Indeed, experiments with relatively long diffusers and
with slots normal to the flow have shown 关5兴 that passive control
can produce modest improvements in pressure recovery and sig-
nificant reductions in flow unsteadiness. However, with passive
Passive Boundary Layer Control With boundary layer control by porous surfaces made of normal holes
or slots the extent of improvements in the performance of the
Slots in Short Diffusers diffuser is limited. This is due to the fact that blowing normal to
the surface thickens the boundary layer. A thick boundary is sen-
S. Raghunathan sitive to adverse pressure gradients and tends to separate down-
stream. To some extent the thickening of the boundary layer can
Professor be reduced by blowing through holes or slots, which are inclined
in the direction of the main flow. Equally, suction holes or slots,
R. K. Cooper which are inclined and facing the main flow, can enhance the
effectiveness of suction downstream. This paper presents results
Senior Lecturer of the application of such concepts to short wide-angle
diffusers.
School of Aeronautical Engineering,
Queen’s University, Belfast, United Kingdom
e-mail: s.raghunathan@qub.ac.uk Experiments
Experiments were conducted on a two-dimensional diffuser
with a smooth contoured inlet 共Fig. 1兲 installed at the end of a 1.2
m long duct with a diffuser inlet contraction ratio of 6. The air
Experiments were conducted in short wide-angle diffusers. The
supply to the diffuser was from a centrifugal fan. The diffuser
control surface was made of slots and a plenum chamber under-
entry had a width of 25 mm, inlet height to width 共aspect ratio兲 of
neath and on both diverging walls of the diffusers. The experi-
6, and length to width ratio of 2.71. The inlet contour was made of
ments included slots normal to the surface and a combination of
cylindrical blocks to which the diverging walls constructed of
slots normal and inclined to the surface. Passive control with
steel plates were attached. With this arrangement, the symmetry at
inclined slots produced modest improvement in pressure recovery
the inlet of the diverging walls was maintained for all diffuser
but significant increases in the stall angle.
angles. The top and bottom nondiverging walls of the diffuser
关S0098-2202共00兲01501-7兴
were made of Perspex sheets. Both diverging walls had 26 evenly
spaced pressure orifices along the centerline of the wall starting
from the throat. Pressure orifices were also located on the parallel
Introduction walls. The diverging side walls of the diffuser also had flush
Diffusers 关1兴, which convert kinetic energy to potential energy mounted Kulite transducers 共type XC6-062兲 at x/L⫽0.9, where x
in the form pressure, are key elements of several fluid systems was the distance measured from the diffuser entry and L was the
such as aircraft engine intakes, combustors, wind tunnels, flow length of the diffuser.
meters, centrifugal compressors, carburetors, and noise suppres-
sors. The performance of a fluid system downstream of diffuser,
for example, the combustion chamber in the case of aircraft en-
gines, depends critically on the diffuser performance. Therefore, a
well-designed diffuser should produce not only a maximum pres-
sure recovery, but also minimum levels to total pressure loss, flow
distortion, and unsteadiness. In practice, short wide-angle diffus-
ers are often used to save space and reduce cost. Such diffusers
have poor performance, due to large boundary-layer separation
共stall兲, asymmetry of flow at the exit, and large unsteady pressure
fluctuations 关1–3兴, all of which are also detrimental to the perfor-
mance of the flow system downstream.
There are several methods by which the onset of stall in the
diffuser can be postponed 共See, for example兲 关4–6兴. These meth-
ods are primarily based on transferring kinetic energy into the

Contributed by the Fluids Engineering Division of THE AMERICAN SOCIETY OF


MECHANICAL ENGINEERS. Manuscript received by the Fluids Engineering Division
October 5, 1998; revised manuscript received September 7, 1999. Associate Techni-
cal Editor: D. R. Williams. Fig. 1 Schematic of diffuser

Journal of Fluids Engineering MARCH 2000, Vol. 122 Õ 177

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 4 Pressure fluctuations at x Õ I Ä0.9

Passive controls were located symmetrically on both diverging


walls. Three types of slot arrangements were selected which all
had slots located at x/L⫽0.15 and 0.6 as shown in Fig. 2. The
configurations tested included diffuser with solid walls 共case 1兲,
slots 1 mm wide and normal 共90 degrees兲 to the surface 共case 2兲,
a combination of slots inclined 90 degrees upstream 45 degrees
downstream 共case 3兲, and slots 2 mm wide and both inclined at 45
degrees to the surface 共case 4兲. The slot width of 2 mm was of the
order of the boundary layer thickness at the inlet. The 2 mm wide
slots resulted in porosity ratio based on the open to closed surface
area of each diverging wall of 2.96%. The static pressures on the
four walls of the diffuser and dynamic pressures at specific loca-
tions on the diffuser walls were measured with scani-valve and
Kulite transducers, respectively. A pitot tube with an opening of
0.2 mm height and 1 mm width was used for measurement of
velocity profiles at the inlet.

Results and Discussion


Fig. 2 Diffuser slot arrangements When compared to earlier work on long diffusers 关5兴 the
present experiments on short diffusers with normal slots 共case 2兲
showed insignificant changes in pressure recovery and stall angle
and therefore, the results of the datum 共case 1兲 and inclined slots
The boundary layer at the inlet was turbulent with an inlet 共cases 3 and 4兲 only are presented here.
blockage 2 ␦ * /W 1 ⫽0.015 where ␦* was boundary layer displace- Figure 3 shows the typical effect of passive control in pressure
ment thickness and W 1 inlet width. This value was comparable to recovery C PR in the diffuser. Here C PR is given by (p e ⫺p i )/q,
the tests of Waitman et al. 关3兴. The Reynolds number based on the where q⫽1/2␳ V i2 and where ␳ is density, V i is velocity, and i and
diffuser width was 0.12⫻106 . The turbulence levels at the inlet e refer to inlet and outlet, respectively. The maximum pressure
measured by a DISA 55M hot-wire anemometer with a 55P14 recovery for the solid wall 共case 1兲 is 0.55 and the stall angle is 15
probe was 0.61%. degrees, which are comparable to results of earlier works
关1,2,4,7,5兴. The general effect of passive control is to reduce the
pressure recovery at small divergence angles but postpone stall.
Configurations with a combination of 45 and 45 degrees wide
slots 共case 4兲 produced a significant postponement of stall to 20
degrees, an increase of over 30%, and a modest improvement in
the pressure recovery. The configurations with 90 and 45 degree
slots 共case 3兲 did not produce any improvement in pressure
recovery.
The variation of nondimensional root mean square pressure
fluctuation levels p̄/q with 2␪ for the datum and case 4 are com-
pared in Fig. 4. The results shown here are the average value of

Fig. 3 Variation of overall pressure recovery with diffuser


angle for various slot configurations Fig. 5 Typical spectra of pressure fluctuations

