J Fluids Engineering 2009 Vol 131 N5

Journal of
Fluids Engineering
Published Monthly by ASME
VOLUME 131 NUMBER 5 MAY 2009

FLUIDS ENGINEERING DIVISION
Editor
J. KATZ 2009
Assistant to the Editor
L. MURPHY 2009
Associate Editors
M. J. ANDREWS 2009
E. M. BENNETT 2012
S. L. CECCIO 2009
D. DRIKAKIS 2012
P. DURBIN 2012
I. EAMES 2010
C. HAH 2010
T. J. HEINDEL 2011
J. KOMPENHANS 2009
YU-TAI LEE 2009
J. A. LIBURDY 2011
R. MITTAL 2010
T. J. OHERN 2009
N. A. PATANKAR 2011
H. PEERHOSSAINI 2011
U. PIOMELLI 2010
Z. RUSAK 2010
D. SIGINER 2009
M. STREMLER 2012
M. WANG 2011
St. T. WERELEY 2011
Y. ZHOU 2009
PUBLICATIONS COMMITTEE
Chair, B. RAVANI
OFFICERS OF THE ASME
President, THOMAS M. BARLOW
Executive Director, THOMAS G. LOUGHLIN
Treasurer, T. D. PESTORIUS
RESEARCH PAPERS
Flows in Complex Systems
051101
Performance and Radial Loading of a Mixed-Flow Pump Under

Non-Uniform Suction Flow
B. P. M. van Esch
051102
Experimental and Computational Study of Oscillating Turbine Cascade

and Influence of Part-Span Shrouds
X. Q. Huang, L. He, and David L. Bell
051103
Experimental Investigation of Flow Field Structure in Mixing Tee

Seyed Mohammad Hosseini, Kazuhisa Yuki, and
Hidetoshi Hashizume
051104
Experimental Study of a New Flow Conditioner on Disturbed Flow

in Orifice Plate Metering
A. Ahmadi
051105
2D NavierStokes Simulations of Microscale Viscous Pump With Slip

Flow
Khaled M. Bataineh and Mohd A. Al-Nimr
051106
Pre-Stall Instability Distribution Over a Transonic Compressor Rotor

A. J. Gannon and G. V. Hobson
Fundamental Issues and Canonical Flows

051201
Front Condition for Gravity Currents in Channels of Nonrectangular

Symmetric Cross-Section Shapes
B. M. Marino and L. P. Thomas
051202
Vortex Dynamics and Shedding of a Low Aspect Ratio, Flat Wing at

Low Reynolds Numbers and High Angles of Attack
Daniel R. Morse and James A. Liburdy
PUBLISHING STAFF
Managing Director, Publishing
P. DI VIETRO
Manager, Journals
C. MCATEER
Production Coordinator
A. HEWITT
Multiphase Flows
051301
Drag Reduction in Turbulent Flow With Polymer Additives

Shu-Qing Yang
051302
Cavitating Turbulent Flow Simulation in a Francis Turbine Based on

Mixture Model
Shuhong Liu, Liang Zhang, Michihiro Nishi, and Yulin Wu
Techniques and Procedures
Transactions of the ASME, Journal of Fluids Engineering

ISSN 0098-2202 is published monthly by
The American Society of Mechanical Engineers,
Three Park Avenue, New York, NY 10016.
Periodicals postage paid at New York, NY
and additional mailing offices.
POSTMASTER: Send address changes to Transactions of the
ASME, Journal of Fluids Engineering, c/o THE AMERICAN
SOCIETY OF MECHANICAL ENGINEERS,
22 Law Drive, Box 2300, Fairfield, NJ 07007-2300.
CHANGES OF ADDRESS must be received at Society
headquarters seven weeks before they are to be effective.
Please send old label and new address.
STATEMENT from By-Laws. The Society shall not be
responsible for statements or opinions advanced in papers
or ... printed in its publications B7.1, Par. 3.
COPYRIGHT 2009 by the American Society of
Mechanical Engineers. Authorization to photocopy material for
internal or personal use under those circumstances not falling
within the fair use provisions of the Copyright Act, contact
the Copyright Clearance Center CCC, 222 Rosewood Drive,
Danvers, MA 01923, tel: 978-750-8400, www.copyright.com.
Request for special permission or bulk copying should be
addressed to Reprints/Permission Department.
Canadian Goods & Services Tax Registration #126148048.
051401
Parallel-Plate Conductive Electrodes for the Fabrication of Larger 2D

Colloidal Photonic Crystals
R. Asmatulu, S. Kim, F. Papadimitrakopoulos, and H. Marcus
051402
Comparison of Turbulence Modeling Strategies for Indoor Flows

Ammar M. Abdilghanie, Lance R. Collins, and David A. Caughey
051403
Unsteady Wall Shear Stress in Transient Flow Using Electrochemical

Method
H. Zidouh, L. Labraga, and M. William-Louis
Contents continued on inside back cover
Downloaded 03 Jun 2010 to 171.66.16.159. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Contents continued
Journal of Fluids Engineering
Volume 131, Number 5
MAY 2009
TECHNICAL BRIEFS
054501
Inviscid Flow Past Two Cylinders

R. S. Alassar and M. A. El-Gebeily
The ASME Journal of Fluids Engineering is abstracted and indexed in

the following:
Applied Science & Technology Index, Chemical Abstracts, Chemical Engineering and
Biotechnology Abstracts (Electronic equivalent of Process and Chemical Engineering),
Civil Engineering Abstracts, Computer & Information Systems Abstracts, Corrosion
Abstracts, Current Contents, Ei EncompassLit, Electronics & Communications
Abstracts, Engineered Materials Abstracts, Engineering Index, Environmental
Engineering Abstracts, Environmental Science and Pollution Management, Excerpta
Medica, Fluidex, Index to Scientific Reviews, INSPEC, International Building Services
Abstracts, Mechanical & Transportation Engineering Abstracts, Mechanical Engineering
Abstracts, METADEX (The electronic equivalent of Metals Abstracts and Alloys Index),
Petroleum Abstracts, Process and Chemical Engineering, Referativnyi Zhurnal, Science
Citation Index, SciSearch (The electronic equivalent of Science Citation Index), Shock
and Vibration Digest, Solid State and Superconductivity Abstracts, Theoretical Chemical
Engineering
Performance and Radial Loading

of a Mixed-Flow Pump Under
Non-Uniform Suction Flow
B. P. M. van Esch
Technische Universiteit Eindhoven,
P.O. Box 513,
5600MB Eindhoven, The Netherlands
e-mail: b.p.m.v.esch@tue.nl
Many centrifugal pumps have a suction velocity profile, which is nonuniform, either by
design like in double-suction pumps, sump pumps, and in-line pumps, or as a result of an
installation close to an upstream disturbance like a pipe bend. This paper presents an
experimental study on the effect of a nonuniform suction velocity profile on performance
of a mixed-flow pump and hydrodynamic forces on the impeller. In the experiments, a
newly designed dynamometer is used, equipped with six full Wheatstone bridges of strain
gauges to measure the six generalized force components. It is placed in between the shaft
of the pump and the impeller and corotates with the rotor system. A high accuracy is
obtained due to the orthogonality of bridge positioning and the signal conditioning electronics embedded within the dynamometer. The suction flow distribution to the pump is
adapted using a pipe bundle situated in the suction pipe. Results of measurements show
the influence of the suction flow profile and blade interaction on pump performance and
forces. Among the most important observations are a backward whirling motion of the
rotor system and a considerable steady radial force. DOI: 10.1115/1.3089539
Keywords: fluid-induced forces, hydrodynamics, mixed-flow pump, nonuniform suction
flow
Introduction
Forces on pump shafts may have several causes. The origin can
be mechanical, such as mass unbalance or misalignment, or induced by the working fluid. Generally, a nonuniform pressure distribution at the periphery of the impeller is regarded as a cause for
a steady lateral force, like the one encountered in pumps equipped
with a volute type of casing. Hydrodynamic excitation forces result from a variety of unsteady flow phenomena such as stall, flow
recirculation, cavitation, and blade interaction, while hydrodynamic reaction forces on the other hand may result from rotor
whirl or precessing motion relative to the casing.1.
Agostinelli et al. 2 and Iversen et al. 3 were the first to
measure the steady lateral force on an impeller in a vaneless volute. Domm and Hergt 4 did similar experiments in which they
demonstrated that impeller forces are dependent on rotor eccentricity. Hergt and Krieger 5 extended these measurements for an
impeller, which was placed eccentrically in a vaned diffuser.
Uchida et al. 6 measured the unsteady lateral forces on an impeller in a volute type of casing. In all of these experiments, force
sensing devices of different designs were used to measure reaction
forces in the bearing support structures. Chamieh et al. 7 used an
external balance to measure the hydrodynamic stiffness matrix for
a centrifugal impeller in a volute, whirling at a very low speed.
Internal balances, or so-called dynamometers, were used by
subsequent investigations to assess the fluid-induced rotordynamic
behavior of centrifugal pumps. These force sensing devices are
located between the impeller and the shaft so as to directly measure forces acting on the impeller. Jerry et al. 8 constructed a
dynamometer with four square posts parallel to the shaft, onto
which were mounted nine full Wheatstone bridges of strain
gauges to measure six generalized force components. Rotordynamic forces were measured for a radial-type centrifugal impeller
in a vaneless volute forced into a whirling motion. Further re-
search focused on the effect of leakage flow on rotordynamic

forces e.g., Refs. 911. Ohashi et al. 12 designed a dynamometer consisting of two pairs of parallel plates with four strain
gauges each, to measure the two lateral forces and associated
force moments. Rotordynamic behavior was assessed for a centrifugal impeller in a vaneless diffuser, set into precessing motion.
The same device was subsequently used in a series of experiments
studying the rotordynamic behavior of several pumps of different
designs and specific speeds, as well as the influence of seal geometry e.g., Refs. 1316.
A nonuniform suction flow to a pump has an influence on pump
performance e.g., Refs. 1719. This type of flow is encountered in many different pump designs, e.g., double-suction pumps,
in-line radial pumps, sump pumps, the suction stage of a boiler
feed pump, and in pumps positioned closely to an upstream flow
disturbance like a pipe bend. Yet, the effect on hydrodynamic
forces has hardly been addressed in literature. Bolleter et al. 20
measured lateral forces on the shaft of a radial pump with different inlet chamber designs. Fujii et al. 21 investigated the blade
loading of an inducer under cavitating conditions, with and without inlet flow distortion.
This paper presents an experimental assessment of the effect of
a nonuniform suction flow on pump performance and radial loading. The pump is a scale model of a waterjet pump, introduced in
Sec. 2. In practice, the suction flow to this pump is highly nonuniform. Even at normal operating conditions, variations in axial
velocity can be as high as 50% of its average value. A general
description of the experimental setup is given in Sec. 3, while the
details of the dynamometer and the flow redistribution device are
given in Secs. 4 and 5. Finally, results of measurements are given
in Sec. 6. Both general performance of the pump and radial loading of the impeller are addressed. Results with and without suction flow distortion are compared.
2
Contributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received June 26, 2008; final manuscript
received January 19, 2009; published online April 1, 2009. Assoc. Editor: Chunill
Hah.
Waterjet Pumps
Figure 1 shows a frequently used design of a waterjet system, a

mixed-flow pump with an intake duct flush-mounted to the hull of
the ship. Water is taken from below the ship and flows through the
Copyright 2009 by ASME
MAY 2009, Vol. 131 / 051101-1
electromagnetic flow meter
gauge and safety valve
air in (max 4 bar)
discharge valve
air
water
outlet pressure inlet pressure

transducer
transducer
Fig. 1 Waterjet propulsion system employing a mixed-flow

pump with stator bowl and nozzle
duct to the pump, is subsequently pressurized, and accelerated in

the nozzle. The inlet flow distribution in systems like these is
inevitably nonuniform, caused by the following:
mixed-flow pump
redistribution device
and pitot tubes
Fig. 3 Experimental facility
suction from the boundary layer

diffusion and/or acceleration in the inlet duct, depending on
operating conditions
the double bend in the inlet duct
the rotating pump shaft protruding into the inlet duct
The nonuniform inflow velocity distribution at normal operating condition is a phenomenon, which is widely recognized
2224.
As an example, measured axial velocity distributions are shown
in Fig. 2 25. Two operating conditions are given, representing a
fully loaded and an unloaded vessel at maximum power. Operating conditions of waterjet propelled ships are customarily classified according to the inlet velocity ratio IVR, which is defined as
IVR = vs/v p
with vs the ship speed and v p the average axial velocity entering
the pump. Thus, a fully loaded ship at maximum velocity has a
medium IVR value while an unloaded ship at maximum velocity
is characterized by a high value of IVR. Operating conditions
during the start-up phase are characterized by low values of IVR.
Experimental Setup
A scale-model version of a waterjet pump manufactured by

Wrtsil Propulsion Netherlands Drunen, The Netherlands is
built into the test facility at the Technische Universiteit Eindhoven
Eindhoven University of Technology Fig. 3. The main characteristics of this setup are a closed-loop circuit of 18 m, a pipe
diameter of 315 mm, an electromagnetic flow meter, a discharge
butterfly valve, air release valves, and a pressure tank. The pump
is of mixed-flow type with a six-bladed impeller and a stator bowl
with seven vanes. The impeller outer diameter is 335 mm. A key
component is the dynamometer. A detailed description of this
force sensing device is given in Sec. 4.
In a normal waterjet pumping unit, the shaft is mounted at the
suction side of the pump Fig. 1. However, in this setup, the shaft
construction is changed and installed at the discharge. The reason

for this alteration is that it enables a comparison between results
obtained at nonuniform suction flow and equivalent results with
uniform inlet flow. This is why the straight suction pipe is
equipped with a flow redistribution device. Further discussion of
this device is postponed until Sec. 5.
Another adaptation to the pump, which is currently not in use,
however, is the dual shaft system by means of which the rotor can
be set into a whirling or precessing motion. This feature is included in the design to allow for future measurements of hydrodynamic reaction forces.
Signals of the dynamometer are transmitted by cables, which
lead to a set of slip rings through a hole in the central shaft. The
shaft of the pump is equipped with an optical encoder providing a
once-per-revolution pulse from which the impeller orientation is
determined for synchronization with the data acquisition system.
Dynamometer
The dynamometer is a force sensing device, which is placed

between the shaft of the pump and the impeller Fig. 4. Forces on
the impeller are measured directly, as opposed to using external
balances in bearings or pump support structures for which measurements are often hampered by unknown properties of dynamically loaded bearings and seals, or suffer from inaccuracies due to
bearing play. The dynamometer has four square posts Fig. 5 onto
which 24 strain gauges are mounted, grouped into six full Wheatstone bridges to enhance temperature compensation. This way, six
generalized force components radial forces, bending moments,
axial force, and torque can be measured. The capability of the
dynamometer to measure radial force as well as its associated
bending moment is essential to this investigation, since the point
of action of the radial force is not known a priori. The design
resembles that of Jerry et al. 8 although positions of strain
gauges are different.
flexible pulley spacing ring
coupling
thrust
bearing
Fig. 2 Examples of measured axial velocity distribution in a

plane just upstream of the impeller. Velocities are normalized
with the average axial velocity.
051101-2 / Vol. 131, MAY 2009
radial
bearing
discharge pipe
double roller
bearing
main
shaft
stator
secondary
shaft
impeller suction pipe
dynamometer
Fig. 4 Cross section of the mixed-flow pump indicating the

position of the dynamometer
Transactions of the ASME
1.0
uniform
0.8
profile A
profile B
y [-] 0.6
0.4
0.2
0
0.0
0.5
1.0
1.5
2.0
v / vaverage [-]
Fig. 5 Drawing of dynamometer showing four posts left and

cross section of dynamometer with spline part and sealing
jackets
To minimize noise and prevent loss of signals, electronics for

stabilized bridge power supply and signal amplification are built
into the dynamometer Fig. 6. Eight wires carrying the supply
voltage and the bridge output signals are connected to a slip ring
assembly with 24 channels. Three channels are used for each wire
connection to minimize the effect of noise in the transmission.
Calibration of the dynamometer is done both statically and dynamically. Static radial loading consists of a series of measurements with different weights applied at different axial distances
from the dynamometer. The accuracy of the calibration is increased further by a dynamic calibration procedure in situ. A disk
of known mass is mounted on the shaft at two different axial
positions. Bridge signals are sampled while the system is running
at the intended shaft speed. From these two measurements and the
known mass and axial distances, the response of all six bridges to
radial force and associated bending moment can be assessed. Additional runs with a disk of different weight at different positions
revealed that the weight and the axial location of its center of
mass are determined with accuracies of 0.1% and 0.2%, respectively. The error in phase angle was less than 1.5 deg. The error in
instantaneous forces, determined from the standard deviation in
the ensemble-averaged signal, was 3%.
Flow Redistribution Device
Since the axial velocity profile in the straight suction pipe of the
setup is approximately uniform, a flow redistribution device of
some sort is required. Its purpose is to modify the velocity distribution into a nonuniform profile as given in Fig. 2. A bundle of
small diameter tubes was shaped according to the method of Kotansky 26, extended to three dimensions. However, measurements showed that the resulting velocity distribution, though
stable and nonuniform, did not resemble the intended shear profile. Velocity measurements were performed with Pitot tubes in a
Fig. 7 Axial velocity profiles along a vertical traverse, with and

without a pipe bundle installed
plane located approximately one pipe diameter upstream of the

impeller and at a distance of three pipe diameters downstream of
the redistribution device. Measured velocity profiles are given in
Fig. 7. Velocity profile B was obtained after rotation of the pipe
bundle over 180 deg.
Although velocity profile B does not closely resemble the situation encountered in waterjet pump installations, the main characteristic of a changing axial velocity in vertical direction is obtained. The resulting variation in velocity is almost in the range of
what is normally encountered at medium values of IVR Fig. 8.
Therefore, it is concluded that many of the phenomena in waterjet
installations, related to the nonuniform entrance flow, can be studied in the current setup.
Results
Experiments are performed with and without velocity redistribution device for different shaft speeds ranging from
450 rpm to 900 rpm. Reynolds numbers are in the range Re
= 5 11 106, and Rev = 6 12 105, based on BEP operation.
Flow rates are varied between 10% QBEP and the maximum flow
rate of 120% QBEP 110% QBEP with pipe bundle installed. For
each run, bridge signals are sampled during 500 shaft revolutions and ensemble averaged. No additional filtering is performed
apart from the 50 Hz line frequency.
In Sec. 6.1 global characteristics of the pump are given for both
uniform and nonuniform suction flows. Results of force measurements are given in Sec. 6.2, for a shaft speed of 700 rpm at BEP
flow rate. The influence of the nonuniform suction flow distribution, as well as blade interaction on radial forces, is shown. A
detailed study of the influence of flow rate on the steady fluidinduced force is presented in Sec. 6.3.
6.1 Global Performance Characteristics. Figures 911
present characteristic curves for manometric head, torque, and
axial force, at different shaft speeds. No suction flow redistribution device was installed. The manometric head was calculated
1.0
medium IVR
0.8
high IVR
profile B
y [-] 0.6
0.4
0.2
0
0.0
0.5
1.0
1.5
2.0
v / vaverage [-]
Fig. 6 Dynamometer with electronics for bridge power supply

and signal conditioning left and dynamometer mounted on
the shaft
Fig. 8 Comparison of velocity profile B with measured profiles

in a waterjet intake duct Fig. 3
MAY 2009, Vol. 131 / 051101-3
3.0
3.0
900 rpm
2.5
2.5
750 rpm
2.0
gHs / (N2D2)
gHm / (N2D2)
700 rpm
2.0
1.5
uniform
750 rpm
800 rpm
1.0
non-uniform
small clearance
large clearance
1.5
0.5
0.7
0.9
2.00
1.0
1.75
0.5
0
0.5
1.50
-0.5
0
0.0
0.4
0.6
0.8
1.0
from measured and averaged static pressure and average axial

velocity. Four pressure taps along the circumference of the pipes
0.16
900 rpm
800 rpm
750 rpm
700 rpm
T / (N2D5)
0.14
0.12
0.10
0.08
0.4
0.6
0.8
1.0
1.2
Q / QBEP
Fig. 10 Impeller torque for uniform suction flow, at different

shaft speeds
1.2
1.0
1.2
were used to measure static pressure at stations approximately one

diameter up- and downstream of the pump. Torque and axial force
were measured by the dynamometer and thus represent the forces
on the impeller only. Axial force was corrected for shaft force by
adding the suction gauge pressure multiplied by the cross sectional area of the shaft. Results scale appropriately, although Reynolds effects are present for these fairly low shaft speeds. Characteristic curves show a trend typical for pumps of mixed-flow
type.
For a nonuniform suction velocity profile, the manometric head
is difficult to determine, since an accurate mass-average of total
pressure should be measured. Therefore, the static head, based on
a difference in average static pressure, is used in stead. Figures
1215 present the results of this comparison, where use is made of
profile B see Fig. 7. It appears that a nonuniform suction flow
reduces pump performance in the sense that both head and torque
are reduced over a large part of the pump curve. For this mild
nonuniformity the decrease in head and torque is limited to
2.5% at most. However, static efficiency is not affected, as
illustrated in Fig. 14. It means that a nonuniform suction flow
does not lead to an increase in power consumption; the reduction
in pump head can be compensated by a slight increase in shaft
speed. This is different from, for example, the effect of tip clearance. Figures 1215 also present results for a small clearance of
1 mm and a large clearance of 2 mm. Clearly, the effect of an
increase in tip clearance on performance is much larger. But, more
important in this respect, it leads to a true reduction in perfor-
900 rpm
0.16
700 rpm
uniform
750 rpm
750 rpm
1.1
non-uniform
0.15
1.0
0.14
T / (N2D5)
Fax / (N2D4)
0.8
Fig. 12 Static head for uniform and nonuniform suction flows,

at small and large tip clearances
800 rpm
0.9
0.13
0.12
0.8
small clearance
0.11
0.7
0.0
0.6
Q / QBEP
Q / QBEP
0.2
0.4
1.2
Fig. 9 Manometric head for uniform suction flow, at different

shaft speeds
0.0
0.2
0.0
0.2
large clearance
0.10
0.2
0.4
0.6
0.8
1.0
1.2
Q / QBEP
Fig. 11 Axial force on the impeller for uniform suction flow, at

different shaft speeds
051101-4 / Vol. 131, MAY 2009
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Q / QBEP
Fig. 13 Impeller torque for uniform and nonuniform suction

flows, at small and large tip clearances
0.5
small clearance
750 rpm
0.5
uniform
non-uniform
0.0
700 rpm
100% Q
BEP
FY / Fg
0.4
large clearance
static efficiency
0.3
-0.5
-1.0
-1.5
0.2
-2.0
-1.5
-1.0
-0.5
0.1
0.0
0.5
1.0
1.5
FX / Fg
Fig. 16 Force vector plot in the inertial frame of reference,

normalized with impeller weight Fg in water, for uniform suction
flow profile
-0.1
0.2
0.4
0.6
0.8
1.0
1.2
Q / QBEP
Fig. 14 Pump efficiency based on static head for uniform and

nonuniform suction flows, at small and large tip clearances
mance since also efficiency is reduced. Figure 15 shows results

for axial force on the impeller, which are consistent with the results for head and torque.
6.2 Hydrodynamic Forces. From the ensemble-averaged
bridge signals, one can calculate instantaneous radial force components FX and FY in the inertial frame of reference, during one
impeller revolution. A typical result is given in Fig. 16 for a shaft
speed of 700 rpm, at 100% QBEP, and uniform suction flow profile. Since the geometry is axisymmetric and the suction flow is
almost uniform, the time-averaged radial force is approximately
equal to the weight of the impeller in water. This means that the
net fluid-induced force on the impeller is small.
The dynamometer bridge signals show spectral content with
main contributions at frequencies equal to the shaft frequency and
higher harmonics. Of these higher frequencies, the seventh harmonic is dominant. This is clearly observed in Figs. 17 and 18,
where forces at different ranges in frequency are shown. In Fig.
17, the influence of impeller weight and centrifugal force is apparent. The latter is caused by mass imbalance and, to a larger
extent, hydraulic imbalance. The centrifugal force component ro-
tates in the direction of shaft revolution. Time-averaged hydrodynamic force is small. Rotor-stator blade interaction force is shown
in Fig. 18, in the inertial frame of reference. Note that this blade
passing excitation force induces a backward whirling motion of
the impeller at a frequency of six times shaft frequency.
In case of a nonuniform suction flow distribution, similar force
plots are obtained. However, an additional steady fluid-induced
force is observed in a direction, which depends on the suction
flow profile. A detailed analysis is given in Sec. 6.3.
6.3 Steady Fluid-Induced Forces. Hydrodynamic forces
were measured for different combinations of shaft speed and flow
rate, and for different suction flow profiles. In case of nonuniform
suction flow, a steady fluid-induced force emerges, which results
from an unbalanced torque loading of the impeller blades 27.
The magnitude and direction of this steady force are given in Fig.
19, for different flow rates. It is apparent that the force magnitude
increases with flow rate, which confirms the findings of Bolleter et
0.5
0.0
FY / Fg
0.0
700 rpm
100% Q
BEP
-0.5
-1.0
-1.5
1.3
-2.0
uniform
750 rpm
-1.5
non-uniform
-1.0
-0.5
0.0
0.5
1.0
1.5
FX / Fg
1.2
1.1
small clearance
1.0
1.0
0.5
FY / Fg
Fax / (N2D4)
Fig. 17 Force vector plot for frequencies up to and including

shaft frequency, for uniform suction flow profile
0.9
large clearance
0.8
0.0
-0.5
-1.0
-1.5
0.7
0.0
700 rpm
100% Q
BEP
0.2
0.4
0.6
0.8
1.0
1.2
Q / QBEP
Fig. 15 Axial force for uniform and nonuniform suction flows,

at small and large tip clearances
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
FX / Fg
Fig. 18 Force vector plot for frequencies above shaft frequency showing rotor-stator blade interaction and backward
whirl of the impeller
MAY 2009, Vol. 131 / 051101-5
106%QBEP
0.02
900 rpm
800 rpm
profile A
700 rpm
profile B
625 rpm
0.03
600 rpm
96%
0.01
FY / (N2D4)
0.04
uniform
700 rpm
118%
0.00
F / (N2D4)
0.03
74%
68%
50%
106%
450 rpm
0.00
-0.02
-0.02
500 rpm
0.02
0.01
82%
-0.01
550 rpm
-0.01
0.00
0.01
0.02
0.03
0.0
0.2
FX / (N2D4)
0.4
0.6
0.8
1.0
1.2
Q / QBEP
Fig. 19 Normalized steady fluid-induced force, for different

flow rates and suction flow profiles
Fig. 21 Magnitude of normalized steady fluid-induced force

for different shaft speeds and suction velocity profile B
al. 20. The direction of the force is opposite for the two nonuniform suction flow profiles A and B, consistent with the two profiles being approximately inversed in vertical direction.
Further measurements for suction profile B indicative of the
suction flow profile in practical waterjet applications are presented in Figs. 2022, for different shaft speeds. Note that the
steady fluid-induced force scales with the square of the shaft
speed. The direction of the force vector with respect to the horizontal axis is 35 45 deg, for flow rates above 50% QBEP. For
lower flow rate the direction deviates, presumably due to inlet
recirculation, which influences the suction velocity profile.
lection of the drive train components should then allow for adjustment of the intended shaft speed in order to compensate for the
reduced head.
A second phenomenon related to nonuniform suction flow is the
radial loading of the impeller. It is of interest to compare the
magnitude of the forces measured in this investigation to forces
mentioned in literature for pumps of the same type. An overview
of force measurements on pump shafts can be found in, e.g., Ref.
28. The magnitude of forces is customarily divided into steady
and unsteady components in the inertial frame of reference. Figure 23 presents a comparison of the steady force components, as a
function of flow rate. Clearly, at a larger flow rate, the measured
force is much higher than the values normally anticipated in these
pumps. A high radial loading not accounted for will lead to excessive wear of bearings and seals and even rubbing of the impeller
at the casing wall. Moreover, the force being steady in the inertial
frame of reference is potentially harmful to the shaft, since it may
lead to damage through fatigue failure.
A second remark on the radial loading of the impeller concerns
the direction of the force vector. From the measurements it was
concluded that the impeller undergoes a radial force with an upward component. Depending on the type of suction velocity profile this force may be in upward direction entirely. For example,
consider a double-suction pump with its suction nozzle in an unfavorable direction, or a waterjet system in a ship during maneu-
Discussion and Conclusion
An important observation in this study is that the global performance of a pump is influenced by the type of inflow velocity
profile. For the relatively mild nonuniformity in the experiments,
the head, torque, and axial force were reduced by a few percent.
And although the efficiency of the pump was not reduced, it still
has an important practical implication. Many pumping systems
that face nonuniform inflow conditions during service are selected
based on performance under uniform inflow, e.g., sump pumps or
cooling water pumps. The reason is that most pump manufacturers
use standard closed-loop test-rigs with straight suction piping. Se0.03
900 rpm
140
800 rpm
700 rpm
450 rpm
0.00
625 rpm
600 rpm
550 rpm
0.01
700 rpm
100
angle [degr]
FY / (N2D4)
600 rpm
500 rpm
800 rpm
120
625 rpm
0.02
900 rpm
80
550 rpm
500 rpm
60
450 rpm
40
20
0
-0.01
-0.01
0.00
0.01
0.02
0.03
FX / (N2D4)
Fig. 20 Vector plot of normalized steady fluid-induced force

for different flow rates and shaft speeds, and suction velocity
profile B
051101-6 / Vol. 131, MAY 2009
-20
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Q / QBEP
Fig. 22 Direction of steady fluid-induced force vector for different shaft speeds and suction velocity profile B
0.05
900 rpm
800 rpm
700 rpm
0.04
625 rpm
600 rpm
F / (N2D4)
550 rpm
500 rpm
0.03
450 rpm
0.02
Glich (mixed-flow,
vaned casing)
0.01
0.00
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Q / QBEP
Fig. 23 Comparison of radial forces as given in literature with

measured values of Fig. 21
vering. If not properly accounted for in the design phase of these

pumps, the effect may lead to the bearings being unloaded at
times, and thus to high levels of vibration and noise.
The measurement of hydrodynamic forces led to the conclusion
that the rotor-stator blade passing excitation force induces a backward whirling motion of the impeller at a frequency of six times
shaft frequency. It is anticipated by several authors that, from a
theoretical analysis, it is to be expected that rotor-stator systems
with a number of stator vanes exceeding the number of rotor
blades by one may exhibit perturbations propagating in the reverse direction at blade passing frequency 29,30. However, the
resulting backward rotor whirl was never confirmed experimentally up to now. A detailed account of hydrodynamic force measurements will be presented in a future paper.
Acknowledgment
The author would like to thank Wrtsil Propulsion Netherlands
for providing the scale-model pump and supporting the publication of this paper.
Nomenclature
gravitational acceleration m / s2
impeller outer diameter m
force N
manometric pump head m
static pump head m
shaft speed rps
flow rate m3 / s
torque N m
Reynolds number based on impeller tip speed,
Re = D2 / v
Rev Reynolds number based on pipe velocity and
diameter
kinematic viscosity m2 / s
mass density kg/ m3
g
D
F
Hm
Hs
N
Q
T
Re
Subscripts and Superscripts

ax axial direction
r radial direction
X , Y x-and y-components in lateral direction, in the
steady frame of reference
References
1 Ehrich, F. F., and Childs, D. W., 1984, Self-Excited Vibrations in High Performance Turbomachinery, Mech. Eng. Am. Soc. Mech. Eng., 106, pp.
6679.
2 Agostinelli, A., Nobles, D., and Mockridge, C. R., 1960, An Experimental
Investigation of Radial Thrust in Centrifugal Pumps, ASME J. Eng. Power,
82, pp. 120126.
3 Iversen, H. W., Rolling, R. E., and Carlson, J. J., 1960, Volute Pressure
Distribution, Radial Force on the Impeller and Volute Mixing Losses of a
Radial Flow Centrifugal Pump, ASME J. Eng. Power, 82, pp. 136144.
4 Domm, U., and Hergt, P., 1970, Radial Forces on Impeller of Volute Casing
Pumps, Flow Research on Blading, L. S. Dzung, ed., Elsevier, Netherlands,
pp. 305321.
5 Hergt, P., and Krieger, P., 1969, Radial Forces in Centrifugal Pumps With
Guide Vanes, Proc. Inst. Mech. Eng., 184, pp. 101107.
6 Uchida, N., Imaichi, K., and Shirai, T., 1971, Radial Force on the Impeller of
a Centrifugal Pump, Bull. JSME, 14, pp. 11061117.
7 Chamieh, D., Acosta, D. S., Brennen, C. E., and Caughey, T. K., 1985, Experimental Measurements of Hydrodynamic Radial Forces and Stiffness Matrices for Centrifugal Pump Impeller, ASME J. Fluids Eng., 1073, pp. 307
315.
8 Jerry, B., Acosta, A. J., Brennen, C. E., and Caughey, T. K., 1985, Forces on
Centrifugal Pump Impellers, Proceedings of the second International Pump
Symposium, Houston, TX, Apr. 30May 2.
9 Guinzburg, A., Brennen, C. E., Acosta, A. J., and Caughey, T., 1993, The
Effect of Inlet Swirl on the Rotordynamic Shroud Forces in a Centrifugal
Pump, ASME J. Eng. Gas Turbines Power, 115, pp. 287293.
10 Uy, R. V., and Brennen, C. E., 1999, Experimental Measurements of Rotordynamic Forces Caused by Front Shroud Pump Leakage, ASME J. Fluids
Eng., 121, pp. 633637.
11 Hsu, Y., and Brennen, C. E., 2002, Effect of Swirl on Rotordynamic Forces
Caused by Front Shroud Pump Leakage, ASME J. Fluids Eng., 124, pp.
10051010.
12 Ohashi, H., Imai, H., and Tsuchihashi, T., 1991, Fluid Force and Moment on
Centrifugal Impellers in Precessing Motion, ASME FED Fluid Machinery
Forum, Vol. 1193, Portland, OR, pp. 5760.
13 Yoshida, Y., Tsujimoto, Y., Ohashi, H., and Kano, F., 1999, The Rotordynamic Forces on an Open-Type Centrifugal Compressor Impeller in Whirling
Motion, ASME J. Fluids Eng., 121, pp. 259265.
14 Yoshida, Y., Tsujimoto, Y., Yokoyama, D., Ohashi, H., and Kano, F., 2001,
Rotordynamic Fluid Force Moments on an Open-Type Centrifugal Compressor Impeller in Precessing Motion, Int. J. Rotating Mach., 7, pp. 237251.
15 Yoshida, Y., Tsujimoto, Y., Morimoto, G., Nishida, H., and Morii, S., 2003,
Effects of Seal Geometry on Dynamic Impeller Fluid Forces and Moments,
ASME J. Fluids Eng., 125, pp. 786795.
16 Suzuki, T., Prunires, R., Horiguchi, H., Tsukiya, T., Taenaka, Y., and
Tsujimoto, Y., 2007, Measurements of Rotordynamic Forces on an Artificial
Heart Pump Impeller, ASME J. Fluids Eng., 129, pp. 14221427.
17 Toyokura, T., 1961, Studies on the Characteristics of Axial-Flow Pumps,
Bull. JSME, 4, pp. 287340.
18 Badowski, H. R., 1970, Inducers for Centrifugal Pumps, Worthington
Canada, Ltd., Internal Report.
19 Del Valle, D., Braisted, D. M., and Brennen, C. E., 1992, The Effects of Inlet
Flow Modification on Cavitating Inducer Performance, ASME J. Fluids Eng.,
114, pp. 360365.
20 Bolleter, U., Leibundgut, E., Sturchler, R., and McCloskey, T., 1989, Hydraulic Interaction and Excitation Forces of High Head Pump Impellers, Joint
ASCE/ASME Pumping Machinery Symposium, San Diego, CA, pp. 187193.
21 Fujii, A., Azuma, S., Yoshida, Y., Tsujimoto, Y., and Laffite, S., 2002, Unsteady Stress of 4-Bladed Inducer Blades and the Effect of Inlet Flow Distortion, JSME Int. J., Ser. B, 45, pp. 4754.
22 Seil, G. J., Fletcher, C. A., and Doctors, L. J., 1995, The Application of
Computational Fluid Dynamics to Practical Waterjet Propulsion System Design and Analysis, Proceedings FAST95, pp. 13791389.
23 Verbeek, R., and Bulten, N. W. H., 1998, Recent Developments in Waterjet
Design, Proceedings Rina Waterjet Propulsion Conference, Amsterdam.
24 Brandner, P., and Walker, G. J., 2001, A Waterjet Test Loop for the Tom Fink
Cavitation Tunnel, Proceedings Waterjet Propulsion III Conference, Gothernburg.
25 Bulten, N. W. H., Verbeek, R., and van Esch, B. P. M., 2006, CFD Simulations of the Flow Through a Waterjet Installation, Int. J. Marit. Eng., 148, pp.
2333.
26 Kotansky, D. R., 1966, The Use of Honeycomb for Shear Flow Generation,
AIAA J., 4, pp. 14901491.
27 Bulten, N. W. H., and van Esch, B. P. M., 2007, Fully Transient CFD Analyses of Waterjet Pumps, Marine Technology, Vol. 443, pp. 185193, Society
of Naval Architects and Marine Engineers SNAME.
28 Glich, J. F., 1999, KreiselpumpenEin Handbuch fr Entwicklung, Anlageplanung und Betrieb, Springer-Verlag, Berlin.
29 Dubas, M., 1984, ber die Erregung infolge der Periodizitt von Turbomachinen, Ing.-Arch., 54, pp. 413426.
30 Brennen, C. E., 1994, Hydrodynamics of Pumps, Oxford University Press,
Oxford.
MAY 2009, Vol. 131 / 051101-7
Experimental and Computational

Study of Oscillating Turbine
Cascade and Influence of
Part-Span Shrouds
X. Q. Huang1
L. He2
University of Durham,
Durham DH1 3LE, UK
David L. Bell
Alstom Power Ltd.,
Rugby CV21 2NH, UK
This paper presents a combined experimental and computational study of unsteady flows
in a linear turbine cascade oscillating in a three-dimensional bending/flapping mode.
Detailed experimental data are obtained on a seven-bladed turbine cascade rig. The
middle blade is driven to oscillate and oscillating cascade data are obtained using an
influence coefficient method. The numerical simulations are performed by using a 3D
nonlinear time-marching NavierStokes flow solver. Single-passage domain computations
for arbitrary interblade phase angles are achieved by using the Fourier shape correction
method. Both measurements and predictions demonstrate a fully 3D behavior of the
unsteady flows. The influence of the aerodynamic blockage introduced by part-span
shrouds on turbine flutter has been investigated by introducing flat plate shaped shrouds
at 75% span. In contrast to practical applications, in the present test configuration, the
mode of vibration of the blades remains unchanged by the introduction of the part-span
shroud. This allows the influence of the aerodynamic blockage introduced by the partspan shroud to be assessed in isolation from the change in mode shape. A simple shroud
model has been developed in the computational solver. The computed unsteady pressures
around the shrouds are in good agreement with the experimental data, demonstrating the
validity of the simple shroud model. Despite of notable variations in local unsteady
pressures around the shrouds, the present results show that the blade aerodynamic damping is largely unaffected by the aerodynamic blockage introduced by part-span shrouds.
DOI: 10.1115/1.3111254
Introduction
In modern blading designs, there has been an increasing need to

adequately predict blade flutter stability, particularly for low pressure compressors and turbines with long blades. In recent years,
3D computational modeling tools for blade flutter predictions
have been actively developed, for example, Gerolymos 1, Hall
and Clark 2, He and Denton 3, and Vahdati and Imregun 4.
There has been some considerable effort in experimental work to
produce test data e.g., Blcs and Fransson 5, Buffum et al. 6,
and Lepicovsky et al. 7 for both code validation and advancing
our physical understanding of flutter mechanisms, but most of the
tests were confined to 2D cascades with data taken at midspans.
Given the significant progress in development in advanced 3D
computational methods, it is, however, recognized that there has
been notable lack of detailed 3D experimental data for method
validations.
Bell and He 8 experimentally illustrated the importance of 3D
effects in aeroelastic applications from test data, which were obtained on a single turbine blade oscillating in a profiled channel.
Yang and He 9 conducted a flutter experiment on a linear compressor cascade oscillating in a 3D bending mode. They pursued a
one oscillating blade strategy and the tuned cascade data were
constructed using a linear superposition based on the influence
coefficient method 10. Their results showed strong effects from
both the 3D unsteady flow behavior and the aerodynamic coupling
1
Present address: Northwestern Polytechnical University, China.

Corresponding author. Present address: Dept. of Engineering Science, Oxford
University, Oxford OX1 3PJ, UK.
JOURNAL OF FLUIDS ENGINEERING. Manuscript received September 10, 2006; final
manuscript received December 8, 2008; published online April 10, 2009. Assoc.
Editor: Yu-Tai Lee.
2
between blades. Vogt 11 also Vogt and Fransson 12 carried

out a test using the influence coefficient method on an annular
sector cascade featuring a 3D background flow. From the observations on influence coefficients on the middle blade 3D tuned
cascade data are not available for their studies, they attributed the
3D behavior of increasing response magnitude from hub to tip
observed at bending modes to the respective blade mode shape.
The locally varying response magnitude observed at a torsion
mode was suggested to be due to the 3D effects of the mean flow
field. To further clarify the main source of 3D unsteady effects, it
would be useful to investigate a case where the baseline mean
flow would be largely of 2D nature. Prismatic blading profiles
have been previously used for a single turbine blade 8 and for a
compressor cascade 9, where the 3D unsteady effects have been
identified to be due to the instantaneous spanwise interactions. But
no such work has been carried out for a turbine cascade. It should
be mentioned that the nominal setting of the present experiment is
similar to that of Yang and He 9. However, high quality of data
and a first of its kind effort on a turbine cascade make it a useful
complement to test cases currently available in the turbomachinery aeroelastic community.
Regarding turbine flutter prediction, another issue of interest is
the modeling requirement for part-span shrouds. To prevent flutter
from occurring, passive measures are often introduced at the final
blade design stage. Of them, using part-span shrouds dampers
or snubbers is one typical option. The presence of part-span
shrouds would introduce blockage to the radial propagation of
unsteady flow disturbances. It is, however, not clear, how a partspan shroud would behave and affect the blade aerodynamic
damping. To the authors knowledge, none of previously published unsteady CFD analyses has included modeling the shrouds
in blade flutter predictions.
The objectives of the present work are twofold. First, the de-
MAY 2009, Vol. 131 / 051102-1
Table 1 Blade profile
Fig. 1 Low-speed turbine cascade rig
tailed 3D experimental data are obtained on a low-speed linear

turbine cascade through applying the influence coefficient method.
The experimental cases are numerically simulated by a 3D timemarching NavierStokes method for validation purposes. Second,
the modeling issue regarding part-span shrouds is addressed also
experimentally and computationally. In Secs. 2.12.3, the experimental rig, instrumentation, the data acquisition/processing, and
the results for the nominal condition will be first described. This
will be followed by the computational methods, results, and discussions for the nominal condition. Finally, the study on part-span
shrouds will be presented.
Experimental Study
2.1 Low-Speed Test Rig. A low-speed test facility, shown in

Fig. 1, is commissioned in the present experimental study. In this
rig, air flow is driven by a centrifugal fan powered by an 11 kW
ac motor; speed of 01500 rpm through a two-stage slatted diffuser and into a large settling chamber. The flow, under a propelling force generated by the high pressure in the settling chamber,
accelerates through a shaped contraction a contraction ratio of
7.5:1 and discharges into the working section. In this setup, uniform flow is developed at inlet to the working section, through
appropriate design of the contraction profile, and the deployment
of screens and honeycomb before and after the contraction, respectively. The wind tunnel is of a variable speed, with the mass
flow being accurately controlled by a digital three-phrase inverter.
The working section is a self-contained linear turbine cascade.
It consists of seven prismatic turbine blades, Perspex sidewalls,
and a wooden frame. The blade profile Table 1 is a bespoke
low-speed design, which is developed within the framework of the
present investigation. The cascade data and operational conditions
are summarized in Table 2. The top and bottom walls of the cascade are profiled to simulate the two adjacent blades in cascade;
therefore, eight blade passages are formed, with four passages on
each side of the middle reference blade. Bleed slots are applied
upstream of blades to break down the boundary layer development on the sidewalls whereas no such effort is made to remove
the boundary layer on the profiled top and bottom walls. Two
adjustable tailboards, acting as the extension of top and bottom
walls, are used to help establish blade-to-blade periodicity. To
achieve a 3D bending condition, the reference blade is hinged at
the hub and driven at the tip end. This allows a linear variation in
bending amplitude along the span, corresponding to the first bending mode. A sinusoidal oscillation rate is enforced by using a
single bar crank connection between a dc motor shaft and a rod
protruding from the blade tip. Since the hinge at the hub is
mounted outside the working section, it should be noted that a
small displacement is introduced to the hub of the oscillating
blade. Indeed, the amplitude at the hub is a tenth of that at the tip.
To allow the small blade translational movement through the wall
051102-2 / Vol. 131, MAY 2009
X
mm
Y
mm
0.00
0.11
0.48
0.74
2.25
6.16
9.91
13.53
16.16
19.58
22.89
26.14
29.32
32.44
35.49
38.49
41.43
44.32
47.15
49.93
52.65
55.32
57.94
60.50
63.01
65.46
67.85
70.17
72.43
74.61
76.73
78.82
80.85
82.85
84.81
86.73
88.62
90.48
92.29
94.95
97.50
99.56
101.55
103.45
105.28
106.69
108.03
109.32
110.28
110.30
110.29
110.18
109.96
109.66
109.32
108.98
108.68
108.45
108.37
107.99
106.46
104.48
102.40
100.25
98.04
95.80
93.54
91.23
88.86
86.43
83.94
92.26
92.91
93.46
93.67
94.46
96.28
97.69
98.76
99.37
99.93
100.26
100.41
100.38
100.20
99.87
99.38
98.76
98.00
97.11
96.08
94.91
93.62
92.19
90.62
88.91
87.05
85.04
82.87
80.53
78.00
75.34
72.56
69.67
66.67
63.59
60.42
57.16
53.82
50.37
45.02
39.44
34.60
29.58
24.37
18.96
14.48
9.85
5.08
1.20
1.02
0.85
0.52
0.25
0.07
0.00
0.05
0.22
0.48
0.64
1.74
5.93
10.94
15.73
20.32
24.79
29.18
33.49
37.70
41.76
45.67
49.42
Table 1
Continued.
X
mm
Y
mm
81.41
78.88
76.34
73.70
70.96
68.12
65.20
62.19
59.10
55.94
52.70
49.39
46.69
43.25
39.75
36.21
33.33
30.39
27.41
24.37
20.49
16.55
12.60
8.66
4.71
1.49
0.85
0.34
0.05
53.07
56.72
60.34
63.73
66.85
69.73
72.39
74.85
77.10
79.16
81.04
82.73
83.96
85.33
86.56
87.68
88.44
89.08
89.58
89.96
90.24
90.35
90.45
90.55
90.65
90.73
90.91
91.33
91.93
and the hub displacement, profiled slots were located on the hub
wall. It should be mentioned that sealing measure was taken to
prevent flow leakage though these slots.
2.2 Instrumentation, Data Acquisition and Reduction. The
oscillating blade and one nonoscillating blade are instrumented
with pressure tappings at six spanwise sections 10%, 30%, 50%,
70%, 80%, and 95% span sections. At each section, 14 pressure
tappings are located on the suction surface SS and 10 on the
pressure surface PS, as indicated in Fig. 2. In total, there are 144
tappings on each blade surface. These pressure tappings are used
for both steady and unsteady pressure measurements. The instrumented nonoscillating blade is interchangeable with all other stationary blades, thereby enabling both steady and unsteady pressure measurements at all blade positions in the cascade.
In this low-speed test facility, realistic reduced frequencies can
be achieved at low frequencies of blade vibration. For example, a
nominal frequency of 10.3 Hz here is equivalent to a reduced
frequency of 0.4, which is based on the blade chord and exit
isentropic velocity. This makes it possible for the unsteady pressure signals to be recorded with off-board pressure transducers.
The rig is equipped with five off-board pressure transducers Sen-
Fig. 2 Pressure tappings
sym 142C01D, 01 psi range, so a total of 30 sets of measurements are required for each blade in one test. Unsteady signals
from the transducers are recorded on a personal computer PC
through an Amplicon PC30G data-logging card. The unsteady
pressure acquisition procedure is synchronized by a signal from
an optical Schmitt trigger.
The unsteady pressure signals are ensemble-averaged over 150
periods. The ensemble-averaged unsteady pressure signals are reduced into their harmonic components using a Fourier series. To
correct the unsteady pressure signal for phase shift and attenuation, along the tubing lengths, a tubing transfer function TTF
scheme, proposed by Irwin et al. 13, is applied. The present
application of the TTF method follows the procedures of SimsWilliams 14 and Yang and He 9.
Once the unsteady pressure measurements are obtained, on
each individual blade, they are superposed to construct the tuned
cascade data by utilizing the influence coefficient method. The
validity of the method for this case has been validated by examining the convergence of influence coefficients on those blades
away from the central oscillating blade and the linearity of the
unsteady pressures at different oscillation amplitudes. As described earlier, all influence coefficients have been reduced into
their harmonic components. Accordingly, the first harmonic component Cp1 for a tuned cascade can be expressed as a sum of the
first harmonic components Cp1n of influence coefficients from
all seven blades:
3
Cp1 =
Blade chord, C
Blade span, h
Pitch length, S
Stagger angle
Inlet flow angle
Vibration mode
Bending amplitude
Reynolds number, Re
Reduced frequency, k
Nominal frequency, f
0.143 m
0.180 m
0.100 m
40.0 deg
0.0 deg
First bending normal to absolute chord
3%C at tip, 0.3%C at hub
2.2 105
0.2, 0.4, 0.6
5.2 Hz, 10.3 Hz, 15.5 Hz
in
n=3
where n is the blade index number, and is the inter blade phase
angle. The direction of blade indexing is the same as that of a
forward traveling wave i.e., that of the rotation if the cascade was
taken from a turbine rotor with index 0 denoting the middle reference blade. Consequently, the sign convention for the interblade
phase angle IBPA is such that a negative IBPA corresponds to a
backward traveling wave mode.
2.3
Table 2 Cascade data and operation conditions
Cp ne
Experimental Results
2.3.1 Steady Flow. The steady flow results are discussed aiming at identifying the main steady flow features. Since the major
unsteady pressure response happens on the central three blades,
which will be verified later, the steady flow results are, therefore,
presented only on the central three blades.
Shown in Figs. 3a3c are the steady flow blade surface pressure distributions for the three middle blades at the flow conditions described in Table 2. The steady flow blade surface pressure
measurements were obtained at six spanwise sections from 10%
to 95% span. For each blade, the blade surface pressure distributions are almost identical for different spanwise sections, although
slight deviations in pressure distribution are observed toward
MAY 2009, Vol. 131 / 051102-3
Fig. 3 Steady pressure distributions at six spanwise sections on three central

blades. a Blade 1, b blade 0, c blade 1, and d midspan section
blade ends over the suction surface. These deviations are believed
to be associated with the endwall effects with slight unloading
present on the suction surface. These slight endwall effects are
attributable to the favorable pressure gradient present in the cascade and the relatively small blade deflection angle. Overall, the
results indicate that the steady flow blade surface pressure response is predominantly two dimensional.
Figure 3d shows the comparison of the steady flow blade
surface pressure distributions at the midspan section on the three
central blades. Because of the predominant two-dimensional feature of the steady flow, this comparison at the midspan section is
sufficient to verify the blade-blade periodicity in the cascade. The
plot demonstrates an excellent comparison on the pressure surface, while a good comparison is achieved on the suction surface
despite some deviation present on the rear half chord of blade 1.
Good blade-blade periodicity achieved in the steady flow surely
forms a sound basis for the unsteady flows presented hereafter.
2.3.2 Unsteady Flow. To assess the uncertainty of unsteady
flow measurements, a series of tests was performed to evaluate the
experimental errors and repeatability. In these tests, five tappings
were chosen from the 70% span of the oscillating blade to provide
a representative range of unsteady pressure response. For each
tapping, 40 sets of measurements were obtained, through the ex-
perimental procedure described in Sec. 2.2. The results showed

that the deviation from mean values fell within a range of 0.08
for the nondimensional amplitude of the first harmonic pressure
and 5 for the phase angle. The corresponding overall standard
deviation is below 0.029 in amplitude and 1.95 deg in phase angle
of unsteady pressure coefficients, indicating an excellent level of
repeatability.
Pitchwise finiteness is an inherent nature of linear cascade.
Therefore, the basic requirement of a linear cascade experiment
when using the influence coefficient method is that the unsteady
pressure response is to attenuate quickly away from the oscillating
blade in the pitchwise direction. This pitchwise convergence in
the present linear cascade is to be evaluated herein.
Figure 4 shows the amplitude normalized by the blade tip
bending amplitude of the first harmonic pressure coefficient at
the midspan section on the five middle blades with blade 0 oscillating at a reduced frequency of 0.4 and a tip bending amplitude
of 3%C. Clearly, the unsteady aerodynamic response converges
quickly away from the reference oscillating blade. On the suction
surface, the amplitude of the first harmonic pressure on the blade
1 the immediate neighbor on the pressure surface side of the
reference blade shows the same order of magnitude as that on
blade 0, while those on blades +1 and 2 have an order of mag-
Fig. 4 Unsteady pressure amplitudes on five central blades
051102-4 / Vol. 131, MAY 2009
Fig. 5 Experimental test for pitchwise convergence:

aerodamping contribution from five central blades k = 0.4,
IBPA= 0 deg
Fig. 7 Experimental test for linearity: relative amplitudes of

the second harmonic pressure coefficient at different spanwise
sections
nitude lower. On the pressure surface, the main unsteady pressure

response happens on blades 0 and 1 and blades 2 sense much
smaller unsteady disturbance. Figure 5 shows the global aerodynamic damping contributions from the middle five blades to the
tuned cascade at a reduced frequency of 0.4 and an IBPA of 0 deg.
The oscillating blade exerts the stabilizing contribution direct
term. At this IBPA, the coupling terms do have the destabilizing
contributions. For different values of IBPA, the direct term would
remain unchanged whereas coupling terms vary as indicated by
Eq. 1. From Fig. 5, it is found that aerodynamic coupling effects
are mainly from the immediate neighbors blades 1 while the
contributions from blade 2 are of negligible magnitudes.
The linearity assumption is the fundamental premise of the application of the influence coefficient method. Moreover, it is a
general interest in the unsteady aerodynamic modeling of oscillating blade flows. Two sets of measurements of unsteady pressure
response at different oscillating amplitudes were performed, in
order to evaluate the linearity assumption of the superposition
employed to reconstruct the tuned cascade data. The values of two
bending oscillation amplitudes at the blade tip were arbitrarily
chosen to be 3% and 5% chord. Figure 6 compares the amplitudes
and phase angles of the first harmonic pressure coefficients on the
reference blade at these two oscillating amplitudes. The results
presented in this figure shows a predominantly linear behavior of
the unsteady pressure response with very good agreement in the
first harmonic pressure coefficients in terms of both amplitude and
phase angle between the two different oscillating amplitudes.
Another measure of linearity of the oscillating blade flows is
the relative amplitude of the second harmonic pressure response.
Figure 7 presents the amplitudes of the second harmonic pressure
coefficients relative to the first on the reference blade for reduced
frequency 0.4. At most locations on the suction surface, the am-
plitudes of the second harmonic pressure reveal to be an order of

magnitude lower than those of the first, which indicates a predominant linear behavior of the unsteady pressure response. On
the suction surface, high relative amplitudes of the second harmonic pressure aft of 80%Cax at the 10% span section, to a less
extent at the 30% span section, are evident, while on other spanwise sections, which have much higher bending amplitudes, do
not reveal similar behavior. Considering this, the high relative
second harmonic pressure response at spanwise sections located
near the hub is believed to be most likely arising from a leakage
flow through the imperfect hub sealing as well as the relative low
pressure response locally near the hub, rather than indicating a
nonlinear unsteady aerodynamic response to the blade oscillation.
In addition, low signal-to-noise ratio is expected toward the trailing edge on the suction surface where the amplitude of unsteady
pressure is small. On the pressure surface, the amplitudes of the
second harmonic pressure are relatively bigger, which certainly is
attributable to the smaller amplitude of the unsteady pressure response and accordingly the lower signal-to-noise ratio. Overall,
the unsteady pressure response to the blade oscillation is largely
linear on both surfaces.
The amplitudes and phase angles of the first harmonic pressure,
presented in Figs. 8 and 9, are for an IBPA of 60 deg. This
particular IBPA is chosen because it is identified to be the least
stable IBPA. The effects of IBPA on the blade stability are shown
in Fig. 10.
Figure 8a shows the amplitudes of the first harmonic pressure
on the suction surface. Along the blade span, the amplitude of the
first harmonic pressure consistently increases away from the hub
except for the area toward the leading edge. However, the increase
in the amplitude of the first harmonic pressure is not proportional
to the increase in the local vibration amplitude along the span. In
Fig. 6 Experimental test for linearity: first harmonic pressure coefficient at 50% span
section on blade 0 at two bending amplitudes k = 0.4
MAY 2009, Vol. 131 / 051102-5
Fig. 8 Amplitudes of first harmonic pressure at IBPA= 60 deg: a suction surface and b
pressure surface
fact, the amplitude of the first harmonic pressure at 95% span is

only about 2.5 times that at 10% span, while the local vibration
amplitude at 95% span appears to be five times that at 10% span.
This nonproportional unsteady response to the local vibration amplitude, at different blade spans, certainly indicates that the unsteady flow response is of a strong three-dimensional nature. On
the pressure surface, displayed in Fig. 8b, a consistent increase
in the amplitude along the span height is noticed for a range from
55% chord to the trailing edge. Toward the leading edge, the
change of the first harmonic pressure amplitude is not in a clear
trend along the span on both surfaces. This behavior may be associated with a high sensitivity to the local incidence. Overall, the
presence of unsteady pressure peak around midchord of the suction surface and toward the trailing edge of the pressure surface
suggests that the unsteady response is largely driven by the variation of the passage flow area.
In Fig. 9a, the phase angle distributions of the first harmonic
pressure along the chord are plotted at six spanwise sections on
the suction surface. Apart from the 95% span section, the phase
angle shows consistent shift along the chord from 10% to 80%
span. The phase angle becomes more diverse near the trailing
edge, particularly at 10% and 95% span sections, which may suggest some 3D end wall effects. Also, the influence of the imperfection of endwall sealing should be recognized. The phase angles
on the pressure surface Fig. 9b are on the whole insensitive to
the local span height.
Figure 10 shows the global aerodynamic damping for three reduced frequencies tested. The global aerodynamic damping,
which gives a direct indication of aeroelastic stability, is an overall
assessment of the unsteady aerodynamic response to the blade
oscillation. It allows easy comparison of the influence of reduced
frequency on the unsteady pressure response and direct examination of the variation in aeroelastic stability with regard to the
IBPA for each reduced frequency.

