J Fluids Engineering 2009 Vol 131 N9

Journal of
Fluids Engineering
Published Monthly by ASME
VOLUME 131 NUMBER 9 SEPTEMBER 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
O. COUTIER-DELGOSHA 2012
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
P. VLACHOS 2012
M. WANG 2011
St. T. WERELEY 2011
Y. ZHOU 2009
RESEARCH PAPERS
Flows in Complex Systems
091101
Pharyngeal Airflow Analysis in Obstructive Sleep Apnea Patients Preand Post-Maxillomandibular Advancement Surgery
John Huynh, Ki Beom Kim, and Mark McQuilling
091102
Flow Characteristics in a Curved Rectangular Channel With Variable

Cross-Sectional Area
Avijit Bhunia and C. L. Chen
091103
Measurements of a Wall Jet Impinging Onto a Forward Facing Step

D. C. Langer, B. A. Fleck, and D. J. Wilson
Fundamental Issues and Canonical Flows

091201
Large-Eddy Simulation of Wake and Boundary Layer Interactions

Behind a Circular Cylinder
S. Sarkar and Sudipto Sarkar
091202
On the Pulsed and Transitional Behavior of an Electrified Fluid

Interface
Paul R. Chiarot, Sergey I. Gubarenko, Ridha Ben Mrad, and
Pierre E. Sullivan
091203
Curved Non-Newtonian Liquid Jets With Surfactants

Jamal Uddin and Stephen P. Decent
091204
Circulation Generation and Vortex Ring Formation by Conic Nozzles

Moshe Rosenfeld, Kakani Katija, and John O. Dabiri
PUBLICATIONS COMMITTEE
Chair, B. RAVANI
OFFICERS OF THE ASME
President, AMOS E. HOLT
Executive Director, THOMAS G. LOUGHLIN
Treasurer, WILBUR MARNER
PUBLISHING STAFF
Managing Director, Publishing
P. DI VIETRO
Manager, Journals
C. MCATEER
Production Coordinator
A. HEWITT
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.
Multiphase Flows
091301
Measurements of High Velocity Gradient Flow Using Bubble Tracers

in a Cavitation Tunnel
Bu-Geun Paik, Kyung-Youl Kim, and Jong-Woo Ahn
091302
Thermodynamic Effect on Rotating Cavitation in an Inducer

Yoshiki Yoshida, Yoshifumi Sasao, Mitsuo Watanabe,
Tomoyuki Hashimoto, Yuka Iga, and Toshiaki Ikohagi
091303
Transition of Bubbly Flow in Vertical Tubes: New Criteria Through

CFD Simulation
A. K. Das, P. K. Das, and J. R. Thome
091304
Transition of Bubbly Flow in Vertical Tubes: Effect of Bubble Size

and Tube Diameter
A. K. Das, P. K. Das, and J. R. Thome
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 9
SEPTEMBER 2009
TECHNICAL BRIEFS
094501
Analytical Solution for Newtonian Laminar Flow Through the Concave and Convex Ducts
M. Firouzi and S. H. Hashemabadi
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
John Huynh
Resident
e-mail: jhuynh1@slu.edu
Ki Beom Kim
Assistant Professor
e-mail: kkim8@slu.edu
Graduate Orthodontic Program,
Center for Advanced Dental Education,
Saint Louis University,
St. Louis, MO 63104
Mark McQuilling1
Assistant Professor
Department of Aerospace and Mechanical
Engineering,
and Center for Fluids at All Scales,
Saint Louis University,
3450 Lindell Boulevard,
St. Louis, MO 63103
e-mail: mmcquil2@slu.edu
Pharyngeal Airflow Analysis in

Obstructive Sleep Apnea Patients
Pre- and Post-Maxillomandibular
Advancement Surgery
The purpose of this study was to evaluate pharyngeal airflow in obstructive sleep apnea
(OSA) patients following maxillomandibular advancement (MMA) surgery using computational fluid dynamics (CFD). Computerized models of four OSA patients, pre- and
postsurgery, were created using cone beam computed tomography scans. CFD was used
to model airflow at inspiration rates of 340 ml/s, 400 ml/s, and 460 ml/s. The relative
pressure, eddy viscosity coefficient, and total area-averaged pressure drops were selected
for comparison. Results show a decrease in airway resistance of over 90% for three out
of four patients. In these three patients, the MMA surgery reduced the constriction along
the airway, which resulted in reduced drag and therefore reduced pressure drop required
to move a constant volumetric flow between pre- and postsurgery models. CFD analyses
on airways of OSA patients provide data that suggest an improvement in airflow following MMA surgery with less effort required for maintaining constant flow.
DOI: 10.1115/1.3192137
Introduction
It is estimated that obstructive sleep apnea affects 2% of middle

aged women and 4% of middle aged men in the United States 1.
Sequelae for this condition include a disturbance of normal sleep
patterns, excessive daytime sleepiness, cognitive dysfunction,
headaches, systemic hypertension, dysrhythmias, depression,
stroke, and angina 2,3. Risk factors for obstructive sleep apnea
include smoking 4, obesity 5, excessive alcohol consumption
4, respiratory depressing drugs 2, nasal congestion, snoring,
and increased neck circumference 6.
Anatomically, when compared with normal patients, apneic patients have a large and/or retropositioned tongue, an elongated soft
palate, enlarged pharyngeal tonsils, large parapharyngeal fat pads,
retrognathic maxilla and/or mandible, short anterior cranial bases,
long anterior facial heights, inferiorly placed hyoid bones, narrow
posterior airway spaces, or narrow lateral pharyngeal spaces
2,715. In obstructive sleep apnea patients, the balance of constricting factors i.e., negative inspirational pressure, collapsing
soft tissues, and retropositioned structures outweigh the dilating
factors i.e., dilating forces of the pharyngeal musculature and
increased tone in pharyngeal structures, thus, resulting in increased upper airway resistance 7. A problem however is that the
sites of airway collapse and constriction are not identical in all
patients, while some may have multiple sites of obstruction 16.
Treatment modalities for obstructive sleep apnea include nonsurgical and surgical treatments. A treatment that has been increasing in popularity is maxillomandibular advancement surgery
MMA. Maxillomandibular advancement surgery has traditionally been considered a Phase II surgical treatment, which is used
when Phase I treatments have been unsuccessful 17. However,
in patients who are diagnosed with obstructions at multiple levels
along the airway or those with craniofacial anomalies, MMA surgery has been recommended as the first surgical option rather than
other Phase I treatments 2,7,1720.
1
Corresponding author.
Contributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received December 17, 2008; final
manuscript received June 27, 2009; published online August 13, 2009. Assoc. Editor:
Rajat Mittal.
Recent studies using three-dimensional 3D imaging have

evaluated the effects of MMA surgery on pharyngeal airway volume. Fairburn et al. 21 used helical CT scans to evaluate 3D
changes in the upper airways of obstructive apnea patients following MMA surgery. Their results showed that there was a significant increase in the lateral and anteroposterior dimensions of the
airway at all levels that measurements were taken at. In a study
done by Burgess 22 evaluating airway volumetric changes following MMA surgery using cone beam computed tomography
CBCT, she observed that there was a significant change in airway volume in the long term following MMA surgery.
While these studies show that there is an increase in airway
volume following MMA surgery, there are few studies showing
how this change in volume affects airflow through the airway. A
reason for this is that studying airflow in the airway can be very
invasive. As a result, airflow studies have been accomplished using indirect methods to evaluate the airflow. Methods such as
anemometry, particle image velocimetry PIV, and computational
fluid dynamics CFD have been used to simulate real life airflows. Although each method has its advantages and disadvantages, the use of CFD is becoming more prominent because it can
allow more detailed and comprehensive information to be obtained when compared with the other methods.
Sung et al. 23 suggested that the major contributor to the
pathophysiology of obstructive sleep apnea is the reduced crosssectional area of the velopharyngeal retropalatal area. In their
study, they simulated the upper airway of a patient with obstructive sleep apnea using CFD. Their results showed that the maximum airflow velocity and maximum negative pressure occurred
within the velopharyngeal area meaning that this was the most
constricted and, thus, flow limiting area of the airway in that patient.
Shome et al. 24 performed numerical modeling of an anatomically accurate model of the human pharynx using CFD. In their
study, they examined the effects of cross-sectional area of the
airway on the relative pressure drop within the airway in response
to different treatments for obstructive sleep apnea. In response to
mandibular advancement, Shome et al. found that there was a
decrease in the relative pressure in relation to the increase in
cross-sectional area of the airway. While this study shows that air
Copyright 2009 by ASME
SEPTEMBER 2009, Vol. 131 / 091101-1
pressure in the airway is affected by surgical treatment, it does not

show how air velocity is affected. There are limited studies showing the effects of MMA surgery on airflow through the airway in
terms of air velocity and pressure changes.
The purpose of the study is to evaluate the pharyngeal airflow
of obstructive apnea patients who have undergone MMA surgery.
Computational fluid dynamics methods will be used to evaluate
both air velocity and pressure changes in these patients following
surgery.
1.1 Current Study. Previous studies have shown the ability
of CFD in analyzing airflows through the upper airways of patients with or without obstructions 2334. However, most of the
studies were based on one single model and few showed the effects of treatment on airflow post-treatment. Some studies used
simplified nonanatomically correct airway models in their CFD
analysis instead of anatomically correct models 30,31. On the
other hand, some studies used anatomically correct anatomy, but
did not use a turbulence model to model their airflow 35.
In the present study, pharyngeal airflow in patients with obstructive sleep apnea is shown before and after treatment with
maxillomandibular advancement surgery. Using cross-sectional
CT scans to construct the airway models ensure that the models
are anatomically correct. The RNG k- turbulence model is also
used to simulate turbulence due to the higher Reynolds numbers
occurring in the airways.
Fig. 1 3D skull depicting X, Y, and Z axes
Computational Models
Sections 2.1 and 2.2 detail the method used to create and analyze CFD models starting from CBCT scans.
2.1 Sample and Imaging. Records of four patients who have
undergone combined orthodontic treatment and MMA surgery
were selected. The sample was comprised of one female and three
males. The mean age was 40.0 years. The inclusion criteria for
these patients included having a history of obstructive sleep apnea, being treated with a combination of orthodontic treatment
and orthognathic surgery, had MMA surgery, and had presurgical
CBCT scans taken within a week prior to surgery T0 and a scan
at least a minimum of 7 weeks following surgery T1. The exclusion criteria were patients with craniofacial syndromes and patients who had surgical expansion of their upper jaws. All surgeries were performed by one of two oral surgeons utilizing similar
surgical techniques. For the mandibular advancement surgeries, a
bilateral sagittal split osteotomy technique was utilized. A nonsegmental Le Fort I osteotomy was utilized for the maxillary advancements. Bone grafts were placed in areas where large surgical
movements were performed and rigid fixation was used to fixate
both jaws. Prior to fixation, the mandibular condyles were seated
in centric relation in all patients.
The CBCT scans of the patients were performed using the same
i-CAT CBCT machine at both pre- and postsurgical time points.
The field of view used was 23 19 cm2. The voxel size of the
CBCT scans taken was 0.4 mm. All scans were taken with the
condyles seated in centric relation. V-WORKS 3D software and DOLPHIN 3D were used view and analyze the CBCT scans.
The amount of surgical movement in the anteroposterior AP
direction was determined using the pre- and postsurgical CBCT
scans. The head position was oriented with Frankfort Horizontal
parallel to the horizontal plane x-axis. The facial midline was
then centered on the vertical plane z-axis. A plane perpendicular
to Frankfort Horizontal, passing through nasion, was created to
represent the y-axis. These three planes are displayed in Fig. 1.
Four skeletal landmarks basion, nasion, point A, and point B
were identified along the z-axis plane; these are shown in Fig. 2.
The AP surgical movement was calculated using the z-axis coordinate for each jaw. Basion was used as a reference point to evaluate the reliability of the position of the other landmarks. The rela091101-2 / Vol. 131, SEPTEMBER 2009
tive positions of points A and B in relation to nasion were

recorded to determine the amount of advancement for the upper
and lower jaw, respectively.
The scanned images were imported into V-WORKS to generate
the volumetric image of the scanned region. Once the volumetric
image was produced, the area of interest could be isolated. The
superior border was defined as the scan that corresponded with a
horizontal line drawn through PNS. The inferior border was defined as the scan that corresponded with a horizontal line drawn
tangent to the lowest point on the third cervical vertebra. Typical
border positions are shown in Fig. 3.
2.2 3D Model and CFD. Using V-WORKS, the scan corresponding with the most superior border of the area of interest was
saved as a picture file for future processing. Using a slice thickness of 1.2 mm, successive scans were saved as picture files until
the most inferior border of the area of interest was reached. Once
these picture files were obtained, they were imported into IMAGEJ
for further processing. Here, airway slice coordinate sets were
generated by clicking on the outline of the airway, as can be seen
in Fig. 4. The coordinate sets were then used to create slices of the
airway in a solid modeling program, PRO/ENGINEER. The slices
were then blended together at distances of 1.2 mm apart to form
the three-dimensional airway, as seen in Fig. 5. Next, a box was
created around the joined slices to produce a hollowed out tunnel
in the shape of the airway. The box was then exported as an stl file
for input into the geometry cleanup software and CFD program,
CADTHRU and SC/TETRA, respectively. SC/TETRA is a finite volume
flow solver based on the SIMPLEC method 36 and MUSCL 37
differentiation technique for solving the NavierStokes equations.
SC/TETRA employs an unstructured tetrahedral mesh 38 and contains all pre- and postprocessing required for CFD analysis. Typical screenshots from the geometry cleanup process are shown in
Fig. 6. After appropriate geometry cleaning, the model was exported into the CFD software to configure the simulation.
A grid convergence test was done to determine an appropriate
grid size for the mesh models. Grid convergence was determined
when a number of velocity profiles at various locations along the
airway showed minimal difference after further grid refinement.
The resulting grid size input into the simulation was based on an
octree size of 0.5 mm. The final mesh models of the airway had
Transactions of the ASME
Fig. 2 Landmarks used to determine surgical movement
between 118,878 and 319,193 tetrahedral elements depending on

the size of the airways; a typical model is shown in Fig. 7. The
surface mesh h-ratio, which is the ratio of radii of the inscribed
circle and the circumscribed circle of a surface mesh triangle, was
Fig. 3 Typical superior and inferior borders of pharyngeal

airway
between 0.43 and 0.46 for all models. This h-ratio describes the
quality of surface triangulation, with the best h-ratio equal to 0.5
for an equilateral triangle. The tetrahedral volume mesh quality is
similarly described by its h-ratio, which is the ratio of radii of the
inscribed sphere and the circumscribed sphere of a tetrahedral
element. The tetrahedral h-ratio was between 0.26 and 0.27 for all
models in this study, where a value of 0.333 is the best possible
h-ratio. These numbers indicate the meshes used for the simulations described herein are of high quality.
In order to model turbulent flow within the pharyngeal airway,
the RNG k- turbulence model was used, as suggested by Sung et
al. 23 for simulations on similar geometries. Therefore, no other
turbulence closure models were compared. The inlet boundary
velocity was set as uniform and was determined for each model to
Fig. 4 Outline of airway using
IMAGEJ
SEPTEMBER 2009, Vol. 131 / 091101-3
Fig. 7 3D mesh model showing tetrahedral elements
Fig. 5
PRO/ENGINEER
model of the airway
correspond with the volumetric flow rates of 340 ml/s, 400 ml/s,
and 460 ml/s using the equation Q = AV. The inlet areas, A, were
computed from the solid modeling program. The outlet boundary
was set to a static pressure of 0 Pa for incompressible flow. A no
slip condition was set for the walls of the airway, which were also
assumed to be rigid and noncompliant. The simulations were performed on a PC with an AMD Athlon64 X2 Dual Core Processor 6400+ 3.21 GHz with 8 GB of RAM. Typical simulation
times were between 5 min and 8 min.
Results
The quantity of maxillary and mandibular surgical advancement

is shown in Table 1 for all patients. Maxillary advancement refers
to the change in point A between pre- and postsurgery, while
mandibular advancement refers to the change in point B between
pre- and postsurgery. Of particular note is the minimal amount of
advancement for patient 3 as compared with the other patients. As
described herein, postsurgery patient 3 behaves differently than
the other patients, and this small amount of advancement is believed to be one of the causes for the differences.
Figure 8 shows the change in hydraulic diameter across the
airway for all four patients. The hydraulic diameter is a measure
of the effective flow area at each location, and is defined as D
= 4A / P, where A is the cross-sectional area and P is the perimeter
24. This figure again shows how patient 3 results are unlike the
others, where the MMA surgery creates more constriction in several locations of the airway by reducing the effective flow area,
represented here by the negative change in hydraulic diameter.
The other three patients experience an increase in hydraulic diameter throughout the airway. The authors suspect the local decreases in hydraulic diameter for patient 3 are the primary reason
for the difference in flow behavior when compared with the other
patients.
The Reynolds number is a dimensionless number used to characterize different types of fluid flow as laminar, transitional, or
turbulent. For laminar flows, the Reynolds number is 2000,
while turbulent flows have a Reynolds number 4000; in between
those two ranges lies the transitional phase of fluid flow 39. The
equation used to calculate the Reynolds number is Re= VD / ,
where is the density of air= 1.2 kg/ m3, V is the mean fluid
velocity, D is the hydraulic diameter, and is the fluid
viscosity= 1.8 105 N s / m2. The volumetric flow rate, Q, is
equal to Q = AV, where A is the cross-sectional area and V is the
mean fluid velocity 40. Substituting these quantities transforms
the Reynolds number equation into Re= 4Q / P. This equation
for the Reynolds number illustrates how for a constant volumetric
flow rate, the flow can become more laminar in nature due to
increasing the perimeter of a cross section in the airway, as done
with the advancement surgeries. Figures 912 show the variation
in Reynolds number throughout the pharyngeal airways for each
patient at T0 presurgery and T1 postsurgery for all three flow
rates. The Reynolds numbers observed for all patients at T0
showed there were significant areas of reduced perimeter, or
Fig. 6 Typical CADTHRU models used for geometry cleanup. a Model after importing into
surfaces removed in cleanup process.
091101-4 / Vol. 131, SEPTEMBER 2009
CADTHRU.
b Model with some
Table 1 Quantity of surgical advancement mm
Maxillary
Mandibular
Pt 1
Pt 2
Pt 3
Pt 4
Mean
6.1
5.8
4.6
8.5
3.3
5.3
2.0
11.3
4.0
7.7
stenosis, along the airway, noted by peaks in the Re plots. For

patients 1, 2, and 4, the Reynolds numbers increased over 2000 in
the areas of stenosis, meaning the laminar flow was changing over
to turbulent flow. Patient 3 also showed smaller margin increases
in Reynolds number in stenotic areas but the values did not go
above 2000. Thus, airflow in this patient was predominantly laminar in nature. As the airflow was increased, there was a corresponding increase in Reynolds number. This was expected as a
Change in Hydraulic Diameter

Pt 1
Pt 2
Pt 3
Pt 4
12
Postdiameter minus Prediameter [mm]
10
2
0
10
20
30
40
Distance from inlet [mm]
50
60
70
Fig. 8 Change in hydraulic diameter along pharyngeal airway

Patient 1 Reynolds Number Variation
5000
Pt 1 Pre, 340
Pt 1 Post, 340
Pt 1 Pre, 400
Pt 1 Post, 400
Pt 1 Pre, 460
Pt 1 Post, 460
4500
4000
Reynolds number
3500
3000
2500
2000
1500
1000
500
10
20
30
40
50
60
70
Fig. 9 Reynolds number along pharyngeal airway of patient 1
SEPTEMBER 2009, Vol. 131 / 091101-5

5000
Pt 2 Pre, 340
Pt 2 Post, 340
Pt 2 Pre, 400
Pt 2 Post, 400
Pt 2 Pre, 460
Pt 2 Post, 460
4500
4000
Reynolds number
3500
3000
2500
2000
1500
1000
500
10
20
30
40
50
60
70
larger volume of air is trying to pass through a confined space. In

general, an increase in Re signifies more turbulent energy, but
again patient 3 Reynolds numbers did not change appreciably like
the others. Comparing pre- and postsurgery Reynolds numbers
show how the advancement surgeries remove stenotic sections of
the airway. Following surgery, the Reynolds numbers decreased
for all patients due to the increase in perimeter. As a result, the
airflow postsurgically tends to be more laminar in nature. In tur-
bulent flow, there is increased mass and energy transfer normal to

the surface the airflow is passing over. As a result, the surface
friction is increased as turbulence increases. Thus, if turbulence is
decreased Reynolds number is lowered, the respiratory force
needed to breathe is also decreased because of the reduction in
friction caused by turbulence 41. As the perimeter was increased
along with the hydraulic diameter after the surgery, the driving
force necessary to respire decreases. The increase in hydraulic

5000
Pt 3 Pre, 340
Pt 3 Post, 340
Pt 3 Pre, 400
Pt 3 Post, 400
Pt 3 Pre, 460
Pt 3 Post, 460
4500
4000
Reynolds Number
3500
3000
2500
2000
1500
1000
500
10
20
30
40
50
60
70
091101-6 / Vol. 131, SEPTEMBER 2009

5000
Pt 4 Pre, 340
Pt 4 Post, 340
Pt 4 Pre, 400
Pt 4 Post, 400
Pt 4 Pre, 460
Pt 4 Post, 460
4500
4000
Reynolds Number
3500
3000
2500
2000
1500
1000
500
10
20
30
40
50
60
70
diameter opens up constricted areas and reduces the pressure drag

forces, while the increase in perimeter lowers the Reynolds number, thus reducing the turbulence and resulting skin friction drag.
Figures 912 show how patients 1 and 4 behave similarly,
while there are significant differences in Reynolds number behavior between 1, 2, and 3. The effects of Reynolds number changes
between patients 1 and 2 are also similar; therefore, in the interest
of remaining concise, no additional figures from patients 1 or 4
will be presented as results are qualitatively similar to patient 2. It
should also be noted that quantitative comparisons drawn in this
paper are between pre- and postsurgery behaviors for a single
patient, and only qualitative comparisons are drawn between different patients. Therefore, figure scales and colorbars are identical
for pre- and postsurgery results for a single patient and are not the
same across patients; they were selected in order to illustrate the
differences between T0 and T1.
As mentioned earlier, two mechanisms act to reduce the pressure force required to drive a constant volumetric flow through the
airwayan increase in hydraulic diameter to reduce pressure
drag, and an increase in perimeter to decrease the Reynolds number and thereby decreasing the skin friction drag. Figure 13 shows
the eddy viscosity coefficient with streaklines along the midsagittal plane for a typical patient having stenotic regions in the passage and illustrates why this is true. The eddy viscosity coefficient
was selected for comparison herein since it contains effects of
both turbulent production and turbulent diffusion. The eddy viscosity coefficient rises in regions with increased turbulence, seen
near the right wall in the figure. Increased swirling motions indicated by the streaklines are also seen in the turbulent regions.
Thus, some of the kinetic energy is no longer in the direction of
flow, and its interaction with the bulk fluid movement dissipates
some of its energy, thereby requiring more pressure force to drive
the volume of fluid through the passage. This deviation from laminar flow can also result in increased shear stresses on the walls of
the airway.
Contours presenting the eddy viscosity coefficient along the
midsagittal plane for T0 and T1 are presented in Figs. 1417 for
patients 2 and 3 at all flow rates. Higher values for the eddy
viscosity coefficient were found in areas of stenosis, as well as

along the anterior and posterior walls of the airway. Figure 14
shows how increasing the flow rate causes a corresponding increase in eddy viscosity, as the faster freestream supplies more
energy into the turbulence process. This trend is true for all patients at T0 and T1, and regions near the walls show more of an
Fig. 13 Eddy viscosity coefficient contour with streaklines
SEPTEMBER 2009, Vol. 131 / 091101-7
Fig. 14 Contour of eddy viscosity coefficient for patient 2

along midsagittal plane at T0. Three flow rates shown: a 340
ml/s, b 400 ml/s, and c 460 ml/s.
Fig. 18 Contour of relative pressure Pa for patient 2 along

midsagittal plane at T0. Three airflow speeds shown: a 340
increase in turbulent activity than regions near the middle of the

passage. Comparing Fig. 14 to Fig. 15 shows how the advancement surgery decreases the turbulence near the walls and out into
the passage, which agrees well with the decreases in Reynolds
number as presented earlier; similar trends are seen with patients 1
and 4. Patient 3, however, shows only minimal difference in eddy
viscosity between T0 and T1, as seen in Figs. 16 and 17, respectively. Only the middle of the passage changes appreciably, with
higher turbulence remaining near the walls. As will be seen later,
the changes in hydraulic diameter and perimeter due to the surgery and the resulting decrease in passage turbulence seen here
actually requires an increase in total pressure to drive the volume
of fluid through the passage. Hence, lowering turbulence is not the

only factor to consider when deciding if the surgery was effective
or not.
Contours presenting the relative pressure along the midsagittal
plane for T0 and T1 are presented in Figs. 1821 for patients 2 and
3 at all flow rates. As expected, all patients experience an increase
in relative pressure with an increase in flow rate because a greater
driving force is required to move more air through the same passage. Figure 18 for patient 2 at T0 shows how areas of stenosis not
only increase turbulence, but also decrease the local pressure. This
is due to conservation of mass; since the flow rate Q = VA must be






091101-8 / Vol. 131, SEPTEMBER 2009
Table 2 Total pressure drop p at three flow rates for T0 and

T1 Pa
Pt
Pt
Pt
Pt
Pt
Pt
Pt
Pt
1,
1,
2,
2,
3,
3,
4,
4,
T0
T1
T0
T1
T0
T1
T0
T1
340 ml/s
400 ml/s
460 ml/s
12.7
1.1
23.6
2.3
2.1
2.4
43.5
1.1
17.1
1.5
31.9
3.1
2.8
3.2
59.3
1.4
22.3
1.9
41.4
4.0
3.5
4.1
77.7
1.8
conserved at all locations along the airways, if the flow area A

decreases, then the mean velocity V must increase, requiring a
greater difference in pressure to drive the flow. Toward the end of
the stenosis, however, the pressure recovers until a decrease toward the next stenosis or the outlet. For patient 2, comparing preand postsurgery in Figs. 18 and 19, respectively, shows how the
surgery causes a reduction in relative pressure all along the airways; the same is true for patients 1 and 4. In contrast, patient 3 in
Figs. 20 and 21 displays an increase in relative pressure along the
airway between T0 and T1.
Table 2 shows the total area-averaged pressure drop along the
airway, p, for all patients pre- and postsurgery at all flow rates.
As expected for all patients, there is an increase in the total pressure drop along the airway as the flow rate is increased. Patients 1,
2, and 4 have significant decreases in the p required from T0 to
T1, meaning it takes much less pressure force after the surgery to
drive the same volume of air through the passage. Patient 3, however, has an increase in the p value from T0 to T1.
The total area-averaged pressure drop, p, can also be used to
,
calculate the airway resistance, R, using the equation R = p / m
is the mass flow rate 25. The mass flow rate, m
, can be
where m
= Q, where Q is the volumetric
determined using the equation m
flow rate and is the density of air. The percentage change in this
resistance from T0 to T1 is shown in Table 3. Patients 1, 2, and 4
showed a minimal tenth of a percent decrease in airway resistance
as flow speed was increased. Patient 3 showed a larger magnitude
increase in resistance with flow rate at 1.2% and 1.4%. For patients 1, 2, and 4, the decrease in resistance from T0 to T1 was
higher than 90% with patient 4 showing the largest decrease in
airway resistance following the surgery. Patient 3 had an increase
in airway resistance from T0 to T1. Considering the amounts of
surgical advancements shown earlier in Table 1 and the negative
changes in hydraulic diameter seen in Fig. 8, this result is not
surprising. It is also noted that while patient 3 had less surgical
advancement, the total pressure drop required to drive the presurgery flow is on the order of magnitude of all other patients postsurgery where significant improvement was seen. Quite possibly,
patient 3 may be in a range of pharyngeal geometries where the
MMA surgery cannot produce results as optimal as others.
Conclusions
In this study, CFD modeling illustrated the effects maxillomandibular advancement surgery had on flow characteristics in the
pharyngeal airway of obstructive sleep apnea patients. This paper
Table 3 Percentage change in airway resistance
Pt
Pt
Pt
Pt
1
2
3
4
340 ml/s
400 ml/s
460 ml/s
91.2
91.2
+11.1
97.5
91.3
91.3
+12.5
97.6
91.4
91.4
+13.7
97.7
demonstrated a method to use cone beam computed tomography

scans to create a three-dimensional model suitable for CFD analysis. The results presented in this paper lead to the following conclusions.
1. Following surgery, the airway resistance decreased by over
90% for three out of the four patients studied. This is a result
of increasing the cross-sectional areas in stenotic regions
with the MMA surgery, thereby increasing the hydraulic diameter and reducing the Reynolds number to require less
total pressure force to drive the flow for a constant volumetric flow rate.
2. If one assumes the pharyngeal airway is similar to a tube, a
generalization using the laminar flow equation p
= 128LQ / d4 can be used to relate p with volumetric
airflow, where is the fluid viscosity, L is the tube length, Q
is the volumetric flow rate, and d is the tube diameter 40. If
an average diameter is used for d and all other variables
remain equal, and the p required for inspiration in the presurgical model was applied to the postsurgical model, then at
least ten times as much air can be moved in the postsurgical
model with equal effort for the three patients reporting improvements.
3. Isono et al. 42 reported that in obstructive sleep apnea
patients there is progressive narrowing of the velopharyngeal airway in flow limited inspirations. This paper showed
how areas of stenosis cause a lower local relative pressure,
and these lower local pressures in the segments distal to the
stenotic area could result in further collapse of the airway if
the tissues are highly compliant in that area.
4. Patient 3 did not exhibit the favorable characteristics seen in
the other three patients between the pre- and postsurgery
models. The authors believe this is due to three reasons: the
small quantities of maxillary and mandibular surgical advancement, the airway regions having a negative change in
hydraulic diameter, and the small change in Reynolds number behavior throughout the pharyngeal passage. This may
suggest that patients having a certain range of pharyngeal
geometry may not benefit from MMA surgery. This could
only be proven with continued discussion with patients and
possible testing after the surgery to see if their apnea conditions have improved.
Acknowledgment
The authors wish to thank George Huang of Wright State University and David Welsh of Software Cradle Co. for their guidance with the computational modeling.
Nomenclature
A
D
m
P
p
Q
R
Re
T0
T1
V
cross-sectional area m2
hydraulic diameter, D = 4A / P m
= Q kg/s
mass flow rate, m
perimeter m
total area-averaged pressure drop Pa
flow rate, Q = VA m3 / s
Pa s / kg
flow resistance, R = p / m
Reynolds number, Re= VD /
time denoting one week prior to surgery
time denoting seven weeks after surgery
mean flow velocity m/s
Greek Symbols
density of air kg/ m3
viscosity of air kg/ m s
References
1 Young, T., Palta, M., Dempsey, J., Skatrud, J., Weber, S., and Badr, S., 1993,
The Occurrence of Sleep-Disordered Breathing Among Middle-Aged
SEPTEMBER 2009, Vol. 131 / 091101-9
Adults, N. Engl. J. Med., 328, pp. 12301235.

2 Fleisher, K., and Krieger, A., 2007, Current Trends in the Treatment of Obstructive Sleep Apnea, J. Oral Maxillofac Surg., 65, pp. 20562068.
3 Padma, A., Ramakrishnan, N., and Narayanan, V., 2007, Management of
Obstructive Sleep Apnea: A Dental Perspective, Indian J. Dent. Res., 18, pp.
201209.
4 Young, T., Skatrud, J., and Peppard, P., 2004, Risk Factors for Obstructive
Sleep Apnea in Adults, JAMA, J. Am. Med. Assoc., 291, pp. 20132016.
5 Tangugsorn, V., Krogstad, O., Espeland, L., and Lyberg, T., 2001, Obstructive
Sleep Apnea: A Canonical Correlation of Cephalometric and Selected Demographic Variables in Obese and Nonobese Patients, Angle Orthod., 71, pp.
2335.
6 Davies, R., Ali, N., and Stradling, J., 1992, Neck Circumference and Other
Clinical Features in the Diagnosis of the Obstructive Sleep Apnea Syndrome,
Thorax, 47, pp. 101105.
7 Prinsell, J., 1999, Maxillomandibular Advancement Surgery in a Site-Specific
Treatment Approach for Obstructive Sleep Apnea in 50 Consecutive Patients,
Chest, 116, pp. 15191529.
8 Guilleminault, C., Riley, R., and Powell, N., 1984, Obstructive Sleep Apnea
and Abnormal Cephalometric Measurements: Implications for Treatment,
Chest, 86, pp. 793794.
9 Lowe, A., Santamaria, J., Fleetham, J., and Price, C., 1986, Facial Morphology and Obstructive Sleep Apnea, Am. J. Orthod. Dentofacial Orthop., 90,
pp. 484491.
10 Lyberg, T., Krogstad, O., and Djupesland, G., 1989, Cephalometric Analysis
in Patients With Obstructive Sleep Apnoea Syndrome: II. Soft Tissue Morphology, J. Laryngol. Otol., 103, pp. 293297.
11 Lyberg, T., Krogstad, O., and Djupesland, G., 1989, Cephalometric Analysis
in Patients With Obstructive Sleep Apnoea Syndrome: I. Skeletal Morphology, J. Laryngol. Otol., 103, pp. 287292.
12 Pracharktam, N., Nelson, S., Hans, M., Broadbent, B., Redline, S., and Rosenberg, C., 1996, Cephalometric Assessment in Obstructive Sleep Apnea, Am.
J. Orthod. Dentofacial Orthop., 109, pp. 410419.
13 Riley, R., Guilleminault, C., Herran, J., and Powell, N., 1983, Cephalometric
Analyses and Flow-Volume Loops in Obstructive Sleep Apnea Patients,
Sleep, 6, pp. 303311.
14 Rodenstein, D., Dooms, G., Thomas, Y., Liistro, G., Stanescu, D., and Culee,
C., 1990, Pharyngeal Shape and Dimensions in Healthy Subjects, Snorers,
and Patients With Obstructive Sleep Apnea, Thorax, 45, pp. 722727.
15 Schwab, R., Gefter, W., Hoffman, E., Gupta, K., and Pack, A., 1993, Dynamic Upper Airway Imaging During Awake Respiration in Normal Subjects
and Patients With Sleep Disordered Breathing, Am. Rev. Respir. Dis., 148,
pp. 13851400.
16 Hoffstein, V., and Wright, S., 1991, Improvement in Upper Airway Structure
and Function in a Snoring Patient Following Orthognathic Surgery, J. Oral
Maxillofac Surg., 49, pp. 656658.
17 Prinsell, J., 2002, Maxillomandibular Advancement Surgery for Obstructive
Sleep Apnea Syndrome, J. Am. Dent. Assoc., 133, pp. 14891497.
18 Prinsell, J., 2000, Maxillomandibular Advancement MMA in a Site-Specific
Treatment Approach for Obstructive Sleep Apnea: A Surgical Algorithm,
Sleep Breath., 4, pp. 147154.
19 Prinsell, J., 2000, Maxillomandibular Advancement Surgery for Obstructive
Sleep Apnea, J. Med. Assoc. Ga., 89, pp. 5457.
20 Conradt, R., Hochban, W., Brandenburg, U., Heitmann, J., and Peter, J., 1997,
Long-Term Follow-Up After Surgical Treatment of Obstructive Sleep Apnea
by Maxillomandibular Advancement, Eur. Respir. J., 10, pp. 123128.
21 Fairburn, S., Waite, P., Vilos, G., Harding, S., Bernreuter, W., and Cure, J.,
2007, Three-Dimensional Changes in Upper Airways of Patients With Obstructive Sleep Apnea Following Maxillomandibular Advancement, J. Oral
Maxillofac Surg., 65, pp. 612.
22 Burgess, B., 2004, Pharyngeal Airway Volume Following Maxillomandibular
091101-10 / Vol. 131, SEPTEMBER 2009
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
Advancement Surgery Utilizing Cone Beam Computed Tomography, MS thesis, Saint Louis University, Saint Louis, MO.
Sung, S., Jeong, S., Yu, Y., Hwang, C., and Pae, E., 2006, Customized ThreeDimensional Computational Fluid Dynamics Simulation of the Upper Airway
of Obstructive Sleep Apnea, Angle Orthod., 76, pp. 791799.
Shome, B., Wang, L., Santare, M., Prasad, A., Szeri, A., and Roberts, D., 1998,
Modeling of Airflow in the Pharynx With Application to Sleep Apnea,
ASME J. Biomech. Eng., 120, pp. 416422.
De Backer, J., Vanderveken, O., Vos, W., Devolder, A., Verhulst, S., and Verbraecken, J., 2007, Functional Imaging Using Computational Fluid Dynamics
to Predict Treatment Success of Mandibular Advancement Devices in SleepDisordered Breathing, J. Biomech., 40, pp. 37083714.
Xu, C., Sin, S., McDonough, J., Udupa, J., Guez, A., and Arens, R., 2006,
Computational Fluid Dynamics Modeling of the Upper Airway of Children
With Obstructive Sleep Apnea Syndrome in Steady Flow, J. Biomech., 39,
pp. 20432054.
Vos, W., Backer, J. D., Devolder, A., Vanderveken, O., Verhulst, S., and Salgado, R., 2007, Correlation Between Severity of Sleep Apnea and Upper
Airway Morphology Based on Advanced Anatomical and Functional Imaging, J. Biomech., 40, pp. 22072213.
Croce, C., Fodil, R., Durand, M., Sbirlea-Apiou, G., Caillibotte, G., and Papon, J., 2006, In Vitro Experiments and Numerical Simulations of Airflow in
Realistic Nasal Airway Geometry, Ann. Biomed. Eng., 34, pp. 9971007.
Wen, J., Inthavong, K., Tu, J., and Wang, S., 2008, Numerical Simulations for
Detailed Airflow Dynamics in a Human Nasal Cavity, Respir. Physiol. Neurobiol., 161, pp. 125135.
Heenan, A., Matida, E., Pollard, A., and Finlay, W., 2003, Experimental
Measurements and Computational Modeling of the Flow Field in an Idealized
Human Oropharynx, Exp. Fluids, 35, pp. 7084.
Huang, Y., White, D., and Malhotra, A., 2005, The Impact of Anatomic
Manipulations on Pharyngeal Collapse: Results From a Computational Model
of the Normal Human Upper Airway, Chest, 128, pp. 13241330.
Huang, Y., White, D., and Malhotra, A., 2007, Use of Computational Modeling to Predict Responses to Upper Airway Surgery in Obstructive Sleep
Apnea, Laryngoscope, 117, pp. 648653.
Jeong, S., Kim, W., and Sung, S., 2007, Numerical Investigation on the Flow
Characteristics and Aerodynamic Force of the Upper Airway of Patient With
Obstructive Sleep Apnea Using Computational Fluid Dynamics, Med. Eng.
Phys., 29, pp. 637651.
Ishikawa, S., Nakayama, T., Watanabe, M., and Matsuzawa, T., 2006, Visualization of Flow Resistance in Physiological Nasal Respiration: Analysis of
Velocity and Vorticities Using Numerical Simulation, Arch. Otolaryngol.
Head Neck Surg., 132, pp. 12031209.
Wang, K., Denney, T., Morrison, E., and Vodyanoy, V., 2005, Numerical
Simulation of Air Flow in the Human Nasal Cavity, Conf. Proc. IEEE Eng.
Med. Biol. Soc., 6, pp. 56075610.
Patanker, S., 1980, Numerical Heat Transfer and Fluid Flow, Hemisphere,
New York.
van Leer, B., 1977, Towards the Ultimate Conservation Difference Scheme 4:
A New Approach to Numerical Convection, J. Comput. Phys., 23, pp. 276
299.
Rhie, C., and Chow, W., 1983, Numerical Study of the Turbulent Flow Past
an Airfoil With Trailing Edge Separation, AIAA J., 21, pp. 15251532.
White, F., 2006, Viscous Fluid Flow, McGraw-Hill, Singapore.
Fox, R., McDonald, A., and Pritchard, P., 2006, Introduction to Fluid Mechanics, Wiley, Hoboken, NJ.
Collins, J., Shapiro, A., Kimmel, E., and Kamm, R., 1993, The Steady Expiratory Pressure-Flow Relation in a Model Pulmonary Bifurcation, ASME J.
Biomech. Eng., 115, pp. 299305.
Isono, S., Feroah, T., Hajduk, E., Brant, R., Whitelaw, W., and Remmers, J.,
1997, Interaction of Cross-Sectional Area, Driving Pressure, and Airflow of
Passive Velopharynx, J. Appl. Physiol., 83, pp. 851859.
Flow Characteristics in a Curved

Rectangular Channel With
Variable Cross-Sectional Area
Avijit Bhunia
e-mail: abhunia@teledyne.com
C. L. Chen
Teledyne Scientific Company,
1049 Camino Dos Rios,
MC A10,
Thousand Oaks, CA 91360
Laminar air flow through a curved rectangular channel with a variable cross-sectional
(c/s) area (diverging-converging channel) is computationally investigated. Such a flow
passage is formed between the two fin walls of a 90 deg bend curved fin heat sink, used
in avionics cooling. Simulations are carried out for two different configurations: (a) a
curved channel with long, straight, constant c/s area inlet and outlet sections (entry and
exit lengths); and (b) a short, curved channel with no entry and exit lengths. Formation
of a complex 3D flow pattern and its evolution in space is studied through numerical flow
visualization. Results show that a secondary motion sets in the radial direction of the
curved section, which in combination with the axial (bulk) flow leads to the formation of
a base vortex. In addition, under certain circumstances the axial and secondary flow
separate from multiple locations on the channel walls, creating Dean vortices and separation bubbles. Velocity above which the Dean vortices appear is cast in dimensionless
form as the critical Dean number, which is calculated to be 129. Investigation of the
friction factor reveals that pressure drop in the channel is governed by both the curvature
effect as well as the area expansion effect. For a short curved channel where area
expansion effect dominates, pressure drop for developing flow can be even less than that
of a straight channel. A comparison with the flow in a constant c/s area, curved channel
shows that the variable c/s area channel geometry leads to a lower critical Dean number
and friction factor. DOI: 10.1115/1.3176970
Introduction
With the ever increasing trend of power density of electronic

devices, thermal management is becoming a growing challenge,
particularly when the application platform precludes liquid cooling. Various fin structures and surface modifications have been
considered in the past few decades to improve air cooling convection heat transfer, as reviewed by Garimella 1. Certain special
structure heat sinks have gained popularity in recent years, particularly for high end applications. Curved fin heat sink, such as
the one shown in Fig. 1, is an example from the avionics electronics industry. Air impingement on a curved fin heat sink has also
been considered for high performance computer chip cooling 2.
Flow in a curved fin heat sink is significantly more complex than
in a traditional, straight fin sink. The fundamental difference arises
due to the formation of secondary flow in the curved section. The
secondary flow is beneficial as it alters the boundary layer formation bringing in cold fluid from the core of the channel toward the
curved walls, thereby enhancing heat transfer 3. However,
curved flow passages also experience a higher pressure drop, compared with straight channels of similar length. An overall system
figure of merit FOM of curved channel over straight channel
may be defined as the ratio of heat transfer enhancement to pressure drop increase. A thermal designer strives to achieve a high
FOM, where the benefit of a curved channel over a straight channel is realized by significant heat transfer enhancement without
excessive rise in pressure drop. An important tool in the design
process is a knowledge base on the fundamental flow and thermal
characteristics in the individual curved flow passages between
fins, as it forms the building block of the overall heat sink performance. Keeping this in mind we have carried out a study of air
JOURNAL OF FLUIDS ENGINEERING. Manuscript received September 10, 2008; final
manuscript received May 29, 2009; published online August 13, 2009. Assoc. Editor:
Rajat Mittal. Paper presented at the 2008 ASME International Mechanical Engineering Congress IMECE2008, Boston, MA, October 31November 6, 2008.
flow through a single passage formed between two curved fins of

the heat sink shown in Fig. 1. The passage has a constant c/s area
straight inlet section, followed by a variable c/s area curved section a 90 deg bend, and another constant c/s area, straight outlet
section. The present paper is the first part of the study that focuses
solely on the flow characteristics. The heat transfer aspects of the
problem will be published in a separate article.
In a curved channel, the fluid elements are subject to centrifugal
force. To balance this extra body force and the local viscous
stresses, a radial inward pressure gradient P / r develops
from the outer curvature wall OCW or the concave wall toward
the inner curvature wall ICW or the convex wall. However, a
perfect balance is not achieved over the entire c/s area. For example, within the boundary layers close to the walls, where local
velocity is significantly less than the bulk fluid, the centrifugal
force is weak and P / r cannot be balanced. The local imbalance
results in a radial or secondary flow from OCW to ICW. The
dimensionless number characterizing flow in a curved channel is
the Dean number De, defined as De= Re Dh / R, where Re is
the Reynolds number, Dh is the channel hydraulic diameter, and R
is the radius of curvature of the inner curvature wall. The structure
and strength of the secondary flow depend on the channel geometry and flow velocity Vi. Figure 2a shows the schematic of a
typical secondary flow created as P / r exceeds the centrifugal
force within the velocity boundary layers near the top and bottom
walls of the curved channel. The resulting secondary flow is from
OCW to ICW near the walls, and back toward the OCW around
the midheight of the channel. This is representative of a low axial
channel flow bulk velocity condition Vi, when only one pair of
secondary flow cells develops and remains conformal to all the
channel walls. Conventionally this is termed as the main secondary flow. With increasing Vi, as the secondary flow strengthens, it
has a tendency to separate from one or more walls. For example,
in the schematic of Fig. 2b, the secondary flow detaches from
the top wall TW, attaches to the ICW, and then again detaches
from the ICW. Each secondary flow separation creates a local
SEPTEMBER 2009, Vol. 131 / 091102-1
Fig. 1 A schematic of curved fin heat sink. Representative air

flow paths between two sets of fins are shown by red arrows.
recirculation cell as shown in Fig. 2c. For higher velocity conditions, in addition to the main secondary flow, multiple secondary flow structures are formed, commonly known as Dean vortices, after the first analysis by Dean 4. The nondimensional
number characterizing the onset of Dean vortices is called the
critical Dean number Decrit.
The curved channel/duct flow is widely encountered in many
engineering applications, such as bent tube heat exchangers, turbine blades, air conditioning ducts, the condenser area of solar
collector heat pipes, passages in biological systems, etc. A significant amount of prior research has investigated the flow and thermal fields of a curved channel. Since this article focuses on the
flow characteristics, we only cite the flow related literature for the
sake of brevity. Cheng et al. 5 experimentally visualized the
secondary flow pattern in curved rectangular channels, 25 mm
wide b, curvature ratio CR of 5, and the aspect ratio AR
varying between 1 and 12. CR is defined as the ratio of radius of
curvature R and the channel width b, while AR is the ratio of
channel height H and width b. Sugiyama et al. 6 reported
flow visualization at the end of 180 deg turn in channels, 20 mm
wide, CR varying between 5 and 8, and AR varying from 0.5 to
2.5. Ligrani and Niver 7 experimented with a nearly 2D channel
AR = 40 for De varying between 40 and 220, and showed the
formation of secondary flow patterns at various angular positions
along the curved section. Effort on the numerical front dates back
to the early 80s. Cheng and Akiyama 8 reported a 2D model of
fully developed laminar flow entering a curved rectangular channel, while Ghia and Sohkey 9 and Humphrey et al. 10 investigated the developing flow. In these calculations, AR varied in the
range of 0.25, and De was limited to 100. Hwang and Chao
11 investigated fully developed laminar flow in a curved square
duct under isothermal condition. More recently, Chandratilleke
and Nursubyakto 12 presented a 3D computational model of
flow through a curved rectangular channel with uniform heating at
the outer concave wall. They considered parametric ranges of
1 AR 8 and 20 De 500, and included the effect of buoyancy. To summarize the findings of these research worksThe
Decrit value; the main secondary flow and the Dean vortices for
De Decrit condition all depend on several geometric and thermal
conditions, such as the channels, AR and CR location along the
curved section and the wall heating pattern.
In spite of 3 decades of research, most prior works report flow
pattern at a specific cross section. Literature on the evolution of
multiple vortices in a curved channel remains scarce. In an experi091102-2 / Vol. 131, SEPTEMBER 2009
Fig. 2 Schematics of secondary flow structures in a curved

channel. a Secondary flow conformal to all the walls occurs at
low axial flow velocity Vi, b secondary flow separation from
multiple walls at higher Vi, and c main/base secondary flow
and additional recirculation cells due to secondary flow
separation.
mental approach, this needs detailed flow visualization at multiple

cross sections, such as in Ref. 7, which is quite challenging. On
the other hand, a computational model to resolve such complicated 3D flow pattern is extremely grid sensitive. Furthermore, all
the curved channel flow works carried out so far consider constant
c/s area only. The present work focuses on a variable c/s area
curved channel, where the flow is governed by a combination of
Fig. 3 Schematic of an individual flow passage between two

fins, a long channel, and the numerical simulation domain
variable channel geometry and curvature effect. Motivation arises

from the practical application platform shown in Fig. 1.
We developed a 3D computational model of laminar air flow
through a 90 deg bend of a variable c/s area curved channel where
the channel area varies along the curvature length. The channel c/s
area at the curvature inlet and outlet are the same. In between, the
channel width b first increases along the length of curvature up
to the 45 deg angular position and then decreases, making it a
diverging-converging curved channel. In this article we present
only the flow characteristics part of the study. The primary focus
is on elucidation of the flow field, i.e., understanding the interaction between the secondary flow and the axial bulk flow, the
resulting vortices and their evolution in space. At this stage we are
not conducting a parametric study of the several important geometric features, such as variable c/s area ratio, CR, AR, channel length, etc., that govern the flow field. Therefore, geometry is
maintained constant in this study, specific to the heat sink Fig. 1.
Simulations are carried out for two configurations: a a divergingconverging curved channel with long, straight, constant c/s area
inlet and outlet sections entry and exit lengths; and b a
diverging-converging curved channel with no entry or exit length,
respectively, marked as the long and short channels in Fig. 1. The
formation, evolution, and dynamics of multiple vortices are investigated, along with the critical Dean number and the channel pressure drop. Comparisons between the variable and constant c/s area
channels are presented.
the inlet section b is used for calculating all the geometric parameters such as hydraulic diameter Dh, AR, and CR.
For both the long and short channels, air at temperature Ti
enters the channel at a uniform flow velocity of Vi. Ambient pressure boundary condition is specified at the channel exit. The channel bottom and the side walls are maintained at a constant temperature of Tw Tw Ti. The isothermal side wall assumption
essentially means that the fins of the heat sink shown in Fig. 1
are 100% efficient. Indeed a computational model of the entire
heat sink made of aluminum shows that the fins are close to 99%
efficient. In a practical avionics application, often the top of the
fins is covered by a plate to avoid flow bypass. To simulate this
condition the top channel wall is assumed adiabatic.
The two critical dimensionless flow parameters are the Reynolds number Re= ViDh / and the Dean number De
= Re Dh / R. Vi values are chosen on the basis of a practical
range of air flow rates used in the heat sink. Air properties at
average temperature Tav = Ti + Tw / 2 are used to report these
numbers. Re and De vary between the range of 183 to 2017 and
92 to 1008, respectively. The last air flow passage, having a length
Lav / Dh = Lcurv / Dh of 6.8, represents one asymptotic condition
where uniform flow directly enters the curved section. Therefore,
it is picked as the short channel in this study. However, the other
asymptotic condition of fully developed flow entering the curved
section does not occur for any of the channels of this heat sink.
The dimensionless entry length Lfd / Dh for fully developed flow
in a constant c/s area straight channel of AR close to 2 is
0.02 Re 13. At the highest velocity condition of Re= 2017,
Lfd / Dh is 40. Not even the longest air flow passage of the heat
sink, shown in Fig. 1, has this much entry length. In other words
for Re= 2017, Lentry Lfd for all channels. Instead of focusing on
the ideal asymptotic condition, we use the current heat sink geometry and arbitrarily pick a long channel having an overall
average length of Lav / Dh = 39.1 Lav = Lentry + Lcurv + Lexit, with
Lentry / Dh = 16.1 and Lexit / Dh = 16.1. Flow entering the curved section of the chosen long channel is fully developed for Re 805.
Above this value, the flow enters the curved section in partially
developed state. Clearly if a different channel were chosen as the
long channel, magnitudes of the flow and thermal parameters
would be different. However, since this study is not intended to be
a parametric study of the geometric features, such as Lentry / Dh,
Lcurv / Dh, Lexit / Dh, CR, AR, etc., the specific values are not so
important. The primary objective here is to understand the complex flow field for a given geometry. Therefore, for both the long
and short channels, the channel geometry is kept constant, with
Dh = 0.007 m, bmax / b = 1.5, AR = 2.3, and CR = 5.6; and the
Lentry / Dh, Lcurv / Dh, and Lexit / Dh values discussed above.
3
2
Physical Model
Figure 3 shows a schematic of the air flow path in the long

channel. The flow passage consists of three segments: a a
straight, constant c/s area inlet section of length Lentry; b a 90
deg bend, a variable c/s area curved section of length Lcurv along
the middle of the cross section; and c another straight, constant
c/s area exit section of length Lexit. For a short channel there are
no entry and exit lengths Lentry = Lexit = 0. Flow directly enters the
variable c/s area curved section and leaves to the ambient at the
end of the channel. For both the long and short channels, the
channel height H remains constant. In the curved section, the
channel width first increases along the length of curvature, up to
an angular position of 45 deg and then decreases between the 45
deg and 90 deg position, making it a diverging-converging curved
channel. Width at the channel inlet and the inlet to the curved
section are b, which increases to 1.3b at the 22.5 deg angular
position, to a maximum of 1.5b at the 45 deg angular position, and
then back to b at the exit of the curved section. Channel width at
Numerical Simulation
Steady, laminar air flow through the passages are computationally simulated using the commercial computational fluid dynamics
CFD code FLUENT. The continuity, momentum, and energy
equations are solved. These equations are coupled due to the use
of temperature dependent air properties. Some of the underlying
assumptions and observations are as follows.
1 Even though all the features of solving compressible flow
are considered, the flow is essentially incompressible. The
Mach number Ma based on the highest inlet velocity Vi
is 0.015.
2 The value of Gr/ Re2 is 9 107 for the lowest flow velocity condition, where Gr is the Grashof number. Therefore, buoyancy effects are neglected. It may be argued that
the local velocity at any point in the channel could be significantly lower than Vi. Therefore, if the above ratio is
calculated with Re value based on the local velocity, the
Gr/ Re2 ratio can actually become orders of magnitude
higher. To address this concern, we carried out a calculation
SEPTEMBER 2009, Vol. 131 / 091102-3
carried out a flow visualization study at the end of the bend and
reported the critical Dean numbers Decrit for the onset of Dean
vortices for various channel aspect ratio AR = H / b and curvature
ratio CR = R / b values. Figure 4b shows a comparison of the
predicted values from our computational model with the experimental data, for a constant CR value of 5. Note that this CR value
is very close to the variable c/s area curved channel geometry
investigated in this article CR = 5.6. This experimental paper calculated De based on channel width b, i.e., De= Re b / R, even
though Re is calculated based on Dh Re= ViDh / . We report De
based on Dh, i.e., De= Re Dh / R. However, for the sake of the
present comparison, just for this section Fig. 4b, we follow
their procedure. An excellent agreement between the prediction
and experimental data in Fig. 4b, for the entire range of AR
values, lends significant confidence about the present computational model.
Fig. 4 Validation of the numerical model against the experimental data of Sugiyama et al. 6. a Schematic of the experimental channel geometryconstant cross-sectional area
curved channel with a 180 deg bend. Lentry = 1000 mm, b
= 20 mm, CR = R / b = 5, and H and AR=H / b variable; b comparison of numerical prediction of the critical Dean number
Decrit with the experimental data 6.
at the lowest velocity condition, with the buoyancy effect

on. The results showed no significant difference with or
without buoyancy.
3 The viscous dissipation term is neglected in the energy
equation as the Brinkman number Br= V2i / k f Tw Ti is
3 104.
4 Temperature dependent air properties are used. For example, a density a linear function, b thermal conductivity k f a linear function, and c viscosity
Power law / o = T / To2/3, where o and To are
1.716 105 kg/ m s and 273 K, respectively.
A coupled pressure velocity solver algorithm is used for a robust and efficient solution. The first few iterations are run using
SIMPLE algorithm, results of which are used as the initial guess
for the coupled solver. A Courant number value between 10 and
20 is used to stabilize the convergence behavior. The convection
terms in the momentum and energy equations are discretized using a second order UPWIND scheme. For the pressure term, the
PRESTO scheme is used, recommended for flow with strongly
curved domains. Further details on the numerical scheme are
available in the FLUENT manual 14.
Validation of the Numerical Methodology
The accuracy of the numerical procedure is ascertained by validating the model against existing experimental data from literature. For this purpose, a computational model is developed with
channel geometry replicating the experimental set up of Sugiyama
et al. 6. Isothermal air flow through a constant c/s area 180 deg
bend, as shown in Fig. 4a, is considered. In the model, a uniform
velocity is prescribed at the inlet of the straight entry section
Lentry, whereas an ambient pressure outlet boundary condition is
specified at the end of the 180 deg bend. Sugiyama et al. 6
091102-4 / Vol. 131, SEPTEMBER 2009
Results and Discussion
Prior to discussing the specific results, we start with outlining a

few basic conceptual points. There are only two fundamental flow
mechanisms in a curved channel: a the axial or bulk flow, i.e.,
the original flow component, and b the secondary flow, i.e., the
created component at the curved section. The secondary flow may
perfectly follow all the channel wall contours as shown in the
schematic of Fig. 2a. Or it may separate from the channel walls
at multiple locations, creating a main/base secondary flow and
additional recirculation cells Fig. 2c. Similarly, depending on
the channel geometry and flow condition, the axial flow may also
separate from a channel wall, causing local axial flow reversal.
Between these two basic mechanisms, a total of four different
combinations are possible toward the evolution of multiple complex 3D flow structures:
A
B
C
D
forward moving axial flow and conformal secondary flow

following the channel walls fully conformal or over a
certain portion of a channel wall
axial flow reversal and conformal secondary flow
forward moving axial flow, and secondary flow separation
both axial and secondary flow separation and reversal
The following acronyms will be used for sake of convenience:

a constant c/s area straight channel CASC, b constant c/s area
curved channel CACC, and c variable c/s area curved channel
VACC.
Grid Independence Study
A grid independent numerical solution has to resolve the flowthermal characteristics of the overall channel as well as all the
secondary recirculation cells, at the highest flow rate condition.
The following parameters are used to quantify the secondary flow
structures at two angular positions, 45 deg and 90 deg: a the
total number of cells Ncells at that c/s area, b minimum helicity
Hmin, and c the dimensionless maximum velocity at that c/s
e.g., cross-sectional Vmax at 45 / Vi. In addition, two points A and
B are chosen, respectively, at the 45 deg and 90 deg angular
positions located at the core of the recirculating flow region. The
local dimensionless velocity such as VA / Vi, and temperature
such as A = TA Ti / Tw Ti are investigated at these points.
Four other dimensionless parameters are used to characterize the
overall channel transport fields. They are: the maximum velocity
at the exit Vmax,outlet / Vi of the long channel, the overall channel
pressure drop 2P / V2i , the average fluid exit temperature
exit = Tav,exit Ti / Tw Ti, and the Nusselt number Nuo
= hoDh / k f , characterizing the total heat transfer from the channel.
Values of these local and global parameters, obtained from computational simulations with four different grid sizes, are presented
in Table 1. Based on the comparison, a 351,000 cell grid is chosen
Table 1 Grid independence study for a long VACC for the highest flow velocity condition. Re= 2017 and De= 1008 based on
uniform flow velocity Vi and hydraulic diameter Dh at channel inlet and properties at average temperature Tav = Ti + Tw / 2.
AR = 2.3, CR = 5.6, Lav / Dh = 39.1, Lentry / Dh = 16.1, and Lexit / Dh = 16.1. Location of point A: axial locationon the 45 deg angular position from the start of the curvature, heightat H/2, and lateralat b45 deg / 4 closer to the ICW. Location of point B: axial
locationon the 90 deg angular position; height and lateral locationat the geometric center of the curvature outlet plane.
Grid size No. of cells
Flow-thermal characteristics
126,720
230,400
351,000
620,880
At 45 deg angular position
No. of recirculation cells Ncells