178 Õ Vol. 122, MARCH 2000 Copyright © 2000 by ASME Transactions of the ASME

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
pressure fluctuations measure on both diverging walls, at x/I A Numerical Study of Vortex
⫽0.9. The onset of stall for both the datum and the slotted diffus-
ers can be observed from these measurements. The measured val- Breakdown in Turbulent Swirling
ues p/q on divergence walls of the datum diffuser with attached
boundary layers 共2 ␪ ⬍14 degrees兲 are in close agreement the val-
Flows
ues quoted in literature for a flat plate boundary layer of p̄/q
⫽0.006. The measured values of p̄/q in the datum diffuser for Robert E. Spall
2 ␪ ⬎14 degrees where the boundary layer is separated agree with
that measured in low speed separated flow 关8兴. The results also Associate Professor
agree with those measured in two-dimensional diffusers 关9兴. It is Mem. ASME
clearly seen that the effect of passive control is to not only post-
pone the onset of separation but also reduce buffet excitation. For Blake M. Ashby
2 ␪ ⬎30 degrees the passive control is ineffective in reducing buf-
fet levels. Research Assistant
Typical spectra of pressure fluctuations in the form of decibels Student Mem. ASME
共dB兲 with an arbitrary reference plotted against nondimensional
frequency n given by f L/V i where f is the frequency of pressure Department of Mechanical and Aerospace Engineering,
fluctuation are shown in Fig. 5. The data shown here are for dif-
fuser divergence of 2 ␪ ⫽20 degrees, which corresponds to a sig-
Utah State University,
nificant separation on the walls of the solid wall diffuser. Maxi- Logan, UT 84322-4130
mum pressure fluctuations occur on the solid wall diffuser at n
0.6–0.7. These values of n are comparable to those measured in
separated flow 关8兴. It can be observed that with passive control Solutions to the incompressible Reynolds-averaged Navier–Stokes
buffet excitation levels are reduced over a wide range of equations have been obtained for turbulent vortex breakdown
frequencies. within a slightly diverging tube. Inlet boundary conditions were
Kline 关1兴 describes the mechanism of stall in a diffuser. It is derived from available experimental data for the mean flow and
understood that maximum pressure recovery occurs in a diffuser turbulence kinetic energy. The performance of both two-equation
where large transitory stall first starts to appear, i.e., slightly above and full differential Reynolds stress models was evaluated. Axi-
the appreciable stall line. It is suggested here that with passive symmetric results revealed that the initiation of vortex breakdown
control the upstream porous surface succeed in preventing the was reasonably well predicted by the differential Reynolds stress
large areas of stall moving downstream. This in turn delays the model. However, the standard K – ␧ model failed to predict the
onset of large transitory stall. The downstream porous surface occurrence of breakdown. The differential Reynolds stress model
reduces the appearance of these large areas of stall by reducing also predicted satisfactorily the mean azimuthal and axial velocity
the thickness of the boundary layer at this location. The reduced profiles downstream of the breakdown, whereas results using the
separation may also reduce the effective area of ratio of the dif- K – ␧ model were unsatisfactory.关S0098-2202共00兲01601-1兴
fuser and so prevent large areas of stall from developing on the
opposite wall by containing their growth.

Introduction
Conclusions The majority of numerical work concerning vortex breakdown
may be divided into two categories: 1兲 those for both confined and
Preliminary experiments were conducted on a two-dimensional unconfined flows containing inflow/outflow boundaries and 2兲
diffuser with a porous surface made of slots and cavity underneath those that consider flows within cylindrical containers driven by
the diverging walls. It can be concluded from these experiments rotating endwalls. In terms of category 共1兲, which represent break-
that passive control with inclined slots can produce a modest im- down flows most similar to those found in practical applications,
provement in pressure recovery but significant increases in the the majority of the existing numerical work investigates low Rey-
stall angle. Further investigations are necessary in order to opti- nolds number, laminar, axisymmetric 关1兴 or three-dimensional
mize geometry and position of the slots and to understand the flows 关2–4兴.
scale effects on the effectiveness of passive control. One of the primary contributions of these numerical works has
been to provide insight into the internal structure of vortex break-
down. However, in most technological applications, the vortex
References breakdown arises within a turbulent swirling flow, and the appli-
cability of laminar results to turbulent breakdown is questionable.
关1兴 Kline, S. J., 1959, ‘‘On the Nature of the Stall,’’ ASME J. Basic Eng., 81, p.
305. In fact, Sarpkaya 关5兴 and Sarpkaya and Novak 关6兴 presented ex-
关2兴 Kline, S. J., Abbott, D. E., and Fox, R. W., 1959, ‘‘Optimum Design of perimental results for vortex breakdown in noncavitating, high
Straight Walled Diffusers,’’ ASME J. Basic Eng., 81, pp. 321–331. Reynolds number 共up to 225,000兲 swirling flows and considered
关3兴 Waitman, B. A., Reneau, L. B., and Kline, S. J., 1961, ‘‘Effect of Inlet Con- the resulting ‘‘conical’’ breakdown fundamentally distinct from
ditions on the Performance Diffusers,’’ ASME J. Basic Eng., 83.
关4兴 Holzhauser, C. A., and Hall, L. P., 1956, ‘‘Explanatory Investigation of the the various forms of laminar breakdown. Using high speed pho-
Use of Area Suction to Eliminate Airflow Separation in Diffusers Having tography 共exposure times on the order of 6 ns兲, Sarpkaya and
Large Expansion Angles,’’ NACA TN 3793. Novak 关6兴 showed that the conical form of the mean flow is ac-
关5兴 Raghunathan, S., McIlwain, S. T., and Mabey, D. G., 1991, ‘‘Wide Angle tually a consequence of a very rapid precessing of the vortex core
Diffusers with Passive Boundary-Layer Control,’’ Aeronaut. J., 95, No. 941,
pp. 28–34. slightly displaced from the tube centerline. Flow visualization also
关6兴 Wong, H. M., and Dowling, A. P., 1994, ‘‘Active Boundary Layer Controls in revealed the existence of a stagnation point and very short region
Diffusers,’’ AIAA J., 32, No. 12, pp. 2409–2413. of reversed flow. These high Reynolds number breakdowns ap-
关7兴 Raghunathan, S., 1988, ‘‘Passive Control of Shock Waves Boundary Layer pear to be the most robust of all the breakdown forms.
Interaction,’’ Prog. Aeronaut. Sci., 25, pp. 271–296.
关8兴 Mabey, D. G., 1978, ‘‘Prediction of Severity of Buffeting,’’ AGARD-LS-94,
pp. 7.1–7.30. Contributed by the Fluids Engineering Division of THE AMERICAN SOCIETY OF
关9兴 Smith, C. R., and Layne, J. L., 1979, ‘‘An Experimental Investigation of Flow MECHANICAL ENGINEERS. Manuscript received by the Fluids Engineering Division
Unsteadiness Generated by Transitory Stall in Plane Wall Diffuser,’’ ASME J. November 23, 1998; revised manuscript received November 2, 1999. Associate
Fluids Eng., 101, p. 181. Technical Editor: U. Ghia.

Journal of Fluids Engineering MARCH 2000, Vol. 122 Õ 179

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
The only existing numerical work appearing in the literature strongly swirling flows, leading to an overly rapid decay of maxi-
aimed at studying turbulent vortex breakdown 共outside of the mum swirl velocities. Nevertheless, K⫺␧ model results are in-
combustion community, where the breakdown is referred to as a cluded, in part, to serve as a point of reference. The advantage of
central toroidal recirculation zone, and where geometries typically Reynolds stress models is that they inherently account for the
include such complicating factors as dilution jets and rapid expan- effects of streamline curvature, body forces, and rotation. In the
sions, cf. 关7兴兲 are those of Bilanin et al. 关8兴 and Spall and Gatski present work, the DRSM closure assumptions are based upon the
关9兴. The work of Bilanin et al., in which the Reynolds-averaged work of Launder et al. 关13兴 and Gibson and Launder 关14兴. In
Navier–Stokes 共RANS兲 equations were solved, followed much particular, the pressure-strain term, ␾ i j , consists of the linear
along the lines of the steady, laminar axisymmetric calculations of return-to-isotropy model for the ‘‘turbulence’’ portion as,