Apparently, the variation in the global aerodynamic damping
with regard to the IBPA reveals an S-curve shape as expected of a
summation of sinusoidal functions for each reduced frequency.
The vertical position of the damping S-curve depends on the value
of reduced frequency. Generally, it shifts consistently upward as
well as rightward as reduced frequency increases. For a reduced
frequency 0.2, the present cascade is unstable based on the aerodynamic damping for a range of IBPA from 150 deg to 10
deg. As the reduced frequency increases to 0.4, the cascade is
unstable at a small range of IBPA values around 60 deg,
whereas the cascade is stable at the whole range of IBPA for a
reduced frequency of 0.6.
The damping variation with IBPA is the direct reflection of the
aerodynamic coupling between blades. The consequence of the
aerodynamic coupling is such that blade flutter only picks up the
least stable IBPA, which has the lowest aerodynamic damping.
Upon examining the least stable IBPA for different reduced frequencies, it shows a consistent shift as indicated by the bold line
in the plot. Indeed, the least stable IBPA is identified to be around
80 deg, 60 deg, and 45 deg for the reduced frequencies of
0.2, 0.4, and 0.6, respectively. This is consistent with a quasisteady consideration, in which the least stable IBPA should be
90 15.
Computational Study
3.1 Method Description. The computational fluid dynamics

CFD code used in this study, TF3D, has been developed at University of Durham 16. It solves the 3D Reynolds averaged unsteady NavierStokes equations in time domain. The governing
equations are discretized in space using the cell-centered finite
volume scheme, together with the blend second- and fourth-order
Fig. 9 Phase angles of first harmonic pressure at IBPA= 60 deg: a suction surface b and pressure surface
051102-6 / Vol. 131, MAY 2009
Fig. 10 Global aerodynamic damping at three reduced frequencies bold line: trend of the least stable IBPA with regard
to reduced frequency
Fig. 12 Steady pressure distributions
NFou
artificial dissipations 17. Temporal integration of the discretized

equations is carried out using the second-order four-step Runge
Kutta approach. The convergence is speeded up by a timeconsistent multigrid technique 16,18.
The computational domain consists of a single passage with the
blade positioned at the middle of the passage, as shown in Fig. 11.
The present single-passage method employs a Fourier decomposition based shape-correction scheme 19 to deal with phaseshifted periodic boundary conditions. The unsteadiness concerned
here is induced by the blade harmonic vibration and satisfies a
temporal and circumferentially spatial periodicity. Being periodic
in time, any flow variable U can be decomposed into a time and an unsteady part U
, which can be approxiaveraged part U
mated by a set of Fourier series in time.
x, ,r + U
x, ,r,t
Ux, ,r,t = U
NFou
x, ,r,t =
U
A x, ,rsinnt + B x, ,rcosnt
n
n=1
where An and Bn are the nth order Fourier coefficients; NFou is the
order of the Fourier series; x, , and r are the axial, circumferential, and radial coordinates, respectively, and t is the physical time.
The phase-shifted periodic conditions are then defined as
Fig. 11 Single-passage computational domain
x, G,r,t =
U
A x, ,rsinnt
n
n=1
+ Bnx, ,rcosnt
where G is the circumferential angular spacing between two

neighboring blades and is the IBPA of the unsteadiness.
During the computation, the Fourier coefficients, which are
used to approximate the flow variables Eq. 3, are stored and
updated at the periodic boundary inner cells IL and IU. To apply
periodic boundary conditions, flow variables at the dummy cells
DL and DU need to be determined at every time step. According
to Eq. 4, the flow variables at dummy cells DL and DU are
obtained from the stored Fourier coefficients at inner cells IU and
IL, respectively.
The shape-correction method is able to deal with multiple disturbances 20 and its applications in multirow configurations can
be found in Refs. 2123. The computational mesh is of a
simple-H type with a density of 107 41 51 nodes in the
streamwise, pitchwise, and spanwise directions, respectively. The
numerical tests on mesh dependence show that this mesh density
is adequate 24.
At the inlet, stagnation parameters and flow angles are specified. At the exit, back pressure is set at a fixed value. It is well
known that the density based compressible flow solver is unsuitable to calculate flows of very low Mach number. The flow velocity is therefore elevated to avoid the difficulty. The exit Mach
number, prescribed in the calculations, is around 0.3, at which the
flow compressibility influence is still negligible. On blade and
endwall surfaces, the tangential velocity is allowed to slip but
restricted by the wall shear stress, which is determined from an
approximation of the logarithmic law 25. For unsteady computations, the tested reduced frequency is matched and the experimental Reynolds number of 2.2 105 retained. The blade oscillating amplitude is, however, reset to 1% chord for a better
convergence of solution.
3.2 Comparison of Results. Figure 12 presents the predicted
and measured steady flow blade pressure distributions at 50% and
95% span sections. These results are representative of the whole
blade surface, at least from 10% to 95% span, and demonstrate
good correlation between the measurements and the predictions.
Figures 13 and 14 show the measurements and numerical predictions of the first harmonic pressure obtained at a reduced frequency of 0.4 for the least and the most stable IBPAs, respectively. The results are shown at three spanwise sections. For each
section, the left plot is the amplitude and the right is the phase
angle of first harmonic pressure. For both the least and most stable
IBPAs, the predicted amplitudes of the first harmonic pressure
demonstrate very good correlation to the experimental measurements at most locations on both pressure and suction surfaces see
MAY 2009, Vol. 131 / 051102-7
Fig. 15 Aerodamping variation along blade span k = 0.4
Fig. 13 Amplitude and phase of unsteady pressure at k = 0.4

and the least stable IBPA = 60 deg: a 10% span section,
b 50% span section, and c 95% span section
Figs. 13 and 14. Discrepancy between the predicted and measured amplitudes of the first harmonic pressure is observed at
Fig. 14 Amplitude and phase of unsteady pressure at k = 0.4

and the most stable IBPA = 120 deg: a 10% span section,
b 50% span section, and c 95% span section
051102-8 / Vol. 131, MAY 2009
sections near the cascade endwalls. The predicted phase angles of

the first harmonic pressure are in good agreement with the test
data over most locations on both pressure and suction surfaces
except the area toward the trailing edge.
The predicted and measured variations in local aerodynamic
damping along the blade span are presented in Fig. 15 for the least
and most stable IBPAs. The local damping gives an aggregate
indication of the quality of the amplitude and phase angle prediction of the first harmonic pressure at different spanwise sections.
Clearly, this plot shows excellent local damping prediction. The
overall aerodynamic damping prediction is shown in Fig. 16 and a
good agreement with the data is achieved over the whole range of
IBPA, although the predicted damping S-curve has a slight shift
from the measured.
Effects of Part-Span Shrouds
The cascade has been modified to include 75% span shrouds in

blade passages, as shown in Fig. 17. The shroud profile is chosen
to be of flat plate for simplicity of application. The specifications
of the part-span shrouds are summarized in Table 3. To accommodate the movement of the oscillating blade, small gaps 2 mm are
allowed in the two immediate neighboring passages of the oscillating blade. At the gap location, a thin flexible sliding card is
applied to minimize the flow leakage through the gap. It should be
noted that, in a real machine, the configuration of part-span
shrouds is more complex and the presence of part-span shrouds
would cause distortion to the steady flow field and result in aerodynamic loss. With respect to the negative impact of part-span
shrouds on the aerothermal performance, it is beyond the scope of
the present study. Also, the potential influence of the baseline
steady loading change resulting from the presence of part-span
shrouds on the unsteady flow response is not included. The simplification made herein is to highlight the interference effects of
part-span shrouds on the unsteady pressure propagation in an oscillating cascade with a 3D bending mode.
Fig. 16 Overall aerodynamic damping prediction k = 0.4
Fig. 17 Part-span shrouded linear turbine cascade endwall

on the blade tip side is removed for clarity
The shroud modeling is achieved by applying solid wall boundary conditions at shroud surface locations. Figure 18 shows the
computational mesh used for the study of effects of part-span
shrouds. It consists of 132 nodes in the axial direction, 41 in the
pitchwise direction and 51 in the spanwise direction. The mesh
clustering is applied toward the shroud location in the spanwise
direction and finer mesh spacing is also used around the shroud
leading and trailing edges. For the comparison purpose, the same
mesh density is utilized in both calculations with and without
shrouds to eliminate the influence of mesh dependency on the
results. During blade oscillation, the shrouds are assumed to
stretch in a plane of constant span height. In other words, the
analyzed shroud configuration is otherwise identical to the tested
one except that the radial movement of shrouds is neglected. The
part-span shroud plates are thin and do not induce noticeable
steady pressure loading change on the blade surface. The results
presented are, therefore, confined to those for the unsteady flow.
Figure 19 shows the amplitudes of the first harmonic pressure at
50%, 70%, 80%, and 95% span sections for settings with and
without part-span shrouds. For each section, the left plot shows
the experimental results and the right shows the predictions. Upon
inspection of the set of the first harmonic pressure, it is found that
the influence of part-span shrouds manifests itself mainly in terms
of the change in the pressure amplitude on the suction surface.
The presence of the part-span shrouds results in reduced ampli-
tudes for the sections below 75% span and increased amplitudes
for those sections above 75% span. It is evident that the computational predictions for settings with and without part-span
shrouds reveal exactly the same behavior as that observed in the
measurement. This demonstrates the overall validity of the present
simple shroud model.
The opposite trends in the amplitude change of the first harmonic pressure for spanwise sections below and above the partspan shrouds reflect a 3D behavior of the unsteady pressure propagation. Recall the observation on the results in Fig. 8 that the
amplitude of the first harmonic pressure is nonproportional to the
local blade vibration amplitude along the blade span. The basic
mechanism is that the spanwise unsteady interaction results in an
instantaneous redistribution of the unsteady pressure. The unsteady loading is alleviated toward the tip and enhanced toward
the hub, which can be simply regarded as an unsteady 3D relief.
The presence of part-span shrouds, on the other hand, blocks a
radial propagation of unsteady pressures and hence reduces the
3D unsteady relief effects. Consequently, the amplitude of the first
harmonic pressure with part-span shrouds is larger than that of the
nominal case for sections above 75% span, whereas the trend in
the amplitude change is reversed for sections below 75% span.
Figure 20 shows the phase angles of first harmonic pressure for
settings with and without part-span shrouds. Regardless the spanwise locations, the phase angles of the unsteady pressure are
largely unaffected.
The overall influence of the part-span shrouds on the blade
flutter is indicated by the integrated aerodynamic damping. The
experimental data show that the part-span shroud has a small stabilizing effect, and this behavior has been also correctly captured
by the CFD results. The differences in the aerodynamic damping
as caused the stabilizing effects are, however, comparable to or
smaller than the discrepancies between the experimental data and
the CFD predictions for the nominal case as shown in Fig. 16.
Hence, the overall influence of the part-span shroud can be regarded as insignificantly small for the present case studied. This is
largely due to the fact that the part-span shrouds have predominantly local effects on unsteady pressures and hence the aerodynamic damping.
Table 3 Part-span shroud specification
5
Thickness
Length in axial direction
Spanwise location
Axial location
3 mm
60 mm
75% span
18 78%Cax
Fig. 18 Mesh used for computation with part-span shrouds
Concluding Remarks
A linear oscillating turbine cascade is investigated experimentally and computationally. Detailed unsteady pressure measurements are made at six spanwise sections from 10% to 95% blade
height. The steady pressure coefficient distributions reveal a predominant 2D steady background flow, while the unsteady pressure
results demonstrate a strong 3D behavior of unsteady pressure
response to a three-dimensional blade flapping oscillation. The
computational solutions are obtained from a 3D unsteady timemarching NavierStokes flow solver using a single-passage domain. The computational results are in good agreement with experimental data.
A further study is performed to investigate the influence of the
aerodynamic blockage that is introduced by the application of
part-span shrouds, on the blade flutter characteristics in the linear
turbine cascade with the same blade oscillation mode shape. The
part-span shrouds are positioned at 75% span height. Notable
variations in local unsteady pressure at spanwise sections near the
shroud location, as shown in the experimental data, are well captured by the computational method with a simple part-shroud
model. The influence of the aerodynamic blockage associated with
the part-span shrouds on the overall aerodynamic damping, however, seems insignificant for the present cascade.
MAY 2009, Vol. 131 / 051102-9
Fig. 19 Effects of part-span shrouds on amplitude of the unsteady pressure response. Left: experiment. Right: calculation k = 0.4, = 60 deg: a 50% span section, b 70% span section, c 80% span section, and d 95% span section.
Fig. 20 Phase angle of first harmonic pressure for settings with and without partspan shrouds: a 50% span section, b 70% span section, c 80% span section, and
d 95% span section
051102-10 / Vol. 131, MAY 2009
Acknowledgment
This work was carried out at University of Durham, Durham
UK, sponsored partially by ALSTOM Power Ltd, which is gratefully acknowledged. The authors would also like to thank Dr.
Peter Walker ALSTOM for continued interest and technical discussions
Nomenclature
Ax bending amplitude, nondimensionalized with
chord
Axtip bending amplitude at blade tip, nondimensionalized with chord
Ap1 amplitude of the first harmonic pressure, Pa
C blade chord length
Cax axial blade chord length
Cp blade surface pressure coefficient, Cp = P
P2 / P01 P2
Cp1 amplitude of the first harmonic pressure coefficient, Cp1 = Ap1 / P01 P2Axtip
h blade span, m
k reduced frequency, k = C / Vref
P2 exit static pressure
P01 inlet stagnation pressure
Re Reynolds number
S pitch length
Vref reference isentropic exit velocity, m/s, Vref
= 2P01 P2 /
x axial coordinate, m
circumferential coordinate, deg
phase angle, deg
IBPA, deg
angular frequency, rad/s
aerodynamic damping coefficient, = 1 / hh0ldr
l local aerodynamic damping coefficient,
l = C AxlCp1sin1 / C Axtipds
Subscript
1 inlet parameter; first harmonic

2 exit parameter; second harmonic
References
1 Gerolymos, G. A., 1993, Advances in the Numerical Integration of the ThreeDimensional Euler Equations in Vibrating Cascades, ASME J. Turbomach.,
115, pp. 781790.
2 Hall, K. C., and Clark, W. S., 1993, Linearized Euler Predictions of Unsteady
Aerodynamic Loads in Cascades, AIAA J., 31, pp. 540550.
3 He, L., and Denton, J. D., 1994, Three-Dimensional Time-Marching Inviscid
and Viscous Solutions for Unsteady Flows Around Vibrating Blades, ASME
J. Turbomach., 116, pp. 469475.

4 Vahdati, M., and Imregum, M., 1994, Nonlinear Aeroelasticity Analysis Using Unstructured Dynamic Meshes, Symposium on Unsteady Aerodynamics
and Aeroelasticity of Turbomachines, Fukuoka, Japan, Sept.
5 Blcs, A., and Fransson, T. H., 1986, Aeroelasticity in Turbomachines Comparison of Theoretical and Experimental Cascade Results, Communication du
Laboratoire de Thermique Applique et de Turbomachines, Ecole Polytechnique Federale de Lausanne, Report No. 13.
6 Buffum, D. H., Capece, V. R., King, A. J., and El-Aini, Y. M., 1998, Oscillating Cascade Aerodynamics at Large Mean Incidence, ASME J. Turbomach., 120, pp. 122130.
7 Lepicovsky, J. and McFarland, E. R. and Capece, V. R. and Hayden, J., 2002,
Unsteady Pressures in a Transonic Fan Cascade Due to a Single Oscillating
Airfoil, ASME Paper No. GT2002-30312.
8 Bell, D. L., and He, L., 2000, Three-Dimensional Unsteady Flow for an
Oscillating Turbine Blade and the Influence of Tip Leakage, ASME J. Turbomach., 122, pp. 93101.
9 Yang, H., and He, L., 2004, Experimental Study on Linear Compressor Cascade With Three-Dimensional Blade Oscillation, J. Propul. Power, 201, pp.
180188.
10 Vogt, D. M., 2005, Experimental Investigation of Three-Dimensional Mechanisms in Low-Pressure Turbine Flutter, Ph.D. thesis, Royal Institute of Technology, Sweden.
11 Vogt, D. M., and Fransson, T. H., 2004, Effect of Blade Mode Shape on the
Aeroelastic Stability of a LPT Cascade, presented at the Ninth National Turbine Engine High Cycle Fatigue HCF Conference, Pinehurst, NC, Mar.
12 Hanamura, Y., Tanaka, H., and Yamaguchi, Y., 1980, A Simplified Method to
Measure Unsteady Forces Acting on the Vibrating Blades in a Cascade, Bull.
JSME, 23180, pp. 880887.
13 Irwin, H. P. A. H., Cooper, K. R., and Girard, R., 1979, Correction of Distortion Effects Caused by Tubing Systems in Measurements of Fluctuation
Pressures, J. Wind. Eng. Ind. Aerodyn., 5, pp. 93107.
14 Sims-Williams, D. R., 2001, Self-Excited Aerodynamic Unsteadiness Associated With Passenger Cars, Ph.D. thesis, School of Engineering, University of
Durham, Durham.
15 Huang, X. Q., He, L., and Bell, D. L., 2008, Effects of Tip Clearance on
Aerodynamic Damping in a Linear Turbine Cascade, J. Propul. Power, 24,
pp. 2633.
16 He, L., 2000, Three-Dimensional Unsteady NavierStokes Analysis of StatorRotor Interactions in Axial-Flow Turbines, Proc. Inst. Mech. Eng., Part A,
214, pp. 1322.
17 Jameson, A., 1983, Numerical Solution of the Euler Solution for Compressible Inviscid Fluids, Princeton University, Report No. MAE 1643.
18 He, L., 1993, New Two-Grid Acceleration Method For Unsteady Navier
Stokes Calculations, J. Propul. Power, 92, pp. 272280.
19 He, L., 1990, An Euler Solution for Unsteady Flows Around Oscillating
Blades, ASME J. Turbomach., 112, pp. 714722.
20 He, L., 1992, Method of Simulating Unsteady Turbomachinery Flows With
Multiple Perturbations, AIAA J., 3011, pp. 27302735.
21 Li, H. D., and He, L., 2005, Blade Aerodynamic Damping Variation With
Rotor-Stator Gap: A Computational Study Using Single-Passage Approach,
ASME J. Turbomach., 127, pp. 573579.
22 Li, H. D., and He, L., 2005, Toward Intra-Row Gap Optimization for One and
Half Stage Transonic Compressor, ASME J. Turbomach., 127, pp. 589598.
23 Huang, X. Q., He, L., and Bell, D. L., 2006, Influence of Upstream Stator on
Rotor Flutter Stability in a Low Pressure Steam Turbine Stage, Proc. Inst.
Mech. Eng., Part A, 220, pp. 2535.
24 Huang, X. Q., 2006, Three-Dimensional Unsteady Flow in Oscillating Turbine Blade Row, Ph.D. thesis, University of Durham, Durham.
25 Denton, J. D., 1992, The Calculation of Three-dimensional Viscous Flow
Through Multistage Turbomachines, ASME J. Turbomach., 114, pp. 1826.
MAY 2009, Vol. 131 / 051102-11
Seyed Mohammad
Hosseini
Department of Energy,
Division of Fluid Mechanics,
Lund University,
Lund 22100, Sweden
e-mail: seyed_mohammad.hosseini@energy.lth.se
Kazuhisa Yuki
Hidetoshi Hashizume
Department of Quantum Science and Energy,
Tohoku University,
6-6-01 Aza-Aoba, Aramaki,
Sendai 980-8579, Japan
Experimental Investigation of
Flow Field Structure in Mixing
Tee
T-junction is one of the familiar components in the cooling system of power plants with
enormous capability of high-cycle thermal fatigue. This research investigates the structure and mixing mechanism of turbulent flow in a T-junction area with a 90 deg bend
upstream. According to the wide distribution of turbulent jets in the T-junction, a reattached jet was selected previously as the best representative condition with the highest
velocity fluctuation and the most complex structure. For considering the mixing mechanism of re-attached jet, T-junction is subdivided into few lateral and longitudinal sections, and each section is visualized separately by particle image velocimetry technique.
Corresponding to the experimental data, the branch flow acts as a finite turbulent jet,
develops the alternative type of eddies, and causes the high velocity fluctuation near the
main pipe wall. Three regions are mainly subject to maximum velocity fluctuation: the
region close to the jet boundaries (fluctuation mostly is caused by KelvinHelmholtz
instability), the region above the jet and along the main flow (fluctuation mostly is caused
by Karman vortex), and the re-attached area (fluctuation mostly is caused by changing
the pressure gradient in the wake area above the jet). Finally, the re-attached area (near
the downstream of wake area above the jet) is introduced as a region with strongest
possibility to high-cycle thermal fatigue with most effective velocity fluctuation on the
main pipe wall above the branch nozzle. DOI: 10.1115/1.3112383
Keywords: fluid mixing structure and interaction, mixing tee, PIV, piping system, turbulent flow
Introduction
This research investigates fluid mixing phenomena and turbulent jet structure in a T-junction area with a 90 deg bend upstream
to understand the basic mechanism of high-cycle thermal fatigue.
The problem of thermal fatigue occurs in the pipes where two
flows with different temperatures mix together. The T-junction is
one of the familiar components with a considerable potential of
thermal fatigue and is used in many thermohydraulic systems such
as combustion engines, turbines, exhaust systems, hydraulicspneumatics, and reheat systems. With due attention to the importance of thermal striping phenomenon in the T-junction area, several experiments and analyses have been performed such as the
evaluation of the thermal fatigue by Faidy 1, the numerical
simulation of the mixing phenomenon by Tanaka 2, the analysis
of the flow field structure by Igarashi 3, and the LES study of the
high-cycle temperature fluctuation by Hu and Kazimi 4. These
researches consider the T-junction as a single component. Since
usually the T-junction is connected to other apparatuses in the
piping systems, the 90 deg bend is chosen as one of the common
components in most piping systems and it was also connected in
the upstream of the T-junction in some familiar leakage accidents
of power plants such as Phenix and Civaux in France. A secondary
flow is formed by this 90 deg bend and has strong effects on the
flow field along with the mixing mechanism. These effects were
studied previously by Hosseini and Yuki 57. The interaction
between main and branch flows including the effects of secondary
flow formed an unstable area above the branch nozzle. This area
was shown to be highly loaded and subjected to the temperature
JOURNAL OF FLUIDS ENGINEERING. Manuscript received February 13, 2008; final manuscript received February 25, 2009; published online April 13, 2009. Assoc. Editor:
Juergen Kompenhans. Paper presented at the 2007 ASME Fluids Engineering Division Summer Meeting and Exhibition FEDSM2007, San Diego, CA, July 30
August 2, 2007.
fluctuation due to some experimental and numerical researches

8,9. Depending on the momentum ratio of the branch flow to the
main flow, mainly four different types of turbulent jets were identified by Hosseini and Yuki 10 such as wall jet, re-attached jet,
turn jet, and impinging jet. Some parameters had main impact on
the classification of turbulent jets, namely, secondary flow, temperature difference, pipe geometries, velocity ratio, and momentum ratio. These parameters were defined experimentally by Yuki
et al. 11, and the following results were pointed out: The turn jet
has the lowest velocity fluctuation near the wall, the 90 deg bend
has positive effects for decreasing fluctuation near the main pipe
wall, and connecting the branch pipe closer to the bend helps
decrease the velocity fluctuation near the wall as well as using a
bend with lower curvature ratio.
After evaluating and categorizing many flow patterns in the
T-junction, the fluid mixing condition of re-attached jet showed
the most complex structure and mechanism in comparison with
the other turbulent jets and was selected to provide the precise
mechanism of velocity fluctuation. It was difficult to completely
figure out the transport mechanism of turbulent flow toward the
wall only by the information of the whole flow field so the measuring area was subdivided into small regions to get more accurate
close-up data. This close-up measurement method was really helpful in explaining the turbulent wall flow in great detail. According
to the strong effect of secondary flow on the mixing mechanism,
several lateral sections beside the longitudinal sections are visualized. The interaction between the main and the branch flows creates the complex turbulent structure including the wide distribution types of eddies. The branch jet has three parts based on the
theory of turbulent jets by Abramovich 12 such as main, transitional, and initial parts. The initial part is formed by the potential
core of jet and acts as flexible obstacle against the main flow.
Unsteady separation flow over the bluff body of initial part develops some vortices that look like Karman vortices. The steep velocity gradient between the main and the branch flows in both
MAY 2009, Vol. 131 / 051103-1
Fig. 1 Experimental apparatuses: 1 main pipe, 2 branch

pipe, 3 tee junction, 4 90 deg bend, 5 main pump, 6
branch pump, 7 heat exchanger, 8 heating tank, 9 mixing
tank, 10 straightener tank, 11 measuring window, and 12
water jacket
initial and transient parts of jet causes KelvinHelmholtz instability and creates the high vorticity field in the same area. In addition
there are many other eddies in the downstream of T-junction that
mainly are created by the following mechanisms: the oscillation
of the flexible body of jet in the transitional part caused by the
frequent drag and lift forces, the structure-interaction between the
flow and piping system, the secondary flow created by 90 deg
bend in the upstream, the self-developing behavior of turbulent jet
in the main part, etc. Finally, the T-junction area was subdivided
into few regions based on their intensity of velocity fluctuation
near the wall, and the turbulent behavior of the most intense region was analyzed in more detail along with the fluctuation
mechanism.
Experimental Apparatuses
Figure 1 shows experimental apparatuses including the fluid

cycle system to mix two fluids with different physical properties.
The mixing area is T-shaped with two main and branch acrylic
pipes that connect to each other at the right angle with 3 mm thick
and square edged. The main pipe diameter, Dm, is 108 mm and the
branch pipe diameter, Db, is 21 mm. The main pipe runs vertically
upward in the test section, which is connected to a measuring
window downstream for visualizing the lateral flow field. The 90
deg bend is installed in the upstream of the T-junction. This bend
is a stainless steel with 1.41 curvature ratio. There is a straightener
tank which is made of acrylic plates with 30 30 30 cm3 size
to straighten the main flow before it is injected to a long main pipe
and the bend 13Dm, respectively. The branch pipe is also long
enough 60Db to inject fully-developed flow into the mixing area.
The mixing tank is placed to insert small tracers and remove
bubbles in the main cycle. This tank is connected to other apparatuses with five pipe connections heat-exchanger pipe, emergency inlet pipe, heating tank pipe, main pump pipe, and drying
outlet pipe. There are two main and branch pumps with 600 l/min
and 75 l/min maximum flow rates. The mean velocity of these
flows is adjusted by a flow-rate control valve and an inverter
individually. A heat-exchanger is used to control the inlet temperature of the main flow and the branch flow in the mixing tee and it
is set behind the mixing tank. The heat-exchanger has a secondary
cooling system for decreasing the temperature with city water, and
the flow is controlled by the inlet heat-exchanger valve. There is a
cube water jacket around the T-junction, which is made of acrylic
plates the same as the piping material to decrease the effect of
refraction on light through the pipe wall during the visualization
of longitudinal sections.
051103-2 / Vol. 131, MAY 2009
Experimental Conditions and Analysis Methods
The particle image velocimetry PIV system is used to visualize the flow field in the T-junction area with water as the working
fluid. In this system, the velocity vectors are derived from subsections of the target area of the particle-seeded flow by measuring
the movement of particles between two light pulses. The flow is
illuminated in the target area with a light laser sheet. The camera
is able to capture each light pulse in separate image frames. Once
a sequence of two light pulses is taken, the images are divided
into small subsections called interrogation areas. The interrogation
areas from each image, frames 1 and 2, are cross correlated with
each other pixel by pixel. The correlation produces a signal peak,
identifying the common particle displacement. An accurate measure of the displacementand thus also the velocityis achieved
with subpixel interpolation. A velocity vector map over the whole
target area is obtained by repeating the cross-correlation for each
interrogation area over the two image frames captured by the
charge coupled device CCD camera. The correlation of the two
interrogation areas, frames 1 and 2, results in the particle displacement represented by a signal peak in the correlation.
The number of particles in the flow is important for obtaining a
good signal peak in the cross-correlation. As a rule of thumb,
1025 particle images should be seen in each interrogation area.
When the size of the interrogation area, the magnification of the
imaging, and the light-sheet thickness are known, the measurement volume can be defined. Setting up a PIV measurement, the
side length of the interrogation area d and the image magnification
sm are balanced against the size of the flow structures to be resolved. One way of expressing this is to require the velocity gradient to be small within the interrogation area:
sm . Umax Umin t
5%
d
The highest measurable velocity is constrained by particles

traveling further than the size of the interrogation area within the
time, t. The result is a lost correlation between the two image
frames and thus loss of velocity information. As a rule of thumb,
sm U t
25%
d
In this research, two conditions are selected to visualize the flow

field: whole flow field long-shot and close-up flow field with
150 150 mm2 and 60 60 mm2 view sections, respectively.
The resolution of each section is 1018 1008 pixels. The interrogation cell is basically divided into 64 64 pixels at each segment for measuring whole flow field and 32 32 pixels for measuring close-up area with 50% area overlap and 100 s time
intervals. The cross section is visualized into two velocity matrices. Both matrices have 30 29 arrays when 64 64 pixels cross
correlated is used and have 62 59 arrays when 32 32 pixels
cross correlated is used. The laser sheet thickness is 13 mm with
200 mJ energy level based on the three parameters such as visualization area, amount of tracer in the fluid, and distance between
visualization cross section and the cameras lens. Two kinds of
tracer with different diameters are selected, both made from nylon
powder with 1.03 g / cm3 density. The tracer with a diameter of
80 m is used for visualizing the whole flow field and the tracer
with a diameter of 20 m for close-up flow field condition. The
camera starts to shoot at 30 Hz trigger input frequency whose
frame rate is 30 fps, and two images are captured in the rapid
succession by the camera in the triggered double exposure mode.
Each shot captures 99 images continually, and 49 velocity vector
maps are obtained with 0.03 s time separation. Five shots are
taken, and a total of 240 frames in the each condition is used for
evaluating both flow fields. In all experimental conditions, the
measurement is done several times and is calibrated really skillfully. There were not any error vector in the experimental data and
data are evaluated without using any filtering, masking, smoothTransactions of the ASME
Fig. 2 Flow field structure: a longitudinal section, parallel to the jet, b

lateral section, and c longitudinal section, perpendicular to the jet; 1
initial part of the jet, 2 transitional part of the jet, 3 main part of the jet, 4
KelvinHelmholtz instability, 5 wake area, 6 secondary flow twin vortex,
7 high velocity area, 8 main flow turning area, 9 interface, 10 vortices
act as Karman vortex, 11 large wake eddies, 12 moving cross section of
the jet, 13 small eddies, 14 large eddies in the downstream, 15 main
flow blow jet, and 16 small area below jet
ing, and peak locking during the process and analysis.

The average flow field and the intensity of velocity fluctuation
are evaluated by the following equations:
Uave =
1
n
2
r,j
2
+ uz,i
1
sj =
u j,i u j2
n i=1
In this figure the different parts of the flow field are numbered,
and the experiment related to each part is described in the following paragraphs. To have comprehensive evaluation of every part,
the T-junction area is subdivided in the several lateral and longitudinal sections A, B, C1, C2, C3, and C4, as shown in Fig. 3,
and each section is visualized separately with high accuracy. All
these experiments are done under the re-attached jet condition
1/2
2
Umix = U2b + Um
I = sr + sz/2/Umix
Here, ur,i and uz,i represent the instantaneous radial and axial
velocities at the i frame, Uave is the averaged absolute velocity,
and n is the total number of the frames. Ub and Um represent the
branch and main velocities, respectively, and sr and sz are the
standard deviations of the velocity variation in the radial and axial
directions. I represents the intensity of velocity fluctuation.
Results and Discussion
The mixing mechanism of two fluids in the T-junction is researched by visualizing the flow field instantaneously, and the
schematic structure of this mixing is drawn in Fig. 2.
Fig. 3 Positions of lateral and longitudinal sections to visualize the flow field
MAY 2009, Vol. 131 / 051103-3
Fig. 5 Average velocity distributions in different longitudinal

sections
Fig. 4 Flow field in longitudinal section A: a average velocity
distribution and b intensity of velocity fluctuation
with almost same branch and main velocities and different Reynolds numbers Re= 53,000 and 10,000, respectively. The complexity of this condition is the basic purpose for choosing it,
which was studied previously by Hosseini and Yuki 10.
Figure 4 shows the average velocity distribution and the velocity fluctuation in the longitudinal section A with 0.48 m/s main
velocity and 0.47 m/s branch velocity. The different parts of
branch jet are easily recognizable in this figure with almost the
same structure as the one introduced by Abramovich 12. The
first part is the wedge shape region including the potential core
that is called the initial part with the minimum velocity fluctuation
and the highest instability compared with other parts. The second
part of the branch jet is the transitional part that is distributed from
the edge of the initial part to the upright direction along the branch
jet axis. The cross section of this area becomes larger in the downstream, and the jet in this part has a flexible structure against the
main flow. The main part of the jet is the last part and spreads
from the transitional part of the jet to the long downstream. This
part is the most flexible, unstable, and self-developed. There is an
interface that is raised by the steep velocity gradient between the
main flow and the jet in the initial and transient parts. The nature
of the interface is caused by the viscous shearing force of two
fluids with different velocities, and its thickness depends on the
boundary condition in each part. The wake area above the jet has
the most steep velocity gradient and has strong temperature fluctuation effects in the case of nonisothermal mixing fluids, and if
the interface touches the main pipe wall frequently then it can
cause the high-cycle thermal fatigue. The effect of temperature
fluctuation on the wall is studied previously by several research
groups such as Hosseini and Yuki 10, Yuki et al. 11, Metzner
and Wilke 8, and Chapuliot and Gourdin 9.
Figure 5 presents the average velocity distribution in four longitudinal sections C1, C2, C3, and C4. The main flow is distributed almost symmetrically in all sections, and the cross section of
the branch jet and the wake area above it become more visible by
getting closer radially to the branch pipe nozzle from section C4
to section C1. The intensity of velocity fluctuation is calculated in
the same longitudinal sections, as drawn in Fig. 6. The fluctuation
is minor in the region with highest velocity because of instabilitys
051103-4 / Vol. 131, MAY 2009
law and it is large in the region around the jet along with the edge
of the wake area because of the variation in steep velocity gradient.
Figure 7 shows the velocity distribution and the velocity fluctuation of secondary flow in the lateral section B. The symmetrical
twin vortexindicated by the average velocity vectors in Fig.
7a and basically created by the 90 deg bend in the upstream
has strong effects on the flow field in the mixing area. Three main
high velocity regions are observed as well as a high velocity fluctuation region between twin vortices in the lateral section.
The mixing progress will be monitored better by analyzing every frame of the time series data repeatedly. Figure 8 describes
four continual velocity vector frames with 0.03 s time separation
Fig. 6 Intensity of velocity fluctuation in different longitudinal

sections
Fig. 7 Lateral section B: a average velocity vectors, b mean

velocity distribution, and c intensity of velocity fluctuation
Fig. 8 Velocity vectors of time series and average flow field

above the jet in C2 section t = 0.03 s, four continual frames,
average of 240 frames, Ub = 0.47 m / s, and Um = 0.485 m / s
Fig. 9 Five continual frames of time series data with close-up visualization
Ub = 0.55 m / s and Um = 0.73 m / s
MAY 2009, Vol. 131 / 051103-5
the main flow and the branch jet, and in the case of nonisothermal
mixing condition can cause high-cycle thermal fatigue in the same
area and sometimes inflict considerable damage on the piping
structures. The second region is distributed between the branch
nozzle and the re-attached area in the edge of the wake area, and
the Karman vortices are the main source of this fluctuation. The
frequency of velocity fluctuation in the second region is lower
compared to the first one as well as the velocity gradient. The
third region is the re-attached region with lowest frequency of the
velocity fluctuation. This fluctuation mainly is generated by the
lift and drag force between the main flow and the jet body. Due to
the fact that the jet body is flexible against the main flow, the high
pressure gradient around the jet changes the cross section of the
jet in the transitional part from the circular shape to the oval shape
periodically. Meanwhile the pressure gradient decreases based on
the oval shape and then automatically the cross section of the jet
changes back to the circular shape again, which means that the lift
and drag forces fluctuate side by side. So the jet body starts to
move up and down as well as right and left frequently. This phenomenon causes the low frequency fluctuation of the velocity in
the re-attached area. Because of an efficient thermal transfer to the
structure, this low frequency fluctuation inflicts the most effective
thermal fatigue damage compared with the other two areas and
can be improved simply just by changing the momentum ratio
between the main and the branch flows or the piping geometry,
which is described previously by Hosseini and Yuki 10.
Fig. 10 Schematic of high velocity fluctuation areas near the

main pipe wall in the T-junction area
together with the average velocity vectors of 240 frames in the

longitudinal section C2. Some vortices are created in the wake
area above the branch jet and they look like Karman vortices. The
different pressures between the wake area and the main flow below the jet create drag force on the jet body and make the jet
move up and down with a low frequency. On the other hand, the
formation of Karman vortex causes lift force and moves the jet
body to the left and right as well.
Figure 9 shows different time series data in the longitudinal
section A with four continual frames. The existence of the wake
area above the jet is clear in this section. The gradient velocity
between the wake area and the jet causes KelvinHelmholtz instability that stretches the flow and produces some anticlockwise
vortices in the wake area.
A part of the jet flows over the edge of the wake area and
touches the main pipe wall in the region, which is called the
re-attached region. By considering the formation mechanism of
vortices in the wake area three-dimensionally, the combination of
the KelvinHelmholtz instability vortex and the Karman vortex
creates a type of arch vortex and develops it eventually in the
downstream. There are also many other large eddies in the downstream of T-junction with different sources such as the interaction
of main flow with the body of the jet, some hidden eddies in the
self-developing part of the jet, the fluctuation of the jet structure
by the drag and lift forces, the 90 deg bend effects, etc.
Conclusion
The fluid mixing mechanism is provided experimentally by particle image velocimetry method to investigate the structure of the
turbulent flow field in a T-junction area with a 90 deg bend upstream. For this purpose, several longitudinal and lateral sections
in the mixing area are visualized by the high frame rate of velocity
distribution, and the statistical analyses of time series data give us
the general view of mixing phenomena.
Figure 10 summarizes three regions with the highest velocity
fluctuation in the T-junction. The first one surrounds the branch
nozzle and is created by the KelvinHelmholtz instability between
051103-6 / Vol. 131, MAY 2009
Nomenclature
t
d
Db
Dm
I
Imax
sm
sj
sr
sz
t
U
Uave
Ub
Um
Umax
Umin
ui,j
uj
ur,i
uz,i
z
time separation
side length of the interrogation area
branch pipe diameter
main pipe diameter
intensity of the velocity fluctuation
maximum velocity fluctuation
image magnification
standard deviation in the j direction
radial velocity standard deviation
axial velocity standard deviation
time separation
mean velocity
average of the absolute velocity
branch velocity
main velocity
maximum velocity
minimum velocity
velocity at the i frame in the j direction
average velocity in the j direction
radial velocity at the i frame
axial velocity at the i frame
longitudinal distance along the axial direction
References
1 Faidy, C., 2004, Thermal Fatigue in Mixing Area, Third International Conference on Fatigue of Reactor Components, EPRI-US NRC-OECD NEA,
Seville, Spain.
2 Tanaka, M., 2004, Sixth International Conference on Nuclear Thermal Hydraulic, Operation and Safety, NUTHOS-6, Nara, Japan, ID No. N6P334.
3 Igarashi, M., 2003, Study on Fluid Mixing Phenomena for Evaluation of
Thermal Striping in a Mixing Tee, Tenth International Topical Meeting on
Nuclear Reactor Thermal Hydraulic, NURETH-10, Seoul, Korea.
4 Hu, L. W., and Kazimi, M. S., 2006, LES Benchmark Study of High Cycle
Temperature Fluctuation Caused by Thermal Striping in a Mixing Tee, Int. J.
Heat Fluid Flow, 271, pp. 5464.
5 Hosseini, S. M., and Yuki, H., 2006, Experimental Investigation of Thermal
Hydraulic Characteristics at a Mixing Tee, International Heat Transfer Conference, Sydney, Australia, FCV-17.
6 Hosseini, S. M., and Yuki, H., 2005, Three-Dimensional Study of Flow Mixing Phenomenon, International Conference Nuclear Energy for New Europe,
Bled, Slovenia, September ID: 037.
7 Yuki, K., and Hosseini, S. M., 2004, 15th International Conference on Nuclear
Thermal Hydraulic, Operation and Safety NUTHOS-6, Nara, Japan, ID No.
N6P082.
8 Metzner, K. J., and Wilke, U., 2005, European THERFAT Project-Thermal
Fatigue Evaluation of Piping System Tee Connections, Nucl. Eng. Des.,
2352-3, pp. 473484.
9 Chapuliot, S., and Gourdin, C., 2005, Hydro-Thermal-Mechanical Analysis of
Thermal Fatigue in a Mixing Tee, Nucl. Eng. Des., 2355, pp. 5755906.
10 Hosseini, S. M., Yuki, H., and Hashizume, H., 2008, Classification of Turbulent Jets in a T-Junction Area With a 90-deg Bend Upstream, Int. J. Heat Mass
Transfer, 519-10, pp. 24442454.
11 Yuki, K., Sugawara, Y., and Hosseini, S. M., 2008, Influence of Secondary
Flow Generated in a 90-Degree Bend on the Thermal-Hydraulic Characteristics in a Mixing Tee, Nucl. Sci. Eng., 158, pp. 194202.
12 Abramovich, G. N., 1963, The Theory of Turbulent Jets, MIT Press, Cambridge, MA, Chap. 1, p. 3.
MAY 2009, Vol. 131 / 051103-7
A. Ahmadi
Department of Mechanical Engineering,
Iran University of Science and Technology-Arak
Branch,
Arak,
Markazi 38181-41167, Iran
e-mail: a_ahmadi@iust.ac.ir
Experimental Study of a New

Flow Conditioner on Disturbed
Flow in Orifice Plate Metering
The sensitivity to poor conditioned and swirling flow of flow measurements using an
orifice plate are subjects of concern to flowmeter users and manufacturers. Measurements
of mass flow rate under different conditions and different Reynolds numbers were used to
establish a change in discharge coefficient relative to the standard one. The experimental
results show that an optimally shaped flow conditioner could attenuate the effects of both
swirl and asymmetrical flows. The optimization of the swirler flow conditioner is a main
outcome of this work. So far the experimental results show that the cone swirler flow
conditioner is the best one for swirling flow. DOI: 10.1115/1.3114677
Keywords: orifice plate, asymmetric, swirling flow, flow conditioner, swirler, flow
measurement
Introduction
The differential pressure flowmeter is one of the most common

forms of flowmeter used in industry. It is mainly because the
orifice plate is simple to construct, has a low maintenance cost,
and wide applicability to different fluids including both liquids
and gases. In addition, there is a great weight of experience to
confirm its operation and installation, which is documented in BS
EN ISO 5167 1.
The International Standards state that the uncertainty of metering is typically 0.51% when supplied with fluid with a wellconditioned velocity profile 2. The most important assumption
of well-conditioned flow is that the flow approaching the orifice
plate must be fully developed and free of any asymmetry or swirl.
In practical applications, however, pipe fittings and pipe installations such as valves, bends, heat exchangers, compressors, and
other devices can generate a distorted and swirling flow profile.
These distortions can alter the accuracy of an orifice plate up to
4% for an asymmetric velocity profile 3 and from 3% up to 7%
error for swirling flow generated by twin 90 deg elbows 4,5. In
order to produce a fully developed flow, free from all disturbances, a long straight pipe must be installed upstream from the
orifice plate. British Standard or ISO recommends a minimum
straight length upstream of the meter that depends on the Reynolds number, pipe diameter, orifice diameter, the ratio of pipe
diameter to hole diameter, and the pipe fitting 6. In general, this
requirement means that at least 10 pipe diameters of smooth
straight pipe is required for orifice plates with small holes increasing to 36 pipe diameters for plates with large ones 6. In many
installations it is impossible to provide a sufficiently long straight
pipe upstream of the meter to remove flow disturbances. In these
circumstances, flow conditioners and flow straighteners can be
placed upstream to remove disturbances in the flow, which reduce
the number of straight upstream pipe lengths required for accurate
flow measurement.
In general, a flow straightener is a device that removes a swirl
from the flow but has little effect on the asymmetrical velocity
profile. On the other hand, a flow conditioner is a device that not
only removes swirl but also produces a repeatable downstream
JOURNAL OF FLUIDS ENGINEERING. Manuscript received May 7, 2007; final manuscript
received March 13, 2009; published online April 13, 2009. Assoc. Editor: Juergen
Kompenhans. Paper presented at the 2005 ASME Fluids Engineering Division Summer Meeting and Exhibition FEDSM2005, Houston, TX, June 1923, 2005.
velocity irrespective of the upstream flow disturbances. Therefore,

most flow conditioners are used to avoid unwanted shifts in the
meters calibration by reducing the effect of upstream disturbances on the meter to an insignificant level. It is desirable for a
good flow conditioner to fulfill its duty within the following requirements:
low-pressure loss across the device

short downstream length from the source of disturbances
the combination of device and conditioner to be as compact
as possible
easy installation
However, even the best performing flow conditioners need to be

installed at least 2 pipe diameters from the disturbances and also 4
or 5 pipe diameters from the orifice plate 7. So, great efforts
have been expended to achieve a flow conditioner that will perform with a lower number of pipe lengths and minimum pressure
drop across it 810.
There are a large number of flow conditioners used in flow
measurement, and some are included in the British and ISO Standards and others can be found in technical reports 1,2. For instance, a flow conditioner introduced by Canada Pipeline CPA
50E 11 is a perforated plate, which can provide a repeatable,
swirl free, fully developed velocity profile for high-pressure natural gas applications 11. A new device, a vane perforated plate
described as meeting the velocity profile requirement of ISO
5167, is positioned 6 pipe diameters upstream of the orifice plate
and produces a pressure loss of about 1.4 dynamic heads 12.
Two kinds of new flow conditioner called the in-line flow conditioning plate 13 and the AS-FC flow conditioner 14 claim to
produce a fully developed velocity profile with a low-pressure
loss. The efficiency of standard flow conditioner in reducing the
effect of disturbed flow consisting of asymmetric velocity profile
and swirling flow has been described in other references 15.
Generally, all flow conditioners are aiming either to supply a
flow in a settled fully developed state through turbulent mixing
devices or to achieve a repeatable constant velocity profile independent of source of disturbances as in vortex action devices.
1.1 Recent Experimental Work. An approach to the flow
conditioner, based on vortex action and developed at Sheffield
University, is to use a device at the short distance upstream of the
orifice plate, which makes a strong swirl or disturbance that will
absorb all the other disturbances in a predetermined manner. This
device creates a predetermined flow condition, independent of the
MAY 2009, Vol. 131 / 051104-1
upstream conditions. Therefore, the strong disturbed flow that is

made by the swirler device can absorb other unknown and unwanted disturbances and achieve a repeatable velocity profile independent of disturbances. An elementary study using a twisted
piece of plastic was positioned 1.5 pipe diameters upstream as a
flow conditioner was carried out in an experimental rig using air
as the working fluid. This experimental work was carried out for a
76 mm pipe with = 0.5 and a variety of Reynolds numbers from
14,000 to 70,000 for both standard and nonstandard velocity profiles. The flow was disturbed upstream of the orifice plate by
fixing different shaped blocks within 3.54.5 pipe diameters of the
orifice plate 16.
The results showed that block disturbances produced up to 4%
shift in measuring flow based on the equation in the ISO Standards 1. The application of using a swirler flow conditioner for
higher Reynolds numbers and different orifice diameters has been
conducted in an experimental water rig with of 0.40.7. The
effect of these disturbances on mass flow showed that the effect of
asymmetric velocity profile is to produce up to 3% shift in discharge coefficient for low Reynolds number and about 1.5% for
high Reynolds number for = 0.4 and = 0.7, respectively 17.
Finally, the effect of an optimum design for the swirler flow conditioner on disturbed flow shows that this flow conditioner has a
positive effect on swirling flow disturbances 18.
Thus the focus of this study is to examine the effect of different
shapes of swirler flow conditioner in limiting disturbances due to
different kinds of disturbances. These experiments were carried
out with water and air flow in two rigs with both high and low
Reynolds numbers.
1.2 Theory. The pressure drop P across the orifice plate
and the mass flow rate qm are linked by an equation from ISO
1.
qm =
1 4 4 d
2p
where C is the discharge coefficient that depends on the exact type

of differential flowmeter being used. For instance, for the D and
D / 2 orifice plate flowmeter, which has been used in this study,
this coefficient is
C = 0.5961 + 0.02612 0.2168 + 0.000521

19,000
Re
0.8
+ 0.0188 + 0.0063
19,000
Re
0.8
+ 0.04289 1 0.11
+ 0.031
0.94
0.94
0.8
1
1
+ 0.0110.75 2.8
3.5
106
Re
106
Re
0.7
0.3
4
1 4
1.1
D
25.4
1.3
where is the ratio of the orifice diameter to the pipe diameter

= d / D.
It can be seen that the flow rate is calculated from the pressure
drop, geometry, and the physical parameters of the fluid and also
the Reynolds number of the flow, which is itself a function of the
flow rate. So, in order to calculate a mass flow rate from differential pressure, an iterative method must be used. An important
point regarding the above correlations is that the profile of the
flow approaching the orifice plate should be fully developed. This
allows the uncertainty of the measured flow rate to be less than
1%.
1.3 Variation in the Discharge Coefficient. In evaluating the
effects of disturbances on mass flow rates, various disturbances
051104-2 / Vol. 131, MAY 2009
and hence variations in the discharge coefficient have been analyzed. According to the ISO Standards the flow rate is determined
from the differential pressure measured through the orifice plate
1. For a fully developed velocity profile, for a known discharge
coefficient, the mass flow rate can be calculated or for a known
mass flow rate the standard discharge coefficient can be calculated. Nevertheless, the mass flow rate for disturbed flow can be
measured by experiment and then new discharge coefficient Cd
can be calculated from Eq. 2. Thus the percentage shift of the
standard discharge coefficient due to disturbances Cd % can be
determined as
Cd =
Cd Cd0
100
Cd0
This can be used to express the effect of disturbances on metering.

1.4 Pressure Loss Coefficient. In order to compare pressure
losses across different flow conditioners, it is customary for this
quantity to be expressed according to a specific term called the
pressure loss coefficient. This coefficient is defined as
PL =
P
1 2
u
2
where P is a pressure drop caused by the flow conditioner and u

is an average velocity in the pipe line or the bulk velocity in the
pipe volume flow divided by the cross-sectional area.
Experimental Facility
In order to assess the effect of disturbed flow on the orifice

plate, two experimental facilities have been built: one for air and
one for water. In these facilities, the mass flow rate of the orifice
plate with both standard and nonstandard velocity profiles has
been measured for different Reynolds numbers and ratios.
Throughout the experimental work the accuracy of the standard
orifice plate without disturbances was also compared with the ISO
Standards 1 using Eq. 3.
2.1 Water Rig. For the water rig, a smooth 32 mm internal
diameter pipe with 20 and 10 pipe diameters upstream and downstream, respectively, was used Fig. 1. For measuring the pressure
difference, two pressure tappings were used: D upstream and D / 2
downstream of the plate. The pressure drop across the orifice plate
was measured by using both a U-tube water manometer and a
pressure transducer.
To measure the mass flow rate, the dynamic weighing scale was
used to measure the mass flow rate. This method requires the mass
flow rate to be under steady state flow rate conditions. As the mass
flow rate was not initially determined, so the pressure drop was
fixed for a particular value. This task was achieved by changing a
valve, which was installed at the end of a pipeline. After adjusting
the pressure drop at a constant value and obtaining steady state
mass flow rate, the weighing tank drain valve was closed and
mass began to accumulate until the counterbalance weight was
lifted. When the balance arm started to move, a timer was triggered to run and the test run began. After a while, the flow rate
was accumulated in the weighting tank for a specific interval time.
At the end of the interval time the time was stopped manually and
the mass flow was measured. Hence, by measuring the discharged
water in the weighting tank and its discharging time, experimental
mass flow rate was calculated. In the next step, the velocity of the
flow was calculated; consequently, the Reynolds number basis of
the pipe diameter was calculated. It followed that with having the
pressure drop, which was fixed during the experiment, the discharge coefficient from Eq. 2 was calculated. On the other hand
by having experimental mass flow rate, the discharge coefficient
was calculated using Eq. 1. Finally, these two discharge coefficients calculated from mass flow rate and from specific correlation stated in the Standards were compared to investigate a variaTransactions of the ASME
Fig. 1 Sketch of the experimental rig for water
tion in discharge coefficient for different experimental conditions.

Orifice plates with values of varying from 0.4 to 0.7 were
employed in this rig but only the results of 0.5 will be reported
here. The experimental mass flow rate for each case was measured
by the dynamic weighting method 6. This mass flow rate was
compared with the mass flow rate calculated according to the ISO
Standards. The difference between these two flow rates has been
investigated in order to estimate likely errors.
2.2 Air Rig. For the air rig, two orifice plates were positioned
in a series of smooth and circular pipes of 76.2 mm diameter. The
first of these was used as a reference meter and the second was the
one where flow approaching the orifice plate was altered. These
two pipes were joined together with a large air box to reverse the
flow and also to allow it to settle. A sketch of this experimental
arrangement is shown in Fig. 2. In the air rig a diameter ratio of
= 0.5 was selected for both orifice meters. The pipes were 3 m
40 diameters long on either side of each orifice plate. This exceeded the ISO Standards requirements of 20 pipe diameters.
However, it should be noted here that the requirement of 20 pipe
diameters is still in doubt by several authors and some of them
mention the need for a 100 pipe diameters requirement to become
fully developed 12.
Liquid filled differential manometers were fitted to either side
of the orifice plates again using ISO Standards D and D / 2 tappings. As the flow rates were calculated using the density of the
air, the atmospheric pressure and temperature were measured for
each set of experiments. The air was sucked through the system
using a fan controlled by a ball valve. All of the measurements

were carried out simultaneously on both meters using Reynolds
numbers of up to 40,000.
2.3 Swirler Flow Conditioners. The initial shape for the
swirler was a piece of plastic with a 90 deg twist in it. The length
of the swirler was about 1.5 times the pipe diameter, which was
similar to the previous experiments 16. In the new series of
experiments, the development of the suitable swirler conditioner
on the disturbed flow was the main goal. The reason for this was
that the previous results showed that the original, one piece,
swirler conditioner could attenuate the effect of asymmetric flow
very well but for swirling flow, it performed less well. So, four
different swirler shapes were designed and also tested. The length
of these was kept the same, as the original 1.5 pipe diameter and
they were positioned in the same position upstream of the orifice
plate. These swirler conditioners can be described as follows in
Fig. 3:
a
b
c
d
cone swirler
joint two-piece of the swirler made of two pieces joined
together
two-piece of the swirler with a gap between these two
parts
four-piece of the swirler with a gap between each part
2.4 Disturbances. Velocity profiles different from those

formed in fully developed flow can be produced using distur-
Fig. 2 Sketch of air experimental rig
MAY 2009, Vol. 131 / 051104-3
distribution. The second one, referred to as the 1/8 disturbance,

had a cross section of 1/8 of the area of the pipe and caused a
weaker disturbance. These block disturbances are shown in Fig. 4.
A swirling flow was produced in the pipe using similar devices
to the one-piece swirler. The length of each was 1D and they had
angles of 90 deg and 180 deg. Also, in the water rig two out of
plane elbows were employed. The two-elbow disturbance was two
successive bends connected together by a flange without a straight
length to set the configuration in and out of plane. The ratio of the
radius of the curve to the pipe diameter was 1.5. Consequently, the
following disturbances were used in experimental rigs:
Fig. 3 Swirler flow conditioner shapes; left to right: a fourpiece swirler, b cone swirler, c two-piece swirler, and d
joint swirler
bances upstream of the orifice plate. These disturbances can provide a combination of an asymmetric velocity profile and a swirling flow. In order to assess the effect of the swirler flow
conditioner on the disturbed flow, both types of disturbances were
used in the experimental facilities. To achieve an asymmetric velocity profile, block disturbances were used. These blocks were
cut from a circular piece of metal and were placed on the bottom
of the pipe. One block, referred to as the 1/4 disturbance, had a
cross section of 1/4 of the area of the pipe and caused a significant
All of the disturbances were 1.5D in length and were positioned

about 3D from the orifice plate. The configuration of the swirler,
disturbances, orifice plate, and their position from the upstream
face of the orifice plate is shown in Fig. 5.
Four different sets of experiments for measuring mass flow rate
were conducted. The first was done for a standard velocity profile
to estimate the accuracy of rig and to compare it to the ISO Standards. Then the effect of disturbances was examined. The effect of
each swirler flow conditioner on the mass flow rate was then
investigated with disturbances. Then, these mass flow rates were
used to calculate new discharge coefficients, and the variation in
C Eq. 3 was calculated to allow a comparison to take place. All
of these results were obtained with a of 0.5.
Fig. 4 Block disturbances left to right: 1/4 and 1/8

disturbances
two elbows both in and out of plane

swirler disturbance with different twist angles
block disturbance
3.1 Performance of the Different Swirler Flow Conditioner on Disturbed Flow. The results for the two-piece and joint
swirler conditioners are shown in Figs. 6 and 7, respectively. By
comparing the two graphs it can be seen that the change in discharge coefficient caused by each swirler conditioner on its own is
as much as when using a block disturbance. This means that the
swirler flow conditioner can attenuate the effect of an asymmetric
velocity profile. Nevertheless, the effect of these two swirler conditioners on the swirling flow caused by the swirler disturbance is
totally different. As the same time employment of swirler disturbance causes about 3% change in standard C, using joint and
two-piece swirler conditioner for this condition causes about 7%
and more than 10% this line has not been shown in the graph
Fig. 5 The positions of disturbance, swirler, and orifice plate
Fig. 6 Change in discharge coefficient with two-piece swirler
051104-4 / Vol. 131, MAY 2009
Fig. 7 Change in discharge coefficient with joint swirler
because it is out of scale. Thus, the swirler flow conditioner generally can dampen the effect of block disturbance; however, the
effective swirler flow conditioner on the swirling flow is a main
task. Therefore, other shapes of the swirler have been designed
and their effects on the swirling flow have been tested.
The effect of the four-piece and cone swirler flow conditioner
on the disturbed flow was examined, and the results on the discharge coefficient are presented in Figs. 8 and 9. It can be seen
from Fig. 8 that the four-piece flow conditioner does not have an
attenuating effect on the asymmetry flow caused by the block
disturbance and neither has any better effect on the swirling flow
relative to previous ones. On the other hand, the result of the cone
swirler shown in Fig. 9 illustrates perhaps the best result for
implementation of the swirler flow conditioner on the disturbed
flow. It can be seen that using the cone swirler by its own changes
the discharge coefficient from 4% for low Reynolds numbers to
about 6% for high Reynolds numbers. The change in discharge
coefficient for both block and swirler disturbance with cone

swirler is similar to the change in C due to cone swirler by its own
with about 1% tolerance. So, it shows that the cone swirler can not
only attenuate the effect of asymmetric velocity profile but can
also dampen the effect of swirling flow. Thus, it could be concluded that the cone swirler is an optimum shape for swirler flow
conditioner, and this device will achieve about a 1% variation in
metering when faced with a disturbed flow. This compares well
with the standard orifice plate without any disturbances, which
can also obtain 1% accuracy 6.
3.2 Implementation of Cone Swirler With Two-Elbow
Disturbance. In order to justify that the cone swirler flow conditioner has a positive effect on dampening swirling flow, the performance of this device was tested on other sources of swirling
flow. According to the majority of references, two elbows in and
out of plane are the main sources of swirling flow 6,1. So, these
Fig. 8 Change in discharge coefficient with four-piece swirler
MAY 2009, Vol. 131 / 051104-5
Fig. 9 Change in discharge coefficient with cone swirler
double elbow disturbances were set in the water rig, and the effect
of these configurations on the standard orifice plate alongside the
cone swirler conditioner was investigated. This result of change in
discharge coefficient is given in Fig. 10.
It can be seen from Fig. 10 that the two-elbow disturbances can
make a 23% shift in the standard discharge coefficient 4. However, the combination of a cone swirler with a two-elbow disturbance causes a change in discharge coefficient up to 46%. It
means that the cone swirler conditioner on its own can cause up to
a 6% change in standard discharge coefficient, and the combination of this device with other disturbances such as block disturbance, swirler disturbance, and two elbow again causes the
change in discharge coefficient within the 1% error bars. In other
words, the cone swirler conditioner can produce a repeatable shift
in discharge coefficient independent of sources of disturbance. On
the other hand, this graph shows that the result of change in C for
using the standard orifice plate falls within the 1% error bars too.
The latter conclusion confirms the accuracy of the standard orifice
metering 1.
3.3 Performance of the Cone Swirler at Low Reynolds
Numbers (Air Rig). After the measurement of the cone swirler on
a disturbed flow at relatively high Reynolds number, which was
achievable for water rigs, this procedure was examined on the air
rig with low Reynolds numbers. In the air rig, low Reynolds numbers up to 20,000 were obtained. The results for the air rig with
different combinations of disturbances and the cone swirler are
shown in Fig. 11.
The graph shows that using a cone swirler conditioner in the
upstream changes the standard discharge coefficient by up to
1.5%. This trend is repeated when a combination of block disturbance and cone swirler is placed upstream. This means that the
Fig. 10 Change in discharge coefficient for two-elbow disturbance with cone swirler
051104-6 / Vol. 131, MAY 2009
Fig. 11 Change in discharge coefficient for different disturbances and cone swirler air
rig
cone swirler can attenuate the effect of asymmetric flow with low
Reynolds number. On the other hand, employment of swirler disturbance makes up to 1.5% error on metering, and this change
when using a cone swirler conditioner can be affected by 0.5%
error. This means that the cone swirler cannot compensate the
effect of swirling flow as much as for asymmetric flow.
3.4 Calibration of the Cone Swirler. While the swirler conditioner is used in front of an orifice plate, the standard discharge
coefficient equation 2 cannot be used to calculate a mass flow
rate. So the appropriate discharge coefficient for the combination
of the swirler and an orifice plate was determined by curve fitting.
Figure 12 shows the calibration curve for the combination of
cone swirler and orifice plate. The experimental mass flow rate in
the presence of the cone swirler is plotted versus the calculated