Min. helicity Hmin in m / s2
Max. dimensionless velocity Vmax / Vi
Dimensionless velocity at A VA / Vi
Dimensionless temperature at A A
2
44,027
1.42
0.17
0.78
2
51,280
1.45
0.08
0.79
6
51,686
1.45
0.10
0.88
6
52,079
1.45
0.09
0.82
At 90 deg angular position
No. of recirculation cells Ncells

Min. helicity Hmin in m / s2
Dimensionless velocity at B VB / Vi
Dimensionless temperature at B B
6
24,775
1.62
1.30
0.31
6
28,313
1.64
1.35
0.32
6
28,610
1.64
1.35
0.32
6
28,776
1.65
1.37
0.31
At channel outlet

Av. dimensionless temperature av
1.77
0.61
1.82
0.62
1.85
0.62
1.85
0.62
2.97
5.90
2.99
6.03
3.03
6.09
3.04
6.10
Dimensionless overall channel pressure drop= 2P / V2i

Dimensionless heat transfer coefficient Nuo based on total heat transfer from the channel
for all calculations on the long VACC. Similar grid independence

studies are carried out for other geometries. The optimal grid size
for a long CACC is also 351,000. For short channels both CACC
and VACC a 65,600 cell grid is used.
Long CACC: Flow in the Curved Section
As the channel geometry plays a pivotal role in the flow dynamics, it is useful to decouple the effect of variable c/s area first.
Therefore, we start with investigating flow through a 90 deg bend
of a long CACC with entry and exit lengths. The channel geometry matches the schematic in Fig. 3, except for the fact that channel width b remains constant through the curved section. These
results serve as the baseline case for comparison with those of
variable c/s area curved channel. Even though flow through
CACC has been widely investigated in the past, studies focusing
on the space evolution of the flow structures remain scarce.
To follow the complex 3D secondary flow pattern, we investigate the cross-sectional view of the secondary flow, i.e., the inplane velocity vectors, at various angular positions Figs.
5a5e, in conjunction with the oil flow path lines on the ICW,
OCW, and TW Figs. 6a6c, respectively. The in-plane velocity vector plots are color coded by dimensionless local velocity
magnitude =local velocity V/uniform channel inlet velocity
Vi. Since the flow is symmetric heightwise, in Fig. 5 the computational results are shown in only half the c/s, while the other
half is used for explanatory schematics. The oil flow lines are
plotted by releasing particles from the same curved section walls
ICW, OCW, or TW on which the flow is visualized. The lines
are color coded with particle ID. Thus the color codes do not have
any physical significance as such. To retain the clarity of the oil
flow line plots, various markings are shown on a separate, identical plot below each figure.
The secondary flow sets in immediately after the fluid elements
enter the curved section. This secondary flow along with the forward moving axial flow, i.e., combination A, creates a pair of base
vortices, at the top and bottom portion of the channel. The traces
of base vortices in the oil flow lines plot Fig. 6 indicate that they
travel through the entire length of the curved section 090 deg
angular position, although their shape and size change along the
path. At the early portion of the curved section, the secondary
flow is fully conformal to the channel walls, as in Fig. 2a. ThereJournal of Fluids Engineering
fore a pair of base vortices covers the entire c/s area of the channel, as seen in Fig. 5a for an angular position of 11 deg. Flow
from the upper and lower base vortices collide and separate from
the ICW and attach to the OCW, both around midheight, respectively, forming separation line S1 in Fig. 6a, and attachment line
A1 in Fig. 6b.
Further downstream, the secondary flow starts to separate from
the walls, in the corner regions between ICW, and the top and
bottom walls. As the inset of Fig. 5b shows, the secondary flow
separates from the TW and attaches to the ICW, forming a local
recirculation similar to that shown in the schematic of Fig. 2c.
The oil flow lines in Fig. 6c indicate that the separation from
TW starts at an angular position P1, just before the 22.5 deg
location. This local, corner region, secondary flow recirculation
can be traced along separation line S2 on TW Fig. 6c and
attachment line A2 on ICW Fig. 6a, both continuing all the way
to the end of the curved section. This secondary flow reversal,
along with the forward moving axial flow combination B leads
to the formation of another local vortex. Since this is in addition
to the base vortex, following the traditional definition, it is categorized as a Dean vortex. We name it the corner region Dean vortex.
The oil flow lines in the inset of Fig. 6a show another interesting aspect of the corner region flow. Between the angular positions P2 and P3, the oil flow lines are backward and upwards.
The backward axial flow can also be seen from the insets B and C
of Fig. 6c. This indicates an axial as well as secondary flow
separation in the region, or combination D, that leads to the formation of a closed separation bubble. Figure 7a shows a schematic of this corner region axial and secondary flow separation,
and the formation of the separation bubble. The corner region
Dean vortex bypasses this separation region by detaching from the
ICW, squeezing toward the separation line S2, and then coming
back to the ICW beyond the separation region.
After attaching to the ICW along attachment line A2, the secondary flow follows the wall contour for a while, flowing toward
the centerline heightwise. However, before reaching the centerline it bounces off the ICW, causing another local secondary flow
separation and flow reversal, as sketched in Fig. 2b. The trace of
this separation line S3 in Fig. 6a, and Fig. 5c indicates that this
flow separation starts just upstream of the 34 deg position and
increases in size along the curved path follow Figs. 5c5e.
SEPTEMBER 2009, Vol. 131 / 091102-5
(a)
(c)
(b)
(d)
(e)
Fig. 5 Formation of secondary flow at various angular positions along a 90 deg bend of a long CACC. Re= 2017 and De= 1008
based on uniform flow velocity Vi and hydraulic diameter Dh at channel inlet, and properties at average temperature Tav
= Ti + Tw / 2. Lav / Dh = 39.1, Lentry / Dh = 16.1, Lexit / Dh = 16.1, AR = 2.3, and CR = 5.6. Views: a angular position 11 deg, b 22.5 deg, c
37 deg, d 56 deg, and e 90 deg outlet from the curved portion of the channel or beginning of the straight constant c/s area exit
channel.
The angular position where this separation started cannot be precisely located from the current data. We tentatively identify an
angular position P4. Coupled with the forward moving axial flow
this secondary flow separation Combination B leads to the formation of another Dean vortex, which we term as the ICW Dean
vortex. Flow from the top and bottom ICW Dean vortices meet at
the heightwise centerline of ICW. Thus, beyond the angular position P4, the ICW centerline becomes an attachment line A3 Fig.
6a.
A comparison of Figs. 5a5e shows that as the flow moves
downstream, the return secondary flow from ICW back toward
OCW gets stronger. As a result the location of the highest velocity in the channel shifts toward the OCW and the velocity boundary layer there gets thinner. An abrupt drop in the centrifugal force
occurs across this thin boundary layer near the OCW. At some
location between the 45 deg and 56 deg angular position, it causes
091102-6 / Vol. 131, SEPTEMBER 2009
another separation of the secondary flow, this time from the OCW.
A local flow reversal follows, as seen in Figs. 5d and 5e. This
along with the forward moving axial flow combination B creates
the third pair of Dean vortices in the channel, now termed as the
OCW Dean vortex. It is bounded by the separation line S4 and
attachment line A4 in the oil flow lines of Fig. 6b.
Figures 5d and 5e show an additional pair of recirculation
cells inside the c/s, adjacent to the base vortex. Following the
conventional definition, this should be another Dean vortex. The
origin of this vortex is not clear. It is possible that this vortex
arises from a local split of the base vortex, somewhere downstream of the 45 deg angular position. This Dean vortex weakens
shrinks in size and gradually shifts toward the center of the
channel in the downstream direction.
To summarize, Fig. 7b shows a schematic of the base vortex
and all the Dean vortices in the top half of the channel, close to
Fig. 6 Oil flow lines on the walls of the curved section 90 deg bend of a long CACC. Re= 2017 and De= 1008, Lav / Dh = 39.1,
Lentry / Dh = 16.1, Lexit / Dh = 16.1, AR = 2.3, and CR = 5.6. a ICW or concave wall, b OCW or convex wall, and c TW. All views are
looking into the wall from front.
the curvature exit. The origin of the multiple vortices and their
evolution in space are also summarized in Table 2.
Long VACC: Flow in the Curved Section
With a basic understanding of the flow dynamics in the CACC,

we now investigate the effect of the variable c/s area. Once again
we consider the in-plane velocity vector plots at various angular
positions Fig. 8 in conjunction with the oil flow lines on the
ICW, OCW, and TW Figs. 9a9c, respectively. A slight
asymmetry between the top and bottom of the channel arises due
to the difference in thermal boundary conditions. The top wall is
insulated while the bottom is isothermal, Tw Ti, and
temperature-dependent properties are used. It is interesting to note
that even though the exact same set of boundary conditions was
used for the CACC, there was no asymmetry. Clearly the temperature dependent property effect is more pronounced for the decelerating flow in the diverging section. Since our focus is on understanding the overall flow field, for all practical purposes we will
ignore the slight asymmetry in the subsequent discussions.
The main/base secondary flow at the early part of the channel
close to the curvature entry is very similar as before. Fully conformal secondary flow, in conjunction with the forward moving
axial flow combination A, creates a pair of base vortices at the
top and bottom half of the channel Fig. 8a. They meet at separation line S1 on ICW Fig. 9a and attachment line A1 on OCW
Fig. 9b. As before, the secondary flow separates from the corner regions between the ICW and the top and bottom walls somewhere between the 11 deg and 22.5 deg angular position Point P1
SEPTEMBER 2009, Vol. 131 / 091102-7
Fig. 7 Schematics of various 3D flow patterns in the curved section 90 deg bend of a long
CACC. Re= 2017 and De= 1008, Lav / Dh = 39.1, Lentry / Dh = 16.1, Lexit / Dh = 16.1, AR = 2.3, and
CR = 5.6. a Axial flow separation from ICW near the corner region and formation of a separation bubble, and b base vortex and other Dean vortices, such as corner region Dean
vortex, ICW Dean vortex, and OCW Dean vortex.
in Fig. 9c and follows separation line S2 all the way to the end
of the curved section. Between positions P2 and P3 inset A of
Figs. 9a and 9c, the axial flow also separates. The overall flow
field in the corner region consists of the corner region Dean vortex
and a separation bubble, similar to that sketched in Fig. 7a.
Flow in the diverging section of the VACC, i.e., between the
angular positions 045 deg, is significantly different from that of
the CACC. The first noticeable effect of the variable c/s area is
near the OCW, just after the flow enters the curved section. The
inset of Fig. 9b shows that at the top half of the channel some of
the oil flow lines are upwards and backward. The upward flow is
indicative of conformal secondary flow at OCW, while the backward flow is due to axial flow separation. The separated axial flow
and the conformal secondary flow lead to combination C, which
leads to the formation of a base vortex, but of a different structure.
Figure 10a shows a sketch of this base vortex structure. Figure
9b shows that the axial flow separation initially starts near the
top and bottom walls, and gradually expands toward the center
angular position P4, along the separation line S3. From the angular position of P5 on the centerline, the axial flow starts to
reattach to the wall along attachment line A2. The reattachment is
triggered by separation of the axial flow from the ICW discussed
later, which pushes the fluid back toward the OCW.
091102-8 / Vol. 131, SEPTEMBER 2009
In Fig. 9a, two angular positions P6 and P7 are identified on

the ICW. Between these two points, a series of oil flow lines
originate from the centerline and flow backward and toward the
top and bottom of the channel. This is a result of both the secondary and axial flow separation from ICW, along separation line S4.
The axial and secondary flow separation, or combination D, leads
to the formation of a closed separation bubble near ICW that
extends from just before the 22.5 deg position to angular position
P7, just after the 45 deg location. Figures 8b shows the presence
of this separation bubble near ICW. The separation bubbles at the
top and bottom of the channel are not symmetric heightwise. This
is due to asymmetry in the boundary condition, discussed earlier.
Flow from the two separation bubbles at the top and bottom of the
channel forms attachment line A3 on ICW Fig. 9a, which significantly deviates from the midheight centerline. The oil flow
lines in Fig. 9a show that the flow structure within this separation bubble is extremely complicated. Apparently there are multiple recirculation cells within the separation bubble. Figure 10b
shows a schematic of the separation, based on the preliminary
interpretation of these oil flow lines.
Another noticeable difference between the CACC and VACC is
the pattern of secondary flow near OCW, particularly in the converging section 4590 deg. In the CACC, the secondary flow
Table 2 Summary of flow pattern along the 90 deg bend of two long curved channels, a CACC and b VACC.
Re= 2017 and De= 1008, AR = 2.3, CR = 5.6, Lav / Dh = 39.1, Lentry / Dh = 16.1, and Lexit / Dh = 16.1.
from OCW to ICW was triggered by an imbalance in the radial

pressure gradient P / r within the boundary layers near the top
and bottom of the channel. As a result, the secondary flow from
OCW to ICW occurred predominantly within this boundary layer
and the flow returned to OCW over rest or most of the c/s area. On
the other hand for a VACC, the continually changing geometry
along the length introduces an inherent radial velocity component
toward ICW. As a result, the secondary flow from OCW to ICW
occurs over the entire channel height see Figs. 8c8e and
compare with Figs. 5d and 5e. An additional pressure gradient
from the top and bottom of the channel toward the center is also
formed near OCW P / z. The combined effect of these two
pressure gradients is a split of the secondary flow into two
streams, shown in Figs. 8c8e. One remains confined close to
the top or bottom wall, similar to the traditional base vortex discussed for CACC. The other branch flows over the rest of the c/s
area toward ICW. This part of the secondary flow leads to the
formation of another vortex shown in the schematic of Fig. 10c.
Since this vortex arises from a conformal secondary flow, it is not
a Dean vortex. We name it split base vortex. The branch-off starts
somewhere between the 45 deg and the 56 deg location, generating a separation line S5 on OCW Fig. 9b. Therefore the midheight centerline of OCW must transition from an attachment line
A1 in the diverging section to a separation line S6 in the converging section. Although the position of this transition cannot be
precisely located from the current data, we tentatively mark a
point P8 in Fig. 9b, the angular position of which is coincident
with the inception of S5.
The axial flow encounters a favorable pressure gradient in the
converging section and reattaches to the ICW at angular position
P7, just after the 45 deg location. The split base vortex starts to
form around the same location. The split base vortices from the
top and bottom of the channel meet at attachment line A3, around
the midheight centerline. The structure of the ICW and OCW
Dean vortices appear very similar to those in the CACC, with a
few exceptions. Unlike CACC where the ICW Dean vortices form
early in the curved channel, in VACC they appear only near 56
deg angular position. It is sandwiched between the split base vorJournal of Fluids Engineering
tex and the base vortex, within the boundaries of attachment line
A4 and separation line S7. The onset of the OCW vortex is also
delayed beyond 79 deg due to the existence of split base vortex,
unlike CACC where the OCW vortex appears just after the 45 deg
angular position. Along the outer curvature wall, since the flow
direction of OCW vortex is similar to that of split base vortex,
there is no additional attachment or separation line on OCW
other than S5 due to this Dean vortex. This is another noticeable
difference with CACC compare Fig. 9b with Fig. 6b.
The evolution of the complex flow structures in a VACC and its
difference with CACC are summarized in Table 2. Before closing
this discussion, it is important to re-emphasize that some features
of the complex flow structures is not yet well understood. Further
studies are much needed to shed light on those aspects.
Long VACC: Effect of Flow Velocity
The in-plane velocity vector plots in Figs. 11a11c, and Fig.

8e show the secondary flow structures at the outlet of the curved
section 90 deg angular position of a VACC, for various inlet
flow velocity Vi conditions. As before, these vector plots are
color coded by the dimensionless velocity local velocity V / Vi.
At the lowest inlet velocity condition of Re= 183 and De= 91.5
Fig. 11a, only the base vortex appears, characterized by a pair
of counter-rotating, recirculating cells covering the entire c/s area
of the channel. With increasing axial flow rate thus Re and De as
the boundary layer thickness decreases, the secondary flow from
OCW to ICW, and thus the base vortex, gets confined toward the
top and bottom walls, as seen in Figs. 11b and 11c. The thinner
boundary layer also progressively lowers the maximum dimensionless velocity in that cross section. At Re= 366 and De= 183 in
Fig. 11b, two additional pairs of vortices are observed, which are
Dean vortices by definition. They are identified as the ICW Dean
vortex and the corner region Dean vortex. Further calculations
show that the critical Dean number Decrit marking the onset of
the Dean vortices is 129. At the onset condition, the first Dean
vortex appears near the ICW. Since this is the first study of a
VACC, these results cannot be directly compared with any of the
SEPTEMBER 2009, Vol. 131 / 091102-9
(a)
(c)
(b)
(d)
(e)
Fig. 8 Formation of secondary flow at various angular positions along a 90 deg bend of a long diverging-converging VACC.
Re= 2017 and De= 1008 based on uniform flow velocity Vi and hydraulic diameter Dh at channel inlet, and properties at average
temperature Tav = Ti + Tw / 2. Lav / Dh = 39.1, Lentry / Dh = 16.1, and Lexit / Dh = 16.1. AR = 2.3, CR = 5.6. Views: a angular position 11 deg,
b 22.5 deg, c 56 deg, d 67.5 deg, and e 90 deg outlet from the curved portion of the channel or beginning of the straight
constant c/s area exit channel.
existing literature. Therefore, we also calculated the Decrit value

for a long CACC, which is 165. This is in the same ballpark range
of results shown in Fig. 4 and the experimental literature 6. In
Fig. 4, we reported the Decrit values for a long CACC at the end of
a 180 deg bend. For AR = 2.5 and CR = 5, the calculated Decrit
value was 212. The difference in the calculated Decrit value between Fig. 4 and the present case 212 and 165, respectively can
be attributed to multiple factors, such as: a the flow entering the
curved section in Fig. 4 and in Ref. 6 is fully developed, while
the present case is not; b the AR and CR values used in the
validation case of Fig. 4 and the present problem are close but
different; and c the validation calculations are isothermal,
whereas temperature dependent properties are used here.
A comparison of the CACC and VACC results reveals that the
later has a lower Decrit value. This is due to the accelerating flow
in the converging section of the VACC that leads to a stronger
secondary flow compared with that in the CACC. The Decrit value
of 129 also marks the onset condition of split base vortex. The
radial pressure gradient P / r remains unbalanced over the entire channel depth. As a result, secondary flow from the OCW to
the ICW occurs over the entire channel height and the flow splits
into two streams. At low axial velocity, e.g., De= 183 in Fig.
091102-10 / Vol. 131, SEPTEMBER 2009
11b, the split in base vortex occurs very close to the centerline
heightwise. With increasing axial flow velocity, the location of
the split shifts away from the centerline and the split base vortex
covers a wider area of the channel. At De= 366 in Fig. 11c, an
early sign of formation of another pair of Dean vortex at the OCW
is also observed. The OCW Dean vortex grows in size with increasing De, as seen in Fig. 8e, De= 1008.
10
Short VACC: Flow Pattern
In a short channel that has no entry and exit lengths, uniform

flow enters the curved section at 0 deg angular position and leaves
the channel at the 90 deg position to an ambient pressure outlet
boundary condition. It is important to emphasize at this point that
the secondary flow structure and pressure drop discussed later in
a short channel is highly dependent on the inlet boundary condition. In reality, the flow entering the short channel of the heat sink
Fig. 1 is neither uniform nor fully developed. Therefore, the
present results for short channel geometry should be treated in a
qualitative sense that underscores the role of curvature and variable c/s area by comparing various geometric configurations, such
(a)
(b)
(c)
Fig. 9 Oil flow lines on the curved section 90 deg bend of a long VACC. Re= 2017 and De= 1008, Lav / Dh = 39.1, Lentry / Dh = 16.1,
Lexit / Dh = 16.1. AR = 2.3, and CR = 5.6. a ICW or concave wall, b OCW or convex wall, and c TW. All views are looking into the
wall from front.
as CACC, VACC, and CASC. The specific quantitative values

shown here may not quite represent the real condition.
Figure 12 shows the nondimensional, in-plane velocity vector
plots of secondary flow at various angular positions. Since the
flow outlet is at 90 deg, we show the secondary flow at a plane
just upstream, at the 89 deg position. Even though the geometry of
the short channel is exactly same as that of the curved portion of
the long channel, the flow pattern is significantly different compare Fig. 12 with Fig. 8. This is mainly due to the boundary layer
thickness. Uniform flow entering the short, curved channel forms
a much thinner boundary layer compared with the long channel
with an entry length Lentry / Dh of 16.1, discussed earlier. As a
result, the main secondary flow from the OCW to ICW, that forms
the base vortex, remains confined close to the top and bottom
walls.
In addition to the base vortex, a few other flow features are
similar to those for the long channel. For example, a The corner
region Dean vortex in the ICW and top/bottom wall corner is
formed here too, although it is much smaller in size and appears

further downstream near 45 deg position as observed in Fig.
12c. b The oil flow lines not shown here indicate that the
axial flow separation from OCW is observed here also, although
only over a short length along the curved wall, between the 22.5
deg and 34 deg angular positions Fig. 12b. c Splitting of
secondary flow from OCW to ICW in two streams, one toward the
top/bottom walls, and another toward the centerline. However, in
the present case the splitting starts before the 45 deg angular position. d Formation of a pair of OCW Dean vortex near the
centerline Fig. 12e, just before the flow exit.
There are two noticeable differences between the flow pattern
of the long and short channels. For example, a Formation of the
additional recirculation cell near OCW at the 45 deg angular position Fig. 12c. An investigation of the oil flow lines indicate
that this is an OCW Dean vortex that originates just downstream
of the 34 deg position and terminates on the wall between the 56
deg and 67.5 deg positions. b For the short channel, there is no
SEPTEMBER 2009, Vol. 131 / 091102-11
Fig. 10 Schematics of various 3D flow patterns in the curved section 90 deg bend of a
long VACC. Re= 2017 and De= 1008, Lav / Dh = 39.1, Lentry / Dh = 16.1, Lexit / Dh = 16.1. AR = 2.3, and
CR = 5.6. a Axial flow separation near OCW and the resulting base vortex structure at the
early part of the curved channel, b axial and secondary flow separation from ICW, and
formation of closed separation bubble, and c base vortex, split base vortex, and multiple
Dean vortices at the converging section of the channel.
axial flow separation from ICW in the diverging section, or secondary flow separation from ICW in the converging section. As a
result, Fig. 12 does not show an ICW Dean vortex and separation
bubble which were observed in Fig. 8.
The effect of flow velocity De on the formation of secondary
flow has also been investigated. The generic conclusions are very
similar to that of a long channel. At the lowest velocity condition
091102-12 / Vol. 131, SEPTEMBER 2009
of De= 92, only the base vortex appeared in the channel. The
onset of additional secondary vortices occurs at Decrit = 129, same
as that of the long channel. However, here the onset occurs in the
corner region.
To summarize, the secondary flow in the short, variable c/s
area, curved channel is predominantly characterized by the base
vortex and the split base vortex. The Dean vortices are smaller in
(a)
(b)
(c)
Fig. 11 Effect of axial flow velocity Re and De on the growth of secondary
flow at the curvature outlet at the 90 deg bend of a long VACC. Lav / Dh
= 39.1, Lentry / Dh = 16.1, Lexit / Dh = 16.1. AR = 2.3, and CR = 5.6. a Re= 183, De
= 92; b Re= 366, De= 183; and c Re= 732, De= 366. Re and De are based on
uniform inlet flow velocity Vi, hydraulic diameter Dh at channel inlet, and
properties at average temperature Tav = Ti + Tw / 2.
size and often localized, i.e., they do not cover the entire length of
the curved section.
11
Pressure Drop and Friction Factor
Pressure drop between the channel inlet and exit P is represented in a dimensionless form of friction factor f, defined as
f=
P Dh
1 2 4L
V
2 i
The effects of flow velocity Vi and channel geometry curvature and variable c/s area effect on f are investigated. For the
long channel P includes pressure drop over the entire length
Lentry, Lcurv, and Lexit, while for the short channel it is over Lcurv
only. For both the long and short channels, three different geometric configurations are investigated: the CASC, the CACC, and the
VACC. Traditionally for developing flow, the product of friction
factor and Reynolds number f Re is plotted against the axial
location x, expressed in dimensionless form as x+ = x / Dh Re.
Figures 13a and 13b show the variation of f Re as a function

of the dimensionless total channel length L+ = L / Dh Re, respectively, for the long and short channel. Since Dh and L are
constant, L+ essentially represents the effect of Re or inlet flow
velocity Vi. The f Re curves for CASC in Fig. 13 closely
follow the standard results of the well established literature, such
as 15. The slight difference of actual f Re values are possibly
due to the use of temperature dependent properties in the present
calculation.
It is well known e.g., review in Ref. 16 that the formation of
secondary flow in a curved channel creates higher P and thus
f, compared with that in a straight channel of the same length.
Consistent with the prior literature, Figs. 13a and 13b show
that f Re for CACC is always higher than that for CASC. In
case of VACC, the velocity boundary layer thickness increases in
the diverging section 045 deg, the axial flow slows down, and
pressure increases in the downstream direction pressure recovery. In the converging section 4590 deg as the flow accelerates, velocity boundary layer starts to decrease in the downstream
direction and P increases. On the whole it can be argued that
SEPTEMBER 2009, Vol. 131 / 091102-13
Fig. 12 Formation of secondary flow at various angular positions along a 90

deg bend of a short VACC. Re= 2017, and De= 1008, Lav / Dh = 6.8, Lentry / Dh
= Lexit / Dh = 0, AR = 2.3, and CR = 5.6. View a angular position 11 deg, b 37 deg,
c 45 deg, d 67.5 deg, and e 89 deg, just prior to the channel and curvature
outlet.
091102-14 / Vol. 131, SEPTEMBER 2009
Fig. 13 The effects of flow velocity and channel geometry on

the pressure drop P between the channel inlet and outlet.
Pressure drop expressed in dimensionless form as product of
friction factor and Reynolds number f Re, and velocity as L+.
a Long channel and b short channel.
over the curved length Lcurv, the average flow area for VACC is
greater than that of the CACC. Therefore, the P for the former is
expected to be lower than its constant area counterpart, as observed in Fig. 13.
Comparison of the short length VACC with CASC in Fig. 13b
is quite striking. P or f Re for VACC is even less than that for
CASC. This is quite counter intuitive considering the presence of
secondary flow in the curved geometry. At the highest flow rate
condition of Re= 2017 corresponding L+ 0.004, f Re for
VACC is 25% lower than that of CASC. Evidently, the results
suggest that the area expansion effect overwhelms the secondary
flow effect. An order of magnitude, scaling analysis of P can
provide a reasonable explanation of this phenomenon. The inplane velocity vector plots in Fig. 12 show that even though the
secondary flow takes place over the entire channel c/s area, the
velocity boundary layer of the axial bulk flow is very thin. As a
result, the axial flow is uniform over most of the channel c/s area.
Short CACC and CASC also show a similar pattern. Therefore for
a short channel, P should scale with the pressure drop in the
core uniform flow area, or the difference in kinetic energy between the channel inlet and exit, i.e., P Vmax2 Vi2, where
Vmax is the maximum velocity at the channel exit. For the three
geometric cases shown in Fig. 13b, at Re= 2017, Vmax / Vi values
are 1.43, 1.39, and 1.30, respectively, for CACC, CASC, and
VACC. Figure 13b shows that at any Re, f Re follows the
decreasing trend in the same order. Based on the above Vmax valJournal of Fluids Engineering
ues, the ratio of PVACC / PCASC should be 0.74. The scaling

analysis estimate is strikingly close to the actual ratio shown in
Fig. 13b for Re= 2017.
A similar scaling analysis cannot be applied to the long channel, though. Figures 5 and 8 show that the boundary layer thickness at various angular positions of the curved section can be as
high as b / 2, or more. This indicates that the viscous effect is
dominant in such flow and P should scale with the viscous
stress, i.e., P / L Umax / 2, where is the velocity boundary
layer thickness. However, since the flow is developing throughout
the channel length, Umax / 2 is a variable parameter, and thus
hard to estimate from a simple scaling analysis. A better insight
into the results of Fig. 13a is obtained by looking into the pressure drop at individual length segments Lentry, Lcurv, and Lexit. It
turns out that for all the three geometric configurations, nearly
50% or more of the total channel P occurs in the inlet section,
over the length Lentry. Only 1215% of the total pressure drop
takes place over Lcurv. The difference in overall P is created over
the exit length Lexit. Since the secondary flow generated in the
curved section continues into the exit length, P over Lexit is
significantly higher for both the curved channels CACC and
VACC, compared with that for the straight channel CASC. This
explains why f Re for CACC and VACC are both higher than
that of CASC, unlike the results for short channel.
For the short channel, we concluded that the area expansion
effect in a VACC is the dominant factor that dictates the P,
which is even lower than that in a CASC. To judge how much of
this conclusion is affected by the uniform velocity inlet boundary
condition, we investigated the P over the length of the curved
section Lcurv only, in the long channel. Here the flow entering
the curved section is partially developed. We observed a similar
trend, i.e., PVACC PCASC PCACC. However the ratio is different. For example, for the highest flow rate condition of Re
= 2017, PVACC / PCASC, and PCASC / PCACC over the length
Lcurv of a long channel are 0.96 and 0.79, respectively, compared
with 0.75 and 0.87 for a short channel. Thus, the conclusion is
valid irrespective of the inlet boundary condition, but the relative
magnitude changes.
In closing, it is important to re-emphasize that the flow and
pressure drop characteristics presented in this article are for a
specific channel geometry picked for this study. The geometric
features, such as channel lengths Lentry / Dh, Lcurv / Dh, and
Lexit / Dh, curvature ratio CR, aspect ratio AR, variable c/s area
ratio bmax / b, etc., all play important roles in governing these
characteristics. Therefore, a further parametric study illustrating
the role of these geometric features is much needed.
12
Conclusion
Variable cross-sectional area curved flow passages are formed

between two fin walls of curved fin heat sinks used in avionics
electronics cooling. This article presents a 3D computational
model of developing laminar air flow through such a flow passage
with a 90 deg bend. The c/s area at the curvature inlet and outlet
are the same. In between, the channel width first increases and
then decreases along the length of curvature, making it a
diverging-converging curved channel. The area expansion ratio is
1.5, while the channel aspect ratio, based on the constant channel
height and width at the flow inlet, is 2.3. Simulations are carried
out for two geometric configurations: a a curved channel with
long, straight, constant c/s area inlet and outlet sections entry and
exit lengths, identified as a long channel; and b a curved channel with no entry or exit length, termed as a short channel. The
computational flow visualization at various c/s areas and different
flow rates reveals the formation of secondary flow, the separation
and attachment of the axial and secondary flow from/to various
channel walls, and the resulting growth of multiple vortices. The
roles of the channel inlet flow velocity and channel geometry,
SEPTEMBER 2009, Vol. 131 / 091102-15
specifically the area expansion and contraction effect, on flow

pattern and pressure drop are investigated. Some of the specific
conclusions are as follows.
For a constant c/s area curved channel, flow in the curved

section consists of a pair of base vortices, and three sets of
Dean vortices, at the inner curvature concave wall, the
outer curvature convex wall, and at the corner of top/
bottom wall and the concave wall.
The flow pattern is significantly more complex in case of a
variable c/s area diverging-converging curved channel.
The secondary flow splits in two streams, creating a split
base vortex, in addition to the traditional base vortex. Similar to the constant c/s area curved channel, three sets of
Dean vortices are also observed here. The expanding geometry at the diverging section of the curved channel also leads
to axial flow separation both from the inner and outer curvature walls, and formation of local separation bubbles.
For both the long and short channels, only the base vortices
appear at low flow velocity. There exists a critical velocity,
or a critical Dean number in the dimensionless form, above
which additional vortices Dean vortices appear in the
channel. The critical Dean number is geometry dependent.
For example, for a long, variable c/s area curved channel the
value is 129, compared with 165 for a constant c/s area
curved channel of the same length.
Pressure drop for the variable c/s area curved channel is
always lower than that of its constant c/s area counterpart,
for both the long and short channels. The pressure drop
characteristics in a short channel is quite counterintuitive. In
spite of the presence of secondary flow in a variable c/s area
curved channel, its pressure drop is lower than even that of
a straight channel.
Acknowledgment
We gratefully acknowledge the support from Rockwell Collins
and the technical interaction with Mr. Matthew Wells and Mr.
Perry Baker. We also thank Mr. Travis Sidebotham and Mr. Benjamine Liu for their help in preparing the sketches and the data
analysis.
Nomenclature
Acs cross-sectional area of the flow channel at any
location variable in the curved section
AR channel aspect ratio= H / b
b channel width; variable, maximum at the 45
deg angular position from the start of the curvature, and for all calculations b value at the
channel inlet is used
CR channel curvature ratio= R / b
De Dean number= Re Dh / R
Decrit critical Dean number for the onset of Dean
vortices= Vi,critDh / Dh / R
Dh channel hydraulic diameter at the flow inlet
= 2bH / b + H
f friction factor= 2P / V2i Dh / 4L
H channel height
ho overall channel heat transfer coefficient= total
heat dissipated/total channel wall surface
areaTw Ti
Hmin magnitude of minimum helicity at a c/s area.
k f thermal conductivity of air temperature
dependent
Lav total channel length along the middle of the
cross section= Lentry + Lcurv + Lexit
L+ dimensionless channel length= Lav / Dh Re
Lcurv average length of the curved section, along the
middle of the cross section
091102-16 / Vol. 131, SEPTEMBER 2009
Lentry length of the constant c/s area entrance section

in a long channel
Lexit length of the constant c/s area exit section in a
long channel
Lfd channel entry length for the flow to be fully
developed
Ncell total number of recirculation cells at any c/s
area
Nuo overall channel Nusselt number= hoDh / k f
P pressure
R radius of curvature of the inner curved convex wall
Re Reynolds number= ViDh /
T local temperature at any location
Tav,exit area averaged temperature at the exit of the
flow channel
Ti ambient temperature of the air entering the
channel
Tw temperature of the channel bottom and side
walls
V velocity magnitude at any location
Vi uniform flow velocity at the inlet of a channel
Vi,crit critical velocity at channel inlet for the onset
of Dean vortices
P pressure drop over the entire channel length L
dynamic viscosity of air
kinematic viscosity of air
dimensionless temperature at any location= T
Ti / Tw Ti
References
1 Garimella, S. V., 1996, Enhanced Air Cooling for Electronic Equipment, Air
Cooling Technology for Electronic Equipment, S. J. Kim and S. W. Lee, eds.,
CRC, Boca Raton, FL, Chap. 5, pp. 173202.
2 Carter, D. P., Crocker, M. T., Broili, B. M., Byquist, T. A., and Llapitan, D. J.,
2003, Electronics Assemblies With High Capacity Curved Fin Heat Sinks,
U.S. Patent No. 6,671,172.
3 McCormack, P. D., Welker, H., and Kelleher, M., 1970, Taylor Goertler Vortices and Their Effect on Heat Transfer, ASME J. Heat Transfer, 92, pp.
101112.
4 Dean, W. R., 1928, Fluid Motion in a Curved Channel, Proc. R. Soc. London, Ser. A, 121, pp. 402420.
5 Cheng, K. C., Nakayama, J., and Akiyama, M., 1977, Effect of Finite and
Infinite Aspect Ratios on Flow Patterns in Curved Rectangular Channels,
Proceedings of the International Symposium of Flow Visualization, Tokyo,
Japan, pp. 181186.
6 Sugiyama, S., Hayashi, T., and Yamazaki, K., 1983, Flow Characteristics in
the Curved Rectangular Channels Visualization of Secondary Flow, Bull.
JSME, 26216, pp. 961969.
7 Ligrani, P. M., and Niver, R. D., 1988, Flow Visualization of Dean Vortices in
a Curved Channel With 40 to 1 Aspect Ratio, Phys. Fluids, 3112, pp.
36053617.
8 Cheng, K. C., and Akiyama, M., 1970, Laminar Forced Convection Heat
Transfer in Curved Rectangular Channels, Int. J. Heat Mass Transfer, 13, pp.
471490.
9 Ghia, K. N., and Sohkey, J. S., 1977, Laminar Incompressible Viscous Flow
in Curved Ducts of Rectangular Cross-Section, ASME J. Fluids Eng., 99, pp.
640648.
10 Humphrey, J. A. C., Taylor, A. M. K., and Whitelaw, J. H., 1977, Laminar
Flow in a Square Duct of Strong Curvature, J. Fluid Mech., 833, pp. 509
527.
11 Hwang, G. J., and Chao, C.-H., 1991, Forced Laminar Convection in a
Curved Isothermal Square Duct, ASME J. Heat Transfer, 113, pp. 4855.
12 Chandratilleke, T. T., and Nursubyakto, 2003, Numerical Prediction of Secondary Flow and Convective Heat Transfer in Externally Heated Curved Rectangular Ducts, Int. J. Therm. Sci., 42, pp. 187198.
13 Beavers, G. S., Sparrow, E. M., and Magnuson, R. A., 1970, Experiments on
Hydrodynamically Developing Flow in Rectangular Ducts of Arbitrary Aspect
Ratio, Int. J. Heat Mass Transfer, 13, pp. 689702.
14 2006, FLUENT 6.3 Users Guide, Lebanon, NH.
15 Curr, R. M., Sharma, D., and Tatchell, D. G., 1972, Numerical Prediction of
Some Three-Dimensional Boundary Layers in Ducts, Comput. Methods Appl.
Mech. Eng., 1, pp. 143158.
16 Ebadin, M. A., and Dong, Z. F., 1998, Forced Convection, Internal Flow in
Ducts, Handbook of Heat Transfer, 3rd ed., W. M. Roshenow, J. P. Hartnett,
and Y. I. Cho, eds., McGraw-Hill, New York, Chap. 5.
D. C. Langer
B. A. Fleck
e-mail: brian.fleck@ualberta.ca
D. J. Wilson
Department of Mechanical Engineering,
University of Alberta,
Edmonton, AB T6G 2G8, Canada
Measurements of a Wall Jet

Impinging Onto a Forward Facing
Step
This study examines a horizontal wall jet impinging onto a forward facing vertical step in
a cross-flow. Planar laser induced fluorescence (PLIF) experiments in a 68 40 mm2
water channel indicate how the wall-jet flow after impinging onto the step becomes a
vertical jet with an elliptical cross section. This study proposes predictive empirical
correlations for the aspect ratio and perimeter of the jets elliptical cross section based on
the step geometry and the inlet flow conditions. A numerical model is also presented,
which was produced from a commercial Reynolds averaged NavierStokes computational
fluid dynamics (CFD) code with the k- closure model. The experimental results were
well represented by correlations for the perimeter P and aspect ratio S using the parameters H (the step height), L (the distance from the jet represented as a point source to the
step), and R (the velocity ratio). The CFD simulation was able to predict the trends in the
perimeter (under different conditions), aspect ratio, and the shape of the concentration
profile, but overpredicted the jets perimeter by approximately 50%. The results of these
tests are required as input parameters when modeling jet trajectories.
DOI: 10.1115/1.3203201
Introduction
This study examines the flow field, which occurs when a wall
jet impinges normally onto a forward facing step and is then deflected into a cross-flow. This geometry has not been examined in
the literature and later may be found to have applications for other
mixing or dispersion studies. The preliminary motivation for
studying this geometry was given by Wilson 1, who examined
the dispersion of gases released from pipeline ruptures. Underground pipelines can carry high concentrations of dangerous fluids
such as sour gas or natural gas. Occasionally these pipes fail,
leading to the release of significant quantities of toxic or flammable gasses. The rupture of a buried pipe produces a crater,
where the gas is released at high velocities and impinges onto the
crater wall. The interactions between the jet and the crater lead to
significant spreading and momentum losses in the jet, which can
drastically change the rate of dilution of the released gases in the
surrounding air. These ruptures can produce significant ill-effects
to the environment and danger for the populous surrounding the
rupture area. The ability to model the concentration distributions
of the pollutant within the atmosphere allows for safe placement
of toxic gas pipelines and proper emergency response in the event
of a poisonous gas release 1. Through better understanding the
shape and size of the jet entering into the cross-flowing fluid,
better predictions of the jets behavior can be made. The results of
this study could be used with Wilsons model 1 and be incorporated into hazard assessment models to more accurately determine
the dispersion of hazardous pollutants after a pipeline rupture.
Planar laser induced fluorescence PLIF measurements were
used for both flow visualization and to measure the concentration
profiles within the jet 2. The vertically deflected jet above the
step was studied to determine the shape of the jet entering into the
cross-flow. In preliminary experiments it was found that the jet
exiting the step was elliptical in nature, and the aspect ratio was
dependent on the jet velocity Vjet, cross-flow velocity V, step
height H, and the distance from the step to the jet L. This study
JOURNAL OF FLUIDS ENGINEERING. Manuscript received September 9, 2008; final manuscript received June 29, 2009; published online August 18, 2009. Assoc. Editor:
Juergen Kompenhans.
aims to determine how these parameters influence the shape and

behavior of the jet as it is deflected into the cross-flow. Future
work will study the elliptical jet as it enters the cross-flow to
predict the jets trajectory based on the shape of the ellipse and
other parameters.
CFD simulations were also undertaken to determine the feasibility of using numerical methods to predict the behavior of the
flow. These simulations were modeled using ANSYS-CFX which
solves the Reynolds averaged NavierStokes RANS equations
with a standard k- turbulence model. The k- model has been
used by several researchers to simulate the flow of free jets 3
and impinging jets 4. It is also a model, which is computationally inexpensive and robust, allowing for a variety of parameters
to be tested. This combination of experimental and CFD results
provides insight into the study of this novel geometry under several different flow conditions.
Experimental Method
Experiments were performed in the recirculating water channel

facility at the University of Alberta 57. The facility consists of
a 5.24 m long rectangular channel with a cross section measuring
47 68 cm2 with a water depth of 40 cm. A centrifugal pump
with a variable outlet diameter was used to control the velocity of
the cross-flow, which was varied from 0.041 m/s to 0.081 m/s,
producing Reynolds numbers based on hydraulic diameter ranging from 2 104 to 4 104. A turbulent boundary layer shear flow
was developed in the channel using a grid composed of square
aluminum tubes, a sawtooth trip fence, and roughness elements
composed of a 1.27 mm diamond shaped steel mesh attached to
acrylic panels on the bottom of the water channel. The shear flow
used in this experiment was measured 5 and was found to produce conditions similar to the atmospheric boundary layer.
The experimental test section consisted of the jet tube and step
shown in Fig. 1 and was located 3 m from the channel inlet. The
jet consisted of a 1.11 m 126d long brass tube with an inner
diameter of 8.75 mm, which was made from the jet tube and
turbulence generator used by Johnston and Wilson 7 with a 1 m
extension tube attached to the end. This ensured that the vertical
inlet tube did not significantly effect the cross-flow, and that there
was fully developed turbulent flow within the pipe at the exit. The
SEPTEMBER 2009, Vol. 131 / 091103-1
Fig. 1 Schematic of the test section. The step height H, the

effective distance L, and the velocity ratio R = Vjet / V were
varied for the parametric study. The shape of the cross section
perpendicular to the direction of the jet is shown at three locations: the pipe exit a, the top of the step b, and downstream
of the step c. The round jet flattens at the step b and forms
a high aspect ratio ellipse which then rolls-up to become
more circular.
pipe was mounted flush with the grid on the bottom of the channel, producing a wall jet, which was centered midway across the
water channel. Based on the logarithmic fit to the velocity profile
performed by Hilderman and Wilson 5, the boundary layer
thickness of the cross-flow at the jet exit location was calculated
to be approximately 15 mm 2d. A more detailed explanation of
the cross-flow is provided by Hilderman and Wilson 5. Dyed
water for the jet was stored in a 75 l tank, which was maintained
at 2.1 bars, and was connected to the pipe using flexible tubing.
Water from the jet and channel were of equal temperature, preventing any buoyancy effects. The velocity of the jet was controlled using a needle valve and a rotameter giving average jet
velocities of 0.47 m/s, 0.78 m/s, and 1.10 m/s. This corresponded
to pipe Reynolds numbers based on diameter of 4.1 103, 7.0
103, and 9.6 103. The step was constructed from a 1.2 m long
sheet of black acrylic, which covered the entire width of the water
channel. The height of the step was adjusted by adding strips of
1.27 cm thick acrylic which ensured that the step remained of
uniform height and was set at 2.54 cm, 3.81 cm, and 5.08 cm.
The front surface was perpendicular to the floor of the water channel and parallel to the face of the jet tube outlet. The jet outlet was
located 5d, 9d, and 15d from the step 43.75 mm, 78.75 mm, and
131.25 mm.
Measurements were taken to determine the shape of the jet
leaving the step with illumination in planes parallel to the water
channel floor. The system was composed of a 4 W Coherent
Innova 70 argon-ion laser, steering optics, a Powell lens, and a
12-bit Cooke SensiCam CCD camera. The laser was run in single
line mode, producing a single beam at 488 nm, with a rated power
of 2.1 W. The beam was passed through a focusing lens, decreasing the thickness of the laser sheet at the center of the water
channel to approximately 1 mm. It was then steered by two mirrors into a 30 deg Powell lens 8. The laser sheet was used to
illuminate fluorescein sodium salt C20H10Na2O5, which was premixed in the jet tank at a concentration of 0.20 0.02 mg/ l. The
absorption peak for this dye is approximately 488 nm, and the
emission peak is at 515 nm 2. A number 12 Kodak wratten
gelatin filter was placed in front of the camera to effectively attenuate the diffracted and reflected laser light from the recorded
images. The filter was rated to have zero transmittance at 488 nm
and approximately 25% transmittance at 515 nm the wavelength
of peak fluorescence emission. The camera was oriented effectively perpendicular to the laser sheet and focused using a 75 mm
Cosmicar TV zoom lens. The resolution of the images was found
to be 0.25 mm/pixel. The experimental signal to noise ratio was
091103-2 / Vol. 131, SEPTEMBER 2009
Fig. 2 Schematic of the experimental apparatus. The laser

beam is steered into a cylindrical focusing lens through the use
of two focusing mirrors. The beam then travels into a Powell
lens forming a laser sheet. The camera is oriented perpendicularly to the laser sheet and located above the water channel. A
sheet of glass is used at the top of the water surface to remove
the distortions caused by waves within the channel.
approximately 50 in the areas of high concentration, and 15 in the

areas of lower concentrations. A schematic of the experimental
optics and measurement apparatus is shown in Fig. 2. More detailed information regarding the optical setup is given by Torres
9.
Calibration was required to remove the effects of the nonuniformity of the laser sheets light intensity, which were caused by
the spreading of the laser, the profile produced by the Powell lens,
the operating mode of the laser, and the heterogeneous CCD sensitivity. The nonuniformities in the light sheet and camera array
sensitivity were factored into the daily calibration using a method
similar to that of 5 but assuming a linear concentrationfluorescence relationship 2. For each inlet condition, 500 images
were acquired at a rate of 7.6 frames per second FPS with an
exposure time of 10 ms. These conditions produced measurement
times of 65.8 s. Background measurements were taken before and
after the jet measurements to determine the buildup of dye in the
water channel. In all cases this buildup was found to be less than
1% of the peak concentration. For each experiment, jet images
were processed and converted to an average concentration image.
The laser sheet was positioned parallel to the floor of the water
channel, with its center 2 mm above the top of the step. Images
were taken with the camera perpendicular to the laser sheet to
show the cross section of the jet. An example of a single image is
shown in Fig. 3. Images were processed to produce an average
image, as shown in Figs. 4, parts a and b. As can be seen in the
figures, the jet at this location resembles a high aspect ratio elliptic
jet. As such, the size and shape of the jet were characterized by
two measurements, the jets depth along the original jet axis and
the jets width perpendicular to the original jet axis. The ellipse
was defined as the contour line, which encompassed 70% of the
maximum measured concentration within the measurement plane.
This 70% contour was determined based on experimental results,
as it was the contour, which gave the best representation of the
shape of the vertical jet entering the cross-flow for all of the flow
conditions tested. It was found that for concentrations less than
70%, the size of the contour varied significantly due to the fluid
entrainment at the boundaries of the jet. At concentrations greater
than 70%, the shape of the contours was found to be dominated by
local maxima and did not have a repeatable shape. To minimize
the effects of noise in the measurement of the maximum concentration, the 30 pixels with the highest concentration were averaged
to determine the maximum. These points typically had a standard
(a)
(b)
Fig. 3 Single experimental image of the concentration profile within the jet.
A linear concentration scale is shown in a and a sawtooth scale used to
emphasize the structures apparent in the jet is shown in b. The cross-flow
moves from left to right, with the jet fluid moving out of the page. The step
is located at 0 mm.
deviation of less than 1% of their mean. Once the maximum value

was determined, a Gaussian fit was used along each row and
column, with a threshold value of 60% of the peak. The two
intersections of each Gaussian curve with the 70% concentration
value were used to determine the depth and width of each row and
column. An example of the Gaussian fits used for jet widths and
jet depths is shown in Fig. 5. This fit produced a series of depth
and width measurements, from which the maximum width and
depth were defined as the major and minor axis of the elliptical
jet. The jet aspect ratio was then found by dividing the maximum
width a by the maximum depth b, and the perimeter was approximated by Eq. 1. The jet aspect ratio is defined in Eq. 2
P = 2a2 + b2
S=
a
b
a b2
2
1/2
1
2
CFD Constraints
3.1 Geometry. The geometry used for the simulations was

identical to that used in the water channel experiments. The crossflow was allowed to develop for 1 m before and after the step to
minimize the effects of the inlet and outlet conditions. A symmetry argument along the center of the channel was used to decrease
the computational effort and time required for each test. This argument is justified because the wall jet, impinging jet, and jet in
cross-flow possess symmetry along the jet centerline. The depth of
the channel in the model was identical to that of the experimental
system to help mimic both the cross-flow boundary layer and the
penetration of the jet into the cross-flow. A dimensioned diagram
of the model is shown in Fig. 6a. The two dimensions, which are
given by letters L and H, were varied in different simulations.
The length, L, was set at 5d, 10d, and 15d 43.75 mm, 87.5 mm,
and 131.25 mm, and the height, H, was set at 25.4 mm, 38.1 mm,
and 50.8 mm. These lengths were chosen to compare the CFD
results with the experimental results. Lengths of 5 diameters and
15 diameters were chosen to be compared directly with the experimental measurements at the same locations. The intermediate
length was chosen to be ten diameters for the CFD results to
provide a central value for curve fitting. The experimental length
of nine diameters differs from the ten diameters used for the CFD
only because of experimental limitations.
3.1.1 Boundary Conditions. The locations of the boundary
conditions are shown in Fig. 6b. The inlet condition for the
cross-flow had three initial normal speeds: 0.041 m/s, 0.061 m/s,
and 0.081 m/s, all with an initial turbulence intensity of 1%. The
inlet was defined as a subsonic flow with a scalar concentration of
0 mg/l. The jet inlet was also a subsonic flow, with an initial
concentration of 0.7 kg/ m3. The jet velocity was set at three different normal speeds: 0.47 m/s, 0.78 m/s, and 1.10 m/s, with a
turbulence intensity of 5%. An average static pressure condition
was used at the outlet, which was set at 0 Pa relative to the flow.
The bottom of the channel, the step, and the side wall were all set
as smooth walls with the no-slip condition. The top of the channel
was set as a wall with free slip to best simulate the free surface at
the top of the channel.
3.1.2 Meshing. Three computational grids were used to determine the effect of grid dependence: the grids were designated as
coarse, medium, and fine. The grid used for all of the simulations
except the grid convergence study is shown in Fig. 6c. Due to
computer limitations, the coarse grid was used for the parametric
study, and only one simulation was run using each the medium
and fine grids. Since the purpose of the CFD study was to determine the feasibility of using computer simulations to predict the
perimeter of the jet, the usage of the coarse grid was justified. It
was further justified by the parametric study that was analyzed,
which required several different simulations, limiting the available
computer time. An analysis of the effects of using the coarse grid
on the accuracy due to grid dependence is discussed in Sec. 3.1.4.
The three grids used had spacings of 2.5 cm, 2 cm, and 1.5 cm
between nodes in the free stream region. Along the floor of the
channel and on the step, inflation was used to increase the number
of elements near to the walls. In all cases the inflated layer had a
SEPTEMBER 2009, Vol. 131 / 091103-3
(a)
(b)
(c)
(d)
Fig. 4 Experimental a and b and CFD c and d concentration profiles. Linear and sawtooth concentration scales are shown so flow structures can be compared. The cross-flow moves from left to right with the jet
fluid moving out of the page. The step is located at 0 mm.
maximum thickness of 10 cm, and the expansion factors used

were 1.1 for the coarse and medium grids and 1.08 for the fine
grid. There were 15 inflated layers in the coarse grid, 20 in the
medium, and 30 in the fine grid. Each grid also had a region of
refinement around the step because this was the region with the
largest flow gradient and also the area of greatest interest. Two
cylindrical regions located along the two corners of the step with
a radius of 8 cm were used for grid refinement. In these regions
the spacing between nodes was 0.8 cm, 0.6 cm, and 0.5 cm, respectively, for the coarse, medium, and fine grids. The expansion
factor in these regions was 1.1 for the coarse and medium grids
and 1.07 for the fine grid. With these characteristics, the number
of nodes for each grid were 99,646, 198,341, and 397,942. This
lead to a grid refinement factor, r, of 2.0.
3.1.3 Solver Parameters. The software used a coupled multigrid solver to solve the linear system of discretized equations.
091103-4 / Vol. 131, SEPTEMBER 2009
The advection scheme selected in the CFX software was the high
resolution advection scheme. For this, CFX defines a blending
factor, which will be referred to as but is in the CFX manual,
which represents the order of the numerical scheme, with = 0
representing an upwind differencing scheme and = 1 representing a second order accurate scheme. The high resolution advection
scheme set as close to 1 as possible without introducing oscillations into the flow. This resulted in being variable throughout
the flow with a solution, which was greater than first order accurate throughout the domain though not fully second order. Values
of for a line along the jet centerline for the vertical velocity
component is shown as an example in Fig. 7.
The standard k- model with a scalable wall function was used
for the turbulence within the flow. The coefficients for this model
were C = 0.09, C1 = 1.44, C2 = 1.92, k = 1.0, and = 1.3. The
scalable wall function 10 limits the value of y to be greater
(a)
(b)
Fig. 5 Measured concentrations along the maximum depth line a and the maximum width line b. The
horizontal line shows the 70% value where the depth and width are measured. A threshold value of 60% of the
maximum concentration was used to determine the Gaussian fit.
than 11.06 to avoid placing mesh points within the viscous sublayer which produces the inconsistencies outlined by Grotjans
and Menter 11. Although the k- method has been shown to
overpredict turbulent spreading in round jets and can fail profoundly for three-dimensional flows 12, it provided fast convergence and was sufficiently robust for all of the flow conditions.
Other methods, such as the shear stress transport model and two
Reynolds stress models Baseline Wilcox Model and SpezialeSarkar-Gatski Model, were tested but produced oscillatory results
for the conditions used and would not converge.
A stationary solution was found for the simulation in order to
minimize its computational cost. Since the experiments were performed to determine averaged concentration intensities, a steady
state approximation was all that was required. The timescale and
length scale used in the solution used the default settings: automatic and conservative, respectively. The required convergence
level was set to an average residual of 106. The effect of the
residual on the jet perimeter is shown in Table 1. This shows that
the choice of residual causes less than a 1% variation in the perimeter. The passive scalar transport equation was used to determine the movement of the scalar throughout the flow. It was assumed the concentration variable was a volumetric scalar. The
remainder of the tuning parameters present in the software remained at their default settings.
(a)
(b)
(c)
Fig. 6 Physical definitions in the CFD code. Dimensions were
chosen to match the water channel facility used for the experiments a. The type and locations of the boundary conditions
are given in b. The coarse mesh, which was used for the
simulations with the exception of the grid refinement study, is
given viewed along the symmetry plane in c.
Fig. 7 Variation in , the blending factor used in the CFD

code, where = 0 represents a first order method and = 1 represents a second order method. Values are for the vertical velocity along the major axis of the elliptical cross section in the
measurement plane. The global order of the solution was found
to be 1.2, which corresponds to a value of approximately 0.2.
SEPTEMBER 2009, Vol. 131 / 091103-5
Table 1 Effect of the average maximum residual used for convergence on perimeter measurement. To save computational
time, a residual of 106 was used for the simulations.
Residual
Perimeter
mm
1 1005
1 1006
1 1007
1 1008
220.6
222.9
220.9
220.9
3.1.4 Grid Independence and Uncertainty. The uncertainty in

the results produced by this simulation was determined primarily
through the methods outlined by Roache 13. The validation
method defines a grid convergence factor GCI which is given in
Eq. 3. The GCI is meant to be used as both an estimate of the
error bars, which surround the simulation data and a method of
reporting grid convergence. The error is approximated for the
coarse grid using the fine grid as a reference 13
= FsE2
GCIcoarse
2
where Fs is a safety factor, for which a value of 3.0 is recommended for conservative estimates of the errors 13. E2 is defined
in Eq. 4
Ecoarse
=
2
r p
1 rp
where = f 2 f 1 gives the variation in the parameter of interest

perimeter for this case between the coarser grid 2 and the finer
grid 1. r is the refinement factor, which is the ratio between the
size of the control volumes between the coarse and fine case. p is
the formal order of accuracy of the algorithm, which is defined in
Eq. 5.
The order of accuracy of the simulation algorithm p can be
calculated through the use of a three grid refinement study, providing the grid refinement ratio is constant 14. The equation
defining the order 14 was used in the formulation of the GCI, so
Roaches 13 notation is used, shown in Eq. 5. In this study, the
parameter of interest is the perimeter of the vertical jet issuing
from the step, denoted by f in the general form 13
p = ln
f3 f2
/lnr
f2 f1
where f is the parameter of interest in the three grids being tested,

with the subscripts 1, 2, and 3 representing the finest, medium,
and coarsest grids studied.
For the study presented here, the order, p was found to be 1.2.
From this, the coarse GCI relating the medium grid to the coarse
grid was calculated for the jets perimeter and found to have a
value of 28 mm, and represents the approximate error bars for the
system due to the discretization error. If more computing resources were available, the fine grid could be used, which had a
GCI when compared with the medium grid of 12.2 mm. It was
found that the grids were all within the range asymptotic convergence, and the solution was found to be 1.2 orders accurate 13.
Truncation errors for low order systems like this simulation have
been shown by researchers 15 to cause significant effects on the
modeled flow behavior. The truncation errors from the scheme
used here 1.2 orders accurate cause an artificial viscosity to be
incorporated into the flow, leading to the overprediction of diffusion within the flow. Thus, it is expected that this model will
overpredict the spreading rate of the jet.
The third primary source of uncertainty in this study is caused
by the assumptions implicit in the k- model. The k- model,
though commonly used in flow studies, was designed primarily
for two dimensional shear flow problems. The model assumes that
091103-6 / Vol. 131, SEPTEMBER 2009
zero velocity gradient is accompanied by zero shear stress 16.