冉 冊
Grabowski and Berger 关1兴. Spall and Gatski also solved the
RANS equations 共employing the algebraic stress model of Gatski ␧ 2
␾ i j1 ⫽⫺C 1 u ⬘u ⬘⫺ ␦ K (1)
and Speziale 关10兴兲, presenting results for the unsteady, 3D turbu- K i j 3 ij
lent breakdown of an unconfined longitudinal vortex. Their results and the isotropization of production model for the mean-strain
showed some qualitative agreement with Sarpkaya’s experimental part,
results 共i.e., relative steadiness of the mean flow, robustness, and a
lack of asymmetries兲, but in the absence of common lateral and
inflow boundary conditions, a closer comparison between the re-
sults was deemed unwarranted.
冉 1
␾ i j2 ⫽⫺C 2 P i j ⫺ ␦ i j P kk ⫺S i j
3 冊 (2)

Several distinct approaches to modeling these high Reynolds 共where ⌽ i j ⫽ ␾ i j1 ⫹ ␾ i j2 ⫹ ␾ iwj ; P i j is the production term, S i j a
number breakdown flows exist, ranging from solutions to the curvature related source term, and ␾ iwj the wall reflection term兲.
Reynolds-averaged Navier–Stokes equations, to large eddy simu- The range of values for C 1 and C 2 employed in the past are
lations 共LES兲, to direct numerical simulations 共DNS兲. However, to usually defined by the relation (1⫺C 2 )/C 1 ⫽0.23, with the most
date, DNS and LES simulations utilizing spectral schemes have commonly assigned values 共and those employed in this work兲
been limited primarily to geometrically simple configurations, given as 1.8 and 0.60, respectively. In addition, the dissipation
and, for the case of DNS, the restrictions include relatively low term was modeled by an isotropic dissipation rate, while diffusion
Reynolds numbers. Higher-order finite-difference techniques en- was modeled by a gradient approximation which accounts for di-
joy more flexibility in terms of geometries and boundary condi- rectional diffusivity 关15兴.
tions, but with the lack of spectral accuracy, their built-in low- For each of the models, standard equilibrium wall functions
pass filter may tend to confuse the issue of resolved scales versus were used to implement the duct wall boundary conditions. This
subgrid-scale motions in LES calculations. Consequently, in the eliminated the need for an overly fine grid near the wall which is
near future it is unlikely that LES or DNS approaches will be far removed from the primary area of interest—the vortex core
available as computational tools to be utilized in the investigation region.
of most technologically important applications of high Reynolds
number swirling flows for which breakdown may occur; for ex- Geometry and Boundary Conditions
ample, flows within Ranque-Hilsch tubes.
The present study represents an initial effort to assess the suit- The course of the numerical study has been guided by experi-
ability of the RANS equations for predicting vortex breakdown in mental data provided to these investigators by Professor Sarpkaya
high Reynolds number confined swirling flows. The geometry is 关6,16兴. In particular, the geometry represents a model of the ex-
modeled after the axisymmetric diverging tube test section em- perimental diverging tube apparatus of Sarpkaya 关5兴 and Sarpkaya
ployed in the experimental work of Sarpkaya 关5兴 and Sarpkaya and Novak 关6兴. The diverging portion of the computational do-
and Novak 关6兴. Several features contribute to the complex nature main extends from x⫽0 to x⫽7.25r 0 , where r 0 is the tube radius
of this flow, including the existence of a mild adverse pressure at the inlet boundary. A section of constant radius, r⫽1.154r 0 ,
gradient 共diverging tube兲, strong streamline curvature, and an in- then extends to x⫽41.6r 0 where zero gradient outflow boundary
ternal separation point 共at breakdown兲. This geometry is some- conditions were imposed.
what simplified from that of, for instance, most combustors; how- Experimental data, in terms of the mean flow and turbulence
ever it represents an excellent test case to ascertain the capabilities kinetic energy profiles, were utilized to derive the inlet boundary
of the RANS equations and associated closure models for strongly conditions. The inlet distributions for the mean velocities were
swirling flows. specified through least squares fit to the experimental data in the
form u(r)⫽A⫹Be ( ⫺ ␣ r ) and w(r)⫽⌫/r(1⫺e ⫺ ␤ r ). A func-
2 2

tional form identical to that of the axial velocity was employed for
Numerical Method and Turbulence Models the distribution of turbulence kinetic energy.
The pressure-based finite-volume code Fluent 共Fluent, Inc., Specification of the inlet turbulence quantities, and, in particu-
Lebanon, NH兲 has been utilized to solve both axisymmetric and lar, the dissipation rate, is problematic. The only experimental
three-dimensional formulations of the Reynolds-averaged data available to the authors was the radial distribution of the
Navier–Stokes equations. For the three-dimensional calculations, turbulence kinetic energy. Hence, in the case of the DRSM calcu-
cylindrical-polar velocity components were employed to reduce lations, the stress distribution was assumed isotropic, with the
numerical diffusion. Interpolation to cell faces for the convection normal stresses taken as 2k/3. One may approximate the dissipa-
terms was performed using a bounded QUICK 关11兴 scheme; tion rate, ␧, via an eddy-viscosity hypothesis when strain rates and
second-order central differencing was used for the viscous terms. Reynolds stresses are known. However, the stresses were unavail-
Pressure-velocity coupling was based upon the SIMPLE proce- able, and in addition, the eddy viscosity is likely to be highly
dure 关12兴. Solutions obtained using the segregated solver were anisotropic. Hence, the alternative employed was to specify the
considered converged when the residual for each of the equations dissipation rate at the inlet through a relation of the form ␧
共based on an L2 norm兲 was reduced by a minimum of four to five ⫽CK 3/2/l where l is the vortex core radius, and C represents the
orders of magnitude. proportionality factor that needs to be specified. If one specifies C
Both differential Reynolds stress 共DRSM兲 and standard K⫺␧ such that the ratio of turbulent to laminar viscosity is approxi-
models have been employed for turbulence closure. It is well mately 200 at the vortex centerline, then the relation appears to
known that in regions of high strain rate the standard K⫺␧ model approximate reasonably well dissipation rate distributions deter-
produces excessive levels of turbulence kinetic energy, leading to mined through analysis of the experimental data 关16兴.
high values of the isotropic turbulent viscosity. As a result, the To assess the effect of the inlet dissipation rate on the solutions,
model may overpredict the radial diffusion of momentum in calculations were also made for levels of inlet dissipation at twice

180 Õ Vol. 122, MARCH 2000 Copyright © 2000 by ASME Transactions of the ASME

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 1 Computational grid in the diverging section of the tube
„every fourth grid point plotted in each direction…

and one-fifth the value given above. In terms of the DRSM, the
resulting mean flow profiles varied only slightly from those pre-
sented in the results section. Differences were slightly greater for
the K⫺␧ model, although subsequent conclusions regarding the
ineffectiveness of the K⫺␧ model were not altered. In addition, it
should also be noted that varying ␧ at the inlet boundary also has
the effect of altering the effective viscosity through the relation-
ship v t ⫽(C ␮ K 2 )/␧. This parameter plays a more significant role
in the two-equation formulation, and consequently it is not unex-
pected that the K⫺␧ results were somewhat more sensitive to
variations in the inlet dissipation rate. The sensitivity of the flow
to a change in mean inlet velocity was also investigated by in-
creasing the mean inlet axial velocity by 10% 共effectively increas-
ing the Reynolds number to 143,000兲. This was done to assess the
sensitivity of the axial location of the stagnation point 共or axial
velocity minimum兲 to inlet axial velocity; solutions revealed no
significant change from those presented in the Results section.