mass flow rate of the standard discharge coefficient. It can be seen
that the orifice plate and cone swirler combination appears to produce about 7% more flow rate than the orifice plate on its own. To
obtain a new discharge coefficient appropriate to the combination
of cone swirler and orifice plate, a straight, trend line has been
fitted to this curve. This line represents the new discharge coefficient C for combination of cone swirler and orifice plate flowmeter is 1.079. It means that the new discharge coefficient is
1.079 times the standard orifice plate on its own. So, it can be
concluded that the correction factor K regarding the employment
of cone swirler upstream of a standard orifice plate is 1.0799 for a
of 0.5.
3.5 Pressure Loss Coefficient for Cone Swirler. As already
mentioned, a good flow conditioner should have not only a good
performance in dampening the disturbances but also a minimum
pressure loss as well. By measuring a pressure difference across
3D in upstream and downstream of the orifice plate with and
without flow conditioner, this coefficient can be calculated for
cone swirler according to Eq. 4. The reported results for pressure
loss coefficient for vortex action devices can be varied from as
high as 8 for the AMCA flow conditioner to about 1 for Swirlvan-tab, and this coefficient for Etoile flow conditioner reported a factor of 2 7. The pressure loss calculation shows this
coefficient for cone swirler to be about 2.3.
Fig. 12 Calibration of the cone swirler
3.6 Source of Errors. A brief review of error sources associated with measurement of the various parameters is considered in
this section. The accuracy of the weighing scale was 0.1% and the
stop watch 0.2%. Also, as already mentioned, the pressure drop
was measured by two kinds of pressure sensors: a pressure transducer and U-tube manometer. The pressure difference between
these two meters was compared, and the difference was about 1.8
cm H2O or 176 Pa in 3 m, which is negligible in metering. On the
other hand, the same procedure was used for measuring the mass
flow rate, and so the inaccuracy was the same for the mass flow
rate. Thus, it can be concluded that the errors in metering are
within those predicted by the Standards.
MAY 2009, Vol. 131 / 051104-7
Conclusions
The present study has shown that the novel idea of using a
swirler flow conditioner for disturbed flow can reduce the error of
metering caused by disturbances to an acceptable standard level.
To discover the best design of the swirler flow conditioner, a
variety of different shapes was considered and tested in experimental rigs. So far the results show that the cone swirler flow
conditioner can reduce the distortion due to asymmetric velocity
profile on metering for high and low Reynolds numbers. It is
concluded that this kind of swirler conditioner also has a positive
effect on swirling flow for high Reynolds numbers but has less
effect for low Reynolds numbers. The new discharge coefficient
appropriate for the cone swirler flow conditioner can vary from
4% for low Reynolds numbers up to 67% of standard discharge
coefficient for high Reynolds number.
Regarding other factors for desired flow conditioner, the results
show the cone swirler to be positioned about 1.5 pipe diameters
upstream of the orifice plate and its pressure loss to be 2.3 of
dynamic head. These values clearly describe the advantages of the
cone swirler flow conditioner in comparison with other vortex
action flow conditioners.
Also, an important feature is that the standard orifice plate is
vulnerable for metering of dirty flow because of accumulation of
dirt in front of the orifice plate. So the accumulated particles can
alter the accuracy of metering dramatically. Thus using the cone
swirler conditioner not only gives a mass flow metering independent of upstream disturbance but also keeps particles in suspension and prevents the accumulation of particles in front of the
orifice plate.
Nomenclature
d
D
C
p
Re
P
u
C
Cd0
Cd
diameter of the orifice

diameter of the pipe
discharge coefficient
differential pressure across the orifice
Reynolds number of the flow related to the
pipe diameter
pressure static
velocity
new discharge coefficient with the swirler
standard discharge coefficient for fully developed flow
discharge coefficient for nonfully developed
flow
051104-8 / Vol. 131, MAY 2009
correction factor for different values of and

for an orifice plate with the swirler
dynamic viscosity
kinematic viscosity
ratio of the orifice diameter on the pipe diameter = d / D
expansion factor
density of the fluid
References
1 BS EN ISO 5167, 2003, Measurement of Fluid Flow by Means of Pressure
Differential Devices Inserted in Circular Cross Section Conduits Running
Full, British Standards Publications.
2 Rogers,
S.,
2003,
Differential
Pressure,
available
on
www.flowcontrolnetwork.com/PastIssues/julaug1999/1.asp.
3 Ouazzane, A. K., and Benhadj, R., 2002, Flow Conditioners Design and Their
Effects in Reducing Flow Metering Errors, Sens. Rev., 223, pp. 223231.
4 Laribi, B., Wauters, P., and Aichouni, M., 2002, Experimental Study of Aerodynamics Behaviour Downstream of the Three Flow Conditioner, ASME
Fluids Engineering Division Summer Meeting, Canada.
5 Reader-Harris, M. J., Hutton, S. P., and Laws, E. M., 1989, Flow Straighteners and Flow Conditioning Devices, Flow Measurement and Instrumentation
Consortium Report No. 7.
6 Miller, R. W., 1996, Flow Measurement Engineering Handbook, 3rd ed.,
McGraw-Hill, New York.
7 National Engineering Laboratory NEL, 1998, Flow Conditioners Performance Review, Guidance Note No. 11.
8 National Engineering Laboratory NEL, 2002, Assessment of the Performance of Flow Conditioner at Elevated Reynolds Number, Guidance Note
No. 29.
9 National Engineering Laboratory NEL, 1999, CFD Techniques Applied to
Differential Pressure Flowmeter Performance, Guidance Note No 20.
10 Frattolillo, A., and Massarotti, N., 2002, Flow Conditioners Efficiency a
Comparison Based on Numerical Approach, Flow Meas. Instrum., 13, pp.
111.
11 Canada Pipeline Accessories, 2004, CPA 50E Flow Conditioner, available
on www.flowmeterdirectory.com/flowmeter_flow_conditioners.html.
12 Laws, E. M., and Ouazzane, A. K., 1995, A Further Investigation Into Flow
Conditioner Design Yielding Compact Installations for Orifice Plate Flow Metering, Flow Meas. Instrum., 63, pp. 187199.
13 2004,
In-Line
Flow
Conditioning
Plate,
available
on
www.sierrainstruments.com/products/pdf/flow-track.pdf.
14 2004, AS-FC Flow Conditioner, available on www.fluidcomponents.com/
Aerospace/A_ProdflowCond_ASFC.htlm.
15 Ouazzane, A. K., and Barigou, M., 1999, A Comparative Study of Two Flow
Conditioners and Their Efficiency to Reduce Asymmetric Swirling Flow Effects on Orifice Meter Performance, Trans. Inst. Chem. Eng., Part A, 77, pp.
747753.
16 Beck, S. B. M., and Mazille, J., 2002, A Study of Pressure Differential Flow
Meter That Is Insensitive to Inlet Conditions, Flow Meas. Instrum., 125-6,
pp. 379384.
17 Ahmadi, A., Beck, S., and Stanway, R., 2003, Effect of Diameter Ratio on the
Swirling Orifice Plate, Seventh International Symposium on Fluid Control,
Flow Measurement and Flow Visualization, Italy.
18 Ahmadi, A., and Beck, S., 2005, Development of the Orifice Plate With a
Cone Swirler Conditioner, Sens. Rev., 251, pp. 6368.
2D NavierStokes Simulations of
Microscale Viscous Pump With
Slip Flow
Khaled M. Bataineh1
e-mail: k.bataineh@just.edu.jo
Mohd A. Al-Nimr
Jordan University of Science and Technology,
Irbid 22110, Jordan
In this paper we provide numerical solution of the NavierStokes equations coupled with
energy equation for gaseous slip flow in two-dimensional microscale viscous pumps. A
first-order slip boundary condition was applied to all internal solid walls. The objectives
are to study the performance of the pumps and to study the effect of velocity slip on its
performance. Mass flow rate and pump efficiency were calculated for various pump
operation conditions when an external pressure load is applied at the pump exit plane.
Geometric parameters were held fixed in this work. Microviscous pump performance was
studied in detail for several values of the Reynolds number, pressure load, eccentricity,
and slip factors. Our numerical results for no-slip were compared with previously published experimental and numerical data and were found to be in very good agreement.
Slip values and eccentricity were found to be major parameters that affect the performance of pump. Pump head decreases with increasing slip factors. Maximum pump
efficiency increases with increasing slip factor up to Kn approaching 0.1. However, the
maximum value of pump efficiency is found to experience a steep degradation for Kn
approaching 0.1. The values of moment coefficient always decrease as both slip factor
and distance of the rotor from the lower wall increase. Also, as slip factors and distance
of the rotor from the lower wall increase, less net flow rate is predicted. For a given fixed
driving force at the rotor surface, there is an optimum value for the behavior of pump
efficiency with distance of the rotor from the lower wall. Future research should be
conducted to modify the current design to make this concept work for higher Knudsen
numbers. DOI: 10.1115/1.3112390
Keywords: microfluidics, viscous micropump, slip flow, low Reynolds number, CFD
Introduction
Micropumps are important devices in various fields such as

medical science, biotechnology, microelectronic cooling, and mechanical engineering, for example, controlled drug delivery systems in medical science, the pressure supplies for micromachines
in mechanical engineering, ink jet printing, environmental testing,
and so on. Advances in the miniaturization of electromechanical
systems have enabled the fabrication of microelectromechanical
systems MEMS that offer several advantages over conventional
devices. The general advantages of using MEMS are small size
features, mass production at low cost, sensitivity, and reliability
1,2. The fluid volumes in these systems are pumped, controlled,
or manipulated during operation. Because of their physical and
engineering significance, micropumps form an active part among
the development of MEMS.
Micropumps can be generally classified as mechanical or nonmechanical micropumps. Mechanical micropumps have mechanical moving parts such as diaphragm and check valves. They require actuators or mechanisms to pump fluids. For example,
electrostatic, piezoelectric, and thermopneumatic actuators are
used in mechanical micropumps. The pumping action for nonmechanical micropumps does not involve moving parts. They mainly
depend on transforming nonmechanical energy into kinetic momentum energy for pumping the fluids. These various types were
proposed, researched, and developed to be used for various envi1
Corresponding author.
JOURNAL OF FLUIDS ENGINEERING. Manuscript received June 10, 2008; final manuscript
received February 2, 2009; published online April 14, 2009. Assoc. Editor: Paul
Durbin.
ronments and conditions. For example, drug biocompatibility,

safety, and desired flow rates must be met for drug delivery micropumps.
Mechanical pumps based on conventional centrifugal and inertia forces will not work at microscales where Reynolds number is
typically small. Microscale reduces the effect of centrifugal forces
and inertia forces and amplifies the effect of viscous forces.
Hence, viscous force is the dominant force at the microscale 3.
Positive-displacement pumps with a membrane or diaphragm actuated in a reciprocating mode are the typically used method for
micropumps. The problem with this type of pumps is that they
require even smaller valves and seal mechanism, which are nottoo-trivial micromanufacturing challenge. In addition, there are
problems of wearing or clogging and consequent leaking around
the valves 4. To overcome the design problems, Stemme and
Stemme 5 suggested replacing the check valves with a nozzle at
the inlet and a diffuser at the exit. The problem with the idea is
that restrictive operating conditions must be met for the pump to
operate properly 5.
Several types of nonmechanical micropumps have been developed. The Knudsen pump uses thermal-creep effect to move rarified gases. Ion-drag is used in electrohydrodynamic pumps. They
relay on the electrical properties of the fluid and are thus not
suitable for many applications 6,7,15. Electrokinetic pumps,
similar to electrohydrodynamic pumps, use the moving electric
field to displace the ions in the electric double layer of the electrolyte, rather than the charges in a dielectric fluid 6. The entire
micropump concepts mentioned above, although applicable, are
complex and require additional components for pumping action to
occur.
It is possible to generate axial fluid motion in open channels
through the rotation of a cylinder in a viscous fluid medium. This
incorporates both applicability and simplicity in design. Odell and
MAY 2009, Vol. 131 / 051105-1
Kovasznay 8 studied a pump based on this principle at high

Reynolds number. Sen et al. 9 was the first to propose a novel
viscous micropump. They carried out an experimental study on
different versions of such pump. The viscous micropump consists
simply of a transverse-axis cylindrical rotor eccentrically placed
in a channel. The pumping action depends on the net viscous
differential resistance between the small and large gaps, which
causes net flow along the duct. It depends on viscous force to
operate. The viscous force is always dominant for microscale devices. Hence, regardless of the viscosity of the liquid, this pump
can pump any kinds of liquids. The study of Sen et al. 9 focused
on the effect of the channel height, rotor eccentricity, and angular
velocity on the pump performance and on the fluid bulk velocity
in the duct. No-slip boundary condition was assumed in their
study.
Sharatchandra et al. 10 performed a numerical simulation of
the viscous micropump solving NavierStokes equations for the
case of the cylindrical rotor. Optimal values for optimum performance were obtained. The optimal values were obtained when the
channel height is around one and a half times the cylinder diameter for a fixed eccentricity or at the maximum eccentricity for a
fixed channel height. The maximum efficiency of the pump was
determined to be approximately 2.5% for the optimum channel
thickness.
Abdelgawad et al. 11,12 investigated the transient performance of the viscous micropump numerically. They determined
that rotor eccentricity is a major parameter that affects the transient performance of the micropump. They assumed no-slip
boundary conditions at solid surfaces. da Silva et al. 13 addressed the effect of geometric parameters such as channel height
and rotor eccentricity on the mass flow rate and power consumption of two-dimensional microscale viscous pumps. They showed
that the rotor eccentricity and the channel height have a major
effect on the mass flow rate generated by the rotor and on the shaft
power demanded by the rotor. Preliminary simulations showed
that mass flow rate is maximized when the eccentricity is small.
They studied three different viscous micropump configurations, a
straight housed pump I-shaped housing, and two curved housed
pumps L- and U-shaped housings. They also presented optimal
geometric dimensions of several values of the Reynolds number
and pressure. Sharatchandra et al. 6 briefly investigated the effect of slip for gas flows. They mentioned that the effect of slip is
insignificant when Knudsen number is greater than 0.1. Matthews
and Hill 14 used the hydrodynamic theory of lubrication to obtain an analytical solution, which relates flow rate, rotation rate,
pressure drop, and applied torque as functions of two geometric
parameters for a viscous pump. Their results were valid for very
small values of Reynolds number where inertia effect could be
neglected. Also they assumed that the gaps between the cylinder
and the channel walls are small compared with the radius of the
rotor for the lubrication theory to be used. They did not include
compressibility effect. Hence, their study could be considered a
special case of our present study. It is the objective of this study to
have detailed investigations about the effect of slip on the pumping performance of viscous micropump.
In this study, we investigate the effect of slip flow on the pump
performance. It has been well documented and shown that fluid
flows in small devices differ from those in macroscopic devices.
The conventional flow models such as the NavierStokes equations with a no-slip boundary condition at a fluid-solid interface
cannot predict the operation of MEMS-based devices. We started
with a description of micropump configuration and numerical
modeling. Next, we validate our numerical results by comparing
our present results to those reported by Sharatchandra et al. 6
and Abdelgawad et al. 11. After the validation of the model we
proceed with determining the effect of slip boundary conditions
on pump performance. We also investigated the performance for
several flow parameters. Finally, we sum up the main findings of
this work.
051105-2 / Vol. 131, MAY 2009
Fig. 1 2D model of viscous micropump
Governing Equation
The viscous micropump is composed of two parallel plates with

transverse-axis rotating shaft placed asymmetrically within the
duct, as shown in Fig. 1. The fluid between the two plates is taken
to be Newtonian, compressible, and isothermal with constant
properties. Both upper and lower plates are held stationary while
the shaft rotates clockwise with constant angular velocity .
We take the diameter of the shaft D as our characteristic length
scale. Nondimensional geometric parameters are obtained by dividing by length scale D.
x =
x
,
D
y =
y
,
D
H =
H
,
D
Lu =
Lu
,
D
Ld =
Ld
,
D
d
D
1
where Ld is the downstream distance, Lu is the upper stream distance, H is the channel height, and is the cylinder eccentricity.
We define the dimensionless x- and y-component of fluid velocity as
u =
u
D/2
v =
v
D/2
We define the density as
= r o
where o is the density at the exit of the channel, which is assumed to be constant, without loss of generality, we assume r to
be equal to unity.
The nondimensional pressure rise pump load is defined as:
p =
Pout Pin
2/D2
where Pout is the exit pressure, Pin is the inlet pressure, is the
density of fluid, is the rotational speed of the shaft, and is the
kinetic viscosity of the fluid.
We define the drag, lift, and moment coefficients as follows:
CD =
CL =
CM =
FD
1
oU 2D
2
FL
1
oU 2D
2
M
1
oU 2D 2
2
where FD, FL, and M are defined as

FD =
Rp cos d
Rr sin d
Table 1 Demonstrates the mesh convergence of dimensionless average flow velocity u, Re= 1, H = 1.5, and = 0.025
Rp sin d
Rr cos d
10
No. of elements
11
639
1563
6413
22,352
0.1001
0.10368
0.10514
0.1058
ui ui+1
100%
ui
M=
R 2 rd
The above equations represent forces per unit length exerted by

the fluid on rotor. R is the rotor radius.
The nondimensional mass and momentum balance equations
are as follows:
V = 0
12
Re2V V = P + Re 2V
13
where Re= D2 / 2 is the Reynolds number based on rotor angular velocity and shaft diameter.
A gas flow is classified to be in the slip regime if the Knudsen
number Kn is 0.001 Kn 0.1. Here the Knudsen number is defined as Kn= / H, where is the mean free path of gas molecules. For macrofluidic devices, Kn is very small, and no-slip
boundary condition at solid boundaries is a valid assumption. In
micromechanics of gases, Kn is not small and there is a slip at
solid boundaries 1517. The traditionally used no-slip boundary
conditions for velocity are no longer valid in the micromechanics
of gases. In the slip regime, the gas velocity at a solid surface
differs from the velocity at which the wall moves. Following
Maxwell 18, it has been proposed by Schaaf and Chambre 19
that at a solid surface in isothermal slip flow, the dimensionless
slip velocity is defined as
vs
vg
2 vg
+
Kn
14
where vs is the dimensionless tangential velocity of the surface,

vg is the dimensionless velocity of the gas, n is the coordinate
normal to the solid surface, and is called the tangentialmomentum accommodation coefficient. The equation above is
written in nondimensional form with velocities being nondimensionlized by the surface speed of the shaft D / 2 and channel
height H.
We define slip factor S as
2
Kn
S=
15
where S takes value from zero to unity. For S = 0 there is no slip

between the wall and the fluid; the slip increases as S becomes
larger.
FL =
N/A
3.466
1.40313
0.6076
viscous action. The differential viscous resistance between small

gap and large gap produces a net flow. Hence, the performance of
a viscous micropump depends on the eccentricity of the shaft with
respect to the duct walls. Thus, we introduce dimension d that
indicates the distance between the lower wall and the shaft see
Fig. 1.
In numerical simulations, the inlet boundary condition was
specified by pressure P = Pin = 0. This value will always be assumed zero gauge pressure at the inlet throughout all simulations.
The outlet boundary condition was specified by pressure P
= Pout. This value will be varied to simulate different pump loads.
In all numerical simulations Lu and Ld were taken to be 8D to
satisfy the fully developed flow conditions 1. The eccentricity
is taken to be 0.025 for all simulations.
FLUENT software adopts Maxwells models for slip velocity.
FLUENT defines the equations below to simulate slip velocity.
Uw Ug =
2
U
2
Kn
U Uc
g
Vg V ng = Vw
16
17
Here, U and V represent the velocity components that are parallel

and normal to the wall, respectively. The subscripts g, w, and c
indicate gas, wall, and cell-center velocities. The parameter is
the distance from the cell-center to the wall, and is the momentum accommodation coefficient of the gas. The mean free path is
computed as follows:
=
k BT
2v2p
18
where kB is the Boltzmann constant equal to 1.38066

1023 J / K. The normal component of gas velocity is the same
as the wall normal velocity no penetration. The tangential component values lie somewhere between the cell-center and the wall
values. Slip velocity conditions were used in all internal surfaces
of the channel. As a convergence criterion, the solver iterated the
equations until the scaled residuals of continuity, x-velocity,
y-velocity, and energy were less than 10 6.
Numerical Modeling
It is often difficult to obtain accurate analytical results using

governing equations for viscous micropump. Hence numerical
simulations become an attractive method to predict the performance of the micropump.
To evaluate the performance of viscous micropumps, the commercial finite volume software FLUENT is used in this study 20.
The 2D NavierStokes equations coupled with energy equation
were numerically solved. The numerical domain consists of two
parallel plates held stationary and an asymmetrically placed shaft
rotating clockwise shown in Fig. 1. Due to viscous action, the
clockwise-turning shaft impels the fluid from left to right against
an external pressure difference defined as P = Pout Pin. This
pressure difference is called backpressure or pump pressure.
It has been shown by Sen et al. 4 that a rotating body, asymmetrically placed within a duct, will produce a net flow due to
Validation of the Model
The mesh density was exhaustively tested in order to guarantee

that the results were mesh-independent. Table 1 illustrates the
mesh convergence of the average fluid velocity. A model that consists of 22,352 elements was used for the entire simulation. Figure
2 shows part of the mesh density numerical domain where a finer
mesh was used around the rotor surface.
Fig. 2 Part of numerical domain mesh density
MAY 2009, Vol. 131 / 051105-3
Table 2 Validation of the simulation results; Re= 1, H = 1.5,

and = 0.025
0
7.5
15
30
u Present work
u 6
0.1058
0.075
0.042
0.0214
0.106
0.0736
0.043
0.021
Difference
upresent
uSharatchandra
100%
uSharatchandra
%
0.19
1.9
2.3
1.9
In order to validate the simulation model, the present results

were compared against previous results obtained by Sharatchandra et al. 6. Table 2 compares the effect of the backpressure on
the average fluid velocity u at the outlet plane. Let Re= 1, H
= 1.5, and = 0.025, where the average fluid velocity is defined as
u =
1
H
uydy
19
Table 2 shows that our present numerical results are in good

agreement with previous numerical results throughout P range
with an average difference on the order of 2%.
5.1 Effects of Slip Factor and Pump Load. The pump load
pump backpressure is modeled by increasing the outlet pressure
of the pump to simulate the pressure increase needed from the
pump to overcome the imposed external pressure. The effect of
slip was simulated by changing the values of slip factor. We investigated four different values of slip factor, namely, S = 0,
0.0027, 0.027, and 0.27. Figure 3 shows the change in the flow
rate at different pump loads and slip factors. It is clear that the
pump flow rate decreases linearly with increasing pump load. It is
also clear that pumping flow rate decreases with increasing slip.
When the external imposed pressure exceeds the maximum load
the pump can deliver, a backflow occurs and the average velocity
becomes negative. The maximum load that pumps can deliver
decreases with increasing slip. For example, the maximum pumping load is around 25 for no-slip. This value decreases dramatically to a value around 4 for Kn approaching 0.1 corresponds to
S = 0.27. Figure 3 shows that for S = 0.0027 corresponds to Kn
= 0.002, the deviation from no-slip is small. This behavior agrees
very well with the accepted classification of flow regime, that is,
for Kn greater than 0.001, the no-slip is a valid condition. From
the definition of S, S = 0.27 corresponds to Kn approaching 0.1 in
our simulations. For Kn approaching 0.1, the pump operating conditions degraded significantly. The maximum backpressure at
zero flow rate is only 0.0252 Pa. The maximum flow rate ob-
Fig. 3 Effect of slip factor on flow rate versus pump load at

Re= 1 and = 0.025, H = 1.5
051105-4 / Vol. 131, MAY 2009
Fig. 4 Total drag coefficient viscous and pressure as a function of pump load for different slip factors = 0.025, H = 1.5, and
Re= 1
tained at zero pump pressure is only 0.132 m3 / s.

Figure 4 shows how the drag coefficients vary as a function of
pump load for different slip factors. Drag coefficients are always
negative and increase with an increase in pump load. The drag
coefficient decreases with increasing slip factor. The deviation between no-slip S = 0.0027 and S = 0.027 is small. However, for S
= 0.27 there is a significant reduction in drag coefficient and hence
reduced traction on the rotor surface. Figure 5 shows how the
moment coefficients vary as a function of pump load for different
slip factors. It is clear that C M did not change significantly over a
wide range of pump loads. However C M decreases with increasing
slip factor. The moment coefficient for no-slip is greater than that
for S = 0.27 by six times. This is because the magnitude of shear
stresses at the rotor surface is proportional to the slip velocity.
In order to determine the efficiency of the micropump and the
effect of flow parameters, the pump efficiency is defined as
flow energy rise

mechanical energy input
20
where the input mechanical energy is M . M is the torque supplied to the rotor and is the angular velocity. The net energy of
the flow is only in the form of pressure increase at the pump
outlet, since the kinetic energy at the inlet and exit are equal and
there is no elevation change. Hence, micropump efficiency can be
obtained by
p
m
=
M
21
And in nondimensionless form as
Fig. 5 Moment coefficient as a function of pump load for different slip factors = 0.025, H = 1.5, and Re= 1
Fig. 8 Efficiency as function of Reynolds number for different

slip factors = 0.025, H = 1.5, and P = 1
Fig. 6 Efficiency as a function of pressure for different slip
factors = 0.025, H = 1.5. and Re= 1
upH
C M Re2
22
Simulations were carried out to determine the micropump efficiency and the effect of the flow parameters on the performance of
the pump. Figure 6 shows how the efficiency varies as a function of pump load and slip factors. The pump efficiency is high
when the flow rate and pump load are high and the coefficient of
moment is low. Since for a fixed slip factor, the coefficients of
moment do not change considerably, the dominant terms are the
pump load and flow rate. Since the flow rate is inversely proportional to the pump load, there should be an optimum operating
condition of the pump. Figure 6 shows that the optimum operating
condition lies in the middle the pump loads. For example, for
no-slip, the pump delivers flow when P varies from 0 to 25 see
Fig. 3. Figure 6 shows that the maximum value of efficiency
corresponding to pump load is equal to half of the maximum
backpressure zero flow rates. The same is true for different values of slip factor. Generally, the effect of increasing the slip factor
is the increase in pump efficiency and the shift of the optimal
point to the left. For given values of H, P, and Re, pump
efficiency reduces to the ratio between u and C M . Increasing slip
factor reduces both u and C M . Numerical results predict that C M
decreases more than u with increasing slip factor. However, when
Kn approaches 0.1, average velocity experiences a steeper reduction, while moment coefficient still decreases smoothly. Figure 6
shows that pump efficiency has been greatly degraded for S
= 0.27 compared with S = 0.027 or smaller. This agrees well with
Ref. 6. For example, for Kn approaching 0.1, maximum efficiency is only about 1.3%. Maximum efficiency for no-slip is
about 2.2%, which matches previously published results.
function of Reynolds number and slip factors. The flow rate increases linearly with increasing Reynolds number. Also, flow rate
decreases with increasing slip factor.
Figure 8 shows that pump efficiency stayed constant as a function of the Reynolds number. This was due to the fact that efficiency is inversely proportional to the product of C M and Re2.
Figure 9 shows that C M and CD decrease with increasing Reynolds number, so the product stayed almost constant.
5.3 Effect of Rotor Eccentricity. Simulations were carried
out to investigate the effect of rotor eccentricity on pump performance for several values of slip factors. External pressure was
fixed to 1. Reynolds number was fixed to unity. Figure 10 shows
flow rate as a function of rotor eccentricity for different slip factors. As S and increase, less net flow rate is predicted. Although
higher values of S creates more forward flow in the upper gap and
more backward flow in the lower gap, but the net effect will be
less net flow as S increases. Also, as increases, the width of the
5.2 Effect of Reynolds Number. To investigate the effect of

Reynolds number on pump performance, numerical simulations
were carried out for various values of Re. External pressure difference was fixed to be 1. Figure 7 shows flow rate variations as a
Fig. 9 Moment coefficient as a function of pump load for different slip factors = .025, H = 1.5, and Re= 1
Fig. 7 Flow rate as a function of Reynolds number for different

slip factors = 0.025, H = 1.5, and P = 1
Fig. 10 Flow rate as a function of rotor eccentricity for different slip factors Re= 1, H = 1.5, and P = 1
MAY 2009, Vol. 131 / 051105-5
Fig. 13 The effect of S on the optimized eccentricity

Fig. 11 Moment coefficient as a function of rotor eccentricity
for different slip factors Re= 1, H = 1.5, and P = 1
lower gap increases and that of the upper gap decreases. This
causes a reduction in the amount of the forward flow in the upper
gap and an increase in the amount of the backward flow in the
lower gap. The net effect will be a reduction in Q with increasing
. Our results agreed well with previously published data that flow
rate increases with decreasing .
Figure 11 shows the moment coefficient as a function of rotor
eccentricity for different slip factors. The values of C M always
decrease as both S and increase. Increasing S implies that the
slipping effects at the rotor surface increases and this yields less
velocity gradients and as a result, less shear forces and C M . In the
same manner, as increases, the slipping effects will decrease due
to the less velocity gradients near the rotor surface. As a result,
less C M values are predicted. However, for S = 0.27, the moment
coefficient reduces slightly with increasing .
Figure 12 shows efficiency variation with S and . For small
as S increases the efficiency increases. For a given fixed driving
force at the rotor surface and as S increases, the flow in both the
upper and lower gaps will be enhanced by the slipping effects due
to the less wall resistance against the flow. This implies that both
the forward flow in the upper gap and backward flow in the lower
gap will increase as S increases. The flow in the upper wider gap
is subject to less slipping effects due to its less velocity gradients,
as compared with the flow in the lower gap. But since the upper
gap has a larger area the net effect will be enhancing the forward
flow as S increases and hence improving pump efficiency. At large
the lower gap has a significant area and the effect of will be to
decrease regardless of the slipping effects. Within this limit of
relatively larger , the increase in will increase the area of the
backward flow and decrease the area of the forward flow. The net
effect will be a reduction in the net flow as increases and this
causes a reduction in regardless of the values of S, because at
large values of the slipping effects become insignificant.
Tracing the behavior of a single curve one may observe that
there is an optimum value for the behavior of with . There are
Fig. 12 Efficiency as a function of rotor eccentricity for different slip factors Re= 1, H = 1.5, and P = 1
051105-6 / Vol. 131, MAY 2009
two opposite effects for increasing . Increasing will increase

the area of the lower gap that allows for more backward flow to
return back. This causes the reduction in due to the reduction in
the net flow rate Q. On the other hand, increasing will reduce
the slipping effects within the lower gap, and this reduces the
amount of backward flow within the gap and improving as a
result. Now, from Fig. 12 one may observe that as small values of
increases, the slipping effects within the lower gap decreases
due to the reduction in Kn and the retardation effects of the wall
will overcome the increase in the gap width. As a result, less flow
will return within the lower gap, and the net effect will improve
the pump efficiency. As continues to increase, the slipping effects within the lower gap diminish lower Kn but the increase in
the gap width allows for more flow to return less net positive
flow rate, and this causes a reduction in . Note that the optimum
behavior for S = 0.27 still exist, but it appears at lower values of
that do not appear in the figure.
A better clarification for the effect of slip factors on pump performance for several values of rotor eccentricity is shown in Fig.
13. Pump performance increases with increasing slip factors for
small values of rotor eccentricity. However, for larger values of
rotor eccentricity pump performance decreases with increasing
slip factors. This agrees well with the explanation given above.
Conclusion
The performance of the viscous micropump with slip flow for

gaseous were investigated numerically. A solution of 2D Navier
Stokes equations coupled with energy equation was numerically
obtained. The geometrical parameters were fixed in this present
work except for rotor eccentricity. The effect of varying flow parameters such as varying Reynolds number and backpressure were
studied separately for various values of slip factor.
The drag, lift, and moment coefficient were studied for different
flow parameters. The drag coefficient was always negative, as
expected, and the pressure contribution to the drag force was always higher than the viscous force contribution, sometimes one
order of magnitude higher. The drag coefficient increases with an
increase in external imposed pressure. The drag coefficient decreases with an increase in Reynolds number. Drag coefficient
decreases with increasing slip factor. The lift coefficient is much
smaller than the drag coefficient. The moment coefficient does not
change significantly with changing external imposed pressure.
However, the moment coefficient decreases with increasing slip
factor, which also decreases with increasing Reynolds number.
Pump efficiency varies with external imposed pressure and slip
factor. However, for fixed slip factor and external pressure, pump
efficiency stays almost constant with varying Reynolds number in
the present study range. For the given Reynolds number, maximum efficiency occurs around half of the maximum pressure that
the pump can deliver. Maximum pump efficiency increases with
increasing slip factor.
Slip velocity degraded pump performance when Kn approaches
0.1. Present pump is able to pump fluid against 25 of the nondimensional external pressure for no-slip. On the other hand, it is
able to deliver against only four of the nondimensional external
pressure for Kn approaching 0.1. Pump performance increases
smoothly and slowly with increasing Knudsen number from 0 to

0.02. However, it experiences a steeper degradation when Kn
changes from 0.02 to 0.1.
For the given fixed driving force and imposed external pressure,
moment coefficient always decreases as both S and increases.
Also, as S and increase, less net flow rate is predicted. Finally,
there is an optimum value for the behavior of pump efficiency
with and S.
Nomenclature
CD
CL
CM
d
D
Kn
H
kB
L
Ld
Lu
M
Pin
Pout
P
Re
S
T
u
u
U
vs
x, y
v
V
vg
drag coefficient
lift coefficient
moment coefficient
rotor distance from lower wall m
rotor diameter m
Knudsen number
channel height m
Boltzmann constant
channel length m
down stream distance m
upperstream distance m
moment N / m2
inlet pressure Pa
outlet pressure Pa
pressure Pa
Reynolds number
slip factor
torque N m
x-component velocity m/s
mean velocity m/s
rotor surface velocity m/s
tangential velocity of the solid surface m/s
Cartesian coordinates m
y-component velocity m/s
velocity vector m/s
velocity of the gas m/s
cylinder eccentricity
pump efficiency
mean free path m
density at channel exit kg/ m3
density kg/ m3
density
tangential-momentum accommodation
coefficient
v Lennard-Jones characteristic length m

rotor angular velocity rad/
shear stress N / m2
indicates dimensionless variables
References
1 Judy, J. W., 2001, Fabrication, Design and Applications, Smart Mater.
Struct., 10, pp. 11151134.
2 Ziaie, B., Baldi, A., Lei, M., Gu, Y., and Siegel, R. A., 2004, Hard and Soft
Micromachining for BioMEMS: Review of Techniques and Examples of Applications in Microfluidics and Drug Delivery, Adv. Drug Delivery Rev., 56,
pp. 145172.
3 Gad-el-Hak, M., 1999, The Fluids Mechanics of MicrodevicesThe Freeman Scholar Lecture, ASME J. Fluids Eng., 121, pp. 533.
4 Sen, M., Wajerski, D., and Gad-el-Hak, M., 1996, A Novel Pump for MEMS
Applications, ASME J. Fluids Eng., 118, pp. 624627.
5 Stemme, E., and Stemme, G., 1993, Valveless Diffuser/Nozzle-Based Fluid
Pump, Sensors and Actuators, Physica, 39, pp. 159167.
6 Sharatchandra, M. C., Sen, M., and Gad-el-Hak, M., 1997, NavierStokes
Simulation of a Novel Viscous Pump, ASME Trans. J. Fluids Eng., 119, pp.
372382.
7 Bart, S. F., Tavrow, L. S., Mehregany, M., and Lang, J. H., 1990, Microfabricated Electrohydrodynamic Pumps, Sens. Actuators, A, 2123, pp. 193
197.
8 Odell, G. M., and Kovasznay, L. S. G., 1971, A New Type of Water Channel
With Density Stratification, J. Fluid Mech., 50, pp. 535543.
9 Sen, M., Wajerski, D., and Gad-el-Hak, M., 1996, A Novel Pump for MEMS
Applications, ASME J. Fluids Eng., 118, pp. 624627.
10 Sharatchandra, M. C., Sen, M., and Gad-el-Hak, M., 1998, Thermal Aspects
of a Novel Viscous Pump, ASME J. Heat Transfer, 120, pp. 99107.
11 Abdelgawad, M., Hassan, I., and Esmail, N., 2004, Transient Behavior of the
Viscous Micropump, Microscale Thermophys. Eng., 8, pp. 361381.
12 Abdelgawad, M., Hassan, I., Esmail, N., and Phutthavong, P., 2005, Numerical Investigation of Multistage Viscous Micropump Configurations, ASME J.
Fluids Eng., 127, pp. 734742.
13 da Silva, A. K., Kobayashi, M. H., and Coimbra, C. F. M., 2007, Optimal
Theoretical Design of 2-D Microscale Viscous Pumps for Maximum Mass
Flow Rate and Minimum Power Consumption, Int. J. Heat Fluid Flow, 28,
pp. 526536.
14 Matthews, M. T., and Hill, J. M., 2008, Lubrication Analysis of the Viscous
Micro/Nano Pump With Slip, Microfluid. Nanofluid., 45, pp. 439449.
15 Fuhr, G., Hagedorn, R., Mller, T., Benecke, W., and Wagner, B., 1992, Microfabricated Electrohydrodynamic EHD Pumps for Liquids of Higher Conductivity, J. Microelectromech. Syst., 1, pp. 141146.
16 Richter, A., Plettner, A., Hofmann, K. A., and Sandmaier, H., 1991, A Micromachined Electrohydrodynamic EHD Pump, Sens. Actuators, A, 29, pp.
159168.
17 Harrison, D. J., Manz, A., and Glavina, P. G., 1991, Electroosmotic Pumping
within a Chemical Sensor System Integrated on Silicon, Proceedings of the
International Conference on Solid-State Sensors and Actuators Transducers,
pp. 792795.
18 Maxwell, J. C., 1879, On Stresses in Rarified Gases Arising From Inequalities
of Temperature, Philos. Trans. R. Soc. London, 170, 231256.
19 Schaaf, S. A., and Chambre, P. L., 1961, Flow of Rarefied Gases, Princeton
University Press, Princeton, NJ.
20 FLUENT 6.0 User Guide Manual, 2002, Fluent, Inc., NH.
MAY 2009, Vol. 131 / 051105-7
Pre-Stall Instability Distribution

Over a Transonic Compressor
Rotor
A. J. Gannon
G. V. Hobson
Turbopropulsion Laboratory,
Naval Postgraduate School,
Monterey, CA 93943
An investigation of the behavior of a transonic compressor rotor when operating close to

stall is presented. The specific areas of interest are the behavior and location of lowfrequency instabilities close to stall. In running close to stall, compressors can begin to
exhibit nonperiodic flow between the blade passages even when appearing to be operating in a stable steady-state condition. The data from the current rotor clearly show that
low-frequency instabilities were present during steady-state operation when stall was
approached. These frequencies are not geometrically fixed to the rotor and typically
appear at 0.30.8 of the rotor speed. The presence of these low-frequency instabilities is
known and detection is reasonably commonplace; however, attempts to quantify the location and strength of these instabilities as stall is approached have proved difficult. In
the current test fast response pressure sensors were positioned in the case-wall; upstream, downstream, and over the rotor blade tips. Simultaneous data from the sensors
were taken at successive steady-state settings with each being closer to stall. A time
domain analysis of the data investigates the magnitude of the instabilities and their
transient effect on the relative inlet flow angle. The data are also presented in the frequency domain to show the development and distribution of the instabilities over the
rotor as stall was approached. Initially the instabilities appeared within the rotor row and
extended downstream but at operation closer to stall they began to protrude upstream as
well. The greatest amplitude of the instabilities was within the blade row in the complex
flow region that contains phenomena such as the tip-vortex/normal-shock interaction and
the shock/boundary-layer interaction. In addition as stall is approached the growth of the
instabilities is nonlinear and not confined to one frequency. DOI: 10.1115/1.3112388
Keywords: transonic compressor, near-stall operation, nonaxisymmetric flows
Introduction
It has been observed experimentally that during operation near

stall in an axial flow fan or compressor rotor the flow can become
nonperiodic between blade passages and low frequency; below
once-per-revolution, instabilities in the flow field can be present.
These instabilities have been measured and reported for the current test rotor 1. Labels such as spike and modal instabilities
have been used to describe them 2,3. In some early studies the
instabilities were detected when stall was imminent and it was
thought that these may be a precursor event. More recent work has
shown that this phenomenon can be present even in so called
steady-state conditions when operating near to stall or surge 4,5.
In this work they are called low-frequency instabilities to avoid
confusion with precursor, spike, and modal instabilities and with
poststall rotating instabilities. While low-frequency instabilities
usually indicate that the machine operates near to stall they do not
necessarily indicate that stall is imminent. Experimental results
indicate that the magnitude of these low-frequency instabilities is
significant 1.
Understanding of these structures has significance in the design
of compressors with regard to their stable operation near to stall.
Simulations that force a periodic assumption may not accurately
predict stall or surge points, either being optimistic or pessimistic
making design optimization difficult. Close to stall this periodic
assumption becomes less accurate and it appears that in order to
JOURNAL OF FLUIDS ENGINEERING. Manuscript received August 18, 2008; final manuscript received January 29, 2009; published online April 23, 2009. Assoc. Editor:
Chunill Hah. Paper presented at the ASME Turbo Expo 2008: Land, Sea and Air
GT2008, Berlin, Germany, June 913, 2008.
properly model the flow field an unsteady simulation of the complete rotor needs to be performed. The instabilities that resulted in
the rotor meant it experienced as much as a 2 deg change in the
inlet flow angle near stall over a 510 revolution period 1.
This study aims to more accurately quantify and present the
magnitude and frequency of these instabilities in a transonic rotor.
To do this simultaneous results of a number of sensors are combined to show the distribution of the instabilities over the rotor.
The frequency and distribution of the instabilities are shown at
various operating points each successively closer to stall. The distributions are of interest as the region of greatest magnitude of the
instabilities occurred in the complex flow region containing the
tip-vortex shock interaction and shock boundary-layer interaction.
In addition as stall was approached the low-frequency instabilities
are detectable well upstream and downstream of the rotor.
The detection of these low-frequency instabilities required the
use of high-speed pressure sensors and good signal processing
techniques. It will be shown that the instabilities do not behave in
either a linear fashion toward stall nor do they occur at predictable
frequencies. In addition the magnitudes of the instabilities were
significant, which resulted in observable changes in the bulk flow
properties.
There may be many possible mechanisms for the production of
these low-frequency instabilities. The fact that the present test
case was a rotor-only machined blisk seemed to indicate that they
were inherent in the flow without the need for external excitation.
Further improvement to stall margin for stable operation of future
compressors close to stall will demand better methods to simulate
these low-frequency instabilities and better understanding of their
causes and effects on machine operation.
MAY 2009, Vol. 131 / 051106-1
Fig. 2 Transonic test rotor
Fig. 1 Transonic test rotor
Experimental Apparatus
The rotor used, as shown in Fig. 1, was designed by Sanger 6

and subsequently tested by Gannon et al. 7. A detailed description of the stage, instrumentation, and test rig, as well as methods
to accurately determine the efficiencies, is contained in the papers
6,7. To summate the rotor consisted of 22 blades of aspect ratio
1.2 with a design tip speed of 396.2 m/s 1300 ft/s, diameter of
0.2794 m 11 in., and axial chord of 22.4 mm 0.884 in.. The
rotor-only and stage design pressure ratios are 1.61 and 1.56, respectively, with a design mass flow rate of 7.75 kg/s 17.05 lb/s.
The rotor was driven by an air turbine with the pressure ratio
controlled by an upstream throttle.
2.1 Tip-Gap. An abradable strip was positioned over the rotor and before any runs took place the cold tip-gap was measured.
Initially the machine was run to a particular low speed, shut down,
and the new increased cold tip-gap measured. While this does
cause a casing groove it ensures a small tip-gap. This process was
repeated in steps until the design speed was reached. At 100%
speed the rotor tip-gap was 25 m 0.001 in.. This allowed the
hot tip-gap to be inferred for different speeds as shown in Table 1.
2.2 Instrumentation. Two types of instrumentation, one set
for steady-state performance measurement and a second for highspeed data capture, were installed on the rig. The steady-state
instrumentation consisted of 2 upstream and 20 downstream Kiel
stagnation-pressure probes. Stagnation-temperature probes to
measure the efficiency consisted of two upstream and nine downstream 7. An additional temperature sensor was embedded in a
static port downstream of the rotor in the case-wall. The number
of temperature probes has been increased from six to nine as this
allows for more accurate efficiency measurement especially at off
design conditions when the downstream temperature profile was
distorted. The Kiel probe angles were also turned away from axial
as no stator row was present. The angles were based on the
streamline curvature code output from the design report of the
rotor 8. The angles are measured from axial, range from 45 deg
at the hub to 39 deg at the tip. The stagnation Kiel probes used to
measure the stagnation pressure have a high accuracy in the re-
gion of 54 deg in yaw and 49 deg in pitch. The probes were

arranged at 15 deg intervals on a plane four axial chord lengths
downstream of the rotor trailing tip as shown in Fig. 2.
The high-speed data system consisted of nine Kulite staticpressure transducers referred to as high-speed sensors over the
rotor as shown in Fig. 3. Kulite sensors 1, 7, 8, and 9 were
upstream of the rotor while sensor 2 was positioned at the leading edge and acted as the position reference. Sensors 35 were
over the rotor and sensor 6 downstream of the rotor. One impact
or stagnation Kulite sensor 10 was positioned upstream of the
rotor facing the flow. It was positioned 0.44 of the rotor diameter
upstream of the rotor at midspan 45 mm upstream of the nose
cone of the blade to measure the strength of the low-frequency
instabilities upstream of the rotor. The sensors were sampled at
196.6 kHz giving 19.6 readings per blade passage at design speed.
This was sufficiently high enough to ensure that the magnitude of
any aliased higher frequencies was small.
Experimental Program
The present test program concentrated on compressor operation

near to stall. The basic performance of the machine and the location of the stall lines were previously determined 7,9. In addition
the operating region for the appearance of the flow instabilities for
the rotor-only case was known 1.
Detailed data at various speeds from the subsonic to the transonic region were taken. The compressor was not stalled at 100%
speed in the rotor-only configuration due to concerns about the
high overspeed during stall. For brevity, only the 70% speed, subsonic operation, and the 90% and 95% speeds, transonic operation
are presented. Although the two transonic speeds are close the
behavior of the low-frequency instabilities close to stall is markedly different.
To measure data for a particular speed line the compressor was
set to operate in a steady-state condition at a point near to stall but
far enough away that low-frequency instabilities were not expected to be significant. A sample of high-speed data was then
taken. The compressor operating point was then moved incrementally closer to the stall point and data taken again. This allowed
the development of these low frequencies to be observed as stall
was approached. Data samples of 0.5 s were taken, which was
sufficient to resolve the low frequencies of interest. At 95% speed
this was equivalent to 214 rotor revolutions and approximately
100 cycles of the low-frequency flow variations.
Figure 4 shows the performance maps for the rotor-only configuration. Points at which the high-speed data were taken are
shown with A being closest to stall and F farthest away. Point
Table 1 Rotor tip gaps

Speed %
Tip-gap
Tip-gap/blade chord
051106-2 / Vol. 131, MAY 2009
70
90
95
127 m 0.005 in.

0.0025
25.4 m 0.001 in.

0.0005
12.7 m 0.0005 in.

0.0002
F was at the full open throttle position of the test rig, point E
was at peak efficiency, and point D was where low-frequency
instabilities first appeared. These points are used for reference
throughout the rest of the paper.
Fig. 3 Kulite sensor positions
Postprocessing
4.1 Time Domain. In order to isolate the signals of interest in

the time domain a selective filtering technique was required.
These are well documented in the fields of signal processing but a
basic outline found to be applicable to the current field is described here. A digital convolution filter method was used 10.
The steps are as follows: 1 A suitable filter was constructed in
the time domain. 2 The sampled signal and filter are both transformed into the frequency domain using a fast Fourier transform
FFT. It was important to note that the FFT of the sampled signal
and filter had to be of the same length. 3 A simple element by
element complex multiplication of the filter and sample in the
frequency domain was performed. 4 The product was converted
back to the time domain using an inverse fast Fourier transform
IFFT again of the same length to give the resultant filtered signal. The method for designing a low-pass digital filter was as
Fig. 4 Rotor-only performance maps
MAY 2009, Vol. 131 / 051106-3
follows. The desired cutoff frequency f c must be nondimensionalized by the sampling frequency
fc =
cuttoff frequency
sampling frequency
The roll-off bandwidth BW was defined as the desired frequency width where the signal dropped from 99% to 1% in
strength. The constant BW was defined as follows
BW =
roll-off width
sampling frequency
The bandwidth of the roll-off determined the number of points

required for the digital filter and BW was used to construct a
vector M from 2/BW to 2/BW in steps of 1 2/BW must be
rounded to an integer value
M=
2
2
:1:
BW BW
The so called sinc function was used to create an ideal low-pass

filter vector hsinc as follows
hsinc =
sin2 f cM
2 f c M
Due to the filter length being finite a truncation error occurs that
requires correction. To achieve this, a Blackman window is used
wBlack and results in near unity gain in the band-pass region
wBlack = 0.42 0.5 cos
2
4
+ 0.08 cos
M
M
An element by element vector multiplication of the sinc filter hc

and Blackman window wBlack gave the filter shape hBlack. The
filter must be normalized to ensure a unity gain in the pass region
to get the final filter h as follows
hBlack = hsinc wBlack h = hBlack/
Black
To create a high-pass filter, all the elements of hBlack were multiplied by 1 and unity added to the central sample. A low-pass,
h1, and high-pass, h2, filter were then combined to form a bandpass filter. Multiple band-pass filters can be added together but
only one was used here. With the desired digital filters constructed
the steps outlined earlier can be used to filter the sampled signal S
to obtain the filtered signal S f
S f = IFFTFFTS FFTh1 FFTh2
MATLAB was used to perform the filtering techniques, which had

built in functions to perform the above listed operations. All frequencies were normalized to the rotor speed in rpm. The bandpass frequency filter was from 0.3 to 0.85 with a roll-off of 0.05
rpm.
4.2 Frequency Domain. Observation of the signals in the frequency domain was also useful. In order to better isolate the lowfrequency signals the signals were transformed into the frequency
domain using FFTs as these were computationally cheap and easily implemented. A spectral analysis of the frequency data showed
the magnitude and frequency of the sampled pressures over the
rotor. The long data samples and high sampling rate ensured minimal aliasing and a good signal to noise ratio. To ensure that the
correct magnitude of a particular frequency was captured the
sampled signals were padded with zeros, usually eight times the
length of the sample. Practically this was done by performing a
longer FFT than the number of points in the sample. If this very
simple method was not used the magnitude would often be
cropped especially at lower frequencies 10, which were of special interest in this research.
051106-4 / Vol. 131, MAY 2009
Fig. 5 Experimental pressure contours near stall, 95% speed

A
Results
Figure 5 shows the experimentally obtained projection of the

pressure contours by the rotor on the casing. Superimposed in this
figure are the measurement locations of the Kulite sensors as the
rotor passes over them. It was obtained using 0.5 s of data at 95%
speed 214 rotations= 4717 blade passages. A full set of these
figures is presented in Ref. 11. The sensor positions can be cross
referenced with the low-speed frequency data presented in the
figures that follow. In addition the detached shock typical of nearstall flow, as well as the track of the tip-vortex as it impinges on
the shock, is indicated.
5.1 Time Domain. Observation of the filtered signals is useful in that is shows that their magnitude is not insignificant as
compared with the raw signal. The three main regions of interest
at near-stall operation are focused on the inlet just upstream of the
rotor, the rear part of the blade where the large low-frequency
instability magnitudes are found, and downstream of the blade as
this would affect any following blade rows.
Figure 6 shows an example of a signal passed through a bandpass filter with a pass band of 0.250.9 of rotor revolution. The
flow condition was at 95% speed at the closest stable operating
point near to stall point A Fig. 4. Results from three Kulite
sensors, 1 upstream, 4 within the blade row, and 6 downstream are shown. Underlying the filtered signals are the raw data
to give a visual comparison of the magnitudes. The data are presented for 50 rotor revolutions. A closer inspection of the raw
signal would show that the blade passing frequency is dominant as
expected.
It should be noted that if the flow was periodic in the blade-toblade sense, the filtered signal would be constant. This is clearly
not the case as shown in Fig. 6. Upstream of the rotor at sensor 1
the filtered signal was fairly small when compared with the raw
signal, which was dominated by the blade passing frequency.
Once inside the blade row at sensor 4 the magnitude of the
instabilities increased while the blade passing amplitude decreases
and the instabilities became an appreciable fraction of the pressure
variations. Once the flow had exited the rotor the instabilities were
still present and would affect downstream blade rows. In addition
to wake effects variations in the bulk flow would affect downstream blade row performance. In Fig. 6, for sensor 6 downstream of the rotor, the effects of the underlying instabilities can
be seen on the raw signal, which correspond to the underlying
low-frequency instabilities. In a simulation where a periodic assumption is made these would not be captured.
Figure 7 presents only the filtered signals on a larger scale with
the upstream impact-sensor included. It can be seen that there is
an amplification of the signal from the upstream impact-sensor
10 to the over-rotor sensor 4 and a slight decrease downstream
Fig. 6 Resultant filtered signal 95%
of the rotor to sensor 6. These signals would be responsible for

variations in the bulk flow not associated with the blade passing
frequencies. The above signals were sampled simultaneously so it
is also interesting to note that there are quite significant differences in the peaks. Where sensor 1 shows a small amplitude,
sensor 4 shows a large one. The spectral analysis shows this in
more detail.
Initial results looking at the variations of the inlet flow angle at
95% were presented by Gannon and Hobson 1. Here too the
assumption was made that the upstream stagnation pressure was
constant. The amplitude of this filtered signal at the inlet is small
when compared with the filtered static signal at the casing and
assuming it is constant results in a more conservative calculation
of the change in relative inlet angle.
Knowing the inlet static and stagnation pressures and using the
measured upstream stagnation temperature allow the inlet Mach
number to be calculated. Here the filtered signals derived from the
Kulite sensors were used. In addition to the flow Mach number the
blade tip speed was known and this allowed the low-frequency
variation in the relative inlet angles to be calculated using a basic
velocity triangle. Zero tangential inlet velocity was assumed but in

the presence of changing circumferential static pressures such as
those shown in Fig. 7 this was not strictly accurate. Figure 8
shows the variation in the rotor relative inlet angle for operation
near to stall. In the figure the low frequency of the variations is
evident with about 5 cycles for every ten rotor revolutions. A very
low-frequency cycle was also observed at 95% speed but this was
thought to be either a system oscillation or beats resulting from
interference between the different frequencies.
Table 2 shows the amplitude of the variation for near 200 revolutions. The table shows that the effects of the low-frequency
instabilities are significant. While it is not thought that the flow
separates at the leading edge of the rotor blade, changes in inlet
flow angle would affect the flow through the blade passage. As the
frequencies of the instabilities are known we see that one peakto-trough is roughly 0.75 of a revolution. Knowing the inlet speed
we can then show through a rough calculation that the blade will
have 14 chord lengths of flow from one peak-to-trough of a lowfrequency signal. This would be sufficient time for the boundary
MAY 2009, Vol. 131 / 051106-5
Fig. 7 Filtered signals through the rotor 95%
layer to respond to these signals and thus each rotor passage

would likely be operating along the relevant machine characteristic for a particular speed.
Tables 35 compare the peak-to-peak values of pressure for the
raw unfiltered signal and the filtered signal for 95%, 90%, and
70% speeds, respectively, when operating close to stall. The tabulated 95% speed data correspond to those from Figs. 6 and 7. The
filtered peak-to-peak values are a measure of the passage unsteadiness. At all speeds once the flow entered the blade row, the
magnitudes of the low-frequency instabilities increased from the
leading edge to the rear of the rotor. This is typical of diffuser type
devices that tend to amplify flow distortions. This contrasted with
the raw signal that was largest near the leading edge and decreased through the rotor thus increasing the ratio between the raw
and filtered peak-to-peak amplitude through the rotor. The ratio of
the peak-to-peak filtered over peak-to-peak raw data shown in the
fifth column allows for comparison between speeds. This is most
evident at probe 6 downstream of the rotor, which grows from
0.097 at 70% speed to 0.241 at 95% speed.
It is probable that it would be a peak and not a mean adverse
pressure gradient that would precipitate a stall. The data in Tables
35 indicate that a downstream blade row may be subjected to a
greater magnitude of instability than the one upstream of it. In the
present rotor forcing periodicity would cause a simulation to fail
to capture the nonaxisymmetry of the flow. It is thought that this
would cause the greatest error in the prediction of the stability
limit of the machine. Each individual passage would still be expected to operate along the unique characteristic of the machine.
5.2 Frequency Domain. A spectral analysis allowed for better observation of the distribution and magnitude of the lowfrequency instabilities over the blade row. The spectral analyses
were performed on the signals once they had been nondimensionalized to the inlet stagnation pressure Figs. 6 and 7.
A brief summary of the method is shown in the following fig051106-6 / Vol. 131, MAY 2009
ures. Figure 9 shows the complete spectral power distribution

from the single Kulite sensor 4 at 95% speed. The lowfrequency region of interest below the onceper-revolution signal
is indicated. The once-per-revolution signal was an indication of
imperfections in the rotor. The largest magnitude was as expected
at the blade passing frequency. Due to the high sampling speed a
significant roll-off at high frequencies was found meaning that
aliasing was minimal.
Figure 10 expands the low-frequency region presented in Fig. 9
on a linear scale. It shows the low-frequency instabilities that are
present when operating close to stall. No clear higher-frequency
harmonics of the signals were observed indicating that the instabilities are fairly pure sinusoidal signals. Recalling that the physical axial position of the Kulite sensors relative to the blade was
known Fig. 5 it is possible to present the distribution of the
signals over the blades as shown in the next sets of figures.
Figure 11 shows the combined signals of the Kulite pressure
sensors over the rotor presented as a waterfall plot. At each sensor
position the relevant low-frequency spectrum similar to that
shown in Fig. 10 is plotted. The four operating points presented
correspond to those in Fig. 4 at 95% speed. All magnitudes from
points A to D are plotted on the same vertical scale. A cursory
observation of the figures indicates that there is a substantial
change in the distribution and frequency of the low-frequency
instabilities as stall is approached.
At point D small instabilities were present at the leading edge
of the rotor row and decayed toward the trailing edge of the blade.
At point C, where the rotor was operating closer to stall, the first
significant low-frequency instabilities occurred in the aft portion
of the rotor at sensors 4 and 5. An inspection of Fig. 5 showed
this to be behind where the normal-shock wave impinged on the
blade suction surface. This region contains a thick boundary layer
especially when operating close to stall. Moving even closer to
stall to point B the dominant instability frequency abruptly
Fig. 8 Relative inlet angle variation close to stall
changed and in addition began to protrude ahead of the rotor

blade. This could be due to the tip-vortex shock interaction becoming stronger and more complex. At point A, the closest
Table 2 Variation in inlet flow angle when operating near to
stall
Speed and point
Mean inlet
flow angle
deg
rms variation
deg
Peak-to-peak variation
deg
95% A
90% A
70% B
70.6
71.5
75.1
0.351
0.341
0.269
2.04
2.21
1.65
Table 3 Variations in sensor pressure at 95% speed
stable operating point before stall, one dominant frequency

emerged. The position of greatest magnitude was found again at
sensors 4 and 5 from midrotor to the trailing edge of the blade
within the thick boundary-layer region. It could also be seen that
Sensor No.
Mean pressure
Raw
peak to peak
Filtered
peak to peak
Ratio
10
1
4
6
1.00
0.92
1.25
1.30
0.0502
0.5649
0.3281
0.1269
0.0048
0.0200
0.0359
0.0234
0.092
0.035
0.109
0.184
Sensor No.
Mean pressure
Raw
peak to peak
Filtered
peak to peak
Ratio
Sensor No.
Mean pressure
Raw
peak to peak
Filtered
peak to peak
Ratio
10
1
4
6
1.00
0.90
1.29
1.35
0.094
0.6069
0.3767
0.1453
0.0096
0.0215
0.0406
0.0351
0.102
0.035
0.107
0.241
10
1
4
6
1.00
0.97
1.14
1.17
0.0335
0.1426
0.1826
0.1029
0.0022
0.0074
0.0193
0.0100
0.065
0.052
0.106
0.097
MAY 2009, Vol. 131 / 051106-7
10
Blade passing
frequency
Once per rev

4
10
Magnitude
10
10
Low frequency region

10
10
95% Speed
Kulite 4
12
10
.1
10
Frequency (Normalized to Rotor)
100
Fig. 9 Spectral analysis from FFT log-log scale
this instability was detectable upstream of the rotor in sensors 1

and 9 and as a small peak at the impact-sensor 10 more than
five axial chords upstream.
For completeness Fig. 12 shows the low-frequency regions for
peak efficiency and choked operation of the rotor at 95% speed. It
can be seen that any low-frequency instabilities are barely detectable above the background noise. There was a slight increase in
their strength as the compressor loading increased from a choked
condition to peak efficiency.
Figure 13 presents the low-frequency instabilities for the 90%
speed case again showing the four points closest to stall Fig. 4.
Again a constant vertical scale was used throughout the series of
figures. The vertical scale is smaller than for 95% speed due to the
lower speed. What was notable is that the relatively small difference in speeds caused a significantly different type of behavior in
the instabilities as stall was approached. Point D was the first
x 10
point where instabilities were significant. Operating slightly closer

to stall at point C the magnitude of the instabilities increased.
The largest magnitudes were again in the midchord region of the
blade and a small instability was seen to protrude upstream. Moving closer to stall to point B four dominant frequencies began to
emerge and the first significant signs of instabilities protruding
upstream began to occur. As at 95% speed the largest region of
instability was in the rear to trailing part of the blade and as stall
was approached the instabilities did not decay behind the blade
row. At point A, the closest stable operating point to stall, two
dominant instabilities emerged at 0.5 and 0.75 of rotor speed. The
lower frequency protruded and was detectable well upstream of
the blade row. The higher frequency had a larger magnitude but
increased toward the rear of the blade row. Unlike at 95% speed,
at 90% speed no single frequency emerged dominant suggesting
that the mechanisms that cause these instabilities can act indepen-
Detailed view of low

3.5 frequency region
Magnitude
2.5
Low frequency instabilities
95% Speed
Kulite 4
1.5
0.5
0.3
0.4
0.5
0.6
0.7
Frequency (Normalized to Rotor)
0.8
0.9
Fig. 10 Low-frequency region linear scale
051106-8 / Vol. 131, MAY 2009
Fig. 11 Near-stall low-frequency instabilities 95%
dently of each other.