In cases with large streamline curvatures and large pressure gradients, such as impinging jets, the k- model can be inaccurate
12. Cusworth 3 found that the k- model significantly overpredicted on the order of 30% the turbulent kinetic energy and thus
the spreading rate of a free round jet. The model can also highly
overpredict the rate of heat or scalar transfer within flows 4.
From this, it can be assumed that the use of the k- model will
lead to the overprediction of diffusion and the spreading of the jet
in the study presented here. Modifications of the coefficient in the
eddy viscosity C term can minimize this overprediction 17 and
could be used later to improve the accuracy of this model.
It has been shown that there are large uncertainties in the solution method used for this study due to modeling, truncation, and
discretization. It should be expected that the CFD results significantly overpredict the spreading rate of the jet. This simulation is,
however, computationally inexpensive; predictions of the of the
trends in the jet perimeter for parametric studies could be very
valuable. Further improvements on the computational model
could be made to improve the predictions of the jet perimeter in
the future. However, the purpose of this study was to determine
the feasibility of using CFD to predict experimental data.
3.2 Perimeter Measurement Method. CFD results were
analyzed to determine the perimeter and aspect ratio of the ellipse
encircling the 70% of the maximum concentration contour lines of
the flow. This was measured by analyzing the contour lines along
a horizontal plane located at the top of the step z = 0 mm. Contours on this plane are shown in Fig. 4 parts c and d for a
standard case. From this, the 70% contour was isolated and measured to determine the width and depth of the contour using the
tools available in CFX. The perimeter and aspect ratio were then
calculated for the CFD results in the same manner as the experimental results.
Experimental Results
Experiments were done to determine the effects of three parameters: L, H, and R. The results of this parametric study are shown
in Fig. 8. The data plotted here show the average of all of the tests
taken at each flow condition typically two runs. The error bars
were determined by taking two times the standard deviation of
one condition where five tests were run. Figure 8a shows the
relation between perimeter and length is nearly linear for all of the
step heights. Figure 8b shows the effect of the step height, which
is also possesses linearity. Figure 8c shows the perimeter compared with the logarithm of the velocity ratio for a constant step
height H = 38.1 mm and a constant length L = 9d. It can be
seen in these figures that the data is dependent on L, H, and R.
These data were used to create the universal scaling law for this
geometry proposed in Sec. 5.
Discussion
5.1 Empirical Perimeter Correlation. An empirical correlation predicting the perimeter of the jet based on R, L, and H was
determined using the data presented above. In order to better fit
the model, the length L was altered to the distance from a point
momentum source to the step L, located five diameters behind
the jet outlet. This produced the empirical correlation given in Eq.
6, and shown in Fig. 9. From the fit, it can be seen that this
model predicts the perimeter within 13% error for all of the data
points. This correlation provides allows for the perimeter of the jet
entering the cross-flow to be predicted for any H, L, and R within
the range tested
P
HL
= 1.84R0.2
d
d2
1/2
5.79
Both Eq. 6 and Fig. 9 normalize the perimeter by the inlet jet
diameter. This scaling is not fully justified by the experiments, as
(a)
(b)
(c)
Fig. 8 Measured values of perimeter compared with the different inlet conditions, normalized with the jet diameter d. The effect of the step position L with different velocity ratios
and a constant step height is shown in a. The effect of the step height H at different
velocity ratios is shown in b. The effect of velocity ratio R for a single step height and
length is shown in c.
only one diameter was tested. The diameter is a convenient factor,

which is used frequently in the literature for nondimensionalizing
length units. Further experiments are required to fully justify this
scaling. Attempts at finding a different length scale to produce
nondimensional groups were unsuccessful, however, it may be

found that a turbulence or friction length scale could be used
instead of the jet diameter.
5.2 Experimental Aspect Ratio Correlation. To fully define
the shape of the ellipse representing the vertical jet entering the
cross-flow, the perimeter, and aspect ratio S are required. The
experimental correlation for S related to the inlet conditions is
given in Eq. 7 and shown in Fig. 10. In this correlation it is
apparent that the ratio of H and L is used instead of their product
as used in Eq. 6. As L increases, the wall jet grows vertically,
which upon impingement leads to increased depth in the ellipse. It
was found that the rate of growth of the depth is significantly
larger with L than H when compared with the growth of the
width. This leads to a decrease in S with increasing L
S = 8.38
Fig. 9 Measured perimeters compared with the experimental

fit outlined in Eq. 6. The error bars represent twice the standard deviation of the condition where five tests were taken.
H
L
0.3
R0.4
5.3 Comparison of CFD to Experimental Data. A comparison of the concentration profiles along the measurement plane is
shown in Fig. 4. Parts a and c show the concentration profile
on a linear scale, which appears to have similar shapes for both
the actual and modeled jets. Parts b and d used a sawtooth
concentration profile, which emphasized the structures in the flow.
From this, it can be seen that the simulation accurately predicted
SEPTEMBER 2009, Vol. 131 / 091103-7
Fig. 10 Experimental correlation for the jet aspect ratio S defined in Eq. 7 based on the step height H, effective distance
to the step L, and velocity ratio R
Fig. 12 Comparison of aspect ratio measurements S for CFD

results and experimental results in terms of the step height H,
effective distance to the step L, and velocity ratio R
the elliptical shape of the vertical jet, though with slightly larger
depth than the experimental case. The location of the counterrotating vortex pair downstream of the jet is also accurately predicted by the simulation. The comparison of the flow visualization
images shows that the CFD code can clearly predict mean characteristics of this flow.
The perimeters of the ellipses formed by the 70% contours for
all of the experiments and simulations are shown in Fig. 11. A
comparison of the aspect ratios is shown in Fig. 12. This figure
shows significant overlap between the CFD and experimental results. It can be seen that the perimeters predicted by the simulation
do not match those determined experimentally within error. As
can be seen in Fig. 11, the jet perimeter predicted through the
numerical model is significantly larger than the measured values.
According to Pope 12, a well-known deficiency of the
k-epsilon model is that it significantly overpredicts the rate of
spreading for the round jet. This increased spreading would lead
to an increase in the jets perimeter. Since the overprediction of
the spreading is equal in all directions, it has little effect on the
jets aspect ratio, as shown in Fig. 11. They do however predict a

similar trend and could be used to provide a reasonable prediction
for the perimeter based on the inlet conditions if the constant
overprediction caused by the artificial viscosity introduced into
the flow is taken into account. The error bars shown here are those
recommended by Roache 13 and equal to the GCI outlined in
Eq. 3. Despite the prediction of the trend, the error bars are
significant and it should be noted that a horizontal line with a
value of approximately P / d = 25 can be drawn through the error
bars.
Through comparing the CFD data to the experimental data, it
can be seen that the CFD model can give a reasonable estimate of
the actual value of the perimeter. The trend in the data is predicted
to within 30%, with a constant overprediction of about 50%. This
gives a reasonable order of magnitude prediction and can also be
used to perform further parametric studies on the effects of different flow conditions.
Conclusions
The findings of a study on a jet impinging onto a forward facing

step in a uniform cross-flow have been presented. It has been
shown that after the initial impingement, the jet leaves the step as
a high aspect ratio elliptical jet and enters into the cross-flow. The
perimeter and aspect ratio of the jet were found to be dependent
on H, L, and R and can be predicted with the relations given in
Eqs. 6 and 7. CFD measurements were found to accurately
predict the structures within the flow and provided a reasonable
estimate of the jet perimeter. The trends of the parametric study
predicted by the CFD model were found to follow those measured
experimentally.
This paper presents predictive correlations for the size and
shape of the jet issuing into the cross-flow after impingement. A
study is in progress, which uses these measurements to predict the
trajectory of the jet in cross-flow. Through that study and the use
of the experimentally measured predictive model developed here,
the trajectory of the mean plume can be predicted using only the
initial conditions of the flow.
Fig. 11 Comparison of CFD results to experimental results using the scaling relation given in Eq. 6. CFD results slightly
overpredict the perimeter due to the artificial viscosity introduced into the flow by the k- turbulence model and the low
order of the simulation.
091103-8 / Vol. 131, SEPTEMBER 2009
Acknowledgment
The work presented here was made possible by NSERC research grants to B.A.F. and D.J.W., and an NSERC graduate
scholarship to D.C.L.
Nomenclature
a
b
d
Fs
H
L
L
p
P
r
R
S
V
Vjet
z
blending factor in CFD code

width of elliptical cross section
depth of elliptical cross section
jet diameter, m
safety factor in GCI calculation
step height, m
distance from jet inlet to step, m
distance from point source to step, m
order of accuracy of CFD code
perimeter of vertical jet, mm
grid refinement ratio
jet to cross-flow velocity ratio
jet aspect ratio
cross-flow velocity, m/s
jet velocity, m/s
height above the step
References
1 Wilson, D. J., 1979, The Release and Dispersion of Gas From Pipeline Ruptures, Alberta Environment, Contract No. 790686.
2 Walker, D. A., 1987, A Fluorescence Technique for Measurement of Concentration in Mixing Liquids, J. Phys. E, 20, pp. 217224.
3 Cusworth, R. A., and Sislian, J. P., 1987, Computation of Turbulent Free
Isothermal Swirling Jets, University of Toronto Institute for Aerospace Studies, Technical Report No. 315.
4 Behnia, M., Parneix, S., and Durbin, P. A., 1998, Prediction of Heat Transfer
6
7
8
9
10
11
12
13
14
15
16
17
in an Axisymmetric Turbulent Jet Impinging on a Flat Plate, Int. J. Heat Mass

Transfer, 4112, pp. 18451855.
Hilderman, T. L., and Wilson, D. J., 2007, Predicting Plume Meandering and
Averaging Time Effects on Mean and Fluctuating Concentrations in Atmospheric Dispersion Simulated in a Water Channel, Boundary-Layer Meteorol.,
122, pp. 535575.
Shahzad, K., Fleck, B. A., and Wilson, D. J., 2007, Small Scale Modeling of
Vertical Surface Jets in Cross-Flow: Reynolds Number and Downwash Effects, ASME J. Fluids Eng., 129, pp. 311318.
Johnston, C. R., and Wilson, D. J., 1997, A Vortex Pair Model for Plume
Downwash Into Stack Wakes, Atmos. Environ., 311, pp. 1320.
Powell, I., 1987, Design of a Laser Beam Line Expander, Appl. Opt.,
2617, pp. 37053709.
Torres, L. A., 2008, Turbulent Round Jet in a Counter Flow: Effects of Small
Inlet Yaw Angle, MS thesis, University of Alberta, Edmonton, Alberta,
Canada.
Launder, B. E., and Spalding, D. B., 1974, Numerical Computation of Turbulent Flows, Comput. Methods Appl. Mech. Eng., 3, pp. 269289.
Grotjans, H., and Menter, F., 1998, Wall Functions for General Application
CFD Codes, Proceedings of the Fourth European Computational Fluid Dynamics Conference, Wiley, New York, Vol. 1, pp. 11121117.
Pope, S. B., 2000, Turbulent Flows, Cambridge University Press, New York.
Roache, P. J., 1997, Quantification of Uncertainty in Computational Fluid
Dynamics, Annu. Rev. Fluid Mech., 29, pp. 123160.
de Vahl Davis, G., 1983, Natural Convection of Air in a Square Cavity: A
Bench Mark Numerical Solution, Int. J. Numer. Methods Fluids, 3, pp. 249
264.
Fletcher, C. A. J., 1991, Computational Techniques for Fluid Dynamics, 2nd
ed., Springer-Verlag, Berlin, Vol. 1.
Andreasson, P., and Svensson, U., 1992, A Note on a Generalized EddyViscosity Hypothesis, ASME J. Fluids Eng., 114, pp. 463466.
Hanjalic, K., 1994, Advanced Turbulence Closure Models: A View of Current
Status and Future Prospects, Int. J. Heat Fluid Flow, 153, pp. 178203.
SEPTEMBER 2009, Vol. 131 / 091103-9
Large-Eddy Simulation of Wake

and Boundary Layer Interactions
Behind a Circular Cylinder
S. Sarkar1
e-mail: subra@iitk.ac.in
Sudipto Sarkar
Indian Institute of Technology Kanpur,
Kanpur-208016, India
Large-eddy simulations (LESs) of flow past a circular cylinder in the vicinity of a flat
plate have been carried out for three different gap-to-diameter G / D ratios of 0.25, 0.5,
and 1.0 (where G signifies the gap between the flat plate and the cylinder, and D signifies
the cylinder diameter) following the experiment of Price et al. (2002, Flow Visualization
Around a Circular Cylinder Near to a Plane Wall, J. Fluids Struct., 16, pp. 175191).
The flow visualization along with turbulent statistics are presented for a Reynolds number
of Re 1440 (based on D and the inlet free-stream velocity U). The three-dimensional
time-dependent, incompressible NavierStokes equations are solved using a symmetrypreserving finite-difference scheme of second-order spatial and temporal accuracy. The
immersed-boundary method is employed to impose the no-slip boundary condition at the
cylinder surface. An attempt is made to understand the physics of flow involving interactions of shear layers shed from the cylinder and the wall boundary layer. Present LES
reveals the shear layer instability and formation of small-scale eddies apart from their
mutual interactions with the boundary layer. It has been observed that G / D ratio has a
large influence on the modification of wake dynamics and evolution of the wall boundary
layer. For a low gap ratio, it is difficult to identify the boundary layer because of its
strong interactions with the shear layers; however, a rapid transition to turbulence of the
boundary layer, which is similar to bypass transition, is observed for a large gap ratio.
DOI: 10.1115/1.3176982
Keywords: wake-boundary layer interactions, shear layer transition, LES, IB method
Introduction
The vortex shedding and the aerodynamic forces of flow past a

circular cylinder near a plane wall have been the subject of many
studies for academic interests and for practical applications. This
kind of study can directly be related to many engineering problems, such as flow past a suspension bridge, pipelines near the
ground, flow through heat exchanger tubes near the wall, forced
air-cooling of board-mounted electronic components, etc. The
most attractive feature of flow over a circular cylinder, when the
Reynolds number Re is close to around 50, is the instability of
symmetric wake and the onset of a time-periodic alternate vortex
shedding two-dimensional 2D, known as the Krmn vortex
street. The formation of this vortex street is generally considered
to be the result of the KelvinHelmholtz instability with separated
shear layers. By further increasing the Re, a transition to threedimensional flow occurs when the Re is close to 180; however, the
vortex shedding phenomenon retains the dominant large-scale
structure at a higher Re in the turbulent wake 2,3. This alternate
shedding of vortices in the near wake leads to large fluctuating
pressure forces in a direction transverse to the flow that may cause
structural vibration, acoustic noise, or resonance.
The presence of a wall close to the cylinder changes the dynamics of vortex shedding as compared with the unbounded condition. The impermeability of the wall offers an irrotational constraint to the cylinder wake that cannot spread without limit
resulting in a finite mean force directed away from the wall. The
boundary layer presents a nonuniform velocity in front of the
cylinder, which causes asymmetry in the strength of vortex shed1
JOURNAL OF FLUIDS ENGINEERING. Manuscript received October 23, 2008; final manuscript received June 16, 2009; published online August 14, 2009. Assoc. Editor: Paul
Durbin.
ding. Furthermore, the vorticity shed from the cylinder interacts

with the vorticity of the boundary layer downstream resulting in
an excited wake-induced boundary layer with complex flow structures. If the gap between the cylinder and the wall is relatively
low, this interaction is strong, provoking the separation of the
boundary layer apart from a drastic change in the entire wake.
Thus, the dynamics of vortex shedding in presence of a wall are
influenced by the approaching boundary layer thickness, Reynolds
number, and gap between the cylinder and the wall. These factors
also control the evolution of boundary layer, its transition under
excitation of migrating wake. In brief, the major modifications of
flow dynamics past a cylinder in the vicinity of a wall can be
identified as 1 deflection of the boundary layer away from the
flat plate, 2 suppression of vortex shedding from the lower half
of the cylinder, 3 presence of separation bubble both upstream
and downstream of the cylinder, and 4 a significant change in
flow parameters, such as lift, drag coefficients, and Strouhal number.
One of the very early experiments in this area was performed
by Taneda 4. A circular cylinder close to a wall was towed
through the stagnant water at a low Reynolds number Re= 170,
where there was no effect of boundary layer. He observed a single
vortex row behind the cylinder for a small gap ratio G / D = 0.1
with re-establishment of a regular double row of vortices as the
gap ratio was increased to G / D = 0.6; the row of vortices became
unstable and broke down downstream. Several experiments were
conducted for flow around a cylinder in the presence of the
boundary layer at moderately high Reynolds number, from Re
= 2 104 to 105 58. In most of these cases, the approaching
boundary layer was turbulent. It was reported that the aerodynamic forces on the cylinder were modified with a slight variation
in shedding frequency. The suppression of vortex shedding was
observed when the body was closer than a critical distance from
the wall such that G / D ratio becomes 0.30.4 5,9. At smaller
SEPTEMBER 2009, Vol. 131 / 091201-1
distances, the wake is almost steady and the periodic shedding is

strongly inhibited with separation bubbles on the wall. Later, Lei
et al. 8 reported that the suppression of vortex shedding occurs
at a gap ratio of about 0.20.3; the value depends on the thickness
of the turbulent boundary layer . This critical gap ratio decreases as the thickness of the boundary layer increases and it
appears relatively independent of Reynolds number.
When the cylinder is placed close to a wall, the frontal stagnation point moves toward the wall producing an upward lift CL on
the cylinder with a consequence of deviation of the wake away
from the wall. This increases the base pressure resulting in a low
mean value of coefficient of drag, CD, although its value depends
on the boundary layer, its character and thickness. Furthermore,
the rms values of both CL and CD decrease as the wall is approached. This may be attributed to the diffusion of velocity fluctuations because of the boundary layer 7,8. Angrilli et al. 10
conducted the experiment of cylinder wake and boundary layer
interactions at Re= 2.8 103 to 7.6 103 and found an influence
of G / D ratio on vortex shedding frequency if the gap ratio is less
than 5.0, whereas others observed that the shedding frequency
was almost independent of gap ratio 58,11. A possible reason
of this difference is the different range of Re used in the previous
experiments.
The flow around a circular cylinder near a wall was demonstrated in Ref. 1 with the help of particle image velocimetry
PIV and hot-film anemometry for Re in the range of 12004960
and G / D ratio of 02. It was concluded that G / D was a major
factor for the change in flow characteristics. The flow interactions
between the cylinder wake and the boundary layer were divided in
four regimes based on G / D ratio. For G / D 0.125, the flow beneath the cylinder was suppressed, and separation of boundary
layer occurred both upstream and downstream of the cylinder.
Although there was no regular vortex shedding, a periodicity associated with the outer shear layer was found. For 0.125 G / D
0.5, the separation region decreased, but the flow characteristics
were the same as that of the previous regime. A strong coupling
between the inner-shear layer shed from the cylinder and the wall
boundary layer was observed because of opposite sign of vorticity.
For 0.5 G / D 0.75, the vortex shedding was first observed and
for G / D 1.0, there was no separation of the wall boundary layer
either upstream or downstream of the cylinder. The paper presented the flow structures and modification of wake dynamics in
low subcritical range of Reynolds number, where the variation in
Strouhal frequency St with gap ratio was mainly dependent on
Re. For Re 2600, the Strouhal number for G / D 2 was significantly greater than that of an isolated cylinder. But as Re increased Re 4000, St became insensitive to the G / D ratio.
However, the aerodynamic forces, the mean velocity field and
turbulence characteristics were not discussed.
As compared with the experiments, there are a few numerical
simulations of flow past a circular cylinder in the vicinity of flat
plate. Vada et al. 12 were the first to perform a 2D simulation
using vortex in cell method at a transcritical Reynolds number,
Re= 3.6 106 and G / D = 0.4, 0.8, and 1.5. The simulation suffered
from overprediction of drag coefficient and underprediction of lift
coefficient as compared with the experiment 11. Liou et al. 13
performed a large-eddy simulation LES of the turbulent wake
behind a square cylinder for a Re= 2.2 104, and the discussion
was focused on the phase-averaged vortex dynamics. The celerity
of the positive vortex shed from the lower side of the cylinder was
smaller than that of the upper side shed vortex due to the interaction with the boundary layer. Recently in Ref. 14, a 2D simulation of cylinder-boundary layer interaction with - formulation
was presented, and the flow features were explained by vortex
dynamics comparing with the experiment 1 for Re= 1200 and
G / D = 0.5 and 1.5. The computation illustrated larger vortical
structures as compared with the experiment with rollup of vortices
closer to the cylinder. This was attributed to the fact that the
091201-2 / Vol. 131, SEPTEMBER 2009
absence of spanwise direction in 2D simulation failed to redistribute the energy in the third direction developing unrealistic large
structures 15.
There are number of experiments illustrating flow around a circular cylinder in the vicinity of a wall, but the discussions were
largely focused on vortex shedding process and modification of
aerodynamic forces on the cylinder. Most of the numerical simulations made so far on this flow configuration are in twodimensional framework and elucidate the flow characteristics with
limited success particularly for high Reynolds numbers. Thus, the
objective of the present paper is to characterize the vortex shedding, either qualitatively or quantitatively, illustrating the flow
physics in terms of vortex stretching, breakdown and turbulence
generation. Here, we try to explain the mutual interactions, i.e.,
modification of wake dynamics due to the presence of the boundary layer and the excitation of boundary layer because of the
wake. It should be noted that the eddy motions and their interactions are expected to be resolved to understand the dynamical
process in turbulent flows for the concerned problem. A direct
numerical simulation DNS would have been ideal for the problem, but it is extremely costly and Reynolds-averaged Navier
Stokes calculations will not provide the answer. Hence, LES can
be pursued as an alternative means with the benefit being a considerably lower computational effort. In the present study, a 3-D
LES with a dynamic subgrid-scale model has been used to investigate the flow around a cylinder in proximity to a wall boundary
layer at Re= 1440 and for G / D of 0.25, 0.5 and 1.0. The Reynolds
number considered here is in the range of shear layer transition
3.
Numerical Methods
2.1 Governing Equations. In the present study, we perform

LES of incompressible flow. The filtered mass and momentum
equations can be expressed as,
u j
=0
xj
1 P 1 2
ui
ij
jui =
ui
u
+
+ fi
+
t xj
xi Re
xj
where, ui denotes the filtered velocity field and ij = uiu j uiu j is

the residual stress tensor also known as subgrid-scale stress,
SGS. The presence of the body forces f i is due to the immersedboundary IB method, which will be discussed later. The above
equations have been made dimensionless using the inlet freestream velocity U and the cylinder diameter D. The resulting
Reynolds number is defined as Re= UD / , being the kinematic
viscosity. The model proposed by Germano et al. 16 and modified by Lilly 17 is used here to include the effect of subgrid
motions in the resolved LES, where the model coefficient is dynamically calculated instead of input a priori. The momentum
advancement is explicit using the second-order Adams-Bashforth
scheme, except for the pressure term, which is solved by a standard projection method 18. The pressure equation is discrete
Fourier transformed in the spanwise direction an infinite cylinder
is assumed neglecting end effects and hence the flow can be considered homogeneous allowing periodicity of flow to be imposed
and is solved by the BI-CG algorithm 19 in the other two directions. The spatial discretization is second-order accurate on a staggered mesh arrangement using a symmetry-preserving centraldifference scheme, which is widely used in LES owing to its
nondissipative and conservative property 20,21.
Here, the simulations of flow past a cylinder in proximity to a
wall have been performed in the Cartesian grid considering the
origin of axes lies on the flat plate and at a distance of 10D from
the leading edge. The coordinates x , y , z denote the streamwise,
wall-normal and spanwise directions respectively and the correTransactions of the ASME
v = 0, u / y = w / y = 0
Computational domain
Blasius
profile
v=w=0
/ t + U / x = 0
Flat plate
5D
u=v=w=0
Boundary
layer
5D
X
25 D
Fig. 1 Computational configuration and boundary conditions
sponding velocities are denoted by u, v and w. The computational

domain and boundary conditions used here are illustrated in Fig.
1. The domain extends from 5D at the inflow to 25D at the
outflow with a spanwise length of 3D. It should be noted that the
same span was used in Refs. 20,22,23, while simulating the flow
around a circular cylinder in unbounded condition for the shear
layer transition region. It was also reported that doubling the domain in the spanwise direction did not affect the results significantly 23. In the wall-normal direction, the domain is extended
up to 9D for G / D = 1.0 and 8.5D for G / D = 0.25, 0.5.
A Blasius profile is imposed at the inlet considering the computational domain begins at 5D distance downstream from the
leading edge of the flat plate, where the boundary layer thickness
is considered as 0.295D calculated from boundary layer theory.
A nonreflecting boundary condition 24 is applied at the exit. The
no-slip boundary condition u = v = w = 0 is applied on the flat
plate and on the cylinder surface. A free-slip velocity v = 0,
u / y = w / y = 0 is imposed along the upper surface of the domain, which does not allow outflow through the top surface. This
causes the flow to accelerate above the boundary layer in order to
compensate for the growth of boundary layer. In our simulations,
the maximum value of the acceleration parameter K / U2
dU / dx due to this effect was 1 106 and 5 107 for
G / D = 0.25 and 1.0, respectively. The value of K at which a turbulent boundary layer is expected to relaminarize is around 3
106 25. Thus in the present case, the values of K assure that
the acceleration of the free-stream would not alter the flow physics of wake and boundary layer interactions. In the spanwise direction, the periodic boundary condition is used.
The IB method helps to impose boundary conditions on a given
surface not coinciding with the computational grid as in the
present case. Thus, to satisfy the no-slip boundary conditions at
the cylinder surface, the immersed-boundary technique is employed following Ref. 26, which belongs to the class of direct
forcing method. The velocity field near the boundary of the body
is modified at each step in such a way that the no-slip boundary
condition is satisfied on the surface. This is done using some
interpolations, which is equivalent to include a body force f i in the
momentum equation. In this paper, a quadratic unidirectional interpolation by Muldoon and Acharya 27 is used. The flow solver
has been extensively validated for a variety of transitional and
turbulent flows 2830, including simulation of vortex dynamics
behind a circular cylinder for a wide range of Reynolds numbers
31,32.
2.2 Computational Details. In the streamwise direction, a
uniform and refined mesh is used near the cylinder to get adequate
immersed-boundary points on the cylinder surface. In the flownormal direction, a refined and uniform mesh is employed near
the wall up to y / D = 2 for G / D = 0.25, 0.5 and y / D = 2.5 for
G / D = 1.0 to resolve the wake behind the cylinder and its interaction with the boundary layer. Away from the cylinder, the mesh
is slowly stretched out. A uniform mesh is used in the spanwise
direction owing to the symmetry of the body.
In order to ensure the accuracy of calculations, a grid resolution
test has been carried out for a gap ratio of G / D = 0.25. The grid
requirements for the simulation of fully turbulent boundary layer

using a second-order accurate LES solver are well known x+
50, y +min 1.0 and z+ 20. However, the grid requirements
to resolve the transitional flow or the boundary layer excited by
external disturbances are less well established. Jacobs and Durbin
33 indicated that the prediction of boundary layer transition induced by free-stream turbulence fst is sensitive to streamwise
resolution. Furthermore, Ovchinnikov et al. 34 illustrated that an
under-resolved calculation predicts a premature and abrupt transition. In the case of bluff-bodies, two types of interactions of
wakes with the boundary layer can be identified 35: i a strong
interaction, where the wake retains coherent eddies in the region
of interaction, and ii a weak interaction, which takes place sufficiently far downstream of the cylinder when the spanwise rolls
have decayed. The instant study falls in the first category, where
the interactions occur on relatively larger scales, and it is anticipated that the grid requirements would be less restrictive as compared with the grid employed in the boundary layer transition by
fst 33. In the present computation, four levels of mesh, such as
288 192 32, 384 160 32, 384 192 32, and 384 192
64 designated as grid 1, grid 2, grid 3, and grid 4 in the
streamwise, wall-normal, and spanwise directions, respectively,
are used. For grid 1, among 288 streamwise grid points, 32 points
are distributed upstream of the cylinder 5D to D, 96 points
surrounding the cylinder D to +D, 108 points near wake region
+D to +10D and rest are distributed in the far field downstream
+10D to +25D, where as for other three mesh levels 384
points, the distribution is 40, 96, 160, and 88 in the respective
regions. In the wall-normal direction, a total of 128 grid points
with equal spacing are used between 0 and 2D, and the remaining
64 points are slowly stretched away from the cylinder +2D to
+8.5D for grid levels 1, 3, and 4. For grid 2, the distribution is 96
and 64 in the respective regions. A total of 146 IB-points for
u-velocity along the cylinder surface are obtained for grid level 3,
as compared with 140 surface points used in Ref. 20 for a LES
of flow past a circular cylinder at Re= 3900. It should be noted
that the boundary layer thickness assuming a laminar flow at the
location of cylinder without its presence is 0.42D, and approximately 27 grid points are employed within the boundary layer for
grid 3.
The mean streamwise velocity u / U and turbulent kinetic energy TKE profiles at x / D = 2, 3, 5, 7, 10, and 15 are depicted in
Fig. 2 for all four grid levels. The data are normalized by the
free-stream velocity and the cylinder diameter. A maximum
change of 1% in the mean velocity profiles is observed between
grid 1 and grid 4 except at x / D = 3 Fig. 2a. However, appreciable changes in TKE profiles are seen between grid levels 1 and
4, Fig. 2b. This indicates that there is a need for a large number
of grid points in the near wake region and inside the boundary
layer. For the subsequent analysis, grid 3 is chosen, considering
the fact that there were no appreciable changes in both the mean
velocity and TKE profiles by further increasing the grid points. In
the near wake x / D = 3 to 15 region, nondimensional mesh spacing for grid 3 varies as x+ = 0.15 15, y + = 0.07 1.5, and z+
= 0.1 7.5. For G / D = 0.5, the same computational domain and
grid distribution are used as considered in G / D = 0.25. The computational domain is extended by 0.5D in the wall-normal direction for G / D = 1, keeping the streamwise domain unchanged.
Here, a total of 160 grid points with equal spacing are used between 0 and 2.5D, and 64 grid points are distributed in the remaining +2.5D to +9D normal direction. Few parameters, such
as the domain, grid distributions, and the values of x+, y +, and
z+, averaged between x / D = 3 to 15 and are summarized in Table
1 for different cases.
The time step was about t = 1.125 103D / U in the nondimensional unit. This kept the Courant number below 0.2 for the
entire simulation, and the viscous stability number was much less.
The flow field was allowed to evolve for 15 vortex shedding
SEPTEMBER 2009, Vol. 131 / 091201-3
G/D = 0.25
4
x/D =
10
15
y/D 2
1
0
3
__
(a)
U /U
Fig. 3 Instantaneous spanwise vorticity z: a and b experiment at Re= 1900 1; c and d present LES at Re= 1440.
a and c G / D = 0.25 and b and d G / D = 0.5.
G/D = 0.25
4
x/D = 2
10
15
y/D 2
1
0
0.1
0.2
(b)
0.3
0.4
0.5
TKE
Fig. 2 Grid independent test for G = 0.25D and Re= 1440 con / U and b turbulent
sidering a mean streamline velocity U
kinetic energy profiles at different streamwise locations:

288 192 32 grid 1, 384 160 32 grid 2, 384
192 32 grid 3, and 384 192 64 grid 4
cycles to get a dynamically steady state solution. Then the data

were collected for statistics for a period of 15 cycles that were
found sufficient to ensure statistically converged results.
Before discussing in detail, a few results of flow past a cylinder

in proximity to a wall from the present LES are compared with the
corresponding experiment 1 for validation. Figure 3 shows the
isocontours of spanwise vorticity z illustrating the flow over
the cylinder for G / D = 0.25 and 0.5. The wake-boundary layer
interactions, formation of separation bubbles both upstream and
downstream of the cylinder, stretching of shear layers shed from
the cylinder, rollup, and then breaking down to smaller eddies are
well represented. The lower half of the shear layer is suppressed
by the boundary layer. The deflection of the boundary layer away
from the wall downstream of the cylinder creates a separation
bubble. The numerical visualization of instantaneous flow field
illustrates a close resemblance with the experimental observations
in terms of vortex dynamics and flow structures.
From the flow visualization, it is difficult to appreciate the vortex shedding particularly for a low gap ratio. However, there exists a strong periodicity component to the predicted velocity signal
in the cylinder wake; almost certainly, this is associated with the
shear layer shed from the outer surface of the cylinder. A spectral
analysis of streamwise velocity component at x / D = 3.3 and y / D
= 1.5 from the cylinder center for G / D = 0.25 is shown in Fig. 4
and is compared with the experimental data 1. This reveals that
the dominant wake frequency occurs at St= 0.4; however, there is
a secondary peak at St= 0.25. The comparison of the power spectra density with the experiment 1 demonstrates a good agree-
ment. As G / D ratio increases, the effect of the wall boundary

layer reduces and the two peaks that appeared in the power spectra density merge into one peak, the value of which approaches
0.2. In the present simulation, St becomes 0.29 and 0.25 for
G / D = 0.5 and 1.0, respectively, whereas the corresponding values
in the experiment 1 are 0.27 and 0.255. The validation illustrates
that the present LES can be used to describe the evolution of shear
layers shed from the cylinder under the influence of wall boundary
layer.
3.1 Instantaneous Flow Structure. To visualize the vortex
dynamics in proximity to a wall during the shedding cycle, the
snapshots of isocontours of spanwise vorticity z are presented
in Figs. 57 for different gap ratios. These figures are drawn between a time-period T of vortex shedding that has been calculated assuming the Strouhal frequency as 0.2 and following the
experiment of Price et al. 1. However, it should not be interpreted as St= 0.2 for all cases; indeed there is no such periodicity
of the wake at all for a low gap ratio, G / D = 0.25 at Re= 1440. The
Strouhal number as 0.2 represents an approximate average value
over a range of G / D. Instantaneous vorticity is span-averaged that
removes the small-scale fluctuations and tends to identify the coherent structures. A brief description of the vortex formation, its
downstream convection, interaction with the boundary layer and
breaking down to smaller eddies are presented in this section.
Figure 5 depicts the instantaneous span-averaged vorticity between t / T = 0.0 to 0.8 for a small gap ratio G / D = 0.25, which
corresponds to G / = 0.6 being the boundary layer thickness at
the location of cylinder without its presence. For t / T = 0.0, the
outer shear layer shed from the cylinder is seen to curl up forming
a vortex denoted by A over the inner-shear layer, which is deflected away from the wall. The inner-shear layer from the cylinder remains frozen for a long period of time, and there is no
apparent rollup. A strong pairing between the inner-shear layer
and the boundary layer is observed being of opposite signs that
suppresses the shedding of the inner-shear layer. This coupling is
also attributed to the lift-off of boundary layer forming a large
separation bubble downstream of the cylinder. The boundary layer
appears to rollup denoted by B beneath the inner-shear layer. For
the next few steps t / T = 0.2 0.6, the inner-shear layer is
stretched further, developing a Krmn vortex sheetlike structure
denoted by C: one end of it remains attached to the cylinder to
form a quasisteady attached wake, while the other end engulfs the
Table 1 Computational grid and box size

G/D
Dimension of Box Lx Ly Lz
0.25
30D 8.5D 3D
0.5
1.0
30D 8.5D 3D
30D 9D 3D
091201-4 / Vol. 131, SEPTEMBER 2009
Grid nx ny nz
288 192 32 grid
384 160 32 grid
384 192 32 grid
384 192 64 grid
384 192 32
384 224 32
1
2
3
4
x+
y +
z+
Comments
7.96
3.82
4.35
4.23
4.07
4.87
0.68
0.86
0.71
0.74
0.68
0.8
4.11
3.85
4.41
2.25
4.12
4.93
Test case
Test case
Useful
Test case
Useful
Useful
t/T = 0.0
y/D
1
0
3
10
10
t/T = 0.2
y/D
1
0
3
Separation
bubble
0
t/T = 0.4
y/D
1
Stretched inner shear layer

0
Fig. 4 Power spectra density of streamwise velocity at x / D

= 3.3, y / D = 1.5 for Re= 1440, and G / D = 0.25: present LES,
experiment 1
10
t/T = 0.6
y/D
1
0
3
t/T = 0.0
Inner shear
2
1
0
0
10
10
t/T = 0.8
y/D
y/D
0
Negative roll
0
t/T = 0.2
10
A+B
0
x/D
B
Fig. 6 Span averaged vorticity contours for G / D = 0.5. For details refer Fig. 5.
y/D
1
0
3
t/T = 0.4
Stretching of
inner vortex sheet (C)
10
10
10
y/D
1
0
3
2
Separation
t/T = 0.6
y/D
1
0
3
2
Negative vortex sheet

t/T = 0.8
y/D
1
0
10
x/D
Fig. 5 Span averaged vorticity contours for G / D = 0.25. A total
of 20 nondimensional contours are considered in between 5
and +5. The negative vorticity is represented by dotted line.
outer roll A, resulting in a cross over of trajectories of the shed

vortices. Thus, the roll of negative vortices from the upper shear
layer merges with the boundary layer vortices and convects downstream before breakdown to form the final wake sheet of the
same-sign vorticity. At t / T = 0.8, neither the vortex A nor the vortex B is identified downstream, but a negative vortex sheet appears that convects with a speed slower than the free-stream.
For G / D = 0.5 that corresponds to G / = 1.2, both the inner and
outer shear layers shed from the cylinder curl up in an alternating
fashion producing two rolls denoted by A and C, although lack
of symmetry is evident, as shown in Fig. 6, at t / T = 0. The lift-off
of wall boundary layer occurs with a tonguelike vortex sheet that
forms a roll denoted by B. However, this deflection of the boundary layer decreases that, in turn, reduces the size of the downstream separation bubble as compared with the earlier. The curling
of the shear layers also appears to form closer to the cylinder. As
the time progresses t / T = 0.2 0.6, a rapid stretching of the innershear layer is observed forming an elongated vortex sheet, and its
coupling with the wall boundary layer lead to form a large negative roll as before. The mutual interaction of the wall vortices with
the shed vortices creates a remarkable difference in the wake dynamics: The trajectories of shed vortices cross each other, and the
final vertical position is opposite with respect to the unbounded
case. Thus, the negative vortices shed from the upper side of the
cylinder occupy the lower position, while the positive vortices
shed from the inner-side of the cylinder occupy the upper position
in the sheet; a similar trend is reported by Zovato and Pedrizzetti
SEPTEMBER 2009, Vol. 131 / 091201-5
t/T = 0.0
Wall vortex
G/D = 0.25
Inner vortex
y/D
Outer vortex
1
y/D
0
3
10
t/T = 0.2
y/D
B
1
0
3
(a)
101
10
x/D
G/D = 0.25
10
t/T = 0.4
100
| z|max
y/D
1
Wall vortex
10-1
Inner vortex
0
3
t/T = 0.6
Interaction of vortex C
and boundary layer
10
10
Roll-up of
boundary layer
t/T = 0.8
y/D
1
0
10
x/D
Fig. 8 a Trajectory of vortex peak and b variation in z peak

for G / D = 0.25 and Re= 1440
10-2
(b)
y/D
0
Outer vortex
10
x/D
Fig. 7 Span averaged vorticity contours for G / D = 1.0. For details refer Fig. 5.
36. The celerity of positive roll and negative vortex sheet is

marginally different. This stretching process decays the circulation
of the wake vortex because of dissipation.
As the gap ratio increases to G / D = 1 that corresponds to G /
= 2.4 when the cylinder remains well outside the boundary layer,
Fig. 7 illustrates that the cylinder behaves more closely to an
isolated cylinder with an alternate vortex shedding. However, this
shedding of the Krmn vortex, particularly the inner-shear layer
and the associated positive vortex rolls denoted by C, influences
the wall boundary layer in a different fashion. The excitation of
the boundary layer downstream appears to be substantial by these
positive migrating rolls. In detail, downstream of the cylinder a
negative vortex from the wall boundary layer eventually elongates
denoted by B forming a vortex sheet that is seen to be engulfed
by the positive convective vortex roll t / T = 0.6. This creates an
unsteady bubble beneath the negative vortex sheet; the periodicity
of which appears to be same as that of the vortex shedding. The
rollup of boundary layer forming small-scale eddies are apparent
downstream of x / D = 5 marked in Fig. 7 because of the convective coherent vortices imparting momentum to the boundary layer.
The flow dynamics demonstrated in the present LES supports the
experimental observations 1.
Figures 8 and 9 illustrate the overall trajectory of vortex peaks
and the variation in vortex peak values along the streamwise direction for G / D = 0.25 and 1.0. The location of the vorticity peak
can be identified as the vortex center. Three vortices emanating
from the outer shear layer, the inner-shear layer, and the wall
boundary layer are traced to reveal the vortex motion. For G / D
091201-6 / Vol. 131, SEPTEMBER 2009
= 0.25, vortices from the inner-shear layer and the wall boundary
layer travel in a parallel trajectory owing to the strong pairing
between them Fig. 8a and move in the wall-normal direction
immediately behind the cylinder x / D 3.5. The outer shear
layer vortices move almost parallel to the wall. Near x / D = 4, a
strong mutual interaction between vortices is reflected by the
cross over of trajectories of the shed vortices emerging a single
negative vortex sheet that travels downstream and remains almost
parallel to the wall. It is difficult to identify whether this negative
vortex is produced by the upper shear layer or by the wall vortices
as explained before. Figure 8b indicates that the peak values of
all vortices decay in same pace with increasing streamwise coordinate; they decay rapidly in the vicinity of the cylinder and at a
slower rate downstream of x / D = 4. This is attributed to the break5
Wall vortex
Inner vortex
Outer vortex
G/D = 1.0
4
3
y/D
2
1
0
(a)
10
x/D
101
10
G/D = 1.0
Wall vortex
Inner vortex
Outer vortex
| z|
max
10-1
10-2
(b)
x/D
10
Fig. 9 a Trajectory of vortex peak and b variation in z peak

for G / D = 1.0 and Re= 1440
Upstream
bubble
Secondary
bubble
2.74 D
Primary
bubble
G/D=0.25
3
2
y/D
1
0
-3
-2
-1
1
1.8 D
10
11
10
11
10
11
G/D=0.5
3
2
y/D
1
0
-3
-2
-1
1.1 D
G/D=1
3
2
y/D
1
0
-3
-2
-1
x/D
Fig. 10 Mean streamlines for G / D = 0.25, 0.5, and 1.0
down of the larger-scale vortex to the random small-scale eddies

resulting in a decrease in the magnitudes of lower frequency and
the increase in higher frequency energy-content eddies. It should
be further noted that the inner-shear layer and the wall boundary
layer do not cancel each other behind the cylinder, although of
different signs of vorticity. This supports the experimental observation of Price et al. 1, while contradicts to the suggestion of
Grass et al. 6 and Taniguchi and Miyakoshi 7.
For G / D = 0.5, trends of the trajectory of the shed and the wall
vortices remain almost the same as before, but the lift-off of the
boundary layer is relatively less severe in this case figures are not
presented. Here, the decay rate of shed vortices is higher than
that of the wall vortices in the vicinity of the cylinder. As the
cylinder is moved farther away from the wall G / D = 1, the outerand inner-side shed vortices along with the wall vortex can be
separately identified far downstream of the cylinder, Fig. 9. The
trajectories of vortex peak illustrate that the vortices migrate parallel to each other. In this case, the effect of the boundary layer is
minimum and so both the outer- and inner-shear layers rollup to
produce almost identical vortices. Further, the outer and inner vortices have the same decay profile rapidly in the vicinity of the
cylinder and at a slower pace downstream of x / D = 4, which
means that the interaction between them is a mutual process. The
wall vortex appears to decay continuously, but at a slower rate
behind the cylinder.
3.2 Time-Averaged Flow Features. This section presents the
time-averaged results describing the mean effect of interactions of
wakes with the boundary layer behind a circular cylinder. The
mean streamlines depict the formation of separation bubbles both
at upstream and downstream of the cylinder for a low gap ratio,
G / D = 0.25, Fig. 10. An elongated separation region is observed
downstream, which is designated as the primary bubble rotating in
the clockwise direction. A secondary bubble rotating in the opposite direction is formed beneath the primary bubble. This large
separation region can be attributed to the strong coupling between

the inner-shear layer and the wall boundary layer, which deflects
the boundary layer away from the wall. Figure 10 also indicates
that when the cylinder is very close to the wall, the shear layers
remain frozen for a considerable time before they curl up to form
the Krmn rolls the center of the outer roll is marked. These
rolls also lost the symmetry because of suppression of the shear
layers by the boundary layer. The downstream separation bubble
rapidly reduces, and the upstream separation bubble disappears as
the gap ratio increases. Furthermore, for a relatively large gap
ratio G / D = 1.0, the inner and outer shear layer appear symmetrical and they roll close to the cylinder.
The isocontours of time-averaged streamwise um and wallnormal velocity vm along with the mean pressure coefficient
C p = P P / 0.5U2 are presented in Fig. 11. The contour with
zero value of um is drawn with the white line to separate the zones
of positive and negative velocity apart from to mark the stagnation
points. The front stagnation point moves toward the wall from the
centerline of the cylinder. Furthermore, the inner separation point
shifts downstream along the shoulder of the lower half of the
cylinder, while the outer separation point moves upstream along
the shoulder of the upper half as the wall is approached. These
movements of stagnation and separation points change the wake
size behind the cylinder and vary the base pressure: The pressure
distribution becomes quite asymmetrical for G / D = 0.25. The
maximum value of C p is attained at the front stagnation point,
whereas the lowest pressure occurs on the top and bottom of the
cylinder near the point of separation of the shear layers, and these
positions depend on the gap ratio. The center of the core of recirculation region coincides with the negative pressure zone downstream that shifts closer to the cylinder as gap ratio decreases. The
stagnation point determined from the present LES was approximately at s = 10.60, 9.00, and 4.90 for G / D = 0.25, 0.5, and
1.0, respectively. These values moderately agree with the experiSEPTEMBER 2009, Vol. 131 / 091201-7
G/D = 0.25
3
2
0.2
y/D
0.3
0.4
-0.1
0.1
y/D
-2
3
2
0.2
y/D
0.3
0.4
-0.1
y/D
y/D
1
0.6
-2
-2
0.3
2
-0.4
0.8
0.6
1
0.8
0.6
-0.6
-2
0.2
-0.2
0
-2
0.2
y/D
0.2
0.4
0.6
0.8
-0.6
-0.4
-0.8
4
0
-0.2
-0.4
0.4
0.6
-0.8
-0.4
G/D = 1.0
0.1
-0.4
0.1
y/D
-0.6
-1
-0.6
0
-0.2
-0.4
-0.6
G/D = 1.0
-0.8
G/D = 0.5
Cp
1
0
-0.6
-0.6
0
0.2
-0.2
-0.6
1
0.8
G/D = 0.5
(a)
0.6
0.1
-0.3
0
-0.4
1
0.1
G/D = 0.25
Cp
-2
1.2
x/D
(b)
x/D
Fig. 11 The mean a streamwise and wall-normal velocity contours and b coefficient of pressure
distributions for different G / D. Streamwise velocity contours are drawn by flood and wall-normal
velocity contours are drawn by line black color diagrams. The thick white line indicates zero contour level for mean streamwise velocity.
ment 8 even at Re= 1.45 104 and / D = 0.48, where s = 9.60,

6.80, and 3.30 for G / D = 0.3, 0.5, and 1.0, respectively. Here, s
is measured from the front point on the horizontal axis of the
cylinder and in the clockwise direction. The movement of stagnation point will be associated with the generation of lift; an effect
of the boundary layer that has already been reported in Refs.
8,36. The deflection of the wake and its departure from the symmetry due to the wall proximity can be understood by the isocontours of streamwise velocity. Negative um values upstream and
downstream of the cylinder depict the flow separations that diminish as the cylinder is moved away from the wall. Asymmetry in
the wake shape and lift-off of the boundary layer are well reflected
in vm values. As the gap ratio increases G / D = 1.0, distribution
of vm approaches a symmetric look, although the effect of the
boundary layer is evident.
The streamwise evolutions of mean skin friction C f = w /
0.5U2, where w = u / y y=0 being the wall shear stress for the
three gap ratios are shown in Fig. 12. The separation and reattachment points along with the length of bubbles both upstream and
downstream of the cylinder are summarized in Table 2. For
G / D = 0.25, C f becomes negative at x / D = 2.58 to 0.71 indicating an upstream separation, which is absent in the other two cases.
The drop followed by a sharp rise in C f upstream of x / D = 0 is due
to the combination of wall proximity and flow acceleration beneath the cylinder. The peak value of C f becomes maximum for
this gap because the flow beneath the cylinder is under the maximum negative pressure. Downstream of the cylinder, C f becomes
negative for all the cases indicating the separation region. The
appearance of a secondary bubble downstream is also reflected in
the streamwise evolution of C f for a low gap ratio. When compared with the skin friction values evaluated from the correlation
091201-8 / Vol. 131, SEPTEMBER 2009
of turbulent flow, Fig. 12 illustrates the relaxation of flow near the

wall. For G / D = 1, an overshoot of C f is observed after reattachment of flow, and it approaches the level of turbulent C f downstream. For other two cases, large regions of negative C f are seen
Fig. 12 Profiles of Cf for G / D = 0.25, 0.5, and 1.0. The solid

lines indicate Cf and the dotted lines are corresponding instantaneous Cf separated by 0.2T. Turbulent Cf profile is obtained
from the correlation Cf = 0.058/ Rex1/5.
Table 2 The onset of separation and reattachment

Downstream separation x / D
Upstream separation x / D
Primary bubble
Secondary bubble
G/D
Start
End
Length
Start
End
Length
Start
End
Length
0.25
0.5
1.0
2.58
-
0.71
-
1.87
-
1.69
1.06
1.34
10.06
4.57
3.04
8.37
3.51
2.7
2.89
1.95
-
3.64
2.52
-
0.75
0.57
-
that slowly attain the level of turbulent C f far downstream. To

highlight the near-wall flow, the skin friction distributions at six
instants within a shedding period are superimposed. Large amplitude oscillations with low and high frequencies are observed behind the cylinder illustrating the unsteady bubble dynamics and its
breakdown. The oscillations in instantaneous C f also start earlier
for a large gap ratio as compared with a low gap. This is attributed
to the delayed instability of shear layers for a low gap ratio that
will be explained in Sec. 3.3.
3.3 Development of 3D Motion and Breakdown to
Turbulence. The isosurface of the spanwise component of instantaneous vorticity for different gap ratios is depicted in Fig. 13 to
visualize the internal growth mechanism of shear layers in proximity to a wall and the 3D flow structures. For a low gap ratio
G / D = 0.25, the outer shear layer shed from the cylinder appears
to remain laminar up to x / D = 2.5 and then the shear layer instability sets in that quickly breaks down to turbulent flow down-
stream. As stated earlier, the boundary layer cannot be identified

separately because of its strong coupling with the inner-shear
layer and a large separation region is observed. The flow is turbulent away from the wall with the appearance of small-scale eddies
and loss of coherent structures leaving a relatively calm near-wall.
With increase of G / D ratio, the effect of wall proximity is
decreased, and the breakdown of both the outer- and the innershear layer occurs relatively closer to the cylinder. The shear layers tend to retain their coherent structures with appearance of
small-scale eddies. Further to note that the deflection of the
boundary layer is inhibited. For G / D = 1, the boundary layer preserves its identity with formation of a thin separation bubble. In
this case, the perturbations created by the turbulent shear layers
are amplified near the wall with appearance of longitudinal
streaky structures, which are the features of flow transition. A
sectional view of instantaneous velocity fluctuations u and v
that are obtained by subtracting the mean part from the instantaneous value along with streamwise velocity u contours in the
x-y plane is presented in Fig. 14a for a large gap ratio G / D
= 1. The flow is characterized by alternating Krmn vortices
behind the cylinder, and these have not been suppressed. From a
zoomed view it is clear that the Krmn vortices interact with the
outer part of the boundary layer resulting in local eddies, which
convect and breakdown downstream. Thus, the formation of these
eddies in the outer part of the boundary through a Kelvin
Helmholtz-like instability is the root to transition under the convective wake. A similar view of boundary layer transition under
(a)
3
y/D = 0.25
z/D
1
0
(b)
Fig. 13 Isosurfaces of spanwise vorticity z = 2.0 for G / D

= 0.25, 0.5, and 1.0
-2
10
x/D
Fig. 14 Streamwise velocity contour and vector plot of velocity fluctuations u , v for a front view x-y plane along with a
zoomed section, b top view x-z plane considering a gap ratio of G / D = 1.0. For better visualization, two vectors are
skipped in the x-direction and four vectors in the y-direction,
respectively.
SEPTEMBER 2009, Vol. 131 / 091201-9
x/D = 7
y/D = 0.18
R uu
y/D = 0.25
R uu R vv
,
1.5
y/D = 0.32
1
0.5
R vv
0
-0.5
0.3
0.6
0.9
1.2
1.5
(a)
z/D
Fig. 15 Spanwise correlation functions of the streamwise and
the wall-normal velocity fluctuations
passing wakes is described in Ref. 37. The top view along a

section near the wall y / D = 0.25 shows the development of lowspeed streaks because of boundary layer excitation by these migrating coherent vortices, Fig. 14b. These near-wall structures
are visible up to x / D = 8.5.
The average spacing between the streamwise streaks, which are
the features of transition under free-stream turbulence can be determined from the spanwise correlation functions of streamwise
Ruu and wall-normal velocity fluctuations Rvv. For a large gap
ratio G / D = 1, where the boundary layer retains its identity and
appears to approach a canonical layer, the variation in Ruu and Rvv
at a streamwise location of x / D = 7.0, and three wall-normal positions are shown in Fig. 15. Both Ruu and Rvv predict an average
streak spacing of 1.6D twice the distance to the first Ruu minimum, or four times the distance to the first Rvv minimum. In
terms of boundary layer thickness , the streak spacing becomes
1.33, being = 1.2D at the location. Here, is calculated from
the location of maximum mean velocity in between the plate and
the wake. The experiments and simulations on boundary layer
transition due to moderate to strong levels of fst 33,38 illustrate
streamwise streaks with mean separation of the order of boundary
layer thickness. Thus, the present study indicates that the transition mechanism of the boundary layer for a large gap ratio is
similar to the bypass transition by fst, although here the excitation
is caused by the Krmn vortices: the observation is consistent
with Ref. 34.
To characterize the boundary layer further under the excitation
of wake for G / D = 1, the profiles of urms are presented in Fig. 16,
where the wall-normal coordinate is scaled with the local displacement thickness and urms with its maximum inside the
/ U dy, where U is the maxiboundary layer. Here = 01 U
mum mean velocity in between the plate and the wake. The experimental data of Matsubara and Alfredsson 38 are superimposed, which were obtained from a transitional boundary layer
subjected to moderate free-stream turbulence. The computed profiles of urms exhibit an approximate self-similarity near the wall
and within x / D = 3.5 to 5.0. High values attributing to the wake
flow are observed in the outer region, Fig. 16a. The agreement
with the measured data 38 appears good illustrating that the
mode of transition and formation of streaks in this case are similar
to those of boundary layer perturbed by strong levels of fst. The
location of urms peak shifts toward the wall depicting the end of
transition downstream of x / D = 7.5, Fig. 16b. In the region of
x / D = 5 7, a large value of urms is noticed, which is attributed to
interactions of large-scale eddies of the shear layers with the
boundary layer.
For G / D = 0.5, some important features of shear layer transition
under the influence of boundary layer can be illustrated by instantaneous velocity fluctuations along with streamwise velocity u
091201-10 / Vol. 131, SEPTEMBER 2009
(b)
Fig. 16 Time-averaged nondimensional urms near the wall: a
between x / D = 3.5 5.0 and b after x / D = 7.5
contours in Fig. 17a: a sectional view along the x-y plane. The
wake flow seems to be turbulent downstream of x / D = 3.0 with
appearance of large- and small-scale structures. The coherent vortices formed due to mutual interactions of shear layers suffer from
(a)
x/D = 1.5
x/D = 5
y/D
x/D = 10
x/D = 15
y/D
(b)
z/D
z/D
Fig. 17 Streamwise velocity contour and vector plot of velocity fluctuations u , v for a front view x-y plane and b side
views y-z plane considering a gap ratio of G / D = 0.5. For details refer Fig. 14.
10 3
10 1
10 2
10
10
10 -1
10 0
Euu
____
-1
10
-2
10
-3
10
-4
Eww
____
2
U D
-2
10 -3
10
10
(-5/3)
2
U D
(-5/3)
10
G/D = 0.5
-4
G/D = 0.5
10 -5
x/D = 3.3
x/D = 3.3
10 -6
x/D = 5.0
x/D = 5.0
10
-5
10
-6
10
x/D = 10.0
-1
(a)
10
10
10
St
10
-7
10
-8
10
x/D = 10.0
-1
10
(b)
10
10
St
Fig. 18 The power spectra at Re= 1440 for G / D = 0.5: a streamwise velocity and b spanwise
velocity
3.4 Turbulence Statistics. The profiles of the streamwise