Results
The axisymmetric calculations were performed on a 332 共axial兲
⫻143 共radial兲 grid with significant stretching toward the duct cen-
terline, duct outer wall, and axially near the breakdown region.
Fig. 3 Comparison of mean azimuthal velocity between model
The diverging portion of the computational domain is shown in
predictions and experimental data †6,16‡. „a… x Ä5.0; „b… x Ä8.3
Fig. 1, where for purposes of clarity, every fourth grid point is
plotted 共in each direction兲. Note due to the large aspect ratio
(L/D⫽41.6) only the diverging portion of the tube is shown. At
the inlet, approximately 30 grid points were contained within the experimental data for the centerline axial velocity over the length
vortex core radius 共defined as the radius of maximum swirl veloc- of the tube. As revealed in the figure, the DRSM results predict
ity兲. A grid resolution study, performed by doubling the number the occurrence of a stagnation point and region of reversed flow of
of grid points in the radial direction, did not significantly alter the very limited extent. This may be contrasted to the K⫺␧ model
results that follow. In addition, calculations for the results that results, which are far from predicting a stagnation point. Although
follow were performed at a Reynolds number, based upon mean relevant experimental data points are not available, flow visualiza-
axial velocity and tube diameter at the inlet, of 130,000. In the tion also indicated the existence of a stagnation point between
figures to follow, lengths have been made dimensionless with re- 2.5⬍x⬍4.0 关5,16兴. Each set of results predicted a drop in axial
spect to the tube radius at the inlet, and velocities with respect to velocity at a rate which exceeded that observed experimentally.
the mean axial velocity at the inlet. However, the minimum reached using the DRSM was consider-
Figure 2 provides a comparison between numerical results and ably below that obtained using the K⫺␧ model. To the extent
possible, one would like to avoid these rapid decelerations near
the inlet boundary; however, the location of available experimen-
tal data precluded placing the boundary further upstream. In terms
of centerline axial velocities then, neither of the models was able
to predict with great accuracy the behavior of the mean flow near
the region of minimum centerline axial velocity. Downstream of
the velocity minimum, the DRSM predicted a slower rate of re-
covery of the centerline axial velocity than did the K⫺␧ model, a
trend that appears consistent with the limited experimental data
available. We further explore the regions downstream of break-
down in the figures that follow. In particular, mean axial and
azimuthal velocities are examined and compared with experimen-
tal results at two locations downstream of breakdown. In these
remaining figures, the experimental data points represent the av-
erage of values taken at ⫾r, for a given axial location, the aver-
aging being performed due to small asymmetries in the experi-
mental profiles.
Mean swirl profiles at x⫽5.0 and x⫽8.3 共downstream of the
Fig. 2 Comparison between experimental data of Sarpkaya breakdown location兲 are shown in Figs. 3共a兲 and 3共b兲, respec-
and Novak †6‡ and Sarpkaya †16‡ and model predictions of cen- tively. Both numerical and experimental data indicate the exis-
terline mean axial velocity tence of a solid body rotation near the vortex centerline. The

Journal of Fluids Engineering MARCH 2000, Vol. 122 Õ 181

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 5 Contours of mean axial velocity for fully three-
dimensional, unsteady calculations

tence of a strong wake region within the core. On the contrary, the
DRSM clearly indicates the existence of a strong wake, the mag-
nitude of which is similar to that observed experimentally 共taking
into the account the possibility of a lateral translation due to a
difference in the origin of the breakdown兲. The rate of recovery of
the DRSM does appear to be somewhat greater than that observed
experimentally. However, Fig. 1 indicates that the rate of recovery
predicted using the DRSM decreases considerably for x⬎7.
Again, the results obtained by doubling the grid are essentially
unchanged.
Undoubtedly, variations of the basic turbulence models em-
ployed would result in slightly different predictions. However,
based upon the limited set of calculations presented herein, the
results indicate that DRSM formulations may be capable of pre-
dicting the mean flow behavior of strongly swirling flows inherent
in vortex breakdown with accuracy that may be acceptable for
certain engineering-type applications. The results do not support
the same conclusion for two-equation K⫺␧ formulations. Al-
though results were presented only for a single Reynolds number
Fig. 4 Comparison of mean axial velocity between model pre- of 130,000, there is no reason to believe this conclusion would
dictions and experimental data †6,16‡. „a… x Ä5.0; „b… x Ä8.3
change at moderately higher or lower Reynolds numbers. 共As
mentioned previously, a calculation at Re⫽143,000 revealed no
significant change in the solution.兲
rotation rate predicted by the DRSM agrees quite well with the For calculations of laminar breakdown within axisymmetric
experimental data at both streamwise locations. However, the K tube geometries, at other than very low Reynolds numbers, axi-
⫺␧ model predicted a rotation rate that was considerably below symmetric bubbles usually become unstable and transition to
the experimental value. Comparisons between experimental and spiral-type breakdown. Consequently, for the present geometry it
numerical results at the given locations provide a level of infor- is of interest to assess the suitability of the dual assumptions of
mation regarding the performance of each model; however, dif- mean flow steadiness and axisymmetry. Toward that end, one un-
ferences in the location of the upstream stagnation point may steady, three-dimensional calculation employing the differential
make anything more than a quantitative comparison suspect. More Reynolds stress model was performed. A 168⫻62⫻62 body fitted
revealing is a comparison between the rates of change in the mean grid was utilized for these calculations, which were carried out
profiles over a specified axial distance. For instance, the experi- over sufficient time to ensure that no unsteadiness of the mean
mental data shows a decrease in the solid body rotation rate be- flow would develop. The results are shown in Fig. 5 in terms of
tween x⫽5 and x⫽8.3 of approximately 10 percent. The DRSM contours of constant axial velocity, and reveal a strongly symmet-
shows essentially no change in the rotation rate over this axial ric mean flow 共where flow is from left to right, and only the
range. However, the K⫺␧ model shows a decrease of approxi- relevant portion of the domain is shown兲. One explanation for the
mately 50 percent. Consequently, one may conclude that even if difference between the Reynolds-averaged and laminar solutions
the ‘‘virtual origin’’ of the breakdown differs between experiment is that solutions to the RANS equations 共at least in 2D兲 are
and numerics, the DRSM is accurately predicting the experimen- equivalent to laminar solutions for a specific prescription of a
tally observed solid body rotation rate; however, the same may variable viscosity. At these low apparent Reynolds numbers dis-
not be said for the K⫺␧ model. In addition, the figures indicate turbances acting on the mean flow are likely damped. These re-
that the DRSM predicts well the free vortex (1/r) behavior of the sults are consistent with the experimental results of Sarpkaya 关5兴
mean flow profile in the region outside the core 共which shows and Sarpkaya and Novak 关6兴 in which the mean flow appeared
minimal variation between stations兲. We also note that results for quite robust, axisymmetric, and steady. In a related matter, the
the DRSM obtained by doubling the grid in the radial direction results indicate that recent theories based on inviscid analyses
appear on Fig. 3共a兲. The profile is indistinguishable from that may not be applicable to high Reynolds turbulent breakdown
obtained on the coarser grid, and thus the solutions may be con- flows.
sidered grid converged. This should not be unexpected, consider-
ing the extremely fine grid employed for the base calculations.
Mean axial velocity profiles are shown in Figs. 4共a兲 and 4共b兲. Acknowledgments
Observations similar to those above may be made regarding the Professor Sarpkaya is thanked for sharing his experimental data
behavior of the axial velocity profiles. In particular, it is clear that with the authors. The National Science Foundation is acknowl-
the two-equation model has failed to correctly predict the exis- edged for providing funding under CTS-9613143.