Figure 14 shows the low-frequency instabilities for a subsonic
case at 70% speed. The magnitude of the instabilities was lower at
70% speed but still detectable. Again a nonlinearity in the growth
of the instabilities toward stall was present. At point D the instabilities first appeared and increased noticeably when approaching stall to point C. Here two peak instabilities appeared, once
again in the rear section of the blade passage. When moving
closer to stall to point B more instability frequencies appeared
but the peak magnitude was reduced. As with the transonic cases
some instabilities protruded upstream of the rotor row.
Fig. 12 Far from stall low frequencies 95%
Discussion
The ability to observe the collective behavior of the lowfrequency instabilities allowed some insight into their possible
causes and the prestall characteristics of some transonic machines.
It is important to recall that the periodic assumption of the flow in
each passage being the same and steady relative to the blade does
not always hold when operating close to stall 1,5. Typical nearstall flow of a transonic machine contains a normal detached
shock ahead of the blades with a tip-vortex that intersects the
shock, which distorts it as shown in Fig. 5. The shock impinges on
the suction surface of the blade ahead of it.
6.1 Protruding Instabilities. The so called protruding instabilities indicated in the previous figures were most likely due to
the movement of the normal shock relative to the blade row as the
shock clearly protruded upstream of the blade. At peak efficiency
the shock in a transonic rotor is attached to the leading edge and
thus geometrically fixed to the rotor. A detached normal shock
would be affected by any change in the passage blockage and a
changing position relative to the blade would have been detected
upstream of the rotor. This movement may be due to multiple
mechanisms as indicated by the more than one protrudinginstability that were sometimes present Fig. 13, A and B. The
tip-vortex may also not be steady near stall and could affect the
normal shock, which of course in turn would affect the following
blades flow.
MAY 2009, Vol. 131 / 051106-9
stall in a transonic rotor is a region of separated flow that spreads

and becomes large enough to cause a change in the operation of
the compressor at a macroscale, usually detectable by a large reduction in the pressure ratio and mass flow rate. This distinction is
important as rotors can operate with localized regions of separated
flow with no adverse affects. The actual stall event itself is usually
very rapid and develops in a stall cell within five to ten revolutions 4,9. The low-frequency instabilities may cyclically increase the risk of a region of flow becoming a large scale separation.
6.2 Peak Instability Region. The largest instabilities that appeared were within the blade passage at sensors 35 at all
speeds. In this region the tip-vortex would have been interacting
with the detached normal shock as mentioned before. This also
corresponded axially to the location of the shock impinging on the
blade suction surface. While the vortex-shock interaction has a
strong influence in the tip region the shock-boundary impingement region occurs for a significant part of the span in a transonic
machine especially at higher Mach number.
6.3 Instability Behavior. The behavior of the instabilities
was generally nonlinear when approaching stall with different frequencies becoming dominant as stall was approached. It is important to note, however, that as long as they remain below a certain
magnitude stall would not occur. The main difference between
95% and 90% speeds was that at 95% speed the two main instabilities coalesced. At 95% speed, one of the instabilities presumably acted as a strong forcing function causing the other to follow
suit Fig. 11. At 90% speed Fig. 13 the two instabilities remained independent.
The exact mechanism that causes the instabilities can only be
hypothesized here. They clearly moved relative to the rotor but
only appeared when the compressor began to approach stall. The
types of flow found in a rotor, such as the tip-vortex, its interaction with the shock, and the detached normal shock would vary
with small disturbances in the external flow field. A variation due
to a small disturbance in one passage would affect the flow in the
following passage akin to a stall cell. This type of behavior could
reinforce itself after a number of trips around the rotor and eventually lead to the larger detectable variations that were observed.
As stall is approached the adverse pressure gradients increase and
flows with adverse pressure gradients such as in diffusers and
rotors tend to amplify flow disturbances.
6.4
Influence on Stall. It is thought that the eventual cause of
051106-10 / Vol. 131, MAY 2009
6.5 Stall Prediction Improvements. The presence of instabilities, especially at multiple frequencies, makes the exact prediction of stall difficult. As noted it may take a number of revolutions
of the machine for the conditions in the passage that cause a stall
to be present. This may be different for each stall event. When
simulating the flow making using of a periodic boundary assumption for the blade rows this in effect numerically damps out the
low-frequency instabilities. Whether this makes the simulated prediction of stall pessimistic or optimistic depends on the machine.
Ideally full rotor simulations are needed and the simulations will
have to be conducted over many revolutions. If a full annulus,
transient simulation was performed the shocks, tip-vortex, and
boundary layers would be allowed to vary between passages. As
was noted though the actual change in the flow of a single passage
is quite slow, around 14 chord lengths peak-to-trough, which may
leave some scope to using existing single passage steady-state
methods to improve designs. A design change that would indicate
a stall-margin improvement in a steady-state simulation is likely
to lead to a stall-margin improvement in the real machine. However, if the low-frequency instabilities increased in magnitude as
this study seems to indicate the magnitude of the improvement is
likely to be less than predicted.
Conclusions
The distribution and magnitude of low-frequency instabilities in

a transonic compressor rotor operating close to stall have been
presented. Using effective sampling techniques, instrumentation,
and appropriate numerical methods they can be isolated and observed. In isolating these low-frequency signals through the use of
filtering techniques their magnitude in the time domain can be
measured. It is shown that their magnitude was significant when
compared with the dominant blade passing frequency. A spectral
analysis of the instability region in the frequency domain was
used to investigate the development and distribution of the instabilities over the rotor blade as stall was approached. Three different speeds, one in the subsonic and two in the transonic region,
were investigated. The different behaviors served to demonstrate
the unpredictable nature of the development of low-frequency instabilities. The instabilities were present upstream of the rotor and
suspected to be due to the normal shock protruding ahead of the
rotor and moving relative to the blade leading edge. The region of
greatest magnitudes was found at midchord on the blade row,
which contained the tip-vortex interacting with the detached normal shock and the normal shock impinging on the blade suction
surface. The clear presence of these low-frequency instabilities
means that the periodic boundary condition commonly used in
simulations becomes less accurate as stall is approached. This
could be one possible reason for the difficulty in predicting stall
points with current methods.
Acknowledgment
The present study was part of the compressor research program
sponsored by the Propulsion and Power Department of the Naval
Air Warfare Centre, Patuxent River, MD with Ravi Ravindranath
as the technical monitor.
References
1 Gannon, A. J., and Hobson, G. V., 2007, Pre-Stall Modal Instabilities in a
Transonic Compressor Rotor, ISABE, Beijing, China.
2 McDougall, N. M., Cumpsty, N. A., and Hynes, T. P., 1990, Stall Inception in
Axial Compressors, ASME J. Turbomach., 112, pp. 116125.
3 Camp, T. R., and Day, I. J., 1997, A Study of Spike and Modal Stall Phenomena in a Low-Speed Axial Compressor, ASME Turbo, Orlando, FL, Paper No. 97-GT-526.
4 Bergner, J., Kinzel, M., Schiffer, H.-P., and Hah, C., 2006, Short LengthScale Rotating Stall Inception in a Transonic Axial Compressor: Experimental
Investigation, ASME Turbo, Barcelona, Spain, Paper No. GT2006-90209.
5 Hah, C., Bergner, J., and Schiffer, H. P., 2007, Rotating Instability in a Transonic Compressor, ISABE, Beijing, China.
6 Sanger, N. L., 1996, Design of a Low Aspect Ratio Transonic Compressor
Stage Using CFD Techniques, ASME J. Turbomach., 1183, pp. 479491.

7 Gannon, A. J., Hobson, G. V., and Shreeve, R. P., 2004, A Transonic Compressor Stage Part 1: Experimental Results, ASME Turbo Expo, Vienna, Austria, Paper No. GT2004-53923.
8 Sanger, N. L., 1999, Design Methodology for the NPS Transonic Compressor, TPL Technical Note 99-01, Naval Postgraduate School, Monterey, CA.
9 Gannon, A., Hobson, G., Shreeve, R., and Villescas, I., 2006, Experimental
Investigation During Stall and Surge in a Transonic Fan Stage and Rotor-Only
Configuration, ASME Turbo, Barcelona, Spain, Paper No. GT2006-90925.
10 Smith, S. W., 1997, The Scientist & Engineers Guide to Digital Signal Processing, California Technical Publication.
11 Gannon, A. J., Hobson, G. V., and Shreeve, R. P., 2005, Measurement of the
Unsteady Casewall Pressures Over the Rotor of a Transonic Fan and Comparison With Numerical Predictions, Paper No. ISABE-2005-1099.
MAY 2009, Vol. 131 / 051106-11
B. M. Marino
CONICET Researcher
Associate Professor
e-mail: bmarino@exa.unicen.edu.ar
L. P. Thomas
CONICET Researcher
Associate Professor
e-mail: lthomas@exa.unicen.edu.ar
Instituto de Fsica Arroyo Seco,
Universidad Nacional del Centro,
Pinto 399,
B7000GHG Tandil, Argentina
Front Condition for Gravity

Currents in Channels of
Nonrectangular Symmetric
Cross-Section Shapes
We study the variation of the Froude number at the front of gravity currents developed in
uniform channels whose cross-section shape depends on a parameter usually used in
many numerical and theoretical models. The thickness and front velocity of the dense
currents running on the bottom are greater for all the cases studied, resulting in a Froude
number greater than that corresponding to the rectangular cross-section shape. The light
currents developing along the upper boundary show the opposite trend. It is found that
the results are not related to the depth and width of the channel. The relationships
obtained agree with the results of laboratory experiments in which open and closed
channels of different cross-section shapes are used. DOI: 10.1115/1.3089537
Introduction
Gravity currents are flows driven by horizontal buoyancy differences of a fluid running inside another fluid of slightly different
density, at the top, at the bottom, or at an intermediate level. As
they are present in numerous important natural and industrial
events, the comprehension of the motion of its frontal region is
particularly important for a variety of practical situations. For example, the analysis of gravity currents is relevant for water quality
management in reservoirs as they carry suspended matters and
dissolved solids across the lake, often determining the distribution
of pollutant substances. Analogous environmental flows are
caused by the spreading of the cooling water from a power plant
in a river, the spreading of cool marine air under the warmer air
overlying land which is known as the sea-breeze, avalanches of
airborne snow particles, fiery avalanches and base surges formed
from gases and solids issuing from volcanic eruptions, and the
spread of accidentally released liquid natural gas. The properties
of this diverse range of dynamically related flows have been described and comprehensively reviewed by Simpson 1.
In particular, estuaries also exhibit a variety of phenomena
driven ultimately by the density differences associated with freshwater discharge, which may be understood in terms of gravity
current theory. One characteristic of the estuarine phenomenon is
the tidal intrusion of brackish water formed under low energy
conditions of the estuary and characterized by a range of small
magnitude tides and weak effects of the wind over the flow, as
typically occurs in narrow canals. The penetration of this saltwater wedge may and frequently does contaminate and become
useless the supply of potable water and water for industrial use in
coastal cities upstream, with serious consequences from the ecological and economical points of view. Although typical of estuaries with constricted connections to the sea, these tidal intrusion
fronts have been observed in reservoirs and lakes where cold turbid river inflow may plunge below the less dense ambient water
2.
The main physic phenomena underlying in many natural situations can be analyzed by means of a simple experimental system
constituted by a channel that is temporarily divided into two sections by a thin vertical barrier. Fresh water fills one section and
received January 21, 2009; published online April 1, 2009. Assoc. Editor: James A.
Liburdy.
salt water the other, and the levels are made equal. As soon as the
barrier is raised, the dense fluid starts to collapse and countercurrents begin to flow in opposite directions, as illustrated in Fig. 1.
However, the geometry for inflow channels to lakes, reservoirs,
and impoundments uses to play a non-negligible role 3. Most of
the experimental works related to these situations have been conducted in channels with a constant width and sloping bottom see
Refs. 4,5 and references therein and few researchers have dealt
with a varying-width channel or diffusers e.g., Refs. 610. It is
well known that the density underflow and plunge line in a diverging channel may have different regimes, and that the flow field
strongly depends on several parameters: inflow Froude number,
inflow aspect ratio width/depth, the divergent angle, and the bed
slope. In addition, a water course may expand its width and depth,
the flow may have a significant momentum, turbulence may be
important, and so on. However, as most studies are conducted in
idealized geometric setups where effects of individual parameters
are analyzed separately e.g., Ref. 9, it is not possible to generalize the available experimental results to fit all cases, especially
for field applications. Supplementing the existing experimental
results and providing insight into processes that are too difficult to
measure in the field or in a laboratory, computational analysis was
also performed see, for example, Ref. 11.
We are particularly interested in the influence of the transversal
variations of the properties of many currents in estuaries, rivers,
and natural or man-made channels. Even though the influence of
the factors mentioned above deserves additional separated investigations, some indication of their effects might be inferred from
previous studies. For example, Britter and Linden 4, and later
Monaghan 12, found that the front velocity parallel to the slope
is constant for angles less than 45 deg on rectangular cross-section
channels; similarly such an unimportant effect might be expected
for nonrectangular cross-section channels. Fronts in river flows
with significant momentum may be analyzed on a moving frame
of reference but taking into account the change in the stress of the
channel contours. Britter and Simpson 13 studied the influence
of the bottom stress on gravity currents generated in rectangular
cross-section channels in the laboratory, and Zhou 14 in estuaries. They found that the bottom stress may dramatically change
the height profile of the dense current, so the influence of the
momentum of the flow should be carefully analyzed for nonrectangular cross-section shapes as well.
Although in most previous studies, authors assumed that the
properties of the flow are independent of the cross-coordinate as
MAY 2009, Vol. 131 / 051201-1
z
Initial position of the gate
u2
h2
h1
u1
h0
x
Fig. 1 Sketch of the lateral view of the lock-exchange gravity

currents generated after removing the gate
occurs, in fact, in flows evolving along rectangular cross-section

channels, some attempts to model theoretically the flow driven by
lateral variation of the bathymetry were made. Nunes and Simpson 15 modeled the secondary transverse circulation in the
Conwy estuary North of Wales with a parabolic cross-section,
while Wong 16 and Wong and Munchow 17 modeled the
gravitational circulation in Delaware Bay with a triangular crosssection, and Valle-Levinson and Lwiza 18 simulated the flow
field in a coastal estuary through a rectangular cross-section with
a triangular depression in the middle. Engqvist and Hogg 19
used a simple relationship for the cross-section shape to derive
analytical solutions, corroborated by three dimensional numerical
simulations, describing unidirectional stratified flow through a
flat-bottomed nonrectangular channel.
Simpson 1 reported the empirical values of the ratio between
the initial front velocities of the salt-water and freshwater flows
and the thickness of the channel provided by laboratory experiments carried out in trapezoidal, triangular, and circular crosssection flumes. More recently, Thomas and Marino 20 presented
an analytical model based on global balances of mass and momentum, which concerns lock-exchange flows generated in open channels of triangular cross-section shape based on a few experiments.
They found that the ratio between the front velocities corresponding to the light and dense currents depends on an empirical parameter related to the ratio of the depths determined at the position of the gate that initially separates the fluids. That is, the front
velocity evolutions for each flow were coupled by means of a
relationship, simplifying the calculus, but the results do not allow
analyzing each front dynamics separately.
The goal of this paper is to introduce a theoretical treatment
describing the dynamics of both counterflows, which only includes the effect of the cross-section shape. The front condition
for steady-state currents generated in nonrectangular cross-section
channels is analytically derived independently for each flow improving the previous model reported by Thomas and Marino 20.
Thus, as the front velocity is related to the height of the following
flow, the results can be used to study situations where the attention
is focused on a single current as in geophysical and engineering
flows. The theoretical findings are compared with those provided
by laboratory experiments where lock-exchange flows in open and
closed channels of rectangular, triangular, and parabolic crosssections are generated.
Theoretical Model
The front condition of an energy-conserving gravity current

propagating into a rectangular channel of finite depth was studied
by Benjamin 21. He used the mass and momentum conservation
equations to obtain the value of the relationship between the velocity u of the frontal zone and the depth h of the following fluid
layer with a density excess over that of the ambient, which is
the Froude number
F=
gh
where g = / g is the reduced gravity, g is the acceleration

due to gravity, and F is of the order of unity. This relationship
coincides with that previously reported by von Karman 22
051201-2 / Vol. 131, MAY 2009
applying erroneously Bernoullis theorem to the lighter fluid

along a streamline in a turbulent region around the head.
The front condition 1 has been extensively applied at the frontal zone of gravity currents propagating at high Reynolds number
Re= uh / , where is the kinematic viscosity of the current fluid.
The value of the Froude number at the front is a function of the
ratio between the thickness of the current and the total depth of
the fluid or fractional depth, being also affected by the friction at
the rigid bottom, mixing, and a large-density-difference nonBoussinesq casesee Refs. 23,24 and references therein.
Now, consider a uniform nonrectangular cross-section channel
divided by a vertical gate in two parts, which initially contain a
fluid of density 1 and a lighter fluid of density 2 = 1 . The
upper contour may be either a solid boundary as in ducts or pipes
or a free surface; the latter case may be also thought as a half of a
cross-section of a duct. Removing the gate, two opposite gravity
flows produce a two-layer stratification, the lower layer with depth
h1x and the upper one with depth h2x, as illustrated in Fig. 1.
Although mixing is generated for high Reynolds number flows,
it is possible to consider an equivalent layer of unmixed dense
fluid for which the main properties of the flow are maintained as
noted by Marino et al. 25, and then the theoretical findings for
the no-mixing case can be extended to realistic cases.
The channel cross-section contour is assumed to be given by
yz =
bz
bz
for
y0
for
y0
where z and y are the vertical and transversal coordinates, respectively, and and b are constants. In particular, b = w / h0 is determined by the width w of the channel and the height h0 of the
denser fluid at the initial time. The cross-section area s that is
occupied by the denser fluid up to a height h is
s=
bz
bz
dydz =
2bh+1
+1
The use of the relationship 2 simplifies the analysis of the

basic properties of the flows generated in channels of nonrectangular cross-section shape, and is similar to the use of other functions as in most theoretical and numerical studies cited above. The
value 0 corresponds to a rectangular cross-section, = 1 defines a triangular cross-section, 1 provides convex crosssections, and 1 indicates sections with a distinctive central
depression.
Figure 2 illustrates the frontal zone of gravity currents developing on the bottom Fig. 2a and along the top of a closed channel
Fig. 2b. The motion is assumed to be steady, and the current
heads are at rest in a frame of reference moving with the corresponding constant velocity; the contact points O are stagnation
points. Control volumes used in the following analysis are marked
by dashed lines.
2.1 Dense Current. We take into account an approximation
in which a mean velocity is considered over the channel crosssection in each layer. In Fig. 2a the light fluid is entering through
the vertical plane AB with velocity u1. It overcomes the head of
the dense current, and the velocity u1D of the light fluid is again
uniform in plane CD located far enough from the front for any
perturbation originated in the head to be neglected. Continuity
implies that
u1 = u1D1 H1+1
where H1 = h1 / h0 is the dimensionless thickness of the current at

CD.
The pressure distribution is assumed to be hydrostatic in both
planes so that in AB the pressure as function of the height is given
by
Fig. 2 Schematic of the frontal zone of dense a and light b gravity currents
p = pA 2gz
The pressure pA is related to the pressure at the stagnation point O

by means of the Bernoulli equation
pA = pO
1
2
2 2u 1
Pressure on section CD is
p=
pC 1gz
for
pC 1gh1 2gz h1 for
0 z h1
h1 z h0
where pC is the pressure at point C in Fig. 2a, with pC = pO if the

internal flow within the dense current is negligible.
The momentum function for a given cross-section is
=
pdydz +
2u2dydz
Using Eqs. 47 on section AB, we obtain
2bh0+1
+1
1
AB =
pO + 2u21
2gh0
2
+1
+2
and
u21
gh0
u21
g 1h 0H 1
F21 =
g 1h 0H 1
1 H1+1 2
=
2
H+1
+2 1
1 + H1+1

1
+2
11
13
1/+1
14
and the Froude number given by Eq. 11 for a dense energyconserving gravity current becomes
where g1 = g1 2 / 2.
Equation 11 is found by relating the velocities and heights
corresponding to two sections, AB and CD, separated by a horizontal distance of the order of the head length. If the length of the
head is much smaller than the total current length, planes AB and
CD may be considered to be near to each other so that Eq. 11 is
valid at a front where the fluid velocity and the current height
are discontinuous functions of the horizontal coordinate x, as
usual in the theoretical analyses.
As expected, Eq. 11 with = 0 becomes the well known relationship for the rectangular cross-section channels proposed first
by Benjamin 21 and confirmed later by other authors see, for
instance, Ref. 26,
= 21 H1+12
F21c = 2 1
10
12
Equating Eqs. 11 and 13, the nontrivial solution for the current
depth results
H1c =
on section CD.
Since there are no external horizontal forces acting on the flow,
the net flux of momentum into the control volume including the
head is zero. Conservation of the horizontal component of the
momentum may then be written as AB = CD. Using the mass
conservation 4, the Froude number F1 for the dense current is
u21
=0
1 1 H12 H1H1
1 + H1
where = 2 / 1.
While Bernoullis equation is valid between points A and O
see Eq. 6 because of the laminar flow existing there, it is not
always valid between O and any point on plane CD since there is
usually energy dissipation in the head. But for currents in which
dissipation can be considered negligible, Bernoullis equation may
be also applied along a streamline between points on planes AB
and CD as, for example, the upper boundary of the channel to find
2bh0+1
+1
+1
pO
2gh0
1 2gh0H1+2
CD =
+1
+2
+2
2
1 2gh0H11 H1+1 + 2u1D
1 H1+1
1
+2
15
2.2 Light Current. The Froude number in this case is also

based on the mass and momentum balances as done in Sec. 2.1.
As shown in Fig. 2b, the denser fluid is entering through the
vertical plane AB with uniform velocity u2 and plunges under the
light fluid. In plane CD, which is located far enough from the
front for any perturbation originated in the head to be neglected,
the fluid velocity u2D is again uniform. Continuity implies that
u2 = u2DH1+1
16
where H1 is now related to the thickness of the dense fluid layer

below the light current Fig. 2b.
Assuming that the pressure at the stagnation point O is pO = 0,
and applying Bernoullis equation between points O and A, the
pressure on plane AB is
1
p = pO 2 1u22 + 1gh0 z
and
p=
17
pO 2gh0 h1 + 1gh1 z for
0 z h1
pO + 2gh0 z
h1 z h0
for
18
on plane CD. Calculating AB and CD by means of Eq. 8 and

using Eqs. 1618, the conservation of the horizontal component of the momentum AB = CD gives the Froude number F2 of
the light current as a function of H2 and g2 = g1 2 / 1 as
MAY 2009, Vol. 131 / 051201-3
F22 =
u22
g 2h 0H 2
2 1 1 H2+21 H2+1
+ 2
2 1 + H2+1H2
19
For rectangular cross-section channels, Eq. 19 is reduced to

u22
gh0
=0
= 1
2 H21 H2H2
1 + H2
20
When = 1 with 1, Eq. 20 differs from Eq. 12 by a

factor O; that is, the front velocities of the dense and light
currents are the same in the Boussinesq approximation as it is well
known. For nonrectangular cross-section channels the front velocities of the flows developed are not the same even in the Boussinesq approximation see Eqs. 11 and 19 with 1 and
0.
If there is no flow dissipation in the head of the light gravity
current, Bernoullis equation applied along the streamline that coincides with the channel bottom boundary suggests that
u22
= 21 H21 H22+1
gh0
21
Equating Eqs. 19 and 21, the height H2c for an energyconserving light gravity current is found to be the solution of the
algebraic equation
+2
+1
+1
1 H1S
+ 21 H1SH1S
2 H1S
=0
22
where H1S = 1 H2c. Thus, the Froude number for a light energyconserving gravity current is
F22c =
2 1 1 H2c+21 H2c+1
+ 2
2 1 + H2c+1H2c
23
Experimental Results
Six sets of laboratory experiments were performed in Perspex

channels with nonrectangular cross-sections, which are considered
as half the shapes defined by Eq. 2 with = 0.44 concave, Fig.
3a, 1.00 triangular, Fig. 3b, and 1.79 convex, Fig. 3c.
This experimental setup therefore includes a vertical wall at the
centerline of the flow to reduce the optical distortion when it is
observed laterally. The effects of this wall were found negligible
after comparing the results obtained using the triangular crosssection channel with those coming from the use of a rectangular
cross-section tank tilted 45 deg. The results for a half crosssection may be extrapolated to the complete cross-section by symmetry without losing validity.
The channels are 2.00 m long, 0.20 m wide, and 0.29 m high. In
addition, two sets of experiments were carried out in a rectangular
cross-section channel, where known results were reproduced.
Four sets of experiments were performed in channels with a Perspex upper boundary closed channels and another four sets with
a free surface. A central gate separates initially one part of the
channel containing fresh water from another with salt water. Each
set includes four or five experiments with density relative difference / 1 = 1 = 0.5%, 0.75%, 1%, 2%, and 4% approximately.
Densities were measured by an Anton Paar 4500 densimeter with
a precision of 5 105 g / cm3.
The nonrectangular channels were placed inside a greater rectangular cross-section tank depicted with dashed lines in Fig. 3
filled with fresh water to reduce the optical distortion due to the
nonparallel or curved lateral walls. The experiment starts when
the gate is removed, giving place to two opposite gravity currents
as illustrated in Fig. 1. The flows were observed laterally with a
charged-coupled device CCD video-camera located 6 m away
from the tank. Experimental setup and image processing are similar, as done by Marino et al. 25. An anamorphic lens is used in
order to magnify the vertical scale, reduce the horizontal scale,
and improve the resolution of the images. A fluorescent strip light
and a light-diffusing screen located behind the tank provide nearly
uniform backlighting. A calibrated quantity of dye is added to the
051201-4 / Vol. 131, MAY 2009
Fig. 3 Experimental setup to generate gravity flows in a channel of concave a, triangular b, and convex c cross-section
shapes
salt water to provide visualization. The light attenuation due to the

dyed water allows the cross-current averaged concentration to be
measured from which the cross-current average density may be
inferred, taking into account the nonrectangular cross-section
shape of the channel. Figure 4 shows an instantaneous image that
corresponds to the lock-exchange flows generated in a parabolic
cross-section open channel for an initial density relative difference
of 2%. Vertical and horizontal scales are different, and the light
intensity is proportional to the cross-current average density and
independent of the width of the channel in each point.
Fig. 4 Instantaneous image of the counterflows generated in a

parabolic cross-section shape tank after applying the image
processing. A dense current white moving to the right and a
light current black running in the opposite direction are
clearly distinguished.
1.5
1.0
=2
=1
=0
.5
=
0
1.2
0.8
H2
0.4
0.6
0.4
0.8
1.2
1.6
Fig. 5 Theoretical thickness of dense solid line and light

dashed line gravity currents for different cross-section
shapes. The experimental results obtained in closed and open
channels are represented by solid and open symbols, respectively, and by triangles and diamonds for dense and light currents, respectively.
From the density distribution, the front velocity and the height
of the opposite flows may be estimated. A few seconds after the
release, the currents enter into the constant velocity or slumping
phase that is maintained practically until the fronts reach the corresponding end-walls. Two layers with the most advanced point
separated from the bottom/top rigid contour are quickly developed. Then the height of each fluid layer, that is, h1 and h2 marked
in Fig. 1, was measured almost at the end of the run and at a
density relative difference of 90% with respect to the initial value.
Other criteria tested do not modify the results significantly more
details can be found in Ref. 25. The values of the front velocity
and current height lead to Reynolds numbers between 7000 and
35,000. Due to Reynolds number independence for density currents moving along boundaries when Re 103 27, laboratory
flows reproduced well many of the problems to which the results
of this paper might be directed. It is found that each set of experiments that is, different relative densities for the same crosssection shape gives similar results when dimensionless variables
H, u / gh01/2, and F are determined; hence they are averaged in
the following.
It is worthwhile noting that the reduced gravity g enters into
Eqs. 1, 11, and 19 through the product gh. That is to say, the
global results will not be strongly affected by mixing that reduces
g but increases h or viceversa, in approximately the same extent
as shown by Marino et al. 25. Therefore, the model is expected
to provide the correct scaling laws even if a strong mixing is
present. So, the Froude numbers F1 and F2 at the front should not
be modified significantly by the particular values of the Richardson numbers after the head.
Figure 5 shows the variation with of the dimensionless thicknesses of the currents obtained from Eqs. 14 and 22. In addition, symbols and vertical bars indicate the average values and the
standard deviations, respectively, calculated from the experimental data for different cross-section shapes. Theoretical and experimental results agree well. Note that H1 + H2 = 1 only for rectangular cross-section = 0; in such a case, both currents occupy half
1
the channel depth H1c = H2c = 2 as well known 1. For a nonrectangular cross-section channel H1c 1 / 2 H2c being greater the
difference between the thicknesses for increasing , and H1 + H2
1. A significant difference between the thicknesses obtained
with open and closed channels was not detected.
0.0
0.0
2.0
0.2
=
=1 0.5
0.4
F2
0.3
0.2
0.0
0.0
F1
0.9
F1(H1) , F2(H2)
0.6
H1
0.6
0.8
1.0
H1 , H2
Fig. 6 Froude numbers of the dense F1 and light F2 currents
as function of the respective heights H1 and H2 for different .
The energy-conserving solutions F1cH1c and F2cH2c given by
Eqs. 15 and 23, respectively, are plotted by thin solid lines.
Figure 6 shows the Froude numbers F1H1 above the line

corresponding to = 0 and F2H2 below the line for = 0, for
different values of obtained from Eqs. 11 and 19, respectively. The standard deviations for F and H shown in Fig. 4 are
similar, but are not included in Fig. 4 as horizontal and vertical
bars for the sake of clarity. The central thick line corresponds to
the classical solution for rectangular cross-section channels
= 0. There is no difference between the predictions for light and
dense currents provided that the reduced gravity for each current
is defined as done in the Sec. 2. This is not the case for nonrectangular cross-section channels 0. In these cases, the dense
fluid flow is faster than the light current of the same depth, and
this effect increases with . This is because, even though both H1
and u1 increase with , u1 increases faster than H1. The opposite
occurs for the light current; in such a case F2 decreases with
increasing .
The experimental values corresponding to closed channels
solid symbols fall near the energy-conserving solutions thin
solid lines obtained from Eq. 15 for the dense currents triangles, and from Eq. 23 for the light ones diamonds. In particular, the values of Froude numbers at the front of dense currents
fall close to the curve representing the energy-conserving relationship derived theoretically as expected. The high slope of this
curve implies that any variation of F1H1 should be noted mainly
in the front velocity.
For 1 the experimental values for dense currents are close
to the theoretical predictions but do not seem to follow them as
well as the light currents do. In these cases the lateral walls are
quite closer to each other as they reach the bottom of the channel
and the fluid finds it difficult to flow there. A complete description
of the flow up to the apex should include the effects of the walls,
but this problem deserves deeper study.
Conclusions
The values of the Froude number obtained at the front of gravity currents developed in uniform nonrectangular cross-section
channels are studied as a function of a simplified cross-section
shape. Mass and momentum balances in the frontal region of both
dense and light currents running on the bottom and along the
upper boundary, respectively, are considered. It is shown that the
usual relationships obtained for rectangular cross-sections have to
be modified in order to include the effects of the cross-section
shape on the flow. For example, for triangular cross-section
MAY 2009, Vol. 131 / 051201-5
shapes = 1, the theoretical values of the front velocity and

thickness corresponding to the dense current are 92% and 83%,
respectively, greater than those for the light current.
The analytical findings are in reasonable agreement with the
results obtained in laboratory experiments performed in closed
and open channels with four cross-section shapes. In open channels, we find faster light currents and slower dense currents than
those developed in closed channels. The Froude number of the
light currents in open channels is 1.07 0.02 times the value corresponding to the currents generated in closed channels. This
value is in agreement with 1.08, approximately, suggested by Hartel et al. 28 see their Fig. 4 for gravity currents developing
along slip and nonslip boundaries for = 0 in the range of experimental Reynolds numbers here reported. In addition, the Froude
number of the bottom currents results 0.98 0.01 times the value
corresponding to closed channels. These factors seem to be independent of the cross-section shape.
The Eqs. 11 and 19 do not depend on the scale of the depth
h0 or width w of the channel that is, the coefficient b in Eq. 2,
but only on the exponent associated with the cross-section contour shape. Thus, the same relationship may be held for wide and
shallow natural estuaries and narrow laboratory channels provided
that the same value of is used in both cases, which is a convenient advantage for the laboratory modeling of natural environments. It is also suggested that if the depth and/or width of a river
changes smoothly maintaining the value of , the dimensionless
Froude number of a gravity current should tend to be the same,
though the actual front velocity scales with the square root of the
corresponding height of the current h, see Eq. 1.
We study here a stratified environment in which the frontal
zones of dense and light fluid flows are distinguished. This happens for a wide range of natural situations where turbulence is not
dominant. Thus the solutions obtained represent an advance from
the theoretical and practical standpoints of the study of the gravity
currents in complex-shaped cross-section channels, which may
help to understand the features of the exchange flows in geophysical and engineering events. Laboratory experiments also give information about mixing in the interface between dense and light
fluids, which are not described here because they do not modify
the front movement appreciably. These aspects are under investigation and will be reported in a near future.
Acknowledgment
This work was supported by CONICET and ANPCyT Argentina under Grant No. PIP 5893 and Grant Nos. PICT 34088/05
and 1185/06, respectively.
Nomenclature
F
g
g
h
H
p
Re
s
u
x, y, z
w
, b
Froude number
acceleration due to gravity
reduced gravity
height of the current behind the head
dimensionless thickness of the current
hydrostatic pressure
Reynolds number
area of the channel cross-section occupied by
the denser fluid
frontal zone velocity
longitudinal, transversal, and vertical
coordinates
width of the laboratory channel
parameters associated with the shape of the
channel cross-section
relationship between light and dense fluid
densities
density
051201-6 / Vol. 131, MAY 2009
kinematic viscosity
Subscripts
0
1
2
AD, CD
c
O
initial value
dense fluid
light fluid
vertical planes limiting the control volume
energy-conserving current
stagnation point
References
1 Simpson, J. E., 1997, Gravity Currents: In the Environment and the Laboratory, Cambridge University, Cambridge, p. 258.
2 Imberger, J., and Hamblin, P. F., 1982, Dynamics of Lakes, Reservoirs and
Cooling Pounds, Annu. Rev. Fluid Mech., 14, pp. 153187.
3 Farrell, G., and Stefan, H., 1988, Mathematical Modeling of Plunging Reservoir Flows, J. Hydraul. Res., 26, pp. 525537.
4 Britter, R. E., and Linden, P. F., 1980, The Motion of the Front of a Gravity
Current Travelling Down an Incline, J. Fluid Mech., 99, pp. 531543.
5 Alavian, V., Jirka, G. H., Denton, R. A., Johnson, M. C., and Stefan, H. G.,
1992, Density Currents Entering Lakes and Reservoirs, J. Hydrol. Eng.,
118, pp. 14641489.
6 Akiyama, J., and Stefan, G., 1987, Onset of Underflow in Slightly Diverging
Channels, J. Hydrol. Eng., 113, pp. 825844.
7 Johnson, T. R., Farell, G. J., Ellis, C. R., and Stefan, H. G., 1987, Negatively
Buoyant Flow in a Diverging Channel. I: Flow Regimes, J. Hydrol. Eng.,
113, pp. 716730.
8 Johnson, T. R., Ellis, C. R., Farell, G. J., and Stefan, H. G., 1987, Negatively
Buoyant Flow in a Diverging Channel. II: 3-d Flow Field Descriptions, J.
Hydrol. Eng., 113, pp. 731742.
9 Johnson, T., Ellis, C., and Stefan, H., 1989, Negatively Buoyant Flow in
Diverging Channel. IV: Entrainment and Dilution, J. Hydrol. Eng., 115, pp.
437456.
10 Stefan, H., and Johnson, T., 1989, Negatively Buoyant Flow in Diverging
Channel. III: Onset of Underflow, J. Hydrol. Eng., 115, pp. 423436.
11 Bournet, P. E., Dartus, D., Tassin, B., and Vinon-Leite, B., 1999, Numerical
Investigation of Plunging Density Current, J. Hydrol. Eng., 125, pp. 584
594.
12 Monaghan, J. J., 2007, Gravity Current Interaction With Interfaces, Annu.
Rev. Fluid Mech., 39, pp. 245261.
13 Britter, R. E., and Simpson, J. E., 1978, Experiments on the Dynamics of a
Gravity Current, J. Fluid Mech., 88, pp. 223240.
14 Zhou, M., 1998, Influence of the Bottom Stress on the Two-Layer Flow
Induced by Gravity Currents in Estuaries, Estuarine Coastal Shelf Sci., 46,
pp. 811825.
15 Nunes, R. A., and Simpson, J. H., 1985, Axial Convergence in a Well Mixed
Estuary, Estuarine Coastal Shelf Sci., 20, pp. 637649.
16 Wong, K. C., 1994, On the Nature of Transverse Variability in a Coastal Plain
Estuary, J. Geophys. Res., 99, pp. 209222.
17 Wong, K. C., and Munchow, A., 1995, Buoyancy Forced Interaction Between
Estuary and Inner Shelf-Observation, Cont. Shelf Res., 15, pp. 5988.
18 Valle-Levinson, A., and Lwiza, K. M. M., 1995, The Effects of Channels and
Shoals on Exchange Between the Chesapeake Bay and the Adjacent Ocean, J.
Geophys. Res., 100, pp. 1855118563.
19 Engqvist, A., and Hogg, A., 2004, Unidirectional Stratified Flow Through a
Non-Rectangular Channel, J. Fluid Mech., 509, pp. 8392.
20 Thomas, L. P., and Marino, B. M., 2004, Lock-Exchange Flows in NonRectangular Cross-Section Channels, ASME J. Fluids Eng., 126, pp. 290
292.
21 Benjamin, T. B., 1968, Gravity Currents and Related Phenomena, J. Fluid
Mech., 31, pp. 209248.
22 von Karman, T., 1940, The Engineer Grapples With Non-Linear Problems,
Bull. Am. Math. Soc., 46, pp. 615683.
23 Lowe, R. J., Rottman, J. W., and Linden, P. F., 2002, A Laboratory Study of
the Velocity Structure in an Intrusive Gravity Current, J. Fluid Mech., 456,
pp. 3348.
24 Birman, V. K., Martin, J. E., and Meiburg, E., 2005, The Non-Boussinesq
Lock-Exchange Problem. Part 2. High Resolution Simulation, J. Fluid Mech.,
537, pp. 125144.
25 Marino, B. M., Thomas, L. P., and Linden, P. F., 2005, The Front Condition
for Gravity Currents, J. Fluid Mech., 536, pp. 4978.
26 Shin, J. O., Dalziel, S. B., and Linden, P. F., 2004, Gravity Currents Produced
by Lock Exchange, J. Fluid Mech., 521, pp. 134.
27 Simpson, J. E., and Britter, R. E., 1979, The Dynamics of the Head of a
Gravity Current Advancing Over a Horizontal Surface, J. Fluid Mech., 94,
pp. 477495.
28 Hartel, C., Meiburg, E., and Necker, F., 2000, Analysis and Direct Numerical
Simulation of the Flow at a Gravity-Current Head. Part 1. Flow Topology and
Front Speed for Slip and No-Slip Boundaries, J. Fluid Mech., 418, pp. 189
212.
Vortex Dynamics and Shedding of

a Low Aspect Ratio, Flat Wing at
Low Reynolds Numbers and High
Angles of Attack
Daniel R. Morse
James A. Liburdy
Mechanical Engineering,
Oregon State University,
Corvallis, OR 9733
This study focuses on the detection and characterization of vortices in low Reynolds
number separated flow over the elliptical leading edge of a low aspect ratio, flat plate
wing. Velocity fields were obtained using the time-resolved particle image velocimetry.
Experiments were performed on a wing with aspect ratio of 0.5 for velocities of 1.1 m/s,
2.0 m/s, and 5.0 m/s corresponding to chord length Reynolds numbers of 1.47 104,
2.67 104, and 6.67 104, respectively, and angles of attack of 14 deg, 16 deg, 18 deg,
and 20 deg. A local swirl calculation was used on proper orthogonal decomposition
filtered data for vortex identification and corresponding vortex centers were tracked to
determine convective velocities. The swirl function was also analyzed for its temporal
frequency response at several discrete points in both the shear layer and in the separated
recirculation region. A peak frequency was detected in the shear layer with a corresponding Strouhal number of approximately 3.4 based on the flow direction projected length
scale. The Strouhal number increases with both angle of attack and Reynolds number.
The shear layer convective length scale, based on the vortex convection velocity, is found
to be consistent with the mean separation distance between vortices within the shear
layer. This length scale decreases with increasing Rec. DOI: 10.1115/1.3112385
Introduction
Low and ultralow Reynolds number aerodynamics are of increasing interest in a number of application areas, such as microair vehicles, autonomous vehicles, as well as animal and insect
flight. Low Reynolds numbers based on the freestream velocity
and chord length are typically characterized as less than 106, while
ultralow is typically less than 103. Of concern in the design of
small aircraft is the ability to maintain acceptable lift characteristics over a wide range of angles of attack and to reduce the susceptibility for unstable operation. McCullough and Gault 1 discussed the generally accepted three main types of airfoil stall:
trailing edge, leading edge, and thin airfoil stall. Flow over thin
airfoils at high angles of attack displays attributes of stall where
by the flow separates at the leading edge with a re-attachment
point that moves downstream as the angle of attack increases.
McCullough and Gault likened this to the separation of flow past
a sharp edge because at sufficiently high angles of attack the stagnation point moves below the leading edge.
In this present study, a low aspect ratio, thin wing at high angles
of attack results in leading edge separation with a recirculation
bubble extending over a significant portion of the wing. The separation results in a strong shear layer that yields to a Kelvin
Helmholtz instability. Consequently, as the disturbances grow into
vortical structures along the shear layer, they are convected downstream. Figure 1 illustrates the flow characteristics for the current
study with time averaged velocity vectors and associated streamlines showing the extent of the separation bubble along the centerline for Rec = 1.47 104 and = 20 deg. For all cases studied,
flow reattachment occurs near x / c = 0.3. A flow visualization image of this using a smoke wire technique is shown in Fig. 2, where
flow is from left to right at a chord Reynolds number of approxiContributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received April 16, 2008; final manuscript received January 26, 2009; published online April 14, 2009. Assoc. Editor:
Juergen Kompenhans. Paper presented at the 2007 ASME International Mechanical
Engineering Congress IMECE2007, Seattle, WA, November 1016, 2007.
mately 1.47 104 and 20 deg angle of attack. The view is of

approximately 20% of the chord length from the leading edge.
The separated region contains larger, generally weaker swirl components, and at high angles of attack is more consistent with the
von Krmn instability of vortex shedding from bluff bodies.
Increased knowledge of the low Reynolds number flow events
leading up to and during separation will be valuable to the development of unmanned and autonomous aircraft. Moreover, detailed
understanding of the time dependent nature of separation conditions may lead to the ability for new control algorithms to improve overall performance. A better understanding of the leading
edge separation may lead to improved techniques for delaying or
counteracting separation effects under these conditions.
1.1 Low Reynolds Number Flow Over a Wing. Separation
of airflow over a wing has been a topic of interest for many
decades. Studies of high Reynolds flows have direct application to
traditional aircraft. In contrast, Carmichael 2 compiled a detailed
survey of low Reynolds number wing performance and identified
various flow regimes based on Reynolds number, as well as separation and characteristics of the separation bubble. In the regime
of Rec from 104 to 3 104, Carmichael noted that small gliders
operate in a completely laminar manner and that trip devices had
little benefit. At larger Reynolds numbers, from 3 104 to 7
104, wings may have significant induced drag penalties. This
regime received additional attention by Re et al. 3 because it
also coincided with aircraft operating in rarefied environments
such as extremely high altitude or the Martian atmosphere.
The investigation of low Reynolds number flows has increased
as a result of interest in autonomous and micro-air vehicle development. In general, these craft operate at Reynolds numbers less
than or equal to 105. Mueller and DeLaurier 4 reviewed the
performance of small air vehicles, which included the aerodynamics for fixed and flapping wings, with discussion of the Reynolds
number effects on the separation bubble. At very low Reynolds
numbers, less than 5 104, there is no re-attachment of the flow
for leading edge separation. They concluded that distinct differ-
MAY 2009, Vol. 131 / 051202-1
Fig. 1 Mean flow PIV data obtained at the wing centerline. Streamlines
illustrate the extent of the separated region. The light gray region represents an area of no data below the wing. An approximation of the wing
contour is shown with a dotted line.
ences in flight characteristics between high Reynolds number,

greater than 5 105, and low Reynolds number operations necessitate a different design approach for emerging micro-air vehicles,
defined based on a characteristic size on the order of 15 cm.
A study by Torres and Mueller 5 details the lift and drag
characteristics of 12 micro-air vehicle wings. All have low aspect
ratio designs, between 0.5 and 2.0 ratio of wing span to chord.
They found that the 0.5 aspect ratio rectangular wing has somewhat superior lift at high angles of attack for Rec = 7 104 1
105. Broeren and Bragg 6 discussed the advantage of thin
wings at low Reynolds numbers, but a consequence is unsteady
stall characteristics. Their results, for 12 thin wing designs and a
single Rec = 3 105, show fluctuations in lift near the stall condition varying on the order of 1220% depending on angle of attack,
. The associated Strouhal number, Sth, based on projected height,
h = c sin , increases almost linearly with angle of attack, a direct
result of increasing frequency, in addition to the increase in the
projected height. The spectra indicate two dominant values of Sth,
one in the range 0.0080.030 and the other 0.120.22 depending
on the wing design. They discussed how this behavior may be
related to separation bubble effects.
The effect of periodic flow can be significant on the body forces
in low Reynolds number regimes, giving rise to von Krmn vortex shedding. Much work has been done to analyze this phenomenon for flow over blunt bodies. Bishop and Hassan 7 and
Berger and Wille 8 described the phase synchronization of fluctuations for a cylinder in cross flow and its wake at low Reynolds
numbers. Mair and Maull 9 discussed the three dimensional effects of vortex formation on the body forces. The strongest effect
was seen when the vortices were large in the spanwise direction
along the length of the cylinder. This phenomenon was also seen
Fig. 2 Flow visualization of leading edge KelvinHelmholtz instability rollup forming spanwise vortices, Rec = 1.47 104,
= 20 deg.
051202-2 / Vol. 131, MAY 2009
for wings with blunt trailing edges. They proposed the introduction of a serrated trailing edge that produces streamwise vortices
to disrupted the formation of longitudinal vortex structures.
Yarusevych et al. 10 studied the effect of periodic excitation
of an airfoil as a method of flow control. They investigated matching the excitation frequency with the dominant frequency of separation to improve performance of an airfoil at Rec = 104 and low
angles of attack. Tinar and Cetiner 11 analyzed the self-induced
vibration frequencies of a spring supported airfoil at Reynolds
numbers between 5.9 104 and 14.8 104. Accelerometer data
were paired with particle image velocimetry PIV data to determine natural vibration frequencies and vorticity patterns during
the vibration cycle. Among others, Cheng and Chen 12 performed a numerical study evaluating flow control for the body
forces on a bluff shape and found that dominant frequencies of
vortex shedding corresponded to imposed body force frequencies.
Abernathy 13 investigated flow over an inclined flat plate and
studied the frequency of shed vortices. When analyzing pressure
in the wake of the flat plate, it was seen that the Strouhal number
was nearly constant at 0.17 for angles of attack greater than 35
deg, which corresponds to completely separated flow. This is consistent with the results of Roshko 14 who observed a low range
of Strouhal numbers 0.14 Sth 0.21 for various bluff bodies
and Reynolds numbers.
Burgmann et al. 15 examined the vortex structure of a low
Reynolds number airfoil in water using two-component, timeresolved PIV 2C-TR PIV and scanning PIV. They identified the
development of convex shaped vortex structures ejecting from the
separation bubble behind the leading edge. Derksen and Rimmer
16 employed a vortex cloud model in the simulation of flow
over a low Reynolds number airfoil. This model used free vortices
in the flow that convect over time. The results highlight the importance of such structures in separated flows when considering
body forces such as lift and drag.
1.2 Flow Structure Identification. During the past 2 decades
with the advance of digital PIV, as well as computational fluid
dynamics, large sets of discrete two dimensional velocity fields
have become more readily available for study. While vorticity is a
commonly, and often an appropriately, used measure for detecting
swirl, it is also sensitive to shear and the smallest fluctuations in
the data because it is resolved at the grid level. Several researchers
have set forth alternative methods for quantitatively identifying
coherent structures. Jeong and Hussain 17 identified criteria using the Hessian of the pressure field to identify a vortex core
based on a local pressure minimum. This approach neglects viscous and advective effects in the flow and is based on instantaneous velocity strain rate data. Adrian et al. 18 discussed large
eddy simulation LES decomposition on experimental data for

visualizing vortices by the removal of the larger structures. This
method employs the use of a low pass Gaussian filter on the flow
field, which is subtracted from the total velocity field leaving the
high frequency, small structures. Graftieaux et al. 19 used a
structure detection technique based on the local swirl velocity
component. This was done by integrating the circumferential velocity component about an area bounded by a closed path of selected diameter for each point in the flow.
Shinneeb et al. 20 recently performed an experimental investigation of structures in the far field of a jet at Rec = 2.25 104.
Traditional PIV was used to determine the velocity flow field. The
proper orthogonal decomposition POD was used to filter high
frequency flow features based on modes containing 40% of the
variance of the velocity field. The local circulation was determined using the concept of closed streamlines and was used to
show that the number of structures decreased along the flow direction. This same algorithm was used by Agrawal and Prasad
21. Troolin et al. 22 performed TR PIV analysis of a wing with
a Gurney flap. Up to 1000 PIV fields were sampled at 1 kHz, 2
kHz, and 4 kHz resulting in total sampling times of 1 s, 0.5 s, and
0.25 s. They show good comparison of spectral analyses of TR
PIV data and hot wire data. Additionally, Druault et al. 23 performed TR PIV measurements of an internal combustion engine at
276 Hz and 423 Hz over approximately 1.6 s, reported as data
covering 32 cycles at 20 counts/s. The TR PIV results were analyzed using POD using 100 modes and phase averaged to examine
cycle to cycle variation of flow in the cylinder. Recently, Weiland
and Vlachos 24 using vortex detection coupled with TR PIV
analyzed the vortical unsteady flow associated with leading edge
blowing control. They used the first several POD modes to distinguish frequency characteristics influenced by blowing. Kutulus et
al. 25 used TR PIV to study the unsteady forces acting on a
square cylinder in cross flow. Through the use of the instantaneous
NavierStokes equations, they calculated the transient lift and
drag forces from the 2D velocity vector field.
1.3 Objectives. The primary goal of this study is to characterize vortex shedding during leading edge separation from a flat
low aspect ratio wing using time-resolved vortex detection. Time
resolved PIV data are collected for a range of high angles of
attack 1420 deg for low values of Rec 1.47 6.67 104. The
POD and a tracking filter algorithm are used to assist in the identification and characterization of these structures. Furthermore, the
temporal frequency of vortex shedding and the characteristic
length scale of vortex separation are analyzed by examining the
time series of local swirl events.
Experimental Method
Tests were carried out in a large recirculating wind tunnel with

internal dimensions of 1.37 1.52 m2. A Laskin nozzle was used
to generate airborne particles in the tunnel using vegetable oil.
Particle size distribution was centered around 6 m, or on the
order of 1 pixel spacing. This corresponds to a particle response
time limit of 5 kHz using a Stokes flow drag model described by
Hinze 26. A 532 nm New Wave Pegasus Laser was used to
illuminate the flow from the downstream direction. A Dantec Dynamics 9080 0651 light sheet module was used to convert the
beam to a 1 mm wide light sheet. This light sheet was placed in
the spanwise center of the wing.
The wing was a flat plate of aspect ratio AR= 0.5 span to chord
ratio and chord length of 20.5 cm. The plate leading and trailing
edges were elliptical with a major to minor axis ratio of 5:1. The
side edges were semicircular. The thickness of the wing was 4
mm, which is approximately 2% of the chord length. Torres and
Mueller 5 studied several wing configurations for small unmanned aerial vehicle UAV applications. They noted that a flat
wing of aspect ratio equal to 0.5 at Rec = 7 105 had an increasing
lift coefficient with an angle of attack up to 40 deg. This high lift
Fig. 3 Downstream view looking upstream of a low aspect

ratio wing using smoke wire visualization for Rec = 1.47 104,
= 20 deg. The wing outline is highlighted in bold and the
separated flow region is outlined with a dashed line. Streamwise tip vortices are shown with smoke lines curving toward
the center and leading edge instabilities are observed to disrupt the smoke lines along the centerline increasing smoke
diffusion.
coefficient at high angles of attack is a feature that makes low

aspect ratio wings beneficial to small UAV applications.
The wing cross-sectional area as seen by the incoming flow is
defined as the product of the projected height, h, and width of the
wing, Ac = hw = c sinw. The total wind tunnel blockage, at the
maximum angle of attack of 20 deg, was 0.3%, and at the lowest
angle of attack of 14 deg was approximately 0.2%. In order to
obtain the desired field of view, the wing was located approximately 20 cm from the near side optical window of the wind
tunnel. This was well outside the wall boundary layer. This wing
was located approximately 6 m downstream from the contraction.
The wind tunnel has a freestream turbulence level of approximately 1% 27.
The wing was supported from the underside trailing edge using
a mounting plate that was 2.5 mm thick by 15 mm wide and 30
mm along the wing. This plate was connected to an aluminum
support rod approximately 70 cm long. The angle was fixed using
a four-bar linkage mechanism. The trailing edge of the wing was
held at a height of approximately 40 cm from the wind tunnel
floor. The total cross-stream width of the rod and linkage was less
than 2 cm. This system was rigid in the relatively low velocities
used in this study, 5.0 m / s. The wing surface was not observed
to oscillate throughout each of the 1000 image pair sequences
obtained in each data set.
Smoke wire flow visualization was performed to understand the
basic flow configuration around the wing. A nickel wire was
placed approximately 10 cm upstream of the leading edge. The
results for Rec = 1.47 104 and = 20 deg are shown in Figs. 2
and 3. The negative image shown in the figures reveals the presence of streamwise tip vortices, as well as the leading edge disturbances, which increased smoke line diffusion. The separated
region on the wing shown in Fig. 3 occupied a roughly triangular
shaped region extending from the leading edge and tapering
downstream to a point at approximately 60 mm, or 0.3c, from the
leading edge along the wing centerline. This re-attachment point
was also confirmed using the mean flow PIV data obtained at the
centerline, as shown in Fig. 1. Typically, thin wings of large aspect ratio develop a leading edge separation bubble just prior to
stall. In the case of this low aspect ratio wing, the streamwise tip
vortices act to reduce the spanwise separation bubble size by increasing the downward flow of air on the suction side of the wing,
and thereby delay stall. Consequently, the dynamic characteristics
of the separated region can be expected to be impacted by the tip
vortices in a low aspect ratio wing.
MAY 2009, Vol. 131 / 051202-3
Hot wire anemometry was used to confirm the TR PIV frequency spectra for a single angle of attack, = 20 deg, at each
Reynolds number. A TSI IFA-300 constant temperature anemometer was used in conjunction with a single axis, TSI 1201-20
model probe. The probe was mounted in the wind tunnel, supported from the downstream direction, and placed in the separation bubble at the centerline approximately 0.25c downstream
from the leading edge. The probe was also placed in the shear
region approximately 0.05c downstream and 0.01c above the
leading edge. Data were obtained using a personal computer PC
with analog capture board and LABVIEW 8.2.
Three freestream velocities were used for this study: 1.1 m/s,
2.0 m/s, and 5.0 m/s, which correspond to chord Reynolds numbers of 1.47 104, 2.67 104, and 6.67 104. Four angles of attack were investigated at each velocity: 14 deg, 16 deg, 18 deg,
and 20 deg, for a total of 12 experimental cases. For each case
images were obtained using an iNanosense high speed digital
camera equipped with an image intensifier. The pixel resolution of
the charge coupled device CCD was 1280 1024 and the field
of view was approximately 55 42 mm2.
Synchronization with the laser pulse was obtained using Dantec
Dynamics FLOWMANAGER software. The time delay between laser
pulses for these data was 50 s for the Rec = 6.67 104 cases and
100 s for the Rec = 1.47 104 and 2.67 104 cases resulting is
an average particle displacement of approximately 6 pixels. The
time-resolved sampling of velocity fields was performed at 500
Hz. The total sample time was 2.0 s due to hardware memory
limitations. This results in a 0.5 Hz frequency resolution when
determining the velocity spectra. It is understood that this may not
be appropriate for analysis of very low frequency events but it
provides adequate results for the higher frequency vortex shedding observed in this study. To help confirm spectral the results,
the TR PIV data were also collected using a 10 s sampling period
for = 20 deg and Rec = 1.32 104, with a sampling rate of 100
Hz. The lower Rec condition was chosen to compare with since it
is the one susceptible to high uncertainty with its lower frequency.
One angle of attack was compared as it has been noted by Abernathy 13 that angle of attack has small influence on normalized
separated flow fluctuations. Overall, the mean velocity vector values and the local standard deviations were all within 2% and 5%,
respectively. Hot wire anemometry, mentioned previously, was
also used to compare low and high frequency events observed in
the TR PIV data and are shown with the results. It should be noted
that the goal of this study is to determine direct spectral characteristics of detected vortical structures rather than velocity components. This data analysis is discussed later.
Postprocessing of the images was done to reduce glare from the
wing surface. This was done by calculating the mean pixel intensity and removing this mean value from each image. The subregion size for each PIV cross-correlation calculation was 32
32 pixels, with an average of 68 particles per subregion. A
50% overlap was used resulting in a vector field of 79 63 grid
points with a spacing of approximately 0.6 mm. The results were
obtained using FLOWMANAGER software from Dantec Dynamic
Denmark using fast Fourier transform FFT cross correlation.
Numerical experiments were carried out with the same seeding
density and particle size as used in the experiments and an adaptive cross-correlation scheme was not found to improve the results
of the swirl detection describe later in a highly curved flow regions 27. The velocity data were filtered using a 3 3 median
filter since large scale flow structures are of interest. No vectors
were removed from the data set. The PIV velocity uncertainties
were investigated in the same wind tunnel, with the same field of
view, while using the same particle generator system and seeding
levels 27. For the same pixel resolution, approximately 6 m
particle diameter, and a 32 32 pixel subregion as used in this
study, the uncertainties were found to be approximately 1% and
velocity gradient bias is small by virtue of estimated particle displacement variation over the subregion domain.
051202-4 / Vol. 131, MAY 2009
Data Analysis
The vortex detection method used in this study was introduced

by Graftieaux et al. 19 and used by Dano and Liburdy 27 for
vortex detection in a pulsed jet in cross flow. This technique is
based on the orientation of the velocity vectors relative to radial
lines from a local point. The local swirl function, , is calculated
for a given subregion, A M , centered about point P. This vortex
detection function is defined as
1
AM
PM U M Z
1
dA =
PM U M
AM
sin M dA
where PM is the vector from the center point P to subregion point