/ U and fluctuating u / U velocities
component of mean U
rms
at different sections x / D = 2, 3, 5, 7, 10, and 15 for the three gap

ratios are shown in Fig. 19. The evolutions of velocity profile
derived from the Blasius solution are also superimposed to illustrate the change in flow field due to the presence of a cylinder near
the wall. For a low gap ratio, G / D = 0.25 corresponding to G /
= 0.6, the velocity profiles indicate a fairly large wake-deficit behind the cylinder x / D = 2 and 3 apart from an accelerated flow
beneath it and a large downstream separation. This large wakedeficit is attributed to the frozen shear layers for a relatively
longer period when the cylinder is submerged within the boundary
layer. The presence of the velocity deficit is visible far downstream even up to x / D = 10 and is almost parallel to the wall. As
explained earlier, the strong coupling between the inner-shear
layer and the boundary layer in this case causes the deflection of
urms / U
(a) G/D = 0.25

4
0.2
0.4
0.6
0.8
1.2
y/D 2
1
0
3
__
U /U
0.8
1.2
urms / U
(b) G/D = 0.5

0
4
8
(a)
0.2
0.4
0.6
y/D 2
1
0
(b)
3
__
U /U
0.8
1.2
urms / U
(c) G/D = 1.0

0
4
8
0.2
0.4
0.6
y/D 2
(c)
3
__
U /U
the intrusion of the deflected boundary layer and hence loss of

symmetry. The velocity contours become negative near the wall
indicating an unsteady separation bubble that breaks down and
convects downstream. The cross-sectional flow at different y-z
planes is presented in Fig. 17b. At x / D = 1.5, both the shear
layers shed from the cylinder and the boundary layer can be separately identified. The shear layers at this section are laminar with
small fluctuations. At x / D = 5, the flow in the outer layer is perturbed with large spanwise vortices that are most possibly attributed to the shear layer instability. These large spanwise vortices,
which constantly eject fluid from the inner layer, are visible even
at x / D = 10. At x / D = 15, the flow field including the near-wall
region is dominant with fine-scale turbulent structures.
The power spectra considering the streamwise and spanwise
velocity fluctuations are depicted in Fig. 18a and 18b for
G / D = 0.5 and Re= 1440. The velocity histories have been collected in the midspan plane at locations of x / D = 3.3, 5, and 10,
y / D = 2.0 over a time-period of TU / D = 15 T being the timeperiod of vortex shedding. At x / D = 3.3, a peak in the power
spectra is observed at St= 0.25 for both Euu and Eww, which can be
considered as the vortex shedding frequency in proximity to a
wall. At x / D = 5 and 10, a wide range of shedding frequency with
more energetic high frequency harmonic are observed. This is
attributed to the breakdown of primary vortices to small-scale
energetic eddies downstream. The mesh used in the present LES
appears good to resolve the wake and boundary layer interactions.
The power spectra Eww ensure the development of spanwise fluctuations.
Fig. 19 Profiles of streamwise mean and rms velocity along

with the Blasius solution for G / D = 0.25, 0.5, and 1.0 at x / D = 2,
/ U , u / U , Blasius
3, 5, 7, 10, and 15; U
rms
solution
SEPTEMBER 2009, Vol. 131 / 091201-11
x/D
7.5
10
15
20
Kim et al. (1968)
Spalart (1988)
0.2
Production
(a)
0.04
0.08
0.12
0.16
0.2
Zone of
high k
3
y/D 2
3
2
G/D = 0.25
1
0
0.1
(a)
0.1
0.2
0.3
0.4
0.5
0.16
0.2
TKE
__
+ + +
<u/v/> dU / dy
0.3
Production
(b)
20
40
60
80
y+
Fig. 20 Time-averaged nondimensional turbulent kinetic energy productions near the wall and after x / D = 7.5
091201-12 / Vol. 131, SEPTEMBER 2009
0.04
0.08
0.12
Zone of
high k
G/D = 0.5
4
3
y/D 2
the boundary layer, and they convect in a parallel trajectory. When

the turbulent fluctuations are compared with the corresponding
velocity profiles, it illustrates that the turbulence generation occurs along the trajectory of the shear layer, which remains away
from the wall. Thus, the mutual interactions between the shed
vortices and the boundary layer originate turbulence. For G / D
= 0.5 or G / = 1.2, the mechanism of turbulence generation remains the same as before with considerably large wake deficit,
although lift-off of the boundary layer is less severe with appearance of a relatively smaller downstream separation.
For G / D = 1 that corresponds to G / = 2.4, the cylinder remains
well outside the boundary layer. Thus, the wake deficit and the
growth of boundary layer can be separately identified in velocity
profiles except in the near wake, where the shear layers govern the
turbulence. The peak values in velocity fluctuation shift toward
the wall downstream illustrating the interactions of large coherent
vortices in the wake with the boundary layer. When compared
with the Blasius solution, the velocity profiles become inflexional
between x / D = 2 and 3, and the boundary layer appears turbulent
downstream of x / D = 7. To characterize the state of boundary
layer after being excited by the coherent vortices in this case,
profiles of the time-averaged turbulent kinetic energy production
p+ = uvu+ / y + normalized by wall parameters at four
streamwise stations are compared with the DNS of Spalart 39
and the experimental data of Kim et al. 40 in Fig. 20. The study
of Spalart on the turbulent boundary layer revealed that profiles of
P+ at three different momentum thickness Reynolds numbers
Re = 300, 670, and 1410 are self-similar. P+ evaluated from the
resolved part of the present LES almost reproduces this feature as
evident in the self-similarity of the profiles downstream of x / D
= 7.5, although the data of Spalart and Kim et al. are under predicted by about 8%. It should be noted that the value of Re
calculated from LES is approximately 150 at x / D = 7.5, where the
boundary layer appears turbulent. The boundary layer becoming
turbulent at low range of Re has also been observed in Ref. 34:
characteristics of wake-boundary layer interactions.
To understand the relative effects of G / D ratio on the wake and
boundary layer interactions, the profiles of turbulent kinetic en1
ergy TKE = 2 uiui along with its production PTKE =
uiuj ui / x j at different locations downstream of the cylinder
are presented in Fig. 21. The zone of high TKE and production is
approximately marked in the figure. As expected, the enhancement of turbulence occurs along the trajectories of the shear layers
leaving a relatively calm near-wall region for low gap ratios G / D
up to 0.5. This is attributed to the strong coupling between the
approaching boundary layer and the inner-shear layer leading to a
rapid growth of disturbances with appearance of small-scale eddies in regions away from the wall: an example of mutual interactions between the boundary layer and the shear layer. The production occurs in regions of high turbulent stress and high spatial
velocity gradient aligned in the same direction. Thus, the high
(b)
0.1
0.2
0.3
0.4
0.5
0.16
0.2
TKE
Production
(c)
0.04
0.08
0.12
G/D = 1.0
Zone of
high k
y/D 2
1
0
1
0
0.1
(c)
0.2
0.3
0.4
0.5
TKE
Fig. 21 Profiles of turbulent kinetic energy and production for

G / D = 0.25, 0.5, and 1.0 at x / D = 2, 3, 5, 7, 10, and 15; TKE,
production
production along the path of the wake is due to the coexistence of

turbulent stress and shear in the wake fluid. For a large gap ratio
G / D = 1.0, the high TKE and production are not confined along
the path of the shear layers, rather the augmentation is migrated
toward the wall as marked. In this case, the inner-shear layer shed
from the cylinder perturbs the inflexional boundary layer during
its convection generating higher frequency energy-content eddies.
Hence, the high production near the surface can be considered as
the combined effect of a local concentration of vorticity and a
growing level of turbulence in the boundary layer under excitation
of the convective wake.
Conclusions
LES of flow past a circular cylinder in proximity to a wall has

been carried out to understand the insight of flow physics involving the wake and boundary layer interactions for different gap
ratios. The resolved flow field elucidates the mutual effects, i.e.,
modification of wake dynamics due to the presence of boundary
layer and excitation of the boundary layer because of the wake
flow. The Reynolds number considered here is in the range of
shear layer transition.
For a low gap ratio, when the lower shear layer lies within the
boundary layer G = 0.6, a strong pairing between the innershear layer and the boundary layer is observed. This delays the
shear layer transition from two-dimensional to three-dimensional
flow in the wake and deflects the boundary layer immediately
behind the cylinder with a relatively large separation. It also inhibits the creation of a vortex pair evolving a unique row of likesigned vortices after breakdown to small-scale eddies downstream
of x / D = 4. In this case, the dominant wake shedding frequency
shifts to 0.4. Furthermore, the inner-shear layer and the boundary
layer do not cancel each other, although of different signs of vorticity. This supports the experimental observation of Price et al.
1.
At a moderate gap ratio when G , the inner- and the outerlayer shed from the cylinder tends to curl up in an alternate fashion, although lack of symmetry is strongly evident. Here also the
boundary layer is stretched because of the attraction of the innershear layer evolving an unsteady bubble, the periodicity of which
appears to be same of vortex shedding. The mutual interactions of
the wall vortices and that with the shed vortices create a remarkable difference in the wake dynamics: trajectories of shed vortices
cross each other and an inversion on the position of vortices occurs with respect to the unbounded case. The presence of a wall
causes movements of stagnation and separation points changing
the wake size and base pressure and thus an upward lift on the
cylinder is generated.
When the cylinder is far apart from the wall G = 2.4, the
vortex shedding is similar to the Krmn sheet and the Strouhal
frequency becomes close to 0.2, but asymmetry in wake shape is
still evident. The Krmn vortices excite the boundary layer in an
alternate fashion that triggers its transition.
An attempt is made to describe the transition mechanism of the
shear layers and that of the boundary layer with respect to the gap
ratio. For a relatively low gap ratio G = 0.6 1.2, the shear
layer instability sets in with appearance of a spanwise wavelength
that becomes three-dimensional and breakdowns to turbulent flow
downstream. Here, the coherent vortices of the shear layers suffer
from the intrusion of the deflected boundary layer. The boundary
layer losses its identity downstream because of its strong coupling
with the shear layers. Thus, the generation of turbulence and production in this case is mainly due to shear in the wake fluid illustrating a high active outer region along the wake trajectory and a
relatively calm near-wall. On the other hand, when the cylinder is
well outside the boundary layer G = 2.4, the shear layers and
the boundary layer can be separately identified. The breakdown of
both the outer- and inner-shear layer occurs relatively closer to the
cylinder and the wake tends to retain its coherent structures with
appearance of small-scale eddies. These coherent structures inject
momentum to the boundary layer during their convection that in
turn evolves streaks leading to transition to turbulence of the
boundary layer, an example of receptivity to external disturbances.
Thus, the transition mechanism of the boundary layer for a large
gap ratio is somewhat similar to the bypass transition by fst, although here the excitation is caused by the Krmn vortices.
Downstream of x / D = 7.5, the boundary layer approaches a canonical turbulent layer. This is noteworthy in terms of differences
in response of the boundary layer close to a migrating wake as gap
ratio changes.
Acknowledgment
The authors wish to express their gratitude to the reviewers for
their valuable comments and suggestions.
References
1 Price, S. J., Sumner, D., Smith, J. G., Leong, K., and Paidoussis, M. P., 2002,
Flow Visualization Around a Circular Cylinder Near to a Plane Wall, J.
Fluids Struct., 16, pp. 175191.
2 Norberg, C., 1994, An Experimental Investigation of the Flow Around a
Circular Cylinder: Influence of Aspect Ratio, J. Fluid Mech., 258, pp. 287
316.
3 Williamson, C. H. K., 1996, Vortex Dynamics in the Cylinder Wake, Annu.
Rev. Fluid Mech., 28, pp. 477539.
4 Taneda, S., 1965, Experimental Investigation of Vortex Streets, J. Phys. Soc.
Jpn., 20, pp. 17141721.
5 Bearman, P. W., and Zdravkovich, M. M., 1978, Flow Around a Circular
Cylinder Near a Plane Boundary, J. Fluid Mech., 89, pp. 3347.
6 Grass, A. J., Raven, P. W. J., Stuart, R. J., and Bray, J. A., 1984, The Influence of Boundary Layer Velocity Gradients and Bed Proximity on Vortex
Shedding From Free Spanning Pipelines, ASME J. Energy Resour. Technol.,
106, pp. 7078.
7 Taniguchi, S., and Miyakoshi, K., 1990, Fluctuating Fluid Forces Acting on a
Circular Cylinder and Interference With a Plane Wall, Exp. Fluids, 9, pp.
197204.
8 Lei, C., Cheng, L., and Kavanagh, K., 1999, Re-Examination of the Effect of
a Plane Boundary on Force and Vortex Shedding of a Circular Cylinder, J.
Wind. Eng. Ind. Aerodyn., 80, pp. 263286.
9 Buresti, G., and Lanciotti, A., 1979, Vortex Shedding From Smooth and
Roughened Cylinders in Cross-Flow Near a Plane Surface, Aeronaut. Q., 30,
pp. 305321.
10 Angrilli, F., Bergamaschi, S., and Cossalter, V., 1982, Investigation of Wall
Induced Modifications to Vortex Shedding From a Circular Cylinder, ASME
J. Fluids Eng., 104, pp. 518522.
11 Zdravkovich, M. M., 1985, Forces on a Circular Cylinder Near a Plane Wall,
Appl. Ocean Res., 7, pp. 197201.
12 Vada, T., Nestegard, A., and Skomedal, N., 1989, Simulation of Viscous Flow
Around a Circular Cylinder in the Boundary Layer Near a Wall, J. Fluids
Struct., 3, pp. 579594.
13 Liou, T. M., Chen, S. H., and Hwang, P. W., 2002, Large Eddy Simulation of
Turbulent Wake Behind a Square Cylinder With a Nearby Wall, ASME J.
Fluids Eng., 124, pp. 8190.
14 Dipankar, A., and Sengupta, T. K., 2005, Flow Past a Circular Cylinder in the
Vicinity of a Plane Wall, J. Fluids Struct., 20, pp. 403423.
15 Mittal, R., and Balachandar, S., 1995, Effect of Three-Dimensionality on the
Lift and Drag of Nominally Two-Dimensional Cylinders, Phys. Fluids, 7, pp.
18411865.
16 Germano, M., Piomelli, U., Moin, P., and Cabot, W. H., 1991, A Dynamic
Subgrid-Scale Eddy Viscosity Model, Phys. Fluids A, 3, pp. 17601765.
17 Lilly, D. K., 1992, A Proposed Modification of the Germano Subgrid-Scale
Closure Method, Phys. Fluids A, 4, pp. 633635.
18 Chorin, A. J., 1968, Numerical Solution of the NavierStokes Equations,
Math. Comput., 22, pp. 745762.
19 Zhang, S. L., 1997, GPBi-CG: Generalized Product-Type Methods Based on
Bi-CG for Solving Nonsymmetric Linear Systems, SIAM J. Sci. Comput.
USA, 18, pp. 537551.
20 Mittal, R., and Moin, P., 1997, Suitability of Upwind-Biased FiniteDifference Schemes for Large-Eddy Simulation of Turbulent Flows, AIAA J.,
35, pp. 14151417.
21 Morinishi, Y., Lund, T. S., Vasilyev, O. V., and Moin, P., 1998, Fully Conservative Higher Order Finite Difference Schemes for Incompressible Flow,
J. Comput. Phys., 143, pp. 90124.
22 Beaudan, P., and Moin, P., 1994, Numerical Experiments on the Flow Past a
Circular Cylinder at Sub-Critical Reynolds Number, Department Mechanical
Engineering, Stanford University, Report No. TF-62.
23 Kravchenko, A. G., and Moin, P., 2000, Numerical Studies of Flow Over a
Circular Cylinder at ReD = 3900 , Phys. Fluids, 12, pp. 403417.
24 Orlanski, I., 1976, Simple Boundary Condition for Unbounded Hyperbolic
Flows, J. Comput. Phys., 21, pp. 251269.
25 Spalart, P. R., 1986, Numerical Study of Sink Flow Boundary Layers, J.
Fluid Mech., 172, pp. 307328.
26 Fadlun, E. A., Verzicco, R., Orlandi, P., and Mohd.-Yusof, J., 2000, Combined Immersed Boundary Finite Difference Methods for Three Dimensional
Complex Flow Simulations, J. Comput. Phys., 161, pp. 3560.
27 Muldoon, F., and Acharya, S., 2005, Mass Conservation in Immersed Boundary Method, Proceedings of the FEDSM, pp. 19, Paper No. FEDSM-77301.
28 Sarkar, S., 2008, Identification of Flow Structures on a LP Turbine Blade Due
to Periodic Passing Wakes, ASME J. Fluids Eng., 130, p. 061103.
29 Sarkar, S., 2007, The Effects of Passing Wakes on a Separating Boundary
Layer Along a Low-Pressure Turbine Blade Through Large-Eddy Simulation,
Proc. Inst. Mech. Eng., Part A, 221, pp. 551564.
30 Sarkar, S., and Voke, P. R., 2006, Large-Eddy Simulation of Unsteady Surface Pressure on a LP Turbine Blade Due to Interactions of Passing Wakes and
Inflexional Boundary Layer, ASME J. Turbomach., 128, pp. 221231.
31 Sarkar, S., and Sarkar, S., 2007, Large-Eddy Simulations of Cylinder Boundary Layer Interactions, Proceedings of the ACFD7, pp. 13491360.
32 Sarkar, S., and Sarkar, S., 2007, Immersed Boundary Method for Simulating
Complex Flows, Proceedings of the ICFD07, University of Reading, UK.
33 Jacobs, R. G., and Durbin, P. A., 2001, Simulations of Bypass Transition, J.
Fluid Mech., 428, pp. 185212.
34 Ovchinnikov, V., Piomelli, U., and Choudhari, M. M., 2006, Numerical
Simulations of Boundary-Layer Transition Induced by a Cylinder Wake, J.
Fluid Mech., 547, pp. 413441.
35 Zhou, M. D., and Squire, L. C., 1985, The Interaction of a Wake With a
Turbulent Boundary-Layer, J. Aeronaut. Sci., 89, pp. 7281.
36 Zovatto, L., and Pedrizzetti, G., 2001, Flow About a Circular Cylinder Between Parallel Walls, J. Fluid Mech., 440, pp. 125.
37 Wu, X., Jacobs, R. G., Hunt, J. C. R., and Durbin, P. A., 1999, Simulation of
Boundary Layer Transition Induced by Periodically Passing Wakes, J. Fluid
Mech., 398, pp. 109153.
38 Matsubara, M., and Alfredsson, H., 2001, Disturbance Growth in Boundary
Layers Subjected to Free-Stream Turbulence, J. Fluid Mech., 430, pp. 149
168.
39 Spalart, P. R., 1988, Direct Simulation of Turbulent Boundary Layer Up to
R = 1410 , J. Fluid Mech., 187, pp. 6198.
40 Kim, H. T., Kline, S. J., and Reynolds, W. C., 1968, An Experimental Study
of Turbulence Production Near a Smooth Wall in a Turbulent Boundary Layer
With Zero Pressure Gradient, Stanford University, CA, Report No. MD-20.
SEPTEMBER 2009, Vol. 131 / 091201-13
Paul R. Chiarot
Sergey I. Gubarenko
Ridha Ben Mrad
Pierre E. Sullivan
e-mail: sullivan@mie.utoronto.ca
Department of Mechanical and Industrial
Engineering,
University of Toronto,
5 Kings College Road,
Toronto, ON M5S 3G8, Canada
On the Pulsed and Transitional

Behavior of an Electrified Fluid
Interface
Transient modes of an electrified fluid interface are investigated, specifically, (a) intermittent or pulsed cone-jet mode and (b) smooth and abrupt transitions of the interface in
response to a step voltage. These modes were studied experimentally by capturing the
motion of the interface and measuring the emitted ion current (via electrospray) as they
occur. The observed phenomena are described using an analytical model for the equilibrium of an electrified fluid interface, and the effect of operational parameters on the
transient modes is discussed. Pressure, which is related to the supplied flow rate, significantly influences the behavior of the transient modes. It is useful to understand transient
modes so they can be avoided in applications that require a stable electrospray. However,
with improved knowledge, the modes studied here can assist in the development of specialized applications. DOI: 10.1115/1.3203203
Introduction
An electrospray will form from an air-fluid interface subject to

a large electric field. The interface deforms into a conical shape,
and the spray, consisting of highly charged droplets, emits from
the apex of the cone. This mechanism is used in electrospray
ionization, where compounds are transferred from solution into
the gas phase. Electrospray is of great importance for mass spectrometry 1, material deposition 2, and sample synthesis 3.
The shape of an electrified interface, the conditions that lead to
the onset of jetting, and the properties of the emitted jet have been
studied for decades. Taylor 4 first reported an analytical model
for the formation and structure of the electrified cone. Basaran and
Scriven 5 and Pantano et al. 6 investigated the shape and stability of droplets in an electric field, and Stone et al. 7 used
slender body theory to study drops with conical ends. The
modes or regimes that an electrified interface can exhibit
were reported by Cloupeau and Prunet-Foch 8. Fernandez de la
Mora 9 accounted for the effect of space charge in the emitted
jet on the cone shape to explain differences in observed cone
angles from that predicted by Taylor 4. Fernandez de la Mora
and Loscertales 10 and Ganan-Calvo et al. 11 studied the spray
current and emitted droplet size of a conical electrified interface,
and introduced scaling laws to predict these two quantities. Cherney 12 used perturbation methods to investigate the structure of
the Taylor cone and the emitted jet. For a complete review on the
subject of Taylor cones, see Ref. 13.
More recently the dynamic and transient behavior of an electrified interface has received greater attention. Collins et al. 14
reported a simulation of cone and jet formation and droplet
breakup. Notz and Basaran 15 investigated drop formation in an
electric field, and Reznik et al. 16 investigated the shape evolution of an electrified interface over time, examining isolated droplets and the impact of electrical Bond number on interface evolution.
In this study, two transient modes are examined: a intermittent
or pulsed cone-jet mode, and b smooth and abrupt transitions of
the interface in response to a step voltage. These modes are studied by capturing the motion of the interface and measuring the
emitted ion current. The observed phenomena are then accounted
JOURNAL OF FLUIDS ENGINEERING. Manuscript received October 17, 2008; final manuscript received June 11, 2009; published online August 17, 2009. Assoc. Editor:
Steven T. Wereley.
for using a recently developed equilibrium model by Gubarenko

et al. 17. The model simplifies the treatment of the complex
modes by assuming that the operating point of the interface parameters moves within an operational space and transitions between equilibrium and quasi-equilibrium states. This allows for
these phenomena to be understood and predicted, and the impact
of operational parameters on the transient behavior is determined.
Several experimental techniques have been used to probe the
dynamic behavior of electrified interfaces. Zhang and Basaran
18 used fast imaging to characterize the dynamics of drop formation from a nozzle under electric fields of weak to moderate
strengths. Marginean et al. 19 used a combination of fast imaging and current measurements to study the same problem under
weak, moderate, and large electric fields. Others used Phase Doppler Anemometry to measure the size and velocity of droplets
produced by an electrospray 20,21. Here, we follow Marginean
et al. 19, and use a two pronged approach based on imaging and
current measurements.
Typically, a steady cone-jet is of primary importance, especially
in applications such as mass spectrometry. Knowledge of the transient modes considered in this study, and the operational parameters that cause them, is useful to avoid the negative aspects of
these modes. However, if better understood, the transient modes
considered here can assist in the further development of useful
applications, such as e-jet printing 22 and material deposition
23.
The Equilibrium Model
A full description of the equilibrium model is given by

Gubarenko et al. 17, and the details are briefly summarized here.
The model describes the impact of the three dominant stresses
surface tension, electric Maxwell stress, and pressure
differenceon the equilibrium of an electrified interface note
that the impact of the viscous stress is small and can be neglected. A three dimensional space of the operating parameters is
described, and regions or subdomains are defined where equilibrium, quasi-equilibrium, and nonequilibrium conditions exist.
These subdomains are explicitly defined below. The surfaces dividing the subdomains are referred to as critical surfaces, and
represent the boundary of equilibrium or quasi-equilibrium and
nonequilibrium states. It is convenient to work in a two dimensional operational space, so the electrode separation distance is
typically fixed, and the impact of the applied voltage and pressure
difference is focused on. In this case, the critical surface is re-
SEPTEMBER 2009, Vol. 131 / 091202-1
Fig. 1 Quasi-equilibrium states for the interface. In all cases,

the interface emits a spray.
duced to a critical curve. The model reveals that the angle at the
center of the interface, the apex angle , dramatically affects the
location of the critical surface in the operational space.
Equilibrium means all forces are balanced on the interface,
the apex angle is 0 deg, and the interface does not emit a spray. In
quasi-equilibrium, all forces are balanced on the interface; however, the apex angle is nonzero and the interface emits a spray see
Fig. 1 for examples of an interface in quasi-equilibrium. Fluids of
high conductivity have a small jet formation region located almost
entirely at the apex of the cone 8. This means that an interface in
quasi-equilibrium can accurately be described as a continuous
function after specifying an apex angle i.e., the jet formation
region can be omitted in the model. Highly conductive fluids are
the focus of both this study and the analytical model 17.
An electrified interface has an operating point that is a function
of the applied voltage, electrode separation, and pressure difference residing inside the operational space. The location of the
operating point defines the important characteristics of the interface. By using the equilibrium model, properties of the interface
can be predicted or selected, including the equilibrium status of
the interface, the presence of an electrospray, and the shape of the
meniscus. The operating point can also move within the operational space and transfer between equilibrium subdomains.
To use the equilibrium model, it is necessary to specify the
value of the apex angle and calculate the pressure difference
across the interface. The sine of the apex angle defines the location of the critical surface, and the pressure difference dictates
along with the applied voltage and electrode separation distance
the location of the interface operating point in the operational
domain. The apex angle term is calculated from the experimental
data by fitting a linear function to several points on the order of
1020 to a portion of the interface in the vicinity of the apex. It
is important to note that the apex angle used here and the classic
Taylor angle T 4 are related as + T = 90 deg.
For an interface under pressure, surface tension, and electric
stress, the pressure term is found by solving a stress balance for
the interface, knowing the curvature of the interface and surface
tension coefficient giving the stress from the surface tension, and
the applied voltage and counterelectrode separation giving the
electric stress. Note that the pressure term is the excess internal
pressure at the interface that balances the surface tension and electric stress. The position of the interface as a function of pressure
difference can be expressed using the stress balance, and compared with the position of the interface at experimentally found
points. By minimizing the difference between the measured and
calculated points, the pressure difference across the interface can
be found 24. The pressure difference for interfaces not under
electric stress can be determined by solving the YoungLaplace
equation. Pressure values are gauge pressures.
The equilibrium model accounts for the observed transient phenomena of an electrified interface. The phenomena considered in
this study are intermittent or pulsed cone-jet mode and smooth
and abrupt transitions of the interface in response to a step voltage. The equilibrium model accounts for these phenomena as
movement of the operating point of the interface within the operational domain and the transition between equilibrium and
quasi-equilibrium states. This approach simplifies the treatment of
these transient phenomena while still accurately describing the
observed behavior.
091202-2 / Vol. 131, SEPTEMBER 2009
Most applications require transient phenomena to be avoided all

together. In mass spectrometry or material deposition, stable conejets emitting a steady stream of material are essential. However,
other more specialized applications find transient phenomena to
be useful. For example, it is possible to use pulsed cone-jet mode
to emit discrete packets of droplets through the electrospray
mechanism. Rather than having a continuous spray, droplets are
emitted intermediately using this mode, and it was proposed to
use pulsed cone-jets in printing 22 and material deposition 23
applications. It might also be possible to use this mechanism to
mimic the function of an ion gate or shutter. A theoretical understanding of intermittent or pulsed cone-jet mode using the analytical model reported in Ref. 17, combined with the results of this
study, will assist in the further development of specialized applications. Application of the equilibrium model to transient phenomena is discussed in Sec. 3.
Experimental Investigation of Transient Phenomena
3.1 Materials and Experimental Setup. Testing was performed using metal tubing with a radius of 150 m. The electrified interface was located at the edge of the metal tubing that was
positioned directly in front of the counterelectrode. For the imaging of the interface, the bulk fluid used in this study was a
100 M solution of Rhodamine B in 70:30 MeOH: H2O. The
surface tension values were determined using Ref. 25.
Rhodamine B is a fluorescent dye and was used to improve the
recording, analysis, and display of the interface. For the experiments that involved measuring the emitted ion current, the bulk
fluid is a 100 M solution of sodium iodide NaI in 90:10
MeOH: H2O. For some experiments, the bulk fluid is modified by
adding 1% AcOH.
The flow rate was controlled using a syringe pump Cole
Parmer, Montreal, QC, Canada, and the interface was observed
and recorded using an inverted fluorescent microscope Leica,
Wetzlar, Germany and charge-coupled device CCD camera
Sony, Tokyo, Japan. Using optical filters and a dichromatic mirror a filter cube, the microscope can illuminate the interface with
light at a wavelength of 515560 nm, and pass light at a wavelength exceeding 590 nm to the CCD camera. These wavelengths
are ideal for the excitation/emission spectrum of Rhodamine B.
Voltage is applied to the metal tubing using a high voltage source
Labsmith HVS448 with an operating range 03000 V. The radius of the interface in all of the captured images is 150 m,
which is the same as the outer radius of the metal tubing. Images
of the interface are captured at a resolution of 640 480 in PNG
format and processed using MATLAB 26.
The use of Rhodamine B allows for the interface to be more
easily visualized and processed. However, it was typically found
that the emitted jet could not be visualized when illuminated by
filtered light. This is likely because of the narrow bandwidth applied and the small volume of the jet. The jet was visualized when
the full spectrum bandwidth of light is applied. All frames are
aligned so that the bottom of the image is aligned with the end of
the metal tubing. This was confirmed visually and simplifies image processing.
Emitted ion current measurements are made using a current
amplifier Keithley, Cleveland, OH, and the output is monitored
using a digital oscilloscope Agilent, Santa Clara, CA. Ions are
emitted from the apex of the Taylor cone and neutralized on the
counterelectrode. This current was measured, converted to a voltage, and then amplified for measurement.
3.2 Intermittent or Pulsed Cone-Jet Mode. The intermittent
or pulsed cone-jet mode was identified by Cloupeau and PrunetFoch 8 who reported that this occurs when the voltage is
slightly lower than for which a single permanent jet is obtained,
and this mode is one focus of this study. Marginean et al. 19
studied transitions between the operating regimes of an electrospray in experimental studies, and Li 27 investigated meniscus
Fig. 2 Two images of an electrified interface in intermittent or

pulse cone-jet mode. The shape of the interface cycles back
and forth between a and b at a frequency of 6 Hz. The apex
angle in b is 63 deg.
deformation under the influence of a pulsed voltagein contrast

to this work where intermittent or pulsed cone-jet mode is observed for constant voltages.
An example of pulsed cone-jet mode is shown in Fig. 2. The
applied voltage was held constant at 3000 V and the flow rate was
0.1 l / min. The time difference between Figs. 2a and 2b is
165 ms. These two interface shapes repeated cyclically at a constant frequency of 6 Hz, transforming from a rounded apex slope
at the middle equal to zero to a typical cone-jet configuration, and
back again. During the brief cone-jet period, the interface emits a
spray, resulting in a pulsed spray being emitted at the same frequency as the interface transition.
As shown in the operational domain in Fig. 3, this pulsed conejet transition will straddle two critical curvesone corresponding
to = 0 deg and the other corresponding to = 63 deg the apex
angle in Fig. 2b. The operating point of the interface in Fig.
2a sits left of the critical curve in equilibrium. As the pressure
increases, the operating point moves to the right and crosses the
critical curve to the point marked . The movement is in a
straight horizontal line as the voltage is held constant. As the
operating point crosses the critical curve to the point marked , it
smoothly and immediately relocates into the quasi-equilibrium region, left of the critical curve for = 63 deg, as shown by the
broken arrow in Fig. 3. Once in the quasi-equilibrium region, the
interface appears as seen in Fig. 2b.
Once in the quasi-equilibrium region, the pressure increases for
a short amount of time until it crosses the critical curve for
= 63 deg to the point marked with a . Once it crosses this curve,
the interface is in nonequilibrium, and it adjusts to find an equi-
Fig. 3 Operational domain for an electrode separation distance of 7 mm and surface tension coefficient of 27.48 mN/m.
The operating point moves left and right in the domain across
the critical curves between equilibrium and quasi-equilibrium.
Fig. 4 Current measurements of emitted ion current. The ion

current is collected at the counterelectrode and converted to a
voltage. The bulk fluid is a 100 M solution of NaI in 90:10
MeOH: H2O. The flow rate in a is 2 l / min and in b is
0.5 l / min. The current peaks represent the quasi-equilibrium
state of the interface.
librium state. Due to the low applied flow rate, the interface collapses and returns to the equilibrium state shown by the broken
arrow. Once back in equilibrium, the pressure builds again and
the process repeats itself. This cyclic process is observed to occur
at a very regular frequency of approximately 6 Hz.
The question remains as to why the interface in Fig. 2b, after
moving out of the quasi-equilibrium region, collapses to an equilibrium state rather than moving to a second quasi-equilibrium
state as seen for the smooth transition below. This is because of
the supplied flow rate. For similar operating conditions at higher
flow rates, a stable cone-jet can form 17. Experimentally, it was
found that higher flow rates create larger pressure differences in
the negative direction across the interface. A similar finding was
reported in other electrospray studies 28. At low flow rates, the
pressure difference across the interface remains small, and the
operating point of the interface is limited by the amount it can
move in the negative pressure direction to the left in Fig. 3. As a
result, the operating pointafter crossing into nonequilibrium
must move to the right, back to the equilibrium region. Thus, the
contention of Cloupeau and Prunet-Foch 8 that pulsed cone-jet
mode occurs at voltages slightly below those required for a stable
cone-jet should also include the importance of sufficient pressure
related to flow rate as well. Fernandez de la Mora and Loscertales 10 and Chen and Pui 29 also reported a required minimum flow rate to achieve a stable electrospray. In their work,
scaling laws for the minimum flow rate Qmin are expressed as a
function of the bulk fluid parameters, but they do not specifically
address the relationship of flow rate and pulsed mode.
Ion current measurements confirm the regularity of pulsed
cone-jet mode and the impact of flow rate pressure on the pulse
frequency. Figure 4 shows the measured ion current emitted by an
interface in pulse cone-jet mode for two different flow rate conditions. Figure 4a shows a flow rate of 2 l / min and Fig. 4b
shows a flow rate of 0.5 l / min. In both cases, the bulk fluid is a
SEPTEMBER 2009, Vol. 131 / 091202-3
Fig. 5 Time lapsed images of an interface undergoing an

abrupt transition. The pressure difference is smoothly increased in a and b, and a step voltage is discontinuously
abruptly applied in c. A loss of mass from the interface can
be seen when the voltage is applied. d The stable final
cone-jet.
100 M solution of NaI in 90:10 MeOH: H2O. The current between pulses is primarily attributed to external electrical noise. In
Fig. 4, the current peaks correspond to a quasi-equilibrium state of
the interface i.e., Fig. 2b, and the zero current condition corresponds to the equilibrium state i.e., Fig. 2a. The pulsed mode
seen in Fig. 4 is the low frequency component of what Juraschek
and Rollgen 28 identified as axial spray mode I. The low
frequency oscillation of the spraying jet corresponds to the jumping i.e., movement of the operating point in the parameter space
between equilibrium and quasi-equilibrium subdomains. This
measurement shows that the magnitude of the emitted current remains constant despite changes in the pulse frequency, and that
the pulse frequency the emitted ions can be controlled via the
pressure difference across the interface i.e., the supplied flow
rate.
3.3 Smooth and Abrupt Transitions in Response to a Step
Voltage. Experimental evidence reveals that it is possible to have
both smooth and abrupt transitions from equilibrium to final
quasi-equilibrium state. The type of transition for fixed electrode
separation depends on how smoothly the pressure difference and
voltage are applied to the interface. Smoothly applying an increasing pressure difference across the interface is done by operating
the syringe pump at a constant flow rate. Smoothly applying the
voltage is more difficult, as most high voltage supplies including
the unit used in the study do not allow gradual transitions over a
wide operating range, instead, allowing only finite steps. Nevertheless, the impact of varying the spray voltage, as well as the
resultant response of the interface, was previously addressed by
Marginean et al. 19.
Examples of transitions are shown in Figs. 5 and 6. Figure 5
represents an abrupt transition where pressure is smoothly applied, and the voltage is abruptly applied a voltage step from 0 V
up to 3000 V. Figure 6 represents a smooth transition where the
voltage is applied before any fluid is present at 3000 V, and then
pressure is smoothly applied. The time delay between each frame
is shown in each figure.
The transition of the interface in Figs. 5 and 6 can be tracked
using the operational domains shown in Figs. 7 and 8, respectively. The solid and dashed arrows represent smooth transitions
Fig. 6 Time lapsed images of an interface undergoing a

smooth transition. The voltage is applied in a before any fluid
is present. The pressure difference smoothly changes from b
to d. The interface in d has been steady for hundreds of
milliseconds. No abrupt loss of mass is seen for this smooth
transition compared with the abrupt transition.
091202-4 / Vol. 131, SEPTEMBER 2009
The solid arrow represents the smoothly increasing pressure
difference, and the dotted arrow represents a discontinuous or
abrupt increase in voltage. The application of voltage is discontinuous and not smooth. The pressures at a, b, and d are
known exactly. There is no pressure to report for c since the
interface has ruptured.
and the dotted arrow represents an abrupt change in voltage. For

both cases, the electrode separation is fixed.
The operating point of the interface shown in Fig. 5 on the
domain shown in Fig. 7 starts with a zero pressure difference and
zero applied voltage S in Fig. 7. As fluid is infused with the
syringe pump, the operating point moves smoothly to the right
increasing positive pressure, up to the point shown in Fig. 5a.
The pressure difference across the interface in Fig. 5a can be
found by measuring the height of the interface 17. Mass is continually infused until nearly the boundary of equilibrium shown in
Fig. 5b. At this point, a step voltage of 3000 V is applied, and an
image was taken Fig. 5c showing the interface undergoing an
abrupt transition. As the interface completely ruptures, a substan-
The solid arrow represents smoothly increasing pressure difference, and the dashed arrow represents smooth relocations
of the operating points. The transition is smooth; therefore
there is no abrupt loss of mass, as seen in Fig. 5.
tial mass of fluid is ejected. Since the interface loses mass, the
pressure difference across the interface moves in the negative direction. The operating point after transition Fig. 5d is to the left
of the critical curve for = 50 deg in a quasi-equilibrium state,
since the interface emits a spray after the voltage is applied.
Reznik et al. 16 also found a total rupturing of the interface,
as in Fig. 5c they defined this as droplets jumping off, but
only for large contact angles in excess of 0.8. Here, rupturing
occurs at much smaller angles. Reznik et al. 16 dealt with isolated dropletswhile here droplets are infused with mass from
the syringe pump.
Qualitatively, it has been observed that when the pressure difference across the interface is large before an abrupt transition, the
sudden loss of mass from the interface in transition is large. As
seen in Fig. 5, when the applied pressure before transition is near
the equilibrium limit, the mass lost when the voltage is applied
was large. For smaller pressure differences before abrupt transition, the amount of mass lost from the interface was observed to
be comparatively less.
For the smoothly transitioning interface in Fig. 6, the operating
point starts at zero pressure difference but at an applied voltage of
3000 V the step voltage is applied before any fluid is present.
Mass is continually infused with the syringe pump, and the pressure difference smoothly increases in the positive direction, causing the operating point on the domain in Fig. 8 to start at S and
move to the right. After crossing the critical curve for = 0 deg,
the operating point immediately relocates to the quasi-equilibrium
region, where the pressure increases again. Here the operating
point moves to the right, crossing into nonequilibrium, and adjusts
to find equilibrium. The flow rate in this case is sufficiently high;
therefore the operating point can move to the left for decreasing
pressure, as also noted in Ref. 28, and the interface moves into
a second quasi-equilibrium state. This process repeats until the
interface reaches a final steady operating location.
For the operating domain in Fig. 8, only three discrete transitions into three different quasi-equilibrium regions are shown and
correspond to the three time delays of the interface shown in Fig.
6. It is possible to take more intermediate images of the interface
and plot additional operating points. This would show a series of
transitions between quasi-equilibrium states until reaching the
steady configuration shown in Fig. 6d, and would mean that the
transitional operating points and critical curves would get increasingly closer together.
Overall, the interface in Fig. 6 smoothly transitions to a final
quasi-equilibrium state: cone-jet mode. On the operating domain,
the operating point moves in a smooth manner. Most notably,
there is no dramatic rupture of the interface that is found for the
abrupt transition found in Fig. 5, just a smooth transition to an
interface in cone-jet mode.
The smooth transition can be compared with the pulsed cone-jet
configuration considered in Sec. 3.2. In this case, the interface
does not collapse and is able to return to a quasi-equilibrium state
after a critical curve is crossed. The most notable difference in
operating conditions is the supplied flow rate, where the smooth
transition in Fig. 6 is an order of magnitude larger than the conditions in Fig. 2. With adequate flow rate, the pressure difference
can decrease sufficiently, such that the operating point can move
to the left. The operating point will relocate in a smooth manner
between quasi-equilibrium regions, until a steady value for pressure difference is reached.
Summary
Two transient modes of operation for an electrified interface,

intermittent or pulsed cone-jet mode and the smooth and abrupt
transitions of the interface in response to a step voltage, have been
investigated. Using an analytical model 17, the behavior was
simplified by assuming that the operating point of the interface
parameters moves within an operational domain and transitions
between equilibrium and quasi-equilibrium states. The experimenJournal of Fluids Engineering
tally measured interface motion and emitted ion current confirmed

this and revealed the impact of the operational parameters on the
transient modes. It was found that pulsed cone-jet mode is not
only influenced by the applied voltagepressure related to the
supplied flow rate also plays a role in its formation and behavior.
When applying a step voltage, it is important to avoid an abrupt
transition in most applications. When forming an electrospray for
use in mass spectrometry, a dramatic loss of mass, as the voltage
is first applied, could result in significant sample loss contained
in the solvent. For material deposition, an abrupt emission of
mass affects the uniformity of the film being applied. The amount
of mass ejected during the abrupt transition is unpredictable, and a
good way to avoid abrupt transitions is to smoothly vary the pressure difference from zero after the step voltage is applied.
Acknowledgment
The authors gratefully acknowledge the financial support of the
Natural Sciences and Engineering Research Council of Canada
NSERC and Engineering Services Inc. ESI.
References
1 Fenn, J. B., Mann, M., Meng, C. K., Wong, S. F., and Whitehouse, C. M.,
1989, Electrospray Ionization for Mass Spectrometry of Large Biomolecules, Science, 2464926, pp. 6471.
2 Satterley, C., Perdigao, L., Saywell, A., Magnano, G., Rienzo, A., Mayor, L.,
Dhanak, V., Beton, P., and, OShea, J., 2007, Electrospray Deposition of
Fullerenes in Ultra-High Vacuum: In Situ Scanning Tunneling Microscopy and
Photoemission Spectroscopy, Nanotechnology, 18, p. 455304.
3 Barrero, A., and Loscertales, I. G., 2007, Micro- and Nanoparticles Via Capillary Flows, Annu. Rev. Fluid Mech., 39, pp. 89106.
4 Taylor, G. I., 1964, Disintegration of Water Drops in Electric Field, Proc. R.
Soc. London, 2801382, pp. 383397.
5 Basaran, O., and Scriven, L. E., 1990, Axisymmetric Shapes and Stability of
Pendant and Sessile Drops in an Electric Field, J. Colloid Interface Sci.,
1401, pp. 1030.
6 Pantano, C., Ganan-Calvo, A. M., and Barrero, A., 1994, Zeroth-Order, Electrohydrostatic Solution for Electrospraying in Cone-Jet Mode, J. Aerosol Sci.,
25, pp. 10651077.
7 Stone, H., Lister, J., and Brenner, M., 1999, Drops With Conical Ends in
Electric and Magnetic Fields, Proc. R. Soc. London, Ser. A, 455, pp. 329
347.
8 Cloupeau, M., and Prunet-Foch, B., 1994, Electrohydrodynamic Spaying
Functioning Modes: A Critical Review, J. Aerosol Sci., 256, pp. 1021
1036.
9 Fernndez de la Mora, J., 1992, The Effect of Charge Emission From Electrified Liquid Cones, J. Fluid Mech., 243, pp. 561574.
10 Fernndez de la Mora, J., and Loscertales, I. G., 1994, The Current Emitted
by Highly Conducting Taylor Cones, J. Fluid Mech., 260, pp. 155184.
11 Ganan-Calvo, A. M., Davila, J., and Barrero, A., 1997, Current and Droplet
Size in the Electrospraying of Liquids. Scaling Laws, J. Aerosol Sci., 282,
pp. 249275.
12 Cherney, L., 1999, Structure of Taylor Cone-Jets: Limit of Low Flow Rates,
J. Fluid Mech., 378, pp. 167196.
13 Fernandez de la Mora, J., 2007, The Fluid Dynamics of Taylor Cones, Annu.
Rev. Fluid Mech., 39, pp. 217243.
14 Collins, R. T., Jones, J. J., Harris, M. T., and Basaran, O., 2008, Electrohydrodynamic Tip Streaming and Emission of Charged Drops From Liquid
Cones, Nat. Phys., 4, pp. 149154.
15 Notz, P. K., and Basaran, O., 1999, Dynamics of Drop Formation in an
Electric Field, J. Colloid Interface Sci., 213, pp. 218237.
16 Reznik, S., Yarin, A., Theron, A., and Zussman, E., 2004, Transient and
Steady Shapes of Droplets Attached to a Surface in a Strong Electric Field, J.
Fluid Mech., 516, pp. 349377.
17 Gubarenko, S., Chiarot, P., Ben Mrad, R., and Sullivan, P., 2008, Plane Model
of Fluid Interface Rupture in an Electric Field, Phys. Fluids, 204, p.
043601.
18 Zhang, X., and Basaran, O., 1996, Dynamics of Drop Formation From a
Capillary in the Presence of an Electric Field, J. Fluid Mech., 326, pp. 239
263.
19 Marginean, I., Nemes, P., and Vertes, A., 2007, A Stable Regime in Electrosprays, Phys. Rev. E, 76, p. 026320.
20 Gomez, A., and Tang, K., 1994, Charge and Fission of Droplets in Electrostatic Sprays, Phys. Fluids, 61, pp. 404414.
21 Venter, A., Sojka, P., and Cooks, R., 2006, Droplet Dynamics and Ionization
Mechanisms in Desorption Electrospray Ionization Mass Spectrometry, Anal.
Chem., 78, pp. 85498555.
22 Choi, H., Park, J., Park, O., Ferreira, P., Georgiadis, J., and Rogers, J., 2008,
Scaling Laws for Jet Pulsations Associated With High-Resolution Electrohydrodynamic Printing, Appl. Phys. Lett., 92, p. 123109.
SEPTEMBER 2009, Vol. 131 / 091202-5
23 Paine, M., Alexander, M., Smith, K., Wang, M., and Stark, J., 2007, Controlled Electrospray Pulsation for Deposition of Femtoliter Fluid Droplets onto
Surfaces, J. Aerosol Sci., 38, pp. 315324.
24 Chiarot, P., Gubarenko, S., Ben Mrad, R., and Sullivan, P., 2008, Application
of an Equilibrium Model for an Electrified Fluid InterfaceElectrospray Using a PDMS Microfluidic Device, J. Microelectromech. Syst., 176, pp.
13621375.
25 Lide, D., ed., 2006, CRC Handbook of Chemistry and Physics, 87th ed., Taylor
and Francis, New York.
091202-6 / Vol. 131, SEPTEMBER 2009
26 The Mathworks Inc., 2002, MATLAB Version 6.5.

27 Li, J. L., 2006, On the Meniscus Deformation When the Pulsed Voltage Is
Applied, J. Electrost., 64, pp. 4452.
28 Juraschek, R., and Rollgen, F., 1998, Pulsation Phenomena During Electrospray Ionization, Int. J. Mass Spectrom., 177, pp. 115.
29 Chen, D., and Pui, D., 1997, Experimental Investigation of Scaling Laws for
Electrospraying: Dielectric Constant Effect, Aerosol Sci. Technol., 27, pp.
367380.
Curved Non-Newtonian Liquid

Jets With Surfactants
Jamal Uddin1
e-mail: uddinj@maths.bham.ac.uk
Stephen P. Decent
School of Mathematics Edgbaston,
University of Birmingham,
Birmingham, B15 2TT, United Kingdom
Applications of the breakup of a liquid jet into droplets are common in a variety of
different industrial and engineering processes. One such process is industrial prilling,
where small spherical pellets and beads are generated from the rupture of a liquid thread.
In such a process, curved liquid jets produced by rotating a perforated cylindrical drum
are utilized to control drop sizes and breakup lengths. In general, smaller droplets are
observed as the rotation rate is increased. The addition of surfactants along the free
surface of the liquid jet as it emerges from the orifice provides a possibility of further
manipulating breakup lengths and droplet sizes. In this paper, we build on the work of
Uddin et al. (2006, The Instability of Shear Thinning and Shear Thickening Liquid Jets:
Linear Theory, ASME J. Fluids Eng., 128, pp. 968975) and investigate the instability
of a rotating liquid jet (having a power law rheology) with a layer of surfactants along its
free surface. Using a long wavelength approximation we reduce the governing equations
into a set of one-dimensional equations. We use an asymptotic theory to find steady
solutions and then carry out a linear instability analysis on these solutions.
DOI: 10.1115/1.3203202
Introduction
The manufacture of small spherical pellets can be achieved

through the industrial prilling process. In such a process a specially perforated large cylindrical drum is made to rotate about its
axis. Molten liquid is pumped into the drum, and long curved jets
are produced from the holes in the surface of the drum, which
break up due to centrifugal and capillary instabilities. Break up
leads to the production of droplets, which subsequently solidify
and form pellets. This process is widely used in industry for the
manufacture of fertilizers, magnesium, and aluminium pellets see
Refs. 1,2.
Previous studies to investigate the prilling process for Newtonian liquids have been investigated by Wallwork et al. 3,4 and
Decent et al. 5. Nonlinear effects have also been considered by
Pru et al. 6,7 and Wong et al. 8. The linear and nonlinear
instability of rotating power law jets were investigated by Uddin
et al. 911. Experimental studies to model the prilling process
for Newtonian jets were performed by Wong et al. 8 with a
similar investigation for shear thinning jets by Hawkins et al. 12.
Previous studies on the effect of surfactants along straight liquid threads or bridges have been carried out by Whitaker 13,
who considered the instability of a liquid thread encased in an
inviscid environment and Hansen et al. 14 who investigated the
linear instability of a liquid thread containing soluble surfactants
with a viscous surrounding fluid. They found, as expected, that
surfactants slow the growth of disturbances and that as the importance of surfactant activity increases that the dominant, or most
unstable, the wavenumber decreases. Timmermans and Lister 15
considered nonlinear approximations to the governing equations
of a surfactant laden thread in an inviscid surrounding. They formulated a one-dimensional nonlinear model to investigate instability of the thread to changes in surface tension gradients. Using
a scaling argument they showed that surfactants are swept away
from pinching regions and thus have little effect on any dynamics
in those regions. An extension of these situations involving coaxial jets have been considered by Kwak and Pozrikidis 16 and
Blyth et al. 17, and a fully nonlinear one-dimensional approxi1
JOURNAL OF FLUIDS ENGINEERING. Manuscript received October 15, 2008; final manuscript received July 2, 2009; published online August 18, 2009. Assoc. Editor:
Neelesh A. Patankar.
mation to the governing equations for liquid bridges has been

performed by Abravaneswaran and Basaran 18. The effects of
surfactants on the deformation of droplets have also been extensively studied see Ref. 19. The stability of a two-layer channel
flow and flow down an inclined plane with insoluble surfactants
has been investigated by Blyth and Pozrikidis 20,21. A review of
the surfactant transport equation for an elongating thread and a
stretched interface is provided by Blyth and Pozrikidis 22.
Problem Formulation
We assume that a large cylindrical container having radius s0 is

made to rotate with angular velocity along its axis of symmetry.
A small orifice is placed at the side of the container having radius
a, which is considered small in relation to the radius of the container. We assume that the emerging liquid jet has a thin layer of
insoluble surfactant of fixed concentration along its free surface see for example Ref. 23. Furthermore, the liquid is assumed incompressible.
We examine the problem by considering a coordinate system
x , y , z, which rotates with the container, having an origin at the
axis of the container and with the position of the center of the
orifice at s0 , 0 , 0. The effects of gravity on the jet can be neglected if the centripetal acceleration of the jet s02 is much
greater than the acceleration due to gravity g, as is the case with
industrial prilling. Then the jet moves solely in the x-z-plane and
thus we can describe the centerline of the jet by coordinates
Xs , t , 0 , Zs , t, where s is the arclength along the centerline of
the jet, measured from the orifice, and t is time.
The rotation of the container causes the jet to curve on leaving
the orifice and any analysis of the jet in the rotating x , y , z coordinate system becomes cumbersome. We therefore introduce a
coordinate system previously used by Wallwork 4, which is
similar to the approach adopted by Ribe 24 and Entov and Yarin
25, where we have one coordinate lying along the centerline of
the jet corresponding to the arc length s of the jet, with the remaining coordinates as the plane polar coordinates n, radial
and azimuthal in any cross section of the jet. The associated unit
vectors are represented by es, en, and e, respectively, and their
derivation is explained further in Ref. 4.
To describe the flow we use the velocity vector u = ues + ven
+ we to arrive at the familiar equations of motion including the
continuity equation given by
SEPTEMBER 2009, Vol. 131 / 091203-1
u=0
ut + u u =
1
1
p + 2 u r
where and p are the density and pressure, t is time, and is the
apparent viscosity such that
=m

:
2
1/2
where m is the flow consistency number, is a dimensionless

number referred to as the flow index number, and = u + uT is
the rate of strain tensor. Fluids with less than or greater than
unity are termed shear thinning and shear thickening, respectively.
The angular velocity vector of the container is given by
= 0 , , 0. The position vector of a particle on the free surface is
given by r = esds + nen.
The free surface of the liquid jet can be described by the function n = Rs , , t hence we have the unit normal vector n to the
free surface as
n=
,v,w
=
u
p =
t + s us + s nu n = S,Bs + Dss2
where s = I nn is the surface gradient operator, Ds is the

surface diffusivity of surfactant, and s n = , where is the
mean curvature of the free surface. S is the surfactant source term,
which takes into account absorption from the free surface and is a
function of the surfactant concentration on the surface and surfactant concentration in the bulk Bs. In this paper we consider
insoluble surfactants so S is taken as zero and assume that the
diffusivity of surfactant is small and negligible i.e., Ds = 0. Under
these assumptions, and after using the properties of the surface
gradient operator, we have
t + u n n u = 0
The boundary conditions, which apply on the free surface of the

jet, are identical to those found in Ref. 9, except here the presence of a variable surface tension which is a consequence of the
addition of surfactants leads to nonzero tangential stresses. Thus,
we have the normal and tangential stress conditions as
ti n = ti
where is the total stress tensor given by = pI + , is the

curvature such that = s n, and ti are the two tangent vectors to
the normal at the free surface. The kinematic condition is given by
D
Rs,t, n = 0
Dt
= ,
m
= 1 n,R,
,R
n
a
and
u = u0s,t + nu1s, ,t + n2u2s, ,t +

v = nv1s, ,t + n2v2s, ,t +
w = n2w2s, ,t +
p = p0s, ,t + np1s, ,t +
The concentration of surfactant along the free surface of the jet is

governed by see Ref. 22
and
,Z
,t, = 1 s,X,Z,Ut,a
,X
s
s0
where U is the exit speed of the jet in the rotating frame, is the
aspect ratio of the jet, and u, v, and w are the tangential, radial,
and azimuthal velocity components relative to the centerline of
the jet, respectively.
We exploit the fact that the radius of the orifice is small when
compared to the radius of the cylinder by expanding u, v, w, and
p in Taylor series in n see Refs. 26,27, and R, X, and Z in
asymptotic series in . We assume that the leading order axial
component of the velocity is independent of and that the centerline of the jet is unaffected by small perturbations to leading
order. Thus we have
n Rs,t,
n Rs,t,
n n =
p
,
U2
1
u, v,w,
U
R = R0s,t + R1s, ,t +
X = X0s + X1s,t + Z = Z0s + Z1s,t +
In addition we have the following expansions for the surfactant
concentration and variable surface tension as
= 0s,t + 1s,t + O2
= 0s,t + 1s,t + O2
n = Rs, ,t
u1s + 3v2 + w2 + 3v1S cos = 0
where is the surface tension of the liquid in the absence

= 0 of any surfactant, m is the maximum packing concentration
of surfactant, Rg is the universal gas constant, and T is the temperature.
We make our equations dimensionless using the transformations
091203-2 / Vol. 131, SEPTEMBER 2009
where S = X0sZ0ss X0ssZ0s. The second tangential stress condition is automatically satisfied at leading order and at order 2 we
have
3R20R1v1 + R40w2 + v2 = 0
10
Since v1 is a function of s only we therefore have