182 Õ Vol. 122, MARCH 2000 Transactions of the ASME

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
References creased. Hence, vortices shed at the trailing edges of the channel
关1兴 Grabowski, W. J., and Berger, S. A., 1976, ‘‘Solutions of the Navier-Stokes surfaces lead to ‘‘vortex pumping,’’ which is consistent with the
Equations for Vortex Breakdown,’’ J. Fluid Mech., 75, pp. 525–544. negative correlation of the fluctuating velocity on the midplanes
关2兴 Spall, R. E., and Gatski, T. B., 1991, ‘‘Computational Study of the Topology between adjacent channels. 关S0098-2202共00兲01701-6兴
of Vortex Breakdown,’’ Proc. R. Soc. London, Ser. A, 435, pp. 321–337.
关3兴 Breuer, M., and Hanel, D., 1993, ‘‘A Dual Time-Stepping Method for 3-D
Viscous Incompressible Vortex Flows,’’ Comput. Fluids, 22, pp. 467–484.
关4兴 Spall, R. E., 1996, ‘‘Transition From Spiral- to Bubble-Type Vortex Break- Introduction
down,’’ Phys. Fluids, 8, pp. 1330–1332.
关5兴 Sarpkaya, T., 1995, ‘‘Turbulent Vortex Breakdown,’’ Phys. Fluids, 7, pp. Arrays of flat or curved vanes are commonly used to condition
2301–2303. or alter flow fields. For example, swirl vanes—often found in gas
关6兴 Sarpkaya, T., and Novak, F., 1998, ‘‘Turbulent Vortex Breakdown Experi-
ments,’’ IUTAM Symposium on Dynamics of Slender Vortices, E. Krause and
turbine combustors—alter the flow by producing a vortical motion
K. Gersten, eds., Kluwer Academic Publishers, pp. 287–296. that aerodynamically stabilizes reactions within combustors.
关7兴 Hogg, S., and Leschziner, M. A., 1989, ‘‘Computation of Highly Swirling The dearth of studies of the effect of vane arrays on flowfields
Confined Flow with a Reynolds Stress Turbulence Model,’’ AIAA J., 27, provides the motivation for the current study, which uses a sim-
57–63.
关8兴 Bilanin, A. J., Teske, M. E. and Hirsh, J. E., 1997, ‘‘Deintensification as a
plified geometry consisting of arrays of flat vanes or plates. The
Consequence of Vortex Breakdown,’’ Proceedings of the Aircraft Wake Vor- goal of this study is to determine the flow characteristics in the
tices Conference, Report No. FAA-RD-77-68. near wake of a multiplate array with centerplane spacing 共H兲 to
关9兴 Spall, R. E., and Gatski, T. B., 1995, ‘‘Numerical Calculations of Three- thickness 共t兲 ratio of H/t⫽3.33, which is a spacing ratio that is
Dimensional Turbulent Vortex Breakdown,’’ Int. J. Numer. Methods Fluids,
20, pp. 307–318.
characteristic of practical flow control vanes.
关10兴 Gatski, T. B., and Speziale, C. G., 1993, ‘‘On Explicit Algebraic Reynolds
Stress Models for Complex Turbulent Flows,’’ J. Fluid Mech., 254, pp. 59–78. Background
关11兴 Leonard, B. P., 1979, ‘‘A Stable and Accurate Convective Modeling Proce-
dure Based on Quadratic Upstream Interpolation,’’ Comput. Methods Appl. Hayashi et al. 关1兴 study the flow around and downstream of
Mech. Eng., 19, pp. 59–98. two-, three-, and four-plate arrays with 0.014⭐c/t⭐0.114, where
关12兴 Patankar, S. V., 1980, Numerical Heat Transfer and Fluid Flow, Washington,
DC, Hemisphere Publishing Corp.
c/t is the plate chord-to-thickness ratio, and 6⫻103 ⭐Re⭐1.9
关13兴 Launder, B. E., Reece, G. J., and Rodi, W., 1975, ‘‘Progress in the Develop- ⫻104 and with plate separation to plate thickness ratios, H/t, of
ment of a Reynolds-Stress Turbulence Closure,’’ J. Fluid Mech., 68, pp. 537– 0⭐H/t⭐3.75. They study ‘‘flopping’’ which is the spontaneous
566. change in relative base pressure and wake size behind an indi-
关14兴 Gibson, M. M., and Launder, B. E., 1978, ‘‘Ground Effects on Pressure Fluc-
tuations in the Atmospheric Boundary Layer,’’ J. Fluid Mech., 86, pp. 491– vidual plate in the array. They find flopping to occur with the
511. two-plate array only when H/t⫽2.75. For the two-plate arrays
关15兴 Lien, F. S., and Leschziner, M. A., 1994, ‘‘Assessment of Turbulence- with H/t⫽2.75 and 0.07⭐H/t⭐3.25 and three- and four-plate
Transport Models Including Non-Linear RNG Eddy-Viscosity Formulation arrays with 0.07⭐H/t⭐3.25, they find that some plates have wide
and Second-Moment Closure for Flow Over a Backward-Facing Step,’’ Com-
put. Fluids, 23, pp. 983–1004. wakes and others exhibit narrow wakes downstream from their
关16兴 Sarpkaya, T. and Novak, F., private communication. trailing edges but no flopping is observed. This pattern of wide
and narrow wakes is ‘‘quasi-stable’’ as it remains unchanged and
does not vary with time unless the flow is strongly perturbed. For
all arrays, when H/t⬎3.25, the wakes observed downstream of
Synchronous Vortex Shedding „Vortex the trailing edges are of equal width.
Pumping… Downstream of a Flat For a two-plate array with 0.014⭐c/t⭐0.114 and 0.07⭐H/t
⬍3.25, Hayashi et al. 关1兴 find two dominant values of the shed-
Plate Array ding frequency downstream of the array on the centerplane be-
tween plates at 2t downstream. The higher value corresponds to
the narrow wake and the lower value corresponds to the wide
D. W. Guillaume wake. For a three-plate array with 0.014⭐c/t⭐0.114 and 0.07
Assistant Professor, ⭐H/t⬍3.25, on the centerline of the plate array at 2t down-
Department of Mechanical Engineering, California State stream, only one dominant shedding frequency is observed and
corresponds to the narrow wake.
University, Los Angeles, Los Angeles, CA 90032 Miau et al. 关2,3兴 study the flow downstream of two-plate arrays
with 1.4⭐H/t⭐3.0, c/t⫽0.150, and 1.3⫻103 ⭐Re⭐1.2⫻104 .
J. C. LaRue They find spontaneous flopping for 1.5⭐H/t⭐1.85 with 1.3
Professor, ⫻103 ⭐Re⭐1.2⫻104 . They find quasi-stable wake behavior for
1.85⭐H/t⬍2.6, but for H/t⬎2.6.
Department of Mechanical and Aerospace Engineering,
In summary, Hayashi et al. 关1兴, for 0.014⭐c/t⭐0.114, and 6
University of California, Irvine, Irvine, CA 92697 ⫻103 ⭐Re⭐1.9⫻104 , report that flopping and quasi-stable be-
havior can be avoided if H/t⬎3.25, whereas Miau et al. 关3兴, for
c/t⫽0.150, and 1.3⫻103 ⭐Re⭐1.2⫻104 , report that these behav-
Flow visualization and statistics, obtained downstream of an ar- iors can be avoided if H/t⬎2.6.
ray of flat plates, are presented. Flow visualization shows that the
mean separation between streaklines for the flow downstream of Experimental Apparatus
adjacent channels is negatively correlated. Consistent with this
Figure 1 shows the flow facility used in this study. Three of the
observation, the zero-time cross correlation of the fluctuating ve-
plates are labeled ‘‘upper,’’ ‘‘plate0,’’ and ‘‘lower.’’ The plates in
locity between adjacent plates is negative. Vortices shed at the
the six-plate array have aspect ratios, c/t, of 6.66.
trailing edges of the bounding surfaces that form a channel in the
The flow visualization is produced using smoke illuminated
plate array are in-phase, but they are nearly 180 deg out-of-phase
with a 2 W argon-ion, 0.8 mm thick, laser light sheet. The smoke
with the vortices shed at the trailing edges of the bounding sur-
faces of adjacent channels. Relative to the mean velocity, the axial
Contributed by the Fluids Engineering Division of THE AMERICAN SOCIETY OF
velocity on the midplane of the channel increased in the region MECHANICAL ENGINEERS. Manuscript received by the Fluids Engineering Division
between each pair of vortices. At downstream positions not near February 4, 1999; revised manuscript received November 3, 1999. Associate Tech-
the vortex pair, the velocity on the midplane is relatively de- nical Editor: P. W. Bearman.