M, U M is the velocity at point M within the subregion, Z is the
unit vector perpendicular to the plane of data, and M is the angle
between position vector PM and velocity vector U M . This locally
normalized function allows the detection of vortical structures in
both the shear layer and the slower recirculating region in the
wake of the wing leading edge.
It should be noted that in using this technique the local swirl is
not equivalent to an average of spanwise vorticity within the domain specified by A M . That is, the function includes all velocity vectors within the domain not just the line integral around the
domain boundary. Consequently, it weights the vorticity near the
central point, P.
The value of varies between 1 and 1 where 0 represents a clockwise vortex and 0 represents a counterclockwise
vortex. This technique allows the user to provide a prescription of
the mask size, A M , used to evaluate the local value of swirl. In this
study the domain was set to a 9 9 grid centered about each data
point, which corresponds to approximately 4.0 mm2 region. This
is equivalent to 0.02c and is equal to the wing thickness. A further
discussion of the effects of the subregion size and the vortex
strength based on velocity magnitude can be found in Ref. 28.
Additional comparison of the function with other structure detection techniques is given by Chen et al. 29 such as the 2
criteria proposed by Jeong and Hussein 17, as well as various
vector field topological structures. It should be noted that is
sensitive to swirl features smaller than the subdomain, Am. Structures that are larger than Am will yield values large at the swirl
center but will decay as the subregion center moves away from
the swirl center. In addition, is sensitive to opposing shear flow
within a subregion. A pure shear flow with opposing flow centered
around zero in the subregion results in = 0.65, whereas pure
unidirectional shear flow produces = 0. A threshold value of
can be set to help filter out opposing flow shear gradients. In the
current study, this type of flow occurs within the separation. A
threshold of 0.7 was used to filter out shear flow while capturing
the strong swirl elements. Additional filters were applied to aid in
vortex tracking and are discussed later.
Since a time series of vector fields are available using the TR
PIV method, individual vortical structures are tracked over time.
In doing this, the local convective velocity of the vortex is approximated as the change in position of the vortex center over a
given time interval: U = x / t and V = y / t, which represent
the x and y convection velocity components, respectively. Here
the vortex center was calculated at each time step based on the
area geometric center of an enclosed structure. The structure is
defined as a contour of at the given threshold value.
The POD was applied to the velocity field to isolate the large
scale energetic flow structures. A brief overview for the sake of
notation is given here. A more detailed discussion of the concept
and its applications include Refs. 3032. The velocity data,
fx , y , t, can be approximated as follows:
fx,y,t
d x,ya t
k
where dkx , y is the relative spatial distribution of velocity for

each mode k and akt is the amplitude time series of that mode.
The modes, beginning at mode 1, are organized from the mode
with the most variance energy, to the least variance, highest
mode. Mode 0 is simply the time averaged velocity field. Typically, as the mode energy decreases, the frequency of fluctuation
increases for its associated amplitude time series, at, and the
spatial size of the features seen in the mode map, dx , y, decreases. Therefore, used as a filter, a low order reconstruction will
capture much of the energy and behavior of the larger, lower
frequency features of the flow and eliminate the smaller, high
frequency variations.
The procedure used here is the same as was applied to TR PIV
data by Morse and Liburdy 33 for separated flows. Determination of the POD modes was carried out in MATLAB. The first step
was removing the time averaged flow. The remaining fluctuating
values of u and v were concatenated into a single design matrix D,
of M locations by N time steps. Using the svd function in MATLAB, a singular value decomposition was performed on the design
matrix. The resultant set of matrices denoted UNxM , S MxM , and
V MxM can be combined as USVT to recreate the design matrix.
In this case the U matrix is the set of modal spatial velocity
distribution maps. The amplitude time series matrix, A, is equal to
VS. Each column of U represents a mode map and each column
of A represents the amplitude time series for the mode map in U.
A truncation to the decomposition was done for efficiency using
the number of modes equal to the limiting dimension, in this case
N = 1000, the number of time steps. The results show that 90% of
the energy is contained in less than 600 modes so this truncation
had negligible effect on modal distribution.
In order to more clearly identify vortical structures, the higher
frequency velocity fluctuations were filtered using the POD of the
velocity data. This was done by reconstructing the flow using the
first 30 modes which accounts for 3055% of the total variance in
the flow. This reconstruction represents the high energy, lower
frequency features of the flow while eliminating the low energy,
high frequency fluctuations.
To better determine of the convective velocity of individual
flow field structures, a tracking filter method was used. A minimum lifespan criterion l = t / tc where t is the detected lifetime of
the vortex and tc = c / U is the convective timescale using the
freestream velocity was used to filter out vortices, which may
appear in only one or two frames but were not observed over a
sufficiently long time. This method determines vortex paths by
seeking out the nearest vortex location in the subsequent time
steps. If there exists no vortex within this specified distance in the
next time step, then that vortex is no longer tracked. The specified
threshold distance was determined by multiplying 150% of the
freestream velocity by the time period between frames, xlimit
= 1.5 U t. This corresponds to a length threshold, lthreshold
= 0.017c, 0.030c, and 0.075c for Rec = 1.47 104, 2.67 104, and
6.67 104, respectively. Similarly, identified vortex locations
without an associated previously identified vortex were marked as
origin points for further tracking in time. In essence, the tracking
filter requires a vortex to exist for a given number of frames and
places an upper limit on the vortex convection velocity based on
the magnitude of the freestream velocity.
Spectral analysis was performed on individual velocity components and the function with the aim of detecting the presence of
any dominant frequencies. Because is a locally normalized
swirl function, its fluctuations are equally detectable in the high
speed shear region and the low speed recirculation region. The TR
PIV data field provides time series data and frequency response at
each location within the grid. The root mean square of fluctuations
of , denoted , indicates locations of the most intense fluctuations of over time.
An alternative to examining the fluctuation of is to look for

cyclic behavior in . One approach is to examine the time based
autocorrelation of , denoted as . A periodic autocorrelation
function represents a cyclic event, and the root mean square of the
autocorrelation of , , at each TR PIV location is used to
provide an indication of where has the most intense cyclic
fluctuations.
A vortex separation length scale was determined and used to
assure consistency between vortex detection frequency data and
vortex spatial separation data. Two lengths scales are defined as i
the average separation distance between the identified vortex centers in all time frames for the entire time series, ltrack note that this
does not require time-resolved data, and ii the length defined by
dividing the convection velocity of vortex centers by the characteristic frequency, lfreq. These two results would be expected to be
reasonably close if the measured convection velocity and frequency are consistent.
In summary, two primary analyses are presented i vortex detection with tracking and ii frequency analysis of the local nondimensional swirl function, . The velocity field was filtered using the POD to remove high frequency variations prior to
calculation of . The function was used to determine areas of
swirl and a threshold was set to identify discrete regions of high
swirl. The centers of these swirl regions were tracked over time to
determine the convective velocity of these structures. A vortex
tracking filter based on vortex lifespan was applied to reduce the
noise in the vortex velocity signal. Frequency analysis of the
function was carried out. A vortex separation length scale in the
shear layer is evaluated using two methods to check the consistency of the measurements.
Results
4.1 Vortex Detection and Tracking. The data consist of the

planar TR PIV results at three chord Reynolds numbers and four
angles of attack. Figures 4a4d illustrate instantaneous velocity vector data at four angles of attack for Rec = 1.47 104 with the
x and y coordinates, originating at the leading edge, are normalized by the chord length, c. The field of view is of the leading
edge comprising approximately 25% of the chord length; this
view is slightly smaller than that shown in Fig. 1. The region
below the wing surface was not illuminated by the laser and contains no velocity data. The vector plots clearly identify the regions
of separated flow and the corresponding large recirculation region.
The mean vector fields are similar for the other Rec values, with
separation occurring at the leading edge and approximately the
same reattachment location.
Figures 4e4h show instantaneous contours of at the
times and conditions corresponding to the vector plots in Figs.
4a4d, respectively. The regions of highest clockwise local
swirl 0 are located in the high shear region beginning near
the leading edge. The vortices in the shear region form very near
the leading edge and extend along a line, which projects further
from the wing surface as increases. That is, the angle of the line
of KelvinHelmholtz instability beginning at the separation location forms an angle greater than with the wing surface as
increases. Close to the surface the shear flow results in a relatively
weak layer of due to the near surface velocity gradient, not
swirl. A threshold applied to was used to filter out weak shear
flows during vortex tracking. The separated recirculation region
beneath the shear layer does not contain strong swirl components,
although the flow is seen to clearly recirculate from the vector
plots shown in Figs. 4a4d.
Figure 5a5e illustrate a time-resolved sequence of TR PIV
vectors and contours sampled at every time step, or 2 ms time
separation, for the case of = 14 deg and Rec = 1.47 104. The
large vortical structures can be seen to convect downstream. Since
their motion is relatively fast and are shown later to be dependent
on Rec, the resolution of the local convective velocity is limited
MAY 2009, Vol. 131 / 051202-5
Fig. 4 TR PIV vector field for Rec = 1.47 104 at four angles of attack: a 14
deg, b 16 deg, c 18 deg, and d 20 deg; grayscale plots of are shown
in eh, which were calculated from the velocity fields in ad.
by the time resolution of the TR PIV, in this case 500 Hz sampling

rate. To illustrate the time evolution of contours, Figs. 5f5j
illustrate a time sequence for the same conditions given in Figs.
5a5e. In the identification of the strongest vortical structures,
a threshold of 0.7 was applied, which eliminates the near
surface results. The magnitude of the large swirl elements does
not change significantly during tracking, which is discussed later.
The position denoted above the contours in Figs. 5f5j is the
downstream position of the centers of the vortices as defined by
the geometric mean position of the contour. The structures in the
shear layer convect downstream in a systematic manner in most
cases, although they occasionally are observed to stall or temporarily reverse direction, as is shown in the final two frames of this
sequence. This unsteady motion of detected vortices occurs regularly and is illustrated further in the vortex tracking results.
The POD of the velocity field was used to filter out high frequency fluctuations in the flow. This was desirable in order to
more clearly track the vortical structures over time. Figure 6
shows the cumulative modal energy distribution for all Rec and
values. In general, the amount of energy in the lower order modes
increases with increasing angle of attack for = 14, 16, and 18
deg. However, for = 20 deg the energy drops for the Rec = 1.47
104 case. This can be seen more clearly in the inset, which
shows only the first 30 modes. The velocity field at higher angles
051202-6 / Vol. 131, MAY 2009
of attack contained more energy, or variance, in fewer modes

when compared with the lower angle of attack cases. The higher
velocity cases show a greater disparity in lower mode energy content for all angles of attack. For example, in Fig. 6a, the energy
contained in the first 100 modes varies about 7% over all angles of
attack. In contrast, in Fig. 6b, the energy contained in the first
100 modes varies as much as 13% depending on the angle of
attack. Additionally, the energy contained in lower modes for the
highest angle of attack, = 20 deg, does not decrease significantly
with increasing Reynolds number. Therefore, increasing Reynolds
number is seen to cause the velocity field to be less ordered for the
lower angles of attack while not showing a significant effect for
the = 20 deg case.
Figures 7a7c show the unfiltered set of vortex locations for
the first 200 ms of a time sequence of 2000 ms for each Reynolds
number at = 14 deg. The progression of detected vortex structures in the downstream direction can be more clearly seen in
Figs. 7a and 7b as the local slope of the line connecting these
positions over time is used to represent the vortex convective
velocity. The highest Rec data shown in Fig. 7c is much more
difficult to use to determine a convection velocity. A vortex tracking filter coupled with the POD reconstruction of the velocity field
was used to more clearly identify the convection of structures
downstream. The results of applying these methods for tracking
for all cases. The dependence on Reynolds number is significant

with a decrease of approximately 50% from Rec = 1.47 104 to
6.67 104. Averaging the convective velocities for all angles of
attack, the freestream normalized convective velocity is shown to
decrease with increasing Reynolds number from 0.43 to 0.22.
These three convective velocities vary approximately as Re0.45
c
although more data are needed to verify this relationship.
The standard deviation of the convective velocity is on the order of the convective velocity and shows an increase for the Rec
= 6.67 104 case. This high standard deviation is in part due to
jitter or stall and reversal of flow of the detected vortices illustrated in Fig. 5. Figure 7f shows the furthest downstream vortex
between t = 50 and 100 ms moving downstream and upstream in
alternating fashion for approximately 40 ms. More variation in
movement is seen in the highest Rec case in Fig. 7f, which is
consistent with a higher standard deviation of U. This is seen in
Table 1 as a decrease in U / U with increasing Rec while / U
increases.
Fig. 5 Time sequence of ae = 0.70 contours and TR PIV

vectors for Rec = 1.47 104 and = 14 deg from t = 64 ms to t
= 72 ms; fj corresponding isolated contours of shown
over a larger field of view with vortex center x-position plotted
above each contour.
are shown in Figs. 7d7f for the = 14 deg cases. The origin
points for tracking are denoted with a shaded star and path lines
are traced through the vortex centers. These results use a POD
filter based on the first 30 modes and a tracking filter time scale,
l = 0.05 or 5% of the convective time scale based on U. By
removing smaller scale fluctuations, the POD filter allows for
more consistent detection and tracking of vortex positions over
time as seen by comparing the vortex position markers circles
between Figs. 7a7c and Figs. 7d7f. This is especially
useful for the highest Rec case.
The centers of the detected swirl regions were determined at all
time steps for each time series. These time series were 2 s long
resulting in 1000 data sets for each flow condition. The centers are
defined based on the geometric mean of contours of = 0.7
similar to those given in Figs. 5f5j. The distance traveled per
sample time was used to calculate the x directional velocity component of these structures. The velocity vector is not strictly only
along the x direction but also the x component is chosen to determine the downstream convection rate for vortical structures generated from leading edge separation. Table 1 lists the calculated
convective velocity, U, normalized by the freestream velocity,
U, for all cases. The standard deviation of U, , normalized by
the freestream velocity is also listed for all cases. Note that no
filtering of the calculated velocities was applied. The convective
velocities show weak variation with angle of attack, less than 5%
4.2 Spectral Analysis. The time dependent nature of the detected vortices was studied using the transient characteristics of
at all grid location with no threshold applied. The strategy for
evaluating the transient nature of was to examine the regions
where the rms of , , are high as well as the regions where
the rms of the autocorrelation of , , are high. An example
of the nature of the temporal variations of is shown in Fig. 8.
The data shown are taken from a point within the vortex shedding
shear layer at x / c = 0.1, y / c = 0.025. Note that the time span is only
shown for the first 200 ms out of a total of 2000 ms. The autocorrelation of indicates a highly cyclic process with a strong
rms value of the autocorrelation. The associated spectra of
indicates a peak at a Strouhal number, Sth of approximately 2.4.
The distribution of for Rec = 1.47 104, = 20 deg is
shown in Fig. 9a and indicates that fluctuations of are maximum in the shear region. A white line has been added to show the
region of maxima of . Figure 9b shows the distribution of
, which are interpreted as regions of high cyclic characteristics of Two regions of high are identified by enclosed
dashed lines, one just above the region of high rms values of
and the other in the recirculation region. Interestingly, the region
of high rms values of has a relative low cyclic characteristic. In
addition, the high cyclic characteristics within the recirculation
region do not correspond to large amplitude fluctuations since Fig.
9a shows relatively low rms values in the recirculation region.
To help determine the frequency of the cyclic processes within
the shear layer, several experimental results were used. Representative spectra are shown in Fig. 10 for Rec = 1.47 104 and
= 20 deg. These plots are the results averaged over four points
located in the high amplitude cyclic region denoted by the dashed
line above the shear layer shown in Fig. 9b. Using the TR PIV
data at 500 Hz, the average spectrum of the vertical velocity fluctuations, v, is shown in Fig. 10a and the horizontal velocity
fluctuation, u, in Fig. 10b. Both spectra show a peak near Sth
= 3.5, with the v spectrum peak being more dominant. The average spectral for the fluctuations, , is shown in Fig. 10c
revealing a similar peak near Sth = 3.5. These spectra show a broad
low frequency peak in the range of Sth = 0.1 0.3. For all cases the
resolution limit based on the length of the sampling period is
Sth 0.006 0.03, depending on the Reynolds number and angle
of attack. For comparison, hot wire data taken at approximately
the same location in the shear layer were obtained at 10 kHz over
10 s and the resulting spectrum is shown in Fig. 10d, showing a
similar high frequency peak.
The average spectra using six points in the recirculation region
identified in Fig. 9b were obtained using the same methods
stated above for Fig. 10. The average spectra for v and are
shown in Figs. 11a and 11b using the TR PIV data at 500 Hz
for Rec = 1.47 104, = 20 deg. To better illustrate the results, a
MAY 2009, Vol. 131 / 051202-7
Fig. 6 The POD distribution of energy across modes and angles of attack for a Rec = 1.47 104, b Rec = 2.67 104, and c
Rec = 6.67 104
linear plot is used. The results are shown only up to Sth = 5.0 since
no peaks at higher frequencies were found. Both spectra show
broad maxima near Sth = 0.2. Comparing Figs. 10 and 11 indicates
that the higher frequency peaks in the shear layer occur at approximately an order of magnitude higher value of Sth than any
peaks in the recirculation region. Comparative tests using the hot
wire results and longer time period TR PIV were carried out for a
limited number of cases. Hot wire data obtained at 10 kHz over 10
s are presented in Fig. 11c and TR PIV data sampled at 100 Hz,
over 10 s, are shown in Fig. 11d. These spectra are similar to
those found for the 500 Hz TR PIV data for both v and with
peaks in the same range of Sth.
For all cases of Rec and , the average spectra of were
determined using the four points in the high shear region, which
correspond with regions of high values of . In addition, the
average spectra of using five points in the recirculation region
were also calculated. To illustrate the results, the spectra for
= 20 deg and Rec = 1.47 104 and 2.67 104 are shown in Fig. 12
using the TR PIV data at 500 Hz. Similar to the results shown in
Fig. 10 for the high shear region, Figs. 12a and 12b show the
spectra with peaks occurring at the high frequencies, with broad
low frequency maxima. The average spectra in the recirculation
region are shown in Figs. 12c and 12d for = 20 deg, Rec
= 1.47 104 and 2.67 104. A broad low frequency maximum occurs near Sth = 0.2 for both values of Rec, which is similar to the
051202-8 / Vol. 131, MAY 2009
lower frequency maxima found in the high shear region. Based on

these results, it can be concluded that there is no pronounced
frequency response in the recirculation region comparable to that
observed for the leading edge separation, equivalent to a Strouhal
number on the order of 3.5. However, the shear layer shedding
frequency spectra does include a lower frequency component in
the same range as that found in the recirculation region.
The results of the identification of dominant high frequency
peaks for the low and middle Rec cases were compiled and the
results versus the angle of attack are shown in Fig. 13. Due to the
TR PIV sampling rate limitation of 500 Hz, the maximum measurable Sth for the Rec = 6.67 104 case is 3.4, whereas the limits
for the lower two Rec cases are 15.5 and 8.6. Consequently, the
higher Rec shear region results are too high to measure accurately
and are not included. Shown in Fig. 13 is a nearly linear increase
in Sth with increasing for both values of Rec, with a shift toward
higher Sth for the higher Rec case. Further data would be required
to substantiate this trend over a larger range for both and Rec.
The previously identified average length scales, ltrack and lfreq,
were calculated for the two lower Rec cases for which shear region fluctuation signals were measured, Rec = 1.47 104 and
2.67 104. The results are given in Table 2 normalized by the
chord length. The length scale, ltrack, was based on POD filtered
reconstructed instantaneous snapshots of TR PIV data. The length
scale, lfreq, was determined using the measured vortex convection
Fig. 7 Detected vortex positions shown for the first 200 ms of a total of
2000 ms with and without filtering. Unfiltered detected vortex positions for
= 0.70 at =14 deg and a Rec = 1.47 104, b Rec = 2.67 104, and c
Rec = 6.67 104. POD reconstructed detected vortex positions filtered for
path lifetime greater than 0.05c for d Rec = 1.47 104, e Rec = 2.67 104,
and f Rec = 6.67 104 at = 14 deg.
velocity and frequency data. The two length scales are shown to
be within 10% for all values for Rec = 1.47 104 and 4% for
Rec = 2.67 104, with no observable trend with angle of attack for
the range studied. Based on the variability of U and the resolution of the peak frequencies, the results for both length scales
show very good agreement. The length scale is shown to decrease
with increasing Rec, which indicates a relative increase in rate of
vortex generation within the shear layer as Rec increases.
Table 1 Vortex convective velocity and its standard deviation

for all Rec and
/ U
U / U
deg
Rec
14,700
26,700
66,700
14,700
26,700
66,700
4
16
18
0
0.44
0.41
0.42
0.45
0.36
0.38
0.35
0.33
0.23
0.22
0.21
0.20
0.41
0.40
0.37
0.42
0.40
0.42
0.41
0.38
0.46
0.49
0.49
0.50
Avg.
0.43
0.36
0.22
0.40
0.40
0.49
Fig. 8 a Time trace of at x / c = 0.1 and y / c = 0.025 for the

first 200 of 2000 total milliseconds of data for the = 14 deg,
Rec = 1.47 104 case. b Corresponding autocorrelation of
shown for time lag values up to 100 ms. c Corresponding
spectrum of shown with observed peak near Sth = 2.4.
MAY 2009, Vol. 131 / 051202-9
Fig. 9 a rms value of , , for the Rec = 1.47 104, = 20 deg case.
The solid white line identifies the region of highest amplitude. b rms value
of the temporal autocorrelation of , . The two areas outlined with
dashed lines represent regions of large amplitude cyclic activity of .
Conclusions
This study presents the results of the detection of strong vortical

structures evolving from the leading edge of a flat wing for a
range of low Reynolds numbers and angles of attack from 14 deg
to 20 deg. These structures were identified based on a local mea-
sure of swirl. The region of interest coincides with the region of

KelvinHelmholtz type instability extending from the separation
point to approximately one-quarter of the chord length. The time
resolved PIV data are used to generate velocity vector maps and
these data were used to detect spanwise vortices sampled at 500
Fig. 10 Average spectral results using four points in the high cyclic shear
region shown in Fig. 9b for Rec = 1.47 104, = 20 deg. TR PIV data at 500
Hz over 2 s: a averaged spectrum of v, b averaged spectrum of u, c
averaged spectrum of , and d hot wire data spectrum for a single point in
the shear region using data at 10 kHz for 10 s.
051202-10 / Vol. 131, MAY 2009
Fig. 11 Average spectra of selected points in the recirculation region for

Rec = 1.47 104 and = 20 deg. a TR PIV data of v at 500 Hz for 2 s, b TR
PIV data of at 500 Hz for 2 s, c hot wire spectrum of v at 10 kHZ for 10
s, and d TR PIV data of at 100 Hz for 10 s.
Fig. 12 Average spectra of for = 20 deg using TR PIV data at 500 Hz

for 2 s of data obtained from a the high cyclic shear region shown in Fig.
10b for Rec = 1.47 104, b the high cyclic shear region shown in Fig. 10b
for Rec = 2.67 104, c the recirculation region shown in Fig. 10b for Rec
= 1.47 104, and d the recirculation region shown in Fig. 10b for Rec
= 2.67 104
Hz. Lower modes of the POD of the velocity field were used to
reconstruct a low-pass filtered velocity field, which assisted in
tracking vortices. A vortex convection velocity was determined
for three relatively low Reynolds numbers and was not seen to
Table 2 Length scales based on i the average separation distance between tracked vortices, ltrack, and ii the frequency and
convection velocity of detected vortices, lfreq
Rec
Fig. 13 Average Strouhal number versus angle of attack for

the high frequency component determined in the shear region
deg
ltrack / c
lfreq / c
ltrack / c
lfreq / c
14
16
18
20
0.034
0.036
0.038
0.032
0.040
0.038
0.035
0.041
0.025
0.028
0.032
0.028
0.030
0.028
0.028
0.023
Avg.
0.035
0.038
0.028
0.027
14,700
26,700
MAY 2009, Vol. 131 / 051202-11
change over angles of attack between 14 deg and 20 deg. The

vortex convection velocity, U, decreases with increasing Rec,
although more data are needed to accurately describe this trend.
The rms values of the autocorrelation of , , were combined with spectral analysis of to determine a region of time
correlated events in the shear region, in which high frequency
events on the order of Sth = 3.4 are observed. These leading edge
vortices are shed at increasing rates for increasing Rec and . The
vortex separation length scale using the instantaneous TR PIV
data compares well with the length scale calculated from the
dominant frequencies and vortex convection velocities observed
in the shear region. Further detailed studies are needed to define
the Rec effects on dominant shear layer vortex frequencies and
convection velocities.
Acknowledgment
Funding for this work was partially provided by AFOSR under
Grant No. FA-9550-05-1-0041 and is gratefully acknowledged.
Nomenclature
A
AM
c
D
h
lfreq
ltrack
M
N
PM
Rec
S
St
Sth
U
UM
U
U *
w
V
Z
Greek Notation

M

tc
l
amplitude time series matrix

area of swirl calculation subregion
wing chord length
M N matrix of velocity data
approach height of wing section, =c sin
vortex separation length scale defined by the
frequency of events and vortex convection velocity, U
vortex separation length scale observed in PIV
frames containing multiple vortices
number of TR PIV grid locations
number of TR PIV time steps
vector from subregion center
Reynolds number based on chord length,
=U c /
singular values of the design matrix
Strouhal number based on chord length,
=fc / U
Strouhal number based on wing approach
height, =fh / U
left singular vectors of design matrix
velocity vector at point M
freestream velocity
convective velocity of vortex structure,
x-direction
wing span
right singular vectors of design matrix
perpendicular unit vector 0, 0, 1
angle of attack
temporal autocorrelation of
root mean square of
root mean square of
angle between PM and U M
swirl detection parameter
convective time scale, =c / U
tracking filter time scale, =t / tc
References
1 McCullough, G. B., and Gault, D. E., 1951, Examples of Three Representative Types of Wing-Section Stall at Low Speeds, NACA Technical Note 2502.
2 Carmichael, B. H., 1981, Low Reynolds Number Wing Survey, Volume 1,
NASA Contractor Report No. CR-165803.
051202-12 / Vol. 131, MAY 2009
3 Re, R. J., Pendergraft, O. C., Jr., and Campbell, R. L., 2006, Low Reynolds
Number Aerodynamic Characteristics of Several Airplane Configurations Designed to Fly in the Mars Atmosphere at Subsonic Speeds, NASA Report No.
214312.
4 Mueller, T. J., and DeLaurier, J. D., 2003, Aerodynamics of Small Vehicles,
Annu. Rev. Fluid Mech., 35, pp. 89111.
5 Torres, G. E., and Mueller, T. J., 2001, Aerodynamic Characteristics of Low
Aspect Ratio Wings at Low Reynolds Numbers, Prog. Astronaut. Aeronaut.,
195, pp. 115142.
6 Broeren, A. P., and Bragg, M. B., 2001, Unsteady Stalling Characteristics of
Thin Wings at Low Reynolds Number, Prog. Astronaut. Aeronaut., 195, pp.
191213.
7 Bishop, R. E. D., and Hassan, A. Y., 1964, The Lift and Drag Forces on a
Circular Cylinder in a Flowing Fluid, Proc. R. Soc. London, Ser. A, 277, pp.
3250.
8 Berger, E., and Wille, R., 1972, Periodic Flow Phenomena, Annu. Rev. Fluid
Mech., 4, pp. 31340.
9 Mair, W. A., and Maull, D. J., 1971, Bluff Bodies and Vortex SheddingA
Report on Euromech 17, J. Fluid Mech., 452, pp. 209224.
10 Yarusevych, S., Sullivan, P. E., and Kawall, J. G., 2005, Wing Boundary
Layer Separation and Control at Low Reynolds Numbers, Exp. Fluids, 38,
pp. 545547.
11 Tinar, E., and Cetiner, O., 2006, Acceleration Data Correlated With PIV
Images for Self-Induced Vibrations of a Wing, Exp. Fluids, 41, pp. 201212.
12 Cheng, M., and Chen, B. K., 2007, A Numerical Study on Fluid Force Reduction of a Square Cylinder by Flow Control, Fluids Engineering Conference, San Diego, CA, Jul. 30Aug. 2, Paper No. FEDSM2007-37025.
13 Roshko, A, 1954, On the Drag and Shedding Frequency of Two Dimensional
Bluff Bodies, NACA Technical Note 3169.
14 Abernathy, F. H., 1962, Flow Over an Inclined Plate, ASME J. Basic Eng.,
84, pp. 380388.
15 Burgmann, S., Brcker, C., and Schrder, W., 2006, Scanning PIV Measurements of a Laminar Separation Bubble, Exp. Fluids, 41, pp. 319326.
16 Derksen, R. W., and Rimmer, J., 2006, Aerodynamic Flow Simulation, Int.
Conference on Advances in Fluid Mechanics VI, Skiathos, Greece, pp. 5967.
17 Jeong, J., and Hussain, F., 1995, On the Identification of a Vortex, J. Fluid
Mech., 285, pp. 6994.
18 Adrian, R. J., Christiansen, K. T., and Liu, Z.-C., 2000, Analysis and Interpretation of Instantaneous Velocity Fields, Exp. Fluids, 29, pp. 275290.
19 Graftieaux, L., Michard, M., and Grosjean, N., 2001, Combining PIV, POD
and Vortex Identification Algorithms for the Study of Unsteady Turbulent
Swirling Flows, Meas. Sci. Technol., 12, pp. 14221429.
20 Shinneeb, A.-M., Balachandar, R., and Bugg, J. D., 2008, Analysis of Coherent Structures in the Far-Field Region of an Axisymmetric Free Jet Identified
Using Particle Image Velocimetry and Proper Orthogonal Decomposition,
ASME J. Fluids Eng., 130, p. 011202.
21 Agrawal, A., and Prasad, A. K., 2002, Properties of Vortices in the SelfSimilar Turbulent Jet, Exp. Fluids, 334, pp. 565577.
22 Troolin, D. R., Longmire, E. K., and Lai, W. T., 2006, Time Resolved PIV
Analysis of Flow Over a NACA 0015 Airfoil With Gurney Flap, Exp. Fluids,
41, pp. 241254.
23 Druault, P., Guibert, P., and Alizon, F., 2005, Use of Proper Orthogonal Decomposition for Time Interpolation From PIV Data, Exp. Fluids, 39, pp.
10091023.
24 Weiland, C., and Vlachos, P., 2007, Analysis of the Parallel Blade Vortex
Interaction With Leading Edge Blowing Flow Control Using the Proper Orthogonal Decomposition, ASME Fluids Engineering Summer Conference,
San Diego, CA, Paper No. FEDSM2007-37275.
25 Kurtulus, D. F., Scarano, F., and David, L., 2007, Unsteady Aerodynamic
Forces Estimation on a Square Cylinder by TR-PIV, Exp. Fluids, 42, pp.
185196.
26 Hinze, J. O., 1959, Turbulence, McGraw-Hill, New York.
27 Dano, B. P. E., and Liburdy, J. A., 2006, Vortical Structures of a 45 Inclined
Pulsed Jet in Crossflow, Fluid Dynamics Conference, San Francisco, CA,
AIAA Paper No. 2006-3543.
28 Morse, D. R., and Liburdy, J. A., 2007, Dynamic Characteristics of Flow
Separation from a Low Reynolds Number Wing, Fluids Engineering Conference, San Diego, CA, Jul. 30Aug. 2, Paper No. FEDSM20070-37083.
29 Chen, G., Lin, Z., Morse, D., Snider, S., Apte, S., Liburdy, J., and Zhang, E.,
2008, Multiscale Feature Detection in Unsteady Separated Flows, IJNAM,
5, pp. 1735.
30 Chatterjee, A., 2000, An Introduction to the Proper Orthogonal Decomposition, Curr. Sci., 787, pp. 808817.
31 Sirovich, L., 1987, Turbulence and the Dynamics of Coherent Structures
Part I: Coherent Structures, Q. Appl. Math., 453, pp. 561571.
32 Berkooz, G., Holmes, P., and Lumley, J. L., 1993, The Proper Orthogonal
Decomposition in the Analysis of Turbulent Flows, Annu. Rev. Fluid Mech.,
25, pp. 539575.
33 Morse, D. R., and Liburdy, J. A., 2008, Experimental Time Resolved Flow
Features of Separation Over an Elliptic Leading Edge, 46th AIAA Aerospace
Sciences Meeting and Exhibit, Reno, NV, Jan. 710, AIAA Paper No. 2008655.
Drag Reduction in Turbulent Flow

With Polymer Additives
Shu-Qing Yang
Associate Professor
School of Civil, Mining and Environmental
Engineering,
University of Wollongong,
New South Wales 2522, Australia
e-mail: shuqing@uow.edu.au
The mean velocity profile and friction factor in turbulent flows with polymer additives are
investigated using Prandtls mixing-length theorem. This study reveals that the mixinglength theorem is valid to express the drag-reducing phenomenon and that the presence
of polymer additives increases the damping factor B in van Driests model; subsequently
reducing the mixing-length, this interprets that the polymer hampers the transfer of turbulent momentum flux, the velocity is increased, and flow drag is reduced. This study also
discusses the onset Reynolds number for drag reduction to occur. The predicted velocity,
friction factor, and onset Reynolds number are in good agreement with the measured data
in the literature. DOI: 10.1115/1.3111255
Keywords: drag reduction, velocity distribution, van Driests model, damping factor,
friction factor
Introduction
It has been widely observed that addition of a small amount of

macromolecular polymer to a Newtonian fluid may lead to a dramatic reduction in fluid resistance. Such phenomenon was first
discovered by Toms 1, and many research studies and publications had been devoted to the subject over the past 50 years. The
drag reduction DR flow has technological applications from fire
engines to oil pipes. Scientists and engineers have conducted
many studies on this phenomenon, including theoretical, experimental, and numerical approaches. However, the fundamental
mechanism has still remained under debate.
For drag-reducing flows, the onset of drag reduction is an interesting phenomenon. Virk 2 observed that if the Reynolds
number is larger than a certain value, the drag reduction occurs,
and the log-law is shifted by an amount B in the turbulent core,
with no change in slope; below which the flow behaves like a
Newtonian fluid, i.e., no drag reduction can be observed in the
flow of dilute polymer solutions, and the flow resistance is similar
to that of Newtonian fluid in the absence of additives. But the
threshold Reynolds number has not been clearly identified; thus it
is useful to determine the onset Reynolds number for drag reduction.
Virk 2 also divided the velocity profile of drag-reducing flow
into three regions, i.e., the viscous sublayer, a buffer or elastic
layer, and the turbulent core. This division appears to provide
better insight for understanding the mechanisms of the dragreducing phenomena; thus Virks three-layer model has been
widely adopted by researchers, such as by Larson 3, Min et al.
4,5, Gasljevic et al. 6, and Ryskin 7.
Since Virks 2 systematical observations, with the advent of
particle image velocimetry PIV Warholic et al. 8, White et al.
9, etc. and laser Doppler anemometry LDA, e.g., Durst et al.
10, Willmarth et al. 11, Luchik and Tiederman 12, Harder
and Tiederman 13, and Wei and Willmarth 14, turbulent structures can be observed, and their results showed that the rootmean-square rms of the fluctuations in the streamwise velocity
increases while the rms of the fluctuations in the wall-normal
direction decreases with drag reduction. Although the obtained
results provide valuable insights into drag reduction for flow containing polymer additives, there is as yet no clear picture of why
or how polymer additives reduce drag, and the physical mechaContributed by the Fluids Engineering Division of ASME for publication in the
received August 2, 2008; published online April 10, 2009. Assoc. Editor: Steven
Ceccio.
nisms of drag reduction are still tentative 15,16. Of the most

important result from these turbulent measurements by LDA and
PIV is the stress deficit, i.e., the total shear stress in dragreduced flow is greater than the sum of viscous shear stress
=du / dy and the measured Reynolds shear stress =uv.
Willmarth et al. 11, Den Toonder et al. 17, Warholic et al. 18,
and Gyr and Tsinober 19 attributed this stress deficit to the effects of the long-chain polymer in the fluid. Gyr and Tsinober 19
expressed this shear deficit as
Gy =
du
+ uv
dy
where is the kinematic viscosity, u is the time-averaged velocity,

is the fluid density, y is the distance from the wall, G is the shear
deficit and G 0, and is the total shear stress. For a pipe/channel
flow = u21 y / h, where u is the friction velocity and h is the
pipe radius or channel water depth. The existence of shear deficit suggests that the viscoelasticity as modified by the presence
of the long-chain polymers should be an important property of a
dilute polymer solution 4,5,17.
Gy was first called the Reynolds stress deficit or elastic
stresses by Schmmer and Thielen 20 who found that Gy is
essentially non-negligible. Gyr and Tsinober 19 commented that
Gy must be positive. Based on experimental data, Gyr and Tsinober 19 and Giesekus 21 expressed the function of shear deficit as
Gy = eff
du
dy
2a
where eff is the effective viscosity.

Using an analogy with Boussinesqs expression for the eddy
viscosity in turbulence, Yang and Dou 22 postulated that the
effective viscosity can be expressed by
eff = uh
2b
where is an elastic factor that only depends on polymer species

and concentration and is independent of turbulent characteristics.
Investigators found that the shear deficit can be derived from
the constitutive equations governing turbulent flows with polymer
additives, e.g., Min et al. 4,5 from the Oldroyd-B model, Dou
23 and Den Toonder et al. 17 from the Maxwell model, and
Berman 24 and Ptasinski et al. 25 from the FENE-P model.
Benzi et al. 26 derived the following momentum equation for
visco-elastic flows from the FENE-P model:
MAY 2009, Vol. 131 / 051301-1
du
du
y
= uv + + pRyy = u2 1
dy
dy
h
where Ryy is the ensemble average dimensionless extension tensor of the dyadic product of the end-to-end vector distance of the
polymer chains in the y direction and p is a viscosity parameter
that is related to the concentration of the polymer. The right-hand
side of Eq. 3 is simply the rate at which momentum is produced
by the pressure head, and on the left-hand side the Reynolds stress
is the momentum flux 2.
Gyr and Tsinobers 19 expression shown in Eq. 1 can be
rewritten as follows:
du
y
du
= uv + + eff = u2 1
dy
dy
h
Equation 5 establishes the relationship between the effective viscosity and properties of polymer solution in drag-reducing flows.
If Ryy and p can be determined from the polymer concentration
and the vector distance of the polymer molecules, then one is able
to determine eff. Equation 4 provides a useful tool to assess the
effective viscosity eff from the measured Reynolds shear stress
and velocity gradient. In other words, if the FENE-P model is
correctly expressed as the interaction of polymer molecules and
turbulence, then the obtained pRyy from the property of polymer and concentration must be the same as eff obtained from the
properties of turbulence using Eq. 4; this conclusion can be extended to other models, such as Oldroyd-B model and Maxwell
model.
Instead of solving Eq. 3 directly, Benzi et al. 2 and Lvov et
al. 16 introduced the energy balance equation and other assumptions, such as u21 y / h u2 and uv 0, then they claimed
that the velocity equation they obtained is valid even for the
whole flow region.
Different from the approach used by Benzi et al. 2 and Lvov
et al. 16, the present study makes an attempt on the direct solution of Eq. 3. By inserting Eq. 2b into Eq. 4, one obtains the
dimensionless Reynolds shear stress
u v
du+
y+
=
1
h+
dy +
u2
where u+ = u / u, y + = uy / , h+ = uh / , and
D = 1 + h +
Equation 6 becomes the governing equation for a Newtonian

flow when eff = 0 or D = 1. Equation 7 shows that the dragreducing parameter D varies linearly with the Reynolds number
h+, and the proportionality factor depends on visco-elastic effect caused by polymer additive.
Based on Maxwell model, Dou 23 obtained the following
expression for :
= ACo exp B10.7

o C w
where A = / 15, B1 = 25,500, is the intrinsic viscosity of the

polymer, C is the concentration of the solution g / cm3, Cw is the
concentration of the solution as a mass ratio g/g, and o is a
coefficient that depends on polymers characteristics for drag reduction.
To solve Eq. 6, one has to model the Reynolds shear stress in
drag-reducing flow. Now there are many equations available in the
literature to express the Reynolds shear stress in Newtonian fluid
flow, and one of the most widely cited is the mixing-length equation developed by Prandtl. Hence, it would be interesting to investigate whether Prandtls formula is valid in drag-reducing
flows.
2
051301-2 / Vol. 131, MAY 2009
Velocity Profile in Drag-Reduced Flows
The Reynolds shear stress in a turbulent flow can be modeled

by Prandtls mixing-length formulation, i.e.,
By comparing Eqs. 3 and 4, one has
eff = pRyy
Yang and Dou 22 investigated the velocity profiles in turbulent boundary layer flows; the emphasis of this study will be
placed on the pipe flows. The objectives of this study are as follows: 1 to theoretically investigate the mean velocity profile and
friction factor of drag reduction flow, 2 to quantitatively express
the onset Reynolds number for drag reduction, and 3 to determine the relation between the damping factor in van Driests 27
model and D.
u v = l 2

du
dy
where l is the mixing-length over which the turbulence characteristics remain unchanged 28. Inserting Eq. 9 into Eq. 6, one
has
l+2

du+
dy +
+ D
du+
1 = 0
dy +
10
where l+ = lu / and = y + / h+.

By solving the quadratic equation shown in Eq. 10, one obtains du+ / dy + as follows:
21
du+
=
dy + D + D2 + 4l+21
11
The velocity profile can be obtained by integrating Eq. 11 with

respect to y +,
+
u =
y+
21
dy
D + D2 + 4l+21
12
The most commonly employed expression of mixing-length for

pipe, channel, and boundary layer flows is the one developed by
van Driest 27:
l+ = ky +y +
13
where k is the von Karman constant. y is van Driests damping function, which can be expressed as follows:
+
y + = 1 exp y +/B
14
Equations 13 and 14 state that increasing B leads to decreasing

l. If B approaches infinity, then y + = 0 and l = 0. Therefore, one
can conclude that if the polymer hampers the transfer of turbulent
momentum flux, i.e., uv, then the mixing-length l will be
reduced and consequently the parameter B becomes larger. In
other words, the parameter B could reflect the damping effect
caused by polymer additives.
If the effective viscosity eff = 0, Eq. 2b gives = 0, and Eq.
7 shows D = 1, subsequently Eq. 12 becomes the velocity profile of the Newtonian fluid flow. Hence, one may conclude that
Prandtls mixing-length is valid for both drag-reducing flow and
Newtonian fluid flow if the damping factor is properly defined.
According to van Driest, the damping factor B needs to be
determined empirically; he suggested that for a Newtonian fluid
flow, the damping factor B is 26 =Bo. Cebeci and Smith 29
found that the damping factor should be a function of dimensionless quantities involving other factors such as mass transfer, compressibility, etc. Thus, it can be inferred that the damping factor
should be a function of D.
In fact, Lumley 15 and Berman 24 realized that polymer
stretching damp cross-stream or wall-normal velocity fluctuations
hamper eddy movement in the wall-normal direction. Hence, it is
natural that the damping factor in drag-reducing flows will be
different from that in Newtonian flows.
Lumley also found that in drag-reducing flows the polymer
molecules are expanded in the flow outside the viscous sublayer
Fig. 1 The velocity distribution in flows containing polymer additives after

Virk 2
due to possible stretching of the polymer molecules; this leads to

a thickening of the viscous sublayer leading to a decrease in the
velocity gradient at the wall. The thickness of the viscous sublayer, +, can be assessed by + = 11.6D3 22. It is well known
that inside the viscous sublayer, the velocity can be expressed by
u+ = y +, but Eq. 11 gives u+ = y + / D if y + 0. Hence Eq. 11 is
valid only in the flow outside the viscous sublayer. White and
Mungal 30 page 239 commented that in wall-bounded turbulent shear flows, polymer are primarily believed to be stretched
just outside the viscous sublayer; this means that in the viscous
sublayer D = 1 should be used.
To test this postulation, the experimental data compiled by
Virks 2 are shown in Fig. 1. The solid lines are calculated from
Eq. 12 using the popular and most accurate fourth-order Runge
Kutta method, the damping factor B is adjusted to fit the experimental data best, and the values of D are obtained using Eq. 7.
It can be seen from Fig. 1 that the difference between D = 1.1
and D = 1.0 falls into the measurement uncertainty; thus the onset
of drag reduction could be defined as D = 1.1. If D 1.1, the
velocity shift becomes discernible. From Eq. 7, one can estimate
the threshold Reynolds number for the occurrence of drag reduction as follows:

u h
0.1
15
where uh / c is the threshold Reynolds number. Equation 15

indicates that the onset Reynolds number for drag reduction depends only on the kind and concentration of polymer additives.
This is consistent with experimenters observations, such as Virk
2 and Gyr and Bewersdorff 31.
It can be seen from Fig. 1 that in the turbulent core, the predicted velocity is shifted by an amount B from the standard
log-law in Newtonian fluid flow, with no change in slope, and the
velocity distribution can be expressed by
u+ = 2.5 ln y + + 5.5 + B
16
It can be seen clearly from Fig. 1 that Virks three-layer model can
be expressed by a single equation, i.e., Eq. 12.
In Fig. 1, the dotted line represents Virks asymptote of maximum drag reduction MDR, i.e.,
u
uy
= 11.7 ln
17.0
u
17
Equation 17 was empirically proposed by Virk 2. But Benzi et

al. 26 and Lvov et al. 16 obtained Eq. 17 by solving Eq. 3.
Prandtls model shown in Fig. 1 gives continuous velocity gradient from the subviscous layer to the elastic layer and the turbulent
core. The theoretical results state that in the turbulent core, the
solid lines are parallel to the log-law and B in Eq. 16 depends
on D.
Virks asymptote or Eq. 17 makes the drag-reducing effect an
intriguing topic for research on turbulent flow because Virk stated
that this asymptote must be a feature of the turbulent flow. However, several experimenters, e.g., Gyr and Bewersdorff 31, show
that Virks assumption of the ultimate velocity profile may not be
generally true, i.e., whether or not a specific additive has its own
MDR asymptote is still an open question. From Fig. 1, one may
find that the lines with D = 2.63 and 2.69 are very close to the
data points that were used by Virk to obtain MDR asymptote or
Eq. 17.
Figure 1 shows that in the so-called buffer layer the calculated
velocity deviates noticeably from that of Newtonian fluid flow
D = 1, and the deviation becomes larger if D is higher. This
transition is achieved because the strain rate and vorticity fields
associated with the buffer layer are suitable to cause full extension
of polymers, leading to a corresponding large increase in the elongational viscosity 32,33; therefore a large increase in the effective viscosity or D just outside the viscous sublayer will suppress
turbulent fluctuations, increase the buffer-layer thickness, and reduce the wall friction 34. Therefore, it can be seen that the
mechanism of drag reduction by polymer additives has been well
expressed mathematically by changing the effective viscosity or
D from the viscosity sublayer to the buffer layer.
In Fig. 1, the value of damping factor B is obtained by the
best-fit technique; the relationship between the obtained B and D
is shown in Fig. 2. As expected, the experimental data show that
the damping factor depends on D, and the relation can be expressed as follows:
B
= Dm
Bo
18
where Bo = 26 and m = 2.5.

MAY 2009, Vol. 131 / 051301-3
Fig. 2 Relation between the damping factor B and D based on

the data shown in Fig. 1
Figure 1 includes the experimental data presented before 1971.

In order to illustrate the validity of Eq. 18, more recently published experimental data are given in Fig. 3. The data sets published by Den Toonder et al. 17, Luchik and Tiederman 12,
Warholic et al. 8, Ptasinski et al. 35, and Harder and Tiederman
13 are included in Fig. 3. The type of polymer, its concentration
C, pipe radius/flow depth h, and DR are listed in Table 1. Since
some experimental researchers did not report the polymer molecu-
lar weight and intrinsic viscosities in their experiments, the parameter D is estimated by the best-fit to experimental data instead
of using Eq. 8, and the damping factor B is calculated from Eq.
18 based on D.
It can be seen from Fig. 3 that the agreements between the
measured velocity profiles in drag-reduced flows and Eq. 12 are
reasonable. Virks asymptote is also included in Fig. 3 for comparison. From Fig. 3, one finds that Virks asymptote does not
represent a universal relationship for the maximum drag reduction; the experimental data of Ptasinski et al. 25,35 begin to
exceed Virks asymptote significantly at y + = 80. Warholic et al.
18 also concluded that Virks asymptote may not be universal.
However, it can be seen from Figs. 1 and 3 that Eq. 12 can
represent the data points. In the turbulent core, the solid lines
shown in Figs. 1 and 3 are parallel to the log-law for Newtonian
fluid flows with a shift of B without change in slope, as stated by
Eq. 16.
This study assumes that the velocity distribution in the viscous
sublayer follows Eq. 11 because in the near wall region the
distribution of the velocity must follow u+ = y +, not u+ = y + / D as
Eq. 11 gives. Thus it is necessary to test the hypothesis using the
velocity data in the near wall region. Figure 4 shows the measured
velocity distribution inside the viscous sublayer. It can be seen
clearly from Fig. 4 that this model can express the velocity distribution very well, indicating that the assumption is reasonable.
Currently, many researchers used direct numerical simulation
DNS to model the mechanism of drag-reducing flow, such as De
Angelis et al. 36, Min et al. 4,5, Housiadas and Beris 37, and
Ptasinski et al. 25, and their results show that DNS is a powerful
tool to understand the drag-reducing flow. This study uses a different approach to express the drag-reducing effect; thus it would
be useful to compare the proposed model with DNS method.
Fig. 3 Comparison of Eq. 12 with the measured velocity profiles

Table 1 Summary of mean velocity profile data shown in Fig. 3
Source
Den Toonder et al. 17
Luchik and Tierderman 12
Warholic et al. 8
Harder and Tiederman 13
Warholic et al. 18
Ptasinski et al. 25
051301-4 / Vol. 131, MAY 2009
Solvent
Polymer
Water
Water
Water
Water
Water
Water
Superfloc A110
Separan AP-273
Percol 727
Separan AP-273
Percol 727
Superfloc A110
Polymer
Pipe radius/water
concentration,
depth,
Drag reduction
C wppm
h cm
DR
20
1.32.1
1.24
5
13
175, 435
4
1.25
2.54
3.0
2.54
4.0
24.2%
22.0%
42.0%
64.0%
65%, 70%
Fig. 4 Velocity distribution inside the near wall region
Ptasinski et al. 35 measured the velocity profiles in dragreducing flows, and the comparison of the measured data with the
present model is included in Fig. 3, in which the discrepancy is
noticeable. Ptasinski et al. 25 modeled their measurements in
2001 using DNS, and the comparison between the same measured
data as shown in Fig. 3 with DNSs results is included in Fig. 5.
Based on the data of Ptasinski et al. 35 in Figs. 3 and 5, one
could find that the present model yields the best agreement relative to DNSs results as well as Virks asymptote of maximum
drag reduction.
3 Relation Between Reynolds Number and DragReducing Parameter

Figures 1 and 2 show that the measured velocity profiles in
drag-reduced flows can be described by van Driests model, and
the difference between the drag-reducing flow and the Newtonian
fluid flow is caused by the effective viscosity or the single param-
eter D, which contains the effects of polymer and turbulence

characteristics. Equation 7 states that D varies linearly with the
Reynolds number. Therefore, it is worthwhile to investigate
whether the linear relationship exists. Reischman and Tiedermans
38 experimental data are selected for this purpose. Their experiment was conducted in a channel 305 mm high and 28.4 mm
wide, and the velocity was measured by LDA. Magnifloc 837A
was used as the polymer additive. The concentration of dilute
polymer solution was maintained at 100 ppm by weight. All measured velocity profiles are shown in Fig. 6 in which DR stands for
drag reduction. The drag-reducing parameter D is determined by
the best-fit to the data. The relation between the obtained D and
the Reynolds number uh / is shown in Fig. 7, in which the data
points collapse into a linear relationship, as predicted by Eq. 7.
The linear relation shown in Fig. 7 is understandable because
Reischman and Tiederman 38 did not change the polymer species and concentration in their experiment, but only changed the
Fig. 5 Comparison of measured velocity profiles in drag-reducing flows

with DNSs model after Ptasinski et al. 25,35
MAY 2009, Vol. 131 / 051301-5
Fig. 6 Mean velocity profiles measured by Reischman and Tiederman 38;

DR stands for drag reduction
Reynolds number. In other words, the straight line in Fig. 7 interprets that polymer species and concentration only affect the value
of . For Magnifloc 837A at 100 ppm used by Reischman and
Tiederman 38, the coefficient = 0.0011. Therefore, one can
conclude from Reischman and Tiedermans 38 measurements
that the effective viscosity is proportional to uh, as shown in Eq.
2b.
Friction Factor
Equations 7 and 8 state that the elastic factor is independent of flow strength, and D varies linearly with the Reynolds
number, i.e., ur / ; this has been proved in Fig. 7. It would be
interesting to investigate how the factor changes with the polymer concentration. To do so, the friction factor of drag-reducing
flows f defined as follows is investigated:
u
f=2
V
u2 1
dul
y
=
dy
h
20
where ul is the velocity in a laminar flow. Integrating Eq. 20 and

imposing the nonslip boundary condition, i.e., ul = 0 at y = 0, one
obtains
ul+ = y + 1
21
For the transitional region from the laminar to the turbulent state,
Yang and Dou 22 obtained the following velocity distribution:
ut+ = rlul+ + 1 rlu+
22
where V is the cross-sectional averaged velocity.

In a laminar flow, uv = 0, Eq. 4 becomes
19
where u+t is the dimensionless velocity in the transitional region,

u+l is determined using Eq. 21, and u+ is determined using Eq.
12. The parameter rl is the probability of laminar occurrences
and may be expressed as follows 22:
Fig. 7 Linear relation between D and Reynolds number based on Reischman and Tiedermans 38 data
051301-6 / Vol. 131, MAY 2009
Fig. 8 Comparison of friction factor between the measured results by Virk

2 and Eq. 24
1
rl = e
n=1
n Rk
n! R
2n
for R Rk
for R Rk
23
where R is the Reynolds number =uh / and Rk is the threshold Reynolds number at which the flow becomes unstable and
turbulence appears. For a pipe flow Rk = 67.82 or 2Vh / = 2300.
Therefore, one is able to obtain the drag coefficient f from the
mean velocity, which is determined by integrating Eq. 12,
1 1 V
1
R 1 rl
=
rl +
=
4
f 2 u 2
R2
2R y +u+dy +
24
The comparison of 1 / f given by Eq. 24 and Virks 39 experimental data in smooth pipes is presented in Fig. 8, which shows
that Eq. 24 = 0 represents the data points obtained in Newtonian fluid flows well. For drag-reducing flows, Virk used polyethyleneoxide N750 o = 0.019 as the polymer additive with the
following polymer concentrations: 43.6 ppm, 98.6 ppm, and 939
ppm; Eq. 8 gives the corresponding elastic factors
= 0.000158, 0.000326, 0.000817. In Fig. 8, the experimental data
for polyethyleneoxide W301 o = 0.166 with a concentration of
18.7 ppm are also included for comparison; Eq. 8 gives
= 0.0033 for this case. Figure 8 indicates that for each line, the
parameter holds constant over the full range of Reynolds numbers, i.e., uh / . The good agreement between the measured and
predicted friction factors indicates that is independent of the
Reynolds number.
For the drag reduction flows shown in Fig. 8, the calculated
onset Reynolds number Re f=22uh / from Eq. 15 is 1790
for = 0.000158, which is very close to the measured value of
1500; for = 0.000326, the calculated Re f is 867 and the observed value is 960; thus, Eq. 15 can express the onset of drag
reduction 4043.
Conclusions
The applicability of Prandtls mixing-length theorem in dragreducing flows is investigated, and the mixing-length equation
developed by van Driest is used. This study finds that the damping
factor B in van Driests model depends on the effective viscosity
caused by polymer additives, and the velocity in visco-elastic fluid
flow can be well predicted by Prandtls theorem. Based on the
comparison between measured and predicted velocity and friction
factors in pipe/channel flows, this study reaches the following
conclusions.
1. The difference between the drag-reducing flow and Newtonian fluid flow is caused by the stress deficit, which can be
well represented by the sole parameters, D or . The experimental data confirm that D is proportional to the Reynolds number.
2. The mixing-length model can be applied to drag-reducing
flows, and the mixing-length is reduced after the polymer is
added to the Newtonian fluid flow, or the damping factor in
van Driests expression increases with D. The calculated
results show that the mean velocity becomes higher with the
increase in D. The measured velocity profiles can be described by Eq. 12 from the viscous sublayer through the
buffer layer to the turbulent core.
3. The measured friction factor shows that the elastic factor
or effective viscosity eff is independent of Reynolds number
ur / . The friction factor can be expressed by the derived
equation.
4. The theoretical results show that the onset Reynolds number
only depends on the drag-reduction parameter . The onset
Reynolds number decreases with the increase in . If D
1.1, the velocity profile and friction factor for dragreducing flows perform as the Newtonian fluid flow.
Acknowledgment
The author would like to express his sincere appreciation to the
anonymous referees for their careful review that greatly improves
this papers quality.
References
1 Toms, B. A., 1949, Some Observation on the Flow of Linear Polymer Solutions Through Straight Tubes at Large Reynolds Numbers, Proceedings of the
First International Congress on Rheology, Vol. 2, pp. 135141.
2 Virk, P. S., 1971, An Elastic Sublayer Model for Drag Reduction by Dilute
Solutions of Linear Macromolecules, J. Fluid Mech., 45, pp. 417440.
3 Larson, R. G., 2003, Analysis of Polymer Turbulent Drag Reduction in Flow
Past a Flat Plate, J. Non-Newtonian Fluid Mech., 11123, pp. 229250.
4 Min, T., Choi, H., and Yoo, J. Y., 2003, Maximum Drag-Reduction in a
Turbulent Channel Flow by Polymer Additives, J. Fluid Mech., 492, pp.
91100.
5 Min, T., Yoo, J. Y., Choi, H., and Joseph, D. D., 2003, Drag Reduction by
Polymer Additives in a Turbulent Channel Flow, J. Fluid Mech., 486, pp.
213238.
6 Gasljevic, K. Aguilar, G., and Matthys, E. F., 2001, On Two Distinct Types of
Drag-Reducing Fluids, Diameter Scaling, and Turbulent Profiles, J. NonNewtonian Fluid Mech., 963, pp. 405425.
7 Ryskin, G., 1987, Turbulent Drag Reduction by Polymers: A Quantitative
Theory, Phys. Rev. Lett., 59, pp. 20592062.
8 Warholic, M. D., Heist, D. K., Katcher, M., and Hanratty, T. J., 2001, A Study
With Particle-Image Velocimetry of the Influence of Drag-Reducing Polymers
on the Structure of Turbulence, Exp. Fluids, 31, pp. 474483.
9 White, C. M., Somandepalli, V. S. R., and Mungal, M. G., 2004, The Turbu-
MAY 2009, Vol. 131 / 051301-7
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
lence Structure of Drag-Reduced Boundary Layer Flow, Exp. Fluids, 36, pp.
6269.
Durst, F., Keck, T., and Kline, R., 1985, Turbulence Quantities and Reynolds
Stresses in Pipe Flow of Polymer Solutions, Proceedings of the First International Conference on Laser Anemometer-Advances and Applications, BHRA,
p. 31.
Willmarth, W. W., Wei, T., and Lee, C. O., 1987, Laser Anemometer Measurements of Reynolds Stress in a Turbulent Channel Flow With Drag Reducing Polymer Additives, Phys. Fluids, 30, pp. 933.
Luchik, T. S., and Tierderman, W. G., 1988, Turbulent Structure in LowConcentration Drag-Reducing Channel Flows, J. Fluid Mech., 190, pp. 241
263.
Harder, K. J., and Tiederman, W. G., 1991, Drag Reduction and Turbulent
Structure in Two-Dimensional Channel Flows, Philos. Trans. R. Soc. London,
Ser. A, 336, pp. 1934.
Wei, T., and Willmarth, W. W., 1992, Modifying Turbulent Structure With
Drag-Reducing Polymer Additives in Turbulent Channel Flows, J. Fluid
Mech., 245, pp. 619641.
Lumley, J. L., 1969, Drag Reduction by Additives, Annu. Rev. Fluid Mech.,
1, pp. 367384.
Lvov, V. S., Pomyalov, A., Procaccia, I., and Tiberkvich, V., 2004, Drag
Reduction by Polymers in Wall Bounded Turbulence, Phys. Rev. Lett., 92, p.
244503-4.
Den Toonder, J. M. J., Hulsen, M. A., Kuiken, G. D. C., and Nieuwstadt, F. T.
M., 1997, Drag Reduction by Polymer Additives in a Turbulent Pipe Flow:
Numerical and Laboratory Experiments, J. Fluid Mech., 337, pp. 193231.
Warholic, M. D., Massah, H., and Hanratty, T. J., 1999, Influence of DragReducing Polymers on Turbulence: Effects of Reynolds Number, Concentration and Mixing, Exp. Fluids, 27, pp. 461472.
Gyr, A., and Tsinober, A., 1997, On the Rheological Nature of Drag Reduction Phenomena, J. Non-Newtonian Fluid Mech., 73, pp. 153162.
Schmmer, P., and Thielen, W., 1980, Structure of Turbulence in Viscoelastic
Fluids, Chem. Eng. Commun., 4, pp. 593606.
Giesekus, H., 1981, Structure of Turbulence in Drag Reducing Fluids, Lecture
Series 1981-86, von Karman Institute for Fluid Dynamics, Rhode-SainGense, Belgium.
Yang, S. Q., and Dou, G., 2005, Drag Reduction in Flat-Plate Turbulent
Boundary Layer Flow by Polymer Additive, Phys. Fluids, 176, p. 065104.
Dou, G. R., 1996, Basic Law in Mechanics of Turbulent Flows, China Ocean
Eng., 101, pp. 144.
Berman, N. S., 1989, Drag Reduction in Fluids Flows: Techniques for Friction
Control, R. H. J. Sellin and R. T. Moses, eds., Ellis Horwood Limited, Chichester.
Ptasinski, P. K., Boersma, B. J., Nieuwstadt, F. T. M., Hulsen, M. A., Van den
Brule, B. H. A. A., and Hunt, J. C. R., 2003, Turbulent Channel Flow Near
Maximum Drag Reduction: Simulations, Experiments and Mechanisms, J.
051301-8 / Vol. 131, MAY 2009
Fluid Mech., 490, pp. 251291.