11
From the first tangential stress condition at leading order we have
The equation of state relating the surfactant concentration to

the surface tension of the liquid-gas interface is given by the nonlinear Szyskowsky equation
m
= + mRgT log
m
where both these quantities are assumed to vary on the free surface of the jet and only along the jet.
The resulting equations contain a group of nondimensional parameters, which include the Reynolds number Re
= / ms0 U2, the Rossby number Rb= U / s0, and the Weber
number We= U2a / .
The continuity equation at leading order gives v1 = u0s / 2 and
at order n we obtain
w 2 + v 2 = 0
for
u1 = u0S cos
12
After differentiating Eq. 11 with respect to we obtain

w2 = v2
13
v2 3v2 = u1s + 3v1S cos
14
Therefore,
and using the expressions for u1 and v1 we see that
v2 3v2 = u0Ss
u0s
S cos
2
15
Periodic solutions for v2 and w2 are

v2 =
w2 =
16
17
1 u0s
S u0Ss cos
4 2
The NavierStokes equation in the axial direction at order is
1 u0s
S u0Ss sin
4 2
The dimensionless apparent viscosity can be expressed as
= + On
2
su0s
Re
18
3 R0s u0ss
Re 0s
+
u2 = u0s
+
2
R0
4
We 2R0
6
u0sS + u0Ss cos
2
u0s
Re
0
,
R0We
p1 =
R1 + R1
0
+ S cos
R0We
R 02
4
1
v2 +
Re
R0We
R0t +
21
0t + u00s +
24
25
26
u0s
=0
2
27
We see that we have now five equations 22 and 2427 for the
five unknowns u0, R0, 0, X, and Z. The dimensionless equation of
state relating the surfactant concentration to the surface tension
of the liquid-gas interface is given by the nonlinear Szyskowsky
equation
= 1 + log1
28
where the parameter = RgT / determines how effective or

important surfactants are.
Steady-State Solutions
We search for steady-state solutions by considering all the variables to be functions of s. In this case we have
uu0s =
+
22
This equation is valid only if the leading order terms in the expansion of X and Z are independent of t. If we had retained leading order translational velocity terms v0 and w0 in our expansions
for v and w, and had X0t 0 and Z0t 0, then the right hand side
of Eq. 22 would contain some additional unsteady terms in E
= ZsXt ZtXs. However E 0 has already been found in Ref. 6 to
be a very accurate approximation between the orifice and the
break up point of the jet. In particular Pru et al. 7 considered
an expansion of the form Xs , t = X0s + Xs , t and Zs , t
= Z0s + Zs , t and then solved the linearized version of the resulting equations. Perturbations of the steady centerline were
found to be within a maximum deviation of order 102 and thus
are relatively small when compared to O1 values of X0s and
Z0s effectively making the trajectory steady. This is backed up
by the experimental observations of Wong et al. 8, which show
that the centerline of the jet is steady in the rotating frame,
giving Xt 0, Zt 0 and E 0 for experimental observations.
u0s
R0 + u0R0s = 0
2
Xs2 + Zs2 = 1
7 u0s
1
2
u0
+
u20 +
Rb
2 Re
R0We
X + 1Zs ZXs 1
+
u0XsZsss XsssZs
Rb2
Re
The arclength equation at leading order gives
20
By substituting the expression for p1 using Eq. 21 into the

expression formed by differentiating Eq. 20 with respect to ,
and noting that R is periodic in , we get
XsZss XssZs
X0 + 1X0s + Z0Z0s
Rb2
And finally the last equation to be found is the convectiondiffusion equation, which at leading order is
For simplicity of notation we will relabel X0 and Z0 with X and Z,

respectively.
The NavierStokes equation in the azimuthal direction gives at
leading order p0 = 0, which is automatically satisfied. At order
we get the above equation differentiated with respect to . From
the normal stress condition we obtain at leading order and order
p0 =
3
20s
R0s
u0ss + 2u0s
+
R
Re
WeR0
19
2
X + 1Zs ZXs
u0 +
cos
p1 = u20S
Rb
Rb2
Re
1 0
We R0
u0t + uu0s =
From the kinematic condition, we obtain at order
From the second NavierStokes equation, we have at leading

order p0n = 0, which is automatically satisfied. At order we get
23
which after using the expressions for u2 and p0 becomes
where the leading order apparent viscosity is given by

= 3u0s1. The first tangential stress condition at order 2 is
given by
X + 1Xs + ZZs
+
u0ss + 4u2 + u2
2
Rb
Re
u0t + u0u0s = p0s +
1 0
We R0
+
s
20s
WeR0
29
X0sZ0ss X0ssZ0s
=
X0 + 1X0s + Z0Z0s 3
R0s
+
u0ss + 2u0s
2
R
Rb
Re
7 u0s
1
2
u0
+
u20 +
Rb
2 Re
R0We
X0 + 1Z0s Z0X0s 1
+
u0X0sZ0sss X0sssZ0s
Rb2
Re
30
u0s
R0 + u0R0s = 0
2
31
u0s
=0
2
32
and
u00s +
From Eqs. 31 and 32 we observe that R20u0 and 20u0 are constant and by using R00 = 1, 00 = where is the initial surfactant concentration at the nozzle and u00 = 1, we have R20u0
= 1 and 20u0 = 2 so that we can substitute R0 and u0 in the previous equations, which now become
SEPTEMBER 2009, Vol. 131 / 091203-3
where e = 1 + log1 + / 1 is the surface tension of

the undisturbed liquid jet corresponding to a uniform surfactant
concentration , and E = / 1 is the Gibbs elasticity.
If the centerline of the rotating jet is assumed to curve over a
lengthscale of s = O1 and perturbations along the jet are of order
a which is comparable to when s = O1 then travelling wave
+ t must be considered, where s
modes of the form expiks
= s / and t = t / , in order to have k = ks = O1 and = s

= O1. We begin by expressing our disturbances about our steady
state obtained from Sec. 3 as
3
2.5
2
1.5
1
0.5
0.5
4
Z
1 + log1 u1/2
u0s
0 u0s
+
We
2Weu
1
u1/2
2 u0
0
0
u2
X + 1Xs + ZZs 3
+
u0ss 0s
2
Rb
Re
u0
X0sZ0ss X0ssZ0s u20
u1/2
0
u0
33
34
35
The unknowns in Eqs. 3335 are X0, Z0, and u0. This nonlinear
system of ordinary differential equations could be solved using a
centered finite difference method as in Ref. 7 for the case where
= 1, however, since it can be shown that viscosity has a negligible influence see Refs. 7,9,28 on the steady-state solutions
for X, Z, and u0 except in the case of highly viscous fluids which
we do not consider here, we will solve Eqs. 3335 in the
simpler inviscid limit i.e., as Re . In this inviscid limit the
resulting nonlinear equations are similar to those found in Ref. 9
except here there are some extra terms due to variable surface
tension. We solve this system using a RungeKutta method with
initial conditions R0 = u0 = X0s = 1 and X0 = Z0 = Z0s = 0 for s = 0. The
effect of changing the initial surfactant concentration on the
steady centerline of the jet is shown in Fig. 1. The trajectory is
seen to curve less when surfactants are present.
In this section we consider the linear stability of disturbances

about our leading order steady-state solution obtained in Sec. 3. In
this linear analysis it is sufficient to consider a linearized version
of the surface tension equation of state. By expanding Eq. 28 in
a Taylor series about the initial surfactant concentration we have
= 1 + log1 + = 1 + log1
= e E
1
091203-4 / Vol. 131, SEPTEMBER 2009
1
1
2R0ss
0
2 1/2
2 3/2
We
R01 + R0s
1 + R0s
20s
WeR0
38
We substitute Eq. 37 into our governing equations 25, 27,

and 38, and drop the subscript zero from our variables. We get at
leading order the eigenvalue relation
3
Re
k2 + + iku + iku2
k 2 R 1
k2 + iku
2We R2
Ek2
+ iku = 0
2RWe
39
where we have rescaled the Reynolds number and based it on the

orifice radius a so that Re = Re to bring in viscous terms and
obtain a distinguished limit. By choosing i = ku so that disturbances are convected along the jet we find that
r2 +
3
Re
k 2 r
which has solution

r =
3
2 Re
k2 +
k
2
1
k2
E
R 2 k2
=0
2We
R
R

3 k
Re
2 1
2E
k2 R
We R2
RWe
40
1/2
41
Differentiating the above equation to find the wavenumber for
which r is a maximum which we shall henceforth refer to as k
we find
Temporal Instability
u0t + u0u0s =
and finally
37
X0 + 1X0s + Z0Z0s 3
R0s
+
+
u0ss + 2u0s
2
R0
Rb
Re
2u0 X0 + 1Z0s Z0X0s
=
+
Rb
Rb2
2
2
+ Z0s
=1
X0s
R eikset
where is some dimensionless small constant. In this case the

symbols with subscripts denote steady state solutions, and k is a
real wavenumber with being complex, so that = r + ii where
r is the growth rate of disturbances and i is the wavenumber of
disturbances along the jet. In order to prevent instability to wavemodes with zero wavelength we replace the leading order pressure
term in Eq. 24 with the full curvature expression this ad-hoc
approach is discussed in detail in the review on liquid jets by
Eggers 26. We thus have
7 u0s
1
1 + log1
We
2 Re
1
+
u0X0sZ0sss X0sssZ0s
Re
u0
R = R0 +
0
Fig. 1 The steady centerline of a rotating Newtonian = 1.0

liquid jet with and without surfactants. Here we have Rb= 5.0,
We= 6.0, = 0.5, and = 0.5. The presence of surfactants leads
to the jet curving less.
u0u0s =
= 0.0 and = 0.0

= 0.5 and = 0.5
36
k =
1
E
2R31/4 2R + 3Oh
1/2
42
where Oh= We/ Re is the Ohnesorge number. In the limiting

case of zero rotation = 0, the steady-state solution for the radius
of the jet R0 is simply a constant so that R0 = 1. With this value of
the steady state for R0 the eigenvalue relation 40 reduces to the
long-wave limit eigenvalue relation obtained by Timmermans and
Lister 15 for the case of a straight surfactant laden liquid thread.
0.12
85
=0.0
80
breakup length / orifice radius
0.1
= 0.5
0.08
r
= 0.5
0.06
0.04
0.02
75
70
= 0.25,
=0.5
65
60
55
50
= 0.0,
=0.0
45
0
0.2
0.4
0.6
k
0.8
40
0.6
1.2
Fig. 2 The growth rate of disturbances of a rotating shear thinning liquid jet subject to arbitrary disturbance wavenumbers k
at a short distance away from the orifice. The effect of increasing the initial surfactant concentration is seen to lower
growth rates and also to lower the most unstable wavenumber
where r attains a minimum. Here the parameters are We
= 10, Rb= 1.0, Re= 24, = 0.2, and = 0.8.
When there is no surfactant present i.e., 0 = 0, we have = 1

and the above expression reduces to the most unstable wavenumber of a viscous power law liquid jet see Ref. 9
k =
1
2R301/4
43
3Oh + 2R0
As with previous works see Ref. 9 we find that due to variations in the steady-state variables R and u, both the most unstable
wavenumber and its associated growth rate vary along the jet. In
Fig. 2 we plot the growth rate of disturbances r as given by Eq.
41 at a distance s = 0.3 from the orifice for a shear thinning
rotating liquid jet as the initial surfactant concentration is altered. We find that the effect of adding surfactants is to reduce the
growth rate of disturbances to all unstable wavenumbers, as well
as reducing the wavenumber k at which the growth rates attain a
maximum. Furthermore, the effect of increasing the initial surfactant concentration is to reduce the range of instability. In Fig. 3 we
plot the growth rate of disturbances to arbitrary k for a surfactant
laden shear thickening and shear thinning rotating liquid jet. For
these parameter values we find the shear thinning fluid to have
0.08
= 1.3
= 0.7
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
0.2
0.4
0.6
0.8
Fig. 3 The growth rate of disturbances of a rotating liquid jet

subject to arbitrary disturbance wavenumbers k at the orifice at
low rotation rates. Shear thinning jets are more stable and the
most unstable wavenumber increases with the flow index number. We= 12, Rb= 4.0, Re= 24, = 0.2, and = 0.5.
0.8
1
1.2
flow index number,
1.4
Fig. 4 The breakup length of a rotating liquid jet against the

flow index number as calculated using the linear theory in Sec.
4. The case where surfactants are present along the interface
and the case without is shown. Here the parameters are We
= 10, Rb= 1.0, Re= 24, = 0.01, = 0.01, = 1.5, = 0.25, and
= 0.5.
larger growth rates at all wavenumbers and a relatively smaller

value for the most unstable wavenumber. In the classical theory of
Rayleigh 29 the resonant or Rayleigh wavenumber is the
wavenumber k = k at which perturbations grow most rapidly and
lead to the shortest distance between the orifice and first droplet.
Droplet sizes can be estimated by determining the wavelength
associated with this wavenumber = 2 / k and then equating the
volume contained within this wavelength into a sphere. Since in
the present case, both the most unstable wavenumber and its associated growth rate vary as we move along the jet, it is necessary
to determine the location at which breakup occurs and then calculate the value of the most unstable wavenumber there in order
to provide a good estimate of the size of droplets produced. In
order to achieve this we substitute the expression for the most
unstable wavenumber and its associated growth rate see Eqs. 42
and 41 into the equation that describes the evolution of small
unsteady perturbations about the steady-state solution see Eq.
37. Although the variable t does not appear explicitly in our
steady-state analysis in Sec. 3, it does appear implicitly as a function of s and u more specifically t = 1 / usds. After choosing
a suitable value of the dimensionless parameter we can then
determine the location at which R = R0s + eikset = 0. We take

this value of s, at which R = 0, to be the breakup point of the jet.
In Fig. 4 we plot the breakup lengths, as determined by the
method outlined above for rotating liquid jets with different flow
index numbers. It can be seen that surfactants impede the growth
of disturbances and thus produce longer jets. The solutions obtained can be put on the physical x-z plane by first determining the
steady centerline solution; this allows us to calculate Xs, Zs,
Xs, and Zs for each s 0. We then use Eqs. 41 and 42 in
Eq. 37 to determine Rs , t. Subsequently the equation of the
free surface of the jet on the x-z plane is given by
Xs Rs , tZs , Zs Rs , tXs for each s. Profiles of the
jets are shown in Fig. 5 for different flow index numbers and in
Fig. 6 for different Rossby numbers. The resonant wavenumber or
is shown in
most unstable wavenumber at the breakup point k = kbr
Fig. 7 for a rotating liquid jet at moderately high rotation rates
when the initial surfactant concentration on the free surface is
is seen to decrease as is increased for both

varied. While kbr
shear thinning and shear thickening jets, we see that irrespective
and
of the value of shear thinning jets have larger values of kbr
that the difference between them decreases as the initial surfactant
concentration is increased. When the jet eventually ruptures it is to
be expected in the context of linear theory that the volume of
SEPTEMBER 2009, Vol. 131 / 091203-5
0.8
=
1.0
0.6
=
1.5
=
0.5
0.5
X
1.9
Droplet radius / radius of orifice
0.7
0.4
0.3
0.2
0.1
0
0.2
0.4
0.6
0.8
1.8
1.7
1.6
1.5
1.4
= 0.0
= 0.5
1.3
Fig. 5 The profile of a rotating liquid jet with surfactants for

different flow index numbers as calculated using the linear
theory. Here the parameters are We= 12, Rb= 1.0, Re= 15,
= 0.01, = 0.01, = 0.2, and = 0.5. The dotted lines represent the
steady centerline of the jet.
fluid containing one wavelength of the disturbance at breakup will

separate and form a droplet in reality the jet will breakup into
main and satellite droplets. In Fig. 8 we estimate the size of
droplets produced by a rotating shear thickening liquid jet with
surfactants as the rotation rate is varied. Larger droplets are pro-
0.8
0.7
Rb = 1.0
Rb = 4
0.6
0.5
1.5
2
2.5
3
Rossby number, Rb
3.5
Fig. 8 Predicted droplet sizes for a rotating shear thickening

= 1.3 liquid jet when the Rossby number is varied. Here the
parameters are We= 10, Re= 24, = 0.005, = 0.01, and = 0.25.
duced as the rotation rate is decreased in agreement with Newtonian results obtained by Pru et al. 7. Moreover, it can be
seen that the there exists a nonmonotonic behavior with regard to
the size of droplets as Rb is changed; in particular, we have that
surfactant free jets produce larger droplets for large rotation rates
as compared with jets with surfactants, however, for smaller rotation rates i.e., larger values of Rb this result is vice versa. A
similar qualitative picture is obtained in Fig. 9 for a shear thinning
liquid jet. Finally, in Fig. 10 we show the effect of changing the
Weber number on droplet sizes. Droplets become smaller as the
Weber number is increased i.e., the effects of surface tension are
reduced.
0.5
X
Rb = 0.5
0.4
0.3
0.2
0.1
0
0.2
0.4
Z
0.6
0.8
Fig. 6 The profile of a rotating shear thickening liquid jet on

the x-z plane for different Rossby numbers. Here the parameters are We= 10, Re= 20, = 1.5, = 0.005, = 0.01, = 0.2, and
= 0.2.
Conclusions
The linear stability of a slender rotating liquid jet having a layer

of surfactants along its free surface has been investigated. The
most unstable wavenumber along with the corresponding growth
rate has been examined for different flow index numbers and different initial surfactant concentrations.
Liquid jet rupture commonly leads to the production of parasitic satellite droplets. The use of linear theory cannot predict the
occurrence of such droplets and a nonlinear analysis is needed to
examine this phenomenon in greater detail. The results within this
work can be used to aid in optimizing the parameter space when
using a fully nonlinear method to determine breakup lengths and
droplet sizes in particular, the disturbance wavenumber k at the
0.8
= 0.6
= 1.4
1.9
0.75
k br
0.7
0.65
0.6
1.8
1.7
1.6
1.5
1.4
= 0.0
= 0.5
1.3
0.55
0.1
0.2
0.3
0.4
0.5
0.6
Fig. 7 The most unstable wavenumber at breakup for a shear

thinning and shear thickening rotating liquid jet for different
initial surfactant concentrations . Here the parameters are
We= 10, Re= 20, Rb= 1.0 = 0.0025, = 0.01, and = 0.4.
091203-6 / Vol. 131, SEPTEMBER 2009
0.5
1.5
2
2.5
3
Rossby number, Rb
3.5
Fig. 9 Predicted droplet sizes for a rotating shear thinning

= 0.7 liquid jet when the Rossby number is varied. Here the
parameters are We= 10, Re= 24, = 0.005, = 0.01, and = 0.25.
1.9
1.88
1.86
1.84
1.82
1.8
1.78
1.76
1.74
1.72
10
15
20
25
Weber number, We
30
35
Fig. 10 Predicted droplet sizes for a rotating shear thinning

= 0.5 liquid jet when the Weber number is varied. Here the
parameters are Rb= 2.0, Re= 24, = 0.005, = 0.01, = 0.25, and
= 0.5.
orifice which results in the shortest jet is of practical importance

and is usually determined by examining jets subject to a range of
k as in Refs. 7,12. Investigating the nonlinear effects of surfactant transport and diffusion along shear thinning rotating liquid
jets is the focus of current work.
Acknowledgment
J.U. would like to thank EPSRC-GB for their financial support.
References
1 Andersen, K. G. and Yttri, G., 1997, Porso vert, Universitets for laget Oslo.
2 Ornek, D., Gurkan, T., and Oztin, C., 2000, Prilling of Aluminium Sulphate,
J. Chem. Technol. Biotechnol., 758, pp. 689694.
3 Wallwork, I. M., Decent, S. P., King, A. C., and Schulkes, R. M. S., 2002,
The Trajectory and Stability of a Spiralling Liquid Jet. Part 1, Inviscid
Theory, J. Fluid Mech., 459, pp. 4365.
4 Wallwork, I. M., 2002, The Trajectory and Stability of a Spiralling Liquid
Jet, Ph.D. thesis, University of Birmingham, Birmingham.
5 Decent, S. P., King, A. C., and Walwork, I. M., 2002, Free Jets Spun From a
Prilling Tower, J. Eng. Math., 42, pp. 265282.
6 Pru, E. I., Decent, S. P., King, A. C., Simmons, M. J. H., and Wong, D. C.
Y., 2006, Nonlinear Travelling Waves on a Spiralling Liquid Jet, Wave Motion, 43, pp. 599618.
7 Pru, E. I., Decent, S. P., Simmons, M. J. H., Wong, D. C. Y., and King, A. C.,
2007, Nonlinear Viscous Liquid Jets From a Rotating Orifice, J. Eng. Math.,
57, pp. 159179.
8 Wong, D. C. Y., Simmons, M. J. H., Decent, S. P., Pru, E. I., and King, A. C.,
2004, Break-Up Dynamics and Drop Sizes Distributions Created From Spiralling Liquid Jets, Int. J. Multiphase Flow, 305, pp. 499520.
9 Uddin, J., Decent, S. P., and Simmons, M. H., 2006, The Instability of Shear
Thinning and Shear Thickening Liquid Jets: Linear Theory, ASME J. Fluids
Eng., 128, pp. 968975.
10 Uddin, J., Decent, S. P., and Simmons, M. H., 2008, Nonlinear Waves Along
Rotating Non-Newtonian Liquid Jets, Int. J. Eng. Sci., 46, pp. 12531265.
11 Uddin, J., Decent, S. P., and Simmons, M. H., 2008, The Effect of Surfactants
on the Instability of a Rotating Liquid Jet, Fluid Dyn. Res., 40, pp. 827851.
12 Hawkins, V. L., Simmons, M. J. H., Brisbane, C. J., Uddin, J., and Decent, S.
P., 2007, Break-Up of Spiralling Non-Newtonian Liquid Jets, Sixth International Conference on Multiphase Flow, Leipzig, Germany.
13 Whitaker, S., 1976, Studies of Drop-Weight Method for Surfactant Solutions
III, J. Colloid Interface Sci., 54, pp. 231248.
14 Hansen, S., Peters, G. W. M., and Meijer, H. E. H., 1999, The Effect of
Surfactant on the Stability of a Fluid Filament Embedded in a Viscous Fluid,
J. Fluid Mech., 382, pp. 331349.
15 Timmermans, M.-L. E., and Lister, J. R., 2002, The Effect of Surfactant on
the Stability of a Liquid Thread, J. Fluid Mech., 459, pp. 289306.
16 Kwak, S., and Pozrikidis, C., 2001, Effect of Surfactants on the Instability of
a Liquid Thread or Annular Layer. Part I: Quiescent Fluids, Int. J. Multiphase
Flow, 17, pp. 137.
17 Blyth, M. G., Luo, H., and Pozrikidis, C., 2006, Stability of Axisymmetric
Core-Annular Flow in the Presence of an Insoluble Surfactant, J. Fluid
Mech., 548, pp. 207235.
18 Ambravaneswaran, B., and Basaran, O. A., 1999, Effects of Insoluble Surfactants on the Nonlinear Deformation and Breakup of Stretching Liquid
Bridges, Phys. Fluids, 115, pp. 9971015.
19 Stone, H. A., and Leal, L. G., 1990, The Effect of Surfactant on Drop Formation and Breakup, J. Fluid Mech., 220, pp. 161186.
20 Blyth, M. G., and Pozrikidis, C., 2004b, The Stability of Two-Layer Channel
Flow With Surfactants, J. Fluid Mech., 505, pp. 5986.
21 Blyth, M. G., and Pozrikidis, C., 2004c, The Effect of Surfactant on the
Stability of Film Flow Down an Inclined Plane, J. Fluid Mech., 521, pp.
241250.
22 Blyth, M. G., and Pozrikidis, C., 2004a, Evolution Equations for the Surface
Concentration of an Insoluble Surfactant; Applications to the Stability of an
Elongating Thread and a Stretched Interface, Theor. Comput. Fluid Dyn., 17,
pp. 147164.
23 Xue, Z., Corvalan, C. M., Dravid, V., and Sojka, P. E., 2008, Breakup of
Shear Thinning Liquid Jets With Surfactants, Chem. Eng. Sci., 63, pp. 1842
1849.
24 Ribe, N. M., 2004, Coiling of Viscous Jets, Proc. R. Soc. London, Ser. A,
460, pp. 32233239.
25 Entov, V. M., and Yarin, A. L., 1984, The Dynamics of Thin Liquid Jets in
Air, J. Fluid Mech., 140, pp. 91113.
26 Eggers, J., 1997, Nonlinear Dynamics and Breakup of Free Surface Flows,
Rev. Mod. Phys., 693, pp. 865929.
27 Hohman, M. M., Shin, M., Rutlidge, G., and Brenner, M. P., 2001, Electrospinning and Electrically Forced Jets. II. Applications, Phys. Fluids, 138,
pp. 22212236.
28 Uddin, J., 2007, An Investigation Into Methods to Control Breakup and Droplet Formation in Single and Compound Liquid Jets, Ph.D. thesis, University
of Birmingham, Birmingham, UK.
29 Rayleigh, L., 1878, On the Instability of Jets, Proc. London Math. Soc.,
s1-10, pp. 413.
SEPTEMBER 2009, Vol. 131 / 091203-7
Circulation Generation and Vortex

Ring Formation by Conic Nozzles
Moshe Rosenfeld
School of Mechanical Engineering,
Tel Aviv University,
Tel Aviv 69978, Israel
Kakani Katija
John O. Dabiri
Graduate Aeronautical Laboratories and
Bioengineering,
California Institute of Technology,
Pasadena, CA 91125
Vortex rings are one of the fundamental flow structures in nature. In this paper, the
generation of circulation and vortex rings by a vortex generator with a static converging
conic nozzle exit is studied numerically. Conic nozzles can manipulate circulation and
other flow invariants by accelerating the flow, increasing the Reynolds number, and by
establishing a two-dimensional flow at the exit. The increase in the circulation efflux is
accompanied by an increase in the vortex circulation. A novel normalization method is
suggested to differentiate between two contributions to the circulation generation: a
one-dimensional slug-type flow contribution and an inherently two-dimensional flow contribution. The one-dimensional contribution to the circulation increases with the square
of the centerline exit velocity, while the two-dimensional contribution increases linearly
with the decrease in the exit diameter. The two-dimensional flow contribution to the
circulation production is not limited to the impulsive initiation of the flow only (as in
straight tube vortex generators), but it persists during the entire ejection. The twodimensional contribution can reach as much as 44% of the total circulation (in the case
of an orifice). The present study offers evidences on the importance of the vortex generator geometry, and in particular, the exit configuration on the emerging flow, circulation
generation, and vortex ring formation. It is shown that both total and vortex ring circulations can be controlled to some extent by the shape of the exit nozzle.
DOI: 10.1115/1.3203207
Keywords: vortex ring, circulation, nozzle, laminar flow, CFD
Introduction
The present paper studies the generation of circulation and vortex rings using conic converging static nozzles in an attempt to
control or manipulate the emerging flow field. The kinematics and
dynamics of vortices formed by starting the flow through parallelwalled tubes were studied extensively in experiments over the
past 70 years 18. Theories and numerical simulations were
implemented to reproduce many of the salient features of the vortex formation process, such as initial vortex sheet rollup 9 and
vortex pinch-off 1014.
A common way of generating vortex rings is by pushing a
piston in a tube 1,35. Two main vortex generator configurations
have been considered: either a tube with an exit protruding into
the flow field, referred to as a nozzle, or a tube with an exit flushed
with a vertical plate, referred to as an orifice. Rosenfeld et al. 10
did not observe significant differences in the maximal total circulation between the two configurations 2%, but the formation
number see definition in Gharib et al. 5 was slightly lower for
the orifice 3.83 versus 3.97 due to vorticity cancellation of the
forming vortex with the side walls. Gharib et al. 5 found experimentally larger differences in the formation number 3.6 and 4.2
for the orifice and nozzle cases, respectively. Mohseni et al. 13
generated numerically vortex rings by applying a nonconservative
force rather than conventional piston/tube arrangements. Nevertheless, the properties of the formed vortex rings were similar to
those generated by a piston in a tube, and they were also affected
by the generation characteristics, such as the amplitude and duration of the forcing.
In addition to these vortex generator types, one can envision
employing nozzles with an axially varying cross section to decrease the exit area, and thus, to control the exit flow conditions,
such as the mean velocity and the velocity profile. Such exit conContributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received April 14, 2009; final manuscript received July 8, 2009; published online August 18, 2009. Assoc. Editor: Mark
Stremler.
figurations can be found in numerous biological systems. Intracardiac blood flow and aquatic propulsion are two areas of active
research in which the dynamics of the starting flow through coniclike nozzle geometries is known to govern the overall system
performance 15,16. The shape of the vortex generator, and in
particular, the exit geometry, may have a significant influence on
the vortex ring formation and pinch-off since the rate of generation of the motion invariants circulation, impulse, and kinetic
energy depend on the details of the flow in the vortex generator.
Indeed, Allen and Naitoh 17 and Dabiri and Gharib 18 manipulated the circulation production rate, vortex circulation, and
energy by dynamically closing an orifice or opening/closing a
nozzle, respectively.
The most widely used analytical tool for predicting the formation kinematics of vortex rings employing parallel wall vortex
generators is the slug model approximation for the vorticity flux
generated by the starting flow 1,4. By assuming the vorticity in
the flow to be confined into a very thin layer near the wall, one
can derive a simple expression relating the circulation flux to the
spatially averaged axial velocity Ue of the jet efflux d / dtt
= 1 / 2U2e t, where t is the time, and is the accumulated circulation. The slug model has had reasonable success in matching
theoretical predictions with empirical measurements of vortex formation from parallel-walled tubes, as long as the discharge duration is short, and the Reynolds number is large 4. However, its
prominent role in current modeling efforts is more closely related
to its ease of use rather than accuracy.
The case of the starting flow through conic nozzle geometries is
inherently beyond the purview of the slug model, due to the existence of nonzero transverse velocity components in the ejected
jet, resulting from the two-dimensional nature of the flow at the
exit. These components are also present during the early stages of
vortex formation from parallel-walled tubes 1,19. However, in
the latter case, their contribution represents a nearly constant shift
in the circulation production, and thus, can be resolved by the
addition of an empirical constant. By contrast, in conic nozzle
SEPTEMBER 2009, Vol. 131 / 091204-1
Fig. 1 Sketch of the domain of computation
geometries, the relationship between the vorticity flux and the

transverse velocity component is nonlinear and time dependent.
The present study is limited to vortex generators employing
static rigid and conic nozzles. The goal is to study circulation
generation and vortex ring formation for this simple, yet unexplored geometry, as a starting point for the full comprehension of
the dynamically varying nozzles. The present NavierStokes numerical simulations demonstrate the ability to control and manipulate the emerging flow.
Fig. 2 Comparison of the circulation evolution of the numerical solid line and experimental symbols results De / Dp
= 0.6
Methods
2.1 Numerical Model. A sketch of the axisymmetric computational domain is given in Fig. 1. The vortex generator has two
sections: a straight tube of length Ls = 40 cm and a constant diameter of D p = 2.5 cm, and a conic nozzle of length Ln = 5.1 cm the
dimensions are chosen to match the experimental setup and exit
diameter De of De / D p = 0.2, 0.4, 0.6, 0.8, or 1. The computational
domain downstream of the nozzle exit has a length of L / D p = 32,
and the outer boundary is at a radial distance of H / D p = 4. Numerical experiments have verified that the downstream and outer
boundaries are placed far enough from the region of interest.
The moving piston is modeled as a uniform velocity inlet. Undocumented simulations performed in the course of the present
study proved that this model produces results that are very similar
to a moving piston, as long as the maximal stroke length is
smaller than the length of the vortex generator. An impulsive velocity program of the piston is used with a constant piston velocity
of U p = 2 cm/ s in all the cases of the present study. Two velocity
programs are used: i a continuous inflow velocity velocity program no. 1 and ii the constant inflow is stopped at a given
dimensionless time velocity program no. 2. On the outer and
downstream boundaries, zero gauge pressure is specified. On all
the other boundaries, wall conditions are given, except on the axis
of symmetry.
The axisymmetric, incompressible, time dependent, and laminar NavierStokes equations have been solved using the finitevolume package of FLUENT 6.23, ANSYS, Inc. Second-order accurate temporal and spatial schemes have been used with pressureimplicit with splitting of operators PISO pressure-velocity
coupling. Mesh-size and time-step independence tests have been
performed, resulting in a mesh of 180,000 nodes with clustering in
the shear layer and near the walls, and a maximal dimensionless
time step of te = Uet / De0.02, where t is the uniform time
step. The difference in the circulation of the selected mesh relative
to a finer mesh with 685,000 nodes is less than 1%, and relative to
a time step, twice as fine, is less than 0.1%.
2.2 Dimensionless Parameters. The choice of nondimensional parameters is not obvious in the present case since two
velocity scales existone is the piston velocity that remains constant in all the cases and the other is the spatial average exit
velocity Ue that depends on the diameter of the nozzle exit. The
corresponding length scales are the piston diameter and the
nozzle-exit diameter, respectively. Gharib et al. 5 suggested for a
straight tube vortex generator a dimensionless formation time defined by
091204-2 / Vol. 131, SEPTEMBER 2009
t =
U pt
Dp
1a
This formation time is equivalent to the dimensionless piston

stroke divided by the piston diameter t = L p / D p. A nondimensional circulation can be defined by
=
U pD p
1b
although other alternatives were suggested 3,13, while Gharib et

al. 5 preferred using the dimensional form of the circulation. In
a similar way, a nondimensional formation time and circulation,
based on the nozzle-exit parameters, can be defined as
te =
U et
De
and
e =
U eD e
where the spatial average velocity is Ue = U pD p / De2. The relationship between the two formation time definitions is te
= t / De / D p3. As De / D p decreases, the inflow duration to reach a
given te is shorter. For example, to reach a formation time of te
= 4 requires a piston stroke of t = L p / D p = 4, 2.1, 0.85, 0.25, and
0.03 for a nozzle with De / D p = 1, 0.8, 0.6, 0.4, and 0.2,
respectively.
The Reynolds number, based on nozzle-exit parameters, is Re
= 2500, 1250, 830, 625, and 500 for the nozzles with De / D p
= 0.2, 0.4, 0.6, 0.8, and 1.0, respectively. Consequently, in all the
simulations, a laminar flow is assumed, although the case with
Re= 2500 might be transitional 2,20.
2.3 Validation With Experiments. A set of digital particle
image velocimetry DPIV experiments with a nozzle of De / D p
= 0.6 have been performed for validating the numerical simulations. The experimental setup is described in Ref. 7. A nearly
impulsive velocity program, as recorded in the experiments with
a mean velocity of U p 5.7 cm/ s, was imposed in the simulations. The circulation is a major quantity of interest in the present
study, and therefore, the measured and calculated evolution of the
circulation are compared in Fig. 2. Excellent agreement is obtained.
The Flow Inside the Vortex Generator
The efflux of circulation and other motion invariants, and the

vortex ring formation are determined by the velocity profile at the
exit of the vortex generator, e.g., Refs. 10,12,21. This emerging
Fig. 3 The evolution of the axial velocity profile at the nozzle exit velocity program no. 1
flow is affected by the dynamics of the flow inside the vortex

generator. In this section, the characteristics of the nozzle-exit
flow are investigated.
In the straight segment of the vortex generator, the flow properties in the initial phase of the flow are similar to that described
in Refs. 21,22. A Stokes layer is created along the straight tube
wall with the instantaneous start of the inflow. As the inflow continues, a developing flow in the upstream part is established, and it
advances downstream with time. By t = 12, a stationary flow is
obtained along the entire straight segment of the tube with an
almost parabolic velocity profile at the downstream end. Another
observation is that the flow entering the conic cross section is
found to be independent of the nozzle geometry.
The converging shape of the nozzle accelerates the mean flow,
influencing the boundary layer thickness, and consequently, the
efflux of circulation, impulse, and kinetic energy. It also changes
the Reynolds number of the emerging flow. The favorable pressure gradient decreases the thickness of the boundary layer. As the
exit diameter decreases, the boundary layer thinning is more pronounced, resulting in a more uniform exit velocity profile and a
thinner shear layer.
To examine more closely the nozzle-exit flow, Fig. 3 plots the
axial velocity profile for formation times of te = 2, 4, and 10, and
for the different exit diameter nozzles in the case of velocity program no. 1. The first formation time te = 2 is in the roll up phase
of the vortex ring, the second te = 4 is approximately equal to the
formation number of a straight tube, and in the latter instant te
= 10, the leading vortex ring has propagated downstream, far
enough to not affect the flow at the nozzle exit see Fig. 4. The
theory of Gharib et al. 5 predicts that the vortex ring properties
are determined by the ejection characteristics up to a formation
time of F = te 4 for the case of a straight tube vortex generator; this value is also referred to as the formation number 5.
Looking at the nozzle-exit profile reveals that for De / D p 0.8, the
flow is i quasisteady in the core and ii a thin Stokes layer exists
in an otherwise uniform velocity profile. For De / D p 0.8, the
axial velocity profile has enough dimensional time to develop into
a nearly parabolic profile for De / D p = 1.
vorticity distribution for all the nozzle cases. In the formation

phase, the more constricted nozzles exhibit a smaller and more
concentrated vortex core, originating from the thinner vortex
sheet. Significant differences in the vorticity distribution are noticeable only after the formation phase te 4. The thicker shear
layers generated at the lower Reynolds number cases i.e., larger
De / D p create a thicker core vortex ring, as was also observed by
Mohseni et al. 13 for a straight tube vortex generator. The decrease in De / D p has a dual effect regarding viscosity. First, the
Reynolds number based on the exit parameters increases linearly; for De / D p = 1, it increases to Re= 500, while for De / D p
= 0.2, it increases to Re= 2500. Second, as De / D p decreases, the
physical time decreases as well for a given te , and consequently,
viscosity has less time to act through vorticity decay and vorticity
cancellation. The combined effect of these two factors is a larger
vorticity magnitude and thinner core vortex ring as De / D p decreases. At the latest formation time shown te = 10, the vorticity
decay and vorticity cancellation have a prominent effect on the
vorticity distribution for De / D p 0.8. In the case of De / D p = 0.2,
the vorticity accumulated in the tail is susceptible to the Kelvin
Helmholtz instability due to the large Reynolds number of the
emerging flow.
4.2 Total Circulation Evolution. Normalization of the total
circulation by piston parameters might be appealing since it is
common to all the cases considered in the present study. However,
the similarity found in the normalized vorticity field, Fig. 4, supports the normalization of the circulation with the nozzle-exit pa-
Vortex Dynamics and Circulation Evolution
4.1 Vorticity Evolution. The vorticity field for te = 2, 4, and

10 is shown in Fig. 4 for all the nozzle cases and for velocity
program no. 2, with the inflow stopping at te = 8. The vorticity is
normalized by Ue / De, and the axial and radial coordinates by De.
This straightforward normalization reveals the similarity in the
Fig. 4 The scaled vorticity contour lines between 0 to 20 with

an increment of 1 for velocity program no. 2. The spatial coordinates are scaled by De.
SEPTEMBER 2009, Vol. 131 / 091204-3
t =
t
U
De
D
U
e
= U
2 is the running mean velocity
where the velocity scale U
cl
2 1 t 2
Ucl = t 0Ucld. The integration of Eq. 3 using the normalization

of Eq. 4 leads to
=
Fig. 5 The evolution of the total circulation, a normalization

by piston parameters, and b normalization by exit parameters
rameters. To test these normalizations of the circulation, the evolution of the total circulation is depicted in Fig. 5 for the
continuous velocity program no. 1. Normalization by the piston or
the exit parameters are plotted in Figs. 5a and 5b, respectively.
The total circulation is calculated by integrating the vorticity field
in the region outside of the straight tube vortex generator.
The circulation that is also proportional to the dimensional
circulation can be significantly increased by decreasing the exit
diameter of the nozzle, Fig. 5a. The dependence of e on De / D p
is smaller. In the latter normalization, the smallest De / D p exhibits
the lowest normalized circulation for a given te . The circulation,
as predicted by the slug model, is also plotted in Fig. 5b. In all
the nozzle cases, the circulation is larger than the slug model
prediction. The deviation decreases as De decreases due to the
more uniform exit velocity profile Fig. 3 that conforms with the
slug model approximation.
None of these normalizations could collapse the evolution of
the total circulation of all the nozzle cases into a single line,
although the same velocity program of the piston is employed. In
an attempt to alleviate this, an alternative normalization of the
circulation and formation time is suggested hereof. It will also
allow separating between two contributions to the ejected circulation, a slug-type one-dimensional flow contribution, and an inherently two-dimensional flow contribution.
4.3 The Two-Dimensional Contribution to Circulation.
The circulation flux ejected out of the nozzle neglecting diffusive
fluxes can be decomposed 1,19 into
d
=
dt
De/2
1 2
+
udr = Ucl
2
De/2
v
dr
x
where u and v are the axial x and radial r velocity components,

respectively, w is the azimuthal vorticity, and Ucl is the exit centerline velocity. The time integration of this circulation is identical
to the total circulation except a small deviation during the early
formation of the vortex due to the generation of vorticity at the
outer lip of the nozzle. This form suggests the following normalization:
091204-4 / Vol. 131, SEPTEMBER 2009
1
t
+
2 U
D
e
De/2
v
drd
x
The first term is the contribution related to the one-dimensional

flow, while the second term originates from the inherently twodimensional flow at the exit.
In the generalized slug model, the two-dimensional contribution
is neglected, resulting in = t / 2, or in dimensional form
2 t, i.e., the circulation ejected out of the vortex generator
= 1 / 2U
cl
depends on the centerline exit velocity rather than the mean velocity that is used in the original slug model. The generalized slug
model extends the original slug model into cases with a nonuniform axial velocity profile, while keeping the assumption of a
one-dimensional flow at the exit, see for example Ref. 19. In
many previous studies, short ejection time and high Reynolds
number cases have been considered, leading to flows with thin
boundary layers with the mean velocity almost identical to the
centerline velocity. Shusser et al. 21 introduced a first-order approximation that corrected the centerline velocity as a result of the
evolution of a thin Stokes layer, leading to an improved estimation
of the circulation generation for intermediate ejection times. In the
cases of long ejection times and/or low Reynolds number flows,
as in the present study, it is necessary to employ the generalized
slug model to account for the enhanced circulation generation, due
to the exit parabolic-like velocity profile.
The second contribution to the circulation Eq. 5 is significant only when the exit flow is two-dimensional, i.e., when the
integral of uv / x is significant. Most existing studies ignore
this inherently two-dimensional contribution. Didden 1 and
Krueger 19 realized the importance of the two-dimensional contribution to circulation generation in the case of straight tube vortex generators. Krueger 19 showed that for nearly impulsive
velocity programs, a noticeable two-dimensional flow is formed at
a very short time during the initiation of the flow. This results in a
small constant circulation offset 1 / to the classical slug
model. Krueger 19 related this offset to the overpressure that
develops during the initial phases of the flow.
In the present case of conic nozzles, a significant twodimensional contribution is expected throughout the ejection
phase due to the shape of the exit nozzle that enforces a twodimensional flow. Figure 6 plots versus t for the continuous velocity program no. 1. Less dependence of on De / D p is
obtained than in the previous normalizations Fig. 5, indicating
the relatively large one-dimensional contribution. The slope, however, does depend on De / D p. The smaller De / D p is, the larger the
slope is and the larger is the deviation of from the generalized
slug model, i.e., the two-dimensional contribution is more significant. To assess the relative two-dimensional contribution to the
circulation flux throughout the continuous ejection, d / dt versus t is plotted in Fig. 7a. Any deviation of d / dt from 1/2
originates from two-dimensional effects at the exit. Moreover, the
magnitude of the difference is proportional to the relative twodimensional contribution to the total circulation generation. It
should be noted, however, that the magnitude itself of the onedimensional contribution cannot be evaluated from the normalized
since 1 / 2 by definition.
Fig. 8 The asymptotic two-dimensional contribution to the total circulation generation %
Fig. 6 The evolution of the double-star normalized circulation
With the impulsive initiation of the flow, a very large d / dt

is obtained, in agreement with the analysis of Krueger 19. These
effects are significant up to t 1 / 2, i.e., in the initial forming
phase of the vortex ring. In contrast to the findings of Krueger
19 for a straight tube, the two-dimensional effects do not vanish
after the initiation of the flow. Rather, d / dt approaches a
constant asymptotic value greater than 1/2 in the converging
nozzle cases. Figure 8 plots the asymptotic ratio of the twodimensional contribution to the total circulation. It increases linearly with the decrease in De / D p up to a value of 17%.
Krueger 19 used the centerline exit pressure pcl to evaluate the
2D contribution to the circulation instead of the double-star normalization employed in the present study. He also pointed out the
equivalence of pcl / is the density to the two-dimensional
e/2
contribution D
0 uv / xdr. Indeed, plotting the calculated
2
pcl / Ue versus te , Fig. 7b, yields a very similar plot to Fig. 7a.
The persisting two-dimensional contribution of the constricted
conic nozzles can be also noticed in the nonvanishing asymptotic
value of pcl for De / D p 1. One can notice for De / D p 0.6 that
for a brief period in the formation phase te 1, the 2D effects
contribute a negative circulation flux d / dt 1 / 2 in Fig.
7a, and pcl / U2e 0 in Fig. 7b. For larger exit diameter cases,
the negative two-dimensional contribution cancels out a portion of
the circulation added in the flow initiation phase. This occurs
when the vortex ring is developed enough to induce a significant
negative radial velocity at the exit, reversing the sign of v / x.
With the translation of the vortex ring farther downstream, the
sign of v / x changes back to positive in the case of De / D p

1.
Vortex Ring Properties
The evolutions of total and vortex ring circulations both normalized by the exit parameters are shown in Fig. 9 for velocity
program no. 2 the inflow is stopped at te = 8. The maximal vortex
ring circulation is given in Table 1 for the various De / D p cases.
For De / D p 0.8, the enhanced circulation efflux increases the
normalized vortex ring circulation as well. For the straight tube
case De / D p = 1, it is e,V = 3.8, significantly larger than the vortex circulation of e,V 2.7 found by Rosenfeld et al. 10 for a
uniform exit velocity profile. In the case of an imposed parabolic
exit velocity profile, they have obtained a larger vortex circulation
of e,V = 3.5, close to the value calculated in the present study for
the straight tube. The relatively low Reynolds number flow in the
present straight tube case Re= 500 results in a parabolic-like exit
velocity profile Fig. 3, and hence, the increase in the vortex
circulation. The augmented vortex circulation is accomplished by
an axis-touching vortex ring that resembles a Hills spherical vortex, Fig. 4. For smaller exit diameter cases De / D p 0.6, the
vortex circulation reduces to e,V = 2.7, identical to the values obtained for a uniform exit velocity profile 10,12.
To further study the formation process, the evolution of the
normalized energy ejected out of the nozzle, as well as the normalized energy of the vortex ring, are plotted in Fig. 10. Following Gharib et al. 5, the normalized energy is defined as e
= E / I3, where E is the kinetic energy, and I is the impulse. The
normalized vortex ring energy is very useful in studying the dy-
Fig. 7 The evolution of a the normalized circulation flux versus the formation time t, and
b the centerline exit pressure versus te
SEPTEMBER 2009, Vol. 131 / 091204-5
Fig. 11 Geometry of the limiting cases a orifice and b

straight tube vortex generators De = 1.5 cm, H = 10 cm, Ls
= 45.1 cm, and L = 80 cm
Fig. 9 The evolutions of the total lines and vortex ring lines
with symbols circulations
namics of vortex rings as it invokes the three motion invariants

being delivered by the vortex generator. The dependence of the
maximal vortex ring energy on the nozzle-exit diameter is also
given in Table 1. Unlike the normalized circulation, the normalized kinetic energy is larger for the more constricted nozzles, resulting in more energetic vortex rings for a given circulation and
impulse. The decrease of e with the increase in De / D p is related to
the shape vorticity distribution of the vortex ring, as can be seen
in Fig. 4. With the increase in De / D p and the associated decrease
in the Reynolds number, the vortex ring transforms from a thin
core concentrated vortex ring into a thick core axis-touching vortex that resembles the Hills spherical vortex. Gharib et al. 5
showed that for thin core vortex rings, e 0.33 decreases to a
value of e 0.16 for the Hill vortex. In the present case, the larg-
est vortex energy of e = 0.27 is found for De / D p = 0.2, while the

smallest is e = 0.21 for De / D p = 1. The decrease in the vortex energy with the increase in the vortex circulation was also found by
Mohseni et al. 13 for a straight vortex generator. Zhao et al. 12
calculated a vortex energy of e = 0.31 0.34 for a set of imposed
hyperbolic tangent exit axial velocity profiles with thin boundary
layers. However, when Zhao et al. 12 imposed a parabolic exit
velocity profile, the calculated vortex energy dropped to e
= 0.195. Obviously, developed velocity profiles with thick shear
layers reduce the normalized energy of the vortex ring.
Despite the dependence of the circulation efflux and vortex ring
circulation on the conic nozzle-exit diameter, the formation number, as defined by Gharib et al. 5, yields an almost constant value
of F = 3.85 0.25, Table 1, well within the range found in previous experimental and computational studies of straight tube vortex
generators, see for example Refs. 5,10,13. The invariance of the
formation number with the exit diameter is a consequence of the
finding that the increase in the ejected circulation, as De / D p increases, is accompanied by an increase in the vortex ring circulation as well.
6
Table 1 The dependence of the vortex ring parameters on the
exit diameter
De / D p
e,V
0.2
0.4
0.6
0.8
1.0
2.73
2.69
2.79
3.10
3.80
4.1
3.9
3.8
3.9
4.1
0.27
0.24
0.22
0.22
0.21
Fig. 10 The evolutions of the total lines and vortex ring

lines with symbols normalized energies
091204-6 / Vol. 131, SEPTEMBER 2009
Effect of the Nozzle Length
So far, only one factor of the conic nozzle has been considered:
the exit diameter, while the length of the nozzle was kept constant.
The shortening of the nozzle is expanded to increase the twodimensional contribution. A limiting case is a zero-length
nozzlei.e., an orifice, see Fig. 11a. Such a case with an orifice
diameter of De = 1.5 cm, equal to the exit diameter of the conic
nozzle De / D p = 0.6, has also been simulated. It should be noticed
that in most previous vortex ring formation studies that refer to an
orifice case, e.g., Refs. 5,10, the geometry is different. It consists
of a flat plate, flushed with the exit of a straight tube with a small
diameter D p rather than 2H, as in the present case. The most
significant difference is that in the latter case, the flow at the exit
is mostly parallel to the wall of the tube, while in the present case,
there is a large radial component of the flow, upstream of the
orifice, that leads to a large axial gradient of the radial velocity
component, and consequently, to a significant two-dimensional
contribution to the circulation production.
Yet another limiting case that has been simulated is that of a
straight tube vortex generator with a constant diameter equal to
the diameter of the orifice De = D p = 1.5 cm, Fig. 11b. The inflow rate identical to previous conic nozzle cases is employed in
these two additional simulations as well, maintaining the same
exit Reynolds number as of the conic nozzle with De / D p = 0.6
Re= 830.
The exit centerline velocity determines the magnitude of the
one-dimensional contribution, Eq. 3. The evolution of the exit
centerline velocity for these three identical exit diameter cases is
shown in Fig. 12. In all the cases, the initiation of the flow te
1 is accompanied by an overshoot of the exit velocity profile
near the walls, leading to a centerline velocity lower than the
Table 2 Vortex ring parameters for three cases with identical

exit diameter De = 1.5 cm
Straight tube
Conic nozzle
Orifice
Fig. 12 The evolution of the exit centerline velocity for the

cases with an equal exit diameter De = 1.5 cm
mean exit velocity. Afterwards, the exit velocity profile of the

orifice case is almost uniform and steady Ucl / Ue 1.03. The
conic nozzle has a less uniform axial velocity but only by a small
amount Ucl / Ue 1.15. However, it does evolve slowly in time.
In the straight tube case, the velocity profile develops into a nearly
parabolic flow Ucl / Ue 1.92 by te = 20.
The evolutions of e and d / dt are plotted in Fig. 13. The
circulation generated by the conic nozzle and straight tube cases
are almost identical during the starting phase of the flow te 3,
when the exit velocity profile is almost uniform. Only later on
te 4, with the increase in the centerline velocity, the straight
tube generates a larger circulation attributed to the onedimensional contribution. The growth of the centerline velocity
increases the slope de / dte as well, until te 20, when an almost
fully developed flow is attained. Subsequently, the exit flow becomes stationary, and the circulation efflux is also steady.
The two-dimensional contribution in the orifice case is larger
than in the other two cases it contributes as much as 44% of the
3.99
2.79
3.83
4.2
3.8
3.7
0.18
0.22
0.24
total contribution, generating a larger overall circulation efflux up

to te = 8. Afterwards, the continuous increase in the centerline exit
velocity of the straight tube case generates a significantly larger
circulation than the orifice case. Yet, the larger two-dimensional
contribution of the orifice leads to the generation of a larger circulation than in the conic nozzle case.
The vortex circulation and normalized energy, as well as the
formation number of these cases, are summarized in Table 2. The
straight tube and orifice cases generate significantly larger circulation vortices: e,V = 3.99 and 3.83, respectively, as a result of the
larger circulation efflux during the vortex formation phase. The
circulation production is larger either due to a larger centerline
velocity i.e., one-dimensional contribution in the straight tube
case, or due to the two-dimensional contribution in the orifice
case. In the conic nozzle case that is characterized by an almost
uniform exit velocity profile and relatively small 9% twodimensional contribution, the vortex circulation is in the range
reported previously for straight tubes with uniform exit velocity
e,V 2.8.
The formation number is within the common range, in agreement with the previous findings of the present study. The almost
uniform exit velocity profile of the orifice case yields the largest
vortex energy e = 0.24, while the straight tube with the nonuniform exit velocity profile generates the least energetic vortex ring
e = 0.18. The lower energy of the vortex ring is consistent with the
higher formation number since pinch-off occurs later the feeding
vortex sheet is disconnected only when the vortex ring energy
exceeds that of the feeding flow 5.
Fig. 13 Evolution of the a circulation and b circulation flux

for the cases with an equal exit diameter De = 1.5 cm
e,V
Conclusions
The present study presented evidences on the importance of the

vortex generator geometry, and in particular, the exit configuration
on the generation of circulation and on the vortex ring properties
in the case of long ejection times. Conic nozzles allow flow manipulation by increasing the exit mean velocity and the Reynolds
number, and by enforcing a two-dimensional flow at the exit.
While the one-dimensional contribution is proportional to the
square of the centerline exit velocity, and therefore, to approximately the fourth power of D p / De keeping the same volumetric
flow rate, the two-dimensional contribution is found to be linear
with D p / De.
The control of the circulation ejection as well as other motion
invariants is a target of any vortex ring optimization procedure.
The one-dimensional contribution to the circulation is determined
by the magnitude of the exit centerline velocity. It is a function of
time and of the geometry of the vortex generator. Consequently,
the control of the one-dimensional contribution is limited and indirect in nature. A more straightforward control can be obtained
by manipulating the two-dimensional contribution through shaped
nozzles, such as conic nozzles. The two-dimensional contribution
persists the entire duration of the ejection, affecting the circulation
rate as well. The double-star normalization has the advantage of
allowing a quantitative assessment of the two-dimensional contribution relative to the one-dimensional contribution, while the alternate form of Krueger 19 cannot provide such a quantitative
comparison. It has been found that the two-dimensional contribution to the circulation increases as the cone opening angle increases i.e., the exit diameter decreases, and consequently, the
SEPTEMBER 2009, Vol. 131 / 091204-7
circulation generation increases as well. In the limiting case of an

orifice with an opening angle of 90 deg, the two-dimensional
contribution is as large as 44% of the total circulation production.
In the conic cases considered in the present study, the maximal
two-dimensional contribution 17% is obtained for the smallest
exit diameter case De / D p = 0.2.
The increase in the circulation generation rate is accompanied
by an increase in the vortex ring circulation as well, keeping the
formation number almost constant, F 4. Thus, conic nozzles
affect the magnitude of circulation or energy but not the strategy
for obtaining optimal values. Thus, optimization of thrust or other
quantities should employ techniques similar to previous studies of
straight tube vortex generators, e.g., Refs. 6,14.
Nomenclature
D
E
e
F
H
I
L
Lp
Ln
Ls
p
Re
t
te
diameter
kinetic energy g cm/ s2
normalized kinetic energy
formation number
distance of the outer boundary from the axis
cm
impulse g cm/ s
length of the computational domain from the
nozzle exit cm
piston stroke cm
length of the conic segment of the tube cm
length of the straight segment of the tube cm
pressure dyn/ cm2
Reynolds number
time s
formation time based on exit parameters Eq.
2
formation time based on piston parameters
Eq. 1
double-star formation time Eq. 4
the axial and radial velocity components cm/s
t
u, v
U
= U
2 , where U
2 = 1 / tt U2 d
velocity scale U
0 cl
cl
cl
U nozzle exit axial velocity cm/s
x, r axial and radial coordinates cm
Greek Symbols

density g / cm3
azimuthal vorticity 1/s
circulation ejected out of the tube cm2 / s
normalized circulation based on piston parameters Eq. 1
e normalized circulation based on exit parameters Eq. 2
091204-8 / Vol. 131, SEPTEMBER 2009
double-star normalized circulation Eq. 4