Journal of Fluids Engineering MARCH 2000, Vol. 122 Õ 183

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 1 Schematic drawing of test facility that defines nomenclature and speci-
fies the dimensions used for this study

is injected at the leading edge of the plate. Digitized images are and spatial correlations are obtained at positions between adjacent
collected with a Sony video camera 共Model CCD-V101兲. plates on the transverse planes located at x/t⫽1.5, 5, 15, and 45
The hot-wire sensors are made by soldering a 2 mm long, downstream of the trailing edge of the plate array. Figure 3 shows
0.00508 mm diameter, platinum wire to TSI-1210 sensor holders. the zero-time, spatial correlation of the fluctuating axial velocity
The sensors are connected to a TSI Model 1050 constant tempera- at positions between the plates on these downstream planes. In
ture anemometer, a low pass filter, a Computer Boards Inc. 共CBI兲 agreement with the interpretation of the flow visualization images,
SSH-16 sample and hold board and a CBI CIO-AD16F 12 bit a high negative correlation is present when the two hot wires are
analogue to digital converter which is controlled by a PC. The placed near the midplanes between adjacent plates at x/t⫽1.5 and
wires are aligned perpendicular to the plane of Fig. 1. 5, i.e., with one probe positioned at y/H⫽⫺0.5 and the other at
Uncertainties of the velocity measurements were estimated to y/H⫽0.5. Further, at x/t⫽1.5 and 5, as the two hot wires ap-
the 95 percent confidence using techniques of Kline and McClin- proach the trailing edge of adjacent plates, i.e., for sensor posi-
tock 关4兴. The velocities obtained using King’s law have a maxi- tions of y/H⫽0.0 and 1.0, the correlation decreases to approxi-
mum difference of 1.0 percent with a standard deviation of less mately zero. At x/t⫽15, the correlation on the midplane remains
than 0.75 percent. negative but is lower by approximately 25 percent than the corre-
lation at x/t⫽5. At x/t⫽45, the correlation is approximately zero
across the array. Thus, a strong negative spatial correlation is
present on the midplanes between plates at x/t⫽1.5 and 5, while
Results no correlation is present at x/t⭓45.
Figure 2 shows a smoke visualization of the streaklines passing Space-time cross-correlation coefficients of the fluctuating ve-
through and downstream of the array with approximately U locity assess the relationship between vortex shedding at the top
⫽2 m/s, which corresponds to a Reynolds number based on chord and bottom surfaces of plate0 and the top and bottom surfaces of
of 4800, and c/t⫽6.66. The streaklines show that the flow sepa- the upper plate. Figure 21共a兲 shows the cross correlation of fluc-
rates at the leading edge of each plate. Between the plates and for tuating velocity with one probe above the top surface of plate0,
x/t⫽0 until x/t⫽1, the streaklines are fairly parallel. At approxi- i.e., at y/H⫽0.305, and the other probe at the same relative loca-
mately x/t⫽1.5, a transverse displacement of the streaklines is
first observed. From about x/t⫽1.5 until x/t⫽3 there is an in-
crease in the mean transverse separation between the streaklines
that are downstream of the channel formed by plate0 and the upper
plate. Simultaneously, there is a decrease in the mean separation
between the streaklines that are downstream of the channel
formed by plate0 and the lower plate. This is also true for the
region from about x/t⫽4.5 until 6. From about x/t⫽3 until 4.5,
the opposite occurs; there is a decrease in the mean separation
between the streaklines that are downstream of the channel
formed by plate0 and the upper plate. Simultaneously, there is an
increase in the mean separation between the streaklines down-
stream of the channel formed by plate0 and the lower plates. Thus,
the mean separation between streaklines for the flow downstream
of adjacent channels is negatively correlated.
Cross-correlation coefficients of the fluctuating downstream ve-
locity with zero time separation are determined using two hot- Fig. 2 Smoke visualization of the streaklines passing through
wire probes transversely separated by a distance equal to the plate and downstream of an array of plates with c Ä2.54 cm, t
centerplane spacing 共H兲. The two hot-wires are moved in tandem Ä0.38 cm, i.e., c Õ t Ä6.66, and U Ä2 mÕs

184 Õ Vol. 122, MARCH 2000 Copyright © 2000 by ASME Transactions of the ASME

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 3 The spatial correlation of the downstream axial velocity as a function of
y Õ H obtained using two hot-wires, transversely separated by a distance H , with
c Ä2.54, t Ä0.38, i.e., c Õ t Ä6.66, and U Ä2 mÕs. Symbols: x Õ t Ä1.5, 䊊; x Õ t Ä5, 䊏;
x Õ t Ä15, 䉱; and x Õ t Ä45, Ã. The normalized uncertainty of the spatial correlation
is estimated as Á0.01.

tion above the top surface of the upper plate, i.e., at y/H Figure 4共b兲 shows the cross correlation of the fluctuating veloc-
⫹⌬y/H⫽1.305. At zero time separation, the cross correlation has ity with one probe placed above the top surface of plate0, i.e., at
a negative value of ⫺0.083 and exhibits a damped periodic be- y/H⫽0.310, and the other probe placed below the bottom surface
havior with increasing time until about 0.07 s. Thereafter, the of the upper plate, i.e., at y/H⫹⌬y/H⫽0.690. At zero time sepa-
peak-to-peak amplitude in the correlation appears to increase. The ration, the cross correlation has a positive value of 0.329 and
zero time correlation is not the maximum or minimum value be- exhibits a damped periodic behavior with increasing time until
cause of a small phase shift between the vortices shed by these about 0.035 s. From about 0.035 s until about 0.06 s, the peak-to-
neighboring plates and because of a slight spatial misalignment of peak amplitude of the oscillations in the cross correlation in-
the sensors with the trailing edge of the plate array. creases with increasing time. Thereafter, the peak-to-peak ampli-
Since the cross-correlation coefficient is negative at zero time tude decreases.
shift, the vortices shed at the trailing edges, near the top surfaces Since the cross-correlation coefficient is positive at zero time
of adjacent plates are negatively correlated; they are nearly 180 shift, the vortices shed at the trailing edges, near the top and
deg out-of-phase. bottom surfaces of the channel formed by the top surface of plate0
and the bottom surface of the upper plate are nearly in-phase.

Summary
Flow visualization shows that the mean separation between
streaklines for the flow downstream of adjacent channels is nega-
tively correlated. Consistent with this observation, the zero-time
cross correlation of the fluctuating velocity between adjacent
plates is negative. Vortices shed at the trailing edges of the bound-
ing surfaces that form a channel in the plate array are in-phase,
but they are nearly 180 deg out-of-phase with the vortices shed at
the trailing edges of the bounding surfaces of adjacent channels.
Relative to the mean velocity, the axial velocity on the midplane
of the channel is increased in the region between each pair of
vortices. At downstream positions not near the vortex pair, the
velocity on the midplane is relatively decreased. Hence, vortices
shed at the trailing edges of the channel surfaces lead to ‘‘vortex
pumping’’ which is consistent with the negative correlation of the
fluctuating velocity on the midplanes between adjacent channels.