26 Benzi, R., Angelis, E. D., Lvov, V. S., Procaccia, I., and Tiberkevich, V.,
2006, Maximum Drag Reduction Asymptotes and the Cross-Over to the Newtonian Plug, J. Fluid Mech., 551, pp. 185195.
27 Van Driest, E. R., 1956, On Turbulent Flow Near a Wall, J. Aeronaut. Sci.,
23, pp. 10071010.
28 Hinze, J. O., 1975, Turbulence, 2nd ed., McGraw-Hill, New York.
29 Cebeci, T., and Smith, A. M. O., 1974, Analysis of Turbulent Boundary Layers,
Academic, New York, Chap. 6.
30 White, C. M., and Mungal, M. G., 2008, Mechanics and Prediction of Turbulent Drag Reduction With Polymer Additives, Annu. Rev. Fluid Mech., 40,
pp. 235256.
31 Gyr, A., and Bewersdorff, H. W., 1995, Drag Reduction of Turbulent Flow by
Additives, Kluwer Academic, Dordrecht.
32 Hinch, E. J., 1977, Mechanical Models of Dilute Polymer Solutions in Strong
Flows, Phys. Fluids, 20, pp. 2230.
33 Metzner, A. B., and Metzner, A. P., 1970, Stress Level in Rapid Extensional
Flow of Polymeric Fluids, Rheol. Acta, 9, pp. 174181.
34 Lumley, J. L., 1973, Drag Reduction in Turbulent Flow by Polymer Additives, J. Polym. Sci. Macromol. Rev., 7, pp. 263290.
35 Ptasinski, P. K., Nieuwstadt, F. T. M., Van Den Brule, B. H. A. A., and Hulsen,
M. A., 2001, Experiments in Turbulent Pipe Flow With Polymer Additives at
Maximum Drag Reduction, Flow, Turbul. Combust., 66, pp. 159182.
36 De Angelis, E., Casciola, C. M., and Piva, R., 2002, DNS of Wall Turbulence:
Dilute Polymers and Self-Sustaining Mechanisms, Comput. Fluids, 31, pp.
495507.
37 Housiadas, K. D., and Beris, A. N., 2003, Polymer-Induced Drag Reduction:
Effects of the Variations in Elasticity and Inertia in Turbulent Viscoelastic
Channel Flow, Phys. Fluids, 15, pp. 23692384.
38 Reischman, M. M., and Tiederman, W. G., 1975, Laser-Doppler Anemometer
Measurements in Drag-Reducing Channel Flows, J. Fluid Mech., 702, pp.
369392.
39 Virk, P. S., 1971, Drag Reductions in Rough Pipes, J. Fluid Mech., 45, pp.
225246.
40 Edwards, M. F., and Smith, R., 1980, The Use of Eddy Viscosity Expressions
for Predicting Velocity Profiles in Newtonian, Non-Newtonian and DragReducing Turbulent Pipe Flow, J. Non-Newtonian Fluid Mech., 7, pp. 153
169.
41 Gust, G., 1976, Observations on Turbulent= Drag Reduction in a Dilute Suspension of Clay in Sea-Water, J. Fluid Mech., 751, pp. 2947.
42 Virk, P. S., Merrill, E. W., Mickley, H. S., and Smith, K. A., 1967, The Toms
Phenomenon: Turbulent Pipe Flow of Dilute Polymer Solutions, J. Fluid
Mech., 302, pp. 305328.
43 Yang, S. Q., and Dou, G., 2008, Modeling of Viscoelastic Turbulent Flow in
Open Channel and Pipe, Phys. Fluids, 206, p. 065105.
Shuhong Liu1
e-mail: liushuhong@mail.tsinghua.edu.cn
Liang Zhang
State Key Laboratory of Hydro Science and
Hydraulic Engineering,
Tsinghua University,
Beijing 100084, China
Michihiro Nishi
Kyushu Institute of Technology,
Sensui-cho 1-1,
Tobata, Kitakyushu 804-8550, Japan
e-mail: nishi@mech.kyutech.ac.jp
Yulin Wu
State Key Laboratory of Hydro Science and
Hydraulic Engineering,
Tsinghua University,
Beijing 100084, China
Cavitating Turbulent Flow

Simulation in a Francis Turbine
Based on Mixture Model
As a numerical method to study the cavitation performance of a Francis turbine, the
mixture model for the cavity/liquid two-phase flow is adopted in the cavitating turbulent
flow analysis together with the re-normalization group (RNG) k- turbulence model in
the present paper. The direct coupling numerical technique is used to solve the governing
equations of the mixture model for the two-phase flow. Unsteady cavitating flow simulation around a hydrofoil of ALE15 is conducted as preliminary evaluation. Then, the
cavitating flow in a Francis turbine is treated from the steady flow simulation since the
feasibility of the cavitation model to the performance prediction of the turbine is the
present major concern. Comparisons of the computational results with the model test
data, i.e., the cavitation characteristics of hydraulic efficiency and the overload vortex
rope at the draft tube inlet being reproduced reasonably, indicate that the present method
has sufficient potential to simulate the cavitating flow in hydraulic turbines. Further, the
unsteady cavitating flow simulation through the Francis turbine is conducted as well to
study the pressure fluctuation characters caused by the vortex rope in the draft tube at
partial load operation. DOI: 10.1115/1.3112382
Keywords: cavitation, turbulent flow simulation, Francis turbine, mixture model
Introduction
A Francis turbine is usually designed under the cavitation-free

condition at the rated load. But, unfavorable phenomena to affect
the reliability and smooth operation of the turbine have frequently
happened due to the occurrence of cavitation since its operation at
the condition far from the rated load is not a rare case. Thus, it is
desirable if we can predict the cavitating flow in the turbine reasonably at the designing stage. One of such studies was made by
Susan-Resiga et al. 1, who showed the predicted results of the
initial cavitation number of a Francis turbine runner. And from
literature survey on this topic, we can see many contributions
summarized as follows.
At the first stage, simplified cavitating flow models have been
demonstrated based on the ideal fluid assumption and the singularity method for simulating the cavity flows around hydrofoils
and in hydraulic machinery. For instance, using a cavitating flow
model proposed by Yamaguchi and Kato 2, Brewer and Kinnas
3 and Pellone and Peallat 4 calculated the flow around 2D, 3D
hydrofoils and predicted the local cavities near the hydrofoil surfaces. De Lange and De Bruin 5 simulated the periodic variation
in bubble in potential flow. Recently, Watanabe et al. 6 used the
singularity method to analyze the thermodynamic effect of partial
cavitation.
There are such papers where the inviscid flow simulation by
solving Euler equations is conducted for analyzing the cavitating
flow. Chen and Heister 7 simulated the cavitating flow around an
axisymmetric body by using the marker and cell method. Ventikos
and Tzabiras 8 calculated the cavitating flow around a hydrofoil
and considered the temperature variation in the flow by using the
pressure-correction method. Horiguchi et al. 9 proposed a
1
JOURNAL OF FLUIDS ENGINEERING. Manuscript received October 20, 2007; final manuscript received February 11, 2009; published online April 13, 2009. Assoc. Editor:
Chunill Hah.
method for predicting the steady cavitation in turbopumps based

on the assumption that the fluid was inviscid and the stream surface is rotationally symmetric.
The single-phase turbulent flow model is regarded as one of the
popular numerical methods. Though it has been developed extensively, it may be applicable to the fixed-bubble cavitating flow and
not to the dissociative bubble flow and bubble cloud since the
momentum and mass transfer between bubble and liquid cannot
be taken into account.
It should be noted that some attractive results were obtained
through a barotropic model 10,11 to capture the main physics of
complex cavitation wake flows. The simulation methodology was
based on large eddy simulation LES, using a barotropic phase
model to couple the continuity and momentum equations.
The other method is the simulation of two-phase flow consisting of the continuous liquid phase and the dispersed phase 7,12,
where the two-fluid model and the mixture model have been developed. In the former model, the dispersed cavity or bubble
phase is treated as a pseudofluid in the Eulerian approach, i.e., the
flow of the dispersed phase is described by conservation equations
of mass, momentum, and energy in continuum mechanics. In this
model, there exists not only the slip between the carrier fluid and
the dispersed phase but also the different diffusion between them.
Rieger 13 applied the two-fluid model to his simulation of cavitating flows. Grogger and Alajbegovic 14 calculated the cavitating flow in the Venturi tube with this model. Liu et al. 15 used
the k--Ac turbulence model i.e., the k- model for liquid phase,
the local following model for cavity phase to calculate the bubble
flow in the Francis turbine.
In the mixture model, it is assumed that there exist the dynamic
balance and the diffusion balance in both the liquid phase and the
cavity phase of the cavitating flow, and velocities, temperatures,
and densities of both phases are identical to each other at every
position in the whole two-phase flow field. There are contributions
to simulate the turbulent cavitating flow by Navier-Stokes N-S
equations of the mixture and an additional equation of cavity or
liquid volume or mass fraction. Singhal et al. 16 used the
mass fraction equation and k- turbulence model to simulate the
cavitating mixture flow around a foil. Kunz et al. 17 and Brewer
MAY 2009, Vol. 131 / 051302-1
and Kinnas 3 used multiple species approaches with additional

establishment of a mass transfer law between liquid and vapor.
This model was applied even when the relative motion between
the phases should be taken into account in the calculation. Senocak and Shyy 18 used the mass fraction equation and k- turbulence model, as well as the pseudocompressibility method to
simulate the 3D and axisymmetric cavitating flows. The mixture
model has been developed further by introducing various mass
transfer expressions. One of these expressions is based on the
bubble dynamics, another is established from the mechanics of
evaporation and condensation between the cavity and liquid
phase. Singhal et al. 19 presented the full cavitation model of
mixture from the RayleighPlesset equation with considering
phase change rate, turbulence, and nondissolved gas effects.
Aschenbrenner et al. 20 used the mixture model to compute the
unsteady cavitating flow and compared it with experiment.
From evaporation and condensation mechanism between vapor
and liquid and the assumption of unchanged pressure in cavity, the
mass change rate of cavity should be proportional to the difference between liquid and vapor pressures. Saito et al. 21 derived
cavitation phase change rate expression from the plane evaporation and evaluated it by calculating the cavitation flow around
NACA0015 foil. Okita and Kajishima 22 showed a cavitation
change expression in volume fraction form. Guo et al. 23 used
this equation to calculate the cavitating flow in the draft tube of
the Francis turbine. Though the Reynolds averaged N-S equations model is widely used to simulate the cavitating flow
2426 at present, there are attempts to apply the direct numerical simulation to bubbly flow calculation 27,28.
Though the nature of the cavitating two-phase flow is highly
dynamic and unstable, its appropriate modeling is still necessary
for engineering application. Liu et al. 29 demonstrated that the
numerical method based on a mixture model for the cavity/liquid
two-phase flow together with the RNG k- turbulence model was
acceptable to simulate the cavitating flow around a hydrofoil.
Thus, for the development of a numerical method to predict the
cavitation performance of a hydraulic turbine, we have examined
the applicability of the proposed method from two kinds of cavitating flow simulation in this paper. One is unsteady cavitating
flow simulation around a hydrofoil of ALE15. The other is the
steady and unsteady numerical simulation of cavitating flow in a
Francis turbine at three operating conditions, i.e., overload, rated,
and part load.
Numerical Simulation of Cavitating Flow
2.1 Governing Equations of Mixture Model. The following

are governing equations of the present mixture model of the
cavity/liquid two-phase flow, including those effects of nondissolved gas, turbulence, tension of interface at cavity, and phase
change rate based on the evaporation and condensation mechanism.
1 Continuity equation of mixture
u j = 0
+
t xj
2 Continuity equation of cavity phase
caca
cacau j = S
+
t
xj
3 Momentum equations of mixture
ui
p ji
u j u i = g i
+
+
t
xj
xi xi
It is noted that Eq. 2 is the cavitation model equation for the

calculation of cavitating flow based on the evaporation and condensation mechanism proposed by Okita and Kajishima 22,
051302-2 / Vol. 131, MAY 2009
where the static pressure p and vapor pressure pv are replaced

with p* and p*v , respectively, in cavities to consider those effects
of the surface tension, the turbulence effect, and the nondissolved
gases in this paper as follows,
If the surface tension is considered, the pressure p is given
by
p* = p +
2
r
Equation 5 for the apparent vapor pressure p*v is introduced to

describe the turbulence effect 16 in the vapor pressure pv.
p = pv + 0.195k
The source term S in Eq. 2 is adopted to express the mass

transfer rate per volume proposed by Cammenga 30
If p pv condensation being dominant,
S=
3ca
M
C1
r
2R
1/2
pv
T
If p pv evaporation being dominant,

S=
31 ca u
M
C2
r
2R
1/2
pv
T
where correction coefficients of C1 = 0.13 and C2 = 0.01 are used to

consider the effect of phase change rate.
And considering the effect of nondissolved gases 19, the density of mixture is expressed as
= caca + uu + 1 ca ul
where ca = v is the density of saturated vapor and u is the density of nondissolved gases, both of which are in the following
forms from ideal gas assumption:
ca =
M p
R T
u =
M u p
R T
10
Equations 13 are closed in laminar flow or in turbulent

flow with additional turbulence model equations, assuming that
the cavity radius r is constant. In the case of water as the working
liquid, it may be determined that the initial mass fraction of cavity
phase f ca = caca = 1 77 mg/ m3 and the initial nuclei radius r0
= 3 10 m. Using them, we can get the number of cavities in a
unit volume, which is assumed to be constant in the computation.
Thus, the new cavity radius is calculated from parameter of ca
data in each step of iteration of the computation.
2.2 Numerical Algorithm. The time-dependent governing
equations are discretized both in space and time. The unsteady
second order implicit formulation is used to calculate the unsteady
flow. And the RNG k- turbulence model is applied for the unsteady turbulent flow with the wall function near walls. The finite
volume algorithm, with the second order central difference for the
source terms and with the second order upstream difference for
the convective terms of the spatial discrete governing equations, is
used for the numerical simulation.
Though the cavitating flow was simulated by using a pressurecorrection method derived from the semi-implicit method for
pressure-linked equations SIMPLE algorithm and a finite volume discretization in the work made by Fortes-Patella et al. 26,
the direct coupling method is used to solve the incompressible
flow in the present simulation 31. The discrete momentum equations and the continuity equations for the complete flow field are
solved together without iteration and corrections. This numerical
method will need large computer storage, but it will improve the
Fig. 2 Cavitating flow around ALE15 hydrofoil: a geometry of

the hydrofoil and calculation coordinate system and b
streamlines around the hydrofoil and bubbly cloud
Fig. 1 The direct coupling procedure for incompressible flow
stability in the numerical procedure, as shown in Fig. 1, where the

iteration is made in the case of nonlinear cavitating turbulent flow.
2.3 Turbulence Model. As the Reynolds averaged Navier
Stokes solver CFX code is used to solve the turbulent flow of
cavity/liquid two-phase flow in the present work, the turbulence
model is necessary. Coutier-Delgosha et al. 32 evaluated the
influence of turbulence model on the numerical results of unsteady cavitating flow. In this paper the RNG k- is used on the
two-phase flow in the mixture model 33, which is similar to
single flow except the definition of viscosity of the mixture as
follows 34:
= caca + uu + 1 ca ul
11
where ca is the viscosity of cavity phase, l is the water viscosity, and u is the viscosity of nondissolved gas in water.
2.4 Boundary Conditions for Steady Flow. The following
boundary conditions are used to get steady solution of cavity/
liquid turbulent flow in the present simulation.
1 Input data of all physical variants are prescribed at the inlet
boundaries of the flow field.
2 The gradients of all physical variants normal to the boundaries are given at the outlet.
3 On solid walls of the domain, the nonslip flow condition is
adopted. The velocity distribution in the boundary layer is
expressed as the wall function near the solid walls.
4 Zero pressure gradient normal to the surface is used for all
boundaries of the domain except a point corresponding to
the reference pressure. In order to satisfy the continuity
condition from the inlet to the outlet of calculation domain,
correction of velocity magnitude to the outlet distribution is
made based on the difference between the inflow and the
outflow during the calculation.
Simulation of the unsteady cavity/liquid turbulent two-phase

flow around the hydrofoil is conducted in this paper, using steady
flow results as the initial condition. An example of calculated
results is shown in Fig. 2, where Fig. 2a is the geometry of
ALE15 hydrofoil with the calculation coordinate system. Figure
2b displays the streamlines around the foil and bubbly cloud
appeared on the foil surface. The boundary of the bubbles is obtained from a cavity volume fraction of 10% proposed by Okita
and Kajishima 22, who explained that the interface between water and bubble could be observed clearly if the cavity volume
fraction was more than 10%.
As an example of steady flow results, the numerical result of
velocity distribution being compared with the particle image velocimetry PIV measurement and the result from the full cavitation model 19 are shown in Fig. 3, where the ordinate is the
velocity and the abscissa is y or the distance from the hydrofoil
surface at the location of x = 13 mm and z = 5 mm. From Fig. 3,
it can be seen that the calculated result from the full cavitation
model underestimates the velocity, and the present simulation reasonably corresponds to the experimental data.
As a typical result of unsteady flow simulation, four snapshots
of cavitation bubbles on the top surface being taken with the time
interval of about 0.1 s are displayed in Fig. 4, where Fig. 4a
corresponds to the view in y b and Fig. 4b is the view in xb. It is
suspected from this result that the area of cavitation bubbles varies
from time to time in the tested condition.
Cavitating Flow in Francis Turbine
4.1 Francis Turbine Parameters. The cavitating turbulent

flow through a model Francis turbine see Table 1 has been simulated. Figure 5a shows the calculation domain of the turbine,
Cavitating Flow Around a Hydrofoil
The steady cavity/liquid turbulent two-phase flow around the

ALE15 hydrofoil was simulated by using the turbulent flow analysis with the cavitating flow mixture model described in Sec. 2.1
29. The hydrofoil parameters for the calculation were set using
the experimental conditions made by Dular et al. 35. The boundary conditions are as follows:
1 inflow plane: uniform velocity of 13 m/s
2 outflow plane: constant pressure of 197.2 kPa
3 solid walls: nonslip condition
Fig. 3 Comparison of velocity distribution at x = 13 mm and

z = 5 mm
MAY 2009, Vol. 131 / 051302-3
Fig. 4 Numerical snapshots of cavitation bubbles on the top surface of

hydrofoil: a view in yb and b view in xb
which is a complete flow passage of the turbine consisting of

spiral casing SC, stay vanes SVs, guide vanes GVs, runner,
and the draft tube DT. The advantages of adopting the complete
flow passage of turbine as the calculated domain are as follows.
1 Reasonably accurate boundary conditions can be prescribed
at the entrance section of spiral casing as the inlet boundary
and at the exit section of draft tube as the outlet boundary.
2 The interaction between a rotating runner and stationary
guide vanes and that between the runner and the draft tube
can be included in the calculation. Figure 5b indicates the
Table 1 Geometrical parameters of the model turbine

Parameter
No. of stay vanes
No. of guide vanes
Reference diameter of GV
Height of GV
Inlet diameter of runner
Outlet diameter of runner
No. of runner blades
051302-4 / Vol. 131, MAY 2009
Value
ZS = 24
Z0 = 24
D0 = 1.111D1
B0 = 0.288D1
D1 = 372.2 mm
D2 = 366.0 mm
Z = 15
Fig. 5 Model Francis turbine: a calculation domain, b interaction surfaces, and c survey points of pressure in draft tube
Table 4 Prediction of runner torque experiment: = 0.142
Table 2 Grid system model with predicted turbine torque

Grid model
Total
elements
Runner
elements
Runner element size

mm
Torque
N m
1
2
3
1,180,000
1,880,000
3,710,000
100,000
780,000
2,630,000
6
3
2
1129.792
1149.769
1149.767
Cases
two fluid-fluid interfaces upstream and downstream of the

runner, which rotate in the reference frame. Namely, the
frozen mesh of the runner was used to consider the interaction between a rotating runner and stationary components
in the steady cavitating flow simulation. And the slipping
mesh of the runner was used to consider the interaction in
the unsteady cavitating flow simulation.
In order to examine the independence of the numerical simulation on the grid system, as shown in Table 2, three unstructured
tetrahedron grid models with different numbers of elements were
tested as the preliminary calculation. Grid model 2 is selected in
the following calculation based on the predicted results of torque
acting on the runner, which are also shown in Table 2.
Table 3 shows three test cases corresponding to those turbine
operating conditions of part load, rated load, and overload, which
are set by guide vane openings in the experiment. Following the
experimental conditions, the constant test head of 30 m and the
rotating speed of 1122 min1 unit rotational speed of 75 rpm are
used in the calculation.
Another parameter to show the operating condition is the cavitation number given by
Ha HVA HS HV
H
12
where HVA is the vacuum head in the draft tank of the test rig,
which indicates the pressure level at the outlet of the draft tube.
Ha is the atmospheric pressure head. HS is the suction head of the
turbine. H is the test head acting on the turbine. HV is the vapor
pressure head at the test temperature.
In the cavitating flow simulation, the following assumptions
were used:
1 radius of cavitation nuclei, r0 = 0.5 m
2 volume fraction of nondissolved gas, u = 5 105
3 temperature at the inflow plane, T = 298 K
4.2 Validation of Simulation. Since applicability of the cavitation model to the performance prediction of a hydroturbine is
our major concern, its validation is conducted from the steady
flow simulation.
Predicted torque. Table 4 shows the predicted values of runner
torque with and without cavities for three cases. We see that the
differences between the results of the single-phase simulation and
those of the cavitating flow are almost negligible. This is because
of the higher cavitation number around = 0.142 at the corresponding operating condition. Though the underestimation is ob-
Experiment
N m
Single-phase
N m
Cavitating flow
N m
744.5
1178.4
1450.3
682.3
1149.7
1393.6
683.9
1149.7
1395.3
Case1 small flow rate

Case2 optimum
Case3 large flow rate
served in the results, the difference of 3% at optimum load Case

2 and that of 4% at overload Case 3 are regarded as tolerable in
the present study. One of the causes for the difference of 8% in
Case 1 may be due to the steady flow simulation since unsteadiness of the flow in the turbine must be large at part load.
Critical cavitation number. Figure 6 is the cavitation characteristics of hydraulic efficiency at another overload operating condition with a unit flow rate of Q11 = 0.99 m3 / s and a unit speed of
n11 = 83.4 rpm, where both numerical and experimental results are
shown for comparison. Similar to the runner torque in Table 4, the
hydraulic efficiency is predicted a little bit smaller than the experimental result. Based on the international code IEC193A-1972,
we determined the critical cavitation number c, which is defined
as the intersection of two approximate curves for = f, namely,
= const and = A + B. The predicted critical cavitation number
is 0.09 see Fig. 6, which is much closer to the experimental
value of 0.08. Thus, we will be able to evaluate that the steady
cavitating flow simulation is reasonably usable to predict the cavitation characteristics of a Francis turbine. To see the macroscopic
appearance of cavities in the runner, Fig. 7 shows those at typical
conditions, i.e., near the initial cavitation i and the severe condition. As suspected, appreciable cavities on the suction side near
the trailing edge of runner blades are predicted at the operating
condition near c.
Cavitated vortex rope. If the flow from the runner has strong
swirl, the cavitated vortex rope is observed in the draft tube under
the low condition. Using the steady flow analysis, the rope at
overload is investigated. To see the configuration of vortex rope in
the inlet cone of the draft tube, it is visualized from the numerical
results by using the criterion that the rope boundary is given by
the cavity volume fraction of 10%, which was proposed by Okita
and Kajishima 22. Figure 8 shows the result for overload of
130% Case 3. And a snapshot photograph taken at the model test
is shown for comparison. It is seen that the nearly straight vortex
rope appears from the runner cone to the draft tube, as shown in
Figs. 8a and 8b. From them, we will be able to say that fairly
well reproduction of cavitated vortex rope is achieved by the
present numerical simulation, though the numerical vortex rope
has a little bit larger diameter and shorter length than those of the
snapshot in Fig. 8a.
Unsteady cavitating flow at part load. It is known that large
pressure fluctuation caused by the cavitated vortex rope is sometimes observed in the draft tube of the Francis turbine at part load
operation, and it deteriorates the stable operation and reliability of
the turbine. As the vortex rope orbits in the draft tube depending
on time, unsteady flow analysis is essential to simulate the flow at
Table 3 Calculation conditions for three cases
Cases
Small flow rate
= 0.143
Optimum
= 0.144
Large flow rate
= 0.132
GV opening
mm
Head
m
Flow rate
m3 / s
Rotational speed
min1
Q11
m3 / s
n11
rpm
10
30
0.322
1122
0.466
75
15
30
0.497
1122
0.678
75
22
30
0.647
1122
0.882
75
MAY 2009, Vol. 131 / 051302-5
Fig. 6 Cavitation characteristics

= 0.99 m3 / s, n11 = 83.4 min1
of
efficiency
Q11
such operating points. To see the usability of the present numerical method for predicting the pressure fluctuation, Case 1 in Table
3 is treated as a typical example under the cavitation condition of
= 0.07. Here, the time step of 0.001 s was adopted in the unsteady turbulent flow analysis. Thus, the time step corresponds to
the runner rotation of 6.72 deg, as the rotational speed of the
runner is 1122 min1. Following IEC-60193, peak-to-peak amplitude and fundamental frequency of pressure fluctuation are evaluated at two survey points, as shown in Fig. 5c. One of them is
expressed as the downstream point 0.3D1, and the other is the
downstream point 1.0D1. They are located at the distances of 0.3
and 1.0 times the runner inlet diameter from the runner exit,
respectively.
The predicted results of dimensionless peak-to-peak amplitude
Fig. 7 Appearance of cavity in the runner: a initial cavitation

i = 0.157 and b severe cavitation = 0.08
Fig. 9 Wall pressure fluctuation at the 1.0D1 point for Case 1

from unsteady cavitating flow analysis = 0.07: a variation
in pressure with time and b frequency spectrum
H / H and the fundamental frequency are listed in Table 5, where

the test data and those results from single-phase flow simulation
are added for comparison. And the pressure in time domain and
that of frequency domain for the point 1.0D1 are shown in Figs.
9a and 9b, respectively. The frequency of test data exactly
corresponds to the rotating frequency of vortex rope in the draft
tube, though it is outside the 0.22 0.31n / 60 range of draft tube
surging. And the amplitude decreases toward downstream. These
features are qualitatively reproduced in the numerical results. It is
seen that much better prediction is achieved in the results being
considered the cavitating flow.
Conclusions
From the numerical study on the cavitating flow in a Francis

turbine, the following conclusions are drawn.
Fig. 8 Cavitated vortex rope in draft tube at large flow rate

Case 3 under = 0.054: a snapshot and b cavitating flow
simulation
1 The critical cavitation number c and hydraulic performance of a Francis turbine are reasonably predicted by the
steady cavitating flow simulation based on a mixture
model.
2 The cavitated vortex rope in the draft tube at overload operation was reasonably displayed from the numerical re-
Table 5 Pressure fluctuation at partial load operation Case 1 at = 0.07

Test data
Single-flow calculation
Cavitating flow calculation
Survey point
Amp. H / H
%
Fundamental
freq. Hz
Amp. H / H
%
Fundamental
freq. Hz
Amp. H / H
%
Fundamental
freq. Hz
0.3D1
1.0D1
3.91
3.66
3.43
3.42
4.32
4.1
4.98
4.98
4.25
3.2
4.12
4.12
051302-6 / Vol. 131, MAY 2009
sults by using the Okita and Kajishima 22 criterion that

the rope boundary is given by the cavity volume fraction of
10%.
3 From a qualitative viewpoint, pressure fluctuation caused
by the cavitated vortex rope at part load operation is reasonably predicted by the present numerical method for unsteady cavitating turbulent flow.
Acknowledgment
The research work was funded by Chinese National Foundation
of Natural Science Contract No. 90410019.
Nomenclature
i
c
ji
correction coefficients
gravitational acceleration
test head acting on the turbine
atmospheric pressure head
suction head of the turbine
saturated vapor pressure head
vacuum head in the draft tank
kinetic energy of turbulence
Moore molecule weight
rotational speed
unit speed, =nD / H
local pressure
flow rate
unit flow rate, =Q / D2H
gas constant
radius of the bubble
mass transfer rate
temperature
time
velocity
coordinates
volume fraction
surface tension
turbulent kinetic energy dispassion rate
efficiency
viscosity
density
cavitation number, see Eq. 12
initial cavitation number
critical cavitation number
stress
Subscripts
ca
l
u
v
cavity phase
liquid
undissolved gas
vapor
C 1, C 2
g
H
Ha
HS
HV
HVA
k
M
n
n11
p
Q
Q11
R
r
S
T
t
u
x,y,z
Appendix: Experimental Uncertainty

The experiment of the turbine performances was carried out in
the high accurate test rig in Harbin Large Electric Machinery Institute in China. The flow parameters were measured in the following measuring range and the accuracy, as shown in Table 6.
Table 6 Measurement accuracy 99%

No.
Performance
Unit
Measuring range
Uncertainty
1
2
3
4
Flow rate Q
Head H
Force on arm K
Angular speed
m3 / s
MPa
E pound
rad/s
01.0
02.07
01000
03000
0.20%
0.075%
0.02%
0.05%
References
1 Susan-Resiga, R. F., Muntean, S., and Anton, I., 2002, Numerical Analysis of
Cavitation Inception in Francis Turbine, Proceedings of the 21st IAHR Symposium on Hydraulic Machinery and Systems, F. Avellan, G. Ciocan, and S.
Kvicinsky, eds., Lausanne, Switzerland, pp. 431438.
2 Yamaguchi, H., and Kato, H., 1983, On Application of Nonlinear Cavity
Flow Theory to Thick Foil Section, Proceedings of the Second International
Conference of Cavitation, ImechE, Paper No. C209/83, pp. 167174.
3 Brewer, W. H., and Kinnas, S. A., 1995, Experimental and Computational
Investigation of Sheet Cavitation on a Hydrofoil, Proceedings of the Second
Joint ASME/JSME Fluids Engineering Conference and ASME/EALA Sixth
International Conference on Laser Anemometry, Hilton Head Island, SC, pp.
115.
4 Pellone, C., and Peallat, J. M., 1995, Non-Linear Analysis of ThreeDimensional Partially Cavitating Hydrofoil, Proceedings of the International
Symposium on Cavitation, Deauville, France, pp. 6367.
5 De Lange, D. F., and De Bruin, G. J., 1998, Sheet Cavitation and Cloud
Cavitation Re-Entrant Jet and Three-Dimensionality, Appl. Sci. Res., 581
4, pp. 91114.
6 Watanabe, S., Hidaka, T., Horiguchi, H., Furukawa, A., and Tsujimoto, Y.,
2007, Steady Analysis of the Thermodynamic Effect of Partial Cavitation
Using the Singularity Method, ASME J. Fluids Eng., 1292, pp. 121127.
7 Chen, Y., and Heister, S. D., 1994, A Numerical Treatment for Attached
Cavitation, ASME J. Fluids Eng., 116, pp. 613618.
8 Ventikos, Y., and Tzabiras, G., 2000, A Numerical Method for the Simulation
of Steady and Unsteady Cavitating Flows, Comput. Fluids, 29, pp. 6388.
9 Horiguchi, H., Arai, S., Fukutomi, J., Nakase, Y., and Tsujimoto, Y., 2004,
Quasi-Three-Dimensional Analysis of Cavitation in an Inducer, ASME J.
Fluids Eng., 126, pp. 709715.
10 Arndt, R. E. A., Kjeldsen, M., Song, C. C. S., and Keller, A., 2002, Analysis
of Cavitation Wake Flows, Proceedings of the 21st IAHR Symposium on
Hydraulic Machinery and Systems, F. Avellan, G. Ciocan, and S. Kvicinsky,
eds., Lausanne, Switzerland, pp. 395402.
11 Qin, Q., Song, C. S. S., and Arndt, R. E. A., 2003, Numerical Study of
Unsteady Turbulent Wake Behind a Cavitating Hydrofoil, Fifth International
Symposium on Cavitation, Osaka, Japan, Paper No. EM 003.
12 Deshpande, M., Feng, J., and Merkle, C. L., 1997, Numerical Modeling of the
Thermodynamic Effects of Cavitation, ASME J. Fluids Eng., 119, pp. 420
427.
13 Rieger, R., 1992, Mehrdimensionale Berechnung Zweiphasiger Stroemungen, Ph.D. thesis, Technical University Graz, Graz, Austria.
14 Grogger, H. A., and Alajbegovic, A., 1998, Calculation of the Cavitating
Flow in Venturi Geometries Using Two Fluid Model, Proceedings of the
ASME Fluids Engineering Division Summer Meeting, Washington, DC, Paper
No. FEDSM98-5295.
15 Liu, S. H., Wu, Y. L., and Luo, X. W., 2005, Numerical Simulation of 3D
Cavitating Turbulent Flow in Francis Turbine, Proceedings of the ASME
FEDSM, Houston, TX, Paper No. FEDSM2005-77017.
16 Singhal, A. K., Vaidya, N., and Leonard, A. D., 1997, Multi-Dimensional
Simulation of Cavitating Flows Using a PDF Model for Phase Change, Proceedings of the ASME Fluids Engineering Division Summer Meeting, Vancouver, BC, Vol. 4, pp. 18.
17 Kunz, R. F., Boger, D. A., Stinebring, D. R., Chyczewski, T. S., Lindau, J. W.,
Gibeling, H. J., Venkateswaran, S., and Govindan, T. R., 2000, A Preconditioned Navier-Stokes Method for Two-Phase Flows With Application to Cavitation Prediction, Comput. Fluids, 29, pp. 849875.
18 Senocak, I., and Shyy, W., 2002, A Pressure-Based Method for Turbulent
Cavitating Flow Computation, J. Comput. Phys., 176, pp. 363383.
19 Singhal, A. K., Athavale, M. M., Li, H. Y., and Jiang, Y., 2002, Mathematical
Basis and Validation of the Full Cavitation Model, ASME J. Fluids Eng.,
1243, pp. 617624.
20 Aschenbrenner, T., Otto, A., and Moser, W., 2006, Classification of Vortex
and Cavitation Phenomena and Assessment of CFD Prediction Capabilities,
Proceedings. of the 23rd IAHR Symposium on Hydraulic Machinery and Systems, J. Kurokawa, ed., Yokohama, Japan, Paper No. F132.
21 Saito, Y., Takami, R., Nakamori, I., and Ikohagi, T., 2007, Numerical Analysis of Unsteady Behavior of Cloud Cavitation Around a NACA0015 Foil,
Comput. Mech., 40, pp. 8596.
22 Okita, K., and Kajishima, T., 2002, Numerical Simulation of Unsteady Cavitating Flow Around a Hydrofoil, Trans. Jpn. Soc. Mech. Eng., Ser. B, 68, pp.
637644.
23 Guo, Y., Kato, C., and Miyagawa, K., 2006, Large-Eddy Simulation of NonCavitating and Cavitating Flow in Venturi Geometries Using Two Fluid
Model, Proceedings of the 23rd IAHR Symposium on Hydraulic Machinery
and Systems, J. Kurokawa, ed., Yokohama, Japan, Paper No. F195.
24 Lindau, J. W., Kunz, R. F., Boger, D. A., Stinebring, D. R., and Gibeling, H. J.,
2002, High Reynolds Number, Unsteady, Multiphase CFD Modeling of Cavitating Flows, ASME J. Fluids Eng., 1243, pp. 607616.
25 Medvitz, R. B., Kunz, R. F., Boger, D. A., Lindau, J. W., and Yocum, A. M.,
2002, Performance Analysis of Cavitating Flow in Centrifugal Pumps Using
Multiphase CFD, ASME J. Fluids Eng., 1242, pp. 377383.
26 Fortes-Patella, R., Coutier-Delgosha, O., Perrin, J., and Reboud, J. L., 2007,
Numerical Model to Predict Unsteady Cavitating Flow Behavior in Inducer
Blade Cascades, ASME J. Fluids Eng., 1292, pp. 128135.
27 Xing, T., Li, Z. Y., and Frankel, S. H., 2005, Numerical Simulation of Vortex
Cavitation in a Three-Dimensional Submerged Transitional Jet, ASME J. Flu-
MAY 2009, Vol. 131 / 051302-7
ids Eng., 1274, pp. 714725.

28 Lu, T. S., Samulyak, R., and Glimm, J., 2007, Direct Numerical Simulation of
Bubbly Flows and Application to Cavitation Mitigation, ASME J. Fluids
Eng., 129, pp. 595604.
29 Liu, S. H., Li, S. C., Zhang, L., and Wu, Y. L., 2008, A Mixture Model With
Modified Mass Transfer Expression for Cavitating Turbulent Flow Simulation, Eng. Comput., 25, pp. 290304.
30 Cammenga, H. K., 1980, Evaporation Mechanisms of Liquids, Curr. Top.
Mater. Sci., 5, pp. 335446.
31 Vanka, S. P., 1986, Block-Implicit Multigrid Solution of Navier-Stokes Equations in Primitive Variables, J. Comput. Phys., 651, pp. 138158.
051302-8 / Vol. 131, MAY 2009
32 Coutier-Delgosha, O., Fortes-Patella, R., and Reboud, J. L., 2003, Evaluation

of the Turbulence Model Influence on the Numerical Simulations of Unsteady
Cavitation, ASME J. Fluids Eng., 1251, pp. 3845.
33 Speziale, C. G., and Thangam, S., 1992, Analysis of an RNG Based Turbulence Model for Separated Flows, Int. J. Eng. Sci., 3010, pp. 13791388.
34 Lam, S. H., 1992, On the RNG Theory of Turbulence, Phys. Fluids, 4, pp.
10071017.
35 Dular, M., Bachert, R., Stoffel, B., and irok, B., 2005, Experimental Evaluation of Numerical Simulation of Cavitating Flow Around Hydrofoil, Eur. J.
Mech. B/Fluids, 24, pp. 522538.
R. Asmatulu
Wichita State University,
1845 Fairmount,
Wichita, KS 67260-0133
S. Kim
F. Papadimitrakopoulos
Nanomaterials Optoelectronics Laboratory,
Polymer Program,
Institute of Materials Science,
University of Connecticut,
Storrs, CT 06269-3136
H. Marcus
Department of Materials Science and
Engineering,
Institute of Materials Science,
University of Connecticut,
Storrs, CT 06269-3136
Parallel-Plate Conductive
Electrodes for the Fabrication of
Larger 2D Colloidal Photonic
Crystals
A new dielectrophoretic force-induced parallel-plate assembly technique was used to
achieve close-packed 2D large colloidal photonic crystals on gold electrodes (200 nm
thick). The electrodes were patterned on a glass substrate using a conventional UV
lithography technique. The experimental tests conducted with 5.3 m carboxyl functionalized polystyrene particles at various ac and dc voltages, frequencies, and particle
concentrations showed that larger size 0.25 3 mm2 colloidal photonic crystals were
fabricated on the ground electrode rather than on the working electrode. To date, this is
the largest colloidal photonic crystal fabricated using this method. The reason behind this
phenomenon can be attributed to the electro-osmotic flow in the colloidal system and
dipole-dipole attractions between the colloidal particles. DOI: 10.1115/1.3111257
Keywords: parallel-plate gold electrodes, colloidal PS particles, 2D photonic crystals
Introduction
Photonic crystals PCs, with periodicity comparable to the

wavelength of light, have been considered one of the most attractive optical materials in a wide range of scientific arenas because
of their unique property known as photonic band gap PBG. This
property forbids light over a range of wavelengths at certain frequencies and permits other light in the crystal 15. These materials could be used in many areas: high reflectance mirrors, narrow lossless waveguides with sharp corners, zero-threshold lasers,
and superprisms, as well as optical telecommunication, switches,
and computing 610. The integration of photonic crystals will
allow a large number of components to be produced from a single
crystal for industrial applications 1114.
Based on selected materials e.g., photoresists, polymeric films,
dielectric rods, symmetric holes, and colloidal particles and crystal dimensions, a number of fabrication techniques have been proposed to achieve photonic crystals. These techniques include lithography, chemical etching, gravity sedimentation, vertical
depositions, LangmuirBlodgett LB, centrifugation, convective
self-assembly, electrophoresis, and dielectrophoresis DEP. There
are a number of advantages of DEP in micro- and nanofabrication
systems over the other methods: i Higher and lower electric field
strengths and gradients can be achieved with a lower potential, ii
electrodes can be integrated in a small circuit without any dissociation at various frequencies, and iii higher electric field can be
applied without creating an excessive heat 15,16. In addition to
these, DEP is a simple, inexpensive, practical, and rapid assembly
technique, and has great potential for the future fabrication of
colloidal PCs in the photonic industry 812.
The phenomenon of DEP involves the creation of electric
forces on particles to generate momentum in nonuniform electric
fields, usually coming from ac electric fields 1315. The direction and magnitude of the forces rely on several factors, including
applied voltage and frequency, polarizability of the particle, conContributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received April 20, 2008; final manuscript received September 22, 2008; published online April 14, 2009. Assoc. Editor:
Theodore Heindel.
ductivity, and permittivity of both medium and particles, as well

as dielectric properties of the particles and the medium. DEP
forces can be utilized to characterize, handle, and manipulate colloidal photonic materials for the purposes of sorting, trapping,
aligning, and separating. Additionally, other micro- and nanoscale
organic cells, viruses, bacteria, and DNA and inorganic nanoparticles, nanotubes, and nanowires objectives can be directed
using the same method 1723.
Although significant progress has been made during the past
decades for increasing the fabrication of colloidal photonic crystals in different dimensions, the problem of fabricating larger size
PCs has not been solved yet. Most of the 2D colloidal photonic
crystals produced to date are below 2 mm in diameter 5,24,25.
Therefore, the lack of efficient fabrication processes is forcing
many government agencies e.g., ARMY and private companies
to find novel solutions in these areas. This paper presents the
fabrication of larger size 2D colloidal photonic crystals on
parallel-plate conductive electrodes using a 5.3 m polystyrene
PS at various ac and dc voltages, frequencies, and solid contents.
Theory
Dielectrophoretic force relies on the difference in the polarizability of the system compared with its surrounding fluid medium
e.g., water. Consider an isolated/dispersed spherical colloidal
particle or biological cell with known diameter and dielectric constant in a liquid medium. Under the nonuniform electric field, the
induced dipole moment drives the particle in a direction. The dielectrophoretic force acting on the spherical body is expressed as
13
FDEP = 2m ReKr3 Erms2
where r is the sphere radius, m is the permittivity of the suspending medium, Erms is the gradient of the root mean square electric
field, and ReK is the real part of the ClausiusMossotti factor
given by
MAY 2009, Vol. 131 / 051401-1
Fig. 1 Schematic outline of the experimental procedure left utilized to assemble larger 2D colloidal photonic crystals on the parallel-plate gold electrodes right connected to a power supplier. The red line is the working electrode, while the dark line is the ground electrode on which the photonic
crystals were formed not to be scaled.
K =
p m
2m
and p are the complex permittivities of the medium and

where m
particle, respectively. The complex permittivity of the medium is
defined as
j
=
where is the conductivity, is the permittivity, and is the

angular frequency of the applied electric field 13. The direction
and intensity of the driving force acting on the particles depend on
the dielectric properties of the particles and the suspending fluid
medium, electrode configurations, applied voltage, and frequency.
The movement of particles toward the strong electric field gradient region is referred to as positive dielectrophoresis p-DEP and
toward the opposite direction as negative dielectrophoresis nDEP.
The dipoles induced in the particles of a suspension by either
field also interact with each other, resulting in a chaining force
Fchain, which depends on the field strength E. A generalized expression for the force between adjacent polarized particles is
given as 13
Fchain = C ReK2r2E2
In this expression, the coefficient C ranges between 3 and 1000,

depending on the distance among the particles and the length of
the particle chains.
ac electro-osmosis is a fluid flow that occurs when a nonuniform ac electric field is applied to the electrical double layer
EDL of particles, which forms as a result of the interaction of
the ionized solution with static charges on dielectric surfaces. The
interaction produces a steady flow over the microelectrode surface. In this process, charged particles tend to migrate toward
less-charged and turbulent areas. Thus, this phenomenon may possibly affect colloidal particles that are concentrated in specific
locations 2631.
The electro-osmotic velocity v prediction of ac electro-osmosis,
which is modeled using the EDL on the surface of parallel electrodes, is given as 3234
=
1 V202
8 a1 + 22
where V0 is the potential applied to the electrodes, is the viscosity of the electrolyte, and a is the distance from the center of
the electrode gap to the point of interest. In this model, the nondimensional frequency is given by
= r
where is the reciprocal Debye length of the EDL, the distance

from the wall where the electrokinetic potential energy becomes
051401-2 / Vol. 131, MAY 2009
equal to the thermal energy. The fluid motion caused by ac

electro-osmosis varies, based on the geometry of the parallel electrodes, and can be numerically calculated using the equations
above.
Experimental
3.1 Material. A sample of 5.3 m carboxyl functionalized

polystyrene particles was purchased from Polysciences, Inc. Warrington, PA. This monodispersed, negatively charged colloidal
sample was diluted to desired concentrations using de-ionized
DI water Barnstead Nanopure Water System, http://www.
thermo.com/com/cda/product/detail/1,,10131329,00.html,
and
then the pH values of the dispersions, determined using a highly
sensitive pH meter, were adjusted to pH 8 with the addition of a
few drops of triethylamine Sigma-Aldrich, http://www.
sigmaaldrich.com/sigma-aldrich/home.html solution before the
tests. The reason for using a slightly alkaline medium was to
decrease nonspecific binding between particles and substrates. All
samples prepared in our facility were refrigerated until used.
3.2 Method. Parallel-plate gold electrodes were patterned using a conventional photolithography technique followed by gold
deposition plasma sputtering. In the surface-patterning step, a
photoresist S1813 Shipley http://www.microchem.com/products/
pmma.htm positive type was spin-coated on a clean glass substrate 5 7.5 cm2 at 5000 rpm for 30 s before 2 min of softbaking at 115 C. An UV light was exposed with the desired mask
for 40 s at 180 W / cm2, and then the sample was soft-baked
again at 115 C for 1 min. After the developing process was completed using a Shipley MF 351 at a 1:3.5 ratio with water, a thin
layer of gold 200 nm was evaporated on the surface using a
plasma-sputtering unit. In the final step, the photoresist was
washed out with acetone in a sonicator for 10 min to create the
parallel-plate gold electrodes. The width and length of the parallel
plates were 280 m and 3 mm, respectively. Schematic views of
the electrode and experimental setup are shown in Fig. 1 not to
scale.
Throughout the experiments, one side of the parallel plates was
connected to the working electrode ac, dc, or both where the
electric current was applied, whereas the other side was connected
to the ground electrode. This geometry allowed for an electrically
driven assembly of 2D colloidal crystals resulting in a long-range
ordering. The electrical fields were applied to the PS particles in
an aqueous media of a fluid chamber separated by a rubber spacer
0.75 mm thickness after the particles had completely settled on
the surface of the substrate. Tests were performed on an optical
table using an optical microscope equipped with a charge-coupled
device CCD camera, close-loop TV, sealed microfluidic device
made of Plexiglas, function generator, and oscilloscope to monitor the sine waves, ac and dc voltages, and frequency levels. During the tests, particle settlement and crystallization times were
optimized at 30 min and 10 min, respectively.
Fig. 2 Photographs showing a particles settled after 30 min,

and crystallized forms at b 0.1 dc V and c 0.5 dc V on top of
the ground electrodes. Note that the white parts show the gold
substrate surface, while the dark stripes show the glass substrate with the PS particles decorating the gold surfaces.
4.1 Effect of dc Voltage. In order to determine the effects of

dc bias voltage on the crystallization of colloidal photonic crystals, a number of tests were conducted with the 5.3 m PS particles. Figure 2 shows the test results performed at 0.4% solid
content, pH 8, and 10 min of crystallization time. The particles
settled on the ground electrodes Fig. 2a were barely gathered
on the ground electrode at 0.1 dc V Fig. 2b. However, when
the voltage was increased to 0.5 V, localized crystallization of
photonic crystals was gradually enhanced. Note that similar effects were also noticed using negative dc charges throughout the
tests. Meanwhile, all the particles that originally settled on the
working electrode started vibrating Fig. 2c. At higher dc voltages over 0.5 V, it was observed that the working electrodes
were debonded from the glass substrate.
In the second step of the experiments, we optimized testing
conditions using dc voltage only. Figure 3 shows the gold electrode Fig. 3a and electrophoretic test results conducted on the
same particles at 0.5% solid content and 0.4 dc V. The PS particles
were initially settled Fig. 3b and crystallized on the ground
electrode to form randomly oriented 2D multicrystals Fig. 3c.
Once again, it was determined that regardless of the sign and
intensity of the applied dc voltage, PS particles were formed on
the ground electrodes, which can be attributed to the long-range
dipole-dipole interactions. However, the other particles on the
working electrode were not collected and crystallized, and mostly
Fig. 3 Photographs showing a parallel-plate gold electrode

white parts patterned on a glass substrate dark parts, b
particles settled on the surface, and c colloidal particle localized crystallization at 0.4 dc V
Fig. 4 Photographs showing a larger colloidal photonic

crystal collected on the electrodes and b crystallization at 2
ac V, 1 MHz frequency, 0.5% solid content, and pH 8
tended to remain at their original positions. Additionally, particles

between the electrodes were not affected much by the applied
voltage.
4.2 Effect of ac Voltage. ac dielectrophoretic tests were performed on the PS particles at various ac voltages 0.14 V, frequencies 10 kHz10 MHz, and solid contents 0.30.5%. Figure 4 shows the fabrication of 2D colloidal crystals at 0.4% solid
content, 2 ac V, and 1 MHz frequency after 10 min of crystallization time. These parameters were the optimum conditions obtained for the ac tests. Test results confirmed that the initially
settled PS particles were laterally squeezed under the ac voltage
and became a larger multidomain PC on top of the ground electrode. No particles were left on both edges of the ground electrodes. In contrast, PS particles on the working electrode were not
crystallized and started vibrating in the same manner as during the
dc voltage tests. The crystallization of particles under the ac voltage can be due to the electro-osmosis flow see Eq. 5 and longrange dipole-dipole attractions on the PS particles 2631.
Parallel-plate DEP tests were also performed on 0.2% solid
content of 5.3 m PS particles at 2 ac V and 1 MHz to further
determine the concentration effects on the accumulation of the
colloidal photonic crystals. Figure 5 shows the experimental test
Fig. 5 Photographs illustrating a PS particles settled on the

electrodes; b parallel-plate DEP tests conducted at 2 ac V, 1
MHz, pH 8.0, and 0.2% solid content; and c formation of pearl
chains between the electrodes and second layer on the first
layer of the crystal on the ground electrode at 4 ac V and 0.5%
solid content
MAY 2009, Vol. 131 / 051401-3
Fig. 6 Crystallization of larger colloidal photonic particles a

after settlement on the electrodes under b 0.4 dc V, and c
both 0.4 dc and 3 ac V with 1 MHz. This is the largest known
colloidal PC to date.
results on the PC fabrication at a lower solid content. Test results

illustrate that the PS particles that settled on the electrodes Fig.
5a showed less tendency to crystallize on the ground electrodes.
This may be because of the lack of lateral forces acting on the PS
particles, or the higher liquid turbulence and Brownian motion at
lower concentrations.
The other interesting finding during the tests was that at higher
ac voltages and solid content, a second layer of colloidal particles
collected on top of the first crystallized layer, which may be because of the higher lateral forces acting on the particles at these
testing conditions. Figure 5c shows the second layer formation
at 4 ac V and 0.5% solid content. Additionally, a chaining effect
pearl chains of PS particles between the working and ground
electrodes see Eq. 4 was clearly observed in the test, which
may be due to the higher polarity of ac voltage applied to the
electrodes and PS particles.
4.3 Effect of Both ac and dc Voltages. For the fabrication of
larger colloidal photonic crystals, varying ac and dc fields were
applied together on the PS particles with 0.5% solid content. Figure 6 shows the test results obtained at 0.4 dc V Fig. 6b and
0.4 dc+ 3 ac V with 1 MHz Fig. 6c. The test results clarified
that the colloidal particles were initially crystallized on the ground
electrode using dc bias voltage, and then small domains became a
larger multidomain about 250 m in width and 3 mm in length
after the application of 3 ac V and 0.4 V dc together. To the best
of our knowledge, this is the largest multicrystalline colloidal PC
produced using this kind of methods. Figure 6 also shows that
there are a few point defects missing particles and second layers
of particles on the crystals, which may be a result of lateral forces
on the edges of the electrodes. Particles on the working electrode
were again lifted from the surface, while the other particles in
between neighboring electrodes formed pearl chains under the
ac/dc voltage.
After the experimental studies were completed, the fabrication
Fig. 7 Photograph illustrating the remaining colloidal particles

in mostly crystalline form after washing several times with DI
water
device was washed out several times using DI water. However, the
evidence indicated that some of the colloidal particles still remained on the ground electrodes. The reason for this may be
attributed to the excess amount of current/charge on the electrodes
and particles, which created a long-range attraction force to hold
the particles together. Figure 7 shows the crystalline PS particles
on the ground electrode photographed after washing with DI water.
4.4 Schematic Illustration of Crystallization Process. In the
electro-osmosis process high-velocity streamlines on the edge of
the electrodes, the charged particles tend to migrate toward the
lesser charged and turbulence areas as is specified in Eq. 5
2830. Thus, this phenomenon may possibly affect the colloidal
particles concentrated in specific locations. In the presence of
electrically charged ions in an aqueous system, the ions possibly
transfer onto the surface of the PS particles and increase the longrange dipole-dipole particle-particle interactions for the crystallization of the polarized particles 31. Based on the test results, it
is believed that this can be one of the major mechanisms of the
colloidal particle accumulation and crystallization on the ground
electrodes. Additional reasons may be electrohydrodynamic, electric field gradient, and thermal effects on the particles 2630.
Figure 8 depicts the possible schematic views of the larger PC
particle crystallization on the ground electrode using dc and ac
voltages.
Crystallization was not achieved on the working electrode,
which could be because of induced charges greatly increasing the
Fig. 8 Schematic illustrations of larger colloidal particle crystallization under

dc followed by dc/ac electric fields
051401-4 / Vol. 131, MAY 2009
fluid flow on the crake, channel, bump surfaces, and/or sharp

edges of the working electrodes, as well as the pulsing of the
electric field strength in the system. Excess charges may also enhance the electrostatic repulsive forces between the particles on
the working electrodes. In order to determine the actual reasons
for the larger colloidal particle crystallization on the ground electrodes, further investigation using new techniques and approaches
can be utilized.
Conclusions
The parallel-plate DEP tests were conducted on carboxyl functionalized PS colloidal particles 5.3 m to create larger 2D colloidal crystals. Test results revealed that colloidal particles could
be easily formed into closely packed larger colloidal crystals on
the ground electrodes for various dc and/or ac voltages. As is
known, the DEP technique is a simple, inexpensive, and more
rapid assembly technique compared with other methods. Therefore, this technique can have great potential for the photonic crystal fabrication, as well as many other biological, organic, and inorganic particle manipulations in the future.
Acknowledgment
The authors wish to acknowledge AFOSR Contract No.
F49620-01-1-0545 and ONR Contract No. N00014-00-1-0333
for their financial support.
References
1 Yablonovitch, E., 2001, Photonic Crystals: Semiconductors of Light, Sci.
Am., 2856, pp. 4755.
2 Johnson, S. G., and Joannopoulos, J. D., 2002, Photonic Crystals: The Road
From Theory to Practice, 1st ed., Kluwer, Dordrecht.
3 Park, S. H., and Xai, Y., 1999, Assembly of Mesoscale Particles Over Large
Areas and Its Application in Fabricating Tunable Optical Filters, Langmuir,
15, pp. 266273.
4 Kitaev, V., and Ozon, G. A., 2003, Self-Assembled Surface Patterns of Binary Colloidal Crystals, Adv. Mater. Weinheim, Ger., 15, pp. 7578.
5 Lumsdon, S. O., Kaler, E. W., and Volve, O. D., 2004, Two-Dimensional
Crystallization of Microspheres by a Coplanar AC Electric Field, Langmuir,
20, pp. 21082116.
6 Trau, M., Sankaran, S., Saville, D. A., and Aksua, I., 1995, Electric-FieldInduced Patterning of Colloidal Dispersions, Nature London, 374, pp. 437
439.
7 Mclachlan, M. A., Johnson, P. N., De La Rue, R., and McComb, D. W., 2004,
Thin Film Photonic Crystals: Synthesis and Characterization, J. Mater.
Chem., 14, pp. 144150.
8 Hamagami, J. H., Hasegawa, K., and Kanamura, K., 2004, Assembly of
Monodispersed Silica Spheres by Micro-Electrophoretic Deposition Process,
J. Ceram. Soc. Jpn., 12, pp. 169172.
9 Ahn, J. S., Hammond, P. T., Rubner, M. F., and Lee, I., 2005, Self-Assembled
Particle Monolayers on Polyelectrolyte Multiplayer: Particle Size Effects on
the Formation, Structure and Optical Properties, Colloids Surf., A, 259, pp.
4551.
10 Dziomkina, N. V., Hempenius, M. A., and Vancso, G. J., 2005, Symmetry
Control of Polymer Colloidal Monolayers and Crystals by Electrophoretic
Deposition Onto Patterned Surfaces, Adv. Mater. Weinheim, Ger., 172,
pp. 237240.
11 Mekis, A., Chen, J. C., Kurland, I., Fan, S., Villeneuve, P. R., and Joannopoulos, J. D., 1996, High Transmission Through Sharp Bends in Photonic Crys-
tal Waveguides, Phys. Rev. Lett., 7718, pp. 37873790.