Subscripts
cl nozzle exit centerline

e nozzle exit
p piston
References
1 Didden, N., 1979, Formation of Vortex RingsRolling-Up and Production of
Circulation, Z. Angew. Math. Phys., 301, pp. 101116.
2 Glezer, A., 1988, The Formation of Vortex Rings, Phys. Fluids, 3112, pp.
35323542.
3 Glezer, A., and Coles, D., 1990, An Experimental-Study of a Turbulent Vortex Ring, J. Fluid Mech., 211, pp. 243283.
4 Shariff, K., and Leonard, A., 1992, Vortex Rings, Annu. Rev. Fluid Mech.,
24, pp. 235279.
5 Gharib, M., Rambod, E., and Shariff, K., 1998, A Universal Time Scale for
Vortex Ring Formation, J. Fluid Mech., 360, pp. 121140.
6 Krueger, P. S., and Gharib, M., 2003, The Significance of Vortex Ring Formation to the Impulse and Thrust of a Starting Jet, Phys. Fluids, 155, pp.
12711281.
7 Dabiri, J. O., and Gharib, M., 2004, Fluid Entrainment by Isolated Vortex
Rings, J. Fluid Mech., 511, pp. 311331.
8 Krueger, P. S., Dabiri, J. O., and Gharib, M., 2006, The Formation Number of
Vortex Rings Formed in Uniform Background Co-Flow, J. Fluid Mech., 556,
pp. 147166.
9 Pullin, D. I., 1979, Vortex Ring Formation at Tube and Orifice Openings,
Phys. Fluids, 223, pp. 401403.
10 Rosenfeld, M., Rambod, E., and Gharib, M., 1998, Circulation and Formation
Number of Laminar Vortex Rings, J. Fluid Mech., 376, pp. 297318.
11 Mohseni, K., and Gharib, M., 1998, A Model for Universal Time Scale of
Vortex Ring Formation, Phys. Fluids, 1010, pp. 24362438.
12 Zhao, W., Frankel, S. H., and Mongeau, L. G., 2000, Effects of Trailing Jet
Instability on Vortex Ring Formation, Phys. Fluids, 123, pp. 589596.
13 Mohseni, K., Ran, H. Y., and Colonius, T., 2001, Numerical Experiments on
Vortex Ring Formation, J. Fluid Mech., 430, pp. 267282.
14 Linden, P. F., and Turner, J. S., 2001, The Formation of Optimal Vortex
Rings, and the Efficiency of Propulsion Devices, J. Fluid Mech., 427, pp.
6172.
15 Gharib, M., Rambod, E., Kheradvar, A., Sahn, D. J., and Dabiri, J. O., 2006,
Optimal Vortex Formation as an Index of Cardiac Health, Proc. Natl. Acad.
Sci. U.S.A., 10316, pp. 63056308.
16 Dabiri, J. O., Colin, S. P., and Costello, J. H., 2006, Fast-Swimming Hydromedusae Exploit Velar Kinematics to Form an Optimal Vortex Wake, J.
Exp. Biol., 20911, pp. 20252033.
17 Allen, J. J., and Naitoh, T., 2005, Experimental Study of the Production of
Vortex Rings Using a Variable Diameter Orifice, Phys. Fluids, 17, p. 061701.
18 Dabiri, J. O., and Gharib, M., 2005, Starting Flow Through Nozzles With
Temporally Variable Exit Diameter, J. Fluid Mech., 538, pp. 111136.
19 Krueger, P. S., 2005, An Over-Pressure Correction to the Slug Model for
Vortex Ring Circulation, J. Fluid Mech., 545, pp. 427443.
20 Naitoh, T., Fukuda, N., Gotoh, T., Yamada, H., and Nakajima, K., 2002, Experimental Study of Axial Flow in a Vortex Ring, Phys. Fluids, 141, pp.
143149.
21 Shusser, M., Gharib, M., Rosenfeld, M., and Mohseni, K., 2002, On the
Effect of Pipe Boundary Layer Growth on the Formation of a Laminar Vortex
Ring Generated by a Piston/Cylinder Arrangement, Theor. Comput. Fluid
Dyn., 155, pp. 303316.
22 Heeg, R. S., and Riley, N., 1997, Simulations of the Formation of an Axisymmetric Vortex Ring, J. Fluid Mech., 339, pp. 199211.
Bu-Geun Paik
Measurements of High Velocity

Gradient Flow Using Bubble
Tracers in a Cavitation Tunnel
e-mail: ppaik@moeri.re.kr
Kyung-Youl Kim
Jong-Woo Ahn
Maritime and Ocean Engineering Research
Institute,
KORDI,
Jang-dong 171,
Yuseong-gu,
Daejeon 305-343, Korea
The objective of the present study is to investigate propeller wake using particle image
velocimetry (PIV) technique with bubble type of tracers, naturally generated by the
decrease in the static pressure in a cavitation tunnel. The bubble can be grown from the
nuclei melted in the water tunnel and the size of bubbles is changed by varying the tunnel
pressure. A series of experiments are conducted in the conditions of the uniform and high
velocity gradient flows to find out the characteristics of bubble tracers and compared the
measurement results using bubbles with those using solid particles. Bubbles showed good
trace ability in the region of 15 ReS 75; however, some discrepancies showed at high
velocity gradient region of ReS 1000. The fitted vorticity reduction rate would give
reference for the prediction in a real flow when bubble tracers are utilized in PIV measurements of a vortical flow. In addition, the characteristics of bubble slip velocity can
provide information on the vortex core center and the reduction in the Reynolds shear
stress caused by bubbles deformability. DOI: 10.1115/1.3192136
Keywords: PIV (particle image velocimetry), bubble, slip velocity, cavitation tunnel,
vorticity, turbulence
Introduction
The flow velocity can be measured with the PIV technique if

there are tracers in the flow field, which is similar to the laser
Doppler velocimetry LDV technique. In the case of LDV, the
measurement volume is about a few mm3 and the velocity measurement is available even though there is small number of tracers
in the flow. On the other hand, the PIV technique, not a pointwise
measurement but a field measurement, needs uniformly distributed tracers in the flow for the appropriate measurements since it
can have the measurement range of a few m to a few meters. In
addition, the tracer has the size of a few nanometers to a few
millimeters with the variation in the experimental conditions and
the size of the measurement plane. As the size of the tracer is
getting larger, the light intensity scattered from the tracer within
the laser light sheet is also getting higher and this can provide
some merits to the PIV measurement. However, larger tracers may
play a role of additional nuclei in the cavitation tunnel and affect
the cavitation patterns to some degree. As a result, the PIV measurements cannot be conducted together with the cavitation observation tests in a tunnel and lead to some harsh works to clean the
whole cavitation tunnel after them. The Maritime and Ocean Engineering Research Institutes MOERIs midsize cavitation tunnel holds water of about 50 tons and the cleaning works to remove
tracer particles were possible after PIV measurements. Recently,
the MOERI has been constructing another cavitation tunnel,
which will have a large test section of 2.8 1.8 12.5 m3 and
will contain about 2400 tons of water. In this new cavitation tunnel, the cleaning works are considered to be nearly impossible
because they will require too much cost and time. Therefore, serious considerations would be necessary for the effective employment of the PIV measurements in the large cavitation tunnel as
well as the midsize one.
The typical experiments carried out in the MOERIs midsize
cavitation tunnel consists of the reproduction of the ships hull
JOURNAL OF FLUIDS ENGINEERING. Manuscript received September 3, 2008; final manuscript received June 23, 2009; published online August 12, 2009. Assoc. Editor:
Steven Ceccio.
wake, propeller performance tests, rudder cavitation observation,

and so on. For the realization of the hull wake, the equipment
called the wake screen made of several steel meshes is used to
simulate the wake flows behind the bare hull. Since this wake
screen has quite fine and dense meshes, many tiny bubbles can be
generated additionally. Although the large cavitation tunnel does
not need the wake screen but the ship model, it would be meaningful to investigate the trace ability of the bubbles in the presence
of the wake screen for the expansion of the PIV application. The
tunnel pressure should be controlled to match the propeller operation condition and lots of bubbles occurred when it was decreased
under the water vapor pressure.
Studies on the bubbly flow have been performed by many researchers. Brenn et al. 1 investigated the unsteady two-phase
flow with small rising bubbles using phase Doppler anemometry
PDA. They revealed information about flow behaviors of the
two-phase system through relative slip velocities. The increase
in drag on bubbles due to interaction between bubbles and turbulence was introduced numerically by Lane et al. 2. The bubble
size effect was investigated on the gravity-driven pumping by
injecting gas in the pipe using four-point fiber probe and laser
Doppler anemometry LDA see Ref. 3.
The PIV measurements for the two-phase flow were performed
to obtain velocity fields in the bubbly flows. Lindken and Merzkirch 4 did the studies to separate each velocity component of
the liquid or the vapor in the bubbly flow. Although Jansen 5
investigated the breaking wave containing bubbles using the fluorescent particles and ultraviolet rays; they did not show good spatial resolution of the velocity field. Govender et al. 6 showed
insufficient explanation of the PIV technique itself although he
tried to get some information on the two-phase flow. It is necessary to study the details about the light source because the intensity of the scattered light from the bubbles is small in the bubbly
flow with low void fraction and small-size bubbles. Ryu et al. 7
conducted successfully PIV measurements based on the silhouette
technique considering the bubbles as the tracer to visualize the
green water over the head deck of a ship. Recently, Paik et al. 8
measured the velocity fields using bubble tracers; however, they
did not give detailed information on the bubble tracers and their
characteristics.
SEPTEMBER 2009, Vol. 131 / 091301-1
Table 1 Test conditions for 2D PIV measurements

Flow speed m/s
Time interval s
Tunnel pressure atm
3
120
1.0
0.8
0.5
0.3
0.2
5
100
1.0
0.8
0.5
0.3
0.2
8
80
1.0
0.8
0.5
0.3
0.2
Fig. 1 Schematic diagram of a cavitation tunnel
In the present study, the propeller wake as a high velocity gradient flow was investigated using bubbles generated naturally by
the decrease in tunnel pressure. In particular, velocity fields were
extracted to confirm the PIV technique with bubble tracers from
both the uniform flow and the propeller wake within the condition
of high Reynolds number over 106. In addition, the reason why
some discrepancies between two tracer cases occurred was investigated in the viewpoint of the slip velocity. Reference data were
also provided to correct vorticity values in the results of bubbly
flow.
Experimental Apparatus and Method
The experiments have been conducted in the midsize cavitation

tunnel of the MOERI. The rectangular test section has dimensions
of 0.6H 0.6W 2.6L m3. The maximum flow speed is 12 m/s,
and the pressure can be varied from 0.1 kg f / cm2 to 2.0 kg f / cm2.
A schematic diagram of the tunnel is shown in Fig. 1. The twoframe PIV system consists of a dual-head neodymium-doped yttrium aluminium garnet Nd:YAG laser 200 mJ/pulse, two
charge-coupled device CCD cameras, a synchronizer, and a
frame grabber, as shown in Fig. 2. The CCD cameras have a
resolution of 1024 1024 pixels, and a thin laser light sheet was
used to illuminate the center plane longitudinally and capture the
pairs of particle images separated by short time intervals by using
the frame-straddling method. Velocity fields were extracted using
the PIV algorithm of a cross-correlation method based on the fast
fourier transform FFT. The field of view was 10 10 cm2 in the
uniform flow and the measurements were conducted to observe
the behaviors of tracers with the variation in experimental conditions, as shown in Table 1. The tunnel flow speed was detected
from the Pitot tube installed on the wall ahead of the test section.
In this study, together with the uniform flow, the propeller wake
was also measured to find out the likelihood of bubbles as tracers
in the 2D PIV technique. The propeller model for the container
ship model has five blades, a mean pitch ratio of 0.990, and a
diameter of 250 mm. The wake screen was arranged in front of the
propeller model to consider the bare hull wake. The significant
decrease in tunnel pressure for the sake of the generation of
bubbles may result in a reduction in the thrust, mismatching the
design condition of the propeller. Therefore, bubble generation
has to be considered after finding appropriate revolution condition. The present study satisfied well the thrust identity.
The analysis of measurement uncertainties or precision errors
was carried out for actual particle images whose flow field was
known in advance. The quiescent flow was tested for the uncertainty evaluation of the present PIV system by following the procedure recommended by Raffel et al. 9 to estimate the measurement errors of PIV systems under the same condition as explained
in the experimental apparatus and method. The standard errors
encountered in measuring the displacement vector were calculated, and the results were summarized in Table 2. To minimize
the measurement uncertainties, 150 instantaneous velocity fields
were ensemble-averaged in the postprocessing. As a result, the
uncertainty levels measured by PIV system were 0.038% and
0.063% for the axial and radial velocity vectors, respectively. Instantaneous velocity fields were obtained for each measurement
phase. They were ensemble-averaged to obtain the time-averaged
in-plane velocity, vorticity, turbulence intensity, and Reynolds
shear stress in terms of tracers for PIV.
In addition, the bubble size was measured by using the shadowgraph technique with the images of bubbles as tracers. Although the shadowgraph technique may give rather less accurate
sizing values than phase Doppler or global rainbow methods, we
employed it because of the absence of other bubble sizing equipments. Two light bulbs were utilized to illuminate the bubbles at
the bottom and side wall. The visualization system consisted of a
high-speed camera Photron, FASTCAM APX-RS, a Nikon 50
mm lens f = 1.4, metal lamps Photron, HVC-SL, an image processor, and a PC. Figure 3 shows one image of the shadowgraph
technique at 0.38 atm. The camera frame rate was varied from
2000 frames/s fps to 6000 fps, since the flow speed was changed
in the range 38 m/s. The spatial resolution and the size of the
measurement plane were 512 512 pixels and 15 15 cm2, respectively. The bubble size was determined from the occupied
pixels by a bubble, the measurement area, and the spatial resolution of the camera.
Table 2 Measurement uncertainties of displacement vectors

measured by the PIV system used in this study
Fluctuations
m
Fig. 2 Schematic diagram of a PIV experimental setup
091301-2 / Vol. 131, SEPTEMBER 2009
x
y
root-mean-square
rms
Min.
Max.
0.050
0.082
6.317
8.052
6.367
8.4562
Fig. 3 One image of shadowgraph technique at 0.38 atm
As the 2D PIV system uses the time interval between laser

pulses and the frame-straddling method of the CCD camera, the
particle images were captured according to the variation in the
time interval and the tunnel flow speed. Generally, the time interval was chosen for the particle displacement of 35 pixels, which
are efficient for the operation of the PIV algorithm, and the particle images were acquired accordingly.
Figure 4a shows the particle image when the flow speed in
the X direction only and the tunnel pressure were 3 m/s and 0.2
atm, respectively. The tunnel water has no polyamide particles of
20 30 m in diameter, which were used in the tunnel for the
typical PIV measurements. These polyamide particles increased
the contamination of the water and made the cavitation observation quite difficult. In addition, heavy cost and much effort are
necessary to clean the remains of the particles attached to the
tunnel walls and the impeller. The picture shows both large and
small particles; the larger ones mean the bubbles and the small
ones are titanium dioxide TiO2 particles of 3 m. The TiO2
particles do not affect the cavitation patterns on the body surfaces
because they do not function as the nuclei of cavitation. The particle images are similar to each other from 1 atm to 0.3 atm in
tunnel pressure. When the tunnel pressure was lowered below 0.3
atm, many bubbles appeared in the tunnel water. The bubble tracer
has a mean diameter of 0.74 0.24 mm as shown in Table 3, and
it could be employed for the PIV measurements. However, the
influences of the bubbles on the near flow field and their trace
ability in the high velocity gradient flow had to be investigated
minutely for the availability of bubble tracers.
Figure 4b shows the image of particles 5 m/s in flow speed
and 0.2 atm in tunnel pressure. As the increase in the tunnel speed
increasingly lowered the water pressure, the bubbles appeared
from the 0.3 atm tunnel pressure. The bubbles have a mean diameter of 1.03 0.13 mm at 5 m/s, without the scattering band of
the bubble, and larger than that in the condition of 3 m/s. The
increase of the tunnel flow speed to 8 m/s raised the number of the
bubbles and increased the mean diameter to 1.60 0.11 mm.
Since the increase in flow speed resulted in the growth of the
bubble occupied region and the number of bubbles, the buoyancy
that might be caused by the increment of bubble size should be
studied by using the velocity field measurements.
The interrogation window size for PIV measurements was 50%
overlapped 48 48 pixels in the uniform flow. To compare the
differences among velocity field results, the original vector maps
were arranged without removing any error vectors or interpolations in Fig. 5. Figure 5a shows the small error vectors in the
velocity field. They indicate that the particle images including the
bubbles with a diameter of less than 1 mm would be useful for the
Fig. 4 Particle images at the tunnel pressure of 0.2 atm: a 3

m/s, b 5 m/s, and c 8 m/s
PIV analysis. The increment in the flow speed raised the number
of the error vectors in the vector map as shown in Fig. 5c. The
empty spaces in the velocity fields mean errors caused by bubbles.
The velocity field of the flow speed of 8 m/s with the tunnel
pressure of 0.3 atm was similar to that of 0.2 atm and 5 m/s. As
the condition of 8 m/s and 0.2 atm largely increased the error
SEPTEMBER 2009, Vol. 131 / 091301-3
Table 3 Results obtained from the velocity fields in the uniform flows
U0
Pt Bubble size Error ratio U mean V mean Standard error
m/s atm
mm
%
pixel pixel 103 pixel
3
5
8
1.0
0.2
1.0
0.2
1.0
0.2
0.74 0.24
1.03 0.13
1.60 0.11
0.5
1.0
0.8
1.5
2.8
3.5
3.080
3.077
4.210
4.215
5.487
5.490
0.03
0.02
0.03
0.02
0.05
0.03
9.51
9.98
12.72
10.43
14.16
18.25
vectors caused by the light scattered from the bubble, the appropriate bubble images should be considered in advance according
to the experimental conditions.
Table 3 shows the results of the 400 velocity fields in each
uniform flow case. Here, U0 and Pt mean the free stream velocity
and tunnel pressure, respectively. Only the error ratio equal to the
number of error vectors/number of total vectors was obtained
from the original velocity fields, and other parameters such as
mean value and standard error were extracted after the postprocessing of vector maps. The mean displacement in the vertical
direction shows a nearly constant value of about 0.03 pixels,
which were independent of the tunnel flow speed and the size of
bubbles. In other words, the buoyancy effect did not occur in the
uniform flow conditions. At the atmospheric pressure condition,
the error ratio was below 2%, indicating a good uniformity in the
tracer distribution. In the case of 3 m/s, the error ratio and standard error had small values regardless of the variation in the tunnel pressure. With the flow speed of 5 m/s, the error ratio and the
standard error had values larger than those of 3 m/s because of the
increase in the bubble tracer. However, the PIV measurement
could be employed in the 5 m/s condition because the error ratio
of less than 2% means the acquisition of good particle images.
Although the error ratio and the standard error at 8 m/s were also
larger than those of at 3 m/s, a fine control of the time interval
between the two laser sheets, the laser intensity and the aperture
of the CCD camera would be helpful in acquiring better images.
In the case of the 8 m/s and 0.2 atm tunnel pressure, the error ratio
was larger than the 1.0 atm tunnel pressure. The increase in the
error ratio is attributed to the secondary effects by the light
sources scattered from the rather larger bubbles, which illuminate
the surrounding flow region needlessly. Most of all, it is necessary
to treat the bubble tracers carefully in the case of high flow speed
and low tunnel pressure so that large bubbles would not fill up the
whole interrogation window area.
To measure the high velocity gradient flow trace ability of
bubbles in a complicated flow by using the 2D PIV technique, the
propeller wake was selected because it contains the vortical structure with a high velocity gradient. The wake screen in front of the
rotating propeller provided many tiny bubbles through the small
meshes on it, and this made it possible to reduce the interrogation
window for PIV measurements to 50% overlapped 32 32
pixels. In the condition of 5 m/s and 0.25 atm, the mean diameter
of the bubble was 1.03 mm and the size of the measurement plane
was 20 10 cm2. Since the propeller is working in fixed conditions of above pressure and flow speed, the variation in bubble
size was not available in the propeller wake case. The propeller
wake has a quite complicated flow structure that consists of hub
vortices, tip vortices, and trailing vortices. The hub vortices were
excluded in this study because the diffused reflection from them
was too strong to capture the tracer images. The tip vortices are
generated from the pressure difference between the suction and
pressure sides at the blade tip region. The trailing edge of a blade
produced the trailing vortices, which looked like a curved layer
connected to a tip vortex. Figures 6 and 7 showed the particle
images and the instantaneous velocity fields behind the propellers
obtained at two tunnel pressures and 5 m/s free stream. The tip
091301-4 / Vol. 131, SEPTEMBER 2009
Fig. 5 Velocity fields at the tunnel pressure of 0.2 atm: a 3

m/s, b 5 m/s, and c 8 m/s
Fig. 6 Particle images of propeller wake at the same free

stream velocity: a atmospheric pressure and b 0.25 atm
vortices and the wake sheets generated from the blade trailing
resemble each other in both of the instantaneous velocity fields. At
the atmospheric pressure, the aperture of the CCD camera was
fully widened to capture the images of the tiny solid particles. On
the other hand, the aperture was narrowed tightly so that it would
not damage the CCD cells at the condition of 0.25 atm and cavitation number of 1.4 because the light intensity scattered from the
bubbles was very strong.
Figure 8a shows the contour plots of phase-averaged axial
velocity at = 0 deg when the solid particles are put into the
tunnel water. The axial velocity component has large values in the
slipstream, which means that the region has high axial momentum
inside the trajectories of the tip vortices and relatively smaller
values near the blade tip and the propeller axis. Along the tip
region 0.5 Y / D 0.4 of each blade, there are pairs of negative and positive isocontours with the shape of a circle because of
the tip vortices. The velocity cores of Rankin type may give the
information on vortex size from the characteristics of vortex. This
tendency is very similar to the contour plots of bubble tracers
Fig. 8b. The result of the bubble tracers shows the contours
location and shapes similar to that of the solid particles. Although
the magnitudes of axial velocity in the slipstream region are a
little different between solid and bubble cases, the bubbles seem
to show a trace ability as good as the solid tracers.
Figure 9 shows the axial velocity profile at several radial locations at = 0 deg. There was a velocity defect, which looks like a
hump in the region 0.3 r / R 0.7 around X / D = 0.3 and 0.6, and
Fig. 7 Instantaneous fluctuating velocity fields subtracted by

a convection velocity at the same free stream velocity: a atmospheric pressure and b 0.25 atm
it is originated from the merging of the boundary layers developed

on two sides of the blade. As the viscous wake is influenced by a
centrifugal force, the hump of velocity defect grew deeper with
the increase in radial distance. However, there was a significant
change in the axial velocity profile because of the influence of the
tip vortices at r / R = 0.9 in the case of both the solid and bubble
tracers before the location of X / D = 0.35. As the axial velocity of
the slipstream is larger than that of the tip vortices and their rotational component reduces the axial velocity near r / R = 0.9, the
humps at r / R = 0.9 show deep valleys and outbreak earlier than
those at 0.3 r / R 0.7.
Figure 9 shows the limitation of the bubble tracers in a high
velocity gradient flow. In the region of X / D 0.35, bubbles follow the wake flow behaviors such as the movement of the trailing
vorticity and the tip vortices very well. On the other hand, bubbles
could not trace the flow behaviors well in the region of X / D
0.35, especially around the position of r / R = 0.9 where the high
velocity gradients appear. These experimental results led us to find
that the bubble tracers would be useful in a wake region of X / D
0.35 because the bubbles show rather good trace ability in terms
of the axial velocity component compared with the solid particles.
The solid tracers have a slightly larger axial velocity at the range
of 0.3 r / R 0.7 in the whole downstream area. Looking into the
region of X / D 0.35 where the bubbles seem to obtain the stability of its movement, the difference in axial velocity ranged from
0.63% to 2.10% in the radial distance 0.3 r / R 0.7. The axial
velocity difference gradually decreased as the radial distance inSEPTEMBER 2009, Vol. 131 / 091301-5
Fig. 8 The comparison of the phase-averaged axial velocity contours: a solid particles and b
bubble tracers
creased from r / R = 0.3 to r / R = 0.7. The difference between both

tracer cases was from 0.3% to 11.6% at r / R = 0.9.
The difference in velocity magnitude between bubble and water
is known as the slip velocity and is computed as follows:
UmagS =
091301-6 / Vol. 131, SEPTEMBER 2009
where Ububble and Uwater mean the total velocity magnitude of

bubble and solid, and Ufree is the free stream velocity measured by
solid tracers. Figure 10a shows the contours of slip velocity
UmagS regarding two propeller wake cases. In the slipstream region of 0.35 Y / D 0.10, the slip velocity was only 3% of the
frees tream velocity near the trailing vortices, and it has values
ranging from 0.15% to 0.3% at the other areas. Reynolds number
based on the propeller diameter and fluid velocity was around
106 in the present study, and the slip velocity was negligible at the
region of small velocity gradient. This result is very similar to that
of Murai et al. 10 who measured the slip velocity in the horizontal bubbly channel flow at the Reynolds number of 2 104
based on the channel height. To investigate the characteristics of a
bubbly flow further in a high Reynolds number, another Reynolds
number ReS was defined by the slip velocity
ReS =
Fig. 9 The comparison of the phase-averaged axial velocity

profiles at several radial locations
Ububble Uwater
Ufree
Ububble UwaterDbubble
Here, Dbubble and are the mean diameter of the bubbles and the
kinematical viscosity of water, respectively. The slipstream region
in Fig. 10a gives the range of 15 ReS 75 when Dbubble is
supposed to be about 1 mm. This Reynolds number range is
higher than that 0 ReS 9 of the rising bubble case see Ref.
1 in spite of the similar diameter of bubbles. In the range 4
ReS 100, the drag coefficient reported by Clift et al. 11 is as
follows:
Fig. 10 Contour plots of slip velocity between bubble and solid tracers: a for velocity magnitude
and b for vertical velocity
CD = 13.725 ReS0.74
and the rising bubble case gives 1 CD 20. However, it decreased to a range of 0.56 CD 1.85 when a bubble appeared in
a high Reynolds number flow above the transition region, experiencing a small velocity gradient.
In the case of the small velocity gradient flow, the drag force on
the bubbles was too small to require serious correction of a bubbly
flow into a real flow. On the other hand, the high velocity gradient, especially the tip vortex region 0.50 Y / D 0.35 shown
in Fig. 10a, gives different characteristics of bubbles. The ReS
was about 1000 because the slip velocity magnitude increased to
0.2U0 at a tip vortex. The rotational velocity components were
very strong around the core of a tip vortex. The solid tracers
followed the rotational trace around the core better than the
bubble tracers because the skin friction of the solid tracers is
higher than that of bubbles, and it kept the inertial force continuously even at the region of very high velocity gradient. The lift
force on individual bubble see Ref. 12 could be one potential
source of the high slip velocity and it provides bubbles slip in a
shear and rotational region. The pressure gradient force owing to
density effect could be another source because it accelerates
bubbles in the radial direction of vortices. Those made bubbles not
to follow the rotational trajectory exactly and have a rather low
axial velocity component instead around the tip vortex core,
which was different from the real flow behaviors.
In addition, each tip vortex has a similar vertical slip velocity
VS pattern though the wake moves downstream. The area in which
the bubble has a more vertical velocity than water is sandwiched
between those of VS 0 as shown in Fig. 10b. The solid and
bubble tracers circulated around the core of the tip vortex tube
because of significant rotational motion and centrifugal force. In
particular, the bubbles have a greater vertical velocity magnitude
because they may move more rapidly and more easily than the
solids in the state of cavitating vortical flow. The region of the
highest value of VS means the center of the tip vortex core, which
can be confirmed in the Z-directional vorticity contours, as shown
in Fig. 11.
The Z-directional vorticity values of the trailing vortices in the
bubble tracer case are similar to those in solid tracers, and the
contours in the bubble case show a slightly broader distribution
than those of solids as shown in Fig. 11. Near the tip vortices, the
vorticity values in the bubble case became smaller because it lost
some of its trace ability owing to an increase in slip velocity.
Actually, bubble concentration around tip vortices is different
from that of the other regions. In the tip vortex region the bubble
concentration was about 184/ 1202 pixels and it reduced to
75/ 1202 pixels at the sandwiched region between tip vortices.
The bubbles accumulated near vortex cores and resulted with the
reduction in velocity magnitudes of bubbles, which is closely related to the decrease in vorticity value at the tip vortex region of
the bubble case. However, the locations of the tip and trailing
vortices are very similar in both tracer cases. These findings led us
to learn that a flow visualization using bubble tracers can be carried out in terms of the spatial configuration of a complicated
flow. For example, the variation in the blade pitch of a propeller
changes the spatial spacing between the tip and trailing vortices in
a wake flow. Although the PIV technique using bubble tracers
cannot measure the vorticity value accurately in a high velocity
SEPTEMBER 2009, Vol. 131 / 091301-7
Fig. 11 The comparison of the phase-averaged vorticity contours: a solid particles and b bubble
tracers
gradient region such as a vortex core, it would be helpful in diagnosing the parameters affecting the spatial evolution of the flows.
Figure 12 shows the phase-averaged vorticity profiles at several
Fig. 12 The comparison of the phase-averaged vorticity profiles at several radial locations
091301-8 / Vol. 131, SEPTEMBER 2009
radial locations. The vorticity distributions resembled one another

except for the region of tip and trailing vortices in both tracer
cases. Although both tracers displayed similar locations of trailing
vortices at r / R = 0.7 and 0.9, the vorticity itself showed larger
values in the solid tracers, and it increased as the radial distance
increased. In particular, the locations of the trailing vortices varied
according to the radial distance. The trailing vortices of 0.5
r / R 0.7 appeared at more downstream locations than those of
r / R = 0.9 after the position of X / D = 0.4 because the axial velocity
in the slipstream is larger than that in the tip vortices.
The core of a tip vortex has the strongest vorticity value in the
propeller wake except the hub vortices. The vorticity value in the
cores decreased gradually up to the third tip vortex in the case of
solid tracers, and the linear fitted line had an equation of Y
= 15.03X 40.8 as shown in Fig. 13, where Y and X mean vorticity
and X / D axis, respectively. The reduction rate of vorticity value
relaxed largely at the fourth tip vortex. In the case of bubble
tracers, the reduction rate of vorticity increased significantly, and
its equation became Y = 34.02X 37.0 up to the third tip vortex.
Both equations regarding the vorticity reduction rate would give
some insight into how much vorticity value was underestimated
when a PIV technique was employed in the measurements of high
velocity gradient flow using bubble tracers. We found that the
differences in the tip vortices vorticity values between both tracers are attributed to the large slip velocity and ReS in the high
velocity gradient region, as shown in Fig. 10. The difference in
the vorticity values between the two tracer cases was getting
larger as the wake moved downstream. However, its difference
Fig. 13 The comparison of the vorticity values at the cores of

tip vortices
maintained a constant value from the third tip vortex. In addition

to the provision of criteria for the adoption of bubble tracers, these
characteristics would be useful references for analyzing vortical
flow with strong vortices.
In general, a bubble slip velocity is generated to make a balance

between buoyancy and drag at the rising bubble in water without
any acceleration. In a turbulent flow, however, a bubble experiences accelerations and decelerations continuously because of the
vortical turbulence. The viscosity in the complicated turbulent
flow is known to increase and the additional turbulent viscosity is
known to affect the Reynolds shear stress see Ref. 13. The
bubbles influenced the turbulent viscosity in the slipstream region
of r / R = 0.5 and 0.7 and increased the Reynolds shear stress as
shown in Fig. 14. The humps and discontinuities indicate the existence of the trailing and tip vortices in the wake flow system.
Although the absolute values of Reynolds shear stress of bubbles
are higher than those of solid tracer, the distribution pattern and its
amplitude are similar to each other, showing a biased trend owing
to the slip velocity. Actually, Moriguchi and Kato 14 reported
that the reduction in Reynolds shear stress induces the reduction
in skin friction in the high shear rate flow when the bubbles appear on the wall without any dependency of the bubble size. The
flow with a high shear rate on the wall was known to increase the
bubbles deformability and reduce the Reynolds shear stress. The
present experiments also showed that the bubbles reduced the profile amplitude in Reynolds shear stress in the region of r / R = 0.8
and 0.9 though they displayed a sort of biased trend of Reynolds
shear stress. It seems that the high shear rate of the tip vortices
provides the bubbles excessive deformability and influences on
the amplitude decrease in the Reynolds shear stress, showing an
absolute value smaller than the solid case at the tip vortices. Actually, it was well known that bubbles experience an oscillatory
deformation when the Weber number of bubbles is larger than
Fig. 14 The comparison of the Reynolds shear stress profiles at several radial locations
SEPTEMBER 2009, Vol. 131 / 091301-9
unity in a flow, especially rotating motion like vortex see Ref.

15. As the Weber number of bubbles was about 1.04 near a tip
vortex core in the present study, the deformability of bubbles
might be affecting the Reynolds shear stress around it.
The contour plots extracted from the ensemble-averaged velocity fields showed good trace ability from the viewpoint of locations of wake sheet and tip vortices. However, there are some
discrepancies between the two results. For example, the vorticity
values are different near the trailing vorticity and the tip vortices
owing to the slip velocity variation. Additionally, an in-depth investigation of the bubbles as cavitation nuclei is called for because
they might strengthen the cavitation or affect the cavitation pattern
itself on the body surfaces. It would also be necessary to study the
influences of error owing to the variation in bubble size in a high
velocity gradient flow in the future though they could not be revealed in the present study because of the limitation of experimental environments.
Conclusions
The trace ability of the bubbles was investigated by using a

two-frame PIV technique at high Reynolds number over 106 in the
midsize cavitation tunnel. The velocity fields in the longitudinal
planes were obtained both in uniform flows and in high velocity
gradient flows of the propeller wake.
In the case of the uniform flow of 0.2 atm tunnel pressure, the
bubbles about 1 mm in size produced good trace ability. The
bubbles showed rather good trace ability in the high velocity gradient flow as well. The bubbles followed the behaviors of the
propeller wake very well in the slipstream region. However, as the
flow closes to a tip vortex in a high velocity gradient region, the
bubbles showed some discrepancies on the flow behaviors.
A comparison of velocity field data in both tracer cases revealed
the reason for the different results. The large slip velocity and ReS
provided a difference especially in the high velocity gradient region. The slipstream region gives a range of 15 ReS 75; however, ReS is about 1000 at a high velocity gradient region of a tip
vortex. Fitted equations regarding the vorticity reduction rate
would give some reference that is useful for the PIV measurements of high velocity gradient flow using bubble tracers. The
highest vertical slip velocity VS appears at the center of the tip
vortex core, and the high shear flow seems to cause the bubbles
deformability to reduce the Reynolds shear stress.
091301-10 / Vol. 131, SEPTEMBER 2009
Acknowledgment
This research was sponsored by the Ministry of Knowledge
Economy, Korea under the project code PNS132B and also supported by the basic research project at the MOERI/KORDI
PES128B. In addition, the authors would like to thank Professor
Yuichi Murai Hokkaido University for his valuable comments
and kind advices.
References
1 Brenn, G., Braeske, H., and Durst, F., 2002, Investigation of the Unsteady
Two Phase Flow With Small Bubbles in a Model Bubble Column Using PhaseDoppler Anemometry, Chem. Eng. Sci., 57, pp. 51435159.
2 Lane, G. L., Schwarz, M. P., and Evans, G. M., 2005, Numerical Modeling of
Gas-Liquid Flow in Stirred Tanks, Chem. Eng. Sci., 60, pp. 22032214.
3 Guet, S., Ooms, G., Oliemans, R. V. A., and Mudde, R. F., 2004, Bubble Size
Effect on Low Liquid Input Drift-Flux Parameters, Chem. Eng. Sci., 59, pp.
33153329.
4 Lindken, R., and Merzkirch, W., 2002, A Novel PIV Technique for Measurements in Multiphase Flows and Its Application to Two-Phase Bubbly Flows,
Exp. Fluids, 33, pp. 814825.
5 Jansen, P. C. M., 1986, Laboratory Observation of the Kinematics in the
Aerated Region of Breaking Waves, Coastal Eng., 9, pp. 453477.
6 Govender, K., Mocke, G. P., and Alport, M. J., 2002, Video-Imaged Surf
Zone Wave and Roller Structures and Flow Fields, J. Geophys. Res., 107, p.
3072.
7 Ryu, Y., Chang, K. A., and Lim, H. J., 2005, Use of Bubble Image Velocimetry for Measurement of Plunging Wave Impinging on Structure and Associated Greenwater, Meas. Sci. Technol., 16, pp. 19451953.
8 Paik, B. G., Kim, K. Y., Ahn, J. W., and Kim, K. S., 2008, Velocity Field
Measurements Using Bubble Tracers in a Cavitation Tunnel, Proceedings of
the 18th International Offshore and Polar Engineering Conference, Vancouver, BC, Canada, Jul. 611, pp. 198203.
9 Raffel, M., Willert, C., and Kompenhans, J., 1998, Particle Image Velocimetry,
Springer, New York.
10 Murai, Y., Oishi, Y., Takeda, Y., and Yamamoto, F., 2006, Turbulent Shear
Stress Profiles in a Bubbly Channel Flow Assessed by Particle Tracking Velocimetry, Exp. Fluids, 41, pp. 343352.
11 Clift, R., Grace, J. R., and Weber, M. E., 1978, Bubbles, Drops and Particles,
Academic, New York.
12 Auton, T. R., 1984, The Dynamics of Bubbles, Drops and Particles in Motion
in Liquids, Ph.D. thesis, University of Cambridge, Cambridge, UK.
13 Kitagawa, A., Hishida, K., and Kodama, Y., 2005, Flow Structure of
Microbubble-Laden Turbulent Channel Flow Measured by PIV Combined
With the Shadow Image Technique, Exp. Fluids, 38, pp. 466475.
14 Moriguchi, Y., and Kato, H., 2002, Influence of Microbubble Diameter and
Distribution on Frictional Resistance Reduction, J. Mar. Sci. Technol., 7, pp.
7985.
15 Risso, F., and Fabre, J., 1998, Oscillations and Breakup of a Bubble Immersed in a Turbulent Field, J. Fluid Mech., 372, pp. 323355.
Yoshiki Yoshida1
Kakuda Space Center,
Japan Aerospace Exploration Agency,
Koganezawa 1, Kimigaya,
Kakuda, Miyagi 981-1525, Japan
e-mail: yoshida.yoshiki@jaxa.jp
Yoshifumi Sasao
Institute of Fluid Science,
Tohoku University,
Katahira 2-1-1, Aoba,
Sendai, Miyagi 980-8577, Japan
Mitsuo Watanabe
Tomoyuki Hashimoto
Kakuda Space Center,
Japan Aerospace Exploration Agency,
Koganezawa 1, Kimigaya,
Kakuda, Miyagi 981-1525, Japan
Yuka Iga
Toshiaki Ikohagi
Institute of Fluid Science,
Tohoku University,
Katahira 2-1-1, Aoba,
Sendai, Miyagi 980-8577, Japan
Thermodynamic Effect on
Rotating Cavitation in an Inducer
Cavitation in cryogenic fluids has a thermodynamic effect because of the thermal imbalance around the cavity. It improves cavitation performances in turbomachines due to the
delay of cavity growth. The relationship between the thermodynamic effect and cavitation
instabilities, however, is still unknown. To investigate the influence of the thermodynamic
effect on rotating cavitation appeared in the turbopump inducer, we conducted experiments in which liquid nitrogen was set at different temperatures (74 K, 78 K, and 83 K)
with a focus on the cavity length. At higher cavitation numbers, supersynchronous rotating cavitation occurred at the critical cavity length of Lc / h 0.5 with a weak thermodynamic effect in terms of the fluctuation of cavity length. In contrast, synchronous rotating
cavitation occurred at the critical cavity length of Lc / h 0.9 1.0 at lower cavitation
numbers. The critical cavitation number shifted to a lower level due to the suppression of
cavity growth by the thermodynamic effect, which appeared significantly with rising
liquid temperature. The unevenness of cavity length under synchronous rotating cavitation was decreased by the thermodynamic effect. Furthermore, we confirmed that the fluid
force acting on the inducer notably increased under conditions of rotating cavitation, but
that the amplitude of the shaft vibration depended on the degree of the unevenness of the
cavity length through the thermodynamic effect. DOI: 10.1115/1.3192135
Keywords: thermodynamic effect, cavitating inducer, rotating cavitation, cavity length,
fluid force
Introduction
In rocket engines, the turbopump inducer, which is actually a

high-precision high-speed sophisticated rotating impeller, is one
of the key components of a liquid propellant rocket engine. The
inducer is installed upstream of the main impeller in the turbopump to achieve high suction performance. Rotating cavitation
in inducers is known as one type of cavitation instability, in which
an uneven cavity pattern propagates in the same direction as the
rotor with a propagating speed ratio of 1.01.2. Unfortunately, this
rotating cavitation generates shaft vibration arising from to the
increase of the unsteady lateral load on the inducer 1,2.
On the other hand, the thermodynamic effect on cavitation is a
favorable phenomenon in cryogenic fluids. It suppresses the
growth of a cavity, since thermal imbalance appears around the
cavity due to heat transfer for evaporation. Because of this heat
transfer, saturated vapor pressure in the cavity decreases and the
degree of cavity growth becomes smaller than that in water without the thermodynamic effect. Therefore, cavitation performance
of the inducer in cryogenic fluids is improved.
However, the relationship between the thermodynamic effect
and cavitation instabilities is poorly understood. To clarify this
relationship, studies using refrigerant R114, hot water, and liquid
nitrogen have been conducted by Franc et al. 3, Cervone et al.
1
Contributing author.
JOURNAL OF FLUIDS ENGINEERING. Manuscript received April 14, 2008; final manuscript received June 21, 2009; published online August 18, 2009. Review conducted
by Joseph Katz. Paper presented at the 2007 ASME Fluids Engineering Division
Summer Meeting and Exhibition FEDSM2007, July 30August 2, 2007, San Diego, CA.
4, and Yoshida et al. 5, respectively. In the present study, we

investigated the relationship between the thermodynamic effect
and supersynchronous/synchronous rotating cavitation with a focus on the cavity length as an indication of cavitation. We conducted experiments with liquid nitrogen set at different temperatures 74 K, 78 K, and 83 K in order to confirm the dependence
of the thermodynamic effect on the temperature.
Experimental Facility and Apparatus
2.1 Experimental Facility. Experiments were conducted in

the Cryogenic Inducer Test Facility CITF at Kakuda Space Center KSC of Japan Aerospace Exploration Agency JAXA. Figure 1 shows a schematic of the test facility. This facility has two
tanks, one upstream run-tank and one downstream catch-tank,
and the temperature of the liquid nitrogen can be varied by regulating the pressure in the run-tank. Thanks to this capability, liquid
nitrogen set at different temperatures 74 K, 78 K, and 83 K can
be used as working fluid. Figure 2 shows the schematic diagram
of the test inducer showing the installed pressure sensors and shaft
displacement sensor. The inducer used in the present experiment
has the same dimensions and geometric configurations as that in
the previous experiment 5. It has three blades with sweep cutback at the leading edge, its solidity is about 2.1 at the tip. The
inducer is driven by an electric motor at a rotation speed of 18,300
rpm, equal to that of an actual turbopump, and the flow rate is set
at Q / Qd = 1.06 Qd: designed flow rate. In the previous study 5,
test flow rate was set at Q / Qd = 1.00. Only synchronous rotating
cavitation was observed, but supersynchronous rotating cavitation
super-SRC did not appear at the flow rate of Q / Qd = 1.00. In
addition, thermodynamic effect had scarcely any influence on the
unevenness of the cavity length under the synchronous rotating
SEPTEMBER 2009, Vol. 131 / 091302-1
Fig. 1 Cryogenic Inducer Test Facility of JAXA
cavitation. To the contrary, in the present experiments at the flow

rate of Q / Qd = 1.06, both supersynchronous and synchronous rotating cavitation were observed. And the thermodynamic effect
affects the unevenness of the cavity length under the synchronous
rotating cavitation. It could even be said that the flow rate of
inducer has considerable influence concerning the occurrence of
supersynchronous/synchronous rotating cavitation and thermodynamic effect on them.
2.2 Experimental Apparatus. To estimate the cavitating
state of the inducer indirectly, eight pressure sensors Pos. 18
are installed along the blade from the leading edge to the trailing
edge on the casing, as shown in Figs. 2 and 3. The sensor is a
charge mode pressure sensor PCB Piezotronics Inc., model
WH113A with high resistance to shock 20,000 g, a wide range
of the temperature for cryogenic fluid 33477 K, linearity
1%FS, and high resonance frequency 250 kHz. Uncertainty
in / speed ratio of rotating cavitation measured by the pressure sensors is 0.005. The pressure sensors detect the cavitation
near the blade tip, i.e., tip leakage vortex cavitation. However,
sensors are unable to detect the cavitation in the backflow upstream of the inducer, i.e., backflow vortex cavitation.
The methodology to observe the cavitating state in the inducer
Fig. 4 a Estimated cavity region under supersynchronous

rotating cavitation and b estimated cavity region under synchronous rotating cavitation
was basically the same as that mentioned in the previous experiment 5. When cavitation develops on the suction side of the
blade at the tip, the pressure sensor shows a particular wave form
in which a flat region appears at the bottom. This region extends
from the suction side to the pressure side within the interblade the
cavitation number decreases. The pressure of this region can be
considered to be the vapor pressure due to the cavitation. Thus, we
judged this domain as the cavity region.
Figure 4 is drawn by the rearrangement of the wave forms of all
the sensors. The dark area in these figures shows estimated cavity
region. Using these figures at each cavitation number, the cavity
length as an indication of cavitation can be observed by this indirect visualization. In a previous work 6, results of this indirect
visualization using the pressure sensors agreed with the direct
optical observation results in water. Uncertainty in Lc / h nondimensional cavitation length measured by these pressure sensors
is 0.03.
The shaft displacement sensor is installed at the backend of the
rotor, as shown in Fig. 2. The sensor for shaft vibration should be
installed near the inducer. However, there is no space around the
inducer due to the limitation of the pump structure
Rotating cavitation can be divided into two patterns based on
the aspect of cavity fluctuations. Figure 4a indicates the cavity
behavior for typical supersynchronous rotating cavitation, and
Fig. 4b indicates typical synchronous rotating cavitation. Unevenness of the cavity length is observed in both figures. The
longer/shorter cavity propagates from blade to blade on the sequence of rotation under supersynchronous rotating cavitation in
Fig. 4a. In contrast, the cavity length on each blade is uneven
but not unsteady under synchronous rotating cavitation in Fig.
4b. It is steady with no propagation during the sequence of rotation.
3
Fig. 2 Schematic diagram of the test inducer showing the installed pressure sensors and shaft displacement sensor
Fig. 3 Development view of the inducer showing location of

pressure sensors along the inducer blade
091302-2 / Vol. 131, SEPTEMBER 2009
Rotating Cavitation and Thermodynamic Effect
3.1 Rotating Cavitation. As we have already mentioned, rotating cavitation is cavitation instability and has two patterns. Figure 5 shows examples of the cavitation state under conditions of
supersynchronous Fig. 5a and synchronous rotating cavitation
Fig. 5b by direct optical observation in water experiments 7.
In Fig. 5a, the unevenness of cavity length medium, short and
long on each blade blades 13 can be clearly observed. However, supersynchronous rotating cavitation is an unsteady phenomenon in which the asymmetric pattern of the cavity length
propagates from blade to blade at a speed 1.11.2 times higher
than that of inducer rotation. In contrast, the unevenness of cavity
length long, long and short on each blade blades 13 can be
also observed in Fig. 5b. Though the cavitation state in the inducer is surely asymmetric and uneven, it does not change
from blade to blade in rotation. From that viewpoint, synchronous
Fig. 6 Variation in the thermodynamic function T of nitrogen and water
RR + 3 R 2 + R t = C p +
2
2
where
=
Fig. 5 a Photographs of supersynchronous rotating cavitation in water from Ref. 7. b Photographs of synchronous
rotating cavitation in water from Ref. 7.
rotating cavitation can be defined as a steady anomalous phenomenon. These direct optical observations in water agree with
the indirect observations in liquid nitrogen shown in Figs. 4a
and 4b.
3.2 Thermodynamic Effect. The introduction of the temperature depression into the simple RayleighPlesset equation for
a spherical cavity yields 3
3
pvT p
RR + R2 + Rt =
2
l
The third term on the left is the thermal term caused by the
thermodynamic effect. Thermodynamic function , which depends only on temperature T, was originally introduced by Brennen 8. The dimension of thermodynamic function is in m / s3/2
T =
vL2
l2C plTl
C
U3
When the nondimensional thermodynamic parameter

= C / U31/2 is larger, the thermodynamic effect becomes stronger and the cavity growth is suppressed. Watanabe et al. 9 analyzed the influence of nondimensional thermodynamic parameter
on cavity growth and cavitation instabilities in 2D-inducers.
The trend of their analyzed results agreed well with 3D-inducer
results reported by Franc et al. 3.
Figure 6 shows variations in thermodynamic function for
nitrogen and water against the normalized temperature T / Tc,
where Tc is the temperature of the critical point. In addition,
points for experimental conditions 74 K , 78 K , and 83 K
and water at room temperature 300 K are also plotted
in this figure. Comparisons of nondimensional thermodynamic parameter = C / U31/2 of the experimental conditions are presented in Table 1. The range of is 4.3 at 74 K26.6 at 83 K.
Thus, the thermodynamic effect is expected to be larger in case of
rising temperature at 83 K than that of falling temperature at 74 K.
Experimental Results
4.1 Head Coefficient and Cavity Length. We defined the

length from the leading edge to the trailing edge of the estimated
cavity region as the cavity length Lc. In other words, the length
of the dark region along the blade in Fig. 4 is the cavity length.
The diamond-shaped symbols in Fig. 7 show variations in
nondimensional cavity length Lc / h normalized by blade spacing
h for each channel of the inducer i.e., three channels for the
three-bladed inducer. The circles in Fig. 7 show variations in
the inducer head / 0 normalized by the reference head coefficient 0. This figure contains results of two typical temperature
conditions 74 K and 83 K . In addition, the triangle-shaped
Table 1 Comparison of the thermodynamic parameter
Equation 1 is made nondimensional using a chord C, inducer tip

speed U, pressure coefficient C p and cavitation number . The
following equation is obtained Franc et al. 3:
Nitrogen
Temperature
K
Rotation N
rpm
m / s3/2
= C / U31/2
74
78
83
18,300
18,300
18,300
1.4 10+04
3.3 10+04
8.6 10+04
4.3
10.2
26.6
SEPTEMBER 2009, Vol. 131 / 091302-3
Fig. 7 Cavitation performances and cavity length of each

channel top, and variations in dominant frequency of shaft
vibration bottom uncertainty in / 0 = 0.01, / 0 = 0.02,
Lc / h = 0.03, and / = 0.005
symbols , in Fig. 7 show the amplitude of the fluctuating

cavity length peak to peak under the supersynchronous rotating
cavitation, and cross-shaped symbols show the time-averaged
cavity length.
As the cavitation number decreases, supersynchronous rotating
cavitation occurs at a cavity length of about Lc / h 0.5. Cavity
length fluctuates on three blades with a certain phase lag. As a
result, it can be seen that unevenness of cavity length propagates
in the same direction as the rotor with a speed ratio of /
= 1.1 1.2 analyzed by fast Fourier transform FFT, as shown at
the bottom of Fig. 7. This aspect is similar to the visualization in
the photographs in Fig. 5a. Meanwhile, the averaged cavity
length increases with decreasing cavitation number. The occurrence range of supersynchronous rotating cavitation at 83 K shifts
only a few cavitation numbers lower than that at 74 K due to the

thermodynamic effect. The amplitude of cavity fluctuation and
propagation speed ratio of the condition at 83 K are almost the
same as those at 74 K. Watanabe et al. 9 analyzed the rotating
cavitation using the singularity method with Katos heat transfer
model 10. They reported that the propagation speed ratio of
supersynchronous rotating cavitation becomes smaller due to the
thermodynamic effect. On the other hand, Zoladz 11 observed
from the comparison between the experimental results in liquid
oxygen and in water that the thermodynamic effect affects the
normalized amplitude of pressure fluctuations caused by rotating
cavitation which becomes larger in liquid oxygen than in water,
but it does not affect the frequency of rotational speed of the
cavity. Unfortunately, a full explanation of the difference between
the present results and Watanabes analytical results/Zoladzs experimental results has not been obtained in the present experiments. In order to explain, an extension of the experiment is underway in liquid nitrogen at several temperatures and flow rates.
In contrast, synchronous rotating cavitations occur at cavity
length of Lc / h 0.9 at 74 K and Lc / h 1.0 at 83 K, and disappears at a cavity length of Lc / h 1.5 in both temperature conditions. Under synchronous rotating cavitation, unevenness of the
cavity length is manifested as three lengths short, medium or
long, and long, similar to the photographs in Fig. 5b. The
degree of unevenness of the condition at 83 K is considerably less
than that at 74 K, and beginning/closing cavitation numbers of
synchronous rotating cavitation at 83 K shift to a lower point than
those at 74 K due to the thermodynamic effect.
In addition, the circles in Fig. 7 show the variations in the
inducer head. Within the range of the occurrence of supersynchronous rotating cavitation, the inducer head has no effect on this
cavitation instability. However, the inducer head temporarily decreases within the range of the occurrence of synchronous rotating
cavitation only at the condition of 74 K. After breaking away from
the synchronous rotating cavitation, cavity lengths become completely equal near Lc / h = 1.5. Hereupon, inducer head drop begins
to gradually occur for both temperature conditions, and the head
drop is nearly equal at the same cavity length. This means that the
inducer head is strongly dependent only on the cavity length. Fur-
Fig. 8 a Fluctuations of cavity lengths for each channel in the case of supersynchronous
rotating cavitation left: 74 K, right: 83 K uncertainty in Lc / h = 0.03 and b fluctuations of
cavity lengths for each channel in the case of synchronous rotating cavitation left: 74 K, right:
83 K uncertainty in Lc / h = 0.03
091302-4 / Vol. 131, SEPTEMBER 2009
Fig. 9 Influence of temperature on cavity length and the region where the rotating cavitations occurs at 74 K, 78 K, and 83
K uncertainty in / 0 = 0.02 and Lc / h = 0.03
thermore, as mentioned above, the onset of cavitation instability is

also governed only by the cavity length through the liquid temperature, namely, the degree of the thermodynamic effect.
4.2 Behavior of Supersynchronous/Synchronous Rotating
Cavitation. Figure 8 shows the fluctuations of the cavity length in
each channel during supersynchronous Fig. 8a/synchronous
Fig. 8b rotating cavitation against the inducer rotation. For
supersynchronous rotating cavitation in Fig. 8a, the cavity
length fluctuates within Lc / h = 0.4 1.2 with 2 / 3 phase difference. The cavity propagates from blade 3 to blade 2, and finally to
blade 1. After about eight rotations, the uneven pattern of cavity
returns to the same situation. This means that the cavity propagates forward with a propagation speed ratio / = 1.13. This
result agrees with the FFT analysis of the measurement of pressure fluctuations in Fig. 7. Comparing the conditions at 74 K and
83 K, not much difference between the behavior of the cavity
fluctuation and that of the oscillating frequency is observed. From
these results, it is inferred that the thermodynamic effect does not
affect the behavior of cavity fluctuation during supersynchronous
rotating cavitation at the higher cavitation numbers.
On the other hand, Fig. 8b shows that the cavity length for
each channel exhibits breaking symmetry, such as supersynchronous rotating cavitation, but that unevenness is manifested against
the inducer rotation during synchronous rotating cavitation. The
cavity length of each channel is relatively steady. However, the
differences in the unevenness of the cavity lengths short, medium
or long, and long are found between liquid temperatures of 74
K and 83 K. Although the unevenness of cavity length at 74 K is
large, i.e., Lc / h = 0.5, 1.5, and 1.7, that at 83 K is considerably
small, i.e., Lc / h = 1.1, 1.3, and 1.5. Thus, it is considered that the
behavior of synchronous rotating cavitation at lower cavitation
numbers is greatly affected by the thermodynamic effect. Although the averaged cavity length under the supersynchronous
rotating cavitation is shorter than the blade spacing, i.e., Lc / h
1.0, and the longer cavity lengths under the synchronous rotating cavitation are longer than the blade spacing, i.e., Lc / h 1.0.
On the other hand, we have observed in the previous work with
another inducer 6 that the temperature depression increases rapidly when the cavity length, Lc / h, becomes longer than 1.1. When
we consider a thermal time for heat transfer, a transit time,
Lc / U, is appropriate for a simple parameter, which indicates the
interval from the bubble inception to the collapse. As the cavity
length in synchronous rotating cavitation is longer than that in
supersynchronous rotating cavitation, the transit time = Lc / U
of synchronous rotating cavitation is considerably longer than that
of supersynchronous rotating cavitation. Thus, it seems that the
amplitude to thermodynamic effect depends on the thermal time
Fig. 10 a Variations in cavity length top, fluid force top,

and shaft vibration bottom for 74 K uncertainty in / 0
= 0.02, Lc / h = 0.03, and F / Fref = 0.03 and b variations in cavity
length top, fluid force top, and shaft vibration bottom for
83 K uncertainty in / 0 = 0.02, Lc / h = 0.03, and F / Fref = 0.03
becomes larger in synchronous rotating cavitation than that in

supersynchronous rotating cavitation. It might be the reason why
the behavior of synchronous rotating cavitation at lower cavitation
numbers is greatly affected by the thermodynamic effect.
It is remarked that thermodynamic effect on cavity unevenness
under synchronous rotating cavitation was not prominent in the
previous experiments at flow rate Q / Qd = 1.00 5, although that is
considerable large in the present experiments at flow rate Q / Qd
= 1.06. When the flow rate decreases in the inducer, the tip leakage
vortex cavitation increases due to the enlargement of the incidence
angle. Also, vortex cavitation in the backflow upstream of the
inducer becomes stronger due to the increase in the circulation at
the inducer inlet 12. These two vortex cavitations interact complexly on each other, and turbulence becomes stronger. Thus, it
could be said that the change in these cavitation aspects through
the decrease in flow rate has the sever influence on the degree of
thermodynamic effect.
SEPTEMBER 2009, Vol. 131 / 091302-5
Fig. 11 a Instant vector orbits of fluid force under typical supersynchronous rotating cavitation at 74 K and 83 K uncertainty in F / Fref = 0.03 and b time-averaged vector orbits of fluid
force during synchronous rotating cavitation at 74 K and 83 K uncertainty in F / Fref = 0.03
Discussions
5.1 Liquid Temperature and Cavity Length. Figure 9

shows the relationships between the temperature of liquid nitrogen
74 K, 78 K, and 83 K and the region of cavitation number where
rotating cavitation occurs, and the cavitation number at a given
cavity length. This figure indicates the dependence of cavitation
number at the occurrence of the rotating cavitations supersynchronous rotating cavitation and synchronous rotating cavitation
and at the given longer cavity length Lc / h = 1.6, 1.7, 1.8, and
30% head break on the temperature. From this figure, it can be
seen that the cavitation number at the onset of rotating cavitation
and a given cavity length become smaller with the increase in the
temperature of liquid nitrogen, since the thermodynamic effect is
strengthened due to the increase in temperature. In addition, focusing on one cavity length, the difference of cavitation number
for three temperatures tends to become smaller with falling temperature. It is considered that the temperature depression in the
cavity i.e., thermodynamic effect decreases with decreasing ther091302-6 / Vol. 131, SEPTEMBER 2009
modynamic function T, and is restricted physically when it

approaches the temperature 63.1 K of triple point of nitrogen
6,13.
5.2 Fluid Force and Shaft Vibration. Figure 10 shows the
variation in cavity length Lc / h, indications of fluid force
F / Fref top, and the component of the supersynchronous superSRC, / = 1.1 1.2/synchronous SRC, / = 1.0 shaft vibration bottom detected by the displacement sensors, for two temperatures 74 K Fig. 10a and 83 K Fig. 10b. Fluid force was
estimated by integrating the pressure distributions around the inducer casing. This estimated force is only an indication to evaluate
the degree of fluid force acting on the inducer. The figure shows
that the fluid force increases significantly during
supersynchronous/synchronous rotating cavitation. Moreover, it
can be found that the shaft vibrations are clearly amplified because of the action of the fluid force caused by supersynchronous/
synchronous rotating cavitation. Thus, it has been considered that
the unbalanced fluid force caused by the rotating cavitation has a

strong influence on the rotor dynamics in turbopumps 1,2,14.
Figure 11 presents the orbit of the fluid force vector under the
typical supersynchronous rotating cavitations Fig. 11a and during the synchronous rotating cavitations Fig. 11b on the rotating coordinate. From this figure, the directions of instant force
vectors under supersynchronous rotating cavitation are seen to
change. The orbit of the force vector is exactly traced on an equilateral triangle due to the three blades. The degree of the equilateral triangle at 74 K Fig. 11a left is almost the same as that at
83 K Fig. 11b right, and the component of the supersynchronous shaft vibration / = 1.1 1.2 at 74 K Fig. 10a is also
the same as that at 83 K Fig. 10b. From these results, it can be
concluded that there is no thermodynamic effect on the degree of
supersynchronous shaft vibration, in which the averaged cavity
length does not reach at the throat Lc / h 1.0.
In contrast, the orbit of the time-averaged force vector during
synchronous rotating cavitation shows the increase in a certain
direction. However, the fluid force at 83 K Fig. 11b right is less
than that at 74 K Fig. 11b left, and the component of the
synchronous shaft vibration / = 1.0 at 83 K Fig. 10b is
also less than that at 74 K Fig. 10a, respectively. Thus, it is
considered that the thermodynamic effect on cavitation suppresses
the synchronous shaft vibration through the decrease in the unevenness of the cavity length during the synchronous rotating
cavitation. It could even be said that the thermodynamic effect on
cavitation has not only a thermal damping effect on the cavity
growth, but also a thermal damping effect on the shaft vibration through the suppression of the cavitation instability at lower
cavitation numbers, in which the cavity length extends over the
throat Lc / h 1.0.
Conclusions
By experiments with liquid nitrogen set at different temperatures and considerations based on the cavity length as an indication of cavitation, the following points were clarified.
1 Rotating cavitation occurs at the same cavity length independent of temperature. The critical cavity length at the
onset of supersynchronous rotating cavitation was Lc / h
0.5, and that at the onset of synchronous rotating cavitation was Lc / h 0.9 1.0 in the present inducer.
2 The thermodynamic effect suppresses the growth of cavity
length. Growth of the cavity is distinctly suppressed with
increasing liquid temperature.
3 Due to the above, the thermodynamic effect shifts the critical cavitation number of the rotating cavitations lower.
4 The differences in the behavior of the cavity fluctuation and
oscillating frequency under supersynchronous rotating
cavitation were not prominent at any of temperatures.
5 However, unevenness of the cavity length under synchronous rotating cavitation was greatly affected by the thermodynamic effect in the present experiments of flow rate
Q / Qd = 1.06.
6 The favorable thermodynamic effect on cavitation instability becomes greater when the cavity length extends over the
throat at lower cavitation numbers.
7 The fluid force acting on the inducer significantly increases
under supersynchronous/synchronous rotating cavitation.
The thermodynamic effect influences considerably the unevenness of the cavity length under synchronous rotating
cavitation. Therefore, synchronous shaft vibration under
synchronous rotating cavitation is much affected by the
temperature of working fluid.
Acknowledgment
Authors would like to thank Mr. Yusuke Kazami of Tohoku
University for his great help with data processing.
Nomenclature
C
Cp
C pl
F
Fref
h
L
Lc
p
pv
R
Tc
T
t
U
l
v
l
chord of inducer
pressure coefficient
specific heat of liquid
fluid force
reference value of fluid force
blade spacing
latent heat
cavity length
pressure
vapor pressure
radius of bubble
critical temperature
ambient temperature
time
tip velocity of inducer
thermal diffusivity of liquid
vapor density
liquid density
thermodynamic function defined by Eq. 2
nondimensional thermodynamic parameter by
Eq. 4
cavitation number
reference value of cavitation number upper
limit of cavitation number in the present
experiment
transit time =Lc / U
head coefficient
reference value of head coefficient
angular velocity of rotor
angular velocity of rotating cavitation
References
1 Rosenmann, W., 1965, Experimental Investigations of Hydrodynamically Induced Shaft Forces With a Three-Bladed Inducer, Proceedings of the Symposium on Cavitation in Fluid Machinery, Winter Annual Meeting, ASME, New
York, pp. 172195.
2 Kamijo, K., Yoshida, M., and Tsujimoto, Y., 1993, Hydraulic and Mechanical
Performance of LE-7 LOX Pump Inducer, AIAA J., 96, pp. 819826.
3 Franc, J. P., Rebattet, C., and Coulon, A., 2004, An Experimental Investigation of Thermal Effects in a Cavitating Inducer, ASME J. Fluids Eng., 126,
pp. 716723.
4 Cervone, A., Testa, R., and dAgostino, L., 2005, Thermal Effects on Cavitation Instabilities in Helical Inducers, AIAA J., 215, pp. 893899.
5 Yoshida, Y., Sasao, Y., Okita, K., Hasegawa, S., Shimagaki, M., and Ikohagi,
T., 2007, Influence of Thermodynamic Effect on Synchronous Rotating Cavitation, ASME J. Fluids Eng., 1297, pp. 871876.
6 Yoshida, Y., Kikuta, K., Hasegawa, S., Shimagaki, M., and Tokumasu, T.,
2007, Thermodynamic Effect on a Cavitating Inducer in Liquid Nitrogen,
ASME J. Fluids Eng., 1293, pp. 273278.
7 Maekawa, Y., 1996, Experimental Study of Unsteady Cavitation on an Inducer, MS thesis, Osaka University, Osaka in Japanese.
8 Brennen, C. E., 1973, The Dynamic Behavior and Compliance of a Stream of
Cavitating Bubbles, ASME J. Fluids Eng., 95, pp. 533541.
9 Watanabe, S., Hidaka, T., Horiguchi, H., Furukawa, A., and Tsujimoto, Y.,
2007, Analysis of Thermodynamic Effects on Cavitation Instabilities,
ASME J. Fluids Eng., 1299, pp. 11231130.
10 Kato, H., 1984, Thermodynamic Effect on Incipient and Development of
Sheet Cavitation, Proceedings of International Symposium on Cavitation Inception New Orleans, LA, Dec. 914, ASME, New York, Vol. 16, pp. 127
136.
11 Zoladz, T., 2000, Observations on Rotating Cavitation and Cavitation Surge
From the Development of the Fastrac Engine Turbopump, Proceedings of the
36th AIAA/ASME/SAE/ASEE Joint Propulsion Conference, Huntsville, AL, Jul.
1719, Paper No. AIAA200-3403.
12 Kimura, T., Yoshisa, Y., Hashimoto, T., and Shimagaki, M., 2008, Numerical
Simulation for Vortex Structure in a Turbopump Inducer: Close Relationship
With Appearance of Cavitation Instabilities, ASME J. Fluids Eng., 130, p.
051104.
13 Watanabe, S., Furukawa, A., and Yoshida, Y., 2008, Theoretical Analysis of
Thermodynamic Effects on Cavitation in Cryogenic Inducer Using Singularity
Method, Proceedings of the 12th International Symposium on Transport Phenomena and Dynamics of Rotating Machinery, Honolulu, HI, Feb. 1722,
Paper No. ISROMAC12-2008-20251.
14 Kobayashi, S., 2006, Effects of Shaft Vibration on Occurrence of Asymmetric
Cavitation in Inducer, JSME Int. J., Ser. B, 494, pp. 12201225.
SEPTEMBER 2009, Vol. 131 / 091302-7
A. K. Das
e-mail: arup@mech.iitkgp.ernet.in
P. K. Das1
Transition of Bubbly Flow in