References
关1兴 Hayashi, M., Sakurai, A., and Ohya, Y., 1986, ‘‘Wake Interference of a Row
of Normal Flat Plates Arranged Side by Side in a Uniform Flow,’’ J. Fluid
Mech., 164, pp. 1–25.
Fig. 4 Cross correlations of the downstream velocity with c 关2兴 Miau, J. J., Wang, H. B., and Chou, J. H., 1992, ‘‘Intermittent Switching of
Ä2.54 cm, t Ä0.38 cm, i.e., c Õ t Ä6.66, U Ä2 mÕs and x Õ t Ä5. „a… Gap Flow Downstream of Two Flat Plates Arranged Side by Side,’’ J. Fluid
Struct., 6, pp. 563–582.
has one probe above the surface of plate0, i.e., at y Õ H Ä0.305, 关3兴 Miau, J. J., Wang, H. B., and Chou, J. H., 1966, ‘‘Flopping Phenomenon of
and the other probe above the surface of the adjacent plate, Flow Behind Two Plate Place Side-by-Side Normal to the Flow Direction,’’
i.e., at y Õ H ¿⌬ y Õ H Ä1.305. „b… has one probe above the surface Fluid Dyn. Res., 17, pp. 311–328.
of plate0, i.e., at y Õ H Ä0.310, and the other probe below the sur- 关4兴 Kline, S. J., and McClintock, F. A., 1953, ‘‘Describing Uncertainties in Single-
face of the adjacent plate, i.e., y Õ H ¿⌬ y Õ H Ä0.690. Sample Experiments,’’ Mech. Eng., 75, pp. 3–9.

Journal of Fluids Engineering MARCH 2000, Vol. 122 Õ 185

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
A Novel Technique to Measure the range depend on the length and diameter of the sensor tube and
the performance characteristics of the hot-wire anemometers. The
Magnitude and Direction of performance specifications quoted in this paper are for the sensor
developed for the radon entry study—a low flow rate, low detec-
Flow in a Tube tion limit application. A larger diameter sensor tube would be
needed for higher flow rate applications.
Allen L. Robinson
Assistant Professor, Determining Flow Direction
Department of Mechanical Engineering and Differences in the signals between the two hot-wire anemom-
Department of Engineering and Public Policy, eters are used to determine the direction of the flow. Under low
Carnegie Mellon University, Pittsburgh, PA 15213 velocity 共mixed-convection兲 conditions, a hot wire located in an
upward flow 共with respect to gravity兲 will report a higher velocity
than a hot wire located in an equivalent magnitude downward
Richard G. Sextro flow. This phenomenon has been well documented in the field of
Staff Scientist, thermal anemometry 共e.g., Tewari and Jaluria 关3兴兲; the contribu-
Energy and Environment Division, E. O. Lawrence tion of this paper is the application of this phenomenon to create a
Berkeley National Laboratory, Berkeley, CA 94720 novel flow sensor with directional sensitivity.
Briefly, a hot-wire anemometer determines velocity by measur-
ing the heat transfer rate from a heated sensing element 共typically
a cylinder or sphere兲. Under constant fluid temperature and pres-
This paper describes a novel in-line sensor that measures the sure conditions, changes in heat transfer rate indicate the instan-
magnitude and direction of gas flow in a tube. The sensor pos- taneous velocity. At high velocities, the heat transfer rate depends
sesses a unique set of performance characteristics: low detection only on the magnitude of the fluid velocity past the hot wire. In
limit, little resistance to flow, and directional sensitivity. The sen- this flow regime, hot-wire anemometers provide an unambiguous
sor consists of two hot wire anemometers mounted in a U-shaped measurement of gas velocity. As the fluid velocity decreases,
tube. Differences in the signals between the two hot wires under natural convection, created by the temperature difference between
low velocity conditions are used to determine the direction of the the hot wire and the surrounding fluid, begins to significantly
flow. Calibration curves of flow rate versus measured velocity are contribute to the heat transfer. Under low velocity conditions
used to determine the magnitude of the flow. The sensor has ap- 共mixed-convection regime兲, the heat transfer rate, and thus the
plications in systems that are characterized by naturally driven measured velocity, depends on the actual fluid velocity and the
oscillating flows. 关S0098-2202共00兲02701-2兴 orientation of the flow with respect to gravity. We use this depen-
dence to determine the direction of the flow.
A simple theoretical model can be used to illustrate the direc-
tional sensitivity of velocity measurements made by a hot-wire
Introduction anemometer and to estimate the velocity range over which these
This paper describes the theory, calibration, and operation of an measurements depend on the orientation of the flow with gravity.
in-line sensor that measures the magnitude and direction of gas The model presented here predicts the heat transfer rate from the
flow in a tube. The sensor possesses a unique set of performance spherical sensing element of an omnidirectional hot-film velocity
characteristics: low detection limit 共0.15 L min⫺1兲, little resistance transducer 共TSI mode 8470兲 used for the radon entry study. These
to flow 共0.1 Pa min L⫺1兲, and directional sensitivity. We devel- transducers operate at a constant temperature difference above the
oped the sensor to investigate soil-gas and radon entry into an gas and have a low velocity detection limit. Neglecting radiation,
experimental basement driven by natural atmospheric pressure the heat transfer rate, q, from the spherical sensing element is
fluctuations 关1,2兴. Potential additional applications include build-
q⫽h̄A 共 T s ⫺T ⬁ 兲 (1)
ing ventilation and other systems with oscillating flows.
A schematic of the flow sensor is shown in Fig. 1. The sensor
incorporates two air velocity transducers 共e.g., hot-wire anemom-
eters or hot-film velocity transducers兲 mounted in a U-shaped
tube. 共In this paper, we refer to these velocity transducers as hot-
wire anemometers.兲 The sensing element of each hot wire is
mounted in the centerline of the flow sensor tube. One hot wire is
mounted in each leg of the U resulting in one transducer being
exposed to an upward flow 共with respect to gravity兲 and the other
being exposed to an equivalent in magnitude downward flow. The
U-shaped tube must have a vertical orientation such that each leg
of the U is parallel to gravity.
The velocity measured by one of the hot wires and a calibration
curve determines the magnitude of the flow. The direction of the
flow is determined by differences in the signals between the two
hot wires under low-velocity conditions 共less than 10 cm s⫺1 for
the hot wires used in this study兲. These differences arise because
velocity measurements made with a hot-wire anemometer are sen-
sitive to the orientation of the flow with respect to gravity under
low velocity conditions.
The sensor detection limit, resistance to flow, and operational

Contributed by the Fluids Engineering Division of THE AMERICAN SOCIETY OF Fig. 1 Schematic of flow sensor. The sensing element of each
MECHANICAL ENGINEERS. Manuscript received by the Fluids Engineering Division air velocity transducer is mounted along the centerline of the
June 1, 1999; revised manuscript received November 30, 1999. Associate Technical tube. The U-shaped tube used for the radon entry study was 1.9
Editor: M. R. Hajj. cm in diameter and 40 cm tall. The figure is not drawn to-scale.