12 John, S., 1987, Strong Localization of Photons in Certain Disordered Dielectric Superlattices, Phys. Rev. Lett., 58, pp. 24862489.
13 Pohl, H. A., 1978, Dielectrophoresis, Cambridge University Press, Cambridge.
14 Morgan, H., and Green, N. G., 2003, AC Electrokinetics: Colloids and Nanoparticles, Research Studies, Baldock, UK.
15 Kawabata, T., and Washizu, M., 2001, Dielectrophoretic Detection of Molecular Bindings, IEEE Trans. Ind. Appl., 376, pp. 16251633.
16 Kawabata, T., and Washizu, M., 1999, Separation of Biomolecules Using
Dielectrophoretic Chromatography, Trans. Inst. Electr. Eng. Jpn., Sect. E,
119-E10, pp. 454459.
17 Suzuki, M., Yasukawa, T., Mase, Y., Oyamatsu, D., Shiku, H., and Matsue, T.,
2004, Dielectrophoretic Micropatterning With Microparticle Monolayers Covalently Linked to Glass Surfaces, Langmuir, 2025, pp. 1100511011.
18 Lumsdon, S. O., Williams, J. P., Kaler, E. W., and Volve, O. D., 2003, Dielectrophoretic Assembly of Oriented and Switchable Two-Dimensional Photonic Crystals, Appl. Phys. Lett., 82, p. 949.
19 Abe, M., Orita, M., Yamazaki, H., Tsukamoto, S., Teshima, Y., Sakai, T.,
Ohkubo, T., Momozawa, N., and Sakai, H., 2004, Three-Dimensional Arrangement of Polystyrene Latex Particles With a Hyperbolic Quadruple Electrode System, Langmuir, 20, pp. 50465051.
20 Asmatulu, R., Kim, S., Bright, R., Yu, P., Papadimitrakopoulos, F., and Marcus, H., 2006, Fabrication and Defect Designs on DNA Linked 2D Colloidal
Photonic Crystals Using a Nd:YAG Pulsed Laser, Mater. Res. Soc. Symp.
Proc., 901E, pp. Ra2102.1-6.
21 Geiss, E. P., 2003, Fabrication and Defect Design in Two-Dimensional Colloidal Photonic Crystals, Ph.D. thesis, University of Connecticut, Storrs, CT.
22 Yu, P., 2005, Controlled Defect Insertion in 2-D Photonic Crystals Using a
Nd:YAG Laser Harmonic at 532 nm, MS thesis, University of Connecticut,
Storrs, CT.
23 Alvarado-Rodriguez, I., 2003, Fabrication of Two-Dimensional Photonic
Crystal Single-Defect Cavities and Their Characterization by Elastic Scattering, Ph.D. thesis, University of California, Los Angeles, CA.
24 Blaaderen, A. V., and Wiltzius, P., 2004, Growing Large, Well-Oriented Colloidal Crystals, Adv. Mater. Weinheim, Ger., 10, pp. 833835.
25 Gupta, M. C., and Ballato, J., 2007, The Handbooks of Photonics, CRC, Boca
Raton, FL.
26 Morgan, H., Izquierdo, A. G., Bakewell, D., Green, N., and Ramos, A., 2001,
The Dielectrophoretic and Traveling Wave Forces Generated by Interdigitated
Electrode Arrays: Analytical Solution Using Fourier Series, J. Phys. D, 34,
pp. 15531561.
27 Green, N., Ramos, A., Gonzales, A., Morgan, H., and Castellanos, A., 2002,
Fluid Flow Induced by Nonuniform ac Electric Fields in Electrolytes on
Microelectrodes. III. Observation of Streamlines and Numerical Simulation,
Phys. Rev. E, 66, p. 026305.
28 Green, N. G., Ramos, A., and Morgan, H., 2002, Numerical Solution of the
Dielectrophoretic and Traveling Wave Forces for Interdigitated Electrode Arrays Using the Finite Element Method, J. Electrost., 56, pp. 235254.
29 Ramos, A., Gonzales, A., Castellanos, A., Green, N., and Morgan, H., 2003,
Pumping of Liquids With ac Voltage Applied to Asymmetric Pairs of Microelectrodes, Phys. Rev. E, 67, p. 056302.
30 Bhatt, K. H., Grego, S., and Velev, O. D., 2005, An AC Electrokinetic Technique for Collection and Concentration of Particles and Cells on Patterned
Electrodes, Langmuir, 2114, pp. 66036612.
31 Adamson, A. W., and Gast, A. P., 1997, Physical Chemistry of Surfaces, 6th
ed., Wiley, New York.
32 Ramos, A., Morgan, H., Green, N. G., and Castellanos, A., 1999, AC
Electric-Field Induced Fluid Flow in Microelectrodes, J. Colloid Interface
Sci., 217, pp. 420422.
33 James, C. D., Okandan, M., Galambos, P., Mani, S. S., Bennett, D., Khusid, B.,
and Acrivos, A., 2006, Surface Micromachined Dielectrophoretic Gates for
the Front-End Device of a Biodetection System, ASME J. Fluids Eng.,
1281, pp. 1419.
34 Hughes, M. P., 2000, AC Electrokinetics: Applications for Nanotechnology,
Nanotechnology, 11, pp. 124132.
MAY 2009, Vol. 131 / 051401-5
Comparison of Turbulence
Modeling Strategies for Indoor
Flows
Ammar M. Abdilghanie
Lance R. Collins
David A. Caughey1
e-mail: dac5@cornell.edu
Sibley School of Mechanical & Aerospace
Engineering,
Cornell University,
Ithaca, NY 14853-7501
Turbulence modeling techniques are compared for the simulation of low speed indoor air
flow in a simple room. The effect of inlet turbulence intensity on the flow field is investigated using the constant coefficient large eddy simulation (LES) model with uniform
mean inlet conditions at several levels of inlet turbulence intensities. The results show
significant differences between the simulations with laminar inflow conditions and those
in which turbulence was introduced at the inlet. For simulations with turbulent inlet
conditions, it is noticed that the jet transitions to a state of fully developed turbulence
wherein the dynamics of the flow become nearly insensitive to any further increase in the
level of inlet turbulence. For laminar flow conditions, it is seen that the jet slowly spreads
and mixes with the quiescent room air. As a result, the jet reaches a fully developed
turbulent state further away from the inlet relative to the simulations with inlet turbulence. The effect of using experimental inlet profiles is also investigated. It is seen that,
close to the inlet, the flow is sensitive to the inflow details, whereas further away from the
inlet, these effects become less pronounced. The results from the constant coefficient and
the dynamic LES models are compared. The most noticeable differences in the flow occur
at the locations where the subgrid-scales contribution to the turbulent kinetic energy is
highest. Finally, the results from the dynamic LES and the k- models are compared. It is
found that there are significant differences between the two models for the zero inlet
turbulence limit where the flow is most probably transitional in nature and turbulence has
not yet reached a fully developed state. It is seen that in the laminar inflow case the k-
model predicts a fully turbulent jet very close to the inlet and thus fails to capture the
slow development of the jet found in LES. Accordingly, the k- model results are nearly
insensitive to the level of inlet turbulence especially far from the origin of the flow. It is
also seen that for cases with nonzero inlet turbulence level, the k- model predicts the
general features of the mean flow reasonably well; however, the k- model overpredicts
the jet spreading rate and the turbulent kinetic energy close to the inlet. Furthermore, the
k- model under predicts the turbulence level near the corner of the ceiling as it fails to
capture the complicated mean velocity and turbulent kinetic energy, most likely because
of the highly intermittent flow pattern found there in LES. DOI: 10.1115/1.3112386
Introduction
The vast majority of ventilation systems are designed assuming

the pollutants and the ventilating air are well mixed within a
room, regardless of its size. In many instances, this is far from
reality. There are numerous cases where the dispersion of pollutants does not correspond to a well-mixed condition 1. Computational fluid dynamics CFD provides a practical option for predicting the airflow and pollutant distributions in buildings. CFD
can be used to determine the best locations of air supply diffusers
and return outlets and the flow rate needed to create an acceptable
indoor air quality.
The Reynolds number for the flow inside a built environment is
most likely high; hence the CFD model must account for turbulence. Turbulence modeling can be done at a variety of levels,
depending on accuracy requirements, the computational facilities,
and the wall-clock time available for the calculation. Direct numerical simulation DNS is a model-free approach that involves
obtaining the three-dimensional solution of the NavierStokes
NS equations with full resolution of all of the spatial and tem1
Corresponding author. Present address: Department of Mechanical & Aerospace
Engineering, Cornell University, 248 Upson Hall, Ithaca, NY 14853-7501.
JOURNAL OF FLUIDS ENGINEERING DIVISION. Manuscript received June 14, 2008; final
manuscript received March 8, 2009; published online April 15, 2009. Review conducted by Paul Durbin.
poral scales of the turbulent motion. With DNS, the complete flow
field throughout the domain is determined with no modeling assumptions, and thus it offers the most accurate characterization of
the turbulence possible. The principal drawback of DNS is that its
computational cost increases in proportion to the cube of the Reynolds number 2. This limits its use mainly to fundamental scientific investigations of turbulence, and makes it impractical for
CFD of the sort required for the present application.
At the other extreme in terms of computational cost is the
Reynolds-averaged NavierStokes RANS modeling. With
RANS, modeled equations for the mean velocity of a statistically
stationary turbulent flow are solved. The higher-order Reynolds
stresses are usually obtained from a turbulent viscosity model,
which is algebraically related to other turbulence quantities such
as the turbulent kinetic energy k and energy dissipation rate for
which modeled transport equations are solved. Because RANS is
not concerned with solving for the fluctuating velocity and pressure fields, the computational cost is much lower by orders of
magnitude than the equivalent DNS. However, the accuracy of
RANS predictions is limited by the fidelity of the closure models
used to represent the effects of the turbulent fluctuations. Despite
decades of research, there remain fundamental limitations to what
can be expected from a RANS simulation 3.
Large eddy simulation LES has emerged as an important intermediate approach, whereby one solves for the instantaneous
filtered velocity of the largest energy-containing eddies of the turbulent motion and models the effect of the subgrid-scale motions
MAY 2009, Vol. 131 / 051402-1
on the resolved scales. As LES directly solves for the large-scale

turbulence, it yields more accurate results than RANS, yet has a
much lower computational cost than DNS, in which all of the
scales of the turbulence must be fully resolved. With the rapid
growth in computer speed and memory over the past 30 years,
LES has moved from being a research tool to a valuable tool for
studying flows in realistic geometries.
The development of LES for turbulent flows has been an active
area of research, and there have been several important advances
in LES since the pioneering work of Smagorinsky 4, Lilly 5,
Deardorff 6, Schumann 7, and others. Advances have been
made in i modeling the unresolved processes; ii accurate numerical methods on structured and unstructured grids; iii detailed comparison of LES calculations with DNS and experimental data in canonical flows; iv extensions to include additional
phenomena, e.g., turbulent combustion; and v computational
power, which has increased by more than four orders of magnitude since the 1970s 8. Various subgrid-scale SS models have
been developed, beginning with the early work that relied on an
eddy viscosity model with a constant coefficient Cs, which is a
function of the flow regime 46. The dynamic model was first
introduced by Germano et al. 9, with important modifications
and extensions provided by Lilly 10 and Meneveau et al. 11.
The dynamic model provides a method for estimating the Smagorinsky coefficient locally. It has proven quite successful, and the
same procedure has been applied in several other contexts 2.
One requirement for truly accurate LES is that the grid adequately
resolve the largest energy-containing eddies so that the subgrid
model is subdominant to the resolved scales. This implies either
having some a priori knowledge of the length scales of the turbulence so that the grid can be designed accordingly or having an
adaptive approach that refines the grid where needed.
In the field of indoor air flow simulation, several researchers
have studied the performance of the k--model 12,13. The authors demonstrated that the model can successfully predict the
mean velocities. Chen 14 compared the performance of four
different k- and Reynolds stress models RSMs for four common problems in air flows: natural convection, forced convection,
mixed convection, and impinging jet flows. The study showed that
all four models predicted the mean velocities reasonably well;
however, they failed to accurately predict the turbulence levels.
While RANS models are strictly applicable to fully developed
turbulent flows 15, it is not uncommon in indoor air flows, especially at low ventilation rates, to have laminar, transitional, and
slowly developing turbulent flows co-existing 15,16.
LES has been successfully applied to airflow simulations
1719. Davison and Nielson 17 compared the performance of
the standard Smagorinisky and dynamic models for a simply ventilated room. It was found that the dynamic model agrees better
with the experimental data. Jiang et al. 19 used LES to study air
flows in and around buildings. They compared the Smagorinsky
SS model, the filtered dynamic FD model by Zhang and Chen
20, and the stimulated small-scale subgrid-scale SSS model by
Shah and Ferziger 21. They concluded that the simple SS model
gives less accurate results in regions where the flow is not fully
developed, as compared with the FD and the SSS models. The SS
model was found to work fairly well for the outdoor flows, where
conditions are fully turbulent. Since the FD and the SSS models
are computationally more intensive, the SS model was recommended for outdoor flows and the SSS and FD models were recommended for indoor flows. This represented an optimum solution to ensure accurate representation of upstream conditions.
Recently, Zhai et al. 22 and Zhang et al. 23 compared eight
different RANS and LES models for simulation of representative
indoor air flows such as forced and mixed convection in ventilated
spaces and natural convection in a tall cavity. They concluded that
the v2f-dav model by Davidson et al. 24 and LES are the most
suitable models in predicting air velocity, temperature, and turbulent quantities for low Rayleigh number natural convection flows.
051402-2 / Vol. 131, MAY 2009
In forced convection with low turbulence levels, the renormalization group RNG k- model by Yakhot and Orszag 25, the low
Reynolds number k- model due to Launder and Sharma LRN-LS
26, the v2f-dav, and LES all performed very well. The v2f
model by Durbin 27 was found to be very promising but the
numerical algorithm has to be constructed so as to avoid some
inherent numerical problems. The RSMs were shown to be able to
capture more flow details than eddy viscosity based models; however, they require more computational time. The authors concluded that LES provided more detailed and possibly more accurate predictions for indoor air flows and that it is a very important
tool for understanding the dynamics of indoor airflow.
In the current investigation, we use LES to model the air flow
inside an experimental flow chamber located in the Indoor-FlowLaboratory IFL at Syracuse University. The hydrodynamic conditions within the facility are typical of displacement ventilation
systems, where the supply air displaces the room air with presumed minimal mixing so as to achieve a high air change efficiency 1.
Consistent with the incremental procedure for simulation of displacement ventilation systems laid out by Chen and Srebric 28
and Chen and Zhai 29, we have neglected thermal convection
effects and geometrical complexities at this stage. The main objective here was limited to modeling and understanding the basic
flow behavior inside the IFL, with an eye toward incorporating
more comprehensive physics and more geometrical details in future investigations.
The sensitivity of the flow field to the inlet turbulence levels
and the details of the flow at the inlet are investigated. The performance of the Smagorinsky LES model with constant coefficient and the dynamic model in predicting the air flow are systematically studied and analyzed. Finally, light is shed on the
performance of the k- model in predicting the mean flow and the
turbulent kinetic energy throughout the flow field at varying inlet
turbulence levels.
Turbulence Models
2.1 RANS Equations. The ensemble average of the constant

density NS equations takes the form 30
ui
=0
+
t
xi
ui
ui u j 2 ul
uiu j =
+
ij
+
t
xj
xi x j
x j xi 3 xl
+
uiuj
xj
where is the density, p is the pressure, is the dynamic viscosity, ui is the velocity component in the xi direction, and the overbar indicates an averaged quantity. The symbol ij denotes the
Kronecker delta, and the Einstein summation convention is used.
The Reynolds-averaged approach to turbulence modeling requires
that the Reynolds stress term uiuj be appropriately modeled. In
the k- model, the Boussinesq approximation is used to relate the
Reynolds stresses to the mean velocity gradients through
uiuj = t
uk
ui u j
2
+
k + t
ij
3
xk
x j xi
where the turbulent or eddy viscosity t must be specified in

order for the above system of equations to be closed.
2.1.1 The Standard k- Model. The simplest complete models
of turbulence are two-equation models in which the solution of
two separate transport equations allows the turbulent velocity and
length scales to be independently determined. According to Ref.
30, the standard k- model in FLUENT falls within this class of
turbulence models that has become the workhorse of practical
engineering flow calculations in the time since it was proposed by

Launder and Spalding 31. The turbulence kinetic energy k and
its rate of dissipation are obtained from the following transport
equations:
i =
ku
+
t xi
xj

ui =
+
t xi
xj
t k
+ Gk
k x j
where Gk represents the generation of turbulence kinetic energy

due to the mean velocity gradients, and C1, and C2 are the model
constants. The details of how these terms are calculated in FLUENT
can be found in Ref. 30 and are omitted here for the sake of
brevity. In these equations, k and are the turbulent Prandtl
numbers for k and , respectively. The turbulent viscosity t is
computed by combining k and as follows:
6
The model constants C1, C2, C, k, and have the following

default values: C1 = 1.44, C2 = 1.92, C = 0.09, k = 1.0, and
= 1.3.
2.1.2 A Two-Layer Model for the Enhanced Wall Treatment.
The two-layer approach is an integral part of the enhanced wall
treatment and is used to specify both and the turbulent viscosity
in the near-wall cells 30. In this approach, the whole domain is
subdivided into a viscosity-affected region and a fully turbulent
region. The demarcation between the two regions is determined
by a wall-distance-based, turbulent Reynolds number Rey. In the
fully turbulent region Rey Rey , the k- model described above
is employed. In the viscosity-affected near-wall region Rey
Rey , the one-equation model of Wolfshtein 32 is employed.
Here the default value of Rey = 200 is used. In the enhanced wall
treatment approach, the turbulent viscosity is smoothly blended
with the high-Reynolds-number t definition from the outer region. The main purpose of the blending function is to allow convergence even when the k- solution in the outer layer does not
match the two-layer formulation 30. In the enhanced wall function approach, FLUENT blends the linear laminar and the logarithmic turbulent laws-of-the-wall for the mean velocity using a
function suggested by Kader 33 to form a single law-of-the-wall
for the entire wall region. The resulting law-of-the-wall has the
correct asymptotic behavior in the viscous sublayer and the loglaw region and also represents the velocity profile in the wall
buffer region reasonably well 30. The k equation is solved in the
whole domain including the wall-adjacent cells. The boundary
condition for k imposed at the wall is
where the tilde indicates a filtered quantity,

ij u
uiu j
iu j
ij 31 kkij = 2tSij
11
12
where Sij is the rate-of-strain tensor for the filtered velocity, defined by
S 1 ui + u j
ij
2 x j xi
13
and t is the subgrid-scale turbulent or eddy viscosity.

2.2.1 Smagorinsky-Lilly Model. In the Smagorinsky model,
the eddy viscosity is modeled by
t = Ls2S
14
where Ls is the mixing length for the subgrid scales and S
2SijSij. In FLUENT, Ls is computed using
Ls = mind,CsV1/3
15
where =0.41 is the von Karman constant, d is the distance to

the closest wall, and V is the volume of the computational cell. A
value of 0.1 for Cs has been found satisfactory for a wide range of
flows and was thus used in the current investigation.
2.2.2 Dynamic SmagorinskyLilly Model. Germano et al. 9
and subsequently Lilly 10 conceived a procedure by which the
Smagorinsky model constant Cs is dynamically computed based
on the information provided by the resolved scales of motion.
They realized that the LES equations are solved for the filtered
velocity field, although this filtering is not always explicitly performed. Using single- and double-filtered residual stress tensors,
Germano 34 derived a closed expression for a tensor referred to
as the resolved stress tensor, in terms of the grid-filtered velocity
field. The resolved stress tensor can be loosely interpreted as the
contribution to the residual stress from the largest unresolved
scales 2. If a Smagorinsky type model for the deviatoric part of
this tensor is constructed, then an adjustable coefficient Cs can be
locally matched to provide the best estimate for the deviatoric part
of the resolved stress tensor. As it is impossible to match the five
independent components of the deviatoric tensor, Lilly 10 derived a specification of Cs that minimizes the mean-square error.
In the implementation of the dynamic LES model in FLUENT, Cs is
clipped at zero and 0.23 by default to eliminate backscatter and to
avoid numerical instabilities 30.
2.2 LES Equations. By filtering the constant density NS

equations, one obtains 30
ui
ij
p ij
uiu j =
+
t
xj
xj xj
xi x j
10
where n is the local coordinate normal to the wall. As we will

illustrate, the dynamics of the flow are dominated by the shear
layers and hence we expect the flow to be relatively insensitive to
the boundary conditions applied at the walls.
ui
=0
+
t
xi
2 ul
ui u j
+
ij
3 xl
x j xi
The subgrid stresses are modeled using the Boussinesq approximation as in the RANS models:
5
k
=0
n
that represents the contributions of the filtered velocity to the viscous stress tensor, and ij is the subgrid-scale stress defined by
2
t
+ C 1 G k C 2
k
k
x j
k2
t = C
ij =
8
9
Numerical Solution Details
The commercial CFD code FLUENT was used to simulate the

flow. The temporally second-order accurate, unsteady solver based
on the noniterative time advancement algorithm was used with the
incompressible pressure-based segregated solver, using the
pressure-implicit with splitting of operator pressure-velocity coupling scheme. The central differencing scheme is an ideal choice
for LES because of its low numerical diffusion; however, it can
result in unphysical oscillations in the solution. This situation is
worsened if combined with the low subgrid-scale turbulent diffusivity used in LES. The bounded central differencing scheme is a
composite normalized variable diagram scheme that consists of
pure central differencing, a blended scheme of the central differencing and the second-order upwind scheme, and the first-order
upwind scheme, which is used only when the convection boundMAY 2009, Vol. 131 / 051402-3
244
15
20
Z
Y
Outlet
46
Inlet
183
(a)
(b)
Fig. 1 a Geometry of Indoor Flowfield Laboratory Chamber; dimensions are 244 183 244 cm3.
b Top view; dimensions are in cm.
edness criterion is violated 30. Provided that the numerical solution converges, this approach leads to pure second-order differencing 30. The bounded central difference scheme was used for
discretization of both the pressure and momentum equations, and
default tolerances were used a residual tolerance of 0.0001 and
relaxation factors of 1 for the pressure and momentum equations.
Air with constant density and viscosity at standard sea level conditions was used. The time step was set so that the resulting maximum Courant number would be no larger than 0.5. The boundary
conditions used in the first stage of this investigation were simple
plug/uniform flow velocity profile at the inlet, with uniform turbulence intensities of 0%, 5%, and 13%, a uniform turbulence
length scale equal to 7% of the inlet hydraulic diameter, a pressure
outlet for the outflow, and solid wall boundary conditions everywhere else. Simulations of the flow were continued up to the time
when the flow was approximately statistically stationary. In the
second stage of the study, the inlet velocity and turbulence profiles
obtained from experiments conducted in the IFL at Syracuse University 35 were used in the simulations.
3.1 Description of the Geometry. The IFL chamber consists
of a 2.44 1.83 2.44 m3 cubicle, with optical access for particle
image velocimetry PIV measurements through the front 1.83
2.44 m2 wall. The origin of the coordinate system is located at
the center of the cubicle, with positive z pointing vertically upward and positive x pointing toward the front wall see Fig. 1.
The chamber half-height L is used to nondimensionalize the vertical distance. The chamber is designed to represent a typical
indoor-flow environment for a single occupancy cubicle. An isometric view of the chamber is shown in Fig. 1a and a top view
is shown in Fig. 1b. The cubicle is ventilated by a low speed,
closed-loop system that is designed to allow an occupants thermal plume to be a significant factor in driving the flow. There is a
0.2 0.46 m2 inlet on the floor, near the front wall, and an identically sized outlet in the ceiling near the rear wall. The average
inlet and outlet flow velocity is 0.2 m/s.
3.2 Description of the Computational Grid. An initial grid
containing 48 32 48 cells was developed using FLUENTs Gambit mesh generator to represent the chamber. The accuracy of the
resulting solution was checked by grid refinement and comparison
of the resulting solution with the original solution on the coarser
grid. The process was repeated until an accurate solution was
obtained. Based on the initial results from the coarse grid, the grid
was locally refined in those regions exhibiting steep gradients of
the solution variables such as the inlet, the shear layers around the
inflow plume, the ceiling, and the outlet. First, a vertical column
051402-4 / Vol. 131, MAY 2009
including the inlet and three cells on each side of it was refined
once by halving the mesh dimensions in each coordinate direction, and another volume around the outlet, extending three cells
on each side of it and four cells below it was refined again, by
halving the mesh dimensions in each coordinate direction. Finally, a volume covering the entire ceiling area and extending four
cells below, it was similarly refined. The resulting grid was again
used to determine the general features of the flow. Based on the
results of the initial calculations on this grid, it was determined
that extra refinement of the grid in the vicinity of the inlet was
required to maintain adequate accuracy. Two additional steps of
refinement were done. The first involved refinement of the volume
covering the inlet area and four cells away from it on all sides, and
extending 18 cells above the inlet. The second refinement covered
the area of the inlet and two cells away from it on all sides, and
extended nine cells above the inlet. The resulting grid, referred to
here as the baseline grid, is shown in Figs. 2a and 2b where a
shows a side view of the vertical plane passing through the center
of the inlet and b shows a top view of the floor and the inlet area.
The baseline grid calculations were performed on a Linux cluster
using ten processors. The wall-clock time per time step was about
10 s corresponding to 800 s of wall-clock compute time per
physical second since the time step on this grid is t = 0.0125 s.
It was found that the wall-clock time for calculations with the
dynamic subgrid-scale model is nearly the same as that for the
constant coefficient Smagorinsky model. To test grid convergence,
one case 13% inlet turbulence intensity was solved on a refined
grid that contained twice the number of grid cells in each direction. The time step used on the refined grid was t = 0.006 25 s,
so the solution for a fixed time interval was approximately 16
times more computationally expensive than the solution on the
baseline grid. The computations on the refined grid were performed on a newer Linux cluster utilizing Infiniband as an interconnect, which was found to significantly decrease the wall-clock
time. Using 20 cores on five nodes each having two dual core
processors reduced the wall-clock time to about 1 s per time step.
3.3 Simulation of Inlet Turbulence. Simulation of inlet turbulence for fully developed turbulent flows can, in principle, be
achieved by adding small perturbations to a laminar flow and by
having a long enough computational domain in the streamwise
direction to allow the turbulence to develop. In order to reduce the
development region, a more viable approach is to introduce an
upstream domain and to solve for the flow assuming periodicity in
the streamwise direction 36. The result of this calculation is then
used as an inflow boundary condition for the main simulation.
Y
(a)
Z
(b)
Fig. 2 Computational grid for the Indoor Flowfield Laboratory Chamber: a side x-z plane and b top x-y plane
views of the baseline grid
This approach, while accurate, is too expensive to be used in

engineering applications. A more practical approach is to superimpose random noise perturbations on mean velocity profiles obtained from a separate RANS simulations 37. This approach can
lead to stability problems due to the unphysical randomness of the
perturbations that depend on the time step 38. A similar approach is to use random perturbations to generate inlet turbulence
with predefined characteristics. Using this technique, Lee et al.
39 built a realistic turbulence spectrum using Fourier modes
with random phase shifts. This method can, however, result in an
artificial transition region between the prescribed turbulent field
and the solution of NS equations inside the domain 40. FLUENT
supports two techniques for generating time-dependent inlet conditions for LES. The first technique is the random flow generation
RFG technique developed by Smirnov et al. 40. This technique
generates a divergence-free, inhomogeneous, and anisotropic flow
field with predefined correlation coefficients. This method requires the specification of the anisotropic velocity correlation tensor, length, and time scales of the turbulence that may be obtained
from a RANS simulation or high-resolution experimental data.
The spatial and temporal variations of the resulting velocity field
follow a Gaussian spectrum with the specified length and time
scales. Smirnov et al. 40 successfully applied this technique to
the simulation of ship wakes. The ship hull part of the domain was
simulated using a RANS calculation. The resulting anisotropic
field was then used as an inflow plane to the wake part of the
domain, which was simulated with LES. The length and time
scales required by the RFG were deduced from the length scales
predicted by the RANS. The RFG is a more realistic representation of the turbulence than a simple Gaussian distribution of velocity as it guarantees that the resulting velocity field satisfies the
continuity and anisotropy constraints. The RFG, however, does
not guarantee that the resulting field satisfies the momentum equations and is thus an approximate method 40. The second technique for inlet turbulence generation in FLUENT is the vortex
method developed by Mathey et al. 41. With this approach, a
perturbation is added to the specified mean velocity profile via a
fluctuating vorticity field two-dimensional in the plane normal to
the mean-flow direction. The use of this method requires the
specification of the mean turbulent kinetic energy and dissipation
rate profiles at the inlet of the computational domain 30. The
accuracy of this method has been investigated using several test

cases 38 such as turbulent channel and pipe flows and the flow
over periodic hills. The flow over periodic hills represents a case
where the dynamics of the flow are controlled to a great extent by
the level of upstream turbulence. Mathey et al. 41 compared two
approaches for specifying the inlet turbulence with a reference
simulation in which periodic boundary conditions were used. In
the first test simulation, the vortex method was used to generate
the inlet boundary conditions and in the second test simulation
random noise was used. They found that the random noise simulation overestimated the re-attachment length by 50% and underestimated the spreading rate of the shear layer. They also found
that the random noise underestimated the turbulence level inside
the shear layers by one order of magnitude. They explained these
results by the low level of kinetic energy in the random noise
simulation due to the damping of the nonphysical random fluctuations at the inlet. This demonstrated the importance of the upstream history of the turbulence on the subsequent production
mechanisms and thus the sensitivity of the turbulence to the accuracy of the inlet boundary conditions 38. We therefore selected FLUENTs built-in vortex method to generate the timedependent inlet conditions.
We begin by showing a comparison of solutions on the refined

and the baseline grids to demonstrate that the solution on the
baseline grid is spatially converged. On this basis, the baseline
grid is used for all the subsequent computations. The effect of the
turbulence level at the inlet is studied assuming a uniform mean
velocity profile and three levels of isotropic turbulence at the inlet.
Then, the 13% inlet turbulence case with uniform mean profiles is
compared with the case using the measured inlet velocities and
turbulence intensities to determine the significance of the detailed
profiles at the inlet. We also compare the performance of different
modeling approaches: the constant coefficient Smagorinsky
subgrid-scale model and the dynamic model for the 13% case with
uniform inlet profiles, and the constant coefficient model with the
k- model at 0% and 13% inlet turbulence intensity.
4.1 Grid Convergence Study. The grid resolution for LES
should be sufficient to resolve the bulk of the turbulent kinetic
MAY 2009, Vol. 131 / 051402-5
Refined grid
Baseline grid
0.2
0.08
Mean velocity magnitude [m/s]
0.175
0.15
0.125
0.1
0.075
0.05
0.025
(a)
0
0.7
Refined grid
Baseline grid
0.09
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0.8
0.9
1.1
1.2
(b)
x [m]
0
0.7
0.8
0.9
1.1
1.2
x [m]
Fig. 3 Comparison of mean velocity magnitude on the baseline and the refined grids at a z / L = 0.2 and b z / L
= 0.9375
energy. If this is achieved, LES should be at most weakly dependent on the particular subgrid-scale model used in the computations. In FLUENT, the mixing length for the subgrid scales Ls and
the subgrid-scale eddy viscosity t = t / are used to construct an
estimate for the subgrid-scale turbulent kinetic energy ks defined
as
ks =
t2
16
Ls2
which is referred to by FLUENT as the subtest kinetic energy.

Figures 3 and 4 compare the mean and root mean square rms of
the velocity magnitude obtained from the solutions on the baseline
and the refined grids. It is seen that the shapes of the profiles agree
reasonably well and that the deviations are generally small, reaching a maximum of about 30%. It is also seen that the deviations at
the stations closest to the inlet Fig. 3a and 4a at z / L = 0.2
occur predominantly in the shear layers. Figure 5 also compares
the turbulent kinetic energy obtained from the solutions on the
two grids including the subgrid-scale contributions. It is clear that
on the refined grid the subtest kinetic energy is less than 5% of the
total turbulent kinetic energy, except very near the inlet, so the
solution on this grid is judged to be well resolved. The subtest
kinetic energy on the baseline grid is somewhat larger, but still
small enough relative to the total turbulent kinetic energy that we
judge the solution on the baseline grid to be adequately resolved
for the subsequent computations. This conclusion is supported by
the earlier comparison of the mean and rms profiles shown in
Figs. 3 and 4.
4.2 Effect of Inlet Turbulence Level Using Plug-Flow Inlet
Conditions. Figures 6a and 6b show the contour plots of the
instantaneous and mean velocity magnitudes, respectively, along
the vertical x-z plane passing through the middle of the inlet for
the 0% inlet turbulence intensity case. It is clear that the jet in this
case is very confined and that there is little mixing with the room
air. The jet flows straight up toward the ceiling, bends 90 deg, and
flows along the ceiling until it reaches the exit. Other than this
basic flow pattern, there are no significant organized motions or
Refined grid
Baseline grid
0.08
0.07
RMS velocity magnitude [m/s]
0.07
0.06
0.05
0.04
0.03
0.02
0.01
(a)
0
0.7
Refined grid
Baseline grid
0.08
0.06
0.05
0.04
0.03
0.02
0.01
0.8
0.9
x [m]
1.1
1.2
(b)
0
0.7
0.8
0.9
1.1
1.2
x [m]
Fig. 4 Comparison of the rms of the velocity magnitude on the baseline and the refined grids at a z / L = 0.2 and b
z / L = 0.9375
051402-6 / Vol. 131, MAY 2009
0.003
0.002
Total T.K.E. , Refined grid

Total T.K.E. , Baseline grid
Subtest T.K.E. , Refined grid
Subtest T.K.E. , Baseline grid
0.0025
Total T.K.E. , Refined grid

Total T.K.E. , Baseline grid
Subtest T.K.E. , Refined grid
Subtest T.K.E. , Baseline grid
0.0015
T.K.E. [m /s ]
T.K.E. [m /s ]
0.002
0.0015
0.001
0.001
0.0005
0.0005
0
0.7
0.8
0.9
(a)
1.1
0
0.7
1.2
x [m]
(b)
0.8
0.9
1.1
1.2
x [m]
Fig. 5 Comparison of turbulent kinetic energy on the baseline and the refined grids at a z / L = 0.8 and b z / L
= 0.9375
circulation patterns in the room. The figures for the instantaneous

and mean velocities are also reasonably similar, especially near
the inlet, suggesting that the levels of turbulence are quite low.
Near the ceiling, however, the jet has spread enough to interact
with the front wall, generating turbulence there and in the ceilings boundary layer. A complicated unsteady flow pattern, dominated by eddying motions, is observed in the ceilings stagnation
layer.
Figures 7 and 8 show the corresponding contour plots for the
cases of 5% and 13% inlet turbulence intensity. It is seen in both
cases that the jet spreads and mixes with the room air more rapidly and starting closer to the inlet than in the laminar inflow case.
The mean flow, however, is still generally confined to a vertical
column around the inlet and a horizontal slab along the ceiling,
with only weak circulation elsewhere in the room. The reason for
the more rapid mixing and spreading of the jet, and the increased
levels of turbulence relative to the laminar inflow case is that the
inlet jet, seeded with turbulence, undergoes a more rapid transition
to a turbulent jet than the laminar inflow case. Figures 7 and 8
show significant similarity between the 5% and 13% turbulence
intensity cases in both the mean and the instantaneous flow velocity magnitudes. It is also noted that the instantaneous contours in
both cases are characterized by the intermittent breaking-off of

blobs of high velocity fluid from the jet core that increase the local
levels of turbulence as they sweep through the domain.
The above observations can be seen more quantitatively by
looking at profiles of the mean velocity magnitude and the rms of
the velocity fluctuations for the different levels of the inlet turbulence intensity. Figures 9 and 10 compare the profiles of the mean
velocity magnitude at stations located at z / L = 0.4, z / L = 0.0, and
z / L = 0.75, plotted along the horizontal lines extending from x
= 0.71 m all the way to the front wall, along the center of the inlet
y = 0. As can be seen, the mean velocity profiles in the 5% and
13% cases are quite similar, exhibiting nearly the same spreading
rate. This is consistent with the notion that, beyond a threshold
level of the inlet turbulence intensity, the subsequent development
of the jet is only weakly dependent on the inlet conditions. It is
also noted that the profiles of the 5% and 13% cases exhibit significant asymmetry due to the front wall. The laminar inflow case,
in contrast, has nearly symmetric profiles up until z / L = 0.75,
where the jet has spread enough to interact with the wall, causing
an asymmetry of the profile to develop.
0.258
0.2451
0.2322
0.2193
0.2064
0.1935
0.1806
0.1677
0.1548
0.1419
0.129
0.1161
0.1032
0.0903
0.0774
0.0645
0.0516
0.0387
0.0258
0.0129
0
0.281
0.26695
0.2529
0.23885
0.2248
0.21075
0.1967
0.18265
0.1686
0.15455
0.1405
0.12645
0.1124
0.09835
0.0843
0.07025
0.0562
0.04215
0.0281
0.01405
0
(a)
(b)
Fig. 6
a Instantaneous and b mean velocity contours on center plane; 0% inlet turbulence units are in m/s
MAY 2009, Vol. 131 / 051402-7
0.262
0.2489
0.2358
0.2227
0.2096
0.1965
0.1834
0.1703
0.1572
0.1441
0.131
0.1179
0.1048
0.0917
0.0786
0.0655
0.0524
0.0393
0.0262
0.0131
0
0.274
0.2603
0.2466
0.2329
0.2192
0.2055
0.1918
0.1781
0.1644
0.1507
0.137
0.1233
0.1096
0.0959
0.0822
0.0685
0.0548
0.0411
0.0274
0.0137
0
(a)
(b)
Fig. 7
0.262
0.2489
0.2358
0.2227
0.2096
0.1965
0.1834
0.1703
0.1572
0.1441
0.131
0.1179
0.1048
0.0917
0.0786
0.0655
0.0524
0.0393
0.0262
0.0131
0
0.304
0.2888
0.2736
0.2584
0.2432
0.228
0.2128
0.1976
0.1824
0.1672
0.152
0.1368
0.1216
0.1064
0.0912
0.076
0.0608
0.0456
0.0304
0.0152
0
(a)
(b)
Fig. 8
Laminar inflow
5% Turbulence
13% Turbulence
0.2
0.175
0.175
0.15
0.125
0.1
0.075
0.05
0.025
(a)
0.7
Laminar inflow
5% Turbulence
13% Turbulence
0.2
0.15
0.125
0.1
0.075
0.05
0.025
0.8
0.9
x [m]
1.1
1.2
(b)
0.7
0.8
0.9
1.1
1.2
x [m]
Fig. 9 Comparison of the mean velocity magnitude for three levels of inlet turbulence intensity at a z / L = 0.4 and b
z / L = 0.0; baseline grid
051402-8 / Vol. 131, MAY 2009
Laminar inflow
5% Turbulence
13% Turbulence
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
0.7
0.8
0.9
1.1
1.2
x [m]
Fig. 10 Comparison of the mean velocity magnitude for three

levels of inlet turbulence intensity at z / L = 0.75; baseline grid
Profiles of the velocity rms are shown in Figs. 11 and 12. Once
again the 5% and 13% cases show quite similar profiles throughout the computational domain. The behavior of the laminar inflow
case needs some elaboration. At the z / L = 0.4 and z / L = 0.0 stations, the level of turbulence is significantly less than for the 5%
and 13% cases because of the laminar state of the entering jet.
However, near the ceiling, the shear layers have grown enough to
begin to interact with the front wall and turbulence is generated
there and also in the ceilings boundary layer. The contours in Fig.
6 show the generation of a blob of high velocity fluid near the
ceiling, which can contribute to the high levels of turbulence near
the ceiling. Accordingly, the rms levels of the velocity fluctuations
in the laminar inflow case are seen to be higher than the rms levels
for the 5% and 13% cases near the ceiling see Figs. 11 and 12.
The significant differences between the simulation results for
the 0% and both the 5% and 13% cases cannot be captured by the
standard k- model, which is designed primarily for fully developed turbulent flows. This very fact led Loomans 42, who investigated the effect of inlet turbulence intensity level on the flow
field of a displacement ventilation system in a full-scale room
using the standard k- model, to conclude that the level of inlet
Laminar inflow
5% Turbulence
13% Turbulence
0.06
0.05
0.05
0.04
0.03
0.02
0.04
0.03
0.02
0.01
(a)
0
0.7
Laminar inflow
5% Turbulence
13% Turbulence
0.06
0.01
0.8
0.9
1.1
1.2
(b)
x [m]
0
0.7
0.8
0.9
1.1
1.2
x [m]
Fig. 11 Comparison of the rms velocity magnitude for three levels of inlet turbulence intensity at a z / L = 0.4 and b
0.065
Laminar inflow
5% Turbulence
13% Turbulence
0.08
0.07
0.06
0.05
0.04
0.7
0.8
0.9
(a)
x [m]
1.1
Laminar inflow
5% Turbulence
13% Turbulence
0.06
0.055
0.05
0.045
0.04
0.7
1.2
(b)
0.8
0.9
1.1
1.2
x [m]
Fig. 12 Comparison of the rms velocity magnitude for three levels of inlet turbulence intensity at a z / L = 0.875 and b
MAY 2009, Vol. 131 / 051402-9
0.25
0.3
0.25
0.2
0.2
Exp.Data/Left side
0.15
Exp.Data/Right side
W [m/s]
W [m/s]
0.15
4th degree poly fit

4th degree polyfit
0.1
0.1
0.05
Experimental data
0.05
4th degree polynomial fit

4th degree polynomial fit
0.05
1
(a)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Y/L
(b)
0
800
850
900
950
1000
1050
1100
X [mm]
Fig. 13 Experimental data and curve fits along the a long and b short sides of the inlet. In a, Ls is the half-length of
the inlet section.
turbulence has a negligible effect on the flow field especially far

from the inlet. As we will demonstrate, the standard k- model is
insensitive to the level of inlet turbulence intensity and accordingly cannot be used to accurately predict the flow behavior for
situations wherein the flow is most probably transitional in nature.
It is also noted that the levels of inlet turbulence intensities investigated by Loomans were 10% and 35% and accordingly the flow
was most probably fully developed even for the lower value. Using the k- model for the simulation of a jet in a two-dimensional
cavity with inlet turbulence intensities of 437.4%, Joubert et al.
43 also concluded that the effect of the inlet turbulence intensity
on the flow field was negligible.
The LES results for the 5% and 13% cases are consistent with
the study of Jiang 44 who used LES and the standard k- model
to investigate the effects of inlet boundary conditions on the flow
field in a full-scale room. At the highest studied ventilation rate of
19.5 air change per hour ACH, corresponding to an inlet mean
velocity of 1.43 m/s and a Reynolds number of 4895, Jiang investigated three cases: 1 uniform inlet velocity profile and uniform
turbulence intensity of 10%, 2 profiled velocity and turbulent
kinetic energy from hot wire measurements, and 3 experimental
velocity profile and uniform inlet turbulence intensity of 10%. In
LES, the turbulence at the inlet was generated using a random
number generator technique. The author concluded that the effect
of inlet conditions on the flow field is confined to the vicinity of
the inlet and negligible elsewhere in the flow field. Although the
average level of inlet turbulence intensity in case 2 could not be
inferred, the high ventilation rate/Reynolds number used in this
study indicate that the flow was fully developed in nature. Our
LES simulation with laminar inflow conditions shed some light on
the potentially significant differences between simulations at low
levels of inlet turbulence and those with high enough levels of
inlet turbulence to cause the flow to rapidly reach a fully developed state. The existence of a threshold inlet turbulence level,
above which the flow becomes independent of the inlet turbulence
level, is a hypothesis that requires further investigation.
4.3 Effects of Realistic Inlet Flow Conditions. Here we consider the importance of the detailed mean and turbulence profiles
at the inlet on the flow inside the room. We replace the idealized
inlet velocity profile used in Sec. 4.2 by the experimentally measured inlet conditions. The first stage of utilizing the PIV experimental data obtained at Syracuse University 45 in our simulations involved interpretation of the data, which contained artifacts
from the measurement technique and experimental noise. In addition, data could not be obtained for that part of the inlet closest to
051402-10 / Vol. 131, MAY 2009
the front wall, due to distortions and reflections from the glass
wall; consequently, assumptions had to be made about the profiles
of the measured quantities. The experimental data consist of the
mean values of the three components of velocity and the rms of
their fluctuations. The mean velocity components in the plane of
the inlet, i.e., the U and V components, were deemed small
enough to neglect, whereas the vertical component W of the velocity was fitted to a smooth polynomial. The variations parallel to
the long side of the inlet were not as significant as those across the
short side, and a fit of the averaged experimental data on both
sides was used. Figure 13a shows two fourth-order polynomials
one covering most of the profile and one representing the shear
layer portion. In the direction of the short side, the interpretation
of the data was complicated by the fact that part of the inlet was
not covered by the measurement window; hence this part of the
data was missing. We extrapolated the missing data using two,
fourth-order, polynomials that represent the profile, including the
missing segment, as shown in Fig. 13b. The turbulent kinetic
energy and the root mean square of the fluctuating velocities were
assumed symmetric across the short side of the duct, with negligible variation in the direction of the long side, as shown in Fig.
14 which shows only half of the profile since it is assumed
symmetric.
FLUENT calculations with these fitted inlet profiles were run and
compared with the 13% case with uniform profiles. Figures 15
and 16 show comparisons of the mean velocity at four vertical
stations. Note that the significant differences at the lowest station
z / L = 0.8 become weaker with increasing distance from the inlet. The maximum deviation between the two cases occurs in the
shear layers, where the deviation of the measured inlet turbulent
kinetic energy from the spatially averaged value is greatest. This
suggests that there is some memory of the inlet conditions
throughout the flow.
Similar behavior of the rms of the velocity is shown in Figs. 17
and 18, where close to the inlet at the z / L = 0.8 and z / L = 0.4
stations the profiles are significantly different, whereas close to
the ceiling, the two cases predict similar levels of turbulence notice the offset of the origin of the vertical axis in Fig. 18 amplifies
the small differences between the profiles. So in general the flow
field exhibits sensitivity to the inlet flow conditions primarily near
the inlet. As the flow evolves spatially, the effect of the inlet
conditions becomes less pronounced. This conclusion is in agreement with Jiangs 44 conclusions from his LES and k- model
calculations for similar comparisons. Joubert et al. 43 also reported that using a parabolic inlet velocity profile as opposed to a
x 10
Experimental data
Second degree polynomial fit
Turbulent kinetic energy (T.K.E.) [m /s ]
0
860
880
900
920
940
960
980
X [mm]
Fig. 14 Fitted profile of the turbulent kinetic energy
uniform profile does not affect the mean velocity in the twodimensional cavity flow that they analyzed with the k- model;
however, they found significant differences in the predicted levels
of turbulence in the cavity, which they explained on the basis of
the contribution of the shearing already present in the inflow in
the parabolic profile case. It should be noted that the wall jet
studied in their two-dimensional simulation is fundamentally different from the present configuration. The close proximity of the
jet to the top boundary may have caused the greater sensitivity.
They also noted that the width of the slot has an important effect
on the turbulence level within the cavity.
4.4 Comparison of the Constant Coefficient and Dynamic
Smagorinsky LES Models. The success of a subgrid-scale model
can be characterized by how well it predicts the large-scale statis-
tics of a given turbulent flow. It is to be expected that the results

should be relatively insensitive to the subgrid-scale model if the
large-scale, energy-containing, turbulent motions are sufficiently
resolved. The dynamic model adjusts the coefficient Cs locally in
space and time based on the resolved scales. In this way, the
model responds to changes in the large-scale turbulent motions,
and thus it tends to be self-correcting in the sense that it adjusts
the eddy viscosity coefficient based on the energy in the resolved
scales. In this section, a comparison is made between the results
obtained with the constant coefficient Smagorinsky subgrid-scale
model and the dynamic model for the case of 13% inlet turbulence
intensity, with uniform mean inlet profiles. Figures 1922 compare the mean and the rms of the velocity magnitudes obtained
from the constant coefficient and the dynamic model in both the
13% Turbulence intensity

Experimental profiles

0.15
0.1
0.05
0
0.7
(a)
0.2
0.2
0.8
0.9
x [m]
1.1
0.15
0.1
0.05
0
0.7
1.2
(b)
0.8
0.9
1.1
1.2
x [m]
Fig. 15 Mean velocity magnitude at a z / L = 0.8 and b z / L = 0.0; comparison of results with plug flow and experimentally determined inlet conditions
MAY 2009, Vol. 131 / 051402-11


0.16
0.1
0.14
0.12
0.1
0.08
0.06
0.04
0.08
0.06
0.04
0.02
0
0.7
0.8
0.9
1.1
0.7
1.2
0.8
0.9
1.1
1.2
(a)
(b)
x [m]
x [m]
Fig. 16 Mean velocity magnitude at a z / L = 0.75 and b z / L = 0.875; comparison of results with plug flow and experimentally determined inlet conditions

0.05
0.05
0.045
0.045
13% Turbulence intensi

0.04
0.035
0.03
0.025
0.02
0.015
0.01
(a)
0.7
0.04
0.035
0.03
0.025
0.02
0.015
0.01
0.8
0.9
1.1
1.2
0.7
0.8
0.9
1.1
1.2
(b)
x [m]
x [m]
Fig. 17 rms velocity magnitude at a z / L = 0.8 and b z / L = 0.4; comparison of results with plug flow and experimentally determined inlet conditions.

0.065
0.06
0.055
0.05
0.045
0.04
0.7
0.8
0.9

0.065
0.07
1.1
1.2
0.06
0.055
0.05
0.045
0.04
0.7
0.8
0.9
1.1
1.2
(b)
x [m]
x [m]
Fig. 18 rms velocity magnitude at a z / L = 0.875 and b z / L = 0.9375; comparison of results with plug flow and experimentally determined inlet conditions
(a)
051402-12 / Vol. 131, MAY 2009
Const coeff model

Dynamic model
Const coeff mod

Dynamic model
0.2
0.2
0.15
0.1
0.05
0
0.7
0.8
0.9
1.1
0.15
0.1
0.05
0
0.7
1.2
0.8
0.9
1.1
1.2
(b)
x [m]
x [m]
Fig. 19 Mean velocity magnitude at a z / L = 0.4 and b z / L = 0.0; comparison of results using constant-coefficient and
dynamic Smagorinsky models
(a)
Const coeff model

Dynamic model
Const coeff mod

Dynamic model
0.08
0.1
0.08
0.06
0.04
0.02
0.7
0.8
0.9
1.1
0.07
0.06
0.05
0.04
0.7
1.2
0.8
0.9
1.1
1.2
(b)
x [m]
x [m]
Fig. 20 Mean velocity magnitude at a z / L = 0.875 and b z / L = 0.9375; comparison of results using constant-coefficient
and dynamic Smagorinsky models
(a)
Const coeff mod

Dynamic model
0.05
0.06
0.045
0.055
Const coeff model

Dynamic model
0.04
0.035
0.03
0.025
0.7
0.045
0.04
0.035
0.03
0.02
(a)
0.05
0.8
0.9
1.1
1.2
0.7
0.8
0.9
1.1
1.2
(b)
x [m]
x [m]
Fig. 21 rms velocity magnitude at a z / L = 0.4 and b z / L = 0.0; comparison of results using constant-coefficient and
dynamic Smagorinsky models
MAY 2009, Vol. 131 / 051402-13
Const coeff mod

Dynamic model
Const coeff model

Dynamic model
0.07
0.06
0.065
0.06
0.055
0.05
0.055
0.05
0.045
0.045
0.7
(a)
0.8
0.9
1.1
1.2
x [m]
(b)
0.04
0.7
0.8
0.9
1.1
1.2
x [m]
Fig. 22 rms velocity magnitude at a z / L = 0.875 and b z / L = 0.9375; comparison of results using constant-coefficient
and dynamic Smagorinsky models
near- and the far-field of the jet. Maximum deviations of about

2030% in the mean velocity magnitude occur near the inlet
where the largest dissipation rates are concentrated, at the z / L
= 0.4 and z / L = 0.0 stations notice the origin offset in the plots
near the ceiling, Figs. 20 and 22. Further from the inlet, as the jet
loses its strength, maximum deviations of about 1015% are seen
at the z / L = 0.875 and z / L = 0.9375 stations see Figs. 19 and 20.
Similar behavior of the rms of the velocity magnitude is observed
in Figs. 21 and 22, where at the z / L = 0.4 and z / L = 0.0 stations,
the maximum deviations are about 20%, whereas at the z / L
= 0.875 and z / L = 0.9375 stations, the deviations are about 10
15%. The differences described above are expected, as it was
shown that the subgrid-scale contribution to the total turbulent
kinetic energy is largest near the inlet and smallest near the ceiling
see Fig. 5a and 5b. These results demonstrate that the constant coefficient model agrees reasonably well with the dynamic
model, at least for the low-order statistics of the flow. This weak
dependence on the subgrid-scale model reflects the fact that the

large energy-containing scales of motion are well resolved.
4.5 k- Model Predictions at Low Inlet Turbulence Levels.
As already alluded to in Sec. 4.2, significant flow field differences
can potentially exist between the LES and the standard k- model
predictions when the flow is slowly developing in space or time.
In this section, we have systematically studied these model differences for the two extreme levels of inlet turbulence, i.e., 0% and
13%. For the 0% inlet turbulence case, LES predicts that the flow
develops slowly in space and, accordingly, the flow field is a
mixed laminar and turbulent state. For the 13% case, the jet rapidly transitions to a fully developed turbulent state close to the
inlet. Figures 23 and 24 compare the mean velocity magnitude
and turbulent kinetic energy from two simulations using the standard k- model with 0% and 13% inlet turbulent levels, respectively. It is seen that the solution is nearly insensitive to the level
k- model, 0% turbulence
0.08
0.2
0.15
0.1
0.05
0
0.7
(a)
0.8
0.9
x [m]
1.1
0.06
0.04
0.02
1.2
(b)
0.8
0.9
1.1
1.2
x [m]
Fig. 23 Mean velocity magnitude from k- model at a z / L = 0.6 and b z / L = 0.875
051402-14 / Vol. 131, MAY 2009
0.0012
0.0025
0.001
0.0008
T.K.E. [m /s ]
T.K.E. [m /s ]
0.002
0.0015
0.001
0.0006
0.0004
0.0005
0.0002
0.7
(a)
0.8
0.9
1.1
1.2
x [m]
0.7
(b)
0.8
0.9
1.1
1.2
x [m]
Fig. 24 Turbulent kinetic energy from k- model at a z / L = 0.6 and b z / L = 0.875
of inlet turbulence especially far from the inlet. The slow development of the jet predicted by LES in the 0% case is not observed
with the k- model. To emphasize this point, we compare the
solution from the k- and LES, using the constant coefficient
model, at 0% and 13% inlet turbulent levels. Figure 25 shows the
comparison for the mean velocity magnitude profile very close to
the inlet z / L = 0.8 at 0% and 13% inlet turbulent intensities. At
0% turbulence intensity, LES is predicting a flat velocity profile
with a sharp gradient separating the potential core from the quiescent fluid outside. The k- model, in contrast, predicts a much
more diffuse profile with curved shape similar to the profile for a
fully developed free jet. The two model predictions are in better
agreement at the 13% turbulence intensity, as the turbulence is
more nearly fully developed leading both models to predict a
smoother profile. The inability of the k- model to predict the
slow development of the jet at 0% inlet turbulence intensity leads
to an overprediction of the spreading rate. As discussed in Ref.
2, a well-known deficiency of the k- model is that it signifi-
cantly overpredicts the rate of spreading for the round jet. Although these differences could be reduced by adjusting the value
of C1 or C2, we did not feel that this would be justified. Figure
26 shows the mean velocity profile at a location well removed
from the inlet z / L = 0.75. The agreement between the LES and
k- predictions is better at the 13% inlet turbulence intensity level.
Nevertheless, it is clear that the k- model fails to capture the
correct qualitative behavior near the upper right corner x
1.22 m, as it shows a nearly stagnant region, whereas the LES
model predicts that there are small patches of intermittent activity
in the corner. Figure 27 shows a comparison of the resolved turbulent kinetic energy from the LES model to the turbulent kinetic
energy from the k- model at 0% and 13% inlet turbulent intensities. In the laminar inflow case, it is clear that LES is predicting
the correct physical behavior of the jet close to the inlet where
there is no active mechanism for turbulence generation yet at
z / L = 0.8 as the shear layer thickness is close to zero and the jet
LES model, 0% turbulence


0.2
0.15
0.1
0.05
0
0.7
(a)
0.2
0.8
0.9
x [m]
1.1
0.15
0.1
0.05
0
0.7
1.2
(b)
0.8
0.9
1.1
1.2
x [m]
Fig. 25 Mean velocity magnitude from k- and LES models at z / L = 0.8: a 0% and b 13% inlet turbulence intensities
MAY 2009, Vol. 131 / 051402-15


0.15
0.1
0.05
0
0.7
0.8
0.9
1.1
0.1
0.05
0
0.7
1.2
x [m]
(a)
0.15
0.8
0.9
1.1
1.2
x [m]
(b)
Fig. 26 Mean velocity magnitude from k- and LES models at z / L = 0.75: a 0% and b 13% inlet turbulence intensities
inflow is laminar. The k- model on the other hand is predicting

unrealistically high levels of turbulence near the inlet. In the 13%
turbulence intensity case, although the k- model qualitatively
captures the double-peaked profile, it still significantly overpredicts the turbulent kinetic energy relative to the LES model. Figure 28 presents similar comparisons at a location close to the
ceiling z / L = 0.75. As noted above, LES predicts a complicated
patch of activity close to the upper right corner dominated by
vortical flow. The k- model, in contrast, merely predicts a decaying jet and accordingly fails to predict the turbulent patch in the
corner, which is consistent with the underprediction of the level of
turbulence observed in Fig. 28.
Conclusions
The ventilation flow inside a simple cubicle was investigated

using LES with both a constant coefficient and dynamic model,
and the k- model. The effect of inlet turbulence intensity has
been investigated using idealized uniform mean inlet conditions
with 0%, 5%, and 13% inlet turbulence intensities. It is seen that
the 5% and 13% cases exhibit much faster spreading and mixing
with the room air than the laminar inflow case and that the statistics predicted in these two cases are close to each other, especially
far from the inlet. We conjecture that beyond a threshold level of
inlet turbulence intensity, the jet develops nearly independently of
the inlet turbulence intensity. This is consistent with the findings
of Jiang 44, Loomans 42, and Joubert et al. 43 who likewise
found little sensitivity to the inlet turbulence levels. However, the
laminar inlet flow case yielded significant deviations of the mean
flow and turbulence levels from the simulations with higher levels
of inlet turbulence. The results of flow simulations with experimentally measured profiles for the inlet conditions are compared
with those for the 13% case with uniform mean inlet profiles. It is
seen that near the inlet the flow is sensitive to the inlet flow
details, but these effects become less pronounced, although nonnegligible, further away from the inlet. The results of simulations
using the constant coefficient Smagorinsky and the dynamic LES


0.0035
0.003
0.003
0.0025
T.K.E. [m /s ]
2
2
T.K.E. [m /s ]
0.0025
0.002
0.0015
0.001
0.7
0.0015
0.001
0.0005
(a)
0.002
0.0005
0.8
0.9
x [m]
1.1
1.2
(b)
0.7
0.8
0.9
1.1
1.2
x [m]
Fig. 27 Turbulent kinetic energy from k- and LES models at z / L = 0.8: a 0% and b 13% inlet turbulence intensities
051402-16 / Vol. 131, MAY 2009
0.003
0.0025
0.0025
2
T.K.E. [m /s ]
0.002

0.003
T.K.E. [m /s ]

0.0015
0.0005
0.0005
(a)
0.0015
0.001
0.001
0.7
0.002
0.8
0.9
1.1
1.2
x [m]
(b)
0.7
0.8
0.9
1.1
1.2
x [m]
Fig. 28 Turbulent kinetic energy from k- and LES models at z / L = 0.75: a 0% and b 13% inlet turbulence intensities
models agree reasonably well especially far from the inlet where
most of the turbulent kinetic energy is resolved. The most significant differences are seen near the inlet, where the subgrid-scales
contribution to the total turbulent kinetic energy is as high as 30%
of the total turbulent kinetic energy. The k- model simulations
are compared at 0% and 13% inlet turbulence intensities. It is seen
that the k- model is nearly insensitive to the level of inlet turbulence especially far from the inlet. Finally, the results of simulations using the k- and LES models with 0% and 13% inlet turbulence intensities are compared. It is seen that the k- model
predictions for the 0% case are significantly different than the
LES predictions, as the k- model fails to capture the slow spatial
development of the jet into a fully turbulent state. At the 13%
turbulence level, it is seen that the k- model predicts the general
features of the mean velocity reasonably well. Even for this case,
the k- overpredicts the spreading rate of the jet relative to the
LES model and fails to capture the complicated unsteady flow
pattern near the ceiling leading to significant overprediction of the
turbulent kinetic near the inlet and underprediction of it near the
ceiling.
Acknowledgment
This work was supported in part by the Environmental Protection Agency, through the Syracuse University NY STAR Center
for Environmental Quality Systems/EPA Indoor Environmental
Research Program Collaboration.
References
1 Spengler, J. J., Samuel, J. M., and McCarthy, J. F., 2001, Indoor Air Quality
Handbook, McGraw-Hill, New York.
2 Pope, S. B., 2000, Turbulent Flows, Cambridge University Press, Cambridge.
3 Wilcox, D. C., 2004, Turbulence Modeling for CFD, DCW Industries, Inc., La
Canada, CA.
4 Smagorinsky, J., 1963, General Circulation Experiments With the Primitive
Equations: I. The Basic Equations, Mon. Weather Rev., 91, pp. 99164.
5 Lilly, D. K., 1967, The Representation of Small-Scale Turbulence in Numerical Simulation Experiments, Proceedings of the IBM Scientific Computing
Symposium on Environmental Sciences, IBM Form 320-1951, pp. 195210.
6 Deardorff, J. W., 1974, Three-Dimensional Numerical Study of the Height
and Mean Structure of a Heated Planetary Boundary Layer, Boundary-Layer
Meteorol., 7, pp. 81106.
7 Schumann, U., 1975, Subgrid-Scale Model for Finite Difference Simulations
of Turbulent Flows in Plane Channels and Annuli, J. Comput. Phys., 18, pp.
376404.
8 Pope, S. B., 2004, Ten Questions Concerning the Large-Eddy Simulation of
Turbulent Flows, New J. Phys., 635, pp. 124.
9 Germano, M., Piomelli, U., Moin, P., and Cabot, W. H., 1991, A Dynamic
Subgrid-Scale Eddy Viscosity Model, Phys. Fluids A, 3, pp. 17601765.