Vertical Tubes: New Criteria
Through CFD Simulation
e-mail: pkd@mech.iitkgp.ernet.in
IIT Kharagpur,
721302, India
J. R. Thome
LCTM,
EPFL,
Lausanne CH-1015, Switzerland
e-mail: john.thome@epfl.ch
The two fluid model is used to simulate upward gas-liquid bubbly flow through a vertical
conduit. Coalescence and breakup of bubbles have been accounted for by embedding the
population balance technique in the two fluid model. The simulation enables one to track
the axial development of the voidage pattern and the distribution of the bubbles. Thereby
it has been possible to propose a new criterion for the transition from bubbly to slug flow
regime. The transition criteria depend on (i) the breakage and coalescence frequency, (ii)
the bubble volume count below and above the bubble size introduced at the inlet, and (iii)
the bubble count histogram. The prediction based on the present criteria exhibits excellent agreement with the experimental data. It has also been possible to simulate the
transition from bubbly to dispersed bubbly flow at a high liquid flow rate using the same
model. DOI: 10.1115/1.3203205
Keywords: bubbly flow, population balance, transition, coalescence, breakage
Introduction
Bubbly flow may be defined as a typical two phase flow pattern

where a gaseous phase is dispersed in a liquid phase in the form of
spherical, oblate spheroid, and small cap bubbles. Though bubbly
flow often exhibits an apparently homogeneous structure, the hydrodynamics is sufficiently complex due to different velocities of
two phases and dynamic processes such as coalescence and break
up of bubbles.
There are numerous examples of bubbly flow covering a diverse range of applications. Chemical and biotechnological processes, operations for environmental pollution control, power generation, and cooling are only a few examples of such flow.
Particularly in process applications, bubbly flow is often desirable
as it is characterized by a thorough mixing of the dispersed phase
in the carrier fluid. The large interface also provides improved rate
of reaction and mass transfer. During gas-liquid two phase flow
through a pipe line, bubbly flow is one of the common flow regimes, which may occur both in adiabatic and nonadiabatic situations. The designer should not only require the knowledge of the
hydrodynamics, he should also have an adequate idea regarding
the operating range of bubbly flow depending on the process parameters, flow geometry, and fluid properties.
Over the decades, untiring efforts have been made for the simulation of bubbly flow. Initially a homogeneous equilibrium model
HEM was used in an attempt to treat bubbly flow as a single
fluid whose properties were determined through different averaging rules. Drift flux model 1 can be considered as the next higher
level model where the importance of slip between the phases was
recognized. Various modifications 24 of this model have been
proposed time to time in order to make it applicable for complex
systems.
As the properties of the two phases are widely different and the
slip between the phases could be considerable, it was appreciated
that separate set of conservation equations are required for each of
the phases. Additionally, the interaction between the phases also
needs to be modeled. This gave rise to the methodology of the two
1
JOURNAL OF FLUIDS ENGINEERING. Manuscript received January 13, 2009; final manuscript received July 4, 2009; published online August 18, 2009. Assoc. Editor: Theodore Heindel.
fluid model 5. There have been many variations of this technique. An important idealization has been made by the researchers
particularly for the cases when one phase is thoroughly dispersed
in the other. This is known as interpenetrating continua or mixed
continua 6. According to the hypothesis of the interpenetrating
continua, both the phases are simultaneously present at each and
every point of the computational domain, while their influence on
the hydrodynamics and the transport processes is given by the
local phase fraction.
In multiphase flow, the phase distribution changes both with
time and space even if the input and the operating conditions are
kept invariant. Though the two fluid model is a considerable success toward the modeling of multiphase flow, it needs augmentation to take care of the variable phase distribution. Several methodologies have been suggested for this. The interfacial area
transport equation was suggested by Ishii 7. In 1998 Wu et al.
8 developed one grouped interfacial area transport equation for
spherical bubbly flow pattern. Further, it is realized that the shape
of the bubbles distorts during motion causing a change in the drag
force and the fluid particle interaction mechanisms. Two-group
spherical/distorted bubble and cap/churn turbulent bubble area
transport equation was developed 9 to circumvent this.
An alternate approach of simulation through population balance
model PBM 10 has also been adopted. PBM essentially considers the secondary phase as the assemblage of discrete class or
subgroup having different characteristics lengths. Additionally,
PBM can meticulously track the birth and death of the members
of these subgroups. Various methodologies, such as method of
discrete classes 10, quadrature method of moments 11, least
square methods 12, etc., have been developed to implement
PBM numerically.
As this model has a wide applicability with significant accuracy, a number of researchers 5,8,13 have adopted the technique
for the simulation of bubbly flow with different modifications.
Chen et al. 14 showed the applicability of the model in case of
bubble columns of different diameters. Similarly, Yeoh and Tu
15 and Cheung et al. 16 solved the population balance equation along with the two fluid model using the method of discrete
classes 10 for pipe flow. An extensive review of the works related to population balance modeling for bubbly flows in the
bubble column can be found in Ref. 17, while Krepper et al. 18
reviewed the work for pipe flow.
SEPTEMBER 2009, Vol. 131 / 091303-1
Furthermore, for a better prediction of the assemblage of different sized entity multiple-sized group models MUSIGs has also
been tried by different researchers 1820. In these models, different groups based on size of bubbles are assumed to have
different velocities. A separate set of conservation equations are
solved for each group of the bubbles, considering each one of
them as one distinct continuum. However, the increase in computation time and the need for a large number of closure relationships cannot be overlooked in this type of simulations.
Efforts have also been made to directly solve the two phase
interacting momentum balance equation by using sufficiently fine
grids and smaller time steps in the problem domain. Esmaelli and
Tryggvason 21 applied direct numerical simulation DNS for
laminar bubbly flow in a vertical channel. After that, continuous
efforts 2224 have been made to improve the numerical scheme
for a better prediction. Nevertheless, due to the volume of computations involved, only small systems with fewer complexities
can be modeled at present 25. As a consequence, models based
on averaging both in time and space, such as the two fluid
model, are very important for simulating problems of practical
importance.
It is clear from the available literature that the application of a
two fluid model along with PBM or interfacial area transport
equation has produced reasonable success in the prediction of
bubbly flow. However, to date not much effort has been made to
predict the transition of cocurrent bubbly flow through a circular
conduit using computational fluid dynamics CFD. On the other
hand, CFD is rigorously used by researchers 2629 to determine
the transition of homogeneous to heterogeneous regime for bubble
column reactors. This motivated us for the numerical investigations of transition regimes of bubbly flow situations. Bubbly flow
through a vertical tube is bound by two neighboring flow regimes.
At the low and moderate liquid flow rate, bubbly flow transforms
into slug flow with the increase in air flow rate, whereas transition
from bubbly to dispersed bubbly flow is observed at high liquid
flow rates 30. Classically, these transition criteria have been derived from the control volume approach using a number of simplifying assumptions 31,32. From these criteria of transition are
predicted based on the phase superficial velocities without taking
other process parameters such as tube diameter, inlet bubble size
distribution, etc. into cognition. Furthermore, a considerable
variation has been reported in the transition boundaries observed
in different experiments. This has motivated the present work to
apply CFD technique to identify the suitable transition criteria,
which are rigorously based on hydrodynamics. The bubbly flow
has been simulated using the two fluid model along with the population balance technique. Though the basic structure of the model
has been retained, a few important modifications have been made
to simplify the computation and to improve the simulation. The
model simulation has been used to investigate the dynamic development of coalescence and to break up processes along the axial
direction. This has been studied noting the breakage and coalescence frequency, bubble volume account below and above the
bubble size at the inlet, and bubble count histogram. It has been
possible to propose unique transition criteria from bubbly to slug
flow. The transition boundary predicted by the present model
matches very well with the experimental data. Finally, the present
model has been extended to the zone of high liquid flow rate to
predict the transition to dispersed bubbly flow. Experimentally
observed bubbly to dispersed bubbly transition is predicted satisfactorily.
1.1 Model Development. Figure 1 depicts the typical up flow
of gas-liquid mixture with low superficial velocities, where the
lighter phase is dispersed in the form of bubbles of different sizes
in a round vertical conduit. The constitutive equations of each
phase are developed based on the volume average properties of
the phase. The total population of bubbles is discretized into subgroups based on volume to avoid intermediate bubble generation
during coalescence. In the present model each group of bubbles
091303-2 / Vol. 131, SEPTEMBER 2009
Fig. 1 Schematic representation of bubbly flow
has its characteristic length and occupies a fixed space at any

instant. At any location, the number of bubbles in each group can
be individually estimated considering their breakage and coalescence with other bubbles.
A Eulerian technique, known as the two fluid model 33, has
been used for simulating the hydrodynamics. In this model, two
competing phases are considered as interpenetrating continua by
introducing the instantaneous phase fraction i. The approach
gives the freedom to use coarser nodes and longer time steps
compared with the EulerianLagrangian approach.
Isothermal bubbly flow in the absence of any phase change can
be described by the conservation of mass and momentum in each
of the phases. Furthermore, cocurrent vertical upflow can be considered to be axisymmetric. The conservation equations in the r-z
coordinate system are given below based on the formalism of two
fluid model.
For continuity equation,
1
riiui + iiwi = 0
i i +
t
r r
z
For r momentum,
+ i ig r
iiui + iiui2 + iiuiwi = i
r
t
r
z
+ i
2
2
i iu i i
iu i + i 2 iu i
+
2 iu i +
2
r
z
r
r r
FWir FLGr Fdispr
For z momentum,
+ i ig z
iiwi + iiuiwi + iiwi2 = i
z
t
r
z
+ i
2
2
i
iw i + i 2 iw i
2 iw i +
r
z
r r
FWiz FLGz Fdispz
Here i = l stands for liquid phase, and i = g stands for gas phase.
P is assumed as the average pressure of the mixture, and it can be
evaluated using the equation of state in the following manner:
P = ll + ggRT
Constitutive relationships for the force terms used in the momentum equations are selected from the published literature. As the
generic nature of the radial and axial forces is the same, directions
of the forces are omitted in the respective equations Eqs.
510. For the vector quantities associated with the forces, we
have used their nondirectional forms by putting a bar over the

symbols of the respective vectors. For the directional equation of
the forces the quantities under the bar will be substituted by the
respective directional form of the vectors.
FLG is the interfacial drag force that can be expressed as 34
FLG =
2CFL
gug ulug ul + lug ug ul
r
D
2
The first term of the force represents the drag force, and the second term is the apparent mass force due to the change in kinetic
energy induced by the bubble motion in the liquid.
In the above expression CFL is the interfacial friction factor and
is given by
CFL = CD1 1.7
l D
g DB
CD is drag coefficient for a single bubble, and DB is average

bubble diameter at a particular location. This is determined
through the population balance technique and constitutes one of
the interfaces between the two fluid model and population balance
technique.
Based on Chisholms 35 correlation wall liquid friction force
FWL Eqs. 2 and 3 can be expressed as
FWl = 1 + Y 2c 1BXc1 Xc2n/2 + X2n
c PL0
Here, PL0 is the single phase pressure drop due to friction and
n = 0.25 for air water system. In Eq. 7 the empirical constant B is
dependent on the fluid pair associated. It can be calculated as
follows:
B=
where
C=
ul
ug
CY c 22n + 2
Y 2c 1
u2gg
l
1 2
g
ul l
For various combinations of superficial velocities within the

reported range, it varies between 11,000 and 14,000 for air water
combination. For calculation of the values of B in the present
model, an average of the possible values of B has been considered, which came out near to 12,500. Xc and Y c can be expressed
as follows:
1
Xc =
1+
1 l ul
g ug
and
Yc =

f G0l
f L0g
0.5
The wall friction factor for the gaseous phase is also calculated
using Eq. 7 after employing the gaseous phase frictional pressure drop PG0 as the multiplier in place of PL0.
The turbulent dispersion force of gaseous phase is considered in
the form of Favre averaged variables 36
Fdisp = CTDCD
tg l g
Sctg l
g
10
CTD and CD are the turbulent dispersion force coefficient and

the drag force coefficient for a single bubble. CTD is taken as 1.0
in the present simulation. tg is the turbulence kinetic viscosity,
and Sctg is the turbulent Schmidt number of the gaseous phase.
A population balance 37 has been incorporated in the present
model to keep account of bubble evolution in the bulk of continuous phase due to coalescence and breakup. As a result of these
processes, bubbles of new groups may appear birth and those of
existing groups may disappear death. Accordingly, the population balance can be obtained in terms of the birth and death rate
Fig. 2 Steps of the breakage and coalescence process
nr,z,t;d
nr,z,t;d
nr,z,t;d
+ wg
= BBr,z,t;d
+ ug
r
z
t
DBr,z,t;d + BCr,z,t;d DCr,z,t;d
11
In the present model, the entire population of bubble is discretized into different subgroups of equal volume interval. Each
subgroup has its own span of bubble characteristic length. The
mean of this span is considered to be the pivotal length scale
associated with the subgroup. As the bubbles are equispaced in
volume, a newly born bubble during coalescence explicitly falls in
one of the subgroup nodes. This exactly satisfies the conservation
of bubble volume. On the other hand, during breakage the daughter bubbles may not have a size corresponding to the pivotal
length scale. Such bubbles are distributed among the neighboring
subgroups, keeping the bubble mass and number fixed. Nucleation
of new bubbles, evaporation, condensation, shrinkage, and elongation are not considered in the present model. Present model
only recognizes the appearance of newly born bubbles and the
disappearance of the extinct bubbles due to dynamic interactions
in a fixed control volume. However, their redistribution is considered to be random and solely guided by the advection of two
phase mixture. Furthermore, only binary coalescence and binary
breakage of the bubbles are considered.
Collision with turbulent eddies are considered as the mechanism of breakup and is schematically shown in Fig. 2a. Breakage of a bubble is possible only when the turbulence kinetic energy of the striking eddy supersedes the effect of surface energy of
the interacting bubble 38. This leads to the surface rupture of the
bubbles and the formation of daughter bubbles. Moreover,
breakup occurs when an eddy of size comparable to the bubble
characteristic length collides with it. If the eddy size is larger than
the bubble characteristic length, collision between them is unable
to make any craters on the bubble. Thus, in Fig. 2a only eddies
denoted by 1, 2, and 3 are effective for the breakup process. The
size of the daughter bubbles due to breakage depends on the
strength of the eddies 39, which, in turn, depend on the local
hydrodynamics.
Birth and death of bubbles due to the breakup process can be
calculated as follows:
BBr,z,t;d =
d d,ddgdnr,z,t;ddd 12
DBr,z,t;d = nr,z,t;dgd
13
Based on the model proposed by Kostoglou and Karabelas 40

bubble breakup frequency can be expressed as
SEPTEMBER 2009, Vol. 131 / 091303-3
gd = k1
2
d
1/3
min
1 + 2 12c /2.04 2/3d5/311/3

1
c
e
ddf v
11/3
14
Here f v is the ratio of the volume of mother and daughter bubbles,

and is the energy dissipation rate per unit mass. can be written
in the following fashion 41, in terms of pipe diameter D and
mixture velocity Um:
=
3
f vU m
2D
15
Um is calculated as follows:
Um = u2g + w2g2 + ul2 + wl22
16
C1 is an empirical constant and can be expressed in terms of f v as

follows:
C1 = f v2/3 + 1 f v2/3 1
3/5
dmin = 1.15 3/5 2/5
l
18
For bubbles of size larger than a critical value, instantaneous

break up may take place because of the instability of the gasliquid interface. According to Carrica and Clausse 44 the probability is given by
d dcm
19
d dcm + dm
c
where dc is the critical bubble diameter, set as 27 mm, and b and

m are model parameters set as 100 s1 and 6.0, respectively 44.
Daughter distribution probability d1 , d2 proposed by Kostoglou and Karabelas 45 is used in the present model.
d1,d2 =
1
1
2z 1 6I
+
+
d1
b + 0.5 d32
d1
+a 1 +b
d2
d2
20
where I is the normalization coefficient and is expressed as

I=
0.5
z1
ln1 + a lnb +
b + 0.5
and
z=
a
4b1 + b1 a
a and b are parameters that define the shape of the daughter drop
size distribution function. In the present model, the values a
= 0.1 and b = 1 45 are used that signifies U shaped bubble size
distribution.
For coalescence of bubbles to occur in the turbulent flow field
of a gas-liquid mixture, the bubbles must first collide with each
other and then remain in contact for sufficient time so that the
processes of liquid film drainage, film rupture, and finally coalescence may occur. After the collision between two bubbles, a successful coalescence through film rupture is depicted in Fig. 2b.
The following expressions can be taken for the birth and death
due to coalescence process, respectively
BCr,z,t;d =
1
2
vd/2
dvv,dvnr,z,t;dvvnr,z,t;dvdv
21
091303-4 / Vol. 131, SEPTEMBER 2009
dv,dvnr,z,t;dvdv 22
Binary coalescence of two equal sized bubbles generates a

daughter bubble having 1.26 times diameter of its parents. Therefore, two bubbles having the dimension greater than 0.8D D is
tube diameter will result a coalesced bubble of diameter greater
than D. This obviously goes beyond the computational domain.
Further, it has been reported 46 that bubbles with diameter 0.6D
represent a transition to cap bubble or a Taylor bubble. With all
these considerations, the maximum bubble diameter is restricted
up to the half of the diameter of the pipe through which the flow
occurs to attain transition between bubble and slug flow.
Coalescence frequency d1 , d2 is the product of the effective
swept volume rate, hd1 , d2 and film rupture efficiency pd1 , d2.
Effective swept volume rate is calculated using the analogy between kinetic theory of gasses and bubble coalescence phenomena. According to Coulaloglou and Tavlarides 47, effective
swept volume rate is calculated as:
17
min is the nondimensional minimum daughter bubble size that can

be obtained from the Kolmogroff microscale, as described by
Tsouris and Tavlarides 42.
The minimum size of the bubble that can be fragmented by a
colliding eddy of strength is calculated as follows using the
conjecture of Troshko and Zdravistch 43:
gd = b
DCr,z,t;d = nr,z,t;dv
hd1,d2 = c2
2/3 1/2
d1 + d22d2/3
1 + d2
1+
23
Film drainage is controlled by inertia and surface tension forces

48. For two bubbles of diameter d1 and d2 Coulaloglou and
Tavlarides 47 proposed the film rupture efficiency as follows:
pd1,d2 = exp
d 1d 2
c 3 l l
21 + 3 d1 + d2
24
The numerical constants C2 and C3 are related to the collision

frequency and the rupture efficiency, respectively. Values of C2
and C3 are taken as 0.00551.3404 and 5.4 108, as referred by
Ioannou et al. 41. They used the experimental data of Lovick
49 to fit the values of the above constants.
Finally, we like to make a couple of comments regarding the
present simulation. First, no rigorous turbulence closure has been
used. Taking the queue from the single phase flow, number of
researchers has used the k- formulation for turbulence modeling
even in case of two phase flow 50,51. The empirical constants
needed for such formulation were determined from single phase
experiments though there is no guarantee that they are applicable
also for the two phase hydrodynamics 52. We have used a rather
simple relationship for turbulent energy dissipation Eq. 15. It
goes without saying that a better physical model may improve the
prediction substantially.
Second, for the sake of simplicity, we have assumed identical
velocities for bubbles of all the subgroups. Strictly speaking this is
a simplification. For a better prediction, different groups of
bubbles may be assigned different characteristic velocities and
closure laws for drag, lift, etc. 53. Within the same framework of
population balance model the above may be implemented.
Boundary Condition and Solution Procedure
The following boundary conditions are employed for the

present simulation.
i.
For both the phases, at the inlet uniform velocity profile in

stream wise direction and zero velocity at the crosswise
direction is assumed. No slip condition is prescribed at the
wall of the conduit.
ii. Void fraction of gaseous phase and pressure at the inlet
are specified.
iii. For a better simulation of the flow field finer node have
been used near the wall. A small velocity as per the logarithmic law of buffer layer is implemented at a very small
distance from the conduit wall 54.
iv. Uniform distribution of equal sized bubble is considered at
the inlet whereas initially t = 0 the conduit is filled up
only with the carrier phase
Fig. 4 Effect of radial mesh refinement on void distribution
Fig. 3 Solution methodology of the proposed model
nr,0,t;d =
Nd,node for d = din

0
for d din
25
and
nr,z,0;d = 0
v.
26
Validation of the Model
Solution of the model based on the algorithm shown in Fig. 3

gives the phase velocity and void fraction at any particular point
of conduit. Computations have been done selecting different mesh
sizes and time steps to check the node independence and to find
out optimal values of node spacing and time step. Results on the
effect of mesh size on void distribution are depicted in Fig. 4.
From this study it is decided that 20 radial nodes will be sufficient
for a compromise between the accuracy and the computational
time. For the flow field to become fully developed total length of
the pipe is assumed to be a minimum of 100D. Grid independent
studies have also been made for the axial direction, and 1000
nodes were found to be sufficient for that. Total number of cells
then turns out to be 18,981.
There is a considerable volume of gas-liquid bubbly two phase
flow data in open literature. Among them one pioneering work by
Serizawa et al. 55 and one recent work by Ohnuki and Akimoto
56 are selected for the verification of the present model. Comparisons of the present simulation against these data are described
below.
In Fig. 5 computational results are compared with the data of
Serizawa et al. 55 for various inlet qualities and reported bubble
diameters. Simulation results are taken at a considerable distance
30D from the inlet plane to avoid the effect of entrance region.
Liquid velocity is kept at 1.03 m/s, same as that in Ref. 55. Void
fractions at different qualities are presented with respect to nondimensionalized radial position. The velocity of the air at the inlet
plane is calculated by multiplying liquid velocity at the inlet with
void fraction. Though there is a mismatch in the profiles near the
where din is the bubble size entering at the inlet plane and
can be selected as desired. This mimics a situation where
the gas phase is introduced into the conduit by a large
number of nozzles uniformly distributed at the inlet section.
Overall possible bubble sizes min d / 2 D / 2 are divided into 40 equal volumes to simulate a realistic flow
phenomenon.
Solving Eqs. 13, the velocity field at each node point is

calculated with 1% convergence criteria and is subsequently used
for determining the mixture velocity at that particular position.
Mixture velocity is used to determine the distribution of bubble
size for the next instant based on the population balance equation
Eq. 11. Figure 3 shows the solution methodology of the proposed model. Simulation is forward marched until either a steady
dynamic solution / tnr , z , t ; d 0 for all d is reached or the
hydrodynamics gives a clear indication toward the transition into
slug flow. The criteria for such transition will be described in
detail in Secs. 3 and 4.
Fig. 5 Comparison of void fraction of present model and experimental data of Serizawa et al. 50
SEPTEMBER 2009, Vol. 131 / 091303-5
Fig. 6 Comparison of void fraction with the present model and

experimental data of Ohnuki and Akimoto 51
wall, it is clear from Fig. 5 that the model predicts the radial void
fraction distribution satisfactorily. Near the wall, an improved
modeling of the motion of the continuous phase, including the
effect of boundary layer and buffer layer, is expected to give a
better result without the stringent requirement of the finer grids
everywhere in the computational domain.
Ohnuki and Akimoto 56 used an optical probe for the measurement of void fraction and a dual sensor resistivity probe for
the measurement of bubble velocity and its fluctuation. In Fig. 6,
radial distribution of void fraction obtained by numerical simulation is compared with the results of Ohnuki and Akimoto 56 for
identical conditions liquid velocity of 0.18 m/s and gas velocity
of 0.11 m/s. A satisfactory agreement has been observed. Estimated void fraction profile for a higher liquid flow rate of 0.35
m/s keeping the air flow rate the same 0.11 m/s is also depicted
in the same figure. The same phase velocities were used by
Ohnuki and Akimoto 56. A good agreement has also been obtained in this case.
Comparisons of the predicted radial profile of volume averaged
bubble diameter with experiment at same initial situations 56 are
depicted in Fig. 7. The developed model predicts the correct trend
though it underpredicts the average bubble diameter toward the
tube wall. Present model is also used to investigate the distribution of bubble Sauter mean diameter in a radial plane. When the
surface area and the volume of the averaged diameter bubbles are
equal to those of the polydisperse bubbles, this averaged diameter
is called the Sauter diameter volume-surface area mean diam-
Fig. 7 Radial distribution of volume averaged bubble diameter; a comparison between present model and experimental
data of Ohnuki and Akimoto 51 at high flow rates
091303-6 / Vol. 131, SEPTEMBER 2009
Fig. 8 Comparison of the computed bubble Sauter diameter

along a radial plane with the experimental result of Shen et al.
52; low gas superficial velocity
eter. It can be calculated based on the values of the scalar fraction

f i of the volume of a particular sized bubble with respect to the
overall gaseous phase and the corresponding discrete bubble diameter di is obtained as
d32 =
1
fi
i=1 di
N
27
The computed result of the bubble Sauter diameter is compared

with the experimental observation of Shen et al. 57 for air water
flow through a tube diameter of 200 mm. Figure 8 shows a very
good prediction of the Sauter mean diameter at a gas superficial
velocity of 0.186 m/s for three different liquid superficial velocities covering a wide range, as considered by Shen et al. 57 in
their experiment. The prediction is also satisfactory for higher gas
superficial velocity, as depicted in Fig. 9. With the confidence
gained from the above comparisons, the present model has been
employed to investigate the development of flow along the axial
direction and the transition from bubbly to slug flow.
Axial Development of the Phase Distribution
It has been mentioned earlier that the gas phase at the inlet
boundary is considered as a uniform distribution of equal sized
bubbles. Therefore, it is obvious that any transformation from
Fig. 9 Comparison of the computed bubble Sauter diameter

along a radial plane with the experimental result of Shen et al.
52; high gas superficial velocity
Fig. 10 Radial bubble count histogram at an axial distance of

10D for low flow rates
bubbly flow is due to the development of the flow of the two

phase mixture downstream of the inlet plane. Due to the coalescence and breakup of the bubbles during their random motion, the
number of bubbles in a particular subgroup may increase or decrease in the downstream. This is solely guided by the hydrodynamics. Some interesting cases are discussed below.
In Figs. 1012 the bubble count of different subgroups in a
radial plane is shown for different axial positions. The bubbles of
tenth subgroup are introduced at the inlet with a liquid velocity of
0.3 m/s and an air velocity of 0.1 m/s. To track the generation of
bubbles bigger than the size entered at the inlet, the tenth sub-

10D for high flow rates
group has been entered. Moreover, a wide gap between the maximum possible bubble sizes and the bubbles to be entered will not
bias the transition criteria. There is bubble growth due to coalescence; rather smaller size bubbles are formed as one moves toward the downstream. This is clear from the bubble count histograms estimated at 10D Fig. 10, 30D Fig. 11, and 60D Fig.
12. The number of tenth subgroup bubbles introduced at the
inlet reduces, while the number of bubbles belonging to smaller
subgroups increases gradually. The coalescence of bubbles is
clearly discernable from Figs. 1315 at a high phase velocities
liquid velocity of 3 m/s and gas velocity of 1 m/s even at an
axial length of 10D. At this position one can see formation of
bubble up to a subgroup of 30 that is considerably larger compared with the bubbles introduced at the inlet plane tenth subgroup. The number of bubbles just double and triple of inlet
bubble size group is high compared with the other higher sized
bubble group as there is a rich tendency of coalescence of two
inlet sized bubbles. At further downstream the number of large
size bubbles Vi V p increases and new bubbles with even larger
diameter come into existence.
To confirm the generation of larger bubbles at higher gas and
liquid flow rates, a total volume count of bubbles above and below
those introduced at the inlet are depicted in Table 1. The generation of larger bubbles at higher phase flow rates can also be confirmed by taking an account of the volume of the bubbles of
different subgroups. For these a nondimensional parameter,
namely, Vhigh has been defined as follows:



SEPTEMBER 2009, Vol. 131 / 091303-7

Fig. 16 Volume of bubbles above tenth subgroup size at inlet
as a function of axial length; low superficial velocities
v in i
Vhigh =
i=p+1
28
v in i
i=1
where pth subgroup of bubbles enters at the inlet. Vhigh at different

axial planes for low and high phase flow rates are presented in
Tables 1 and 2. At low phase flow rates Table 1 the value of
Vhigh is relatively low and assumes a constant value after some
distance from the inlet plane. This indicates the existence of a
steady bubbly flow for the given hydrodynamic parameters. As the
flow develops, downstream of the inlet plane coalescence and
breakup occurs, the process of break up being more dominant
compared with coalescence. However, these two competing processes reach a dynamic steady state, where the maximum bubble
size can never cross a certain limit to perturb the flow toward the
slug flow regime. The flow essentially remains bubbly though
there is a readjustment of bubble size and void distribution compared with the inlet plane.
Table 2 presents a different picture. Vhigh increases continuously
along the downstream. This signifies the domination of coalescence over break up. Importantly, it indicates gradual growth of
some bubbles as they move up. This is conducive for the generation of cap bubbles to start with and for the formation of the
Taylor bubble finally.
To investigate the bubble size evolution with axial length, a
cumulative estimate of bubble volume over the input sized bubble
is shown in Fig. 16. The inlet velocity of air is taken as 0.1 m/s,
and the inlet velocity of liquid is 0.3 m/s. The figure clearly supports the statement made above from the observations in Table 2.
Though coalescence is present, it is not strong enough to generate

new bubbles of larger diameter di d p in sufficient quantity. On
the other hand, the process of breakup is more significant and is
able to create smaller bubbles di d p. Another interesting feature may be noted from the figure. Beyond a l / D of 100, the
volume count above and below the bubble size introduced at the
inlet does not change with the pipe length. This gives an idea
regarding the pipe length required for the development of the
flow. There is enough controversy regarding the developing length
in the case of two phase flow. CFD simulation is capable of resolving this issue. Obviously the criterion for the developing
length depends on the operating conditions, as will be clear from
the results presented below.
In Fig. 17 an assessment of the breakage and coalescence process along the axial direction has been made by taking an account
of the bubbles with size below and above those d = d P entering
the inlet plane. At a liquid velocity of 0.3 m/s and a gas velocity of
0.1 m/s, the coalescence rate is very low and remains more or less
constant along the axial length. On the other hand, downstream of
the inlet plane, the local breakage rate undergoes a large fluctuation. The fluctuation continues along the axial direction with diminishing amplitude and ultimately stabilizes into a steady periodic nature. Referring back to Figs. 1012, one can appreciate the
presence of a stronger breakup process compared with the process
of coalescence for the selected phase velocities. It is also evident
in Figs. 16 and 17 that the flow of the two phase mixture assumes
Table 1 Liquid velocity 0.3 m/s and gas velocity 0.1 m/s
Axial
location
10D
30D
60D
Total volume above

inlet diameter m3
Total volume below

inlet diameter m3
Vhigh
0.00004588
0.00004689
0.00004701
0.002962
0.00296
0.00298
0.01525
0.01559
0.015526
Table 2 Liquid velocity 3 m/s and gas velocity 1 m/s

Axial
location
10D
20D
30D
Total volume above

inlet diameter m3
Total volume below

inlet diameter m3
Vhigh
0.000667
0.001399
0.001895
0.002331
0.001609
0.001104
0.2225
0.4663
0.6317
091303-8 / Vol. 131, SEPTEMBER 2009
Fig. 17 Evolution of breakage and coalescence with axial location at lower flow rates
Fig. 18 Evolution of breakage and coalescence with axial location at higher flow rates
a dynamic steady state downstream of the inlet plane where the

time averaged flow characteristics do not change axially.
The morphology of the dispersed phase goes through a different
developmental pattern at higher phase flow rates. Figure 18 shows
that a vigorous process of coalescence starts immediately after the
bubbles enter the conduit. However, the coalescence rate reduces
and ultimately stabilizes to a fluctuating pattern with a small amplitude. The process of break up on the other hand is rather insignificant. The net effect of coalescence and breakup on the bubble
size distribution is depicted in Fig. 19, which clearly indicates the
zone for flow development.
The evolution of bubble size along the conduit length as a function of inlet flow rates is depicted in Fig. 20. From the figure it is
evident that at lower flow rates of gas and liquid 0.1 m/s and 0.3
m/s, respectively, the maximum bubble size remains unaltered
more or less identical to the bubbles at the inlet. This clearly
shows that in the flow field, bubble break up is dominant leading
toward the flow of different sized spherical bubbles, i.e., bubbly
flow. At a higher flow rate the bubble size reaches the prescribed
maximum limit of bubble diameter. This definitely indicates a
possible transition from bubbly flow.
4.1 Transition From Bubbly Flow. To date, the identification of flow regime boundary has been achieved either through
experimental 58,59 or by control volume based analytical modeling 60. Assuming the transition to slug flow takes place due to
high frequency of collisions at an increased density of dispersed
Fig. 19 Volume of bubbles above the tenth subgroup size at

inlet as a function of axial length; high superficial velocities
Fig. 20 Change of maximum bubble diameter along axial

length for air water two phase flow
bubble Radovicich and Moissis 60 proposed a semitheoretical

analysis. They have considered a cubic lattice of spherical bubbles
fluctuating around their mean position and suggested that the transition occurs at a maximum limiting void fraction of 0.3 for bubbly flow. This physical model has also been supported by Dukler
and Taitel 31 and Mishima and Ishii 32. However, there is a
controversy regarding the value of the maximum limiting void
fraction, and a value of 0.25 is well accepted in the literature.
A slightly different view is proposed by Bilicki and Kestin 61.
The spherical bubbles dispersed in a bubbly flow do not move
with identical velocities. The wake of a preceding bubble induces
acceleration to its successors. A bubble may catch up to its predecessors and may coalesce to form a larger bubble. According to
them this is the root cause of transition. For the success of the
coalescence process, the distance between the two bubbles should
be below a characteristics length, which is stochastic in nature
depending on the flow phenomena. Based on this argument they
proposed a transition criterion, which depends on a limiting void
fraction, phase superficial velocities, local friction velocity, and
some numerical constants. The numerical constants were determined by experiments.
Different researchers have proposed void fraction wave as the
cause of transition from bubbly to slug flow. Marcedier 62 observed that void fraction waves are damped in bubbly flows where
the damping decreases with the increase in the mean void fraction.
It has further been hypothesized that at some value of mean void
fraction, the damping could disappear and the instability of the
void fraction wave may give rise to bubble-slug transition. Matuszkiewicz et al. 63 took the queue from this postulation and
experimentally demonstrated that transition from bubbly to slug
flow occurs for a range of average void fraction 0.1 mean
0.45. It may be noted that this value is different from the limiting void fraction reported by Taitel et al. 30.
Sun et al. 64 conducted an experiment on the gas-liquid two
phase up flow in a tube diameter of 112.5, mm where the disturbance was created by a swinging metallic plate at the upstream of
the flow. Bubbly flow was observed up to a mean void fraction of
19.2%. The sudden emergence of Taylor bubble was observed at a
gas concentration of 21.5%. This induces a transition to slug flow.
They also observed at high liquid velocity that the intense turbulence suppresses the formation of Taylor bubbles. As a result an
increase in mean void fraction transforms the bubbly flow directly
to churn flow.
Alternatively, efforts have been made to characterize the termination of bubbly flow based on the maximum bubble diameter.
Again different criteria have been used. Tomiyama et al. 46 considered a bubble to be a slug bubble when its equivalent diameter
SEPTEMBER 2009, Vol. 131 / 091303-9
Fig. 21 Transition of bubbly flow; present simulation and experimental results of Olmos et al. 27 and Mercadier 62
was 0.6 of the tube diameter. Krussenberg et al. 65 and Lucas et

al. 59 also suggested a differentiation between the bubbly flow
and slug flow based on the equivalent diameter of the largest
bubble.
It may be noted that the transition criteria predicted from experimental observations are system specific and are dependent on
operating conditions. On the other hand theoretical predictions
developed so far uses mechanistic models on suitable control
volumes. They typically rely on a specific mechanism and a
number of simplified assumptions. To date little effort has been
made to apply CFD simulation for the identification of flow regime boundary. In the present work we present a methodology for
the same.
It has been accepted beyond doubt that coalescence is the
mechanism for the transition from bubbly to slug flow. However,
coalescence is a spatio-temporal process. It is necessary to investigate the hydrodynamics both locally and axially to bring out a
clear picture about the transition. Furthermore, as coalescence and
breakup simultaneously occur, we propose monitoring of the three
different characteristics.
The formation of a large size bubble is of crucial importance.
The number of large bubbles should increase, and individual
large bubbles should further grow in size
Finally, the consistency of the above two process should also be
reflected in a diminishing rate of bubble breakup.
To monitor these three aspects, we have studied i the bubble
histogram at different axial levels, ii bubble volume account
below and above the bubble size introduced at the inlet, and iii
breakage and coalescence frequency along axial length.
Based on the above criteria transition from bubbly to slug flow
has been predicted and compared with the experimental observation of Dukler and Taitel 31 in Fig. 21. An excellent match
between our prediction and experimental result has been observed. It can also be noted that the prediction by the present
method is better than that of Taitel et al. 30.
A second type of transition from bubbly flow at high liquid
velocity was proposed by Taitel et al. 30. This flow regime is
different from bubbly flow at low liquid velocity and is termed as
dispersed bubbly flow. Hinze 66 proposed a mechanism for estimating the maximum bubble size from a balance between turbulence kinetic energy and drop surface energy. Hinzes 66 analysis is applicable for dilute suspension. Brauner 67 extended
Hinzes 66 model for dense dispersion and obtained a modified
expression for maximum bubble diameter. They have further suggested considering the largest of the diameter value calculated by
the original Hinze 66 theory and its extension. For the present
system the diameter of the largest dispersed bubble is given by the
following equation:
091303-10 / Vol. 131, SEPTEMBER 2009
Fig. 22 Radial bubble count histogram for the high liquid flow
rate showing dispersed bubbly flow
cU2c D P
Dsurf
= 0.55
DP

0.6
m f
c1
0.4
29
To investigate the bubbly to dispersed bubbly transition in the

perspective of the present CFD model, the simulations have been
done for higher liquid flow rates. A typical result of bubble count
histogram can be seen in Fig. 22. The prediction shows a good
concurrence with the model of Taitel et al. 30. Almost complete
disappearance of tenth subgroup bubbles indicate that the bubbles
introduced at the inlet plane undergoes a thorough disintegration
process due to high rate of turbulent dissipation. As a result, bubbly flow at the channel inlet transforms into dispersed bubbly flow
further downstream.
To get an overall picture of the flow development with the
change in liquid velocity, one may refer to Fig. 23, where simulation results for a single gas velocity at three different liquid
velocities have been shown. The volume histogram clearly indicates slug flow, bubbly flow, and dispersed bubbly flow with the
increase in liquid flow rates. Finally, the occurrence of critical
dispersed bubble Eq. 29 as a function of phase superficial velocities has been depicted in Fig. 21. The simulation result shows
a close match with the regime boundary mechanistically predicted
by Brauner 67.
Conclusion
In the present paper the population balance equation coupled

with two fluid model is used for the prediction of bubble evolution
in gas-liquid two phase flow through a vertical conduit. Coales-
Fig. 23 Indication of different flow regimes by the bubble volume histogram obtained with a variation in liquid velocity for a
fixed gas flow rate
cence of two similar or different sized bubbles and homogeneous

and nonhomogeneous binary break up of a bubble has been considered in the PBM. The model has been validated rigorously
against experimental results from various sources.
Using the present model an effort has been made to propose a
new transition criterion from bubbly flow to slug flow. By tracking the coalescence frequency and breakup frequency along the
conduit axis, the dominance of any of the processes over the other
at a given condition of flow superficial velocities can easily be
made. Volume count of the bubbles over and below the bubble
size introduced at the inlet can be used as a criterion for the above.
Additionally, the bubble count histogram obtained from the
present model at different axial locations also helps to ascertain
the transition from bubbly to slug flow. The transition boundary
obtained by such procedure gives an excellent match with the
published experimental data. At a higher liquid velocity the simulation results show a suppression of coalescence and a dominance
of breakup processa situation conducive for dispersed bubbly
flow. Simulation results have been generated to obtain the maximum bubble size Eq. 29 needed for a transition to disperse
bubbly flow. A good agreement has been obtained with the model
of Brauner 67.
The present work demonstrates the strength of CFD simulation
for the identification of flow regime. Even though no special turbulence model has been considered in the formulation, the model
successfully defines the boundaries of bubbly flow. Furthermore,
as the axial development of the flow phenomena can be studied,
the model can be utilized for determining the developing length
for flow through a conduit.
Nomenclature
b
B
BB
BC
c1
c2
c3
CD
CFL
CTD
dc
DB
DC
D
d
f Go
f Lo
fv
Fdispr
Fdispz
FLGz
FLGr
FGIz
FGIr
FLIz
FLIr
empirical constant of Eq. 19

empirical constant of Eq. 7
number of new born bubbles due to break up
number of new born bubbles due to
coalescence
constant in Eqs. 14 and 17
empirical constant in Eq. 23
empirical constant in Eq. 24
drag coefficient for a single bubble
interfacial friction factor
turbulent dispersion force coefficient
critical bubble diameter for bubble breakup
frequency m
number of dying bubbles due to break up
number of dying bubbles due to coalescence
pipe diameter m
diameter of the bubbles m
wall friction factor of gaseous phase
wall friction factor of liquid phase
volume fraction of daughter bubble to mother
bubble
turbulent dispersion force in the radial direction kg m s2
turbulent dispersion force in the axial direction
kg m s2
interfacial force to gaseous phase in the axial
direction kg m s2
interfacial force to gaseous phase in the radial
direction kg m s2
mass transfer force to gaseous phase in the
axial direction kg m s2
mass transfer force to gaseous phase in the
radial direction kg m s2
mass transfer force to liquid phase in the axial
direction kg m s2
mass transfer force to liquid phase in the radial
direction kg m s2
FWGz wall gas friction force in the axial direction

kg m s2
FWGr wall gas friction force in the radial direction
kg m s2
FWLz wall liquid friction force in the axial direction
kg m s2
FWLr wall liquid friction force in the radial direction
kg m s2
gx gravitational acceleration in the x direction
m s2
ga breakage frequency of the bubbles of diameter
a s1
ha , b effective swept volume rate for bubble diameter a and b s1
k empirical constant in Eq. 14
m empirical constant in Eq. 19
n empirical constant in Eq. 7
nr , z , t : d number of bubbles at r , z of size d at time t
pa , b film rupture efficiency of bubbles of diameter
a and b
P pressure of the mixture phase kg m s2
RB average bubble diameter m
Sctg turbulent Schmidt number for gaseous phase
t time s
Um mixture velocity m s1
wg axial velocity of dispersed phase m s1
wl axial velocity of continuous phase m s1
d
v
volume of the bubble of diameter d m3
ug radial velocity of dispersed phase m s1
ul radial velocity of continuous phase m s1
z axial direction of the conduit m
r radial direction of the conduit m
Greek Alphabet
dispersed phase void fraction
PLO single phase frictional pressure drop kg m s2
energy dissipation rate per unit mass m2 s3
tg turbulent kinetic viscosity m2 s1
a , b daughter bubble probability distribution
a , b coalescence frequency of bubble diameter a
and b s1
g viscosity of discrete phase kg m1 s1
l viscosity of continuous phase kg m1 s1
a number of bubbles formed from the breakage
of a bubble diameter a
min nondimensional minimum daughter bubble size
l density of continuous phase kg m3
g density of dispersed phase kg m3
surface tension kg s2
References
1 Zuber, N., and Findlay, J. A., 1965, Average Volumetric Concentration in
Two Phase Flow Systems, ASME J. Heat Transfer, 87, pp. 453468.
2 Ishii, M., and Zuber, N., 1970, Thermally Induced Flow Instabilities in TwoPhase Mixtures, Proceedings of the Fourth International Heat Transfer Conference, Paris, France.
3 Saha, P., and Zuber, N., 1978, An Analytical Study of the Thermally Induced
Two Phase Flow Instabilities Including the Effects of Thermal NonEquilibrium, Int. J. Heat Mass Transfer, 21, pp. 415426.
4 Kim, C., and Roy, R. P., 1981, Two Phase Flow Dynamics by a Five Equation
Drift Flux Model, Lett. Heat Mass Transfer, 8, pp. 5768.
5 Hibiki, T., and Ishii, M., 2000, Two-Group Interfacial Area Transport Equations at Bubbly-to-Slug Flow Transition, Nucl. Eng. Des., 202, pp. 3976.
6 Rakhmatulin, K. A., 1956, Fundamentals of Gas Dynamics of Interpenetrating Motions of Compressible Media, Prikl. Mat. Mekh., 202, pp. 184195.
7 Ishii, M., 1977, One Dimensional Drift Flux Model and Constitutive Equations for Relative Motion Between Phases in Various Two-Phase Flow Regimes, Argonne National Laboratory, Technical Report No. ANL-77-47.
8 Wu, Q., Ishii, M., and Uhle, J., 1998, Frame Work of Two-Group Model for
SEPTEMBER 2009, Vol. 131 / 091303-11
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
Interfacial Area Transport in Vertical Two-Phase Flows, Trans. Am. Nucl.