186 Õ Vol. 122, MARCH 2000 Copyright © 2000 by ASME Transactions of the ASME

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Table 1 Parameters used to estimate heat-transfer rate from
spherical sensing element of a TSI Model 8470 omnidirectional
hot-film velocity transducer

Property Symbol Value


Air:
Kinematic viscosity ␯ 1.5⫻10⫺5 m2 s⫺1
Volumetric expansion coefficient ␤ 3.2⫻10⫺3 K⫺1
Thermal conductivity k 2.6⫻10⫺2 W m⫺1 K⫺1
Ambient temperature T⬁ 293 K
Sphere:
Temperature Ts T ⬁ ⫹30 K
Diameter d 3.175⫻10⫺3 m
Surface area A 3.2⫻10⫺3 m2
Grashof number Gr 130
Gravitational acceleration g 9.8 m s⫺2

where h̄ is the average heat transfer coefficient for the sphere, A is


the surface area of the sphere, T s is the temperature of the sphere,
and T ⬁ is the fluid temperature. Chen and Armaly 关4兴 present
semi-empirical correlations for the heat transfer coefficient of a
sphere in mixed-convection flow. For upward flows 共flow oppo-
site to the direction of gravity兲,

Nu⫽共Nr3.5⫹Ng3.5兲1/3.5⫹2 (2)

where Nr⫽0.493 Re1/2, and Ng⫽0.392 Gr1/4. Nu is the average Fig. 2 „a… Predictions of the heat transfer rate from a 3.175-
Nusselt number, Re is the Reynolds number, and Gr is the mm-diameter spherical sensing element of an omnidirectional
Grashof number: hot-film velocity transducer „TSI model 8470… as a function of
velocity. „b… Difference between the predicted heat transfer rate
h̄d vd g ␤ 共 T s ⫺T ⬁ 兲 d 3 from the sensing element in an upward and a downward flow.
Nu⫽ , Re⫽ , Gr⫽ , (3) An upward flow is in the opposite direction to gravity; a down-
k ␯ ␯2
ward flow is in the same direction as gravity.
where d is the sphere diameter, k is the thermal conductivity of the
fluid, v is the fluid velocity, ␯ is kinematic viscosity of the fluid, g
is the gravitational acceleration, and ␤ is the volumetric thermal Calibration
expansion coefficient of the fluid. For downward flows 共flows in
Calibration curves of flow rate versus velocity are required to
the same direction as gravity兲,
determine the gas flow rate through the sensor tube. Two calibra-
Nu⫽共Nr3⫺Ng3兲1/3⫹2 if Ng/Nr⬍1 (4a) tion curves are needed, one for each direction of flow. These
calibration curves account for the tube diameter and the location
or of the velocity transducer inside the tube. Calibrating the sensor
Nu⫽共Ng6⫺Nr6兲1/6⫹2 if Ng/Nr⬎1. (4b) using measurements of flow versus velocity is much more accu-
rate than calculating the flow rate based on the measured velocity
Using Eqs. 共1兲–共4兲 and the physical properties listed in Table 1 and an assumed velocity profile.
we calculate the heat transfer rate from the spherical sensing ele- We determined the calibration curves by imposing a constant
ment of the TSI velocity transducer used for the radon entry study flow rate through the sensor tube, measured with a primary-
共Fig. 2兲. The results indicate that for air velocities between 1 and standard flow meter 共Gilibrator, Gilian Inc.兲, and recording the
10 cm s⫺1, the heat transfer rate, and thus the measured velocity, velocity indicated by each velocity transducer. This procedure is
depends significantly on the orientation of the flow with respect to repeated for the entire range of expected flows and for each flow
gravity. In this range, the heat transfer rate from the transducer direction. The velocities measured by the transducer located in the
located in the upward flow is higher than the heat transfer rate upward flow 共flow in the opposite direction to gravity兲 are used to
transducer located in the equivalent downward flow. This results determine the flow rate through the sensor tube.
in the transducers reporting different velocities depending on the As an example, Fig. 3共a兲 shows one of the calibration curves
orientation of the flow with respect to gravity. This phenomenon for the flow sensor developed for the radon entry study. The cali-
enables us to determine the direction of the flow. bration curve illustrates the principles underlying the operation of
The predictions shown in Fig. 2 indicate that at high velocities the flow sensor. The detection limit of this sensor is ⬃0.15 L
the dependence of the heat transfer rate on the orientation of the min⫺1. This value is determined by the velocity detection limit of
flow with gravity disappears. Under these conditions, the flow the hot wire and the diameter of the sensor tube. As the flow rate
sensor is not directionally sensitive. Therefore, the diameter of the increases above this detection limit, the upward velocity trans-
sensor tube must be sized such that the minimum expected flow ducer begins to detect the flow. The upward transducer reports a
rates create velocities between 1 and 10 cm s⫺1. In addition, the higher velocity than the downward transducer until the velocity
sampling frequency of the flow sensor must be higher than the through the tube approaches 15 cm s⫺1 共a flow rate of 1.5 L min⫺1
frequency of any changes in flow direction. Rapid sampling al- for the sensor tube used in the radon entry study兲. The disappear-
lows us to determine the direction of the flow by recording the ance of the directional distinction at a velocity of ⬃15 cm s⫺1 is
lower velocities at which the flow sensor is directionally sensitive. consistent with the theoretical predictions shown in Fig. 2.
The fluctuations in flow rate must also be slower than the response The response time, accuracy, and resolution of the flow sensor
time of the velocity transducers. used for the radon entry study are 2 s, 5 percent of reading, and

Journal of Fluids Engineering MARCH 2000, Vol. 122 Õ 187

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 4 Comparison between flow from an oscillating flow
source to that measured by the flow sensor

different sinusoids, each with the same magnitude and periods of


16, 8, 4, and 2 min., are shown in Fig. 4. The results indicate the
flow sensor accurately measured the sinusoidally oscillating flow.
The largest absolute error in the flow rate is 0.2 L min⫺1 with an
average absolute error of 0.04 L min⫺1. As expected, the largest
errors occur at low flowrates, around the detection limit of the
sensor. The relative error in the absolute value of the integrated
flow is less than 1 percent.

Fig. 3 „a… Calibration curves for the flow sensor developed for Acknowledgments
the radon entry study. „b… Difference in the velocity reported by
the hot-wire anemometers. The upward hot-wire is in the leg of We thank W. Fisk and A. Gadgil for their critical review of this
the U-shaped tube in which the flow is in the opposite direction work. This work was supported by the Director, Office of Energy
to gravity; the downward hot-wire is in the leg of the U-shaped Research, Office of Health and Environmental Research, Environ-
tube in which the flow is in the same direction as gravity. As mental Sciences Division, and by the Assistant Secretary for En-
described in the text, the velocity measured by the upward ergy Efficiency and Renewable Energy, Office of Building Tech-
transducers is used to determine the flow rate through the nology of the U.S. Department of Energy 共DOE兲 under Contract
tube. Note: the specific velocity versus flow rate relationship No. DE-AC03-763SF0098.
shown in this figure is for the sensor developed for the radon
entry study.
References
关1兴 Robinson, A. L., and Sextro, R. G., 1997, ‘‘Radon Entry into Buildings Driven
⫺1 by Atmospheric Pressure Fluctuations,’’ Environ. Sci. Technol., 31, pp. 1742–
0.02 L min , respectively. The response time and resolution are 1748.
determined by the performance characteristics of the velocity 关2兴 Robinson, A. L., Sextro, R. G., and Fisk, W. J., 1997, ‘‘Soil-Gas Entry Into an
transducers. The accuracy was determined by the repeatability of Experimental Basement Driven by Atmospheric Pressure Fluctuations—
the calibrations. The sensor was regularly recalibrated; there was Measurements, Spectral Analysis, and Model Comparison,’’ Atmos. Environ.,
31, pp. 1477–1485.
less than a 5 percent variation between each of these calibrations. 关3兴 Tewari, S. S., and Jaluria, Y., 1990, ‘‘Calibration of Constant-Temperature
Hot-Wire Anemometers for Very Low Velocities in Air,’’ Rev. Sci. Instrum.,
Operation 61, pp. 3834–3845.
关4兴 Chen, T. S., and Armaly, B. F., 1987, ‘‘Mixed Convection in External Flow,’’
To illustrate the operation of the flow sensor, the sensor was Handbook of Single-Phase Convective Heat Transfer, S. Kakaç, R. K. Shah,
used to measure a sinusoidally oscillating flow. Results for four and W. Aung, eds. Wiley, NY, pp. 14-1–14-35.

188 Õ Vol. 122, MARCH 2000 Transactions of the ASME

Downloaded 16 Jul 2010 to 210.212.205.135. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Вам также может понравиться