10 Lilly, D. K., 1992, A Proposed Modification of the Germano Subgrid-Scale
Closure Method, Phys. Fluids A, 4, pp. 633635.
11 Meneveau, C., Lund, T. S., and Cabot, W. H., 1996, A Lagrangian Dynamic
Subgrid-Scale Model of Turbulence, J. Fluid Mech., 319, pp. 353385.
12 Nielson, P. V., Restive, A., and Whitelaw, J. H., 1978, The Velocity Characteristics of Ventilated Room, ASME J. Fluids Eng., 1009, pp. 291298.
13 Murakami, S., Kato, S., and Suyama, Y., 1992, Three Dimensional Numerical
Simulation of Turbulent Airflow in a Ventilated Room by Means of a TwoEquation Model, ASHRAE Trans., 981, pp. 8297.
14 Chen, Q., 1995, Comparison of Different k- Models for Indoor Airflow
Computations, Numer. Heat Transfer, Part B, 28, pp. 353369.
15 Nielson, P. V., 1998, Selection of Turbulence Models for Prediction of Room
Airflow, ASHRAE Trans., 104, pp. 11191127.
16 Chen, Q., and Jiang, Z., 1992, Significant Questions in Predicting Room Air
Motion, ASHRAE Trans., 981, pp. 929939.
17 Davidson, L., and Nielson, P. V., 1996, Large Eddy Simulations of the Flow
in Three-Dimensional Ventilated Room, Proceedings of the Fifth International
Conference on Air Distribution in Rooms, Yokohama, Japan, pp. 161168.
18 Zhang, W., and Chen, Q., 2000, Large Eddy Simulation of Indoor Airflow
With a Filtered Dynamic Subgrid Scale Model, Int. J. Heat Mass Transfer,
43, pp. 32193231.
19 Jiang, Y., and Chen, Q., 2003, Using Large Eddy Simulation to Study Airflows in and Around Buildings, ASHRAE Trans., 1092, pp. 517526.
20 Zhang, W., and Chen, Q., 2000, Large Eddy Simulation of Natural and Mixed
Convection Airflow Indoors With Two Simple Filtered Dynamic Subgrid Scale
Models, Numer. Heat Transfer, Part A, 375, pp. 447463.
21 Shah, K. B., and Ferziger, J. H., 1995, A New Non-Eddy Viscosity Sub-GridScale Model and Its Application to Channel Flow, Annual Research Briefs,
Center for Turbulence Research, Stanford, CA.
22 Zhai, Z., Zhang, Z., Zhang, W., and Chen, Q., 2007, Evaluation of Various
Turbulence Models in Predicting Airflow and Turbulence in Enclosed Environments by CFDPart 1: Summary of Prevalent Turbulence Models,
HVAC&R Res., 136, pp. 853870.
23 Zhang, Z., Zhang, W., Zhai, Z., and Chen, Q., 2007, Evaluation of Various
Turbulence Models in Predicting Airflow and Turbulence in Enclosed Environments by CFDPart 2: Comparison With Experimental Data From the
Literature, HVAC&R Res., 136, pp. 871886.
24 Davidson, L., Nielsen, P., and Sveningsson, A., 2003, Modification of the v2f
Model for Computing the Flow in a 3D Wall Jet, Turbulence Heat and Mass
Transfer, K. Hanjalic, Y. Nagano, and M. J. Tummers, eds., Begell House,
Inc., New York, Wallingford, UK, Vol. 213, pp. 6572.
25 Yakhot, V., and Orszag, S. A., 1986, Renormalization Group Analysis of
Turbulence, J. Sci. Comput., 1, pp. 351.
26 Launder, B. E., and Sharma, B. I., 1974, Application of the Energy Dissipation Model of Turbulence to Calculation of Flow Near a Spinning Disk, Lett.
Heat Mass Transfer, 1, pp. 131138.
27 Durbin, P. A., 1995, Separated Flow Computations With the k--v2 Model,
AIAA J., 33, pp. 659664.
28 Chen, Q., and Srebric, J., 2002, A Procedure for Verification, Validation, and
Reporting of Indoor Environment CFD Analyses, HVAC&R Res., 82, pp.
201216.
29 Chen, Q., and Zhai, Z., 2004, The Use of CFD Tools for Indoor Environmental Design, Advanced Building Simulation, A. Malkawi and G. Augenbroe,
eds., Spon, New York, pp. 119140.
MAY 2009, Vol. 131 / 051402-17
30 Fluent Inc., 2006, FLUENT Users Guide, NH, Lebanon.

31 Launder, B. E., and Spalding, D. B., 1972, Lectures in Mathematical Models of
Turbulence, Academic, New York.
32 Wolfshtein, M., 1969, The Velocity and Temperature Distribution of OneDimensional Flow With Turbulence Augmentation and Pressure Gradient, Int.
J. Heat Mass Transfer, 12, pp. 301318.
33 Kader, B., 1981, Temperature and Concentration Profiles in Fully Turbulent
Boundary Layers, Int. J. Heat Mass Transfer, 249, pp. 15411544.
34 Germano, M., 1992, Turbulence: The Filtering Approach, J. Fluid Mech.,
286, pp. 229255.
35 Marr, D., 2007, Velocity Measurements in the Breathing Zone of a Moving
Thermal Manikin Within the Indoor Environment, Ph.D. thesis, Syracuse
University, Syracuse.
36 Friedrich, R., and Arnal, M., 1990, Analysing Turbulent Backward-Facing
Step Flow With the Lowpass-Filtered NavierStokes Equations, J. Wind.
Eng. Ind. Aerodyn., 35, pp. 101128.
37 Le, H., Moin, P., and Kim, J., 1997, Direct Numerical Simulation of Turbulent Flow Over a Backward-Facing Step, J. Fluid Mech., 330, pp. 349374.
38 Mathey, F., Cokljat, D., Bertoglio, J. P., and Sergent, E., 2006, Assessment of
the Vortex Method for Larged Eddy Simulation Inlet Conditions, Prog. Comput. Fluid Dyn., 6, pp. 5867.
051402-18 / Vol. 131, MAY 2009
39 Lee, S., Lele, S., and Moin, P., 1992, Simulation of Spatially Evolving Turbulence and the Applicability of Taylors Hypothesis in Compressible Flow,
Phys. Fluids A, 4, pp. 15211530.
40 Smirnov, A., Shi, S., and Celik, I., 2001, Random Flow Generation Technique
for Large Eddy Simulation and Particle Dynamics Modeling, ASME J. Fluids
Eng., 123, pp. 359371.
41 Mathey, F., Cokljat, D., Bertoglio, J. P., and Sergent, E., 2003, Specification
of LES Inlet Boundary Condition Using Vortex Method, Fourth International
Symposium on Turbulence, Heat and Mass Transfer, K. Hanjalic, Y. Nagano,
and M. Tummers, eds., Begell House, Inc., Antalya, Turkey.
42 Loomans, M., 1998, The Measurement and Simulation of Indoor Air Flow,
Ph.D. thesis, Eindhoven University of Technology, Eindhoven, The Netherlands.
43 Joubert, P., Sandu, A., Beghein, C., and Allard, F., 1996, Numerical Study of
the Influence of Inlet Boundary Conditions on the Air Movement in a Ventilated Enclosure, Proceedings of the Roomvent, Yokohama, Japan, Vol. 1, pp.
235242.
44 Jiang, J., 2007, Experimental and Numerical Study of Air Flows in a FullScale Room, Ph.D. thesis, University of Illinois at Urbana-Champaign, Urbana.
45 Marr, D., 2007, private communication.
H. Zidouh
L. Labraga1
e-mail: llabraga@univ-valenciennes.fr
M. William-Louis
Universit Lille Nord de France,
F-59000 Lille, France;
UVHC, LME,
59313 Valenciennes, France
Unsteady Wall Shear Stress in

Transient Flow Using
Electrochemical Method
Experimental measurements of the wall shear stress combined with those of the velocity
profiles via the electrochemical technique and ultrasonic pulsed Doppler velocimetry are
used to analyze the flow behavior in transient flows caused by a downstream short pipe
valve closure. The Reynolds number of the steady flow based on the pipe diameter is
Re 148,600. The results show that the quasisteady approach of representing unsteady
friction is valid during the initial phase for relatively large decelerations. For higher
decelerations, the unsteady wall shear stress is consistently higher than the quasisteady
values obtained from the velocity profiles. Attention has been focused on the friction
acceleration model. The results obtained from this study show the ability of the electrochemical method in determining the local unsteady wall shear stress even in severe
decelerating transient flows. DOI: 10.1115/1.3112387
Introduction
Transient flows associated with the water hammer phenomenon

are commonly encountered in both natural and engineering systems. Examples include water supply and distribution system, oil
transportation systems, and human arterial network. The dramatic
changes in velocity and pressure arising from these transient
events can cause pipe breakages, flooding, and other damage hazards.
In engineering analysis of such flows, it is assumed that phenomenological expressions relating the wall shear stress to cross
sectional averaged velocity in steady-state flows remain valid under unsteady conditions. In other words, a new Reynolds number
is computed each time the velocity is altered and the wall shear
stress is then adjusted to the value corresponding to stationary
flow at the new Reynolds number. For example, the form of the
DarcyWeisbach equation used in water hammer models is
wt =
tVtVt
2
where wut is zero for steady flows, small for slow transients,
and significant for fast transient flows. Daily et al. 1 conducted
laboratory experiments and found wut to be positive for accelerating flows and negative for decelerating flows. They argued
that during acceleration, the central portion of the stream moved
somehow bodily so that the velocity profile steepened, giving
1
JOURNAL OF FLUIDS ENGINEERING. Manuscript received August 18, 2008; final manuscript received February 24, 2009; published online April 15, 2009. Assoc. Editor:
Hassan Peerhossaini.
w = qs +
kD dV
8 dt
where k, an empirical parameter, needs be determined either from

experiments or analysis. Daily et al. 1 showed that k = 0.01 for
accelerating flows and k = 0.62 for decelerating flows. The research of Shuy 2 led to k = 0.0825 for accelerating flows and
k = 0.13 for decelerating flows. This illustrates that empirical
constant k is flow case dependent. To explain these conflicting
results, Vardy and Brown 3 argued that the different behaviors
observed by the authors may be attributed to different time-scales.
The most successful formulation of this approach is the expression proposed by Brunone et al. 4 as follows:
where is the friction factor. The application of such a simplified

wall shear stress model is satisfactory only for very slow transients, in which the shape of the instantaneous velocity profiles
does not differ markedly from the corresponding steady-state
ones. During fast transients or high frequency periodic flows, on
the other hand, velocity profiles change, in particular, and more
complex patterns, featuring greater gradients, hence, greater shear
stresses, than the corresponding steady flow values.
A simple modification of Eq. 1 involves the introduction of an
unsteady component wut such that
wt = qst + wut
higher shear. The relation postulated by Daily et al. 1 can be

reformulated as follows:
w = qs + k3
V
R V
a
x
2 t
where the unsteady friction term depends on both instantaneous

local acceleration and instantaneous convective acceleration. The
measurement of the unsteady wall shear stress so far was conducted in previous studies by measuring the wall drag force 2 or
by determining the transient friction coefficient from the instantaneous mean flow velocity 5,6. Whereas the above mentioned
approaches are acceptable for flows accelerating or decelerating
slowly at a uniform rate, they are questionable during fast transients. Moreover, the studies so far performed about transient
flows are limited to quite narrow accelerations and deceleration
ranges, and the results obtained by these studies are very different
from one another.
From this brief review, it is shown that a critical lack of characterizing turbulent transient flow is a more precise knowledge of
the local unsteady wall shear. It is shown that no conclusive result
was obtained because of the scarcity of experimental data available. A nonintrusive, local, and quantitative method is actually
greatly required to determine the important unsteady phenomena
involved under transient conditions. The primary aim of this work
was to examine the possibility of measuring the local unsteady
wall shear stress within a decelerating turbulent flow in a short
circular pipe. This is made possible by the use of the electrochemical method combined with the ultrasound velocimetry. Although the pipe was rather short, it was able to provide fast tran-
MAY 2009, Vol. 131 / 051403-1
Fig. 1 Concentration fields and velocity profile over a single

probe
sient flows characterized by very high decelerations and complex

shapes of instantaneous velocity profiles featuring annular effects
leading to local flow reversal.
Theory
Shear stress is determined experimentally using an electrochemical method based on the reduction in the ferricyanide ion on
an electrode surface. A thorough review of this technique is provided by Hanratty and Campbell 7. This method is based on the
determination of the limiting diffusion current at the surface of an
electrode under conditions for which the chemical reaction rate is
fast enough so that the concentration of the reacting ions is zero at
the surface of the working electrode. If the length of this electrode
is very small in the flow direction, the concentration boundary
layer on the electrode is thin; thus the flow velocity varies linearly
throughout the thickness of this boundary layer.
A diagram of a single electrochemical probe is given in Fig. 1.
The current I flowing to the test electrode of area A is related to
the mass transfer coefficient K by
K=
I
AneFC0
where ne is the number of electrons involved in the reaction, F is

the Faraday constant, and C0 is the bulk concentration of active
species. If the dimensionless velocity gradient at the wall S+
= S2 / D where is electrode length and D is diffusion coefficient is high enough, it is then possible to neglect the longitudinal
diffusion, and a solution of the steady-state mass balance equation
gives
K+ =
K
= cS+1/3
D
The integration of the mass transfer equation by several authors

gives a theoretical value for the coefficient c in the case of a
rectangular probe, c = 0.807.
When flush mounted mass transfer probes are used to measure
wall shear rate of unsteady flows, featuring low velocity amplitude variations, a pseudo-steady-state assumption can be made
and the one-third-power law can be used to compute the instantaneous wall shear rate 8. However, in many cases, the concentration boundary layer inertia cannot be neglected and the quasisteady-state assumption is not valid. Most studies of the dynamic
behavior of electrochemical probes assume that the amplitude of
the shear rate fluctuations is small compared with the time averaged shear rate. With this assumption, the mass balance equation
can be linearized and solved analytically 9 and numerically
10,11. Then, the transfer function between the mass flux and the
wall shear rate at different frequencies can be deduced. The difficulty of this approach is that the linearization assumption is not
051403-2 / Vol. 131, MAY 2009
Fig. 2 Pulsed ultrasonic Doppler velocimetry technique: measurement setup
always valid. For example, in flows involving large amplitude

oscillations, high frequency velocity variations, or reverse flows,
the above approach cannot apply.
Sobolik et al. 12 introduced a technique based on the correction of the wall shear rate obtained from the Levque solution by
adding a term estimated from the known solution of the unsteady
diffusion at the beginning of the potential transient process. They
established that the corrected wall shear rate Sct is
2 Sqt
Sct = Sqt + t0
3
t
where t0 = 0.4862/3D1/3S2/3
and Sqt = D / 2K+t / 0.8073.
q
This method is limited both to rather moderate frequencies and
nonreverse unsteady flows. In the presence of large unsteady or
reverse flows, the inverse method 1315 seems to be more appropriate to calculate the wall shear rate from the mass transfer
signal measured. This technique consists in solving the direct
convection-diffusion equation and in estimating sequentially the
unknown wall shear rate by minimizing the difference between
measured and simulated mean concentration gradients on the
probe surface. In this study both the inverse method and that
suggested by Sobolik et al. 12 were applied for the determination of the true wall shear rate of the transient turbulent flow in
a pipe.
The ultrasonic pulsed Doppler velocimetry 1618 was used
for velocity profile measurements. Figure 2 shows a common configuration of the ultrasound beam and flow. This method is based
on detecting and processing the backscattered echoes originating
from moving targets suspended in the flowing liquid that is to be
investigated. This technique provides measurements of local onedimensional velocity and related distance from the transducer,
leading to the construction of an almost instantaneous velocity
profile along the acoustic beam. Information on the position from
which the ultrasound is reflected is extracted from the time delay
TD between the start of the pulse burst and its reception as Y
= aTD / 2.
If the particle is moving at an angle regarding the axis of the
ultrasonic beam, its velocity u is measured by computing the
variation of its depth between two emissions separated by the time
interval between two pulses as
u=
fD
a
2f cos
where f is the emitting frequency, and f D is the Doppler frequency

shift that is due to the particle motion. According to the Nyquist
theorem, a maximum detectable Doppler frequency exists for each
pulse repetition frequency f prf, over which measurements are
affected by aliasing
Table 1 Physical properties of the electrochemical solution at

25 C
Density, kg/ m3
Kinematic viscosity, m2 s1
Diffusivity, D m2 s1
1024
1.025 106
7.45 1010
The chemical reaction at the electrodes is described as

4
FeCN3
6 + e FeCN6
Fig. 3 Experimental setup
f D max f prf/2
Equation 9 leads to a maximum detectable velocity for each

pulse repetition frequency
umax =
a
4Tprf f cos
10
In addition to the velocity, the maximum measurement depth is

determined by the repetition frequency of the pulses according to
the relationship Y max = a / 2f prf. This depth can be reduced in order to increase the spatial resolution.
Material and Method
Experiments were conducted in a vertical water channel Fig.

3. The main component is the L = 2.62 m long polypropylene
pipe with a 61.4 mm internal diameter and an e = 6.8 mm wall
thickness. The pipe is connected to a supply and recycling system.
A constant level tank is used to keep upstream pressures constant
throughout measurements. A butterfly valve at the end of the pipe
is used to create controlled opening and closing actions by adjusting the discharge into the free surface tank. More details of the
experimental apparatus are reported in Ref. 19. The working
fluid is either water seeded with copolyamide particles for the
velocity profile measurement or an electrochemical solution for
the wall shear stress measurement.
Pressure, velocity, and wall shear stress are measured at a distance of 3.27 m from the free surface.
The electrolytic solution is a mixture of potassium ferricyanide
10 mol/ m3, potassium ferrocyanide 10 mol/ m3, and potassium sulfate. The potassium sulfate acts as a low resistance vehicle for current flows and ensures that the transfer at the cathodic
surface is controlled by diffusion only.
The conversions of ferrycyanide and ferrocyanide that take place

between the cathode and the anode maintain a constant concentration of both chemicals. The electric current I flowing between
the electrodes is related to the transfer coefficient according to Eq.
5.
Two kinds of probes were used: single circular probe and sandwich rectangular probes. The latter is used to sense reversal flows.
The test electrodes were constructed by inserting circular platinum
wires or platinum sheet through the wall and gluing them in place.
The protruding metal was smoothed flush with the pipe wall. The
platinum wire was 0.4 mm in diameter. The sandwich probe was
made with two platinum sheet strips. The probe is 0.5 mm long,
0.1 mm wide, and with a 0.008 mm gap. The platinum sheets and
the cellophane tape are glued together and inserted into a Perspex
block. The block is then mounted flush inside a Plexiglas module
inserted in the pipe between two collars. The polarizations curves,
not reported here, show that the electrode is polarized by an applied voltage of 200 to 600 mV over a Reynolds-number
range of about 35,000100,000. An operating voltage of 400
mV was selected.
The active surface of the electrochemical probes is determined
with the Cotrell method. This technique is based on the study of
the transient response of the probes to a voltage step from 0 to the
diffusional plateau potential 20; before the stationary regime is
obtained, the current decreases following the Cotrell asymptote.
During this period of time, the evolution of the electrical intensity
as a function of t1/2 is linear. The slope permits to deduce the
active electrode area and therefore the value of the effective diameter of the probe.
All the experiments were carried out at 25 C. Under these
conditions the physical properties of the electrochemical solution
take the values given in Table 1. According to these values, the
Schmidt number is 1375.
Two pressure transducers of stainless diaphragm type with a
recording range of 070 m of water, an accuracy of 0.5%, and of
the full-range scale are mounted at the L / 2 and 3L / 4 sections
along the pipeline.
The ultrasound Doppler velocimeter was a DOP2000 from Signal Processing SA with a 4 MHz ultrasound transducer. The transversal dimension of the sampling volume is determined by the
ultrasonic field geometry. The longitudinal dimension, however,
depends on the burst length and on the bandwidth of the receiving
unit. The minimum value of the sampling time is about 4 ms. The
ultrasonic field continuity between the probe and the pipe was
maintained with a coupling gel. The basic characteristics of the
ultrasound Doppler velocimetry used in this study are presented in
Table 2.
Table 2 Ultrasound Doppler velocimeter specification

Emitting frequency, f e MHz
Emitting PRF, f kHz
Radius of the transducers piezo mm
The angle of incidence, deg
Longitudinal resolution mm
4
10
5
68
0.69
MAY 2009, Vol. 131 / 051403-3
Fig. 4 Synoptic of the acquisition system
Simultaneous pressure and wall shear stress data acquisition is

realized with a sample and hold device. Digitizing is done by a 12
bit A/D converter. The reference time is based on the electromechanical valve closure. The starting time t = 0 is set by switching
off a dc power supply that simultaneously triggers the valve closure and acquisition of the wall shear stress or velocity profiles
and the pressure signals.
The sampling frequency of pressure and electrochemical signals was 10 kHz during 250 ms. The relatively high sampling rate
is needed for the electrochemical signal dynamic response to be
accurately corrected. When the measurement is finished, data are
read by a PC high-speed data bus for final processing and storage.
The detailed synoptic of the acquisition system is presented in
Fig. 4.
Before undertaking unsteady flow measurements, a preliminary

set of steady-state measurements was carried out for different
Reynolds numbers. The velocity measurements were taken at a
distance of 1.945 m from the inlet of the tube. The velocity profiles provided by the ultrasonic pulsed Doppler velocimetry
UDV technique were corrected in order to remove bias errors
due to the crossing of media with different acoustic properties,
ultrasonic field shapes, intensity spatial variability, and finite dimension of sampling volume especially in areas close to the wall.
The correction procedure suggested by Wunderlich and Brunn
21 was applied to the data. The averaged velocity profiles were
obtained from instantaneous measurements, during 100 s of acquisition. Numerous experiments including velocity profiles, pressure, and wall shear stress in unsteady flow conditions caused by
a fast valve closure were conducted. However, due to the lack of
space, only a few representative results are reported hereby.
The experiment is being carried out in two phases. First, the
instantaneous velocity profiles and pressure were simultaneously
collected in seeded water. The time origin corresponds to the start
of the maneuver initiated by the electromechanical valve signal.
Second, since wall shear stress data required specific electrochemical solution, they were acquired later on in the same time
base; thanks to the electromechanical valve signal that triggers the
data acquisition system therefore allowing perfectly repeatable
tests.
Figure 5 shows the response of a pressure transducer and six
Fig. 5 Experimental results showing pressure time history and associated velocity profiles
051403-4 / Vol. 131, MAY 2009
Fig. 6 Mean velocity during deceleration

Fig. 7 Instantaneous deceleration
corresponding velocity profiles during the transient events. A

rapid decay of pressure peaks after the end of a complete valve
closure is observed. The transient event associated with Fig. 5 has
the following characteristics: The initial mean flow velocity is
2.42 m/s, and the valve is operated from a fully open to a fully
closed position in about 70 ms. The time history of the pressure
and the velocities are measured at a distance of 0.675 m upstream
from the valve. The time origin, t = 0, corresponds to the start of
the maneuver of the valve. The points labeled with small numbers
on the time history in the upper plot label the instants corresponding to the displayed velocity profiles. The pressure wave celerity
was calculated from the recorded data based on the analysis in the
time and frequency domains of the pressure waveform. The mean
value of this quantity amounted to about a = 500 m / s. This mean
pressure wave speed is also calculated from the following analytical solution:
a=
Kw p
Kw D
1 + 1 p
E e
11
where Kw, p, p, E, and e are bulk modulus, Poissons coefficient,

mass density, Youngs modulus, and pipe-wall thickness, respectively. Equation 11 gives a pressure wave celerity value of 496
m/s, in good agreement with the experimental findings. The velocities were obtained by time-averaging 50 velocity profiles.
The velocity profiles confirm the complex features of the flow
field as it was suggested by Brunone et al. 4. During transient
flow the instantaneous velocity profiles deviate significantly from
the corresponding steady-state as it is shown in Fig. 5 4. A
reverse flow occurs within a near wall annulus corresponding to
0.1 y / R 0.2 Fig. 5, 5 and 6 and may be interpreted by the
so called annular effect that is different from that usually found in
previous studies, characterized by negative shear stresses at the
wall. For this particular event, it can be found that the mean velocity is approximately null while the velocity gradient near the
wall is substantial. In other words, under such a condition, even
with a significant wall shear stress, steady or quasisteady criteria
would give a null friction term.
The time dependence of the sectional mean velocity is shown in
Fig. 6. The mean velocity exhibits several phases. It decreases
slowly from 0 s to 0.04 s and undergoes a steep decrease in the
time range from 0.05 s to 0.07 s. The minimum mean velocity
corresponding to the highest pressure peak results from the reflected wave from the valve. A slight increase in the instantaneous
mean velocity occurring at about 0.085 s is strictly linked to the
minimum pressure value.
The instantaneous acceleration is shown in Fig. 7. The acceleration is negative throughout the transient event except in the
0.075 t 0.085 time range corresponding to a steep decrease in
the pressure. The present values of the acceleration range from
0 ms2 to 100 ms2, much higher values indeed if compared
with those found in literature 2,4,6.
Preliminary measurements of the wall shear stress performed

with a split probe have shown that no reverse flow occurs after a
fast valve closing within the measurement section. This result was
made possible, thanks to the electrochemical method, and could
not be found easily from the velocity profiles provided by UDV
technique.
The unsteady wall shear stress derived from the Levque solution is compared with that corrected with the inverse method and
the method of Sobolik et al. 12 Fig. 8. The wall shear stress is
normalized with that obtained at t = 0 at the beginning of the transient flow. It is clear from Fig. 8 that the probe inertia correction
is necessary, especially in the part of the transient event where the
shear rate variation is significant. Figure 8 shows that the method
of Sobolik et al. 12 gives nearly the same instantaneous wall
shear rate. This result is not so surprising because large Peclet
numbers are observed during the transient flow. Moreover, no reverse flow occurs during the valve closure. For this particular
unsteady flow, the method of Sobolik et al. 12 is quite suitable as
compared with the inverse method, because no additional parameters are required and no initial guess of the wall shear stress is
needed.
Figure 9 shows the measured wall shear stress via the electrochemical technique. About 40 tests were performed for the measurement of the transient wall shear stress. The wall shear stress is
normalized with that obtained at t = 0 at the beginning of the transient event, therefore removing uncertainties due to the difficulty
to estimate the actual probe surface. In the same figure, the absolute pressure and the normalized wall shear stress predicted by
using a quasisteady assumption are displayed. Here we define the
quasisteady friction coefficient q using the instantaneous Reynolds number as follows:
Fig. 8 Comparison between wall shear stress obtained with

different methods
MAY 2009, Vol. 131 / 051403-5
Fig. 9 Wall shear stress and pressure distributions during

deceleration
Fig. 10 Unsteady
acceleration
q =
64
laminar
Re
0.316
turbulent
R0.25
e
12
The results show that the wall shear stress measured from the
electrochemical signals oscillates around that predicted by a quasisteady model for the time range between t = 0 s, corresponding
to the start of the valve closure, up to t = 0.055 s. The ability of
the electrochemical method to sense the local instantaneous flow
unsteadiness is clearly shown in Fig. 9. Indeed, the wall shear
stress exhibits a wavy behavior probably due to the pressure wave
originating from the valve closure. As for the quasisteady model,
it tends to smoothen out the actual wall shear stress fluctuations.
Shuy 2 found that the unsteady wall shear stress is higher than
the quasisteady values computed from the KarmanNikuradse
equation based on instantaneous velocity, for a value of the acceleration parameter = 2D / sV2dV / dt up to 2. The values of
the acceleration parameter of the present study range within 0
100 for t 0.05 s. For this severe decelerating flow, the
present results show that the unsteady wall shear stress may be
either greater or lower than that obtained from a quasisteady approach, following the local fluctuating acceleration with a phase
shift, as it is shown below.
For 0.06 s t 0.07 s, the quasisteady assumption is no
longer valid to determine the wall shear stress. The effective wall
velocity gradient is considerably higher than the one resulting
from the pseudosteady approach. The unsteady wall shear stress
inferred from the electrochemical probe increases up to a maximum value corresponding to that of the pressure. It is shown that
the maximum unsteady wall shear stress value reaches the steady
one at the end of this phase.
For t 0.07 s, the unsteady wall shear stress tends asymptotically toward zero with large oscillations resulting from all reflected pressure waves.
Figure 10 gives a synthesis of the results of both unsteady wall
shear stress component and acceleration wu. According to Eq.
3, it is expected that the changes in the unsteady wall shear
stress component wu correspond closely to those of the acceleration. It is shown that the sudden change in the core of the pipe first
affects the acceleration before unsteady wall shear stress is
changed. The change in flow rate is then transferred toward the
wall by viscous shear. As a result, changes in the velocity field of
the fluid adjacent to the pipe wall, and in resulting wall shear
stresses, lag behind flow rate changes in the core of the pipe 22.
This should explain the discrepancies in phase shift between experimental and numerical head traces obtained in some previous
studies with the friction model based on instantaneous local acceleration for later time of the transient flows.
051403-6 / Vol. 131, MAY 2009
wall
shear
stress
component
and
Axworthy et al. 23 suggested that the unsteady friction formulas based on instantaneous accelerations such as Eq. 3 are
applicable to transient flow problems in which the wave passage
time-scale L / a = 0.0052 s in the present study is significantly
shorter than the vorticity diffusion time-scale 2D / 2u = 0.46 s
in the present study. According to Axworthy et al. 23, the
changes in wall shear stress in the conditions mentioned above
should correspond closely to the acceleration. In the light of the
present study, it can be stated that the time-scale arguments do not
allow any definitive conclusion to be made to clarify the applicability of the instantaneous-acceleration approach in both attenuation and phase shift of the transient wall shear rate.
Shuy 2 rewrote Eq. 3 in the following dimensionless form:
k
w
=1+
ws
2
13
It seems interesting to check the validity of Eq. 13 using the

present data. Figure 11 shows the dependence of the ratio between
unsteady and steady wall shear stresses on the acceleration parameter .
For 40, corresponding to the beginning of the transient,
the ratio between unsteady and steady wall shear stresses scatters
about a mean value of 1, confirming a quasi-steady-state behavior
observed in Fig. 9. Over the range of 40 150 the data take
a nearly constant value and show clearly that the unsteady wall
shear stress is lower than the pseudosteady one.
For 150 the data are remarkably aligned and correspond to
an unsteady wall shear stress that is higher than the quasisteady
one. The highest acceleration parameter 400 is reached at t
0.06 s just before the complete valve closure. The acceleration
parameter of the present study takes very high values because of
Fig. 11 Dependence of the ratio between unsteady and steady

wall shear stresses on the acceleration parameter
and energy dissipation. Further experimental measurements

should be performed in a longer pipe allowing both short and long
time-scales to take advantage of this powerful method.
Nomenclature
a
A
C
d
D
F
f
Fig. 12 Dependence of k on the acceleration parameter
the short length of the pipe and the relatively high Reynolds number. The electrochemical technique is actually the only method
able to measure with a fine time resolution the unsteady wall shear
stress in these severe experimental conditions, not found in literature.
The computed values of the empirical coefficient k, obtained
from Eq. 13, are displayed against in Fig. 12. The values of k
are consistent with the distribution of the ratio between unsteady
and steady wall shear stresses analyzed above. The present results
show that the values of k are rather low compared with those
found in literature. Moreover, k is variable in time and therefore
depends on the acceleration parameter as it was already suggested
by Axworthy et al. 23.
Further experiments are performed in order to check the validity of the well known model of Brunone 4 Eq. 4, the most
widely used due to its ability to produce reasonable agreement
with experimental pressure history.
Conclusion
The electrochemical technique is combined with ultrasonic

pulsed Doppler velocimetry for measurements of the unsteady
transient wall shear stress and velocity profiles. Acquired unsteady
wall shear stress by using the electrochemical method provides
complementary and additional near wall information to the velocity profiles and demonstrates the complex nature of the near wall
flow field.
The conclusions are summarized as follows.
1 The dynamic response of the electrochemical probes is being corrected by using the inverse method and the method
of Sobolik et al. 12. Both methods provide remarkable
similar results despite complex flow conditions characterized by very high decelerations.
2 The velocity profiles measured with ultrasonic pulsed Doppler velocimetry clearly show, during transient events, regions of reverse flow occurring at a finite distance from the
wall and associated with high values of the velocity gradient at the wall.
3 The quasisteady approach for predicting the wall shear
stress is valid on average for a wide ranging acceleration
parameter, much more higher than that found in the previous studies.
4 The unsteady wall shear stress component lags behind the
instantaneous acceleration, suggesting that the average velocity field responds more quickly to changes in the flow at
the core of the pipe.
For more meaningful progress in the numerical modeling of
unsteady flows, much more series of experimental tests must be
first carried out, using a reliable technique for the measurement of
the wall shear stress. In the light of this study, the electrochemical
method seems to be the most appropriate technique that could
contribute to a better understanding of transient flow dynamics
Fprf
I
k
K
ne
p
R
Re
S
t
Tprf
V
sound velocity m/s

active surface of the probe m2
bulk concentration mol/ m3
pipe diameter m
diffusion coefficient m2 / s
Faraday number
frequency Hz
pulse repetition frequency Hz
limiting diffusion current A
empirical coefficient of unsteady friction
transfer coefficient m/s
probe width m
number of electrons exchanged in a reaction
fluid pressure m
pipe radius m
Reynolds number: Vd /
velocity gradient at the wall 1/s
time s
1 / f PRF s
mean axial velocity m/s
Special Characters
friction factor
fluid density kg/ m3
w wall shear stress Pa
ws quasisteady wall shear stress Pa
wu unsteady wall shear stress Pa
References
1 Daily, J. W., Hankey, W. L., Olive, R. W., and Jordan, J. M., 1956, Resistance
Coefficient for Accelerated and Decelerated Flows Through Smooth Tubes and
Orifices, Trans. ASME, 78, pp. 10711077.
2 Shuy, E. B., 1996, Wall Shear Stress in Accelerating and Decelerating Turbulent Pipe Flows, J. Hydraul. Res., 342, pp. 173183.
3 Vardy, A., and Brown, J. M. B., 1997, Discussion on Wall Shear Stress in
Accelerating and Decelerating Pipe Flows, J. Hydraul. Res., 351, pp. 137
139.
4 Brunone, B., Golia, U. M., and Greco, M., 1991, Some Remarks on the
Momentum Equation for Transients, International Meeting on Hydraulic
Transients With Column Separation, Ninth Round Table, IAHR, Valencia,
Spain, pp. 201209.
5 Carstens, M. R., and Roller, J. E., 1959, Boundary-Shear Stress in Unsteady
Turbulent Pipe Flow, ASCE J. Hydr. Div., 85HY2, pp. 6781.
6 Kurokawa, J., and Morikawa, M., 1986, Accelerated and Decelerated Flows
in a Circular Pipe 1st. Report, Velocity Profiles and Friction Coefficient,
Bull. JSME, 29249, pp. 758765.
7 Hanratty, T. J., and Campbell, J. A., 1983, Measurement of Wall Shear
Stress, Fluid Mechanics Measurements, R. J. Goldstein, ed., Hemisphere,
Washington, DC, pp. 559615.
8 Reiss, L. P., and Hanratty, T. J., 1963, An Experimental Study of the Unsteady Nature of the Viscous Sub-Layer, AIChE J., 9, pp. 154160.
9 Mitchell, J. E., and Hanratty, T. J., 1966, A Study of Turbulence at a Wall
Using an Electrochemical Wall Shear Stress Meter, J. Fluid Mech., 26, pp.
199221.
10 Fortuna, G., and Hanratty, T. J., 1971, Frequency Response of the Boundary
Layer on the Wall Transfer Probes, Int. J. Heat Mass Transfer, 14, pp. 1499
1507.
11 Deslouis, C., Gil, O., and Tribollet, B., 1990, Frequency Response of Electrochemical Sensors to Hydrodynamic Fluctuations, J. Fluid Mech., 215, pp.
85100.
12 Sobolik, V., Wein, O., and Cermac, J., 1987, Simultaneous Measurement of
the Film Thickness and Wall Shear Stress in Wavy Flow of Non-Newtonian
Liquids, Collect. Czech. Chem. Commun., 52, pp. 913928.
13 Labraga, L., Bourabaa, N., and Berkah, T., 2002, Wall Shear Stress From a
Rotating Cylinder in Cross Flow Using the Electrochemical Technique, Exp.
Fluids, 33, pp. 488496.
14 Zhuoxiong, M., and Hanratty, T. J., 1991, Analysis of Wall Shear Stress
Probes in Large Amplitude Unsteady Flows, Int. J. Heat Mass Transfer, 34,
pp. 281290.
15 Rehimi, F., Aloui, F., Ben Nasrallah, S., Doubliez, L., and Legrand, J., 2006,
MAY 2009, Vol. 131 / 051403-7
16
17
18
19
20
Inverse Method for Electrodiffusional Diagnostic of Flows, Int. J. Heat Mass

Transfer, 49, pp. 12421254.
Takeda, Y., 1995, Velocity Profile Measurements by Ultrasonic Doppler
Method, Exp. Therm. Fluid Sci., 10, pp. 444453.
Berni, A., 2005, Unsteady Velocity Profiles and Energy Dissipation in Turbulent Pipe Flow, Ph.D. thesis, Dipartimento di Ingegneria Civile ed Ambientale, Perugia, Italy.
Nowak, M., 2002, Wall Shear Stress Measurement in a Turbulent Pipe Flow
Using Ultrasound Doppler Velocimetry, Exp. Fluids, 33, pp. 249255.
Zidouh, H., 2007,Etude exprimentale du frottement parital instationnaire,
Ph.D. thesis, Universit de Valenciennes et du Hainaut-Cambrsis, France.
Sobolik, V., Tihon, J., Wein, O., and Wichterle, K., 1998, Calibration of
051403-8 / Vol. 131, MAY 2009
Electrodiffusion Friction Probes Using a Voltage-Step Transient, J. Appl.

Electrochem., 28, pp. 329335.
21 Wunderlich, T., and Brunn, P. O., 2000, A Wall Layer Correction for Ultrasound Measurements in Tube Flow: Comparison Between Theory and Experiments, Flow Meas. Instrum., 11, pp. 6369.
22 Washington, P. A., 1997, Discussion on Wall Shear Stress in Accelerating
and Decelerating Turbulent Pipe Flows, J. Hydraul. Res., 351, pp. 140
142.
23 Axworthy, D. H., Ghidaoui, M. S., and McInnis, D. A., 2000, Extended
Thermodynamics Derivation of Energy Dissipation in Unsteady Pipe Flow, J.
Hydraul. Eng., 1264, pp. 276287.
Inviscid Flow Past Two Cylinders

R. S. Alassar
the perpendicular z-direction. The two-dimensional version is

composed of Apollonian circles, which are two sets of circles
where every circle in the first set intersects every circle in the
second set orthogonally Fig. 1. The transformation equations of
the bipolar cylindrical coordinates are
M. A. El-Gebeily
x=
Department of Mathematics and Statistics,

King Fahd University of Petroleum & Minerals,
Dhahran 31261, Saudi Arabia
A simple solution of the problem of inviscid flow past two circular

cylinders is presented. The two cylinders may be of different diameters and located at any distance from each other. The solutions of the two main cases, namely, when the flow is perpendicular to the center-to-center line and when the flow is parallel to it
(tandem cylinders), lead to a solution of the problem when the
flow is in an arbitrary direction. DOI: 10.1115/1.3114678
Introduction
The study of the flow past two circular cylinders is important

due to the related scientific applications. One can find numerous
papers written on the solution of the flow past a single sphere or a
single cylinder. A review of the flow past spheres can be found in
Ref. 1, past cylinders in Ref. 2, and past oblate and prolate
spheroids in Refs. 3,4. The potential flow past a single spheroid,
as a generalization of a sphere, was treated by Alassar and Badr
5. Exact and closed form solutions in terms of the potential and
stream functions were obtained. There is, however, far less work
conducted on the effect of the existence of one body onto another.
While we are not aware of any simple closed form solution to the
potential flow past two interacting spheres, some typical studies
that deal with the problem are those by Hicks 6, Basset 7,
Herman 8, Lamb 9, Rouse 10, Bentwich 11, and Sun and
Chwang 12. The mutual force between two pulsating spherical
bubbles far apart was studied by Bjerknes 13. Lagally 14,15
considered the problem of the potential flow induced by two stationary circular cylinders in a uniform stream. The method of
obtaining a complicated solution was based on conformal mappings. The work of Lagally 14,15 was extended by Wang 16 to
the unsteady problem of two circular cylinders expanding and
translating arbitrarily in a still fluid. Crowdy et al. 17 published
a follow-up to Wangs paper describing the inviscid flow solution
for two arbitrary shaped cylinders. Crowdy 18 also found a solution for uniform flow past multiple cylinders, again using conformal mapping techniques.
In this paper, we show that a simple solution of the problem of
inviscid flow past two circular cylinders can be obtained by hosting the problem in bipolar coordinates. The inviscid flow solution
provides a good approximation for high Reynolds number flows.
There are two interesting arrangements of the cylinders, namely,
tandem when the flow is parallel to the center-to-center line and
side-by-side when the flow is perpendicular to the center-to-center
line. The two cylinders may be of different diameters and located
at any distance from each other.
The Inviscid Flow Problem
The bipolar cylindrical coordinate system , , z is a threedimensional orthogonal coordinate system that results from projecting the two-dimensional bipolar coordinate system , , in
JOURNAL OF FLUIDS ENGINEERING. Manuscript received April 2, 2008: final manuscript
received March 11, 2009; published online April 16, 2009. Review conducted by
Dimitris Drikakis.
y=
a sin
,
cosh cos
and
z=z
h =
a
,
cosh cos
and
hz = 1 2
a sinh
,
cosh cos
with scale factors given by

h =
a
,
cosh cos
Eliminating from Eq. 1, we get x2 + y a cot 2 = a2 csc 2,

which defines the first coordinate surfaces = constant, 0
2 as circular cylinders centered at 0 , a cot and of radius
a csc . Eliminating , on the other hand, we get the equation
x a coth 2 + y 2 = a2 csc h2, which defines the second coordinate surfaces = constant, as circular cylinders centered at a coth , 0 and of radius a csc h. As , the
circles degenerate into the two focal points a , 0, the cylinders
to lines. The third coordinate surfaces are obviously the planes
z = constant.
The problem considered here is that of two cylinders, possibly
of different diameters, placed in a frictionless incompressible fluid
of infinite extent. The fluid far away from the two cylinders moves
with a velocity Ux or Uy. It can be shown that specifying the
radius of each of the two cylinders and the center-to-center distance fixes a particular bipolar coordinate system in the sense that
1 first circle, 2 second circle, and a are uniquely determined.
We introduce the stream function using the following traditional definition:
h h zw = +
h h zw =
where w and w are the velocity components in the directions of

and , respectively. This definition identically satisfies the continuity equation. The condition of vanishing vorticity can be written as
h
h
+
=0
h h z
hhz
Flow Perpendicular to the Center-to-Center Line
Consider the case when the freestream flow is perpendicular to

the center-to-center line. The boundary conditions to be satisfied
are the impermeability of the surfaces of the cylinders and the
freestream conditions. The freestream boundary conditions can be
obtained by defining two unit vectors on the directions and ,
then taking the projection of the freestream velocity onto these
orthogonal directions. One can show that these conditions can be
written in terms of the stream function as
aUycosh cos 1
cosh cos 2
and
aUy sinh sin
cosh cos 2
5
or simply,
aUy sinh
cosh cos
as
, 0,0
The complete boundary conditions in terms of the velocity and the

stream function are shown in Figs. 2a and 2b, respectively.
Due to symmetry, only the upper half plane is considered. It is
interesting to observe that the freestream conditions are expressed
in bipolar coordinates at a single point, namely, , 0 , 0.
This is exactly the major difficulty in hosting the problem in bipolar coordinates. The single point written in bipolar coordinates
MAY 2009, Vol. 131 / 054501-1
x=0
x = x2
x = x1
Ux
q=0
q=0
q=p
x=0
Uy
Fig. 1 The problem configuration
carefully as , 0 , 0 is the huge circle with infinite radius

that engulfs the whole domain.
The present problem is not separable due to boundary conditions. Our idea is to solve a perturbed separable problem and then
mend the solution with a nonseparable part to satisfy the
freestream conditions. We define the two functions as follows:
, =
cos n sinh n 1
n=1
, =
cn cos n sinh n 1 +
n=1
sinh n1
n=1
10
Similarly, at 1 we set
=
=1
nd
sin n sinh n1 2 =
n=1
aUy sinh 1 sin

cosh 1 cos 2
11
, =
cos n sinh n 2
n=1
These two functions are harmonic and, respectively, satisfy the

bottom and the top boundaries in Fig. 2b. Furthermore, they
both satisfy the right and left boundaries except at the point
, 0 , 0. We use the top boundary to get the coefficients cn
by setting / =2 equal to the similarly defined derivative
boundary condition at the far field Eq. 5. Thus,
=
=2
nc
sin n sinh n2 1 =
n=1
aUy sinh 2 sin

cosh 2 cos 2
and
, =
n=1
sinh n2
n=1
12
Using the orthogonality properties of trigonometric functions, the
coefficients cn and dn can be determined as
cn =
9
The fact that does not vanish as , 0 , 0 can be dealt with
by adding to it the constant term n=1

cn sinh n1 so that
dn cos n sinh n 2 +
aUy sinh 2
n sinh n2 1
n2
2aUy e
sinh n2 1
sin sin n
d
cosh 2 cos 2
13
Fig. 2 Boundary conditions
054501-2 / Vol. 131, MAY 2009
0.5
0.25
0
-0.25
-0.5
-0.75
-1
0
0.5
1.5
2.5
Fig. 3 Streamline patterns for the case r1 = 1, r2 = 2, and H = 4
aUy sinh 1
n sinh n1 2
dn =
sin sin n
d
cosh 1 cos 2
+n1
= +
2aUy e
sinh n1 2
14
Now the function y = aUy sinh / cosh cos has

all the required properties in the sense that it is harmonic and
satisfies all boundary conditions including the freestream. The solution of the problem may be written compactly as
sinh
y,
=
2aUy
2cosh cos
cos ne
n=1
n=1
n2
+n1
sinh n 2
As H , 2 . Using Eq. 18 to calculate P , i.e., the

dimensionless pressure around the right cylinder when the two
cylinders are far apart, one finds that
P, =
15
Typical streamline patterns are shown in Fig. 3 for both the transformed and actual physical domains. The case shown in Fig. 3 is
for r1 = 1, r2 = 2, and H = 4 where H is the center-to-center distance. The streamlines are orthogonal to the boundaries = 0 and
= . The two lines originally the cylinders = 1 and = 2 are
streamlines themselves. The point , 0 , 0 works like a
black hole where all streamlines are sucked into.
Once the flow field has been determined, the distribution of the
pressure P along the surfaces of the cylinders can be obtained
from the equation of conservation of momentum. The
-component of the momentum equation can be written as
1
U2y
w
d
17
which measures the scaled pressure at any location relative to that

at = 0. P , may be evaluated on the surface of either cylinder. Figure 4 shows the distribution of P , 2 for the case r1
= 1, r2 = 1. Since the bipolar coordinate variable is neither fixed
nor uniformly distributed over the surface of the cylinder, we use
16
where is the fluid density. We define the dimensionless pressure

P , as
1
U2y
d = 1 cos 2
19
This result is identical to that obtained from analyzing the inviscid

flow past a single cylinder using cylindrical coordinates. Figure 4
shows that as the distance between the two cylinders becomes
large, the negative pressure gradient created by the existence of
the other cylinder at = disappears and the pressure approaches
that for a single cylinder given by Eq. 19.
To have an idea of the truncation error involved in the series
solution, we estimate the error that results from considering N
terms of the series in Eq. 18. It can be shown that the error
decays exponentially and has the following bound:
n=N+1
w
P
= w
U2y
18
sinh n 1 + e
sinh n2 1
P P0
en2 sinh n cos n

y,
sinh
=
+
sinh n2
2aUy
2cosh cos n=1
sinh n2 + 1
sinh n2 1
P , =
the uniform angle , which is used in the normal polar cylindrical

coordinates system. A polar plot of the pressure is also provided in
Fig. 4b.
As the distance between the two cylinders increases, the effect
of the existence of one cylinder on the other becomes negligible.
Consider, for example, two cylinders having the same diameters
1 = 2. Equation 15 reduces to
en2 sinh n cos n

eN+122
sinh n2
1 e22
20
4 Flow Parallel to the Center-to-Center Line (Tandem

Cylinders)
The freestream boundary conditions Fig. 5 in bipolar coordinates can be written in terms of the stream function as
aUx sinh sin
cosh cos 2
and
aUx1 cosh cos
cosh cos 2
21
or simply,
MAY 2009, Vol. 131 / 054501-3
Fig. 4 The pressure distribution around the right cylinder for the case r1 = 1, r2 = 1
aUx sin
cosh cos
as
, 0,0
22
The following solution, which we denote by x , , is harmonic

and satisfies all boundary conditions including the freestream:
fn =
= +
aUx sin
f n sin n sinh n 1
cosh cos n=1
gn sin n sinh n 2
23
gn =
n=1
The solution is obtained in a similar way to the previous case. It is

noted here that the two sums in the solution vanish as ,
0 , 0 and no adjusting constant terms are needed.
The coefficients f n and gn are given by
cos n1 cos cosh 2

d
cosh 2 cos 2
2aUxe
sinh n2 1
n2
x , = +
aUx
n sinh n2 1
aUx
n sinh n1 2
24
cos n1 cos cosh 1

d
cosh 1 cos 2
+n1
= +
2aUxe
sinh n1 2
25
The solution can now be written as
Fig. 5 Boundary conditions for tandem cylinders
054501-4 / Vol. 131, MAY 2009
1
0.8
0.6
0.4
0.2
0
-0.2
0.5
1.5
2.5
Fig. 6 Streamline patterns for the case r1 = 4, r2 = 1, and H = 6
sin
x ,
= +
2aUx
2cosh cos
n=1
The far field boundary conditions can be written in terms of the

stream function as
sin nen2 sinh n 1 e+n1 sinh n 2

sinh n2 1
26
Typical streamlines are given in Fig. 6 for the case r1 = 4, r2 = 1,

and H = 6. Figure 7 shows the distribution of P , 2 for the case
r2 = 1, H = 6 for different sizes of the left cylinder.
Flow in an Arbitrary Direction
Consider a freestream with velocity Uo, which makes angle

with the positive x-axis. The only boundary conditions to be satisfied here are the impermeability conditions on the surfaces of the
cylinders and the freestream. No conditions are set at = 0 or at
= . We, then, must consider the whole plane as there is no
symmetry involved in this case.
aUo cos 1 cosh cos aUo sin sinh sin
cosh cos 2
cosh cos 2
aUo cos sinh sin aUo sin cosh cos 1
cosh cos 2
cosh cos 2
27
Or simply,
aUo cos sin aUo sin sinh
cosh cos
cosh cos
as
, 0,0
28
It is easy now to see that the solution , can be written as
Fig. 7 The pressure distribution around the right cylinder for the case r2 = 1, H = 6
MAY 2009, Vol. 131 / 054501-5
0.5
-0.5
-1
-1.5
0
Fig. 8 Streamline patterns for the case r1 = 1, r2 = 4, H = 7, and = / 4
, = y , + x ,
29
where the velocities Uy and Ux are to be replaced, respectively, by

Uo sin and Uo cos in the expressions of y , and x , .
The function is harmonic and satisfies all boundary conditions.
Figure 8 shows a typical solution for the case r1 = 1, r2 = 4, H = 7,
and = / 4.
Acknowledgment
The authors would like to thank King Fahd University of Petroleum & Minerals KFUPM for supporting this research under
Grant No. MS/Inviscid/375. R.S.A. is grateful to Professor M.
Amara and the Applied Mathematics Laboratory at the Pau University in France for such a pleasant visit that made this research
possible.
References
1 Alassar, R. S., Badr, H. M., and Allayla, R., 2000, Viscous Flow Over a
Sphere With Fluctuations in the Free-Stream Velocity, Comput. Mech.,
265, pp. 409418.
2 Badr, H. M., 1994, Oscillating Inviscid Flow Over Elliptic Cylinders With
Flat Plates and Circular Cylinders as Special Cases, Ocean Eng., 211, pp.
105113.
3 Alassar, R. S., and Badr, H. M., 1999, Oscillating Flow Over Oblate Spheroids, Acta Mech., 13734, pp. 237254.
4 Alassar, R. S., and Badr, H. M., 1999, Oscillating Viscous Flow Over Prolate
Spheroids, Trans. Can. Soc. Mech. Eng., 231A, pp. 8393.
054501-6 / Vol. 131, MAY 2009
5 Alassar, R. S., and Badr, H. M., 1997, Analytical Solution of Oscillating

Inviscid Flow Over Oblate Spheroids With Spheres and Flat Disks as Special
Cases, Ocean Eng., 243, pp. 217225.
6 Hicks, W. M., 1880, On the Motion of Two Spheres in a Fluid, Philos. Trans.
R. Soc. London, 171, pp. 455492.
7 Basset, A. B., 1886, On the Motion of Two Spheres in a Liquid, and Allied
Problems, Proc. London Math. Soc., s118, pp. 369378.
8 Herman, R. A., 1887, On the Motion of Two Spheres in Fluid and Allied
Problems, Q. J. Math., 22, pp. 204231.
9 Lamb, H., 1932, Hydrodynamics, 6th ed., Cambridge University Press, Cambridge, UK, Chap. 6.
10 Rouse, H., 1976, Advanced Mechanics of Fluids, Krieger Publishing Company, Malabar, FL, Chap. 3.
11 Bentwich, M., 1978, Exact Solution for 2-Sphere Problem in Axisymmetrical
Potential Flow, ASME Trans. J. Appl. Mech., 45, pp. 463468.
12 Sun, R., and Chwang, A. T., 2001, Hydrodynamic Interaction Between a
Slightly Distorted Sphere and a Fixed Sphere, Theor. Comput. Fluid Dyn.,
15, pp. 1122.
13 Bjerknes, V. F. K., 1906, Fields of Force, Columbia University Press, New
York.
14 Lagally, M., 1929, The Frictionless Current in the Outer Areas of Double
Circuits, Z. Angew. Math. Mech., 9, pp. 299305.
15 Lagally M., 1931, The Frictionless Flow in the Region Around Two Circles,
N.A.C.A. Technical Memorandum No. 626.
16 Wang, Q. X., 2004, Interaction of Two Circular Cylinders in Inviscid Flow,
Phys. Fluids, 1612, pp. 44124425.
17 Crowdy, D. G., Surana, A., and Yick, K.-Y., 2007, The Irrotational Motion
Generated by Two Planar Stirrers in Inviscid Fluid, Phys. Fluids, 19, p.
018103.
18 Crowdy, D. G., 2006, Analytical Solution for Uniform Potential Flow Past
Multiple Cylinders, Eur. J. Mech. B/Fluids, 25, pp. 459470.

J Fluids Engineering 2009 Vol 131 N5

Uploaded by

Copyright:

Available Formats

J Fluids Engineering 2009 Vol 131 N5

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

J Fluids Engineering 2009 Vol 131 N5

Uploaded by

Copyright:

Available Formats

Journal of

VOLUME 131 NUMBER 5 MAY 2009

Performance and Radial Loading of a Mixed-Flow Pump Under

Experimental and Computational Study of Oscillating Turbine Cascade

Experimental Investigation of Flow Field Structure in Mixing Tee

Experimental Study of a New Flow Conditioner on Disturbed Flow

2D NavierStokes Simulations of Microscale Viscous Pump With Slip

Pre-Stall Instability Distribution Over a Transonic Compressor Rotor

Fundamental Issues and Canonical Flows

Front Condition for Gravity Currents in Channels of Nonrectangular

Vortex Dynamics and Shedding of a Low Aspect Ratio, Flat Wing at

Drag Reduction in Turbulent Flow With Polymer Additives

Cavitating Turbulent Flow Simulation in a Francis Turbine Based on

Techniques and Procedures

Transactions of the ASME, Journal of Fluids Engineering

Parallel-Plate Conductive Electrodes for the Fabrication of Larger 2D

Comparison of Turbulence Modeling Strategies for Indoor Flows

Unsteady Wall Shear Stress in Transient Flow Using Electrochemical

Contents continued on inside back cover

Volume 131, Number 5

Inviscid Flow Past Two Cylinders

The ASME Journal of Fluids Engineering is abstracted and indexed in

Performance and Radial Loading

search focused on the effect of leakage flow on rotordynamic

Journal of Fluids Engineering

Figure 1 shows a frequently used design of a waterjet system, a

Copyright 2009 by ASME

MAY 2009, Vol. 131 / 051101-1

electromagnetic flow meter

gauge and safety valve

air in (max 4 bar)

outlet pressure inlet pressure

Fig. 1 Waterjet propulsion system employing a mixed-flow

duct to the pump, is subsequently pressurized, and accelerated in

Fig. 3 Experimental facility

suction from the boundary layer

A scale-model version of a waterjet pump manufactured by

construction is changed and installed at the discharge. The reason

The dynamometer is a force sensing device, which is placed

Fig. 2 Examples of measured axial velocity distribution in a

051101-2 / Vol. 131, MAY 2009

impeller suction pipe

Fig. 4 Cross section of the mixed-flow pump indicating the

Transactions of the ASME

Fig. 5 Drawing of dynamometer showing four posts left and

To minimize noise and prevent loss of signals, electronics for

Flow Redistribution Device

Fig. 7 Axial velocity profiles along a vertical traverse, with and

plane located approximately one pipe diameter upstream of the

Fig. 6 Dynamometer with electronics for bridge power supply

Journal of Fluids Engineering

Fig. 8 Comparison of velocity profile B with measured profiles

MAY 2009, Vol. 131 / 051101-3

from measured and averaged static pressure and average axial

Fig. 10 Impeller torque for uniform suction flow, at different

were used to measure static pressure at stations approximately one

Fig. 12 Static head for uniform and nonuniform suction flows,

Fig. 9 Manometric head for uniform suction flow, at different

Fig. 11 Axial force on the impeller for uniform suction flow, at