Soc., 79, pp. 351352.
Fu, X. Y., and Ishii, M., 2003, Two-Group Interfacial Area Transport in Vertical Air-Water Flow I. Mechanistic Model, Nucl. Eng. Des., 219, pp. 143
168.
Kumar, S., and Ramkrishna, D., 1996, On the Solution of Population Balance
Equations by DiscretizationI. A Fixed Pivot Technique, Chem. Eng. Sci.,
518, pp. 13111332.
Marchisio, D. L., Vigil, R. D., and Fox, R. O., 2003, Quadrature Method of
Moments for Aggregation-Breakage Processes, J. Colloid Interface Sci., 258,
pp. 322334.
Dorao, C. A., and Jakobsen, H. A., 2006, A Least Squares Method for the
Solution of Population Balance Problems, Comput. Chem. Eng., 30, pp. 535
547.
Yao, W., and Morel, C., 2004, Volumetric Interfacial Area Prediction in Upwards Bubbly Two-Phase Flow, Int. J. Heat Mass Transfer, 47, pp. 307328.
Chen, P., Sanyal, J., and Dudukovic, M. P., 2005, Numerical Simulation of
Bubble Columns Flows: Effect of Different Breakup and Coalescence Closures, Chem. Eng. Sci., 604, pp. 10851101.
Yeoh, G. H., and Tu, J. Y., 2006, Two-Fluid and Population Balance Models
for Subcooled Boiling Flow, Appl. Math. Model., 30, pp. 13701391.
Cheung, S. C. P., Yeoh, G. H., and Tu, J. Y., 2007, On the Modelling of
Population Balance in Isothermal Vertical Bubbly FlowsAverage Bubble
Number Density Approach, Chem. Eng. Process., 46, pp. 742756.
Jakobsen, H. A., Lindborg, H., and Dorao, C. A., 2005, Modeling of Bubble
Column Reactors: Progress and Limitations, Ind. Eng. Chem. Res., 44, pp.
51075151.
Krepper, E., Lucas, D., and Prasser, H. M., 2005, On the Modelling of Bubbly Flow in Vertical Pipes, Nucl. Eng. Des., 235, pp. 597611.
Lo, S., 1996, Application of the MUSIG Model to Bubbly Flows, AEA
Technology, Technical Report No. AEAT-1096.
Lucas, D., Krepper, E., and Prasser, H. M., 2001, Prediction of Radial Gas
Profiles in Vertical Pipe Flow on Basis of the Bubble Size Distribution, Int. J.
Therm. Sci., 40, pp. 217225.
Esmaeeli, A., and Tryggvason, G., 1998, Direct Numerical Simulations of
Bubbly Flows. Part 1Low Reynolds Number Arrays, J. Fluid Mech., 377,
pp. 313345.
Bunner, B., and Tryggvason, G., 2002, Dynamics of Homogeneous Bubbly
Flows: Part 1. Rise Velocity and Microstructure of the Bubbles, J. Fluid
Mech., 466, pp. 1752.
Bunner, B., and Tryggvason, G., 2002, Dynamics of Homogeneous Bubbly
Flows: Part 2. Velocity Fluctuations, J. Fluid Mech., 466, pp. 5384.
Bunner, B., and Tryggvason, G., 2003, Effect of Bubble Deformation on the
Properties of Bubbly Flows, J. Fluid Mech., 495, pp. 77118.
Biswas, S., Esmaeeli, A., and Tryggvason, G., 2005, Comparison of Results
From DNS of Bubbly Flows With a Two Fluid Model for the TwoDimensional Laminar Flows, Int. J. Multiphase Flow, 31, pp. 10361048.
Shnip, A. I., Kolhatkar, R. V., Swamy, D., and Joshi, J. B., 1992, Criteria for
Transition From the Homogeneous to Heterogeneous Regime in TwoDimensional Bubble Column Reactors, Int. J. Multiphase Flow, 185, pp.
705726.
Olmos, E., Gentric, C., Vial, C., Wild, G., and Midoux, N., 2001, Numerical
Simulation of Multiphase Flow in Bubble Column Reactors. Influence of
Bubble Coalescence and Break-Up, Chem. Eng. Sci., 56, pp. 63596365.
Olmos, E., Gentric, C., Ponsin, S., and Midoux, N., 2003, Description of
Flow Regime Transitions in Bubble Columns Via Laser Doppler Anemometry
Signal Processing, Chem. Eng. Sci., 58, pp. 17311742.
Sankaranarayanan, K., and Sundaresan, S., 2002, Lift Force in Bubbly Suspensions, Chem. Eng. Sci., 57, pp. 35213542.
Taitel, Y., Bornea, D., and Dukler, A. E., 1980, Modelling Flow Pattern
Transitions for Steady Upward Gas-Liquid Flow in Vertical Tubes, AIChE J.,
263, pp. 345354.
Dukler, A. E., and Taitel, Y., 1977, Flow Regime Transitions for Vertical
Upward Gas Liquid Flow, Houston University, Progress Report No. 2,
NUREG-0163.
Mishima, K., and Ishii, M., 1984, Flow Regime Transition Criteria for Upward Two-Phase Flow in Vertical Tubes, Int. J. Heat Mass Transfer, 27, pp.
723737.
Anderson, T. B., and Jackson, R., 1967, A Fluid Mechanical Description of
Fluidized Beds: Equations of Motion, Ind. Eng. Chem. Fundam., 6, pp. 527
539.
Richter, H. J., 1983, Separated Two-Phase Flow Model: Application to Critical Two Phase Flow, Int. J. Multiphase Flow, 95, pp. 511530.
Chisholm, D., 1973, Pressure Gradient Due to Friction During the Flow of
Evaporating Two Phase Mixtures in Smooth Tubes and Channel, Int. J. Heat
Mass Transfer, 16, pp. 347358.
Burns, A. D., Frank, T., Hamill, I., and Shi, J., 2004, The Farve Averaged
Drag Model for Turbulent Dispersion in Eulerian Multi-Phase Flows, Proceedings of the Fifth International Conference on Multiphase Flow, Yokohama,
Japan.
Ramkrishna, D., 2000, Population BalancesTheory and Applications to Particulate Systems in Engineering, Academic, San Diego, CA.
Wang, T., Wang, J., and Jin, J., 2003, A Novel Theoretical Breakup Kernel
Function for Bubbles/Droplets in a Turbulent Flow, Chem. Eng. Sci., 58, pp.
091303-12 / Vol. 131, SEPTEMBER 2009
46294637.
39 Luo, H., and Svendsen, H. F., 1996, Theoretical Model for Drop and Bubble
Breakup in Turbulent Dispersions, AIChE J., 42, pp. 12251233.
40 Kostoglou, M., and Karabelas, A. J., 2005, Towards a Unified Framework for
the Derivation of Breakage Functions Based on the Statistical Theory of Turbulence, Chem. Eng. Sci., 60, pp. 65846595.
41 Ioannou, K., Hu, B., Matar, O. K., Hewitt, G. F., and Angeli, P., 2004, Phase
Inversion in Dispersed LiquidLiquid Pipe Flows, Proceedings of the Fifth
International Conference on Multiphase Flow, Yokohama, Japan.
42 Tsouris, C., and Tavlarides, L. L., 1994, Breakage and Coalescence Models
for Drops in Turbulent Dispersions, AIChE J., 40, pp. 395406.
43 Troshko, A. A., and Zdravistch, F., 2009, CFD Modeling of Slurry Bubble
Column Reactors for FisherTropsch Synthesis, Chem. Eng. Sci., 64, pp.
892903.
44 Carrica, P. M., and Clausse, A. A., 1993, Mathematical Description of the
Critical Heat Flux as Nonlinear Dynamic Instability, Instabilities in Multiphase Flow, G. Gouesbet and A. Berlemont, eds., Plenum, New York.
45 Kostoglou, M., and Karabelas, A. J., 1998, Theoretical Analysis of Steady
State Particle Size Distribution in Limited Breakage Process, J. Phys. A, 31,
pp. 89058921.
46 Tomiyama, A., Nakahara, Y., Adachi, Y., and Hosokawa, S., 2003, Shapes
and Rising Velocities of Single Bubbles Rising Through an Inner Subchannel,
J. Nucl. Sci. Technol., 40, pp. 136142.
47 Coulaloglou, C. A., and Tavlarides, L. L., 1977, Description of Interaction
Processes in Agitated Liquid-Liquid Dispersions, Chem. Eng. Sci., 32, pp.
12891297.
48 Chesters, A. K., 1991, The Modelling of Coalescence Processes in FluidLiquid Dispersions: A Review of Current Understanding, Trans. Inst. Chem.
Eng., 69, pp. 259270.
49 Lovick, J., 2004, Horizontal Oil-Water Flows in the Dual Continuous Flow
Regime, Ph.D. thesis, University College London, England.
50 Hu, H. G., and Zhang, C., 2007, A Modified k Turbulence Model for the
Simulation of Two-Phase Flow and Heat Transfer in Condensers, Int. J. Heat
Mass Transfer, 50, pp. 16411648.
51 Ekambara, K., Sanders, R. S., Nandakumar, K., and Masliyah, J. H., 2008,
CFD Simulation of Bubbly Two-Phase Flow in Horizontal Pipes, Chem.
Eng. J., 144, pp. 277288.
52 Prosperetti, A., and Tryggvason, G., 2007, Introduction: A Computational
Approach to Multiphase Flow, A. Prosperetti and G. Tryggvason, eds., Computational Methods for Multiphase Flow, Cambridge University Press, Cambridge, England.
53 Tomiyama, A., and Shimada, N., 2001, A Numerical Method for Bubbly
Flow Simulation Based on a Multi-Fluid Model, ASME J. Pressure Vessel
Technol., 123, pp. 510516.
54 Schumann, U., 1975, Subgrid Scale Model for Finite Difference Simulations
of Turbulent Flows in Plane Channels and Annuli, J. Comput. Phys., 18, pp.
376404.
55 Serizawa, A., Kataoka, I., and Michiyoshi, I., 1975, Turbulence Structure of
Air-Water Bubbly FlowII. Local Properties, Int. J. Multiphase Flow, 2, pp.
235246.
56 Ohnuki, A., and Akimoto, H., 2000, Experimental Study on Transition of
Flow Pattern and Phase Distribution in Upward Air-Water Two-Phase Flow
Along a Large Vertical Pipe, Int. J. Multiphase Flow, 263, pp. 367386.
57 Shen, X., Mishima, K., and Nakamura, H., 2005, Two-Phase Phase Distribution in a Vertical Large Diameter Pipe, Int. J. Heat Mass Transfer, 481, pp.
211225.
58 Nakoryakov, V. E., Kashinsky, O. N., Randin, V. V., and Timkin, L. S., 1996,
Gas Liquid Bubbly Flow in Vertical Pipes. Data Bank Contribution, ASME
J. Fluids Eng., 118, pp. 377382.
59 Lucas, D., Krepper, E., and Prasser, H. M., 2005, Development of Co-Current
Air-Water Flow in a Vertical Pipe, Int. J. Multiphase Flow, 31, pp. 1304
1328.
60 Radovicich, N. A., and Moissis, R., 1962, The Transition From Twophase
Bubble Flow to Slug Flow, MIT Technical Report No. 7-7633-22.
61 Bilicki, Z., and Kestin, J., 1987, Transition Criteria for Two Phase Flow
Patterns in Vertical Upward Flow, Int. J. Multiphase Flow, 13, pp. 283294.
62 Mercadier, Y., 1981, Contribution aletude des propagations de perturbations
de taux de vide dans les ecoulements diphasiques eau-air a bulles, Ph.D.
thesis, Institut National Polytechnique de Grenoble, Universite Scientifique et
Medicale, France.
63 Matuszkiewicz, A., Flamand, J. C., and Boure, J. A., 1987, The Bubble Slug
Flow Pattern Transition and Instabilities of Void Fraction Waves, Int. J. Multiphase Flow, 13, pp. 199217.
64 Sun, B., Wang, R., Zhao, X., and Yan, D., 2002, The Mechanism for the
Formation of Slug Flow in Vertical Gas-Liquid Two Phase Flow, Solid-State
Electron., 46, pp. 23232329.
65 Krussenberg, A. K., Prasser, H. M., and Schaffrath, A., 1999, A New Criterion for the Bubble Slug Transition in Vertical Tubes, Proceedings of the
Ninth International Topical Meeting on Nuclear Reactor Thermal Hydraulics
NURETH-9, San Francisco, CA.
66 Hinze, J., 1955, Turbulence, McGraw-Hill, New York.
67 Brauner, N., 2001, The Prediction of Dispersed Flows Boundaries in LiquidLiquid and Gas-Liquid Systems, Int. J. Multiphase Flow, 27, pp. 885910.
A. K. Das
e-mail: arup@mech.iitkgp.ernet.in
P. K. Das
e-mail: pkd@mech.iitkgp.ernet.in
IIT Kharagpur,
721302, India
J. R. Thome
LCTM,
EPFL,
Lausanne CH-1015, Switzerland
e-mail: john.thome@epfl.ch
Transition of Bubbly Flow in

Vertical Tubes: Effect of Bubble
Size and Tube Diameter
In a companion paper (Modelling Bubbly Flow by Population Balance Technique Part
I: Axial Flow Development and Its Transitions, ASME J. Fluids Eng), a two fluid model
along with a multiclass population balance technique has been used to find out comprehensive criteria for the transition from bubbly to slug flow, primarily through a study of
axial flow development. Using the same basic model the transition mechanism has been
investigated in the present paper covering a wide range of process parameters. Though
the dominating rate of bubble coalescence during the axial development of the flow acts
as the main cause for the transition to slug flow, the simultaneous transformation of the
radial voidage pattern cannot be overlooked. Appearance of core, intermediate, wall, and
two peaks are observed in the radial voidage distribution depending on the phase superficial velocities. A map has been developed indicating the boundaries of the above subregimes. It has been observed that not only the size of the bubbles entering the inlet plane
but also the size distribution (monodispersion or bidispersion) changes the voidage peak
and shifts the transition boundary. It is interesting to note that the bubbly flow only with
a core peak void distribution transforms into slug flow with a change in the operating
parameters. Transition boundary is also observed to shift with a change in the tube
diameter. The simulation results have been compared with experimental data taken from
different sources and very good agreements have been noted. DOI: 10.1115/1.3203206
Keywords: bubbly flow, population balance, transition, wall peaking, core peaking
Introduction
A bubbly flow often appears as a homogeneous dispersion of

gas bubbles in a liquid medium. In reality, the flow regime is
rather complex due to the presence of different spatial-temporal
structures and the interplay of various complex mechanisms. The
motion of a bubble is the result of a large number of forces,
namely fluid-fluid interaction, bubble transverse dispersion, wall
lubrication, Magnus force, etc. Depending on the relative magnitude of these forces, different variations are possible in the bubbly
flow. It is well known that the pattern of void distribution in a
cross sectional plane could be wall peaked or core peaked depending on the operating conditions 14. It has also been reported
that the conduit size has a distinct effect on the bubbly flow pattern. Further, the bubbly flow is influenced by the characteristics
of the bubble inputs. Development of the void profile is not identical in the case of monodispersed and polydispersed bubble inputs.
Modeling of the bubbly flow with all its complexities is a formidable task. Nevertheless, due to the occurrence of the bubbly
flow in widespread industrial application, numerous efforts have
been made to model it. Based on generalized two fluid approach
5,6, Antal et al. 7 developed a model for two-phase bubbly
flow. Troshko and Hassan 8 presented a two-phase flow model
including the near wall region and developed a two-phase wall
law. To model the turbulence, they extended the single-phase standard k- model for two-phase flow, neglecting the turbulence in
the dispersed phase. But their model is valid only for monodispersed flow of gaseous bubbles in the bulk liquid. Politano et al.
9 used the two fluid model, considering the bubble size distri1
JOURNAL OF FLUIDS ENGINEERING. Manuscript received January 13, 2009; final manuscript received June 23, 2009; published online August 18, 2009. Assoc. Editor:
Theodore Heindel.
bution with groups of constant mass. Their model can predict the
presence of wall and core peakings but remains silent about the
possibility of any other voidage profile. Carrica and Clausse 10
solved the behavior of polydispersed bubble formation using a
multigroup model along with a two fluid model. Recently,
Kreeper et al. 11 used the N-M MUSIG model for solving bubbly flow having N velocity group and M bubble group size. Yeoh
and Tu 12 solved the population balance equation in axial direction along with a two fluid model using the method of discrete
classes in order to incorporate the effect of bubble size.
In a companion paper 13, using the two-fluid model and the
population balance technique, the hydrodynamics of bubbly flow
has been simulated. Emphasis has been given on the analysis of
axial flow development. The population balance technique enables
one to detect the bubble population at any axial level and to assess
the process of coalescence and breakage. An objective and comprehensive criterion for the transition from bubbly to slug flow has
been proposed. The simulation shows excellent agreement with
published results.
Certainly the importance of axial development cannot be overlooked as it has direct bearing to the transition. Nevertheless, hydrodynamics of bubbly flow strongly depends also on the phase
distribution and bubble population along the radial direction. This
gives rise to different peaked structure 1,3,14 of the void fraction and also influences the transition. To date the prediction of
bubbly-slug transition based on superficial phase velocities
15,16 is well accepted by many researchers. However, such a
criterion does not take into cognition the effect of inlet bubble size
17 and tube diameter 18 on the transition from bubbly to slug
flow. In the present paper, the basic two fluid and population balance model 13 for bubbly flow has been employed to investigate
the above issues. In the first part of the paper, we have made a
thorough study on the radial distribution of dispersed phase. In the
second part, the effect of operating variables other than the phase
velocities on the flow regime transition has been investigated.
SEPTEMBER 2009, Vol. 131 / 091304-1
Fig. 1 Comparison of the void distribution of the present

model and the experimental observations of Lucas et al. 20 at
low phase flow rates
Excellent agreement has been obtained between the simulation

and the experimental data obtained from various sources.
In a number of experiments on gas liquid upflow through vertical conduits, nonuniform void distribution has been observed at
the downstream while gas bubbles were introduced uniformly at
the inlet plane. Theoretical simulations of the bubbly flow also
corroborate the peaked characteristic of the void profile 15,19. It
is important to note that not only the absolute value of the void
fraction varies in the different flow regimes, each of the regimes
has a typical pattern for radial distribution of the void fraction.
Therefore, in the present paper, we want to investigate further into
the transformation of the void profile due to the operating and
inlet conditions.
The present model is compared with the experimental void
fraction obtained by Zun 3, Song et al. 4, and Lucas et al. 20
in the case of a two-phase flow mixture of air and water in vertical
ducts. To simulate the experiments, the ducts were discretized into
40 radial nodes with finer discretization near the wall of the tube.
In the axial direction 500 uniform spaced grids were generated to
track the evolution of the bubbly flow. Bubble mass is divided into
40 equal groups by volume to eradicate the complexity of the
volume discretization during the coalescence of bubbles. In case
of bubble breakage, if the daughter bubbles do not fall in an exact
size group they are distributed among its immediate neighbors,
maintaining the conservation of mass and number.
An extensive study of the bubbly flow 20 has been made in
the MTLoop facility. The test section in this loop has an inner
diameter of 52.3 mm and a length of 3.5 m. Wire mesh sensors
have been used for the measurement of void fractions at a distance
of 60D from the inlet. Figure 1a shows the simulation results at
an air flow rate of 0.0096 m/s and a water flow rate of 0.0405 m/s
as has been used by Lucas et al. 20. Most of the bubbles are
gathered near the wall of the tube. Results from the present model
also show a similar pattern of the void fraction distribution. Similar experimental studies were made by Prasser et al. 21. They
used a larger tube with 195.3 mm diameter and 8.78 m length. At
an air flow rate of 0.53 m/s and water flow rate of 1.02 m/s, they
observed that the void distribution pattern changed Fig. 2 into a
parabolic shape exhibiting core peak. Simulation results for the
same conditions satisfactorily agree Fig. 2 with the experimental
observations. This shows that the present model can handle both
inward and outward lateral migrations of the bubbles, depending
on the flow velocities and tube diameter.
Zun 3 proposed that uniform input of big bubbles at the inlet
produces a core peak while smaller bubbles at the inlet produce a
wall peak. Interestingly, when a mixture of big and small bubbles
091304-2 / Vol. 131, SEPTEMBER 2009
Fig. 2 Comparison of the void distribution of the present

model and the experimental observations of Prasser et al. 21
at high phase flow rates
is introduced at the inlet, simultaneous occurrence of wall and

core peaks is possible at further downstream. According to Zun
3, the bubbles, having high intrinsic oscillatory motion, stay near
the wall and the others settle at the core of the conduit as the
amplitude of the oscillation is a strong function of the bubble
diameter. To demonstrate this, Zun 3 conducted an experiment
using a vertical conduit of square cross section having 25.4 mm
sides. Downstream gas void fraction was measured using microresistivity probes. Experiments were conducted for different inlet
bubble sizes and void fractions. In general, experiments were conducted for low gas flow rates. Results are depicted in Fig. 3. As
the present model is an axisymmetric one, an exact simulation of
the results of Zun 3 is not possible. However, simulations have
been done for a circular tube using identical operating conditions.
Simulation results exhibit excellent trend matched with the experimental results of Zun 3.
The shape of the void distribution in the downstream due to the
injection of two different bubble sizes at the inlet has also been
investigated by Song et al. 4. They used a vertical tube with 29
mm inner diameter and 3.7 m length to study the development of
the voidage profile in a mineral oil-air flow. Nozzles of different
sizes were used to produce bubbles of two different diameters. 3D
photographic method was used for measuring the position of the
bubbles and the void fraction profile was derived from it. Figure 4
shows the comparison of the estimated void distribution against
the experimental observation by Song et al. 4 for the simultaneous entry of 3.4 mm and 2.5 mm diameter bubbles. Liquid flow
Fig. 3 Comparison of the two peak void distributions; present

model versus experimental observations of Zun 3
Fig. 4 Comparison of the two peak void distributions; present

model versus experimental observations of Song et al. 4
rate is kept at 0.079 m/s and air flow rate is maintained at 0.0082
m/s, which are identical to the experimental conditions of Song et
al. 4. Generated results at an axial length of 60D were presented
to minimize the entrance effect. Agreement between the simulation and experiment is reasonably well.
Occurrence of different peaked structures of the voidage profile
in the bubbly flow has been observed in different independent
experiments 1. The reason for this has loosely been attributed to
the forces acting on individual bubbles and the interactions between neighboring bubbles. As the evolution of the voidage profile is a result of the interplay of various complex phenomena, its
prediction is not possible through a control volume based model
prevalent in the analysis of the two-phase flow. On the contrary,
a computational fluid dynamics CFD simulation with an inbuilt
capability to model the bubble interactions can simulate this phenomenon reasonably well. The capability of the present model for
predicting bubbly-slug transition has been presented in an earlier
communication 13. The model prediction depicts a good match
with the regime boundary proposed by Taitel et al. 22 experimentally. It is interesting to note that a unique regime boundary
was not observed by different investigators. In Fig. 5 bubbly-slug
transitions as observed in different experiments point out a substantial difference between themselves. While the trend of the
transition boundary of Griffith and Wallis 23, Dukler and Taitel
15, and Mishima and Ishii 24 are similar which also matches
with the prediction of the present simulation, the nature of the
other transition boundaries 22,2528 are different. It may be
quite possible that the difference in the inlet condition of the spe-
Fig. 5 Comparison of the existing flow pattern maps with the

present prediction
Fig. 6 Effect of bubble size homogeneous distribution at the

inlet on the flow pattern map
cific experiments is mainly responsible for this mismatch. Unfortunately, in most of the experimental results the inlet conditions,
particularly the bubble size distribution, have not been specified
explicitly. Therefore, it is difficult to substantiate the above proposition directly. As an alternative, we have tried to predict the transition boundary for different inlet conditions. The results are discussed below.
To see the effect of bubble size at the inlet on the flow pattern
map, results were generated using the present model for homogeneous inputs of various bubble sizes. Figure 6 shows the change in
the flow regime map due to the change in the bubble size at the
inlet from 0.5 mm to 1 mm and then to 3 mm. Bubbly flow turns
into slug flow at a lower flow rate of liquid for a fixed gas flow
rate with the increase in inlet bubble size. It is obvious that
bubbles of higher size at the input takes less effort to merge into a
big gaseous slug. Results generated from the developed model
efficiently support the intuitive concept about bubble coalescence.
Flow pattern map of Bilicki and Kestin 29 for different sized
bubble inputs are also plotted in Fig. 6 to support the results
generated from the present model. A plexiglass tube with 20 mm
diameter and 1.5 m length has been used in Ref. 29 to view the
two-phase flow patterns for air and water. Tests were carried out
for bubble diameters of 1 mm and 3 mm. The observations are in
close concordance with the simulation results.
In most of the experiments monosized bubbles are rarely injected at the inlet. In general there could be a distribution of
bubble size. Simulations are made to study the effect of a mixed
bubble population at the inlet on the flow regime transition. Different inlet populations consisting mainly of a 0.5 mm bubble
diameter, b 1 mm bubble diameter and c 2 mm bubble diameter as well as d 50% bubbles of 0.5 mm diameter+ 50% of 1
mm diameter, and e 50% bubbles of 1 mm diameter+ 50% of 2
mm diameter were considered at the inlet, as shown in Fig. 7. In
all the cases, the inlet volume fraction is kept constant. It can be
seen that with the mixing of higher sized bubbles the flow pattern
map for 1 mm diameter bubbles at the inlet shifts downward,
causing the transition at a lower air flow rate. Similarly, when 0.5
mm diameter bubbles are mixed with 1 mm diameter bubbles,
slug bubbles appear at a higher gas flow rate for a fixed liquid
flow rate compared with the homogeneous input of 1 mm diameter bubbles. This analysis can be extended further to investigate
the effect of polydispersed bubble input at the inlet plane on the
void distribution profiles and flow pattern map. It is also clear
from the results generated from the simulation that along with the
superficial velocities, the bubble diameter is an important parameter for the transition of bubbly flow to slug flow. It is interesting
to note that curves a and d, and curves c and e intersect each other,
indicating an implicit effect of the inlet bubble size on the transition boundary. Effect of mixing two different bubble sizes cannot
SEPTEMBER 2009, Vol. 131 / 091304-3
Fig. 7 Effect of bubble size nonhomogeneous distribution at

the inlet on the flow pattern map
be predicted even qualitatively from the logic of linear averaging. The input bubble diameter distribution influences the hydrodynamics in a number of ways. It has got direct effects on the rate
of breakage and coalescence as well as on the drag force. The
laws governing these phenomena are strongly nonlinear. The regime boundaries depicted in Fig. 7 is a reflection of that fact.
It is a well recognized fact that flow regime transition is not a
discrete event. It is more likely that a fully developed bubbly flow
transforms into a fully developed slug flow through a gradual
change. In this regard the different peaked structures of the voidage distributions in bubbly flow may have a role to play. The
present model has been used to investigate this. For a case study,
uniform distribution of bubbles of 0.05D has been considered at
the inlet plane and the void fraction profile is investigated at 60D
from the inlet. Figure 8 shows the boundary obtained between the
wall and core peaking void fraction profiles based on the liquid
and gaseous phase flow velocities. To make a comparison with the
available void distribution, map results of Serizawa and Kataoka
30, Ohnuki and Akimoto 14, and Lucas et al. 31 are depicted
in the same figure. Serizawa and Kataoka 30 proposed a band of
flow velocities beyond which a void fraction profile with core
peaking changes into a profile with wall peaking. Present results
of the simulation falls well in the band proposed by Serizawa and
Kataoka 30. The boundaries proposed by Ohnuki and Akimoto
14 and Lucas et al. 31, though fall within the band, do not
match well with the present simulation. In their model, Lucas et
al. 31 assumed that the whole bubble mass is divided into three
separate bubble groups, whereas in the present simulation 40 subgroups have been considered. This may be one of the reasons in
the discrepancy of the results.
Fig. 8 Ranges of wall and core peakings
091304-4 / Vol. 131, SEPTEMBER 2009
Fig. 9 Subregimes in the bubbly flow based on the voidage

profile
To appreciate further the development of the voidage profile

and its effect on the flow regime transition, one may refer to Fig.
9. Here, a line PQ drawn parallel to the abscissa, indicating increasing gas superficial velocities at a constant liquid superficial
velocity, is tracked. Point A on this curve indicates a peak very
near the wall at a low gas flow rate. With the increase in gas flow
rate point B the peak shifts away from the wall. With a further
increase in gas flow rate point C, two peaksone at the pipe
center and another near the wallappears. Subsequent increase in
gas superficial velocity point D strengthens the core peak while
the wall peak disappears. Finally, one gets slug flow by increasing
the air velocity further point E. This also reveals the physics of
flow regime transition. Slug flow constitutes of a pseudoperiodic
movement of Taylor bubbles and spherical bubble laden liquid
slugs through any cross sections. From the above observation it
can be said that the formation of the Taylor bubbles is pre-empted
by a dense voidage as well as larger bubbles at the core of the
conduit. This is also supported by experimental observations 32.
Over the years, numerous efforts 22,33,34 have been made to
predict the bubbly-slug transition criteria for concurrent upflow
through a vertical tube. Different dimensional and nondimensional
parameters have been selected as the coordinate axes for representing the flow regime boundary on a two dimensional plane.
Out of all such efforts the model of Dukler and Taitel 35 is one
of the most popular. Based on the argument of maximum packing
density of the spherical bubbles, the maximum void fraction has
been predicted in this model. While doing so, a uniform distribution of bubbles in an infinite mixture medium has been considered
without taking any cognition of the tube wall effect. With a simplifying assumption, the limiting void fraction achieved was 25%.
The authors have further used the material conservation and the
relationship for slip velocity to obtain the transition criteria in
terms of the phase superficial velocities. This is a simple yet elegant derivation, which showed reasonable agreements with a
number of experimental results. The idealizations made in the
model, namely, one dimensionality and invariance with bubble
and tube diameters were essential for obtaining a simple closed
form expression as the transition criteria.
It has already been demonstrated that the bubble size influences
the transition boundary. Evidence is also available in the literature
indicating different flow structure and transition boundaries for
different tube diameters. To examine exclusively the effect of the
tube diameter on the transition boundary, one needs to conduct
experiments with different tube sizes and identical inlet bubble
populations. However, such results are readily not available. We
have simulated bubbly flow for different conduit diameters 25.4
mm, 50.8 mm, and 76.2 mm with uniform distribution of 2.54
mm diameter bubbles at the inlet. For 25.4 mm and 50.8 mm pipe
geometries, 40 grids have been taken in radial direction whereas
Fig. 10 Effect of tube diameter on the void fraction profile at

low phase velocities
Fig. 12 Bubbly to slug transition boundary for different tube

diameters
3
60 grids have been taken for the 76.2 mm diameter tube. Axial
length is kept constant for all the tube diameters at 5 m. All the
results depicted in the present paper are taken at a distance of 40D
from the inlet plane. Results are depicted in Figs. 1012. Void
fraction distribution at an air flow rate of 0.1 m/s and a liquid flow
rate of 1 m/s is depicted in Fig. 10 for all the three tube sizes. In
all the cases, sharp peak can be visible at the wall of the conduit.
At a 25.4 mm diameter the rise of the peak is highest at the
vicinity of the wall while the deepest valley is observed at the
central region. With the increase in the diameter the peak sharpness decreases, making the voidage profile relatively uniform. The
position of the peak near the wall also shifts toward the center of
the tube geometry. Results of the void fraction distribution for
various tube diameters at higher liquid 3 m/s and air 2 m/s flow
rates are depicted in Fig. 11. In all the cases, core peaking is
observed. However, the profile becomes flatter with the increase
in the tube diameter. The figures bring out the diminishing wall
effect on the gaseous phase with the increase in the tube diameter.
As the change in tube diameter causes a change in the voidage
profile, it is not unwise to expect that the tube diameters may have
an effect on the flow regime transition. Flow regime boundary has
been constructed using the present simulation for the three different tube diameters, as shown in Fig. 12. There is a distinct upward
shift of the transition line with the increase in the conduit size.
The data available for the 25.4 mm 18 and 50.8 mm 22 tube
diameters exhibit reasonable agreement with the simulation results.
Conclusion
A two fluid model along with a population balance technique

13 has been employed for an in depth study of the hydrodynamics of the bubbly flow in a vertical conduit. The observations are
reported in two papers. In the present paper, the distribution of the
dispersed phase in radial plane and the effect of the inlet bubble
size and tube diameter on the transition to slug flow have been
investigated in details. The salient observations are enumerated
below as follows:
1. The capability of the model for predicting different peaked
structures of the void distribution in the bubbly flow has
been demonstrated. A regime map has been constructed,
showing near wall, wall, two, and core peaks with a gradual
increase in the superficial gas velocity. This diagram makes
a clear demonstration of the formation of slug flow from a
core peak rich in gas concentration.
2. The model has convincingly demonstrated that the phase
superficial velocities are not the sole parameters to dictate
the transition of the flow regimes. The inlet bubble size
plays a crucial role in the axial development of the bubbly
flow and its transition to slug flow. When uniform sized
bubbles are introduced at the inlet, the regime boundary experiences an upward shift with a decrease in the bubble size.
3. When bubbles of different sizes are considered at the inlet
plane, the shift in regime boundary becomes complex, as can
be seen from Fig. 7. The crossover between the different
regime boundaries probably indicates a change in the bubble
hydrodynamics at various ranges of the phase velocities.
4. Present simulation also depicts that the tube diameter has an
influence on both the void distribution and transition
boundary.
The present paper along with the companion paper 13 establishes the strength of the CFD technique in the simulation of complex two-phase hydrodynamics. Though the simulation is based
on an averaged formulation and uses no special model for turbulence, the results agree well with experimental data taken from a
diverse source. The simulation brings out a better physical picture
of the bubbly flow, which was hitherto impossible through control
volume based or one dimensional model. Though there is a scope
for the improvisation of the model, the need for new experimental
data for its validation cannot be overlooked.
References
Fig. 11 Effect of tube diameter on the void fraction profile at

high phase velocities
1 Serizawa, A., Kataoka, I., and Michiyoshi, I., 1975, Turbulence Structure of
Air-Water Bubbly FlowII Local Properties, Int. J. Multiphase Flow, 2, pp.
235246.
2 Sekoguchi, K., Fukui, H., and Sato, Y., 1981, Flow Characteristics and Heat
Transfer in Vertical Bubble Flow, Proceedings of Japan-U.S. Seminar on
SEPTEMBER 2009, Vol. 131 / 091304-5
Two-Phase Flow Dynamics, Hemisphere, Washington, DC, pp. 5974.

3 Zun, I., 1990, The Mechanism of Bubble Non-Homogeneous Distribution in
Two-Phase Shear Flow, Nucl. Eng. Des., 118, pp. 155162.
4 Song, Q., Luo, R., Yang, X. Y., and Wang, Z., 2001, Phase Distributions for
Upward Laminar Dilute Bubbly Flows With Non-Uniform Bubble Sizes in
Vertical Pipe, Int. J. Multiphase Flow, 27, pp. 379390.
5 Ishii, M., 1977, One Dimensional Drift Flux Model and Constitutive Equations for Relative Motion Between Phases in Various Two-Phase Flow Regimes, Argonne National Laboratory, Technical Report No. ANL-7747.
6 Drew, D. A., and Lahey, R. T., 1979, Application of General Constitutive
Principles to the Derivation of Multidimensional Two-Phase Flow Equations,
Int. J. Multiphase Flow, 5, pp. 243264.
7 Antal, S. P., Lahey, R. T., Jr., and Flaherty, J. E., 1991, Analysis of Phase
Distribution in Fully Developed Laminar Bubbly Two-Phase Flow, Int. J.
Multiphase Flow, 17, pp. 635652.
8 Troshko, A. E., and Hassan, Y. A., 2001, A Two-Equation Turbulence Model
of Turbulent Bubbly Flows, Int. J. Multiphase Flow, 27, pp. 19652000.
9 Politano, M. S., Carrica, P. M., and Converti, J., 2003, A Model for Turbulent
Polydisperse Two Phase Flow in Vertical Channels, Int. J. Multiphase Flow,
29, pp. 11531182.
10 Carrica, P. M., and Clausse, A. A., 1993, Mathematical Description of the
Critical Heat Flux as Nonlinear Dynamic Instability, Instabilities in Multiphase Flow, G. Gouesbet and A. Berlemont, eds., Plenum, New York.
11 Krepper, E., Lucas, D., and Prasser, H. M., 2005, On the Modelling of Bubbly Flow in Vertical Pipes, Nucl. Eng. Des., 235, pp. 597611.
12 Yeoh, G. H., and Tu, J. Y., 2006, Two-Fluid and Population Balance Models
for Subcooled Boiling Flow, Appl. Math. Model., 30, pp. 13701391.
13 Das, A. K., Das, P. K., and Thome, J. R., 2009, Transition of Bubbly Flow in
Vertical Tubes: New Criteria Through CFD Simulation, ASME J. Fluids Eng.,
In Press.
14 Ohnuki, A., and Akimoto, H., 2000, Experimental Study on Transition of
Flow Pattern and Phase Distribution in Upward Air-Water Two-Phase Flow
Along a Large Vertical Pipe, Int. J. Multiphase Flow, 263, pp. 367386.
15 Dukler, A. E., and Taitel, Y., 1977, Flow Regime Transitions for Vertical
Upward Gas Liquid Flow, Houston University, Technical Report No. 2,
NUREG-0163.
16 Fu, X. Y., and Ishii, M., 2003, Two-Group Interfacial Area Transport in Vertical Air-Water Flow I. Mechanistic Model, Nucl. Eng. Des., 219, pp. 143
168.
17 Bilicki, Z., and Kestin, J., 1988, Experimental Investigation of Certain Aspects of Upward Vertical Bubble and Slug Flows, Exp. Fluids, 6, pp. 455
460.
18 Vince, M. A., and Lahey, R. T., 1982, On the Development of an Objective
Flow Regime Indicator, Int. J. Multiphase Flow, 8, pp. 93124.
091304-6 / Vol. 131, SEPTEMBER 2009
19 Cheung, S. C. P., Yeoh, G. H., and Tu, J. Y., 2007, On the Modelling of
Population Balance in Isothermal Vertical Bubbly FlowsAverage Bubble
Number Density Approach, Chem. Eng. Process., 46, pp. 742756.
20 Lucas, D., Krepper, E., and Prasser, H. M., 2007, Use of Models for Lift,
Wall and Turbulent Dispersion Forces Acting on Bubbles for Poly-Disperse
Flows, Chem. Eng. Sci., 6215, pp. 41464157.
21 Prasser, H. M., Beyer, M., Carl, H., Gregor, S., Lucas, D., Pietruske, H.,
Schtz, P., and Weiss, F. P., 2007, Evolution of the Structure of a Gas-Liquid
Two Phase Flow in a Large Vertical Pipe, Nucl. Eng. Des., 237, pp. 1848
1861.
22 Taitel, Y., Bornea, D., and Dukler, A. E., 1980, Modelling Flow Pattern
Transitions for Steady Upward Gas-Liquid Flow in Vertical Tubes, AIChE J.,
263, pp. 345354.
23 Griffith, P., and Wallis, G. B., 1961, Two-Phase Slug Flow, ASME J. Heat
Transfer, 83, pp. 307320.
24 Mishima, K., and Ishii, M., 1984, Two-Fluid Model and Hydrodynamic Constitutive Relations, Nucl. Eng. Des., 8223, pp. 107126.
25 Sternling, V. C., 1965, Two-Phase Theory and Engineering Decision, Award
Lecture, AIChE Annual Meeting.
26 Govier, G. W., and Aziz, K., 1972, The Flow of Complex Mixtures in Pipes,
Van Nostrand Reinhold, New York, pp. 388389.
27 Gould, T. L., 1974, Vertical Two-Phase Steam Water Flow in Geothermal
Wells, J. Pet. Technol., 26, pp. 833842.
28 Oshinowo, T., and Charles, M. E., 1974, Vertical Two-Phase Flow. Part 1:
Flow Pattern Correlations, Can. J. Chem. Eng., 52, pp. 2535.
29 Bilicki, Z., and Kestin, J., 1987, Transition Criteria for Two Phase Flow
Patterns in Vertical Upward Flow, Int. J. Multiphase Flow, 13, pp. 283294.
30 Serizawa, A., and Kataoka, I., 1987, Phase Distribution in Two-Phase Flow,
Proceedings of ICHMT International Seminar on Transient Two-Phase Flow,
Dubrovnik, Yugoslavia, pp. 2430.
31 Lucas, D., Krepper, E., and Prasser, H. M., 2005, Development of Co-Current
Air-Water Flow in a Vertical Pipe, Int. J. Multiphase Flow, 31, pp. 1304
1328.
32 Rao, N. M., 2002, Investigations on Buoyancy Induced Circulation Loops,
Ph.D. thesis, IIT, Kharagpur, India.
33 Serizawa, A., and Kataoka, I., 1988, Phase Distribution in Two Phase Flow,
Transient Phenomena in Multiphase Flow, N. Afgan ed., Hemisphere, Washington, DC, pp. 179224.
34 Hibiki, T., and Ishii, M., 2000, Two-Group Interfacial Area Transport Equations at Bubbly-to-Slug Flow Transition, Nucl. Eng. Des., 202, pp. 3976.
35 Dukler, A. E., and Taitel, Y., 1986, Flow Pattern Transitions in Gas-Liquid
Systems: Measurement and Modeling, Multiphase Science and Technology,
Vol. 2, G. F. Hewitt, J. M. Delhaye, and N. Zuber, eds., Hemisphere, Washington, DC, pp. 194.
Analytical Solution for Newtonian

Laminar Flow Through the Concave
and Convex Ducts
M. Firouzi
e-mail: hashemabadi@iust.ac.ir
Computational Fluid Dynamics Research Laboratory,
School of Chemical Engineering,
Iran University of Science and Technology,
Narmak, Tehran 16846, Iran
In this paper, the motion equation for steady state, laminar, fully
developed flow of Newtonian fluid through the concave and convex ducts has been solved both numerically and analytically.
These cross sections can be formed due to the sedimentation of
heavy components such as sand, wax, debris, and corrosion products in pipe flows. The influence of duct cross section on dimensionless velocity profile, dimensionless pressure drop, and friction
factor has been reported. Finally based on the analytical solutions
three new correlations have been proposed for the product of
Reynolds number and Fanning friction factor C f Re for these
geometries. DOI: 10.1115/1.3184026
Keywords: analytical, numerical, Newtonian, concave, convex,
noncircular passage, Fanning friction factor, sedimentation
Introduction
A range of noncircular duct configurations is employed for internal flow and heat transfer applications in chemical, petroleum,
pharmaceutical, food, and plastics industries. It is of great practical interest to be able to predict the pressure drop and heat transfer
characteristics of fluid flows in different duct geometries. One of
the applications of fluid flow in irregular ducts is in the piping
systems that encountered with sedimentation, and most fluid flows
in environmental and industrial applications are directly influenced by sediment. Sediment transport is important to river,
shoreline, and harbor projects in piping systems. It can cause to
form geometric domains that are far from ideal in shape. It is not
difficult to imagine subsea pipelines blocked by accumulated wax
or by hydrate plugs, bundled pipes with debris settlement, or
heavily clogged cross sections with grease or hard water scaling
Fig. 1 1. Few exact solutions are available only for simple fluid
models and circular pipe cross sections. Many attempts were
made in modeling Newtonian and non-Newtonian fluids in various noncircular cross sectional ducts, which most of them are
numerical or experimental. Several investigations have been reported for flow characteristics in trapezoidal cross section 24,
in rectangular cross section 57, non-Newtonian fluid flow
through triangular channels 8,9, friction factor, pressure drop,
and Nusselt number of Newtonian fluid flowing in ducts of arbitrary cross sections 10, and an approximate solution for determining the pressure drop in microchannels of arbitrary cross section 11.
Here we focused on analytical solution for laminar fully developed Newtonian fluid flow through a family of noncircular passage with concave and convex cross sections, which is selected
JOURNAL OF FLUIDS ENGINEERING. Manuscript received May 20, 2008; final manuscript
received June 21, 2009; published online August 14, 2009. Assoc. Editor: Ye Zhou.
Mathematical Modeling
2.1
S. H. Hashemabadi
for the solid buildup boundary family of curves, since some of

solids such as wax surfaces are expected to be more curved than
flat and this buildup increases with time.
Analytical Solution
2.1.1 Velocity Profile. Figure 2a shows cross section of three

different noncircular horizontal ducts. The appropriate coordinate
system for solving the flow problem in this case is the well-known
bipolar coordinate system. Coordinate represents the view angle
of the two poles from an arbitrary point in the flow domain, and
coordinate relates to the ratio of the radius vectors r1 and r2,
= lnr1 / r2, where r1 and r2, as illustrated in Fig. 2b, are distances of arbitrary point to two poles. The transformation functions for Cartesian and bipolar coordinates are 12 as follows:
x=
R sin 1 sinh
,
cosh cos
y=
R sin 1 sin
cosh cos
For steady state, laminar, and fully developed Newtonian fluid

flow, while all body forces are negligible the motion equation can
be simplified as follows 13:
cosh cos
sin 1
2v 2v
+
=4
2 2
where the velocity has been nondimensionalized by the maximum

velocity of Newtonian fluid flow through the pipe vmax
= R2 / 4dp / dz. No-slip condition on boundaries can be defined as follows:
v1 = v2 = v= = 0
where 2 is the bottom wall angle Fig. 2a. Equation 1 is

nonhomogeneous, so the general solution of dimensionless velocity profile v is composed of particular solution vp and the homogeneous solution vh:
v = vh + vp
The velocity profile of Newtonian fluid through the circular pipe

is used as a particular solution. Therefore the particular solution
vp in the bipolar coordinate is 15
vp =
2 sin 1sin1
cosh cos
By substituting the particular solution into Eq. 1, the following

homogeneous equation can be obtained:
2vh 2vh
+
=0
2
2
And for boundary conditions, we have
vh=2 +
vh=1 = 0
vh= = 0
2 sin2 1sin1
=0
cosh cos 2
The homogeneous solution vh can be obtained by using Fourier

integral method 14:
vh =
sinh 1cosd
Obviously this solution satisfies Eqs. 5 and 6. Equation 7 is

an underlying requirement for the legitimate use of Fourier integral in obtaining the solution. By substituting Eq. 9 into Eq. 8
we have
SEPTEMBER 2009, Vol. 131 / 094501-1
2 sinh cosh cos

cosd = 1 13
sin sinh
which for the special case of flat bottom 2 = :
2cosh cos
cosd = 1
sinh
14
While the bottom is convex or concave 2 , by comparing

Eq. 11 with Eqs. 13 and 14, the nondimensionalized spectral
function can be derived as follows:
2K sinh2 cosh cos 2

sin 2 sinhsinh2 1
15
and for plane bottom 2 = ,

Fig. 1 Schematic descriptions of a flow passage cross section and b bipolar coordinate
2K1 + cosh
sinhsinh 1
16
Finally the dimensionless velocity distribution can be obtained:
sinh2 1cosd +
2 sin2 1sin1
=0
cosh cos 2
v , =
cosd = 1
11
where
= sinh2 1 ,
K =
sinh 1cosd
+ 2 sin1
This can be simplified to
10
sin 1
cosh cos
17
2.1.2 Pressure Drop. The corresponding flow rates of fluid

can be obtained by integrating the velocity over the flow area; to
calculate the pressure drop, the flow rate is assumed to be equal to
the flow rate of Newtonian flow in the circular cross section duct:
2 sin2 1sin1
cosh cos 2
Q=
12
for curved bottom 2

To calculate the spectral function
, the following definite integral is used 12:
v,
R sin 1
cosh cos
dd
R4 sin2 1 dp
F1, 2
dz
4
18
19
where
+
F1, 2 =
v ,
dd
cosh cos 2
The flow pressure drop through the noncircular ducts can be calculated from Eq. 18:
4Q
dp
F1, 2
=
dz R4 sin2 1
20
The pressure gradient, Eq. 20, can be nondimensionalized with

HagenPoiseuille relationship for the pressure drop of Newtonian
fluid flow through the pipes:
dp
dp
dz
=
=
dz
dp
2 sin21F1, 2
dz pipe
Fig. 2 Some examples of the clogged pipe: a mud sediment

in pipes, b clogged pipe with grease, c fouling in heat exchangers, and d deposit of the sludge pipe 2
094501-2 / Vol. 131, SEPTEMBER 2009
21
2.1.3 Friction Factor Calculation. Once the problem is solved

for the velocity profile, the local shear stress can be derived, the
pipe walls are described by surfaces where = const, or in other
words in the bipolar coordinate system, this boundary is isolines
of coordinate 14,15, hence the shear stress can be calculated
as
z =
1 cosh cos v,
2
sin 1
22
where the shear stress is nondimensionalized by R / 2dp / dz. By

substituting the dimensionless velocity derivative, the following
can be obtained:
C f Re = 4
sin sin 1
cosh cos
cosh cos
2 sin 1
cosh 1d 23
v =

D2h
R
z , 1
Dh =
z, 2
2R 1 + 1/2 sin21
1 + sin1
28
1
2 + sin22
2
sin 1
1 + 2
sin 2
v ,
z, 2
24
R sin 1
d
cosh cos 1
R sin 1
d
cosh cos 2
R sin 1
cosh cos
R sin 1
d
cosh cos
dd
v =
25
dd
26
v
,
R2 dp
4 dz
x =
x
,
D
y =
27
y
D
32
The top and bottom walls are subjected to no-slip condition.
sin 1
1
2R 1 + sin21
2
sin 2
1
A
R sin 1
v ,
cosh cos
Dh
R
where A and P are the area and the perimeter of the duct cross
section, respectively. By substituting Eqs. 25 and 26 into Eq.
24, it yields the following:
R sin 1
d +
cosh cos
and for curved bottom 2 ,
z, 1
For flat bottom 2 = , the hydraulic diameter can be calculated

as follows:
Dh =
C f Re =
1
P
+
Then the friction factor, which is one of the most important characteristics in fluid flows, can be calculated. The definition of Fanning friction factor is 13 as follows:
w
v
Here w and v are the average dimensionless wall shear stress

and average dimensionless velocity, respectively, and can be obtained as follows:
w =
z = cos 1
Figures 3 and 4 show the comparison of dimensionless velocity

profile that is obtained from numerical and analytical solutions for
two different cross sections 1 = / 3 and 2 = and semicircular
cross section duct 1 = / 2 and 2 = as expected by increasing
Fig. 4; the velocity approaches zero in the vicinity of the wall.
29
2.2 Numerical Solution. In this section, to recheck the results
of the analytical solution because of insufficiency of related established results for more validation of obtained analytical results,
the momentum equation is solved numerically with finite element
method FEM. The momentum equation is preferred to be solved
in the Cartesian coordinate system, which, by considering the
mentioned assumptions in analytical solution, can be simplified as
follows:
zx zy dp
+
+
=0
x
y
dz
30
The nondimensional form of the motion equation Eq. 30 for

Newtonian fluid is as follows:
v
v
= 16
+
x x
y x
where
31
Fig. 3 Comparison of numerical and analytical dimensionless

velocity profiles for the case of the semicircle cross section
1 = / 2 and 2 = at = 0 along the meridian
SEPTEMBER 2009, Vol. 131 / 094501-3
Fig. 4 Comparison of numerical and analytical solutions for

the dimensionless velocity profile at = 2 / 3 with cross section specification: 1 = / 3 and 2 =
Fig. 6 Velocity variation with for different upper wall shapes,

2 = 2 / 3
sin 1 sin 2 sin 1

h
= 0.5
D
1 cos 2
1
tan
2
33
The analytical and numerical results illustrate a good agreement.

Figure 5 shows the comparison of numerical and analytical solutions for dimensionless axial velocity profile in different cross
section ducts concave, plane, and convex; as is distinct they are
also in good agreement with the analytical solution less than
0.8% of error on the average. As well, the results show that by
increasing the concavity the velocity promotes and it decreases by
changing the lower wall shape to convexity. Figure 6 depicts the
axial nondimensional velocity profile variation with ; as expected
the maximum dimensionless velocity occurs at = 0 and the velocity decreases by increasing and finally it approaches zero and
indeed it validates the no-slip boundary condition at poles.
Figure 7 illustrates the dimensionless pressure drop versus duct
aspect ratio h / D in different cross section shapes. The duct aspect ratio h / D can be related to 1 and 2 as follows:
Obviously while the duct aspect ratio is equal to 1 circular cross

section the dimensionless pressure drop approaches 1, and the
reduction in aspect ratio decreasing flow area for a given flow
rate affects an increase in the pressure drop and this promotion is
more noticeable for the concave lower wall. 20% irregularity in
the cross section h / D = 0.8 for the case of 2 = 2 / 3 compared
to a circular cross section causes an approximately 30% increase
in pressure drop, which is important in analyzing the pressure
drop of the flow through the clogged pipe.
The product of Reynolds number and Fanning friction factor
C f Re can be obtained for different cross sections by using Eq.
27. For the special case of a circular cross section 1 = 0 C f Re
equals to 16 and for the case of semicircle ducts 1 = / 2 and
2 = the product of Reynolds number and Fanning friction factor C f Re is calculated to be 15.768891, which matches well
with the reported numerical solution C f Re= 15.768832 16.
Tables 1 and 2 illustrate the product of Reynolds number and
Fanning friction factor C f Re for some of these cross sections.
While the upper wall view angle 1 of flat bottom ducts 2
Fig. 5 Comparison of numerical and analytical solutions for

various duct cross sections due to different bottom wall
shapes: 1 = / 3 and = 0
Fig. 7 Variation in the dimensionless pressure drop with aspect ratio h / D for various duct cross sections different 1
and 2
094501-4 / Vol. 131, SEPTEMBER 2009
Table 1 Cf Re for some of duct cross section with plane bottom 2 =
= approaches Table 1, the product of Reynolds number and

Fanning friction factor C f Re tends to the value for the two
parallel plates that equals to 24.
Due to the complexity of analytically solving the integrals into
Eq. 27, two simple correlations for calculating the Fanning friction factor for the Newtonian fluid flow through the ducts with an
irregular cross section are proposed, which is applied by process
designers and analyzers. For the case of the lower plane wall
2 = ,
C f Re = 16 1 +
n
n 1
n=1
and for the curved bottom wall 2 ,
C f Re = 16 1 +
n
n 1
+ 12a4 + a52 + a61
n=1
34
35
Table 3 shows all coefficients of Eqs. 34 and 35. In other

words, we have proposed two geometry correction factors that can
convert the product of Reynolds number and Fanning friction factor C f Re for the laminar flow through the pipes to the laminar
flow through the concave and convex ducts. Figure 8 depicts the
correlation Eqs. 34 and 35 accuracy with respect to analytical
solution results; as seen, they are in very good agreement.
Conclusion
An analytical study was carried out to calculate the velocity

profile, pressure drop, and friction factor of steady state, laminar,
Table 2 Cf Re for some of duct cross section with concave

2 > and convex 2 < bottoms
and fully developed flow of Newtonian fluid through the concave

and convex ducts. The analytical solution has been validated by
the numerical solution of the simplified motion equation with
above mentioned assumptions. The results show that the dimensionless velocity profiles are in good agreement with the analytical solution. As a result, by increasing the concavity of the lower
wall for a given upper wall angle, the velocity profile increases
and by increasing the lower wall convexity the velocity decreases.
The product of Reynolds number and Fanning friction factor
C f Re has been calculated for the different duct geometries,
which for the special case of a circular cross section 1 = 0
C f Re equals to 16 and by increasing the upper wall view angle
1 for flat bottom ducts, the results show that C f Re approaches 24, which is the C f Re value for the two parallel plates.
For simple application of this work, three analytical based correlations have been proposed for three different cross sections flat,
concave, and convex bottom. The prediction of correlations has
good accuracy with respect to analytical solution results.
Nomenclature
A
Cf
D
Dh
F
h
L
p
P
Q
flow area
Fanning friction factor
pipe diameter
hydraulic diameter
defined by Eq. 19
duct height
pipe length
pressure
perimeter of the duct
flow rate
SEPTEMBER 2009, Vol. 131 / 094501-5
Table 3 Coefficients of Eqs. 34 and 35

a
Bottom angle
2 =
2
2
R
Re
v
x
y
Y
z
0.180
2.170
0.191
1.576
0.216
0.356
4.970
0.096
0.003
7.467
1.449
0.200
6.021
0.252
0.043
2.673
0.136
0.112
0.616
-
0.057
-
pipe radius
Reynolds number
velocity
related to the Cartesian coordinate
related to the Cartesian coordinate
distance from the bottom wall
axial direction
Greek Symbols
dynamic viscosity
ratio of the radius vectors in the bipolar
coordinate
density
stress
view angle of the interface in the bipolar
coordinate
Subscripts
1 upper wall
2 bottom wall
max maximum
Superscripts
dimensionless property
Fig. 8 Comparison of analytical solution and correlation results for Cf Re
094501-6 / Vol. 131, SEPTEMBER 2009
- average value
h homogeneous
p particular
References
1 Barat, R., Bouzas, A., Mart, N., Ferrer, J., and Seco, A., 2009, Precipitation
Assessment in Wastewater Treatment Plants Operated for Biological Nutrient
Removal: A Case Study in Murcia, Spain, J. Environ. Manage., 90, pp.
850857.
2 Wu, P. Y., and Little, W. A., 1983, Measurement of Friction Factor of the
Flow of Gases in Very Fine Channels Used for Microminiaturize JouleThomson Refrigerators, Cryogenics, 248, pp. 273277.
3 Wu, H. Y., and Cheng, P., 2003, Friction Factors in Smooth Trapezoidal
Silicon Microchannels With Different Aspect Ratios, Int. J. Heat Mass Transfer, 46, pp. 25192525.
4 Weilin, Q., Mala, G. M., and Dongqing, L., 2000, Pressure-Driven Water
Flows in Trapezoidal Silicon Microchannels, Int. J. Heat Mass Transfer,
433, pp. 353364.
5 Peng, X. F., Peterson, G. P., and Wang, B. X., 1994, Frictional Flow Characteristics of Water Flowing Through Microchannels, Exp. Heat Transfer, 74,
pp. 249264.
6 Pfund, D., Rector, D., Shekarriz, A., Popescu, A., and Welty, J., 2000, Pressure Drop Measurements in a Micro Channel, AIChE J., 468, pp. 1496
1507.
7 Hrnjak, P., and Tu, X., 2007, Single Phase Pressure Drop in Microchannels,
Int. J. Heat Fluid Flow, 281, pp. 214.
8 Etemad, S. Gh., Mujumdar, A. S., and Nassef, R., 1996, Simultaneously
Developing Flow and Heat Transfer of Non-Newtonian Fluids in Equilateral
Triangular Duct, Appl. Math. Model., 2012, pp. 898908.
9 Hashemabadi, S. H., Etemad, S. Gh., Golkar Naranji, M. R., and Thibault, J.,
2003, Laminar Flow of Non-Newtonian in Right Triangular Ducts, Int.
Commun. Heat Mass Transfer, 301, pp. 5360.
10 Etemad, S. Gh., and Bakhtiari, F., 1999, General Equations for Fully Developed Fluid Flow and Heat Transfer Characteristics in Complex Geometries,
Int. Commun. Heat Mass Transfer, 262, pp. 229238.
11 Bahrami, M., Yovanovich, M., and Richard Culham, J., 2007, A Novel Solution for Pressure Drop in Singly Connected Microchannels of Arbitrary CrossSection, Int. J. Heat Mass Transfer, 501314, pp. 24922502.
12 Firouzi, M., and Hashemabadi, S. H., 2008, Analytical Solution for
Newtonian-Bingham Plastic Two-Phase Pressure Driven Stratified Flow
Through the Circular Ducts, Int. Commun. Heat Mass Transfer, 35, pp. 666
673.
13 Bird, R. B., Armstrong, R. C., and Hassager, O., 1978, Dynamics of Polymeric
Liquids, Vol. 1, Wiley, New York, pp. 592595.
14 Brauner, N., Rovinsky, J., and Maron, D. M., 1996, Analytical Solution for
Laminar-Laminar Two Phase Stratified Flow in Circular Conduit, Chem. Eng.
Sci., 141, pp. 103143.
15 Biberg, D., and Halvorsen, G., 2000, Wall and Interfacial Shear Stress in
Pressure Driven Two-Phase Laminar Stratified Pipe Flow, Int. J. Multiphase
Flow, 2610, pp. 16451673.
16 Syrjl, S., 2002, Accurate Prediction of Friction Factor and Nusselt Number
for Some Duct Flows of Power-Law Non-Newtonian Fluid, Numer. Heat
Transfer, Part A, 411, pp. 89100.

J Fluids Engineering 2009 Vol 131 N9

Uploaded by

Copyright:

Available Formats

J Fluids Engineering 2009 Vol 131 N9

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 N9

Uploaded by

Copyright:

Available Formats

Journal of

VOLUME 131 NUMBER 9 SEPTEMBER 2009

Flow Characteristics in a Curved Rectangular Channel With Variable

Measurements of a Wall Jet Impinging Onto a Forward Facing Step

Fundamental Issues and Canonical Flows

Large-Eddy Simulation of Wake and Boundary Layer Interactions

On the Pulsed and Transitional Behavior of an Electrified Fluid

Curved Non-Newtonian Liquid Jets With Surfactants

Circulation Generation and Vortex Ring Formation by Conic Nozzles

Transactions of the ASME, Journal of Fluids Engineering

Measurements of High Velocity Gradient Flow Using Bubble Tracers

Thermodynamic Effect on Rotating Cavitation in an Inducer

Transition of Bubbly Flow in Vertical Tubes: New Criteria Through

Transition of Bubbly Flow in Vertical Tubes: Effect of Bubble Size

Contents continued on inside back cover

Volume 131, Number 9

The ASME Journal of Fluids Engineering is abstracted and indexed in

Pharyngeal Airflow Analysis in

It is estimated that obstructive sleep apnea affects 2% of middle

Journal of Fluids Engineering

Recent studies using three-dimensional 3D imaging have

Copyright 2009 by ASME

SEPTEMBER 2009, Vol. 131 / 091101-1

pressure in the airway is affected by surgical treatment, it does not

tive positions of points A and B in relation to nasion were

Fig. 2 Landmarks used to determine surgical movement

between 118,878 and 319,193 tetrahedral elements depending on

Fig. 3 Typical superior and inferior borders of pharyngeal

Journal of Fluids Engineering

Fig. 4 Outline of airway using

SEPTEMBER 2009, Vol. 131 / 091101-3

Fig. 7 3D mesh model showing tetrahedral elements

model of the airway

The quantity of maxillary and mandibular surgical advancement

091101-4 / Vol. 131, SEPTEMBER 2009

b Model with some

Transactions of the ASME

Table 1 Quantity of surgical advancement mm

stenosis, along the airway, noted by peaks in the Re plots. For

Change in Hydraulic Diameter

Postdiameter minus Prediameter [mm]

Fig. 8 Change in hydraulic diameter along pharyngeal airway

Fig. 9 Reynolds number along pharyngeal airway of patient 1

Journal of Fluids Engineering

SEPTEMBER 2009, Vol. 131 / 091101-5

Patient 2 Reynolds Number Variation

Fig. 10 Reynolds number along pharyngeal airway of patient 2

larger volume of air is trying to pass through a confined space. In

bulent flow, there is increased mass and energy transfer normal to

Patient 3 Reynolds Number Variation

Fig. 11 Reynolds number along pharyngeal airway of patient 3

091101-6 / Vol. 131, SEPTEMBER 2009

Transactions of the ASME

Patient 4 Reynolds Number Variation

Fig. 12 Reynolds number along pharyngeal airway of patient 4

diameter opens up constricted areas and reduces the pressure drag

viscosity coefficient were found in areas of stenosis, as well as

Fig. 13 Eddy viscosity coefficient contour with streaklines

SEPTEMBER 2009, Vol. 131 / 091101-7